| @@ -1,2912 +1,2898 @@ | | | @@ -1,2912 +1,2898 @@ |
1 | /* $NetBSD: pmap.c,v 1.157 2012/01/28 08:57:09 cherry Exp $ */ | | 1 | /* $NetBSD: pmap.c,v 1.158 2012/01/29 11:37:08 cherry Exp $ */ |
2 | | | 2 | |
3 | /*- | | 3 | /*- |
4 | * Copyright (c) 2008, 2010 The NetBSD Foundation, Inc. | | 4 | * Copyright (c) 2008, 2010 The NetBSD Foundation, Inc. |
5 | * All rights reserved. | | 5 | * All rights reserved. |
6 | * | | 6 | * |
7 | * This code is derived from software contributed to The NetBSD Foundation | | 7 | * This code is derived from software contributed to The NetBSD Foundation |
8 | * by Andrew Doran. | | 8 | * by Andrew Doran. |
9 | * | | 9 | * |
10 | * Redistribution and use in source and binary forms, with or without | | 10 | * Redistribution and use in source and binary forms, with or without |
11 | * modification, are permitted provided that the following conditions | | 11 | * modification, are permitted provided that the following conditions |
12 | * are met: | | 12 | * are met: |
13 | * 1. Redistributions of source code must retain the above copyright | | 13 | * 1. Redistributions of source code must retain the above copyright |
14 | * notice, this list of conditions and the following disclaimer. | | 14 | * notice, this list of conditions and the following disclaimer. |
15 | * 2. Redistributions in binary form must reproduce the above copyright | | 15 | * 2. Redistributions in binary form must reproduce the above copyright |
16 | * notice, this list of conditions and the following disclaimer in the | | 16 | * notice, this list of conditions and the following disclaimer in the |
17 | * documentation and/or other materials provided with the distribution. | | 17 | * documentation and/or other materials provided with the distribution. |
18 | * | | 18 | * |
19 | * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS | | 19 | * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS |
20 | * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED | | 20 | * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED |
21 | * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR | | 21 | * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR |
22 | * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS | | 22 | * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS |
23 | * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR | | 23 | * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR |
24 | * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF | | 24 | * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF |
25 | * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS | | 25 | * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS |
26 | * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN | | 26 | * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN |
27 | * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) | | 27 | * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) |
28 | * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE | | 28 | * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE |
29 | * POSSIBILITY OF SUCH DAMAGE. | | 29 | * POSSIBILITY OF SUCH DAMAGE. |
30 | */ | | 30 | */ |
31 | | | 31 | |
32 | /* | | 32 | /* |
33 | * Copyright (c) 2007 Manuel Bouyer. | | 33 | * Copyright (c) 2007 Manuel Bouyer. |
34 | * | | 34 | * |
35 | * Redistribution and use in source and binary forms, with or without | | 35 | * Redistribution and use in source and binary forms, with or without |
36 | * modification, are permitted provided that the following conditions | | 36 | * modification, are permitted provided that the following conditions |
37 | * are met: | | 37 | * are met: |
38 | * 1. Redistributions of source code must retain the above copyright | | 38 | * 1. Redistributions of source code must retain the above copyright |
39 | * notice, this list of conditions and the following disclaimer. | | 39 | * notice, this list of conditions and the following disclaimer. |
40 | * 2. Redistributions in binary form must reproduce the above copyright | | 40 | * 2. Redistributions in binary form must reproduce the above copyright |
41 | * notice, this list of conditions and the following disclaimer in the | | 41 | * notice, this list of conditions and the following disclaimer in the |
42 | * documentation and/or other materials provided with the distribution. | | 42 | * documentation and/or other materials provided with the distribution. |
43 | * | | 43 | * |
44 | * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR | | 44 | * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR |
45 | * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES | | 45 | * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES |
46 | * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. | | 46 | * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. |
47 | * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, | | 47 | * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, |
48 | * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT | | 48 | * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT |
49 | * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, | | 49 | * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, |
50 | * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY | | 50 | * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY |
51 | * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT | | 51 | * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT |
52 | * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF | | 52 | * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF |
53 | * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. | | 53 | * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. |
54 | * | | 54 | * |
55 | */ | | 55 | */ |
56 | | | 56 | |
57 | /* | | 57 | /* |
58 | * Copyright (c) 2006 Mathieu Ropert <mro@adviseo.fr> | | 58 | * Copyright (c) 2006 Mathieu Ropert <mro@adviseo.fr> |
59 | * | | 59 | * |
60 | * Permission to use, copy, modify, and distribute this software for any | | 60 | * Permission to use, copy, modify, and distribute this software for any |
61 | * purpose with or without fee is hereby granted, provided that the above | | 61 | * purpose with or without fee is hereby granted, provided that the above |
62 | * copyright notice and this permission notice appear in all copies. | | 62 | * copyright notice and this permission notice appear in all copies. |
63 | * | | 63 | * |
64 | * THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES | | 64 | * THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES |
65 | * WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF | | 65 | * WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF |
66 | * MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR | | 66 | * MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR |
67 | * ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES | | 67 | * ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES |
68 | * WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN | | 68 | * WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN |
69 | * ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF | | 69 | * ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF |
70 | * OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE. | | 70 | * OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE. |
71 | */ | | 71 | */ |
72 | | | 72 | |
73 | /* | | 73 | /* |
74 | * Copyright (c) 1997 Charles D. Cranor and Washington University. | | 74 | * Copyright (c) 1997 Charles D. Cranor and Washington University. |
75 | * All rights reserved. | | 75 | * All rights reserved. |
76 | * | | 76 | * |
77 | * Redistribution and use in source and binary forms, with or without | | 77 | * Redistribution and use in source and binary forms, with or without |
78 | * modification, are permitted provided that the following conditions | | 78 | * modification, are permitted provided that the following conditions |
79 | * are met: | | 79 | * are met: |
80 | * 1. Redistributions of source code must retain the above copyright | | 80 | * 1. Redistributions of source code must retain the above copyright |
81 | * notice, this list of conditions and the following disclaimer. | | 81 | * notice, this list of conditions and the following disclaimer. |
82 | * 2. Redistributions in binary form must reproduce the above copyright | | 82 | * 2. Redistributions in binary form must reproduce the above copyright |
83 | * notice, this list of conditions and the following disclaimer in the | | 83 | * notice, this list of conditions and the following disclaimer in the |
84 | * documentation and/or other materials provided with the distribution. | | 84 | * documentation and/or other materials provided with the distribution. |
85 | * | | 85 | * |
86 | * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR | | 86 | * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR |
87 | * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES | | 87 | * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES |
88 | * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. | | 88 | * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. |
89 | * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, | | 89 | * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, |
90 | * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT | | 90 | * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT |
91 | * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, | | 91 | * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, |
92 | * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY | | 92 | * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY |
93 | * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT | | 93 | * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT |
94 | * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF | | 94 | * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF |
95 | * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. | | 95 | * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. |
96 | */ | | 96 | */ |
97 | | | 97 | |
98 | /* | | 98 | /* |
99 | * Copyright 2001 (c) Wasabi Systems, Inc. | | 99 | * Copyright 2001 (c) Wasabi Systems, Inc. |
100 | * All rights reserved. | | 100 | * All rights reserved. |
101 | * | | 101 | * |
102 | * Written by Frank van der Linden for Wasabi Systems, Inc. | | 102 | * Written by Frank van der Linden for Wasabi Systems, Inc. |
103 | * | | 103 | * |
104 | * Redistribution and use in source and binary forms, with or without | | 104 | * Redistribution and use in source and binary forms, with or without |
105 | * modification, are permitted provided that the following conditions | | 105 | * modification, are permitted provided that the following conditions |
106 | * are met: | | 106 | * are met: |
107 | * 1. Redistributions of source code must retain the above copyright | | 107 | * 1. Redistributions of source code must retain the above copyright |
108 | * notice, this list of conditions and the following disclaimer. | | 108 | * notice, this list of conditions and the following disclaimer. |
109 | * 2. Redistributions in binary form must reproduce the above copyright | | 109 | * 2. Redistributions in binary form must reproduce the above copyright |
110 | * notice, this list of conditions and the following disclaimer in the | | 110 | * notice, this list of conditions and the following disclaimer in the |
111 | * documentation and/or other materials provided with the distribution. | | 111 | * documentation and/or other materials provided with the distribution. |
112 | * 3. All advertising materials mentioning features or use of this software | | 112 | * 3. All advertising materials mentioning features or use of this software |
113 | * must display the following acknowledgement: | | 113 | * must display the following acknowledgement: |
114 | * This product includes software developed for the NetBSD Project by | | 114 | * This product includes software developed for the NetBSD Project by |
115 | * Wasabi Systems, Inc. | | 115 | * Wasabi Systems, Inc. |
116 | * 4. The name of Wasabi Systems, Inc. may not be used to endorse | | 116 | * 4. The name of Wasabi Systems, Inc. may not be used to endorse |
117 | * or promote products derived from this software without specific prior | | 117 | * or promote products derived from this software without specific prior |
118 | * written permission. | | 118 | * written permission. |
119 | * | | 119 | * |
120 | * THIS SOFTWARE IS PROVIDED BY WASABI SYSTEMS, INC. ``AS IS'' AND | | 120 | * THIS SOFTWARE IS PROVIDED BY WASABI SYSTEMS, INC. ``AS IS'' AND |
121 | * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED | | 121 | * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED |
122 | * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR | | 122 | * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR |
123 | * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL WASABI SYSTEMS, INC | | 123 | * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL WASABI SYSTEMS, INC |
124 | * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR | | 124 | * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR |
125 | * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF | | 125 | * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF |
126 | * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS | | 126 | * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS |
127 | * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN | | 127 | * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN |
128 | * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) | | 128 | * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) |
129 | * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE | | 129 | * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE |
130 | * POSSIBILITY OF SUCH DAMAGE. | | 130 | * POSSIBILITY OF SUCH DAMAGE. |
131 | */ | | 131 | */ |
132 | | | 132 | |
133 | /* | | 133 | /* |
134 | * This is the i386 pmap modified and generalized to support x86-64 | | 134 | * This is the i386 pmap modified and generalized to support x86-64 |
135 | * as well. The idea is to hide the upper N levels of the page tables | | 135 | * as well. The idea is to hide the upper N levels of the page tables |
136 | * inside pmap_get_ptp, pmap_free_ptp and pmap_growkernel. The rest | | 136 | * inside pmap_get_ptp, pmap_free_ptp and pmap_growkernel. The rest |
137 | * is mostly untouched, except that it uses some more generalized | | 137 | * is mostly untouched, except that it uses some more generalized |
138 | * macros and interfaces. | | 138 | * macros and interfaces. |
139 | * | | 139 | * |
140 | * This pmap has been tested on the i386 as well, and it can be easily | | 140 | * This pmap has been tested on the i386 as well, and it can be easily |
141 | * adapted to PAE. | | 141 | * adapted to PAE. |
142 | * | | 142 | * |
143 | * fvdl@wasabisystems.com 18-Jun-2001 | | 143 | * fvdl@wasabisystems.com 18-Jun-2001 |
144 | */ | | 144 | */ |
145 | | | 145 | |
146 | /* | | 146 | /* |
147 | * pmap.c: i386 pmap module rewrite | | 147 | * pmap.c: i386 pmap module rewrite |
148 | * Chuck Cranor <chuck@netbsd> | | 148 | * Chuck Cranor <chuck@netbsd> |
149 | * 11-Aug-97 | | 149 | * 11-Aug-97 |
150 | * | | 150 | * |
151 | * history of this pmap module: in addition to my own input, i used | | 151 | * history of this pmap module: in addition to my own input, i used |
152 | * the following references for this rewrite of the i386 pmap: | | 152 | * the following references for this rewrite of the i386 pmap: |
153 | * | | 153 | * |
154 | * [1] the NetBSD i386 pmap. this pmap appears to be based on the | | 154 | * [1] the NetBSD i386 pmap. this pmap appears to be based on the |
155 | * BSD hp300 pmap done by Mike Hibler at University of Utah. | | 155 | * BSD hp300 pmap done by Mike Hibler at University of Utah. |
156 | * it was then ported to the i386 by William Jolitz of UUNET | | 156 | * it was then ported to the i386 by William Jolitz of UUNET |
157 | * Technologies, Inc. Then Charles M. Hannum of the NetBSD | | 157 | * Technologies, Inc. Then Charles M. Hannum of the NetBSD |
158 | * project fixed some bugs and provided some speed ups. | | 158 | * project fixed some bugs and provided some speed ups. |
159 | * | | 159 | * |
160 | * [2] the FreeBSD i386 pmap. this pmap seems to be the | | 160 | * [2] the FreeBSD i386 pmap. this pmap seems to be the |
161 | * Hibler/Jolitz pmap, as modified for FreeBSD by John S. Dyson | | 161 | * Hibler/Jolitz pmap, as modified for FreeBSD by John S. Dyson |
162 | * and David Greenman. | | 162 | * and David Greenman. |
163 | * | | 163 | * |
164 | * [3] the Mach pmap. this pmap, from CMU, seems to have migrated | | 164 | * [3] the Mach pmap. this pmap, from CMU, seems to have migrated |
165 | * between several processors. the VAX version was done by | | 165 | * between several processors. the VAX version was done by |
166 | * Avadis Tevanian, Jr., and Michael Wayne Young. the i386 | | 166 | * Avadis Tevanian, Jr., and Michael Wayne Young. the i386 |
167 | * version was done by Lance Berc, Mike Kupfer, Bob Baron, | | 167 | * version was done by Lance Berc, Mike Kupfer, Bob Baron, |
168 | * David Golub, and Richard Draves. the alpha version was | | 168 | * David Golub, and Richard Draves. the alpha version was |
169 | * done by Alessandro Forin (CMU/Mach) and Chris Demetriou | | 169 | * done by Alessandro Forin (CMU/Mach) and Chris Demetriou |
170 | * (NetBSD/alpha). | | 170 | * (NetBSD/alpha). |
171 | */ | | 171 | */ |
172 | | | 172 | |
173 | #include <sys/cdefs.h> | | 173 | #include <sys/cdefs.h> |
174 | __KERNEL_RCSID(0, "$NetBSD: pmap.c,v 1.157 2012/01/28 08:57:09 cherry Exp $"); | | 174 | __KERNEL_RCSID(0, "$NetBSD: pmap.c,v 1.158 2012/01/29 11:37:08 cherry Exp $"); |
175 | | | 175 | |
176 | #include "opt_user_ldt.h" | | 176 | #include "opt_user_ldt.h" |
177 | #include "opt_lockdebug.h" | | 177 | #include "opt_lockdebug.h" |
178 | #include "opt_multiprocessor.h" | | 178 | #include "opt_multiprocessor.h" |
179 | #include "opt_xen.h" | | 179 | #include "opt_xen.h" |
180 | #if !defined(__x86_64__) | | 180 | #if !defined(__x86_64__) |
181 | #include "opt_kstack_dr0.h" | | 181 | #include "opt_kstack_dr0.h" |
182 | #endif /* !defined(__x86_64__) */ | | 182 | #endif /* !defined(__x86_64__) */ |
183 | | | 183 | |
184 | #include <sys/param.h> | | 184 | #include <sys/param.h> |
185 | #include <sys/systm.h> | | 185 | #include <sys/systm.h> |
186 | #include <sys/proc.h> | | 186 | #include <sys/proc.h> |
187 | #include <sys/pool.h> | | 187 | #include <sys/pool.h> |
188 | #include <sys/kernel.h> | | 188 | #include <sys/kernel.h> |
189 | #include <sys/atomic.h> | | 189 | #include <sys/atomic.h> |
190 | #include <sys/cpu.h> | | 190 | #include <sys/cpu.h> |
191 | #include <sys/intr.h> | | 191 | #include <sys/intr.h> |
192 | #include <sys/xcall.h> | | 192 | #include <sys/xcall.h> |
193 | #include <sys/kcore.h> | | 193 | #include <sys/kcore.h> |
194 | | | 194 | |
195 | #include <uvm/uvm.h> | | 195 | #include <uvm/uvm.h> |
196 | | | 196 | |
197 | #include <dev/isa/isareg.h> | | 197 | #include <dev/isa/isareg.h> |
198 | | | 198 | |
199 | #include <machine/specialreg.h> | | 199 | #include <machine/specialreg.h> |
200 | #include <machine/gdt.h> | | 200 | #include <machine/gdt.h> |
201 | #include <machine/isa_machdep.h> | | 201 | #include <machine/isa_machdep.h> |
202 | #include <machine/cpuvar.h> | | 202 | #include <machine/cpuvar.h> |
203 | #include <machine/cputypes.h> | | 203 | #include <machine/cputypes.h> |
204 | | | 204 | |
205 | #include <x86/pmap.h> | | 205 | #include <x86/pmap.h> |
206 | #include <x86/pmap_pv.h> | | 206 | #include <x86/pmap_pv.h> |
207 | | | 207 | |
208 | #include <x86/i82489reg.h> | | 208 | #include <x86/i82489reg.h> |
209 | #include <x86/i82489var.h> | | 209 | #include <x86/i82489var.h> |
210 | | | 210 | |
211 | #ifdef XEN | | 211 | #ifdef XEN |
212 | #include <xen/xen-public/xen.h> | | 212 | #include <xen/xen-public/xen.h> |
213 | #include <xen/hypervisor.h> | | 213 | #include <xen/hypervisor.h> |
214 | #endif | | 214 | #endif |
215 | | | 215 | |
216 | /* | | 216 | /* |
217 | * general info: | | 217 | * general info: |
218 | * | | 218 | * |
219 | * - for an explanation of how the i386 MMU hardware works see | | 219 | * - for an explanation of how the i386 MMU hardware works see |
220 | * the comments in <machine/pte.h>. | | 220 | * the comments in <machine/pte.h>. |
221 | * | | 221 | * |
222 | * - for an explanation of the general memory structure used by | | 222 | * - for an explanation of the general memory structure used by |
223 | * this pmap (including the recursive mapping), see the comments | | 223 | * this pmap (including the recursive mapping), see the comments |
224 | * in <machine/pmap.h>. | | 224 | * in <machine/pmap.h>. |
225 | * | | 225 | * |
226 | * this file contains the code for the "pmap module." the module's | | 226 | * this file contains the code for the "pmap module." the module's |
227 | * job is to manage the hardware's virtual to physical address mappings. | | 227 | * job is to manage the hardware's virtual to physical address mappings. |
228 | * note that there are two levels of mapping in the VM system: | | 228 | * note that there are two levels of mapping in the VM system: |
229 | * | | 229 | * |
230 | * [1] the upper layer of the VM system uses vm_map's and vm_map_entry's | | 230 | * [1] the upper layer of the VM system uses vm_map's and vm_map_entry's |
231 | * to map ranges of virtual address space to objects/files. for | | 231 | * to map ranges of virtual address space to objects/files. for |
232 | * example, the vm_map may say: "map VA 0x1000 to 0x22000 read-only | | 232 | * example, the vm_map may say: "map VA 0x1000 to 0x22000 read-only |
233 | * to the file /bin/ls starting at offset zero." note that | | 233 | * to the file /bin/ls starting at offset zero." note that |
234 | * the upper layer mapping is not concerned with how individual | | 234 | * the upper layer mapping is not concerned with how individual |
235 | * vm_pages are mapped. | | 235 | * vm_pages are mapped. |
236 | * | | 236 | * |
237 | * [2] the lower layer of the VM system (the pmap) maintains the mappings | | 237 | * [2] the lower layer of the VM system (the pmap) maintains the mappings |
238 | * from virtual addresses. it is concerned with which vm_page is | | 238 | * from virtual addresses. it is concerned with which vm_page is |
239 | * mapped where. for example, when you run /bin/ls and start | | 239 | * mapped where. for example, when you run /bin/ls and start |
240 | * at page 0x1000 the fault routine may lookup the correct page | | 240 | * at page 0x1000 the fault routine may lookup the correct page |
241 | * of the /bin/ls file and then ask the pmap layer to establish | | 241 | * of the /bin/ls file and then ask the pmap layer to establish |
242 | * a mapping for it. | | 242 | * a mapping for it. |
243 | * | | 243 | * |
244 | * note that information in the lower layer of the VM system can be | | 244 | * note that information in the lower layer of the VM system can be |
245 | * thrown away since it can easily be reconstructed from the info | | 245 | * thrown away since it can easily be reconstructed from the info |
246 | * in the upper layer. | | 246 | * in the upper layer. |
247 | * | | 247 | * |
248 | * data structures we use include: | | 248 | * data structures we use include: |
249 | * | | 249 | * |
250 | * - struct pmap: describes the address space of one thread | | 250 | * - struct pmap: describes the address space of one thread |
251 | * - struct pv_entry: describes one <PMAP,VA> mapping of a PA | | 251 | * - struct pv_entry: describes one <PMAP,VA> mapping of a PA |
252 | * - struct pv_head: there is one pv_head per managed page of | | 252 | * - struct pv_head: there is one pv_head per managed page of |
253 | * physical memory. the pv_head points to a list of pv_entry | | 253 | * physical memory. the pv_head points to a list of pv_entry |
254 | * structures which describe all the <PMAP,VA> pairs that this | | 254 | * structures which describe all the <PMAP,VA> pairs that this |
255 | * page is mapped in. this is critical for page based operations | | 255 | * page is mapped in. this is critical for page based operations |
256 | * such as pmap_page_protect() [change protection on _all_ mappings | | 256 | * such as pmap_page_protect() [change protection on _all_ mappings |
257 | * of a page] | | 257 | * of a page] |
258 | */ | | 258 | */ |
259 | | | 259 | |
260 | /* | | 260 | /* |
261 | * memory allocation | | 261 | * memory allocation |
262 | * | | 262 | * |
263 | * - there are three data structures that we must dynamically allocate: | | 263 | * - there are three data structures that we must dynamically allocate: |
264 | * | | 264 | * |
265 | * [A] new process' page directory page (PDP) | | 265 | * [A] new process' page directory page (PDP) |
266 | * - plan 1: done at pmap_create() we use | | 266 | * - plan 1: done at pmap_create() we use |
267 | * uvm_km_alloc(kernel_map, PAGE_SIZE) [fka kmem_alloc] to do this | | 267 | * uvm_km_alloc(kernel_map, PAGE_SIZE) [fka kmem_alloc] to do this |
268 | * allocation. | | 268 | * allocation. |
269 | * | | 269 | * |
270 | * if we are low in free physical memory then we sleep in | | 270 | * if we are low in free physical memory then we sleep in |
271 | * uvm_km_alloc -- in this case this is ok since we are creating | | 271 | * uvm_km_alloc -- in this case this is ok since we are creating |
272 | * a new pmap and should not be holding any locks. | | 272 | * a new pmap and should not be holding any locks. |
273 | * | | 273 | * |
274 | * if the kernel is totally out of virtual space | | 274 | * if the kernel is totally out of virtual space |
275 | * (i.e. uvm_km_alloc returns NULL), then we panic. | | 275 | * (i.e. uvm_km_alloc returns NULL), then we panic. |
276 | * | | 276 | * |
277 | * [B] new page tables pages (PTP) | | 277 | * [B] new page tables pages (PTP) |
278 | * - call uvm_pagealloc() | | 278 | * - call uvm_pagealloc() |
279 | * => success: zero page, add to pm_pdir | | 279 | * => success: zero page, add to pm_pdir |
280 | * => failure: we are out of free vm_pages, let pmap_enter() | | 280 | * => failure: we are out of free vm_pages, let pmap_enter() |
281 | * tell UVM about it. | | 281 | * tell UVM about it. |
282 | * | | 282 | * |
283 | * note: for kernel PTPs, we start with NKPTP of them. as we map | | 283 | * note: for kernel PTPs, we start with NKPTP of them. as we map |
284 | * kernel memory (at uvm_map time) we check to see if we've grown | | 284 | * kernel memory (at uvm_map time) we check to see if we've grown |
285 | * the kernel pmap. if so, we call the optional function | | 285 | * the kernel pmap. if so, we call the optional function |
286 | * pmap_growkernel() to grow the kernel PTPs in advance. | | 286 | * pmap_growkernel() to grow the kernel PTPs in advance. |
287 | * | | 287 | * |
288 | * [C] pv_entry structures | | 288 | * [C] pv_entry structures |
289 | */ | | 289 | */ |
290 | | | 290 | |
291 | /* | | 291 | /* |
292 | * locking | | 292 | * locking |
293 | * | | 293 | * |
294 | * we have the following locks that we must contend with: | | 294 | * we have the following locks that we must contend with: |
295 | * | | 295 | * |
296 | * mutexes: | | 296 | * mutexes: |
297 | * | | 297 | * |
298 | * - pmap lock (per pmap, part of uvm_object) | | 298 | * - pmap lock (per pmap, part of uvm_object) |
299 | * this lock protects the fields in the pmap structure including | | 299 | * this lock protects the fields in the pmap structure including |
300 | * the non-kernel PDEs in the PDP, and the PTEs. it also locks | | 300 | * the non-kernel PDEs in the PDP, and the PTEs. it also locks |
301 | * in the alternate PTE space (since that is determined by the | | 301 | * in the alternate PTE space (since that is determined by the |
302 | * entry in the PDP). | | 302 | * entry in the PDP). |
303 | * | | 303 | * |
304 | * - pvh_lock (per pv_head) | | 304 | * - pvh_lock (per pv_head) |
305 | * this lock protects the pv_entry list which is chained off the | | 305 | * this lock protects the pv_entry list which is chained off the |
306 | * pv_head structure for a specific managed PA. it is locked | | 306 | * pv_head structure for a specific managed PA. it is locked |
307 | * when traversing the list (e.g. adding/removing mappings, | | 307 | * when traversing the list (e.g. adding/removing mappings, |
308 | * syncing R/M bits, etc.) | | 308 | * syncing R/M bits, etc.) |
309 | * | | 309 | * |
310 | * - pmaps_lock | | 310 | * - pmaps_lock |
311 | * this lock protects the list of active pmaps (headed by "pmaps"). | | 311 | * this lock protects the list of active pmaps (headed by "pmaps"). |
312 | * we lock it when adding or removing pmaps from this list. | | 312 | * we lock it when adding or removing pmaps from this list. |
313 | */ | | 313 | */ |
314 | | | 314 | |
315 | const vaddr_t ptp_masks[] = PTP_MASK_INITIALIZER; | | 315 | const vaddr_t ptp_masks[] = PTP_MASK_INITIALIZER; |
316 | const int ptp_shifts[] = PTP_SHIFT_INITIALIZER; | | 316 | const int ptp_shifts[] = PTP_SHIFT_INITIALIZER; |
317 | const long nkptpmax[] = NKPTPMAX_INITIALIZER; | | 317 | const long nkptpmax[] = NKPTPMAX_INITIALIZER; |
318 | const long nbpd[] = NBPD_INITIALIZER; | | 318 | const long nbpd[] = NBPD_INITIALIZER; |
319 | pd_entry_t * const normal_pdes[] = PDES_INITIALIZER; | | 319 | pd_entry_t * const normal_pdes[] = PDES_INITIALIZER; |
320 | | | 320 | |
321 | long nkptp[] = NKPTP_INITIALIZER; | | 321 | long nkptp[] = NKPTP_INITIALIZER; |
322 | | | 322 | |
323 | struct pmap_head pmaps; | | 323 | struct pmap_head pmaps; |
324 | kmutex_t pmaps_lock; | | 324 | kmutex_t pmaps_lock; |
325 | | | 325 | |
326 | static vaddr_t pmap_maxkvaddr; | | 326 | static vaddr_t pmap_maxkvaddr; |
327 | | | 327 | |
328 | /* | | 328 | /* |
329 | * XXX kludge: dummy locking to make KASSERTs in uvm_page.c comfortable. | | 329 | * XXX kludge: dummy locking to make KASSERTs in uvm_page.c comfortable. |
330 | * actual locking is done by pm_lock. | | 330 | * actual locking is done by pm_lock. |
331 | */ | | 331 | */ |
332 | #if defined(DIAGNOSTIC) | | 332 | #if defined(DIAGNOSTIC) |
333 | #define PMAP_SUBOBJ_LOCK(pm, idx) \ | | 333 | #define PMAP_SUBOBJ_LOCK(pm, idx) \ |
334 | KASSERT(mutex_owned((pm)->pm_lock)); \ | | 334 | KASSERT(mutex_owned((pm)->pm_lock)); \ |
335 | if ((idx) != 0) \ | | 335 | if ((idx) != 0) \ |
336 | mutex_enter((pm)->pm_obj[(idx)].vmobjlock) | | 336 | mutex_enter((pm)->pm_obj[(idx)].vmobjlock) |
337 | #define PMAP_SUBOBJ_UNLOCK(pm, idx) \ | | 337 | #define PMAP_SUBOBJ_UNLOCK(pm, idx) \ |
338 | KASSERT(mutex_owned((pm)->pm_lock)); \ | | 338 | KASSERT(mutex_owned((pm)->pm_lock)); \ |
339 | if ((idx) != 0) \ | | 339 | if ((idx) != 0) \ |
340 | mutex_exit((pm)->pm_obj[(idx)].vmobjlock) | | 340 | mutex_exit((pm)->pm_obj[(idx)].vmobjlock) |
341 | #else /* defined(DIAGNOSTIC) */ | | 341 | #else /* defined(DIAGNOSTIC) */ |
342 | #define PMAP_SUBOBJ_LOCK(pm, idx) /* nothing */ | | 342 | #define PMAP_SUBOBJ_LOCK(pm, idx) /* nothing */ |
343 | #define PMAP_SUBOBJ_UNLOCK(pm, idx) /* nothing */ | | 343 | #define PMAP_SUBOBJ_UNLOCK(pm, idx) /* nothing */ |
344 | #endif /* defined(DIAGNOSTIC) */ | | 344 | #endif /* defined(DIAGNOSTIC) */ |
345 | | | 345 | |
346 | /* | | 346 | /* |
347 | * Misc. event counters. | | 347 | * Misc. event counters. |
348 | */ | | 348 | */ |
349 | struct evcnt pmap_iobmp_evcnt; | | 349 | struct evcnt pmap_iobmp_evcnt; |
350 | struct evcnt pmap_ldt_evcnt; | | 350 | struct evcnt pmap_ldt_evcnt; |
351 | | | 351 | |
352 | /* | | 352 | /* |
353 | * PAT | | 353 | * PAT |
354 | */ | | 354 | */ |
355 | #define PATENTRY(n, type) (type << ((n) * 8)) | | 355 | #define PATENTRY(n, type) (type << ((n) * 8)) |
356 | #define PAT_UC 0x0ULL | | 356 | #define PAT_UC 0x0ULL |
357 | #define PAT_WC 0x1ULL | | 357 | #define PAT_WC 0x1ULL |
358 | #define PAT_WT 0x4ULL | | 358 | #define PAT_WT 0x4ULL |
359 | #define PAT_WP 0x5ULL | | 359 | #define PAT_WP 0x5ULL |
360 | #define PAT_WB 0x6ULL | | 360 | #define PAT_WB 0x6ULL |
361 | #define PAT_UCMINUS 0x7ULL | | 361 | #define PAT_UCMINUS 0x7ULL |
362 | | | 362 | |
363 | static bool cpu_pat_enabled __read_mostly = false; | | 363 | static bool cpu_pat_enabled __read_mostly = false; |
364 | | | 364 | |
365 | /* | | 365 | /* |
366 | * global data structures | | 366 | * global data structures |
367 | */ | | 367 | */ |
368 | | | 368 | |
369 | static struct pmap kernel_pmap_store; /* the kernel's pmap (proc0) */ | | 369 | static struct pmap kernel_pmap_store; /* the kernel's pmap (proc0) */ |
370 | struct pmap *const kernel_pmap_ptr = &kernel_pmap_store; | | 370 | struct pmap *const kernel_pmap_ptr = &kernel_pmap_store; |
371 | | | 371 | |
372 | /* | | 372 | /* |
373 | * pmap_pg_g: if our processor supports PG_G in the PTE then we | | 373 | * pmap_pg_g: if our processor supports PG_G in the PTE then we |
374 | * set pmap_pg_g to PG_G (otherwise it is zero). | | 374 | * set pmap_pg_g to PG_G (otherwise it is zero). |
375 | */ | | 375 | */ |
376 | | | 376 | |
377 | int pmap_pg_g __read_mostly = 0; | | 377 | int pmap_pg_g __read_mostly = 0; |
378 | | | 378 | |
379 | /* | | 379 | /* |
380 | * pmap_largepages: if our processor supports PG_PS and we are | | 380 | * pmap_largepages: if our processor supports PG_PS and we are |
381 | * using it, this is set to true. | | 381 | * using it, this is set to true. |
382 | */ | | 382 | */ |
383 | | | 383 | |
384 | int pmap_largepages __read_mostly; | | 384 | int pmap_largepages __read_mostly; |
385 | | | 385 | |
386 | /* | | 386 | /* |
387 | * i386 physical memory comes in a big contig chunk with a small | | 387 | * i386 physical memory comes in a big contig chunk with a small |
388 | * hole toward the front of it... the following two paddr_t's | | 388 | * hole toward the front of it... the following two paddr_t's |
389 | * (shared with machdep.c) describe the physical address space | | 389 | * (shared with machdep.c) describe the physical address space |
390 | * of this machine. | | 390 | * of this machine. |
391 | */ | | 391 | */ |
392 | paddr_t avail_start __read_mostly; /* PA of first available physical page */ | | 392 | paddr_t avail_start __read_mostly; /* PA of first available physical page */ |
393 | paddr_t avail_end __read_mostly; /* PA of last available physical page */ | | 393 | paddr_t avail_end __read_mostly; /* PA of last available physical page */ |
394 | | | 394 | |
395 | #ifdef XEN | | 395 | #ifdef XEN |
396 | #ifdef __x86_64__ | | 396 | #ifdef __x86_64__ |
397 | /* Dummy PGD for user cr3, used between pmap_deactivate() and pmap_activate() */ | | 397 | /* Dummy PGD for user cr3, used between pmap_deactivate() and pmap_activate() */ |
398 | static paddr_t xen_dummy_user_pgd; | | 398 | static paddr_t xen_dummy_user_pgd; |
399 | #endif /* __x86_64__ */ | | 399 | #endif /* __x86_64__ */ |
400 | paddr_t pmap_pa_start; /* PA of first physical page for this domain */ | | 400 | paddr_t pmap_pa_start; /* PA of first physical page for this domain */ |
401 | paddr_t pmap_pa_end; /* PA of last physical page for this domain */ | | 401 | paddr_t pmap_pa_end; /* PA of last physical page for this domain */ |
402 | #endif /* XEN */ | | 402 | #endif /* XEN */ |
403 | | | 403 | |
404 | #define VM_PAGE_TO_PP(pg) (&(pg)->mdpage.mp_pp) | | 404 | #define VM_PAGE_TO_PP(pg) (&(pg)->mdpage.mp_pp) |
405 | | | 405 | |
406 | #define PV_HASH_SIZE 32768 | | 406 | #define PV_HASH_SIZE 32768 |
407 | #define PV_HASH_LOCK_CNT 32 | | 407 | #define PV_HASH_LOCK_CNT 32 |
408 | | | 408 | |
409 | struct pv_hash_lock { | | 409 | struct pv_hash_lock { |
410 | kmutex_t lock; | | 410 | kmutex_t lock; |
411 | } __aligned(CACHE_LINE_SIZE) pv_hash_locks[PV_HASH_LOCK_CNT] | | 411 | } __aligned(CACHE_LINE_SIZE) pv_hash_locks[PV_HASH_LOCK_CNT] |
412 | __aligned(CACHE_LINE_SIZE); | | 412 | __aligned(CACHE_LINE_SIZE); |
413 | | | 413 | |
414 | struct pv_hash_head { | | 414 | struct pv_hash_head { |
415 | SLIST_HEAD(, pv_entry) hh_list; | | 415 | SLIST_HEAD(, pv_entry) hh_list; |
416 | } pv_hash_heads[PV_HASH_SIZE]; | | 416 | } pv_hash_heads[PV_HASH_SIZE]; |
417 | | | 417 | |
418 | static u_int | | 418 | static u_int |
419 | pvhash_hash(struct vm_page *ptp, vaddr_t va) | | 419 | pvhash_hash(struct vm_page *ptp, vaddr_t va) |
420 | { | | 420 | { |
421 | | | 421 | |
422 | return (uintptr_t)ptp / sizeof(*ptp) + (va >> PAGE_SHIFT); | | 422 | return (uintptr_t)ptp / sizeof(*ptp) + (va >> PAGE_SHIFT); |
423 | } | | 423 | } |
424 | | | 424 | |
425 | static struct pv_hash_head * | | 425 | static struct pv_hash_head * |
426 | pvhash_head(u_int hash) | | 426 | pvhash_head(u_int hash) |
427 | { | | 427 | { |
428 | | | 428 | |
429 | return &pv_hash_heads[hash % PV_HASH_SIZE]; | | 429 | return &pv_hash_heads[hash % PV_HASH_SIZE]; |
430 | } | | 430 | } |
431 | | | 431 | |
432 | static kmutex_t * | | 432 | static kmutex_t * |
433 | pvhash_lock(u_int hash) | | 433 | pvhash_lock(u_int hash) |
434 | { | | 434 | { |
435 | | | 435 | |
436 | return &pv_hash_locks[hash % PV_HASH_LOCK_CNT].lock; | | 436 | return &pv_hash_locks[hash % PV_HASH_LOCK_CNT].lock; |
437 | } | | 437 | } |
438 | | | 438 | |
439 | static struct pv_entry * | | 439 | static struct pv_entry * |
440 | pvhash_remove(struct pv_hash_head *hh, struct vm_page *ptp, vaddr_t va) | | 440 | pvhash_remove(struct pv_hash_head *hh, struct vm_page *ptp, vaddr_t va) |
441 | { | | 441 | { |
442 | struct pv_entry *pve; | | 442 | struct pv_entry *pve; |
443 | struct pv_entry *prev; | | 443 | struct pv_entry *prev; |
444 | | | 444 | |
445 | prev = NULL; | | 445 | prev = NULL; |
446 | SLIST_FOREACH(pve, &hh->hh_list, pve_hash) { | | 446 | SLIST_FOREACH(pve, &hh->hh_list, pve_hash) { |
447 | if (pve->pve_pte.pte_ptp == ptp && | | 447 | if (pve->pve_pte.pte_ptp == ptp && |
448 | pve->pve_pte.pte_va == va) { | | 448 | pve->pve_pte.pte_va == va) { |
449 | if (prev != NULL) { | | 449 | if (prev != NULL) { |
450 | SLIST_REMOVE_AFTER(prev, pve_hash); | | 450 | SLIST_REMOVE_AFTER(prev, pve_hash); |
451 | } else { | | 451 | } else { |
452 | SLIST_REMOVE_HEAD(&hh->hh_list, pve_hash); | | 452 | SLIST_REMOVE_HEAD(&hh->hh_list, pve_hash); |
453 | } | | 453 | } |
454 | break; | | 454 | break; |
455 | } | | 455 | } |
456 | prev = pve; | | 456 | prev = pve; |
457 | } | | 457 | } |
458 | return pve; | | 458 | return pve; |
459 | } | | 459 | } |
460 | | | 460 | |
461 | /* | | 461 | /* |
462 | * other data structures | | 462 | * other data structures |
463 | */ | | 463 | */ |
464 | | | 464 | |
465 | static pt_entry_t protection_codes[8] __read_mostly; /* maps MI prot to i386 | | 465 | static pt_entry_t protection_codes[8] __read_mostly; /* maps MI prot to i386 |
466 | prot code */ | | 466 | prot code */ |
467 | static bool pmap_initialized __read_mostly = false; /* pmap_init done yet? */ | | 467 | static bool pmap_initialized __read_mostly = false; /* pmap_init done yet? */ |
468 | | | 468 | |
469 | /* | | 469 | /* |
470 | * the following two vaddr_t's are used during system startup | | 470 | * the following two vaddr_t's are used during system startup |
471 | * to keep track of how much of the kernel's VM space we have used. | | 471 | * to keep track of how much of the kernel's VM space we have used. |
472 | * once the system is started, the management of the remaining kernel | | 472 | * once the system is started, the management of the remaining kernel |
473 | * VM space is turned over to the kernel_map vm_map. | | 473 | * VM space is turned over to the kernel_map vm_map. |
474 | */ | | 474 | */ |
475 | | | 475 | |
476 | static vaddr_t virtual_avail __read_mostly; /* VA of first free KVA */ | | 476 | static vaddr_t virtual_avail __read_mostly; /* VA of first free KVA */ |
477 | static vaddr_t virtual_end __read_mostly; /* VA of last free KVA */ | | 477 | static vaddr_t virtual_end __read_mostly; /* VA of last free KVA */ |
478 | | | 478 | |
479 | /* | | 479 | /* |
480 | * pool that pmap structures are allocated from | | 480 | * pool that pmap structures are allocated from |
481 | */ | | 481 | */ |
482 | | | 482 | |
483 | static struct pool_cache pmap_cache; | | 483 | static struct pool_cache pmap_cache; |
484 | | | 484 | |
485 | /* | | 485 | /* |
486 | * pv_entry cache | | 486 | * pv_entry cache |
487 | */ | | 487 | */ |
488 | | | 488 | |
489 | static struct pool_cache pmap_pv_cache; | | 489 | static struct pool_cache pmap_pv_cache; |
490 | | | 490 | |
491 | #ifdef __HAVE_DIRECT_MAP | | 491 | #ifdef __HAVE_DIRECT_MAP |
492 | | | 492 | |
493 | extern phys_ram_seg_t mem_clusters[]; | | 493 | extern phys_ram_seg_t mem_clusters[]; |
494 | extern int mem_cluster_cnt; | | 494 | extern int mem_cluster_cnt; |
495 | | | 495 | |
496 | #else | | 496 | #else |
497 | | | 497 | |
498 | /* | | 498 | /* |
499 | * MULTIPROCESSOR: special VA's/ PTE's are actually allocated inside a | | 499 | * MULTIPROCESSOR: special VA's/ PTE's are actually allocated inside a |
500 | * maxcpus*NPTECL array of PTE's, to avoid cache line thrashing | | 500 | * maxcpus*NPTECL array of PTE's, to avoid cache line thrashing |
501 | * due to false sharing. | | 501 | * due to false sharing. |
502 | */ | | 502 | */ |
503 | | | 503 | |
504 | #ifdef MULTIPROCESSOR | | 504 | #ifdef MULTIPROCESSOR |
505 | #define PTESLEW(pte, id) ((pte)+(id)*NPTECL) | | 505 | #define PTESLEW(pte, id) ((pte)+(id)*NPTECL) |
506 | #define VASLEW(va,id) ((va)+(id)*NPTECL*PAGE_SIZE) | | 506 | #define VASLEW(va,id) ((va)+(id)*NPTECL*PAGE_SIZE) |
507 | #else | | 507 | #else |
508 | #define PTESLEW(pte, id) (pte) | | 508 | #define PTESLEW(pte, id) (pte) |
509 | #define VASLEW(va,id) (va) | | 509 | #define VASLEW(va,id) (va) |
510 | #endif | | 510 | #endif |
511 | | | 511 | |
512 | /* | | 512 | /* |
513 | * special VAs and the PTEs that map them | | 513 | * special VAs and the PTEs that map them |
514 | */ | | 514 | */ |
515 | static pt_entry_t *csrc_pte, *cdst_pte, *zero_pte, *ptp_pte, *early_zero_pte; | | 515 | static pt_entry_t *csrc_pte, *cdst_pte, *zero_pte, *ptp_pte, *early_zero_pte; |
516 | static char *csrcp, *cdstp, *zerop, *ptpp, *early_zerop; | | 516 | static char *csrcp, *cdstp, *zerop, *ptpp, *early_zerop; |
517 | | | 517 | |
518 | #endif | | 518 | #endif |
519 | | | 519 | |
520 | int pmap_enter_default(pmap_t, vaddr_t, paddr_t, vm_prot_t, u_int); | | 520 | int pmap_enter_default(pmap_t, vaddr_t, paddr_t, vm_prot_t, u_int); |
521 | | | 521 | |
522 | /* PDP pool_cache(9) and its callbacks */ | | 522 | /* PDP pool_cache(9) and its callbacks */ |
523 | struct pool_cache pmap_pdp_cache; | | 523 | struct pool_cache pmap_pdp_cache; |
524 | static int pmap_pdp_ctor(void *, void *, int); | | 524 | static int pmap_pdp_ctor(void *, void *, int); |
525 | static void pmap_pdp_dtor(void *, void *); | | 525 | static void pmap_pdp_dtor(void *, void *); |
526 | #ifdef PAE | | 526 | #ifdef PAE |
527 | /* need to allocate items of 4 pages */ | | 527 | /* need to allocate items of 4 pages */ |
528 | static void *pmap_pdp_alloc(struct pool *, int); | | 528 | static void *pmap_pdp_alloc(struct pool *, int); |
529 | static void pmap_pdp_free(struct pool *, void *); | | 529 | static void pmap_pdp_free(struct pool *, void *); |
530 | static struct pool_allocator pmap_pdp_allocator = { | | 530 | static struct pool_allocator pmap_pdp_allocator = { |
531 | .pa_alloc = pmap_pdp_alloc, | | 531 | .pa_alloc = pmap_pdp_alloc, |
532 | .pa_free = pmap_pdp_free, | | 532 | .pa_free = pmap_pdp_free, |
533 | .pa_pagesz = PAGE_SIZE * PDP_SIZE, | | 533 | .pa_pagesz = PAGE_SIZE * PDP_SIZE, |
534 | }; | | 534 | }; |
535 | #endif /* PAE */ | | 535 | #endif /* PAE */ |
536 | | | 536 | |
537 | extern vaddr_t idt_vaddr; /* we allocate IDT early */ | | 537 | extern vaddr_t idt_vaddr; /* we allocate IDT early */ |
538 | extern paddr_t idt_paddr; | | 538 | extern paddr_t idt_paddr; |
539 | | | 539 | |
540 | #ifdef _LP64 | | 540 | #ifdef _LP64 |
541 | extern vaddr_t lo32_vaddr; | | 541 | extern vaddr_t lo32_vaddr; |
542 | extern vaddr_t lo32_paddr; | | 542 | extern vaddr_t lo32_paddr; |
543 | #endif | | 543 | #endif |
544 | | | 544 | |
545 | extern int end; | | 545 | extern int end; |
546 | | | 546 | |
547 | #ifdef i386 | | 547 | #ifdef i386 |
548 | /* stuff to fix the pentium f00f bug */ | | 548 | /* stuff to fix the pentium f00f bug */ |
549 | extern vaddr_t pentium_idt_vaddr; | | 549 | extern vaddr_t pentium_idt_vaddr; |
550 | #endif | | 550 | #endif |
551 | | | 551 | |
552 | | | 552 | |
553 | /* | | 553 | /* |
554 | * local prototypes | | 554 | * local prototypes |
555 | */ | | 555 | */ |
556 | | | 556 | |
557 | static struct vm_page *pmap_get_ptp(struct pmap *, vaddr_t, | | 557 | static struct vm_page *pmap_get_ptp(struct pmap *, vaddr_t, |
558 | pd_entry_t * const *); | | 558 | pd_entry_t * const *); |
559 | static struct vm_page *pmap_find_ptp(struct pmap *, vaddr_t, paddr_t, int); | | 559 | static struct vm_page *pmap_find_ptp(struct pmap *, vaddr_t, paddr_t, int); |
560 | static void pmap_freepage(struct pmap *, struct vm_page *, int); | | 560 | static void pmap_freepage(struct pmap *, struct vm_page *, int); |
561 | static void pmap_free_ptp(struct pmap *, struct vm_page *, | | 561 | static void pmap_free_ptp(struct pmap *, struct vm_page *, |
562 | vaddr_t, pt_entry_t *, | | 562 | vaddr_t, pt_entry_t *, |
563 | pd_entry_t * const *); | | 563 | pd_entry_t * const *); |
564 | static bool pmap_is_active(struct pmap *, struct cpu_info *, bool); | | 564 | static bool pmap_is_active(struct pmap *, struct cpu_info *, bool); |
565 | static bool pmap_remove_pte(struct pmap *, struct vm_page *, | | 565 | static bool pmap_remove_pte(struct pmap *, struct vm_page *, |
566 | pt_entry_t *, vaddr_t, | | 566 | pt_entry_t *, vaddr_t, |
567 | struct pv_entry **); | | 567 | struct pv_entry **); |
568 | static void pmap_remove_ptes(struct pmap *, struct vm_page *, | | 568 | static void pmap_remove_ptes(struct pmap *, struct vm_page *, |
569 | vaddr_t, vaddr_t, vaddr_t, | | 569 | vaddr_t, vaddr_t, vaddr_t, |
570 | struct pv_entry **); | | 570 | struct pv_entry **); |
571 | | | 571 | |
572 | static bool pmap_get_physpage(vaddr_t, int, paddr_t *); | | 572 | static bool pmap_get_physpage(vaddr_t, int, paddr_t *); |
573 | static void pmap_alloc_level(pd_entry_t * const *, vaddr_t, int, | | 573 | static void pmap_alloc_level(pd_entry_t * const *, vaddr_t, int, |
574 | long *); | | 574 | long *); |
575 | | | 575 | |
576 | static bool pmap_reactivate(struct pmap *); | | 576 | static bool pmap_reactivate(struct pmap *); |
577 | | | 577 | |
578 | /* | | 578 | /* |
579 | * p m a p h e l p e r f u n c t i o n s | | 579 | * p m a p h e l p e r f u n c t i o n s |
580 | */ | | 580 | */ |
581 | | | 581 | |
582 | static inline void | | 582 | static inline void |
583 | pmap_stats_update(struct pmap *pmap, int resid_diff, int wired_diff) | | 583 | pmap_stats_update(struct pmap *pmap, int resid_diff, int wired_diff) |
584 | { | | 584 | { |
585 | | | 585 | |
586 | if (pmap == pmap_kernel()) { | | 586 | if (pmap == pmap_kernel()) { |
587 | atomic_add_long(&pmap->pm_stats.resident_count, resid_diff); | | 587 | atomic_add_long(&pmap->pm_stats.resident_count, resid_diff); |
588 | atomic_add_long(&pmap->pm_stats.wired_count, wired_diff); | | 588 | atomic_add_long(&pmap->pm_stats.wired_count, wired_diff); |
589 | } else { | | 589 | } else { |
590 | KASSERT(mutex_owned(pmap->pm_lock)); | | 590 | KASSERT(mutex_owned(pmap->pm_lock)); |
591 | pmap->pm_stats.resident_count += resid_diff; | | 591 | pmap->pm_stats.resident_count += resid_diff; |
592 | pmap->pm_stats.wired_count += wired_diff; | | 592 | pmap->pm_stats.wired_count += wired_diff; |
593 | } | | 593 | } |
594 | } | | 594 | } |
595 | | | 595 | |
596 | static inline void | | 596 | static inline void |
597 | pmap_stats_update_bypte(struct pmap *pmap, pt_entry_t npte, pt_entry_t opte) | | 597 | pmap_stats_update_bypte(struct pmap *pmap, pt_entry_t npte, pt_entry_t opte) |
598 | { | | 598 | { |
599 | int resid_diff = ((npte & PG_V) ? 1 : 0) - ((opte & PG_V) ? 1 : 0); | | 599 | int resid_diff = ((npte & PG_V) ? 1 : 0) - ((opte & PG_V) ? 1 : 0); |
600 | int wired_diff = ((npte & PG_W) ? 1 : 0) - ((opte & PG_W) ? 1 : 0); | | 600 | int wired_diff = ((npte & PG_W) ? 1 : 0) - ((opte & PG_W) ? 1 : 0); |
601 | | | 601 | |
602 | KASSERT((npte & (PG_V | PG_W)) != PG_W); | | 602 | KASSERT((npte & (PG_V | PG_W)) != PG_W); |
603 | KASSERT((opte & (PG_V | PG_W)) != PG_W); | | 603 | KASSERT((opte & (PG_V | PG_W)) != PG_W); |
604 | | | 604 | |
605 | pmap_stats_update(pmap, resid_diff, wired_diff); | | 605 | pmap_stats_update(pmap, resid_diff, wired_diff); |
606 | } | | 606 | } |
607 | | | 607 | |
608 | /* | | 608 | /* |
609 | * ptp_to_pmap: lookup pmap by ptp | | 609 | * ptp_to_pmap: lookup pmap by ptp |
610 | */ | | 610 | */ |
611 | | | 611 | |
612 | static struct pmap * | | 612 | static struct pmap * |
613 | ptp_to_pmap(struct vm_page *ptp) | | 613 | ptp_to_pmap(struct vm_page *ptp) |
614 | { | | 614 | { |
615 | struct pmap *pmap; | | 615 | struct pmap *pmap; |
616 | | | 616 | |
617 | if (ptp == NULL) { | | 617 | if (ptp == NULL) { |
618 | return pmap_kernel(); | | 618 | return pmap_kernel(); |
619 | } | | 619 | } |
620 | pmap = (struct pmap *)ptp->uobject; | | 620 | pmap = (struct pmap *)ptp->uobject; |
621 | KASSERT(pmap != NULL); | | 621 | KASSERT(pmap != NULL); |
622 | KASSERT(&pmap->pm_obj[0] == ptp->uobject); | | 622 | KASSERT(&pmap->pm_obj[0] == ptp->uobject); |
623 | return pmap; | | 623 | return pmap; |
624 | } | | 624 | } |
625 | | | 625 | |
626 | static inline struct pv_pte * | | 626 | static inline struct pv_pte * |
627 | pve_to_pvpte(struct pv_entry *pve) | | 627 | pve_to_pvpte(struct pv_entry *pve) |
628 | { | | 628 | { |
629 | | | 629 | |
630 | KASSERT((void *)&pve->pve_pte == (void *)pve); | | 630 | KASSERT((void *)&pve->pve_pte == (void *)pve); |
631 | return &pve->pve_pte; | | 631 | return &pve->pve_pte; |
632 | } | | 632 | } |
633 | | | 633 | |
634 | static inline struct pv_entry * | | 634 | static inline struct pv_entry * |
635 | pvpte_to_pve(struct pv_pte *pvpte) | | 635 | pvpte_to_pve(struct pv_pte *pvpte) |
636 | { | | 636 | { |
637 | struct pv_entry *pve = (void *)pvpte; | | 637 | struct pv_entry *pve = (void *)pvpte; |
638 | | | 638 | |
639 | KASSERT(pve_to_pvpte(pve) == pvpte); | | 639 | KASSERT(pve_to_pvpte(pve) == pvpte); |
640 | return pve; | | 640 | return pve; |
641 | } | | 641 | } |
642 | | | 642 | |
643 | /* | | 643 | /* |
644 | * pv_pte_first, pv_pte_next: PV list iterator. | | 644 | * pv_pte_first, pv_pte_next: PV list iterator. |
645 | */ | | 645 | */ |
646 | | | 646 | |
647 | static struct pv_pte * | | 647 | static struct pv_pte * |
648 | pv_pte_first(struct pmap_page *pp) | | 648 | pv_pte_first(struct pmap_page *pp) |
649 | { | | 649 | { |
650 | | | 650 | |
651 | if ((pp->pp_flags & PP_EMBEDDED) != 0) { | | 651 | if ((pp->pp_flags & PP_EMBEDDED) != 0) { |
652 | return &pp->pp_pte; | | 652 | return &pp->pp_pte; |
653 | } | | 653 | } |
654 | return pve_to_pvpte(LIST_FIRST(&pp->pp_head.pvh_list)); | | 654 | return pve_to_pvpte(LIST_FIRST(&pp->pp_head.pvh_list)); |
655 | } | | 655 | } |
656 | | | 656 | |
657 | static struct pv_pte * | | 657 | static struct pv_pte * |
658 | pv_pte_next(struct pmap_page *pp, struct pv_pte *pvpte) | | 658 | pv_pte_next(struct pmap_page *pp, struct pv_pte *pvpte) |
659 | { | | 659 | { |
660 | | | 660 | |
661 | KASSERT(pvpte != NULL); | | 661 | KASSERT(pvpte != NULL); |
662 | if (pvpte == &pp->pp_pte) { | | 662 | if (pvpte == &pp->pp_pte) { |
663 | KASSERT((pp->pp_flags & PP_EMBEDDED) != 0); | | 663 | KASSERT((pp->pp_flags & PP_EMBEDDED) != 0); |
664 | return NULL; | | 664 | return NULL; |
665 | } | | 665 | } |
666 | KASSERT((pp->pp_flags & PP_EMBEDDED) == 0); | | 666 | KASSERT((pp->pp_flags & PP_EMBEDDED) == 0); |
667 | return pve_to_pvpte(LIST_NEXT(pvpte_to_pve(pvpte), pve_list)); | | 667 | return pve_to_pvpte(LIST_NEXT(pvpte_to_pve(pvpte), pve_list)); |
668 | } | | 668 | } |
669 | | | 669 | |
670 | /* | | 670 | /* |
671 | * pmap_is_curpmap: is this pmap the one currently loaded [in %cr3]? | | 671 | * pmap_is_curpmap: is this pmap the one currently loaded [in %cr3]? |
672 | * of course the kernel is always loaded | | 672 | * of course the kernel is always loaded |
673 | */ | | 673 | */ |
674 | | | 674 | |
675 | bool | | 675 | bool |
676 | pmap_is_curpmap(struct pmap *pmap) | | 676 | pmap_is_curpmap(struct pmap *pmap) |
677 | { | | 677 | { |
678 | #if defined(XEN) && defined(__x86_64__) | | 678 | #if defined(XEN) && defined(__x86_64__) |
679 | /* | | 679 | /* |
680 | * Only kernel pmap is physically loaded. | | 680 | * Only kernel pmap is physically loaded. |
681 | * User PGD may be active, but TLB will be flushed | | 681 | * User PGD may be active, but TLB will be flushed |
682 | * with HYPERVISOR_iret anyway, so let's say no | | 682 | * with HYPERVISOR_iret anyway, so let's say no |
683 | */ | | 683 | */ |
684 | return(pmap == pmap_kernel()); | | 684 | return(pmap == pmap_kernel()); |
685 | #else /* XEN && __x86_64__*/ | | 685 | #else /* XEN && __x86_64__*/ |
686 | return((pmap == pmap_kernel()) || | | 686 | return((pmap == pmap_kernel()) || |
687 | (pmap == curcpu()->ci_pmap)); | | 687 | (pmap == curcpu()->ci_pmap)); |
688 | #endif | | 688 | #endif |
689 | } | | 689 | } |
690 | | | 690 | |
691 | /* | | 691 | /* |
692 | * pmap_is_active: is this pmap loaded into the specified processor's %cr3? | | 692 | * pmap_is_active: is this pmap loaded into the specified processor's %cr3? |
693 | */ | | 693 | */ |
694 | | | 694 | |
695 | inline static bool | | 695 | inline static bool |
696 | pmap_is_active(struct pmap *pmap, struct cpu_info *ci, bool kernel) | | 696 | pmap_is_active(struct pmap *pmap, struct cpu_info *ci, bool kernel) |
697 | { | | 697 | { |
698 | | | 698 | |
699 | return (pmap == pmap_kernel() || | | 699 | return (pmap == pmap_kernel() || |
700 | (pmap->pm_cpus & ci->ci_cpumask) != 0 || | | 700 | (pmap->pm_cpus & ci->ci_cpumask) != 0 || |
701 | (kernel && (pmap->pm_kernel_cpus & ci->ci_cpumask) != 0)); | | 701 | (kernel && (pmap->pm_kernel_cpus & ci->ci_cpumask) != 0)); |
702 | } | | 702 | } |
703 | | | 703 | |
704 | /* | | 704 | /* |
705 | * Add a reference to the specified pmap. | | 705 | * Add a reference to the specified pmap. |
706 | */ | | 706 | */ |
707 | | | 707 | |
708 | void | | 708 | void |
709 | pmap_reference(struct pmap *pmap) | | 709 | pmap_reference(struct pmap *pmap) |
710 | { | | 710 | { |
711 | | | 711 | |
712 | atomic_inc_uint(&pmap->pm_obj[0].uo_refs); | | 712 | atomic_inc_uint(&pmap->pm_obj[0].uo_refs); |
713 | } | | 713 | } |
714 | | | 714 | |
715 | /* | | 715 | /* |
716 | * pmap_map_ptes: map a pmap's PTEs into KVM and lock them in | | 716 | * pmap_map_ptes: map a pmap's PTEs into KVM and lock them in |
717 | * | | 717 | * |
718 | * there are several pmaps involved. some or all of them might be same. | | 718 | * there are several pmaps involved. some or all of them might be same. |
719 | * | | 719 | * |
720 | * - the pmap given by the first argument | | 720 | * - the pmap given by the first argument |
721 | * our caller wants to access this pmap's PTEs. | | 721 | * our caller wants to access this pmap's PTEs. |
722 | * | | 722 | * |
723 | * - pmap_kernel() | | 723 | * - pmap_kernel() |
724 | * the kernel pmap. note that it only contains the kernel part | | 724 | * the kernel pmap. note that it only contains the kernel part |
725 | * of the address space which is shared by any pmap. ie. any | | 725 | * of the address space which is shared by any pmap. ie. any |
726 | * pmap can be used instead of pmap_kernel() for our purpose. | | 726 | * pmap can be used instead of pmap_kernel() for our purpose. |
727 | * | | 727 | * |
728 | * - ci->ci_pmap | | 728 | * - ci->ci_pmap |
729 | * pmap currently loaded on the cpu. | | 729 | * pmap currently loaded on the cpu. |
730 | * | | 730 | * |
731 | * - vm_map_pmap(&curproc->p_vmspace->vm_map) | | 731 | * - vm_map_pmap(&curproc->p_vmspace->vm_map) |
732 | * current process' pmap. | | 732 | * current process' pmap. |
733 | * | | 733 | * |
734 | * => we lock enough pmaps to keep things locked in | | 734 | * => we lock enough pmaps to keep things locked in |
735 | * => must be undone with pmap_unmap_ptes before returning | | 735 | * => must be undone with pmap_unmap_ptes before returning |
736 | */ | | 736 | */ |
737 | | | 737 | |
738 | void | | 738 | void |
739 | pmap_map_ptes(struct pmap *pmap, struct pmap **pmap2, | | 739 | pmap_map_ptes(struct pmap *pmap, struct pmap **pmap2, |
740 | pd_entry_t **ptepp, pd_entry_t * const **pdeppp) | | 740 | pd_entry_t **ptepp, pd_entry_t * const **pdeppp) |
741 | { | | 741 | { |
742 | struct pmap *curpmap; | | 742 | struct pmap *curpmap; |
743 | struct cpu_info *ci; | | 743 | struct cpu_info *ci; |
744 | uint32_t cpumask; | | 744 | uint32_t cpumask; |
745 | lwp_t *l; | | 745 | lwp_t *l; |
746 | | | 746 | |
747 | /* The kernel's pmap is always accessible. */ | | 747 | /* The kernel's pmap is always accessible. */ |
748 | if (pmap == pmap_kernel()) { | | 748 | if (pmap == pmap_kernel()) { |
749 | *pmap2 = NULL; | | 749 | *pmap2 = NULL; |
750 | *ptepp = PTE_BASE; | | 750 | *ptepp = PTE_BASE; |
751 | *pdeppp = normal_pdes; | | 751 | *pdeppp = normal_pdes; |
752 | return; | | 752 | return; |
753 | } | | 753 | } |
754 | KASSERT(kpreempt_disabled()); | | 754 | KASSERT(kpreempt_disabled()); |
755 | | | 755 | |
756 | l = curlwp; | | 756 | l = curlwp; |
757 | retry: | | 757 | retry: |
758 | mutex_enter(pmap->pm_lock); | | 758 | mutex_enter(pmap->pm_lock); |
759 | ci = curcpu(); | | 759 | ci = curcpu(); |
760 | curpmap = ci->ci_pmap; | | 760 | curpmap = ci->ci_pmap; |
761 | if (vm_map_pmap(&l->l_proc->p_vmspace->vm_map) == pmap) { | | 761 | if (vm_map_pmap(&l->l_proc->p_vmspace->vm_map) == pmap) { |
762 | /* Our own pmap so just load it: easy. */ | | 762 | /* Our own pmap so just load it: easy. */ |
763 | if (__predict_false(ci->ci_want_pmapload)) { | | 763 | if (__predict_false(ci->ci_want_pmapload)) { |
764 | mutex_exit(pmap->pm_lock); | | 764 | mutex_exit(pmap->pm_lock); |
765 | pmap_load(); | | 765 | pmap_load(); |
766 | goto retry; | | 766 | goto retry; |
767 | } | | 767 | } |
768 | KASSERT(pmap == curpmap); | | 768 | KASSERT(pmap == curpmap); |
769 | } else if (pmap == curpmap) { | | 769 | } else if (pmap == curpmap) { |
770 | /* | | 770 | /* |
771 | * Already on the CPU: make it valid. This is very | | 771 | * Already on the CPU: make it valid. This is very |
772 | * often the case during exit(), when we have switched | | 772 | * often the case during exit(), when we have switched |
773 | * to the kernel pmap in order to destroy a user pmap. | | 773 | * to the kernel pmap in order to destroy a user pmap. |
774 | */ | | 774 | */ |
775 | if (!pmap_reactivate(pmap)) { | | 775 | if (!pmap_reactivate(pmap)) { |
776 | u_int gen = uvm_emap_gen_return(); | | 776 | u_int gen = uvm_emap_gen_return(); |
777 | tlbflush(); | | 777 | tlbflush(); |
778 | uvm_emap_update(gen); | | 778 | uvm_emap_update(gen); |
779 | } | | 779 | } |
780 | } else { | | 780 | } else { |
781 | /* | | 781 | /* |
782 | * Toss current pmap from CPU, but keep a reference to it. | | 782 | * Toss current pmap from CPU, but keep a reference to it. |
783 | * The reference will be dropped by pmap_unmap_ptes(). | | 783 | * The reference will be dropped by pmap_unmap_ptes(). |
784 | * Can happen if we block during exit(). | | 784 | * Can happen if we block during exit(). |
785 | */ | | 785 | */ |
786 | cpumask = ci->ci_cpumask; | | 786 | cpumask = ci->ci_cpumask; |
787 | atomic_and_32(&curpmap->pm_cpus, ~cpumask); | | 787 | atomic_and_32(&curpmap->pm_cpus, ~cpumask); |
788 | atomic_and_32(&curpmap->pm_kernel_cpus, ~cpumask); | | 788 | atomic_and_32(&curpmap->pm_kernel_cpus, ~cpumask); |
789 | ci->ci_pmap = pmap; | | 789 | ci->ci_pmap = pmap; |
790 | ci->ci_tlbstate = TLBSTATE_VALID; | | 790 | ci->ci_tlbstate = TLBSTATE_VALID; |
791 | atomic_or_32(&pmap->pm_cpus, cpumask); | | 791 | atomic_or_32(&pmap->pm_cpus, cpumask); |
792 | atomic_or_32(&pmap->pm_kernel_cpus, cpumask); | | 792 | atomic_or_32(&pmap->pm_kernel_cpus, cpumask); |
793 | cpu_load_pmap(pmap); | | 793 | cpu_load_pmap(pmap); |
794 | } | | 794 | } |
795 | pmap->pm_ncsw = l->l_ncsw; | | 795 | pmap->pm_ncsw = l->l_ncsw; |
796 | *pmap2 = curpmap; | | 796 | *pmap2 = curpmap; |
797 | *ptepp = PTE_BASE; | | 797 | *ptepp = PTE_BASE; |
798 | #if defined(XEN) && defined(__x86_64__) | | 798 | #if defined(XEN) && defined(__x86_64__) |
799 | KASSERT(ci->ci_normal_pdes[PTP_LEVELS - 2] == L4_BASE); | | 799 | KASSERT(ci->ci_normal_pdes[PTP_LEVELS - 2] == L4_BASE); |
800 | ci->ci_normal_pdes[PTP_LEVELS - 2] = pmap->pm_pdir; | | 800 | ci->ci_normal_pdes[PTP_LEVELS - 2] = pmap->pm_pdir; |
801 | *pdeppp = ci->ci_normal_pdes; | | 801 | *pdeppp = ci->ci_normal_pdes; |
802 | #else /* XEN && __x86_64__ */ | | 802 | #else /* XEN && __x86_64__ */ |
803 | *pdeppp = normal_pdes; | | 803 | *pdeppp = normal_pdes; |
804 | #endif /* XEN && __x86_64__ */ | | 804 | #endif /* XEN && __x86_64__ */ |
805 | } | | 805 | } |
806 | | | 806 | |
807 | /* | | 807 | /* |
808 | * pmap_unmap_ptes: unlock the PTE mapping of "pmap" | | 808 | * pmap_unmap_ptes: unlock the PTE mapping of "pmap" |
809 | */ | | 809 | */ |
810 | | | 810 | |
811 | void | | 811 | void |
812 | pmap_unmap_ptes(struct pmap *pmap, struct pmap *pmap2) | | 812 | pmap_unmap_ptes(struct pmap *pmap, struct pmap *pmap2) |
813 | { | | 813 | { |
814 | struct cpu_info *ci; | | 814 | struct cpu_info *ci; |
815 | struct pmap *mypmap; | | 815 | struct pmap *mypmap; |
816 | | | 816 | |
817 | KASSERT(kpreempt_disabled()); | | 817 | KASSERT(kpreempt_disabled()); |
818 | | | 818 | |
819 | /* The kernel's pmap is always accessible. */ | | 819 | /* The kernel's pmap is always accessible. */ |
820 | if (pmap == pmap_kernel()) { | | 820 | if (pmap == pmap_kernel()) { |
821 | return; | | 821 | return; |
822 | } | | 822 | } |
823 | | | 823 | |
824 | ci = curcpu(); | | 824 | ci = curcpu(); |
825 | #if defined(XEN) && defined(__x86_64__) | | 825 | #if defined(XEN) && defined(__x86_64__) |
826 | /* Reset per-cpu normal_pdes */ | | 826 | /* Reset per-cpu normal_pdes */ |
827 | KASSERT(ci->ci_normal_pdes[PTP_LEVELS - 2] != L4_BASE); | | 827 | KASSERT(ci->ci_normal_pdes[PTP_LEVELS - 2] != L4_BASE); |
828 | ci->ci_normal_pdes[PTP_LEVELS - 2] = L4_BASE; | | 828 | ci->ci_normal_pdes[PTP_LEVELS - 2] = L4_BASE; |
829 | #endif /* XEN && __x86_64__ */ | | 829 | #endif /* XEN && __x86_64__ */ |
830 | /* | | 830 | /* |
831 | * We cannot tolerate context switches while mapped in. | | 831 | * We cannot tolerate context switches while mapped in. |
832 | * If it is our own pmap all we have to do is unlock. | | 832 | * If it is our own pmap all we have to do is unlock. |
833 | */ | | 833 | */ |
834 | KASSERT(pmap->pm_ncsw == curlwp->l_ncsw); | | 834 | KASSERT(pmap->pm_ncsw == curlwp->l_ncsw); |
835 | mypmap = vm_map_pmap(&curproc->p_vmspace->vm_map); | | 835 | mypmap = vm_map_pmap(&curproc->p_vmspace->vm_map); |
836 | if (pmap == mypmap) { | | 836 | if (pmap == mypmap) { |
837 | mutex_exit(pmap->pm_lock); | | 837 | mutex_exit(pmap->pm_lock); |
838 | return; | | 838 | return; |
839 | } | | 839 | } |
840 | | | 840 | |
841 | /* | | 841 | /* |
842 | * Mark whatever's on the CPU now as lazy and unlock. | | 842 | * Mark whatever's on the CPU now as lazy and unlock. |
843 | * If the pmap was already installed, we are done. | | 843 | * If the pmap was already installed, we are done. |
844 | */ | | 844 | */ |
845 | ci->ci_tlbstate = TLBSTATE_LAZY; | | 845 | ci->ci_tlbstate = TLBSTATE_LAZY; |
846 | ci->ci_want_pmapload = (mypmap != pmap_kernel()); | | 846 | ci->ci_want_pmapload = (mypmap != pmap_kernel()); |
847 | mutex_exit(pmap->pm_lock); | | 847 | mutex_exit(pmap->pm_lock); |
848 | if (pmap == pmap2) { | | 848 | if (pmap == pmap2) { |
849 | return; | | 849 | return; |
850 | } | | 850 | } |
851 | | | 851 | |
852 | /* | | 852 | /* |
853 | * We installed another pmap on the CPU. Grab a reference to | | 853 | * We installed another pmap on the CPU. Grab a reference to |
854 | * it and leave in place. Toss the evicted pmap (can block). | | 854 | * it and leave in place. Toss the evicted pmap (can block). |
855 | */ | | 855 | */ |
856 | pmap_reference(pmap); | | 856 | pmap_reference(pmap); |
857 | pmap_destroy(pmap2); | | 857 | pmap_destroy(pmap2); |
858 | } | | 858 | } |
859 | | | 859 | |
860 | | | 860 | |
861 | inline static void | | 861 | inline static void |
862 | pmap_exec_account(struct pmap *pm, vaddr_t va, pt_entry_t opte, pt_entry_t npte) | | 862 | pmap_exec_account(struct pmap *pm, vaddr_t va, pt_entry_t opte, pt_entry_t npte) |
863 | { | | 863 | { |
864 | | | 864 | |
865 | #if !defined(__x86_64__) | | 865 | #if !defined(__x86_64__) |
866 | if (curproc == NULL || curproc->p_vmspace == NULL || | | 866 | if (curproc == NULL || curproc->p_vmspace == NULL || |
867 | pm != vm_map_pmap(&curproc->p_vmspace->vm_map)) | | 867 | pm != vm_map_pmap(&curproc->p_vmspace->vm_map)) |
868 | return; | | 868 | return; |
869 | | | 869 | |
870 | if ((opte ^ npte) & PG_X) | | 870 | if ((opte ^ npte) & PG_X) |
871 | pmap_update_pg(va); | | 871 | pmap_update_pg(va); |
872 | | | 872 | |
873 | /* | | 873 | /* |
874 | * Executability was removed on the last executable change. | | 874 | * Executability was removed on the last executable change. |
875 | * Reset the code segment to something conservative and | | 875 | * Reset the code segment to something conservative and |
876 | * let the trap handler deal with setting the right limit. | | 876 | * let the trap handler deal with setting the right limit. |
877 | * We can't do that because of locking constraints on the vm map. | | 877 | * We can't do that because of locking constraints on the vm map. |
878 | */ | | 878 | */ |
879 | | | 879 | |
880 | if ((opte & PG_X) && (npte & PG_X) == 0 && va == pm->pm_hiexec) { | | 880 | if ((opte & PG_X) && (npte & PG_X) == 0 && va == pm->pm_hiexec) { |
881 | struct trapframe *tf = curlwp->l_md.md_regs; | | 881 | struct trapframe *tf = curlwp->l_md.md_regs; |
882 | | | 882 | |
883 | tf->tf_cs = GSEL(GUCODE_SEL, SEL_UPL); | | 883 | tf->tf_cs = GSEL(GUCODE_SEL, SEL_UPL); |
884 | pm->pm_hiexec = I386_MAX_EXE_ADDR; | | 884 | pm->pm_hiexec = I386_MAX_EXE_ADDR; |
885 | } | | 885 | } |
886 | #endif /* !defined(__x86_64__) */ | | 886 | #endif /* !defined(__x86_64__) */ |
887 | } | | 887 | } |
888 | | | 888 | |
889 | #if !defined(__x86_64__) | | 889 | #if !defined(__x86_64__) |
890 | /* | | 890 | /* |
891 | * Fixup the code segment to cover all potential executable mappings. | | 891 | * Fixup the code segment to cover all potential executable mappings. |
892 | * returns 0 if no changes to the code segment were made. | | 892 | * returns 0 if no changes to the code segment were made. |
893 | */ | | 893 | */ |
894 | | | 894 | |
895 | int | | 895 | int |
896 | pmap_exec_fixup(struct vm_map *map, struct trapframe *tf, struct pcb *pcb) | | 896 | pmap_exec_fixup(struct vm_map *map, struct trapframe *tf, struct pcb *pcb) |
897 | { | | 897 | { |
898 | struct vm_map_entry *ent; | | 898 | struct vm_map_entry *ent; |
899 | struct pmap *pm = vm_map_pmap(map); | | 899 | struct pmap *pm = vm_map_pmap(map); |
900 | vaddr_t va = 0; | | 900 | vaddr_t va = 0; |
901 | | | 901 | |
902 | vm_map_lock_read(map); | | 902 | vm_map_lock_read(map); |
903 | for (ent = (&map->header)->next; ent != &map->header; ent = ent->next) { | | 903 | for (ent = (&map->header)->next; ent != &map->header; ent = ent->next) { |
904 | | | 904 | |
905 | /* | | 905 | /* |
906 | * This entry has greater va than the entries before. | | 906 | * This entry has greater va than the entries before. |
907 | * We need to make it point to the last page, not past it. | | 907 | * We need to make it point to the last page, not past it. |
908 | */ | | 908 | */ |
909 | | | 909 | |
910 | if (ent->protection & VM_PROT_EXECUTE) | | 910 | if (ent->protection & VM_PROT_EXECUTE) |
911 | va = trunc_page(ent->end) - PAGE_SIZE; | | 911 | va = trunc_page(ent->end) - PAGE_SIZE; |
912 | } | | 912 | } |
913 | vm_map_unlock_read(map); | | 913 | vm_map_unlock_read(map); |
914 | if (va == pm->pm_hiexec && tf->tf_cs == GSEL(GUCODEBIG_SEL, SEL_UPL)) | | 914 | if (va == pm->pm_hiexec && tf->tf_cs == GSEL(GUCODEBIG_SEL, SEL_UPL)) |
915 | return (0); | | 915 | return (0); |
916 | | | 916 | |
917 | pm->pm_hiexec = va; | | 917 | pm->pm_hiexec = va; |
918 | if (pm->pm_hiexec > I386_MAX_EXE_ADDR) { | | 918 | if (pm->pm_hiexec > I386_MAX_EXE_ADDR) { |
919 | tf->tf_cs = GSEL(GUCODEBIG_SEL, SEL_UPL); | | 919 | tf->tf_cs = GSEL(GUCODEBIG_SEL, SEL_UPL); |
920 | } else { | | 920 | } else { |
921 | tf->tf_cs = GSEL(GUCODE_SEL, SEL_UPL); | | 921 | tf->tf_cs = GSEL(GUCODE_SEL, SEL_UPL); |
922 | return (0); | | 922 | return (0); |
923 | } | | 923 | } |
924 | return (1); | | 924 | return (1); |
925 | } | | 925 | } |
926 | #endif /* !defined(__x86_64__) */ | | 926 | #endif /* !defined(__x86_64__) */ |
927 | | | 927 | |
928 | void | | 928 | void |
929 | pat_init(struct cpu_info *ci) | | 929 | pat_init(struct cpu_info *ci) |
930 | { | | 930 | { |
931 | uint64_t pat; | | 931 | uint64_t pat; |
932 | | | 932 | |
933 | if (!(ci->ci_feat_val[0] & CPUID_PAT)) | | 933 | if (!(ci->ci_feat_val[0] & CPUID_PAT)) |
934 | return; | | 934 | return; |
935 | | | 935 | |
936 | /* We change WT to WC. Leave all other entries the default values. */ | | 936 | /* We change WT to WC. Leave all other entries the default values. */ |
937 | pat = PATENTRY(0, PAT_WB) | PATENTRY(1, PAT_WC) | | | 937 | pat = PATENTRY(0, PAT_WB) | PATENTRY(1, PAT_WC) | |
938 | PATENTRY(2, PAT_UCMINUS) | PATENTRY(3, PAT_UC) | | | 938 | PATENTRY(2, PAT_UCMINUS) | PATENTRY(3, PAT_UC) | |
939 | PATENTRY(4, PAT_WB) | PATENTRY(5, PAT_WC) | | | 939 | PATENTRY(4, PAT_WB) | PATENTRY(5, PAT_WC) | |
940 | PATENTRY(6, PAT_UCMINUS) | PATENTRY(7, PAT_UC); | | 940 | PATENTRY(6, PAT_UCMINUS) | PATENTRY(7, PAT_UC); |
941 | | | 941 | |
942 | wrmsr(MSR_CR_PAT, pat); | | 942 | wrmsr(MSR_CR_PAT, pat); |
943 | cpu_pat_enabled = true; | | 943 | cpu_pat_enabled = true; |
944 | aprint_debug_dev(ci->ci_dev, "PAT enabled\n"); | | 944 | aprint_debug_dev(ci->ci_dev, "PAT enabled\n"); |
945 | } | | 945 | } |
946 | | | 946 | |
947 | static pt_entry_t | | 947 | static pt_entry_t |
948 | pmap_pat_flags(u_int flags) | | 948 | pmap_pat_flags(u_int flags) |
949 | { | | 949 | { |
950 | u_int cacheflags = (flags & PMAP_CACHE_MASK); | | 950 | u_int cacheflags = (flags & PMAP_CACHE_MASK); |
951 | | | 951 | |
952 | if (!cpu_pat_enabled) { | | 952 | if (!cpu_pat_enabled) { |
953 | switch (cacheflags) { | | 953 | switch (cacheflags) { |
954 | case PMAP_NOCACHE: | | 954 | case PMAP_NOCACHE: |
955 | case PMAP_NOCACHE_OVR: | | 955 | case PMAP_NOCACHE_OVR: |
956 | /* results in PGC_UCMINUS on cpus which have | | 956 | /* results in PGC_UCMINUS on cpus which have |
957 | * the cpuid PAT but PAT "disabled" | | 957 | * the cpuid PAT but PAT "disabled" |
958 | */ | | 958 | */ |
959 | return PG_N; | | 959 | return PG_N; |
960 | default: | | 960 | default: |
961 | return 0; | | 961 | return 0; |
962 | } | | 962 | } |
963 | } | | 963 | } |
964 | | | 964 | |
965 | switch (cacheflags) { | | 965 | switch (cacheflags) { |
966 | case PMAP_NOCACHE: | | 966 | case PMAP_NOCACHE: |
967 | return PGC_UC; | | 967 | return PGC_UC; |
968 | case PMAP_WRITE_COMBINE: | | 968 | case PMAP_WRITE_COMBINE: |
969 | return PGC_WC; | | 969 | return PGC_WC; |
970 | case PMAP_WRITE_BACK: | | 970 | case PMAP_WRITE_BACK: |
971 | return PGC_WB; | | 971 | return PGC_WB; |
972 | case PMAP_NOCACHE_OVR: | | 972 | case PMAP_NOCACHE_OVR: |
973 | return PGC_UCMINUS; | | 973 | return PGC_UCMINUS; |
974 | } | | 974 | } |
975 | | | 975 | |
976 | return 0; | | 976 | return 0; |
977 | } | | 977 | } |
978 | | | 978 | |
979 | /* | | 979 | /* |
980 | * p m a p k e n t e r f u n c t i o n s | | 980 | * p m a p k e n t e r f u n c t i o n s |
981 | * | | 981 | * |
982 | * functions to quickly enter/remove pages from the kernel address | | 982 | * functions to quickly enter/remove pages from the kernel address |
983 | * space. pmap_kremove is exported to MI kernel. we make use of | | 983 | * space. pmap_kremove is exported to MI kernel. we make use of |
984 | * the recursive PTE mappings. | | 984 | * the recursive PTE mappings. |
985 | */ | | 985 | */ |
986 | | | 986 | |
987 | /* | | 987 | /* |
988 | * pmap_kenter_pa: enter a kernel mapping without R/M (pv_entry) tracking | | 988 | * pmap_kenter_pa: enter a kernel mapping without R/M (pv_entry) tracking |
989 | * | | 989 | * |
990 | * => no need to lock anything, assume va is already allocated | | 990 | * => no need to lock anything, assume va is already allocated |
991 | * => should be faster than normal pmap enter function | | 991 | * => should be faster than normal pmap enter function |
992 | */ | | 992 | */ |
993 | | | 993 | |
994 | void | | 994 | void |
995 | pmap_kenter_pa(vaddr_t va, paddr_t pa, vm_prot_t prot, u_int flags) | | 995 | pmap_kenter_pa(vaddr_t va, paddr_t pa, vm_prot_t prot, u_int flags) |
996 | { | | 996 | { |
997 | pt_entry_t *pte, opte, npte; | | 997 | pt_entry_t *pte, opte, npte; |
998 | | | 998 | |
999 | KASSERT(!(prot & ~VM_PROT_ALL)); | | 999 | KASSERT(!(prot & ~VM_PROT_ALL)); |
1000 | | | 1000 | |
1001 | if (va < VM_MIN_KERNEL_ADDRESS) | | 1001 | if (va < VM_MIN_KERNEL_ADDRESS) |
1002 | pte = vtopte(va); | | 1002 | pte = vtopte(va); |
1003 | else | | 1003 | else |
1004 | pte = kvtopte(va); | | 1004 | pte = kvtopte(va); |
1005 | #ifdef DOM0OPS | | 1005 | #ifdef DOM0OPS |
1006 | if (pa < pmap_pa_start || pa >= pmap_pa_end) { | | 1006 | if (pa < pmap_pa_start || pa >= pmap_pa_end) { |
1007 | #ifdef DEBUG | | 1007 | #ifdef DEBUG |
1008 | printf_nolog("%s: pa 0x%" PRIx64 " for va 0x%" PRIx64 | | 1008 | printf_nolog("%s: pa 0x%" PRIx64 " for va 0x%" PRIx64 |
1009 | " outside range\n", __func__, (int64_t)pa, (int64_t)va); | | 1009 | " outside range\n", __func__, (int64_t)pa, (int64_t)va); |
1010 | #endif /* DEBUG */ | | 1010 | #endif /* DEBUG */ |
1011 | npte = pa; | | 1011 | npte = pa; |
1012 | } else | | 1012 | } else |
1013 | #endif /* DOM0OPS */ | | 1013 | #endif /* DOM0OPS */ |
1014 | npte = pmap_pa2pte(pa); | | 1014 | npte = pmap_pa2pte(pa); |
1015 | npte |= protection_codes[prot] | PG_k | PG_V | pmap_pg_g; | | 1015 | npte |= protection_codes[prot] | PG_k | PG_V | pmap_pg_g; |
1016 | npte |= pmap_pat_flags(flags); | | 1016 | npte |= pmap_pat_flags(flags); |
1017 | opte = pmap_pte_testset(pte, npte); /* zap! */ | | 1017 | opte = pmap_pte_testset(pte, npte); /* zap! */ |
1018 | #if defined(DIAGNOSTIC) | | 1018 | #if defined(DIAGNOSTIC) |
1019 | /* XXX For now... */ | | 1019 | /* XXX For now... */ |
1020 | if (opte & PG_PS) | | 1020 | if (opte & PG_PS) |
1021 | panic("%s: PG_PS", __func__); | | 1021 | panic("%s: PG_PS", __func__); |
1022 | #endif | | 1022 | #endif |
1023 | if ((opte & (PG_V | PG_U)) == (PG_V | PG_U)) { | | 1023 | if ((opte & (PG_V | PG_U)) == (PG_V | PG_U)) { |
1024 | #if defined(DIAGNOSTIC) | | 1024 | #if defined(DIAGNOSTIC) |
1025 | printf_nolog("%s: mapping already present\n", __func__); | | 1025 | printf_nolog("%s: mapping already present\n", __func__); |
1026 | #endif | | 1026 | #endif |
1027 | /* This should not happen. */ | | 1027 | /* This should not happen. */ |
1028 | kpreempt_disable(); | | 1028 | kpreempt_disable(); |
1029 | pmap_tlb_shootdown(pmap_kernel(), va, opte, TLBSHOOT_KENTER); | | 1029 | pmap_tlb_shootdown(pmap_kernel(), va, opte, TLBSHOOT_KENTER); |
1030 | kpreempt_enable(); | | 1030 | kpreempt_enable(); |
1031 | } | | 1031 | } |
1032 | } | | 1032 | } |
1033 | | | 1033 | |
1034 | void | | 1034 | void |
1035 | pmap_emap_enter(vaddr_t va, paddr_t pa, vm_prot_t prot) | | 1035 | pmap_emap_enter(vaddr_t va, paddr_t pa, vm_prot_t prot) |
1036 | { | | 1036 | { |
1037 | pt_entry_t *pte, opte, npte; | | 1037 | pt_entry_t *pte, opte, npte; |
1038 | | | 1038 | |
1039 | KASSERT((prot & ~VM_PROT_ALL) == 0); | | 1039 | KASSERT((prot & ~VM_PROT_ALL) == 0); |
1040 | pte = (va < VM_MIN_KERNEL_ADDRESS) ? vtopte(va) : kvtopte(va); | | 1040 | pte = (va < VM_MIN_KERNEL_ADDRESS) ? vtopte(va) : kvtopte(va); |
1041 | | | 1041 | |
1042 | #ifdef DOM0OPS | | 1042 | #ifdef DOM0OPS |
1043 | if (pa < pmap_pa_start || pa >= pmap_pa_end) { | | 1043 | if (pa < pmap_pa_start || pa >= pmap_pa_end) { |
1044 | npte = pa; | | 1044 | npte = pa; |
1045 | } else | | 1045 | } else |
1046 | #endif | | 1046 | #endif |
1047 | npte = pmap_pa2pte(pa); | | 1047 | npte = pmap_pa2pte(pa); |
1048 | | | 1048 | |
1049 | npte = pmap_pa2pte(pa); | | 1049 | npte = pmap_pa2pte(pa); |
1050 | npte |= protection_codes[prot] | PG_k | PG_V; | | 1050 | npte |= protection_codes[prot] | PG_k | PG_V; |
1051 | opte = pmap_pte_testset(pte, npte); | | 1051 | opte = pmap_pte_testset(pte, npte); |
1052 | } | | 1052 | } |
1053 | | | 1053 | |
1054 | /* | | 1054 | /* |
1055 | * pmap_emap_sync: perform TLB flush or pmap load, if it was deferred. | | 1055 | * pmap_emap_sync: perform TLB flush or pmap load, if it was deferred. |
1056 | */ | | 1056 | */ |
1057 | void | | 1057 | void |
1058 | pmap_emap_sync(bool canload) | | 1058 | pmap_emap_sync(bool canload) |
1059 | { | | 1059 | { |
1060 | struct cpu_info *ci = curcpu(); | | 1060 | struct cpu_info *ci = curcpu(); |
1061 | struct pmap *pmap; | | 1061 | struct pmap *pmap; |
1062 | | | 1062 | |
1063 | KASSERT(kpreempt_disabled()); | | 1063 | KASSERT(kpreempt_disabled()); |
1064 | if (__predict_true(ci->ci_want_pmapload && canload)) { | | 1064 | if (__predict_true(ci->ci_want_pmapload && canload)) { |
1065 | /* | | 1065 | /* |
1066 | * XXX: Hint for pmap_reactivate(), which might suggest to | | 1066 | * XXX: Hint for pmap_reactivate(), which might suggest to |
1067 | * not perform TLB flush, if state has not changed. | | 1067 | * not perform TLB flush, if state has not changed. |
1068 | */ | | 1068 | */ |
1069 | pmap = vm_map_pmap(&curlwp->l_proc->p_vmspace->vm_map); | | 1069 | pmap = vm_map_pmap(&curlwp->l_proc->p_vmspace->vm_map); |
1070 | if (__predict_false(pmap == ci->ci_pmap)) { | | 1070 | if (__predict_false(pmap == ci->ci_pmap)) { |
1071 | const uint32_t cpumask = ci->ci_cpumask; | | 1071 | const uint32_t cpumask = ci->ci_cpumask; |
1072 | atomic_and_32(&pmap->pm_cpus, ~cpumask); | | 1072 | atomic_and_32(&pmap->pm_cpus, ~cpumask); |
1073 | } | | 1073 | } |
1074 | pmap_load(); | | 1074 | pmap_load(); |
1075 | KASSERT(ci->ci_want_pmapload == 0); | | 1075 | KASSERT(ci->ci_want_pmapload == 0); |
1076 | } else { | | 1076 | } else { |
1077 | tlbflush(); | | 1077 | tlbflush(); |
1078 | } | | 1078 | } |
1079 | | | 1079 | |
1080 | } | | 1080 | } |
1081 | | | 1081 | |
1082 | void | | 1082 | void |
1083 | pmap_emap_remove(vaddr_t sva, vsize_t len) | | 1083 | pmap_emap_remove(vaddr_t sva, vsize_t len) |
1084 | { | | 1084 | { |
1085 | pt_entry_t *pte, xpte; | | 1085 | pt_entry_t *pte, xpte; |
1086 | vaddr_t va, eva = sva + len; | | 1086 | vaddr_t va, eva = sva + len; |
1087 | | | 1087 | |
1088 | for (va = sva; va < eva; va += PAGE_SIZE) { | | 1088 | for (va = sva; va < eva; va += PAGE_SIZE) { |
1089 | pte = (va < VM_MIN_KERNEL_ADDRESS) ? vtopte(va) : kvtopte(va); | | 1089 | pte = (va < VM_MIN_KERNEL_ADDRESS) ? vtopte(va) : kvtopte(va); |
1090 | xpte |= pmap_pte_testset(pte, 0); | | 1090 | xpte |= pmap_pte_testset(pte, 0); |
1091 | } | | 1091 | } |
1092 | } | | 1092 | } |
1093 | | | 1093 | |
1094 | __strict_weak_alias(pmap_kenter_ma, pmap_kenter_pa); | | 1094 | __strict_weak_alias(pmap_kenter_ma, pmap_kenter_pa); |
1095 | | | 1095 | |
1096 | #if defined(__x86_64__) | | 1096 | #if defined(__x86_64__) |
1097 | /* | | 1097 | /* |
1098 | * Change protection for a virtual address. Local for a CPU only, don't | | 1098 | * Change protection for a virtual address. Local for a CPU only, don't |
1099 | * care about TLB shootdowns. | | 1099 | * care about TLB shootdowns. |
1100 | * | | 1100 | * |
1101 | * => must be called with preemption disabled | | 1101 | * => must be called with preemption disabled |
1102 | */ | | 1102 | */ |
1103 | void | | 1103 | void |
1104 | pmap_changeprot_local(vaddr_t va, vm_prot_t prot) | | 1104 | pmap_changeprot_local(vaddr_t va, vm_prot_t prot) |
1105 | { | | 1105 | { |
1106 | pt_entry_t *pte, opte, npte; | | 1106 | pt_entry_t *pte, opte, npte; |
1107 | | | 1107 | |
1108 | KASSERT(kpreempt_disabled()); | | 1108 | KASSERT(kpreempt_disabled()); |
1109 | | | 1109 | |
1110 | if (va < VM_MIN_KERNEL_ADDRESS) | | 1110 | if (va < VM_MIN_KERNEL_ADDRESS) |
1111 | pte = vtopte(va); | | 1111 | pte = vtopte(va); |
1112 | else | | 1112 | else |
1113 | pte = kvtopte(va); | | 1113 | pte = kvtopte(va); |
1114 | | | 1114 | |
1115 | npte = opte = *pte; | | 1115 | npte = opte = *pte; |
1116 | | | 1116 | |
1117 | if ((prot & VM_PROT_WRITE) != 0) | | 1117 | if ((prot & VM_PROT_WRITE) != 0) |
1118 | npte |= PG_RW; | | 1118 | npte |= PG_RW; |
1119 | else | | 1119 | else |
1120 | npte &= ~PG_RW; | | 1120 | npte &= ~PG_RW; |
1121 | | | 1121 | |
1122 | if (opte != npte) { | | 1122 | if (opte != npte) { |
1123 | pmap_pte_set(pte, npte); | | 1123 | pmap_pte_set(pte, npte); |
1124 | pmap_pte_flush(); | | 1124 | pmap_pte_flush(); |
1125 | invlpg(va); | | 1125 | invlpg(va); |
1126 | } | | 1126 | } |
1127 | } | | 1127 | } |
1128 | #endif /* defined(__x86_64__) */ | | 1128 | #endif /* defined(__x86_64__) */ |
1129 | | | 1129 | |
1130 | /* | | 1130 | /* |
1131 | * pmap_kremove: remove a kernel mapping(s) without R/M (pv_entry) tracking | | 1131 | * pmap_kremove: remove a kernel mapping(s) without R/M (pv_entry) tracking |
1132 | * | | 1132 | * |
1133 | * => no need to lock anything | | 1133 | * => no need to lock anything |
1134 | * => caller must dispose of any vm_page mapped in the va range | | 1134 | * => caller must dispose of any vm_page mapped in the va range |
1135 | * => note: not an inline function | | 1135 | * => note: not an inline function |
1136 | * => we assume the va is page aligned and the len is a multiple of PAGE_SIZE | | 1136 | * => we assume the va is page aligned and the len is a multiple of PAGE_SIZE |
1137 | * => we assume kernel only unmaps valid addresses and thus don't bother | | 1137 | * => we assume kernel only unmaps valid addresses and thus don't bother |
1138 | * checking the valid bit before doing TLB flushing | | 1138 | * checking the valid bit before doing TLB flushing |
1139 | * => must be followed by call to pmap_update() before reuse of page | | 1139 | * => must be followed by call to pmap_update() before reuse of page |
1140 | */ | | 1140 | */ |
1141 | | | 1141 | |
1142 | void | | 1142 | void |
1143 | pmap_kremove(vaddr_t sva, vsize_t len) | | 1143 | pmap_kremove(vaddr_t sva, vsize_t len) |
1144 | { | | 1144 | { |
1145 | pt_entry_t *pte, opte; | | 1145 | pt_entry_t *pte, opte; |
1146 | vaddr_t va, eva; | | 1146 | vaddr_t va, eva; |
1147 | | | 1147 | |
1148 | eva = sva + len; | | 1148 | eva = sva + len; |
1149 | | | 1149 | |
1150 | kpreempt_disable(); | | 1150 | kpreempt_disable(); |
1151 | for (va = sva; va < eva; va += PAGE_SIZE) { | | 1151 | for (va = sva; va < eva; va += PAGE_SIZE) { |
1152 | if (va < VM_MIN_KERNEL_ADDRESS) | | 1152 | if (va < VM_MIN_KERNEL_ADDRESS) |
1153 | pte = vtopte(va); | | 1153 | pte = vtopte(va); |
1154 | else | | 1154 | else |
1155 | pte = kvtopte(va); | | 1155 | pte = kvtopte(va); |
1156 | opte = pmap_pte_testset(pte, 0); /* zap! */ | | 1156 | opte = pmap_pte_testset(pte, 0); /* zap! */ |
1157 | if ((opte & (PG_V | PG_U)) == (PG_V | PG_U)) { | | 1157 | if ((opte & (PG_V | PG_U)) == (PG_V | PG_U)) { |
1158 | pmap_tlb_shootdown(pmap_kernel(), va, opte, | | 1158 | pmap_tlb_shootdown(pmap_kernel(), va, opte, |
1159 | TLBSHOOT_KREMOVE); | | 1159 | TLBSHOOT_KREMOVE); |
1160 | } | | 1160 | } |
1161 | KASSERT((opte & PG_PS) == 0); | | 1161 | KASSERT((opte & PG_PS) == 0); |
1162 | KASSERT((opte & PG_PVLIST) == 0); | | 1162 | KASSERT((opte & PG_PVLIST) == 0); |
1163 | } | | 1163 | } |
1164 | kpreempt_enable(); | | 1164 | kpreempt_enable(); |
1165 | } | | 1165 | } |
1166 | | | 1166 | |
1167 | /* | | 1167 | /* |
1168 | * p m a p i n i t f u n c t i o n s | | 1168 | * p m a p i n i t f u n c t i o n s |
1169 | * | | 1169 | * |
1170 | * pmap_bootstrap and pmap_init are called during system startup | | 1170 | * pmap_bootstrap and pmap_init are called during system startup |
1171 | * to init the pmap module. pmap_bootstrap() does a low level | | 1171 | * to init the pmap module. pmap_bootstrap() does a low level |
1172 | * init just to get things rolling. pmap_init() finishes the job. | | 1172 | * init just to get things rolling. pmap_init() finishes the job. |
1173 | */ | | 1173 | */ |
1174 | | | 1174 | |
1175 | /* | | 1175 | /* |
1176 | * pmap_bootstrap: get the system in a state where it can run with VM | | 1176 | * pmap_bootstrap: get the system in a state where it can run with VM |
1177 | * properly enabled (called before main()). the VM system is | | 1177 | * properly enabled (called before main()). the VM system is |
1178 | * fully init'd later... | | 1178 | * fully init'd later... |
1179 | * | | 1179 | * |
1180 | * => on i386, locore.s has already enabled the MMU by allocating | | 1180 | * => on i386, locore.s has already enabled the MMU by allocating |
1181 | * a PDP for the kernel, and nkpde PTP's for the kernel. | | 1181 | * a PDP for the kernel, and nkpde PTP's for the kernel. |
1182 | * => kva_start is the first free virtual address in kernel space | | 1182 | * => kva_start is the first free virtual address in kernel space |
1183 | */ | | 1183 | */ |
1184 | | | 1184 | |
1185 | void | | 1185 | void |
1186 | pmap_bootstrap(vaddr_t kva_start) | | 1186 | pmap_bootstrap(vaddr_t kva_start) |
1187 | { | | 1187 | { |
1188 | struct pmap *kpm; | | 1188 | struct pmap *kpm; |
1189 | pt_entry_t *pte; | | 1189 | pt_entry_t *pte; |
1190 | int i; | | 1190 | int i; |
1191 | vaddr_t kva; | | 1191 | vaddr_t kva; |
1192 | #ifndef XEN | | 1192 | #ifndef XEN |
1193 | pd_entry_t *pde; | | 1193 | pd_entry_t *pde; |
1194 | unsigned long p1i; | | 1194 | unsigned long p1i; |
1195 | vaddr_t kva_end; | | 1195 | vaddr_t kva_end; |
1196 | #endif | | 1196 | #endif |
1197 | #ifdef __HAVE_DIRECT_MAP | | 1197 | #ifdef __HAVE_DIRECT_MAP |
1198 | phys_ram_seg_t *mc; | | 1198 | phys_ram_seg_t *mc; |
1199 | long ndmpdp; | | 1199 | long ndmpdp; |
1200 | paddr_t dmpd, dmpdp, pdp; | | 1200 | paddr_t dmpd, dmpdp, pdp; |
1201 | vaddr_t tmpva; | | 1201 | vaddr_t tmpva; |
1202 | #endif | | 1202 | #endif |
1203 | | | 1203 | |
1204 | pt_entry_t pg_nx = (cpu_feature[2] & CPUID_NOX ? PG_NX : 0); | | 1204 | pt_entry_t pg_nx = (cpu_feature[2] & CPUID_NOX ? PG_NX : 0); |
1205 | | | 1205 | |
1206 | /* | | 1206 | /* |
1207 | * set up our local static global vars that keep track of the | | 1207 | * set up our local static global vars that keep track of the |
1208 | * usage of KVM before kernel_map is set up | | 1208 | * usage of KVM before kernel_map is set up |
1209 | */ | | 1209 | */ |
1210 | | | 1210 | |
1211 | virtual_avail = kva_start; /* first free KVA */ | | 1211 | virtual_avail = kva_start; /* first free KVA */ |
1212 | virtual_end = VM_MAX_KERNEL_ADDRESS; /* last KVA */ | | 1212 | virtual_end = VM_MAX_KERNEL_ADDRESS; /* last KVA */ |
1213 | | | 1213 | |
1214 | /* | | 1214 | /* |
1215 | * set up protection_codes: we need to be able to convert from | | 1215 | * set up protection_codes: we need to be able to convert from |
1216 | * a MI protection code (some combo of VM_PROT...) to something | | 1216 | * a MI protection code (some combo of VM_PROT...) to something |
1217 | * we can jam into a i386 PTE. | | 1217 | * we can jam into a i386 PTE. |
1218 | */ | | 1218 | */ |
1219 | | | 1219 | |
1220 | protection_codes[VM_PROT_NONE] = pg_nx; /* --- */ | | 1220 | protection_codes[VM_PROT_NONE] = pg_nx; /* --- */ |
1221 | protection_codes[VM_PROT_EXECUTE] = PG_RO | PG_X; /* --x */ | | 1221 | protection_codes[VM_PROT_EXECUTE] = PG_RO | PG_X; /* --x */ |
1222 | protection_codes[VM_PROT_READ] = PG_RO | pg_nx; /* -r- */ | | 1222 | protection_codes[VM_PROT_READ] = PG_RO | pg_nx; /* -r- */ |
1223 | protection_codes[VM_PROT_READ|VM_PROT_EXECUTE] = PG_RO | PG_X;/* -rx */ | | 1223 | protection_codes[VM_PROT_READ|VM_PROT_EXECUTE] = PG_RO | PG_X;/* -rx */ |
1224 | protection_codes[VM_PROT_WRITE] = PG_RW | pg_nx; /* w-- */ | | 1224 | protection_codes[VM_PROT_WRITE] = PG_RW | pg_nx; /* w-- */ |
1225 | protection_codes[VM_PROT_WRITE|VM_PROT_EXECUTE] = PG_RW | PG_X;/* w-x */ | | 1225 | protection_codes[VM_PROT_WRITE|VM_PROT_EXECUTE] = PG_RW | PG_X;/* w-x */ |
1226 | protection_codes[VM_PROT_WRITE|VM_PROT_READ] = PG_RW | pg_nx; | | 1226 | protection_codes[VM_PROT_WRITE|VM_PROT_READ] = PG_RW | pg_nx; |
1227 | /* wr- */ | | 1227 | /* wr- */ |
1228 | protection_codes[VM_PROT_ALL] = PG_RW | PG_X; /* wrx */ | | 1228 | protection_codes[VM_PROT_ALL] = PG_RW | PG_X; /* wrx */ |
1229 | | | 1229 | |
1230 | /* | | 1230 | /* |
1231 | * now we init the kernel's pmap | | 1231 | * now we init the kernel's pmap |
1232 | * | | 1232 | * |
1233 | * the kernel pmap's pm_obj is not used for much. however, in | | 1233 | * the kernel pmap's pm_obj is not used for much. however, in |
1234 | * user pmaps the pm_obj contains the list of active PTPs. | | 1234 | * user pmaps the pm_obj contains the list of active PTPs. |
1235 | * the pm_obj currently does not have a pager. it might be possible | | 1235 | * the pm_obj currently does not have a pager. it might be possible |
1236 | * to add a pager that would allow a process to read-only mmap its | | 1236 | * to add a pager that would allow a process to read-only mmap its |
1237 | * own page tables (fast user level vtophys?). this may or may not | | 1237 | * own page tables (fast user level vtophys?). this may or may not |
1238 | * be useful. | | 1238 | * be useful. |
1239 | */ | | 1239 | */ |
1240 | | | 1240 | |
1241 | kpm = pmap_kernel(); | | 1241 | kpm = pmap_kernel(); |
1242 | for (i = 0; i < PTP_LEVELS - 1; i++) { | | 1242 | for (i = 0; i < PTP_LEVELS - 1; i++) { |
1243 | mutex_init(&kpm->pm_obj_lock[i], MUTEX_DEFAULT, IPL_NONE); | | 1243 | mutex_init(&kpm->pm_obj_lock[i], MUTEX_DEFAULT, IPL_NONE); |
1244 | uvm_obj_init(&kpm->pm_obj[i], NULL, false, 1); | | 1244 | uvm_obj_init(&kpm->pm_obj[i], NULL, false, 1); |
1245 | uvm_obj_setlock(&kpm->pm_obj[i], &kpm->pm_obj_lock[i]); | | 1245 | uvm_obj_setlock(&kpm->pm_obj[i], &kpm->pm_obj_lock[i]); |
1246 | kpm->pm_ptphint[i] = NULL; | | 1246 | kpm->pm_ptphint[i] = NULL; |
1247 | } | | 1247 | } |
1248 | memset(&kpm->pm_list, 0, sizeof(kpm->pm_list)); /* pm_list not used */ | | 1248 | memset(&kpm->pm_list, 0, sizeof(kpm->pm_list)); /* pm_list not used */ |
1249 | | | 1249 | |
1250 | kpm->pm_pdir = (pd_entry_t *)(PDPpaddr + KERNBASE); | | 1250 | kpm->pm_pdir = (pd_entry_t *)(PDPpaddr + KERNBASE); |
1251 | for (i = 0; i < PDP_SIZE; i++) | | 1251 | for (i = 0; i < PDP_SIZE; i++) |
1252 | kpm->pm_pdirpa[i] = PDPpaddr + PAGE_SIZE * i; | | 1252 | kpm->pm_pdirpa[i] = PDPpaddr + PAGE_SIZE * i; |
1253 | | | 1253 | |
1254 | kpm->pm_stats.wired_count = kpm->pm_stats.resident_count = | | 1254 | kpm->pm_stats.wired_count = kpm->pm_stats.resident_count = |
1255 | x86_btop(kva_start - VM_MIN_KERNEL_ADDRESS); | | 1255 | x86_btop(kva_start - VM_MIN_KERNEL_ADDRESS); |
1256 | | | 1256 | |
1257 | /* | | 1257 | /* |
1258 | * the above is just a rough estimate and not critical to the proper | | 1258 | * the above is just a rough estimate and not critical to the proper |
1259 | * operation of the system. | | 1259 | * operation of the system. |
1260 | */ | | 1260 | */ |
1261 | | | 1261 | |
1262 | #ifndef XEN | | 1262 | #ifndef XEN |
1263 | /* | | 1263 | /* |
1264 | * Begin to enable global TLB entries if they are supported. | | 1264 | * Begin to enable global TLB entries if they are supported. |
1265 | * The G bit has no effect until the CR4_PGE bit is set in CR4, | | 1265 | * The G bit has no effect until the CR4_PGE bit is set in CR4, |
1266 | * which happens in cpu_init(), which is run on each cpu | | 1266 | * which happens in cpu_init(), which is run on each cpu |
1267 | * (and happens later) | | 1267 | * (and happens later) |
1268 | */ | | 1268 | */ |
1269 | | | 1269 | |
1270 | if (cpu_feature[0] & CPUID_PGE) { | | 1270 | if (cpu_feature[0] & CPUID_PGE) { |
1271 | pmap_pg_g = PG_G; /* enable software */ | | 1271 | pmap_pg_g = PG_G; /* enable software */ |
1272 | | | 1272 | |
1273 | /* add PG_G attribute to already mapped kernel pages */ | | 1273 | /* add PG_G attribute to already mapped kernel pages */ |
1274 | if (KERNBASE == VM_MIN_KERNEL_ADDRESS) { | | 1274 | if (KERNBASE == VM_MIN_KERNEL_ADDRESS) { |
1275 | kva_end = virtual_avail; | | 1275 | kva_end = virtual_avail; |
1276 | } else { | | 1276 | } else { |
1277 | extern vaddr_t eblob, esym; | | 1277 | extern vaddr_t eblob, esym; |
1278 | kva_end = (vaddr_t)&end; | | 1278 | kva_end = (vaddr_t)&end; |
1279 | if (esym > kva_end) | | 1279 | if (esym > kva_end) |
1280 | kva_end = esym; | | 1280 | kva_end = esym; |
1281 | if (eblob > kva_end) | | 1281 | if (eblob > kva_end) |
1282 | kva_end = eblob; | | 1282 | kva_end = eblob; |
1283 | kva_end = roundup(kva_end, PAGE_SIZE); | | 1283 | kva_end = roundup(kva_end, PAGE_SIZE); |
1284 | } | | 1284 | } |
1285 | for (kva = KERNBASE; kva < kva_end; kva += PAGE_SIZE) { | | 1285 | for (kva = KERNBASE; kva < kva_end; kva += PAGE_SIZE) { |
1286 | p1i = pl1_i(kva); | | 1286 | p1i = pl1_i(kva); |
1287 | if (pmap_valid_entry(PTE_BASE[p1i])) | | 1287 | if (pmap_valid_entry(PTE_BASE[p1i])) |
1288 | PTE_BASE[p1i] |= PG_G; | | 1288 | PTE_BASE[p1i] |= PG_G; |
1289 | } | | 1289 | } |
1290 | } | | 1290 | } |
1291 | | | 1291 | |
1292 | /* | | 1292 | /* |
1293 | * enable large pages if they are supported. | | 1293 | * enable large pages if they are supported. |
1294 | */ | | 1294 | */ |
1295 | | | 1295 | |
1296 | if (cpu_feature[0] & CPUID_PSE) { | | 1296 | if (cpu_feature[0] & CPUID_PSE) { |
1297 | paddr_t pa; | | 1297 | paddr_t pa; |
1298 | extern char __data_start; | | 1298 | extern char __data_start; |
1299 | | | 1299 | |
1300 | lcr4(rcr4() | CR4_PSE); /* enable hardware (via %cr4) */ | | 1300 | lcr4(rcr4() | CR4_PSE); /* enable hardware (via %cr4) */ |
1301 | pmap_largepages = 1; /* enable software */ | | 1301 | pmap_largepages = 1; /* enable software */ |
1302 | | | 1302 | |
1303 | /* | | 1303 | /* |
1304 | * the TLB must be flushed after enabling large pages | | 1304 | * the TLB must be flushed after enabling large pages |
1305 | * on Pentium CPUs, according to section 3.6.2.2 of | | 1305 | * on Pentium CPUs, according to section 3.6.2.2 of |
1306 | * "Intel Architecture Software Developer's Manual, | | 1306 | * "Intel Architecture Software Developer's Manual, |
1307 | * Volume 3: System Programming". | | 1307 | * Volume 3: System Programming". |
1308 | */ | | 1308 | */ |
1309 | tlbflushg(); | | 1309 | tlbflushg(); |
1310 | | | 1310 | |
1311 | /* | | 1311 | /* |
1312 | * now, remap the kernel text using large pages. we | | 1312 | * now, remap the kernel text using large pages. we |
1313 | * assume that the linker has properly aligned the | | 1313 | * assume that the linker has properly aligned the |
1314 | * .data segment to a NBPD_L2 boundary. | | 1314 | * .data segment to a NBPD_L2 boundary. |
1315 | */ | | 1315 | */ |
1316 | kva_end = rounddown((vaddr_t)&__data_start, NBPD_L1); | | 1316 | kva_end = rounddown((vaddr_t)&__data_start, NBPD_L1); |
1317 | for (pa = 0, kva = KERNBASE; kva + NBPD_L2 <= kva_end; | | 1317 | for (pa = 0, kva = KERNBASE; kva + NBPD_L2 <= kva_end; |
1318 | kva += NBPD_L2, pa += NBPD_L2) { | | 1318 | kva += NBPD_L2, pa += NBPD_L2) { |
1319 | pde = &L2_BASE[pl2_i(kva)]; | | 1319 | pde = &L2_BASE[pl2_i(kva)]; |
1320 | *pde = pa | pmap_pg_g | PG_PS | | | 1320 | *pde = pa | pmap_pg_g | PG_PS | |
1321 | PG_KR | PG_V; /* zap! */ | | 1321 | PG_KR | PG_V; /* zap! */ |
1322 | tlbflushg(); | | 1322 | tlbflushg(); |
1323 | } | | 1323 | } |
1324 | #if defined(DEBUG) | | 1324 | #if defined(DEBUG) |
1325 | aprint_normal("kernel text is mapped with %" PRIuPSIZE " large " | | 1325 | aprint_normal("kernel text is mapped with %" PRIuPSIZE " large " |
1326 | "pages and %" PRIuPSIZE " normal pages\n", | | 1326 | "pages and %" PRIuPSIZE " normal pages\n", |
1327 | howmany(kva - KERNBASE, NBPD_L2), | | 1327 | howmany(kva - KERNBASE, NBPD_L2), |
1328 | howmany((vaddr_t)&__data_start - kva, NBPD_L1)); | | 1328 | howmany((vaddr_t)&__data_start - kva, NBPD_L1)); |
1329 | #endif /* defined(DEBUG) */ | | 1329 | #endif /* defined(DEBUG) */ |
1330 | } | | 1330 | } |
1331 | #endif /* !XEN */ | | 1331 | #endif /* !XEN */ |
1332 | | | 1332 | |
1333 | #ifdef __HAVE_DIRECT_MAP | | 1333 | #ifdef __HAVE_DIRECT_MAP |
1334 | | | 1334 | |
1335 | tmpva = (KERNBASE + NKL2_KIMG_ENTRIES * NBPD_L2); | | 1335 | tmpva = (KERNBASE + NKL2_KIMG_ENTRIES * NBPD_L2); |
1336 | pte = PTE_BASE + pl1_i(tmpva); | | 1336 | pte = PTE_BASE + pl1_i(tmpva); |
1337 | | | 1337 | |
1338 | /* | | 1338 | /* |
1339 | * Map the direct map. Use 1GB pages if they are available, | | 1339 | * Map the direct map. Use 1GB pages if they are available, |
1340 | * otherwise use 2MB pages. | | 1340 | * otherwise use 2MB pages. |
1341 | */ | | 1341 | */ |
1342 | | | 1342 | |
1343 | mc = &mem_clusters[mem_cluster_cnt - 1]; | | 1343 | mc = &mem_clusters[mem_cluster_cnt - 1]; |
1344 | ndmpdp = (mc->start + mc->size + NBPD_L3 - 1) >> L3_SHIFT; | | 1344 | ndmpdp = (mc->start + mc->size + NBPD_L3 - 1) >> L3_SHIFT; |
1345 | dmpdp = avail_start; avail_start += PAGE_SIZE; | | 1345 | dmpdp = avail_start; avail_start += PAGE_SIZE; |
1346 | | | 1346 | |
1347 | if (cpu_feature[2] & CPUID_P1GB) { | | 1347 | if (cpu_feature[2] & CPUID_P1GB) { |
1348 | for (i = 0; i < ndmpdp; i++) { | | 1348 | for (i = 0; i < ndmpdp; i++) { |
1349 | pdp = (paddr_t)&(((pd_entry_t *)dmpdp)[i]); | | 1349 | pdp = (paddr_t)&(((pd_entry_t *)dmpdp)[i]); |
1350 | *pte = (pdp & PG_FRAME) | PG_V | PG_RW; | | 1350 | *pte = (pdp & PG_FRAME) | PG_V | PG_RW; |
1351 | pmap_update_pg(tmpva); | | 1351 | pmap_update_pg(tmpva); |
1352 | | | 1352 | |
1353 | pde = (pd_entry_t *)(tmpva + (pdp & ~PG_FRAME)); | | 1353 | pde = (pd_entry_t *)(tmpva + (pdp & ~PG_FRAME)); |
1354 | *pde = ((paddr_t)i << L3_SHIFT) | | | 1354 | *pde = ((paddr_t)i << L3_SHIFT) | |
1355 | PG_RW | PG_V | PG_U | PG_PS | PG_G; | | 1355 | PG_RW | PG_V | PG_U | PG_PS | PG_G; |
1356 | } | | 1356 | } |
1357 | } else { | | 1357 | } else { |
1358 | dmpd = avail_start; avail_start += ndmpdp * PAGE_SIZE; | | 1358 | dmpd = avail_start; avail_start += ndmpdp * PAGE_SIZE; |
1359 | | | 1359 | |
1360 | for (i = 0; i < NPDPG * ndmpdp; i++) { | | 1360 | for (i = 0; i < NPDPG * ndmpdp; i++) { |
1361 | pdp = (paddr_t)&(((pd_entry_t *)dmpd)[i]); | | 1361 | pdp = (paddr_t)&(((pd_entry_t *)dmpd)[i]); |
1362 | *pte = (pdp & PG_FRAME) | PG_V | PG_RW; | | 1362 | *pte = (pdp & PG_FRAME) | PG_V | PG_RW; |
1363 | pmap_update_pg(tmpva); | | 1363 | pmap_update_pg(tmpva); |
1364 | | | 1364 | |
1365 | pde = (pd_entry_t *)(tmpva + (pdp & ~PG_FRAME)); | | 1365 | pde = (pd_entry_t *)(tmpva + (pdp & ~PG_FRAME)); |
1366 | *pde = ((paddr_t)i << L2_SHIFT) | | | 1366 | *pde = ((paddr_t)i << L2_SHIFT) | |
1367 | PG_RW | PG_V | PG_U | PG_PS | PG_G; | | 1367 | PG_RW | PG_V | PG_U | PG_PS | PG_G; |
1368 | } | | 1368 | } |
1369 | for (i = 0; i < ndmpdp; i++) { | | 1369 | for (i = 0; i < ndmpdp; i++) { |
1370 | pdp = (paddr_t)&(((pd_entry_t *)dmpdp)[i]); | | 1370 | pdp = (paddr_t)&(((pd_entry_t *)dmpdp)[i]); |
1371 | *pte = (pdp & PG_FRAME) | PG_V | PG_RW; | | 1371 | *pte = (pdp & PG_FRAME) | PG_V | PG_RW; |
1372 | pmap_update_pg((vaddr_t)tmpva); | | 1372 | pmap_update_pg((vaddr_t)tmpva); |
1373 | | | 1373 | |
1374 | pde = (pd_entry_t *)(tmpva + (pdp & ~PG_FRAME)); | | 1374 | pde = (pd_entry_t *)(tmpva + (pdp & ~PG_FRAME)); |
1375 | *pde = (dmpd + (i << PAGE_SHIFT)) | | | 1375 | *pde = (dmpd + (i << PAGE_SHIFT)) | |
1376 | PG_RW | PG_V | PG_U; | | 1376 | PG_RW | PG_V | PG_U; |
1377 | } | | 1377 | } |
1378 | } | | 1378 | } |
1379 | | | 1379 | |
1380 | kpm->pm_pdir[PDIR_SLOT_DIRECT] = dmpdp | PG_KW | PG_V | PG_U; | | 1380 | kpm->pm_pdir[PDIR_SLOT_DIRECT] = dmpdp | PG_KW | PG_V | PG_U; |
1381 | | | 1381 | |
1382 | tlbflush(); | | 1382 | tlbflush(); |
1383 | | | 1383 | |
1384 | #else | | 1384 | #else |
1385 | if (VM_MIN_KERNEL_ADDRESS != KERNBASE) { | | 1385 | if (VM_MIN_KERNEL_ADDRESS != KERNBASE) { |
1386 | /* | | 1386 | /* |
1387 | * zero_pte is stuck at the end of mapped space for the kernel | | 1387 | * zero_pte is stuck at the end of mapped space for the kernel |
1388 | * image (disjunct from kva space). This is done so that it | | 1388 | * image (disjunct from kva space). This is done so that it |
1389 | * can safely be used in pmap_growkernel (pmap_get_physpage), | | 1389 | * can safely be used in pmap_growkernel (pmap_get_physpage), |
1390 | * when it's called for the first time. | | 1390 | * when it's called for the first time. |
1391 | * XXXfvdl fix this for MULTIPROCESSOR later. | | 1391 | * XXXfvdl fix this for MULTIPROCESSOR later. |
1392 | */ | | 1392 | */ |
1393 | | | 1393 | |
1394 | early_zerop = (void *)(KERNBASE + NKL2_KIMG_ENTRIES * NBPD_L2); | | 1394 | early_zerop = (void *)(KERNBASE + NKL2_KIMG_ENTRIES * NBPD_L2); |
1395 | early_zero_pte = PTE_BASE + pl1_i((vaddr_t)early_zerop); | | 1395 | early_zero_pte = PTE_BASE + pl1_i((vaddr_t)early_zerop); |
1396 | } | | 1396 | } |
1397 | | | 1397 | |
1398 | /* | | 1398 | /* |
1399 | * now we allocate the "special" VAs which are used for tmp mappings | | 1399 | * now we allocate the "special" VAs which are used for tmp mappings |
1400 | * by the pmap (and other modules). we allocate the VAs by advancing | | 1400 | * by the pmap (and other modules). we allocate the VAs by advancing |
1401 | * virtual_avail (note that there are no pages mapped at these VAs). | | 1401 | * virtual_avail (note that there are no pages mapped at these VAs). |
1402 | * we find the PTE that maps the allocated VA via the linear PTE | | 1402 | * we find the PTE that maps the allocated VA via the linear PTE |
1403 | * mapping. | | 1403 | * mapping. |
1404 | */ | | 1404 | */ |
1405 | | | 1405 | |
1406 | pte = PTE_BASE + pl1_i(virtual_avail); | | 1406 | pte = PTE_BASE + pl1_i(virtual_avail); |
1407 | | | 1407 | |
1408 | #ifdef MULTIPROCESSOR | | 1408 | #ifdef MULTIPROCESSOR |
1409 | /* | | 1409 | /* |
1410 | * Waste some VA space to avoid false sharing of cache lines | | 1410 | * Waste some VA space to avoid false sharing of cache lines |
1411 | * for page table pages: Give each possible CPU a cache line | | 1411 | * for page table pages: Give each possible CPU a cache line |
1412 | * of PTE's (8) to play with, though we only need 4. We could | | 1412 | * of PTE's (8) to play with, though we only need 4. We could |
1413 | * recycle some of this waste by putting the idle stacks here | | 1413 | * recycle some of this waste by putting the idle stacks here |
1414 | * as well; we could waste less space if we knew the largest | | 1414 | * as well; we could waste less space if we knew the largest |
1415 | * CPU ID beforehand. | | 1415 | * CPU ID beforehand. |
1416 | */ | | 1416 | */ |
1417 | csrcp = (char *) virtual_avail; csrc_pte = pte; | | 1417 | csrcp = (char *) virtual_avail; csrc_pte = pte; |
1418 | | | 1418 | |
1419 | cdstp = (char *) virtual_avail+PAGE_SIZE; cdst_pte = pte+1; | | 1419 | cdstp = (char *) virtual_avail+PAGE_SIZE; cdst_pte = pte+1; |
1420 | | | 1420 | |
1421 | zerop = (char *) virtual_avail+PAGE_SIZE*2; zero_pte = pte+2; | | 1421 | zerop = (char *) virtual_avail+PAGE_SIZE*2; zero_pte = pte+2; |
1422 | | | 1422 | |
1423 | ptpp = (char *) virtual_avail+PAGE_SIZE*3; ptp_pte = pte+3; | | 1423 | ptpp = (char *) virtual_avail+PAGE_SIZE*3; ptp_pte = pte+3; |
1424 | | | 1424 | |
1425 | virtual_avail += PAGE_SIZE * maxcpus * NPTECL; | | 1425 | virtual_avail += PAGE_SIZE * maxcpus * NPTECL; |
1426 | pte += maxcpus * NPTECL; | | 1426 | pte += maxcpus * NPTECL; |
1427 | #else | | 1427 | #else |
1428 | csrcp = (void *) virtual_avail; csrc_pte = pte; /* allocate */ | | 1428 | csrcp = (void *) virtual_avail; csrc_pte = pte; /* allocate */ |
1429 | virtual_avail += PAGE_SIZE; pte++; /* advance */ | | 1429 | virtual_avail += PAGE_SIZE; pte++; /* advance */ |
1430 | | | 1430 | |
1431 | cdstp = (void *) virtual_avail; cdst_pte = pte; | | 1431 | cdstp = (void *) virtual_avail; cdst_pte = pte; |
1432 | virtual_avail += PAGE_SIZE; pte++; | | 1432 | virtual_avail += PAGE_SIZE; pte++; |
1433 | | | 1433 | |
1434 | zerop = (void *) virtual_avail; zero_pte = pte; | | 1434 | zerop = (void *) virtual_avail; zero_pte = pte; |
1435 | virtual_avail += PAGE_SIZE; pte++; | | 1435 | virtual_avail += PAGE_SIZE; pte++; |
1436 | | | 1436 | |
1437 | ptpp = (void *) virtual_avail; ptp_pte = pte; | | 1437 | ptpp = (void *) virtual_avail; ptp_pte = pte; |
1438 | virtual_avail += PAGE_SIZE; pte++; | | 1438 | virtual_avail += PAGE_SIZE; pte++; |
1439 | #endif | | 1439 | #endif |
1440 | | | 1440 | |
1441 | if (VM_MIN_KERNEL_ADDRESS == KERNBASE) { | | 1441 | if (VM_MIN_KERNEL_ADDRESS == KERNBASE) { |
1442 | early_zerop = zerop; | | 1442 | early_zerop = zerop; |
1443 | early_zero_pte = zero_pte; | | 1443 | early_zero_pte = zero_pte; |
1444 | } | | 1444 | } |
1445 | #endif | | 1445 | #endif |
1446 | | | 1446 | |
1447 | /* | | 1447 | /* |
1448 | * Nothing after this point actually needs pte. | | 1448 | * Nothing after this point actually needs pte. |
1449 | */ | | 1449 | */ |
1450 | pte = (void *)0xdeadbeef; | | 1450 | pte = (void *)0xdeadbeef; |
1451 | | | 1451 | |
1452 | #ifdef XEN | | 1452 | #ifdef XEN |
1453 | #ifdef __x86_64__ | | 1453 | #ifdef __x86_64__ |
1454 | /* | | 1454 | /* |
1455 | * We want a dummy page directory for Xen: | | 1455 | * We want a dummy page directory for Xen: |
1456 | * when deactivate a pmap, Xen will still consider it active. | | 1456 | * when deactivate a pmap, Xen will still consider it active. |
1457 | * So we set user PGD to this one to lift all protection on | | 1457 | * So we set user PGD to this one to lift all protection on |
1458 | * the now inactive page tables set. | | 1458 | * the now inactive page tables set. |
1459 | */ | | 1459 | */ |
1460 | xen_dummy_user_pgd = avail_start; | | 1460 | xen_dummy_user_pgd = avail_start; |
1461 | avail_start += PAGE_SIZE; | | 1461 | avail_start += PAGE_SIZE; |
1462 | | | 1462 | |
1463 | /* Zero fill it, the less checks in Xen it requires the better */ | | 1463 | /* Zero fill it, the less checks in Xen it requires the better */ |
1464 | memset((void *) (xen_dummy_user_pgd + KERNBASE), 0, PAGE_SIZE); | | 1464 | memset((void *) (xen_dummy_user_pgd + KERNBASE), 0, PAGE_SIZE); |
1465 | /* Mark read-only */ | | 1465 | /* Mark read-only */ |
1466 | HYPERVISOR_update_va_mapping(xen_dummy_user_pgd + KERNBASE, | | 1466 | HYPERVISOR_update_va_mapping(xen_dummy_user_pgd + KERNBASE, |
1467 | pmap_pa2pte(xen_dummy_user_pgd) | PG_u | PG_V, UVMF_INVLPG); | | 1467 | pmap_pa2pte(xen_dummy_user_pgd) | PG_u | PG_V, UVMF_INVLPG); |
1468 | /* Pin as L4 */ | | 1468 | /* Pin as L4 */ |
1469 | xpq_queue_pin_l4_table(xpmap_ptom_masked(xen_dummy_user_pgd)); | | 1469 | xpq_queue_pin_l4_table(xpmap_ptom_masked(xen_dummy_user_pgd)); |
1470 | #endif /* __x86_64__ */ | | 1470 | #endif /* __x86_64__ */ |
1471 | idt_vaddr = virtual_avail; /* don't need pte */ | | 1471 | idt_vaddr = virtual_avail; /* don't need pte */ |
1472 | idt_paddr = avail_start; /* steal a page */ | | 1472 | idt_paddr = avail_start; /* steal a page */ |
1473 | /* | | 1473 | /* |
1474 | * Xen require one more page as we can't store | | 1474 | * Xen require one more page as we can't store |
1475 | * GDT and LDT on the same page | | 1475 | * GDT and LDT on the same page |
1476 | */ | | 1476 | */ |
1477 | virtual_avail += 3 * PAGE_SIZE; | | 1477 | virtual_avail += 3 * PAGE_SIZE; |
1478 | avail_start += 3 * PAGE_SIZE; | | 1478 | avail_start += 3 * PAGE_SIZE; |
1479 | #else /* XEN */ | | 1479 | #else /* XEN */ |
1480 | idt_vaddr = virtual_avail; /* don't need pte */ | | 1480 | idt_vaddr = virtual_avail; /* don't need pte */ |
1481 | idt_paddr = avail_start; /* steal a page */ | | 1481 | idt_paddr = avail_start; /* steal a page */ |
1482 | #if defined(__x86_64__) | | 1482 | #if defined(__x86_64__) |
1483 | virtual_avail += 2 * PAGE_SIZE; | | 1483 | virtual_avail += 2 * PAGE_SIZE; |
1484 | avail_start += 2 * PAGE_SIZE; | | 1484 | avail_start += 2 * PAGE_SIZE; |
1485 | #else /* defined(__x86_64__) */ | | 1485 | #else /* defined(__x86_64__) */ |
1486 | virtual_avail += PAGE_SIZE; | | 1486 | virtual_avail += PAGE_SIZE; |
1487 | avail_start += PAGE_SIZE; | | 1487 | avail_start += PAGE_SIZE; |
1488 | /* pentium f00f bug stuff */ | | 1488 | /* pentium f00f bug stuff */ |
1489 | pentium_idt_vaddr = virtual_avail; /* don't need pte */ | | 1489 | pentium_idt_vaddr = virtual_avail; /* don't need pte */ |
1490 | virtual_avail += PAGE_SIZE; | | 1490 | virtual_avail += PAGE_SIZE; |
1491 | #endif /* defined(__x86_64__) */ | | 1491 | #endif /* defined(__x86_64__) */ |
1492 | #endif /* XEN */ | | 1492 | #endif /* XEN */ |
1493 | | | 1493 | |
1494 | #ifdef _LP64 | | 1494 | #ifdef _LP64 |
1495 | /* | | 1495 | /* |
1496 | * Grab a page below 4G for things that need it (i.e. | | 1496 | * Grab a page below 4G for things that need it (i.e. |
1497 | * having an initial %cr3 for the MP trampoline). | | 1497 | * having an initial %cr3 for the MP trampoline). |
1498 | */ | | 1498 | */ |
1499 | lo32_vaddr = virtual_avail; | | 1499 | lo32_vaddr = virtual_avail; |
1500 | virtual_avail += PAGE_SIZE; | | 1500 | virtual_avail += PAGE_SIZE; |
1501 | lo32_paddr = avail_start; | | 1501 | lo32_paddr = avail_start; |
1502 | avail_start += PAGE_SIZE; | | 1502 | avail_start += PAGE_SIZE; |
1503 | #endif | | 1503 | #endif |
1504 | | | 1504 | |
1505 | /* | | 1505 | /* |
1506 | * now we reserve some VM for mapping pages when doing a crash dump | | 1506 | * now we reserve some VM for mapping pages when doing a crash dump |
1507 | */ | | 1507 | */ |
1508 | | | 1508 | |
1509 | virtual_avail = reserve_dumppages(virtual_avail); | | 1509 | virtual_avail = reserve_dumppages(virtual_avail); |
1510 | | | 1510 | |
1511 | /* | | 1511 | /* |
1512 | * init the static-global locks and global lists. | | 1512 | * init the static-global locks and global lists. |
1513 | * | | 1513 | * |
1514 | * => pventry::pvh_lock (initialized elsewhere) must also be | | 1514 | * => pventry::pvh_lock (initialized elsewhere) must also be |
1515 | * a spin lock, again at IPL_VM to prevent deadlock, and | | 1515 | * a spin lock, again at IPL_VM to prevent deadlock, and |
1516 | * again is never taken from interrupt context. | | 1516 | * again is never taken from interrupt context. |
1517 | */ | | 1517 | */ |
1518 | | | 1518 | |
1519 | mutex_init(&pmaps_lock, MUTEX_DEFAULT, IPL_NONE); | | 1519 | mutex_init(&pmaps_lock, MUTEX_DEFAULT, IPL_NONE); |
1520 | LIST_INIT(&pmaps); | | 1520 | LIST_INIT(&pmaps); |
1521 | | | 1521 | |
1522 | /* | | 1522 | /* |
1523 | * ensure the TLB is sync'd with reality by flushing it... | | 1523 | * ensure the TLB is sync'd with reality by flushing it... |
1524 | */ | | 1524 | */ |
1525 | | | 1525 | |
1526 | tlbflushg(); | | 1526 | tlbflushg(); |
1527 | | | 1527 | |
1528 | /* | | 1528 | /* |
1529 | * calculate pmap_maxkvaddr from nkptp[]. | | 1529 | * calculate pmap_maxkvaddr from nkptp[]. |
1530 | */ | | 1530 | */ |
1531 | | | 1531 | |
1532 | kva = VM_MIN_KERNEL_ADDRESS; | | 1532 | kva = VM_MIN_KERNEL_ADDRESS; |
1533 | for (i = PTP_LEVELS - 1; i >= 1; i--) { | | 1533 | for (i = PTP_LEVELS - 1; i >= 1; i--) { |
1534 | kva += nkptp[i] * nbpd[i]; | | 1534 | kva += nkptp[i] * nbpd[i]; |
1535 | } | | 1535 | } |
1536 | pmap_maxkvaddr = kva; | | 1536 | pmap_maxkvaddr = kva; |
1537 | } | | 1537 | } |
1538 | | | 1538 | |
1539 | #if defined(__x86_64__) | | 1539 | #if defined(__x86_64__) |
1540 | /* | | 1540 | /* |
1541 | * Pre-allocate PTPs for low memory, so that 1:1 mappings for various | | 1541 | * Pre-allocate PTPs for low memory, so that 1:1 mappings for various |
1542 | * trampoline code can be entered. | | 1542 | * trampoline code can be entered. |
1543 | */ | | 1543 | */ |
1544 | void | | 1544 | void |
1545 | pmap_prealloc_lowmem_ptps(void) | | 1545 | pmap_prealloc_lowmem_ptps(void) |
1546 | { | | 1546 | { |
1547 | int level; | | 1547 | int level; |
1548 | paddr_t newp; | | 1548 | paddr_t newp; |
1549 | #ifdef XEN | | 1549 | #ifdef XEN |
1550 | paddr_t pdes_pa; | | 1550 | paddr_t pdes_pa; |
1551 | | | 1551 | |
1552 | pdes_pa = pmap_pdirpa(pmap_kernel(), 0); | | 1552 | pdes_pa = pmap_pdirpa(pmap_kernel(), 0); |
1553 | level = PTP_LEVELS; | | 1553 | level = PTP_LEVELS; |
1554 | for (;;) { | | 1554 | for (;;) { |
1555 | newp = avail_start; | | 1555 | newp = avail_start; |
1556 | avail_start += PAGE_SIZE; | | 1556 | avail_start += PAGE_SIZE; |
1557 | HYPERVISOR_update_va_mapping ((vaddr_t)early_zerop, | | 1557 | HYPERVISOR_update_va_mapping ((vaddr_t)early_zerop, |
1558 | xpmap_ptom_masked(newp) | PG_u | PG_V | PG_RW, UVMF_INVLPG); | | 1558 | xpmap_ptom_masked(newp) | PG_u | PG_V | PG_RW, UVMF_INVLPG); |
1559 | memset(early_zerop, 0, PAGE_SIZE); | | 1559 | memset(early_zerop, 0, PAGE_SIZE); |
1560 | /* Mark R/O before installing */ | | 1560 | /* Mark R/O before installing */ |
1561 | HYPERVISOR_update_va_mapping ((vaddr_t)early_zerop, | | 1561 | HYPERVISOR_update_va_mapping ((vaddr_t)early_zerop, |
1562 | xpmap_ptom_masked(newp) | PG_u | PG_V, UVMF_INVLPG); | | 1562 | xpmap_ptom_masked(newp) | PG_u | PG_V, UVMF_INVLPG); |
1563 | if (newp < (NKL2_KIMG_ENTRIES * NBPD_L2)) | | 1563 | if (newp < (NKL2_KIMG_ENTRIES * NBPD_L2)) |
1564 | HYPERVISOR_update_va_mapping (newp + KERNBASE, | | 1564 | HYPERVISOR_update_va_mapping (newp + KERNBASE, |
1565 | xpmap_ptom_masked(newp) | PG_u | PG_V, UVMF_INVLPG); | | 1565 | xpmap_ptom_masked(newp) | PG_u | PG_V, UVMF_INVLPG); |
1566 | /* Update the pmap_kernel() L4 shadow */ | | 1566 | /* Update the pmap_kernel() L4 shadow */ |
1567 | xpq_queue_pte_update ( | | 1567 | xpq_queue_pte_update ( |
1568 | xpmap_ptom_masked(pdes_pa) | | 1568 | xpmap_ptom_masked(pdes_pa) |
1569 | + (pl_i(0, level) * sizeof (pd_entry_t)), | | 1569 | + (pl_i(0, level) * sizeof (pd_entry_t)), |
1570 | xpmap_ptom_masked(newp) | PG_RW | PG_u | PG_V); | | 1570 | xpmap_ptom_masked(newp) | PG_RW | PG_u | PG_V); |
1571 | /* sync to per-cpu PD */ | | 1571 | /* sync to per-cpu PD */ |
1572 | xpq_queue_pte_update( | | 1572 | xpq_queue_pte_update( |
1573 | xpmap_ptom_masked(cpu_info_primary.ci_kpm_pdirpa + | | 1573 | xpmap_ptom_masked(cpu_info_primary.ci_kpm_pdirpa + |
1574 | pl_i(0, PTP_LEVELS) * | | 1574 | pl_i(0, PTP_LEVELS) * |
1575 | sizeof(pd_entry_t)), | | 1575 | sizeof(pd_entry_t)), |
1576 | pmap_kernel()->pm_pdir[pl_i(0, PTP_LEVELS)]); | | 1576 | pmap_kernel()->pm_pdir[pl_i(0, PTP_LEVELS)]); |
1577 | pmap_pte_flush(); | | 1577 | pmap_pte_flush(); |
1578 | level--; | | 1578 | level--; |
1579 | if (level <= 1) | | 1579 | if (level <= 1) |
1580 | break; | | 1580 | break; |
1581 | pdes_pa = newp; | | 1581 | pdes_pa = newp; |
1582 | } | | 1582 | } |
1583 | #else /* XEN */ | | 1583 | #else /* XEN */ |
1584 | pd_entry_t *pdes; | | 1584 | pd_entry_t *pdes; |
1585 | | | 1585 | |
1586 | pdes = pmap_kernel()->pm_pdir; | | 1586 | pdes = pmap_kernel()->pm_pdir; |
1587 | level = PTP_LEVELS; | | 1587 | level = PTP_LEVELS; |
1588 | for (;;) { | | 1588 | for (;;) { |
1589 | newp = avail_start; | | 1589 | newp = avail_start; |
1590 | avail_start += PAGE_SIZE; | | 1590 | avail_start += PAGE_SIZE; |
1591 | #ifdef __HAVE_DIRECT_MAP | | 1591 | #ifdef __HAVE_DIRECT_MAP |
1592 | memset((void *)PMAP_DIRECT_MAP(newp), 0, PAGE_SIZE); | | 1592 | memset((void *)PMAP_DIRECT_MAP(newp), 0, PAGE_SIZE); |
1593 | #else | | 1593 | #else |
1594 | pmap_pte_set(early_zero_pte, (newp & PG_FRAME) | PG_V | PG_RW); | | 1594 | pmap_pte_set(early_zero_pte, (newp & PG_FRAME) | PG_V | PG_RW); |
1595 | pmap_pte_flush(); | | 1595 | pmap_pte_flush(); |
1596 | pmap_update_pg((vaddr_t)early_zerop); | | 1596 | pmap_update_pg((vaddr_t)early_zerop); |
1597 | memset(early_zerop, 0, PAGE_SIZE); | | 1597 | memset(early_zerop, 0, PAGE_SIZE); |
1598 | #endif | | 1598 | #endif |
1599 | pdes[pl_i(0, level)] = (newp & PG_FRAME) | PG_V | PG_RW; | | 1599 | pdes[pl_i(0, level)] = (newp & PG_FRAME) | PG_V | PG_RW; |
1600 | level--; | | 1600 | level--; |
1601 | if (level <= 1) | | 1601 | if (level <= 1) |
1602 | break; | | 1602 | break; |
1603 | pdes = normal_pdes[level - 2]; | | 1603 | pdes = normal_pdes[level - 2]; |
1604 | } | | 1604 | } |
1605 | #endif /* XEN */ | | 1605 | #endif /* XEN */ |
1606 | } | | 1606 | } |
1607 | #endif /* defined(__x86_64__) */ | | 1607 | #endif /* defined(__x86_64__) */ |
1608 | | | 1608 | |
1609 | /* | | 1609 | /* |
1610 | * pmap_init: called from uvm_init, our job is to get the pmap | | 1610 | * pmap_init: called from uvm_init, our job is to get the pmap |
1611 | * system ready to manage mappings... | | 1611 | * system ready to manage mappings... |
1612 | */ | | 1612 | */ |
1613 | | | 1613 | |
1614 | void | | 1614 | void |
1615 | pmap_init(void) | | 1615 | pmap_init(void) |
1616 | { | | 1616 | { |
1617 | int i; | | 1617 | int i; |
1618 | | | 1618 | |
1619 | for (i = 0; i < PV_HASH_SIZE; i++) { | | 1619 | for (i = 0; i < PV_HASH_SIZE; i++) { |
1620 | SLIST_INIT(&pv_hash_heads[i].hh_list); | | 1620 | SLIST_INIT(&pv_hash_heads[i].hh_list); |
1621 | } | | 1621 | } |
1622 | for (i = 0; i < PV_HASH_LOCK_CNT; i++) { | | 1622 | for (i = 0; i < PV_HASH_LOCK_CNT; i++) { |
1623 | mutex_init(&pv_hash_locks[i].lock, MUTEX_NODEBUG, IPL_VM); | | 1623 | mutex_init(&pv_hash_locks[i].lock, MUTEX_NODEBUG, IPL_VM); |
1624 | } | | 1624 | } |
1625 | | | 1625 | |
1626 | /* | | 1626 | /* |
1627 | * initialize caches. | | 1627 | * initialize caches. |
1628 | */ | | 1628 | */ |
1629 | | | 1629 | |
1630 | pool_cache_bootstrap(&pmap_cache, sizeof(struct pmap), 0, 0, 0, | | 1630 | pool_cache_bootstrap(&pmap_cache, sizeof(struct pmap), 0, 0, 0, |
1631 | "pmappl", NULL, IPL_NONE, NULL, NULL, NULL); | | 1631 | "pmappl", NULL, IPL_NONE, NULL, NULL, NULL); |
1632 | #ifdef PAE | | 1632 | #ifdef PAE |
1633 | pool_cache_bootstrap(&pmap_pdp_cache, PAGE_SIZE * PDP_SIZE, 0, 0, 0, | | 1633 | pool_cache_bootstrap(&pmap_pdp_cache, PAGE_SIZE * PDP_SIZE, 0, 0, 0, |
1634 | "pdppl", &pmap_pdp_allocator, IPL_NONE, | | 1634 | "pdppl", &pmap_pdp_allocator, IPL_NONE, |
1635 | pmap_pdp_ctor, pmap_pdp_dtor, NULL); | | 1635 | pmap_pdp_ctor, pmap_pdp_dtor, NULL); |
1636 | #else /* PAE */ | | 1636 | #else /* PAE */ |
1637 | pool_cache_bootstrap(&pmap_pdp_cache, PAGE_SIZE, 0, 0, 0, | | 1637 | pool_cache_bootstrap(&pmap_pdp_cache, PAGE_SIZE, 0, 0, 0, |
1638 | "pdppl", NULL, IPL_NONE, pmap_pdp_ctor, pmap_pdp_dtor, NULL); | | 1638 | "pdppl", NULL, IPL_NONE, pmap_pdp_ctor, pmap_pdp_dtor, NULL); |
1639 | #endif /* PAE */ | | 1639 | #endif /* PAE */ |
1640 | pool_cache_bootstrap(&pmap_pv_cache, sizeof(struct pv_entry), 0, 0, | | 1640 | pool_cache_bootstrap(&pmap_pv_cache, sizeof(struct pv_entry), 0, 0, |
1641 | PR_LARGECACHE, "pvpl", &pool_allocator_kmem, IPL_NONE, NULL, | | 1641 | PR_LARGECACHE, "pvpl", &pool_allocator_kmem, IPL_NONE, NULL, |
1642 | NULL, NULL); | | 1642 | NULL, NULL); |
1643 | | | 1643 | |
1644 | pmap_tlb_init(); | | 1644 | pmap_tlb_init(); |
1645 | | | 1645 | |
1646 | evcnt_attach_dynamic(&pmap_iobmp_evcnt, EVCNT_TYPE_MISC, | | 1646 | evcnt_attach_dynamic(&pmap_iobmp_evcnt, EVCNT_TYPE_MISC, |
1647 | NULL, "x86", "io bitmap copy"); | | 1647 | NULL, "x86", "io bitmap copy"); |
1648 | evcnt_attach_dynamic(&pmap_ldt_evcnt, EVCNT_TYPE_MISC, | | 1648 | evcnt_attach_dynamic(&pmap_ldt_evcnt, EVCNT_TYPE_MISC, |
1649 | NULL, "x86", "ldt sync"); | | 1649 | NULL, "x86", "ldt sync"); |
1650 | | | 1650 | |
1651 | /* | | 1651 | /* |
1652 | * done: pmap module is up (and ready for business) | | 1652 | * done: pmap module is up (and ready for business) |
1653 | */ | | 1653 | */ |
1654 | | | 1654 | |
1655 | pmap_initialized = true; | | 1655 | pmap_initialized = true; |
1656 | } | | 1656 | } |
1657 | | | 1657 | |
1658 | /* | | 1658 | /* |
1659 | * pmap_cpu_init_late: perform late per-CPU initialization. | | 1659 | * pmap_cpu_init_late: perform late per-CPU initialization. |
1660 | */ | | 1660 | */ |
1661 | | | 1661 | |
1662 | #ifndef XEN | | 1662 | #ifndef XEN |
1663 | void | | 1663 | void |
1664 | pmap_cpu_init_late(struct cpu_info *ci) | | 1664 | pmap_cpu_init_late(struct cpu_info *ci) |
1665 | { | | 1665 | { |
1666 | /* | | 1666 | /* |
1667 | * The BP has already its own PD page allocated during early | | 1667 | * The BP has already its own PD page allocated during early |
1668 | * MD startup. | | 1668 | * MD startup. |
1669 | */ | | 1669 | */ |
1670 | if (ci == &cpu_info_primary) | | 1670 | if (ci == &cpu_info_primary) |
1671 | return; | | 1671 | return; |
1672 | | | 1672 | |
1673 | #ifdef PAE | | 1673 | #ifdef PAE |
1674 | cpu_alloc_l3_page(ci); | | 1674 | cpu_alloc_l3_page(ci); |
1675 | #endif | | 1675 | #endif |
1676 | } | | 1676 | } |
1677 | #endif | | 1677 | #endif |
1678 | | | 1678 | |
1679 | /* | | 1679 | /* |
1680 | * p v _ e n t r y f u n c t i o n s | | 1680 | * p v _ e n t r y f u n c t i o n s |
1681 | */ | | 1681 | */ |
1682 | | | 1682 | |
1683 | /* | | 1683 | /* |
1684 | * pmap_free_pvs: free a list of pv_entrys | | 1684 | * pmap_free_pvs: free a list of pv_entrys |
1685 | */ | | 1685 | */ |
1686 | | | 1686 | |
1687 | static void | | 1687 | static void |
1688 | pmap_free_pvs(struct pv_entry *pve) | | 1688 | pmap_free_pvs(struct pv_entry *pve) |
1689 | { | | 1689 | { |
1690 | struct pv_entry *next; | | 1690 | struct pv_entry *next; |
1691 | | | 1691 | |
1692 | for ( /* null */ ; pve != NULL ; pve = next) { | | 1692 | for ( /* null */ ; pve != NULL ; pve = next) { |
1693 | next = pve->pve_next; | | 1693 | next = pve->pve_next; |
1694 | pool_cache_put(&pmap_pv_cache, pve); | | 1694 | pool_cache_put(&pmap_pv_cache, pve); |
1695 | } | | 1695 | } |
1696 | } | | 1696 | } |
1697 | | | 1697 | |
1698 | /* | | 1698 | /* |
1699 | * main pv_entry manipulation functions: | | 1699 | * main pv_entry manipulation functions: |
1700 | * pmap_enter_pv: enter a mapping onto a pv_head list | | 1700 | * pmap_enter_pv: enter a mapping onto a pv_head list |
1701 | * pmap_remove_pv: remove a mapping from a pv_head list | | 1701 | * pmap_remove_pv: remove a mapping from a pv_head list |
1702 | * | | 1702 | * |
1703 | * NOTE: Both pmap_enter_pv and pmap_remove_pv expect the caller to lock | | 1703 | * NOTE: Both pmap_enter_pv and pmap_remove_pv expect the caller to lock |
1704 | * the pvh before calling | | 1704 | * the pvh before calling |
1705 | */ | | 1705 | */ |
1706 | | | 1706 | |
1707 | /* | | 1707 | /* |
1708 | * insert_pv: a helper of pmap_enter_pv | | 1708 | * insert_pv: a helper of pmap_enter_pv |
1709 | */ | | 1709 | */ |
1710 | | | 1710 | |
1711 | static void | | 1711 | static void |
1712 | insert_pv(struct pmap_page *pp, struct pv_entry *pve) | | 1712 | insert_pv(struct pmap_page *pp, struct pv_entry *pve) |
1713 | { | | 1713 | { |
1714 | struct pv_hash_head *hh; | | 1714 | struct pv_hash_head *hh; |
1715 | kmutex_t *lock; | | 1715 | kmutex_t *lock; |
1716 | u_int hash; | | 1716 | u_int hash; |
1717 | | | 1717 | |
1718 | hash = pvhash_hash(pve->pve_pte.pte_ptp, pve->pve_pte.pte_va); | | 1718 | hash = pvhash_hash(pve->pve_pte.pte_ptp, pve->pve_pte.pte_va); |
1719 | lock = pvhash_lock(hash); | | 1719 | lock = pvhash_lock(hash); |
1720 | hh = pvhash_head(hash); | | 1720 | hh = pvhash_head(hash); |
1721 | mutex_spin_enter(lock); | | 1721 | mutex_spin_enter(lock); |
1722 | SLIST_INSERT_HEAD(&hh->hh_list, pve, pve_hash); | | 1722 | SLIST_INSERT_HEAD(&hh->hh_list, pve, pve_hash); |
1723 | mutex_spin_exit(lock); | | 1723 | mutex_spin_exit(lock); |
1724 | | | 1724 | |
1725 | LIST_INSERT_HEAD(&pp->pp_head.pvh_list, pve, pve_list); | | 1725 | LIST_INSERT_HEAD(&pp->pp_head.pvh_list, pve, pve_list); |
1726 | } | | 1726 | } |
1727 | | | 1727 | |
1728 | /* | | 1728 | /* |
1729 | * pmap_enter_pv: enter a mapping onto a pv_head lst | | 1729 | * pmap_enter_pv: enter a mapping onto a pv_head lst |
1730 | * | | 1730 | * |
1731 | * => caller should adjust ptp's wire_count before calling | | 1731 | * => caller should adjust ptp's wire_count before calling |
1732 | */ | | 1732 | */ |
1733 | | | 1733 | |
1734 | static struct pv_entry * | | 1734 | static struct pv_entry * |
1735 | pmap_enter_pv(struct pmap_page *pp, | | 1735 | pmap_enter_pv(struct pmap_page *pp, |
1736 | struct pv_entry *pve, /* preallocated pve for us to use */ | | 1736 | struct pv_entry *pve, /* preallocated pve for us to use */ |
1737 | struct pv_entry **sparepve, | | 1737 | struct pv_entry **sparepve, |
1738 | struct vm_page *ptp, | | 1738 | struct vm_page *ptp, |
1739 | vaddr_t va) | | 1739 | vaddr_t va) |
1740 | { | | 1740 | { |
1741 | | | 1741 | |
1742 | KASSERT(ptp == NULL || ptp->wire_count >= 2); | | 1742 | KASSERT(ptp == NULL || ptp->wire_count >= 2); |
1743 | KASSERT(ptp == NULL || ptp->uobject != NULL); | | 1743 | KASSERT(ptp == NULL || ptp->uobject != NULL); |
1744 | KASSERT(ptp == NULL || ptp_va2o(va, 1) == ptp->offset); | | 1744 | KASSERT(ptp == NULL || ptp_va2o(va, 1) == ptp->offset); |
1745 | | | 1745 | |
1746 | if ((pp->pp_flags & PP_EMBEDDED) == 0) { | | 1746 | if ((pp->pp_flags & PP_EMBEDDED) == 0) { |
1747 | if (LIST_EMPTY(&pp->pp_head.pvh_list)) { | | 1747 | if (LIST_EMPTY(&pp->pp_head.pvh_list)) { |
1748 | pp->pp_flags |= PP_EMBEDDED; | | 1748 | pp->pp_flags |= PP_EMBEDDED; |
1749 | pp->pp_pte.pte_ptp = ptp; | | 1749 | pp->pp_pte.pte_ptp = ptp; |
1750 | pp->pp_pte.pte_va = va; | | 1750 | pp->pp_pte.pte_va = va; |
1751 | | | 1751 | |
1752 | return pve; | | 1752 | return pve; |
1753 | } | | 1753 | } |
1754 | } else { | | 1754 | } else { |
1755 | struct pv_entry *pve2; | | 1755 | struct pv_entry *pve2; |
1756 | | | 1756 | |
1757 | pve2 = *sparepve; | | 1757 | pve2 = *sparepve; |
1758 | *sparepve = NULL; | | 1758 | *sparepve = NULL; |
1759 | | | 1759 | |
1760 | pve2->pve_pte = pp->pp_pte; | | 1760 | pve2->pve_pte = pp->pp_pte; |
1761 | pp->pp_flags &= ~PP_EMBEDDED; | | 1761 | pp->pp_flags &= ~PP_EMBEDDED; |
1762 | LIST_INIT(&pp->pp_head.pvh_list); | | 1762 | LIST_INIT(&pp->pp_head.pvh_list); |
1763 | insert_pv(pp, pve2); | | 1763 | insert_pv(pp, pve2); |
1764 | } | | 1764 | } |
1765 | | | 1765 | |
1766 | pve->pve_pte.pte_ptp = ptp; | | 1766 | pve->pve_pte.pte_ptp = ptp; |
1767 | pve->pve_pte.pte_va = va; | | 1767 | pve->pve_pte.pte_va = va; |
1768 | insert_pv(pp, pve); | | 1768 | insert_pv(pp, pve); |
1769 | | | 1769 | |
1770 | return NULL; | | 1770 | return NULL; |
1771 | } | | 1771 | } |
1772 | | | 1772 | |
1773 | /* | | 1773 | /* |
1774 | * pmap_remove_pv: try to remove a mapping from a pv_list | | 1774 | * pmap_remove_pv: try to remove a mapping from a pv_list |
1775 | * | | 1775 | * |
1776 | * => caller should adjust ptp's wire_count and free PTP if needed | | 1776 | * => caller should adjust ptp's wire_count and free PTP if needed |
1777 | * => we return the removed pve | | 1777 | * => we return the removed pve |
1778 | */ | | 1778 | */ |
1779 | | | 1779 | |
1780 | static struct pv_entry * | | 1780 | static struct pv_entry * |
1781 | pmap_remove_pv(struct pmap_page *pp, struct vm_page *ptp, vaddr_t va) | | 1781 | pmap_remove_pv(struct pmap_page *pp, struct vm_page *ptp, vaddr_t va) |
1782 | { | | 1782 | { |
1783 | struct pv_hash_head *hh; | | 1783 | struct pv_hash_head *hh; |
1784 | struct pv_entry *pve; | | 1784 | struct pv_entry *pve; |
1785 | kmutex_t *lock; | | 1785 | kmutex_t *lock; |
1786 | u_int hash; | | 1786 | u_int hash; |
1787 | | | 1787 | |
1788 | KASSERT(ptp == NULL || ptp->uobject != NULL); | | 1788 | KASSERT(ptp == NULL || ptp->uobject != NULL); |
1789 | KASSERT(ptp == NULL || ptp_va2o(va, 1) == ptp->offset); | | 1789 | KASSERT(ptp == NULL || ptp_va2o(va, 1) == ptp->offset); |
1790 | | | 1790 | |
1791 | if ((pp->pp_flags & PP_EMBEDDED) != 0) { | | 1791 | if ((pp->pp_flags & PP_EMBEDDED) != 0) { |
1792 | KASSERT(pp->pp_pte.pte_ptp == ptp); | | 1792 | KASSERT(pp->pp_pte.pte_ptp == ptp); |
1793 | KASSERT(pp->pp_pte.pte_va == va); | | 1793 | KASSERT(pp->pp_pte.pte_va == va); |
1794 | | | 1794 | |
1795 | pp->pp_flags &= ~PP_EMBEDDED; | | 1795 | pp->pp_flags &= ~PP_EMBEDDED; |
1796 | LIST_INIT(&pp->pp_head.pvh_list); | | 1796 | LIST_INIT(&pp->pp_head.pvh_list); |
1797 | | | 1797 | |
1798 | return NULL; | | 1798 | return NULL; |
1799 | } | | 1799 | } |
1800 | | | 1800 | |
1801 | hash = pvhash_hash(ptp, va); | | 1801 | hash = pvhash_hash(ptp, va); |
1802 | lock = pvhash_lock(hash); | | 1802 | lock = pvhash_lock(hash); |
1803 | hh = pvhash_head(hash); | | 1803 | hh = pvhash_head(hash); |
1804 | mutex_spin_enter(lock); | | 1804 | mutex_spin_enter(lock); |
1805 | pve = pvhash_remove(hh, ptp, va); | | 1805 | pve = pvhash_remove(hh, ptp, va); |
1806 | mutex_spin_exit(lock); | | 1806 | mutex_spin_exit(lock); |
1807 | | | 1807 | |
1808 | LIST_REMOVE(pve, pve_list); | | 1808 | LIST_REMOVE(pve, pve_list); |
1809 | | | 1809 | |
1810 | return pve; | | 1810 | return pve; |
1811 | } | | 1811 | } |
1812 | | | 1812 | |
1813 | /* | | 1813 | /* |
1814 | * p t p f u n c t i o n s | | 1814 | * p t p f u n c t i o n s |
1815 | */ | | 1815 | */ |
1816 | | | 1816 | |
1817 | static inline struct vm_page * | | 1817 | static inline struct vm_page * |
1818 | pmap_find_ptp(struct pmap *pmap, vaddr_t va, paddr_t pa, int level) | | 1818 | pmap_find_ptp(struct pmap *pmap, vaddr_t va, paddr_t pa, int level) |
1819 | { | | 1819 | { |
1820 | int lidx = level - 1; | | 1820 | int lidx = level - 1; |
1821 | struct vm_page *pg; | | 1821 | struct vm_page *pg; |
1822 | | | 1822 | |
1823 | KASSERT(mutex_owned(pmap->pm_lock)); | | 1823 | KASSERT(mutex_owned(pmap->pm_lock)); |
1824 | | | 1824 | |
1825 | if (pa != (paddr_t)-1 && pmap->pm_ptphint[lidx] && | | 1825 | if (pa != (paddr_t)-1 && pmap->pm_ptphint[lidx] && |
1826 | pa == VM_PAGE_TO_PHYS(pmap->pm_ptphint[lidx])) { | | 1826 | pa == VM_PAGE_TO_PHYS(pmap->pm_ptphint[lidx])) { |
1827 | return (pmap->pm_ptphint[lidx]); | | 1827 | return (pmap->pm_ptphint[lidx]); |
1828 | } | | 1828 | } |
1829 | PMAP_SUBOBJ_LOCK(pmap, lidx); | | 1829 | PMAP_SUBOBJ_LOCK(pmap, lidx); |
1830 | pg = uvm_pagelookup(&pmap->pm_obj[lidx], ptp_va2o(va, level)); | | 1830 | pg = uvm_pagelookup(&pmap->pm_obj[lidx], ptp_va2o(va, level)); |
1831 | PMAP_SUBOBJ_UNLOCK(pmap, lidx); | | 1831 | PMAP_SUBOBJ_UNLOCK(pmap, lidx); |
1832 | | | 1832 | |
1833 | KASSERT(pg == NULL || pg->wire_count >= 1); | | 1833 | KASSERT(pg == NULL || pg->wire_count >= 1); |
1834 | return pg; | | 1834 | return pg; |
1835 | } | | 1835 | } |
1836 | | | 1836 | |
1837 | static inline void | | 1837 | static inline void |
1838 | pmap_freepage(struct pmap *pmap, struct vm_page *ptp, int level) | | 1838 | pmap_freepage(struct pmap *pmap, struct vm_page *ptp, int level) |
1839 | { | | 1839 | { |
1840 | lwp_t *l; | | 1840 | lwp_t *l; |
1841 | int lidx; | | 1841 | int lidx; |
1842 | struct uvm_object *obj; | | 1842 | struct uvm_object *obj; |
1843 | | | 1843 | |
1844 | KASSERT(ptp->wire_count == 1); | | 1844 | KASSERT(ptp->wire_count == 1); |
1845 | | | 1845 | |
1846 | lidx = level - 1; | | 1846 | lidx = level - 1; |
1847 | | | 1847 | |
1848 | obj = &pmap->pm_obj[lidx]; | | 1848 | obj = &pmap->pm_obj[lidx]; |
1849 | pmap_stats_update(pmap, -1, 0); | | 1849 | pmap_stats_update(pmap, -1, 0); |
1850 | if (lidx != 0) | | 1850 | if (lidx != 0) |
1851 | mutex_enter(obj->vmobjlock); | | 1851 | mutex_enter(obj->vmobjlock); |
1852 | if (pmap->pm_ptphint[lidx] == ptp) | | 1852 | if (pmap->pm_ptphint[lidx] == ptp) |
1853 | pmap->pm_ptphint[lidx] = TAILQ_FIRST(&obj->memq); | | 1853 | pmap->pm_ptphint[lidx] = TAILQ_FIRST(&obj->memq); |
1854 | ptp->wire_count = 0; | | 1854 | ptp->wire_count = 0; |
1855 | uvm_pagerealloc(ptp, NULL, 0); | | 1855 | uvm_pagerealloc(ptp, NULL, 0); |
1856 | l = curlwp; | | 1856 | l = curlwp; |
1857 | KASSERT((l->l_pflag & LP_INTR) == 0); | | 1857 | KASSERT((l->l_pflag & LP_INTR) == 0); |
1858 | VM_PAGE_TO_PP(ptp)->pp_link = l->l_md.md_gc_ptp; | | 1858 | VM_PAGE_TO_PP(ptp)->pp_link = l->l_md.md_gc_ptp; |
1859 | l->l_md.md_gc_ptp = ptp; | | 1859 | l->l_md.md_gc_ptp = ptp; |
1860 | if (lidx != 0) | | 1860 | if (lidx != 0) |
1861 | mutex_exit(obj->vmobjlock); | | 1861 | mutex_exit(obj->vmobjlock); |
1862 | } | | 1862 | } |
1863 | | | 1863 | |
1864 | static void | | 1864 | static void |
1865 | pmap_free_ptp(struct pmap *pmap, struct vm_page *ptp, vaddr_t va, | | 1865 | pmap_free_ptp(struct pmap *pmap, struct vm_page *ptp, vaddr_t va, |
1866 | pt_entry_t *ptes, pd_entry_t * const *pdes) | | 1866 | pt_entry_t *ptes, pd_entry_t * const *pdes) |
1867 | { | | 1867 | { |
1868 | unsigned long index; | | 1868 | unsigned long index; |
1869 | int level; | | 1869 | int level; |
1870 | vaddr_t invaladdr; | | 1870 | vaddr_t invaladdr; |
1871 | pd_entry_t opde; | | 1871 | pd_entry_t opde; |
1872 | #ifdef XEN | | | |
1873 | struct pmap *curpmap = vm_map_pmap(&curlwp->l_proc->p_vmspace->vm_map); | | | |
1874 | #ifdef MULTIPROCESSOR | | | |
1875 | vaddr_t invaladdr2; | | | |
1876 | #endif | | | |
1877 | #endif | | | |
1878 | | | 1872 | |
1879 | KASSERT(pmap != pmap_kernel()); | | 1873 | KASSERT(pmap != pmap_kernel()); |
1880 | KASSERT(mutex_owned(pmap->pm_lock)); | | 1874 | KASSERT(mutex_owned(pmap->pm_lock)); |
1881 | KASSERT(kpreempt_disabled()); | | 1875 | KASSERT(kpreempt_disabled()); |
1882 | | | 1876 | |
1883 | level = 1; | | 1877 | level = 1; |
1884 | do { | | 1878 | do { |
1885 | index = pl_i(va, level + 1); | | 1879 | index = pl_i(va, level + 1); |
1886 | opde = pmap_pte_testset(&pdes[level - 1][index], 0); | | 1880 | opde = pmap_pte_testset(&pdes[level - 1][index], 0); |
1887 | #if defined(XEN) | | 1881 | #if defined(XEN) |
1888 | # if defined(__x86_64__) | | 1882 | # if defined(__x86_64__) |
1889 | /* | | 1883 | /* |
1890 | * If ptp is a L3 currently mapped in kernel space, | | 1884 | * If ptp is a L3 currently mapped in kernel space, |
1891 | * on any cpu, clear it before freeing | | 1885 | * on any cpu, clear it before freeing |
1892 | */ | | 1886 | */ |
1893 | if (level == PTP_LEVELS - 1) { | | 1887 | if (level == PTP_LEVELS - 1) { |
1894 | /* | | 1888 | /* |
1895 | * Update the per-cpu PD on all cpus the current | | 1889 | * Update the per-cpu PD on all cpus the current |
1896 | * pmap is active on | | 1890 | * pmap is active on |
1897 | */ | | 1891 | */ |
1898 | xen_kpm_sync(pmap, index); | | 1892 | xen_kpm_sync(pmap, index); |
1899 | | | 1893 | |
1900 | } | | 1894 | } |
1901 | # endif /*__x86_64__ */ | | 1895 | # endif /*__x86_64__ */ |
1902 | invaladdr = level == 1 ? (vaddr_t)ptes : | | 1896 | invaladdr = level == 1 ? (vaddr_t)ptes : |
1903 | (vaddr_t)pdes[level - 2]; | | 1897 | (vaddr_t)pdes[level - 2]; |
1904 | pmap_tlb_shootdown(curpmap, invaladdr + index * PAGE_SIZE, | | 1898 | pmap_tlb_shootdown(pmap, invaladdr + index * PAGE_SIZE, |
1905 | opde, TLBSHOOT_FREE_PTP1); | | 1899 | opde, TLBSHOOT_FREE_PTP1); |
1906 | # if defined(MULTIPROCESSOR) | | | |
1907 | invaladdr2 = level == 1 ? (vaddr_t)PTE_BASE : | | | |
1908 | (vaddr_t)normal_pdes[level - 2]; | | | |
1909 | if (pmap != curpmap || invaladdr != invaladdr2) { | | | |
1910 | pmap_tlb_shootdown(pmap, invaladdr2 + index * PAGE_SIZE, | | | |
1911 | opde, TLBSHOOT_FREE_PTP2); | | | |
1912 | } | | | |
1913 | # endif /* MULTIPROCESSOR */ | | | |
1914 | #else /* XEN */ | | 1900 | #else /* XEN */ |
1915 | invaladdr = level == 1 ? (vaddr_t)ptes : | | 1901 | invaladdr = level == 1 ? (vaddr_t)ptes : |
1916 | (vaddr_t)pdes[level - 2]; | | 1902 | (vaddr_t)pdes[level - 2]; |
1917 | pmap_tlb_shootdown(pmap, invaladdr + index * PAGE_SIZE, | | 1903 | pmap_tlb_shootdown(pmap, invaladdr + index * PAGE_SIZE, |
1918 | opde, TLBSHOOT_FREE_PTP1); | | 1904 | opde, TLBSHOOT_FREE_PTP1); |
1919 | #endif /* XEN */ | | 1905 | #endif /* XEN */ |
1920 | pmap_freepage(pmap, ptp, level); | | 1906 | pmap_freepage(pmap, ptp, level); |
1921 | if (level < PTP_LEVELS - 1) { | | 1907 | if (level < PTP_LEVELS - 1) { |
1922 | ptp = pmap_find_ptp(pmap, va, (paddr_t)-1, level + 1); | | 1908 | ptp = pmap_find_ptp(pmap, va, (paddr_t)-1, level + 1); |
1923 | ptp->wire_count--; | | 1909 | ptp->wire_count--; |
1924 | if (ptp->wire_count > 1) | | 1910 | if (ptp->wire_count > 1) |
1925 | break; | | 1911 | break; |
1926 | } | | 1912 | } |
1927 | } while (++level < PTP_LEVELS); | | 1913 | } while (++level < PTP_LEVELS); |
1928 | pmap_pte_flush(); | | 1914 | pmap_pte_flush(); |
1929 | } | | 1915 | } |
1930 | | | 1916 | |
1931 | /* | | 1917 | /* |
1932 | * pmap_get_ptp: get a PTP (if there isn't one, allocate a new one) | | 1918 | * pmap_get_ptp: get a PTP (if there isn't one, allocate a new one) |
1933 | * | | 1919 | * |
1934 | * => pmap should NOT be pmap_kernel() | | 1920 | * => pmap should NOT be pmap_kernel() |
1935 | * => pmap should be locked | | 1921 | * => pmap should be locked |
1936 | * => preemption should be disabled | | 1922 | * => preemption should be disabled |
1937 | */ | | 1923 | */ |
1938 | | | 1924 | |
1939 | static struct vm_page * | | 1925 | static struct vm_page * |
1940 | pmap_get_ptp(struct pmap *pmap, vaddr_t va, pd_entry_t * const *pdes) | | 1926 | pmap_get_ptp(struct pmap *pmap, vaddr_t va, pd_entry_t * const *pdes) |
1941 | { | | 1927 | { |
1942 | struct vm_page *ptp, *pptp; | | 1928 | struct vm_page *ptp, *pptp; |
1943 | int i; | | 1929 | int i; |
1944 | unsigned long index; | | 1930 | unsigned long index; |
1945 | pd_entry_t *pva; | | 1931 | pd_entry_t *pva; |
1946 | paddr_t ppa, pa; | | 1932 | paddr_t ppa, pa; |
1947 | struct uvm_object *obj; | | 1933 | struct uvm_object *obj; |
1948 | | | 1934 | |
1949 | KASSERT(pmap != pmap_kernel()); | | 1935 | KASSERT(pmap != pmap_kernel()); |
1950 | KASSERT(mutex_owned(pmap->pm_lock)); | | 1936 | KASSERT(mutex_owned(pmap->pm_lock)); |
1951 | KASSERT(kpreempt_disabled()); | | 1937 | KASSERT(kpreempt_disabled()); |
1952 | | | 1938 | |
1953 | ptp = NULL; | | 1939 | ptp = NULL; |
1954 | pa = (paddr_t)-1; | | 1940 | pa = (paddr_t)-1; |
1955 | | | 1941 | |
1956 | /* | | 1942 | /* |
1957 | * Loop through all page table levels seeing if we need to | | 1943 | * Loop through all page table levels seeing if we need to |
1958 | * add a new page to that level. | | 1944 | * add a new page to that level. |
1959 | */ | | 1945 | */ |
1960 | for (i = PTP_LEVELS; i > 1; i--) { | | 1946 | for (i = PTP_LEVELS; i > 1; i--) { |
1961 | /* | | 1947 | /* |
1962 | * Save values from previous round. | | 1948 | * Save values from previous round. |
1963 | */ | | 1949 | */ |
1964 | pptp = ptp; | | 1950 | pptp = ptp; |
1965 | ppa = pa; | | 1951 | ppa = pa; |
1966 | | | 1952 | |
1967 | index = pl_i(va, i); | | 1953 | index = pl_i(va, i); |
1968 | pva = pdes[i - 2]; | | 1954 | pva = pdes[i - 2]; |
1969 | | | 1955 | |
1970 | if (pmap_valid_entry(pva[index])) { | | 1956 | if (pmap_valid_entry(pva[index])) { |
1971 | ppa = pmap_pte2pa(pva[index]); | | 1957 | ppa = pmap_pte2pa(pva[index]); |
1972 | ptp = NULL; | | 1958 | ptp = NULL; |
1973 | continue; | | 1959 | continue; |
1974 | } | | 1960 | } |
1975 | | | 1961 | |
1976 | obj = &pmap->pm_obj[i-2]; | | 1962 | obj = &pmap->pm_obj[i-2]; |
1977 | PMAP_SUBOBJ_LOCK(pmap, i - 2); | | 1963 | PMAP_SUBOBJ_LOCK(pmap, i - 2); |
1978 | ptp = uvm_pagealloc(obj, ptp_va2o(va, i - 1), NULL, | | 1964 | ptp = uvm_pagealloc(obj, ptp_va2o(va, i - 1), NULL, |
1979 | UVM_PGA_USERESERVE|UVM_PGA_ZERO); | | 1965 | UVM_PGA_USERESERVE|UVM_PGA_ZERO); |
1980 | PMAP_SUBOBJ_UNLOCK(pmap, i - 2); | | 1966 | PMAP_SUBOBJ_UNLOCK(pmap, i - 2); |
1981 | | | 1967 | |
1982 | if (ptp == NULL) | | 1968 | if (ptp == NULL) |
1983 | return NULL; | | 1969 | return NULL; |
1984 | | | 1970 | |
1985 | ptp->flags &= ~PG_BUSY; /* never busy */ | | 1971 | ptp->flags &= ~PG_BUSY; /* never busy */ |
1986 | ptp->wire_count = 1; | | 1972 | ptp->wire_count = 1; |
1987 | pmap->pm_ptphint[i - 2] = ptp; | | 1973 | pmap->pm_ptphint[i - 2] = ptp; |
1988 | pa = VM_PAGE_TO_PHYS(ptp); | | 1974 | pa = VM_PAGE_TO_PHYS(ptp); |
1989 | pmap_pte_set(&pva[index], (pd_entry_t) | | 1975 | pmap_pte_set(&pva[index], (pd_entry_t) |
1990 | (pmap_pa2pte(pa) | PG_u | PG_RW | PG_V)); | | 1976 | (pmap_pa2pte(pa) | PG_u | PG_RW | PG_V)); |
1991 | #if defined(XEN) && defined(__x86_64__) | | 1977 | #if defined(XEN) && defined(__x86_64__) |
1992 | /* | | 1978 | /* |
1993 | * In Xen we must enter the mapping in kernel map too | | 1979 | * In Xen we must enter the mapping in kernel map too |
1994 | * if pmap is curmap and modifying top level (PGD) | | 1980 | * if pmap is curmap and modifying top level (PGD) |
1995 | */ | | 1981 | */ |
1996 | if(i == PTP_LEVELS && pmap != pmap_kernel()) { | | 1982 | if(i == PTP_LEVELS && pmap != pmap_kernel()) { |
1997 | /* | | 1983 | /* |
1998 | * Update the per-cpu PD on all cpus the current | | 1984 | * Update the per-cpu PD on all cpus the current |
1999 | * pmap is active on | | 1985 | * pmap is active on |
2000 | */ | | 1986 | */ |
2001 | xen_kpm_sync(pmap, index); | | 1987 | xen_kpm_sync(pmap, index); |
2002 | } | | 1988 | } |
2003 | #endif /* XEN && __x86_64__ */ | | 1989 | #endif /* XEN && __x86_64__ */ |
2004 | pmap_pte_flush(); | | 1990 | pmap_pte_flush(); |
2005 | pmap_stats_update(pmap, 1, 0); | | 1991 | pmap_stats_update(pmap, 1, 0); |
2006 | /* | | 1992 | /* |
2007 | * If we're not in the top level, increase the | | 1993 | * If we're not in the top level, increase the |
2008 | * wire count of the parent page. | | 1994 | * wire count of the parent page. |
2009 | */ | | 1995 | */ |
2010 | if (i < PTP_LEVELS) { | | 1996 | if (i < PTP_LEVELS) { |
2011 | if (pptp == NULL) | | 1997 | if (pptp == NULL) |
2012 | pptp = pmap_find_ptp(pmap, va, ppa, i); | | 1998 | pptp = pmap_find_ptp(pmap, va, ppa, i); |
2013 | #ifdef DIAGNOSTIC | | 1999 | #ifdef DIAGNOSTIC |
2014 | if (pptp == NULL) | | 2000 | if (pptp == NULL) |
2015 | panic("pde page disappeared"); | | 2001 | panic("pde page disappeared"); |
2016 | #endif | | 2002 | #endif |
2017 | pptp->wire_count++; | | 2003 | pptp->wire_count++; |
2018 | } | | 2004 | } |
2019 | } | | 2005 | } |
2020 | | | 2006 | |
2021 | /* | | 2007 | /* |
2022 | * ptp is not NULL if we just allocated a new ptp. If it's | | 2008 | * ptp is not NULL if we just allocated a new ptp. If it's |
2023 | * still NULL, we must look up the existing one. | | 2009 | * still NULL, we must look up the existing one. |
2024 | */ | | 2010 | */ |
2025 | if (ptp == NULL) { | | 2011 | if (ptp == NULL) { |
2026 | ptp = pmap_find_ptp(pmap, va, ppa, 1); | | 2012 | ptp = pmap_find_ptp(pmap, va, ppa, 1); |
2027 | #ifdef DIAGNOSTIC | | 2013 | #ifdef DIAGNOSTIC |
2028 | if (ptp == NULL) { | | 2014 | if (ptp == NULL) { |
2029 | printf("va %" PRIxVADDR " ppa %" PRIxPADDR "\n", | | 2015 | printf("va %" PRIxVADDR " ppa %" PRIxPADDR "\n", |
2030 | va, ppa); | | 2016 | va, ppa); |
2031 | panic("pmap_get_ptp: unmanaged user PTP"); | | 2017 | panic("pmap_get_ptp: unmanaged user PTP"); |
2032 | } | | 2018 | } |
2033 | #endif | | 2019 | #endif |
2034 | } | | 2020 | } |
2035 | | | 2021 | |
2036 | pmap->pm_ptphint[0] = ptp; | | 2022 | pmap->pm_ptphint[0] = ptp; |
2037 | return(ptp); | | 2023 | return(ptp); |
2038 | } | | 2024 | } |
2039 | | | 2025 | |
2040 | /* | | 2026 | /* |
2041 | * p m a p l i f e c y c l e f u n c t i o n s | | 2027 | * p m a p l i f e c y c l e f u n c t i o n s |
2042 | */ | | 2028 | */ |
2043 | | | 2029 | |
2044 | /* | | 2030 | /* |
2045 | * pmap_pdp_ctor: constructor for the PDP cache. | | 2031 | * pmap_pdp_ctor: constructor for the PDP cache. |
2046 | */ | | 2032 | */ |
2047 | static int | | 2033 | static int |
2048 | pmap_pdp_ctor(void *arg, void *v, int flags) | | 2034 | pmap_pdp_ctor(void *arg, void *v, int flags) |
2049 | { | | 2035 | { |
2050 | pd_entry_t *pdir = v; | | 2036 | pd_entry_t *pdir = v; |
2051 | paddr_t pdirpa = 0; /* XXX: GCC */ | | 2037 | paddr_t pdirpa = 0; /* XXX: GCC */ |
2052 | vaddr_t object; | | 2038 | vaddr_t object; |
2053 | int i; | | 2039 | int i; |
2054 | | | 2040 | |
2055 | #if !defined(XEN) || !defined(__x86_64__) | | 2041 | #if !defined(XEN) || !defined(__x86_64__) |
2056 | int npde; | | 2042 | int npde; |
2057 | #endif | | 2043 | #endif |
2058 | #ifdef XEN | | 2044 | #ifdef XEN |
2059 | int s; | | 2045 | int s; |
2060 | #endif | | 2046 | #endif |
2061 | | | 2047 | |
2062 | /* | | 2048 | /* |
2063 | * NOTE: The `pmaps_lock' is held when the PDP is allocated. | | 2049 | * NOTE: The `pmaps_lock' is held when the PDP is allocated. |
2064 | */ | | 2050 | */ |
2065 | | | 2051 | |
2066 | #if defined(XEN) && defined(__x86_64__) | | 2052 | #if defined(XEN) && defined(__x86_64__) |
2067 | /* fetch the physical address of the page directory. */ | | 2053 | /* fetch the physical address of the page directory. */ |
2068 | (void) pmap_extract(pmap_kernel(), (vaddr_t) pdir, &pdirpa); | | 2054 | (void) pmap_extract(pmap_kernel(), (vaddr_t) pdir, &pdirpa); |
2069 | | | 2055 | |
2070 | /* zero init area */ | | 2056 | /* zero init area */ |
2071 | memset (pdir, 0, PAGE_SIZE); /* Xen wants a clean page */ | | 2057 | memset (pdir, 0, PAGE_SIZE); /* Xen wants a clean page */ |
2072 | /* | | 2058 | /* |
2073 | * this pdir will NEVER be active in kernel mode | | 2059 | * this pdir will NEVER be active in kernel mode |
2074 | * so mark recursive entry invalid | | 2060 | * so mark recursive entry invalid |
2075 | */ | | 2061 | */ |
2076 | pdir[PDIR_SLOT_PTE] = pmap_pa2pte(pdirpa) | PG_u; | | 2062 | pdir[PDIR_SLOT_PTE] = pmap_pa2pte(pdirpa) | PG_u; |
2077 | /* | | 2063 | /* |
2078 | * PDP constructed this way won't be for kernel, | | 2064 | * PDP constructed this way won't be for kernel, |
2079 | * hence we don't put kernel mappings on Xen. | | 2065 | * hence we don't put kernel mappings on Xen. |
2080 | * But we need to make pmap_create() happy, so put a dummy (without | | 2066 | * But we need to make pmap_create() happy, so put a dummy (without |
2081 | * PG_V) value at the right place. | | 2067 | * PG_V) value at the right place. |
2082 | */ | | 2068 | */ |
2083 | pdir[PDIR_SLOT_KERN + nkptp[PTP_LEVELS - 1] - 1] = | | 2069 | pdir[PDIR_SLOT_KERN + nkptp[PTP_LEVELS - 1] - 1] = |
2084 | (pd_entry_t)-1 & PG_FRAME; | | 2070 | (pd_entry_t)-1 & PG_FRAME; |
2085 | #else /* XEN && __x86_64__*/ | | 2071 | #else /* XEN && __x86_64__*/ |
2086 | /* zero init area */ | | 2072 | /* zero init area */ |
2087 | memset(pdir, 0, PDIR_SLOT_PTE * sizeof(pd_entry_t)); | | 2073 | memset(pdir, 0, PDIR_SLOT_PTE * sizeof(pd_entry_t)); |
2088 | | | 2074 | |
2089 | object = (vaddr_t)v; | | 2075 | object = (vaddr_t)v; |
2090 | for (i = 0; i < PDP_SIZE; i++, object += PAGE_SIZE) { | | 2076 | for (i = 0; i < PDP_SIZE; i++, object += PAGE_SIZE) { |
2091 | /* fetch the physical address of the page directory. */ | | 2077 | /* fetch the physical address of the page directory. */ |
2092 | (void) pmap_extract(pmap_kernel(), object, &pdirpa); | | 2078 | (void) pmap_extract(pmap_kernel(), object, &pdirpa); |
2093 | /* put in recursive PDE to map the PTEs */ | | 2079 | /* put in recursive PDE to map the PTEs */ |
2094 | pdir[PDIR_SLOT_PTE + i] = pmap_pa2pte(pdirpa) | PG_V; | | 2080 | pdir[PDIR_SLOT_PTE + i] = pmap_pa2pte(pdirpa) | PG_V; |
2095 | #ifndef XEN | | 2081 | #ifndef XEN |
2096 | pdir[PDIR_SLOT_PTE + i] |= PG_KW; | | 2082 | pdir[PDIR_SLOT_PTE + i] |= PG_KW; |
2097 | #endif | | 2083 | #endif |
2098 | } | | 2084 | } |
2099 | | | 2085 | |
2100 | /* copy kernel's PDE */ | | 2086 | /* copy kernel's PDE */ |
2101 | npde = nkptp[PTP_LEVELS - 1]; | | 2087 | npde = nkptp[PTP_LEVELS - 1]; |
2102 | | | 2088 | |
2103 | memcpy(&pdir[PDIR_SLOT_KERN], &PDP_BASE[PDIR_SLOT_KERN], | | 2089 | memcpy(&pdir[PDIR_SLOT_KERN], &PDP_BASE[PDIR_SLOT_KERN], |
2104 | npde * sizeof(pd_entry_t)); | | 2090 | npde * sizeof(pd_entry_t)); |
2105 | | | 2091 | |
2106 | /* zero the rest */ | | 2092 | /* zero the rest */ |
2107 | memset(&pdir[PDIR_SLOT_KERN + npde], 0, | | 2093 | memset(&pdir[PDIR_SLOT_KERN + npde], 0, |
2108 | (NTOPLEVEL_PDES - (PDIR_SLOT_KERN + npde)) * sizeof(pd_entry_t)); | | 2094 | (NTOPLEVEL_PDES - (PDIR_SLOT_KERN + npde)) * sizeof(pd_entry_t)); |
2109 | | | 2095 | |
2110 | if (VM_MIN_KERNEL_ADDRESS != KERNBASE) { | | 2096 | if (VM_MIN_KERNEL_ADDRESS != KERNBASE) { |
2111 | int idx = pl_i(KERNBASE, PTP_LEVELS); | | 2097 | int idx = pl_i(KERNBASE, PTP_LEVELS); |
2112 | | | 2098 | |
2113 | pdir[idx] = PDP_BASE[idx]; | | 2099 | pdir[idx] = PDP_BASE[idx]; |
2114 | } | | 2100 | } |
2115 | | | 2101 | |
2116 | #ifdef __HAVE_DIRECT_MAP | | 2102 | #ifdef __HAVE_DIRECT_MAP |
2117 | pdir[PDIR_SLOT_DIRECT] = PDP_BASE[PDIR_SLOT_DIRECT]; | | 2103 | pdir[PDIR_SLOT_DIRECT] = PDP_BASE[PDIR_SLOT_DIRECT]; |
2118 | #endif | | 2104 | #endif |
2119 | | | 2105 | |
2120 | #endif /* XEN && __x86_64__*/ | | 2106 | #endif /* XEN && __x86_64__*/ |
2121 | #ifdef XEN | | 2107 | #ifdef XEN |
2122 | s = splvm(); | | 2108 | s = splvm(); |
2123 | object = (vaddr_t)v; | | 2109 | object = (vaddr_t)v; |
2124 | for (i = 0; i < PDP_SIZE; i++, object += PAGE_SIZE) { | | 2110 | for (i = 0; i < PDP_SIZE; i++, object += PAGE_SIZE) { |
2125 | (void) pmap_extract(pmap_kernel(), object, &pdirpa); | | 2111 | (void) pmap_extract(pmap_kernel(), object, &pdirpa); |
2126 | /* FIXME: This should use pmap_protect() .. */ | | 2112 | /* FIXME: This should use pmap_protect() .. */ |
2127 | pmap_kenter_pa(object, pdirpa, VM_PROT_READ, 0); | | 2113 | pmap_kenter_pa(object, pdirpa, VM_PROT_READ, 0); |
2128 | pmap_update(pmap_kernel()); | | 2114 | pmap_update(pmap_kernel()); |
2129 | /* | | 2115 | /* |
2130 | * pin as L2/L4 page, we have to do the page with the | | 2116 | * pin as L2/L4 page, we have to do the page with the |
2131 | * PDIR_SLOT_PTE entries last | | 2117 | * PDIR_SLOT_PTE entries last |
2132 | */ | | 2118 | */ |
2133 | #ifdef PAE | | 2119 | #ifdef PAE |
2134 | if (i == l2tol3(PDIR_SLOT_PTE)) | | 2120 | if (i == l2tol3(PDIR_SLOT_PTE)) |
2135 | continue; | | 2121 | continue; |
2136 | #endif | | 2122 | #endif |
2137 | | | 2123 | |
2138 | #ifdef __x86_64__ | | 2124 | #ifdef __x86_64__ |
2139 | xpq_queue_pin_l4_table(xpmap_ptom_masked(pdirpa)); | | 2125 | xpq_queue_pin_l4_table(xpmap_ptom_masked(pdirpa)); |
2140 | #else | | 2126 | #else |
2141 | xpq_queue_pin_l2_table(xpmap_ptom_masked(pdirpa)); | | 2127 | xpq_queue_pin_l2_table(xpmap_ptom_masked(pdirpa)); |
2142 | #endif | | 2128 | #endif |
2143 | } | | 2129 | } |
2144 | #ifdef PAE | | 2130 | #ifdef PAE |
2145 | object = ((vaddr_t)pdir) + PAGE_SIZE * l2tol3(PDIR_SLOT_PTE); | | 2131 | object = ((vaddr_t)pdir) + PAGE_SIZE * l2tol3(PDIR_SLOT_PTE); |
2146 | (void)pmap_extract(pmap_kernel(), object, &pdirpa); | | 2132 | (void)pmap_extract(pmap_kernel(), object, &pdirpa); |
2147 | xpq_queue_pin_l2_table(xpmap_ptom_masked(pdirpa)); | | 2133 | xpq_queue_pin_l2_table(xpmap_ptom_masked(pdirpa)); |
2148 | #endif | | 2134 | #endif |
2149 | splx(s); | | 2135 | splx(s); |
2150 | #endif /* XEN */ | | 2136 | #endif /* XEN */ |
2151 | | | 2137 | |
2152 | return (0); | | 2138 | return (0); |
2153 | } | | 2139 | } |
2154 | | | 2140 | |
2155 | /* | | 2141 | /* |
2156 | * pmap_pdp_dtor: destructor for the PDP cache. | | 2142 | * pmap_pdp_dtor: destructor for the PDP cache. |
2157 | */ | | 2143 | */ |
2158 | | | 2144 | |
2159 | static void | | 2145 | static void |
2160 | pmap_pdp_dtor(void *arg, void *v) | | 2146 | pmap_pdp_dtor(void *arg, void *v) |
2161 | { | | 2147 | { |
2162 | #ifdef XEN | | 2148 | #ifdef XEN |
2163 | paddr_t pdirpa = 0; /* XXX: GCC */ | | 2149 | paddr_t pdirpa = 0; /* XXX: GCC */ |
2164 | vaddr_t object = (vaddr_t)v; | | 2150 | vaddr_t object = (vaddr_t)v; |
2165 | int i; | | 2151 | int i; |
2166 | int s = splvm(); | | 2152 | int s = splvm(); |
2167 | pt_entry_t *pte; | | 2153 | pt_entry_t *pte; |
2168 | | | 2154 | |
2169 | for (i = 0; i < PDP_SIZE; i++, object += PAGE_SIZE) { | | 2155 | for (i = 0; i < PDP_SIZE; i++, object += PAGE_SIZE) { |
2170 | /* fetch the physical address of the page directory. */ | | 2156 | /* fetch the physical address of the page directory. */ |
2171 | (void) pmap_extract(pmap_kernel(), object, &pdirpa); | | 2157 | (void) pmap_extract(pmap_kernel(), object, &pdirpa); |
2172 | /* unpin page table */ | | 2158 | /* unpin page table */ |
2173 | xpq_queue_unpin_table(xpmap_ptom_masked(pdirpa)); | | 2159 | xpq_queue_unpin_table(xpmap_ptom_masked(pdirpa)); |
2174 | } | | 2160 | } |
2175 | object = (vaddr_t)v; | | 2161 | object = (vaddr_t)v; |
2176 | for (i = 0; i < PDP_SIZE; i++, object += PAGE_SIZE) { | | 2162 | for (i = 0; i < PDP_SIZE; i++, object += PAGE_SIZE) { |
2177 | /* Set page RW again */ | | 2163 | /* Set page RW again */ |
2178 | pte = kvtopte(object); | | 2164 | pte = kvtopte(object); |
2179 | xpq_queue_pte_update(xpmap_ptetomach(pte), *pte | PG_RW); | | 2165 | xpq_queue_pte_update(xpmap_ptetomach(pte), *pte | PG_RW); |
2180 | xpq_queue_invlpg((vaddr_t)object); | | 2166 | xpq_queue_invlpg((vaddr_t)object); |
2181 | } | | 2167 | } |
2182 | splx(s); | | 2168 | splx(s); |
2183 | #endif /* XEN */ | | 2169 | #endif /* XEN */ |
2184 | } | | 2170 | } |
2185 | | | 2171 | |
2186 | #ifdef PAE | | 2172 | #ifdef PAE |
2187 | | | 2173 | |
2188 | /* pmap_pdp_alloc: Allocate a page for the pdp memory pool. */ | | 2174 | /* pmap_pdp_alloc: Allocate a page for the pdp memory pool. */ |
2189 | | | 2175 | |
2190 | static void * | | 2176 | static void * |
2191 | pmap_pdp_alloc(struct pool *pp, int flags) | | 2177 | pmap_pdp_alloc(struct pool *pp, int flags) |
2192 | { | | 2178 | { |
2193 | return (void *)uvm_km_alloc(kernel_map, | | 2179 | return (void *)uvm_km_alloc(kernel_map, |
2194 | PAGE_SIZE * PDP_SIZE, PAGE_SIZE * PDP_SIZE, | | 2180 | PAGE_SIZE * PDP_SIZE, PAGE_SIZE * PDP_SIZE, |
2195 | ((flags & PR_WAITOK) ? 0 : UVM_KMF_NOWAIT | UVM_KMF_TRYLOCK) | | 2181 | ((flags & PR_WAITOK) ? 0 : UVM_KMF_NOWAIT | UVM_KMF_TRYLOCK) |
2196 | | UVM_KMF_WIRED); | | 2182 | | UVM_KMF_WIRED); |
2197 | } | | 2183 | } |
2198 | | | 2184 | |
2199 | /* | | 2185 | /* |
2200 | * pmap_pdp_free: free a PDP | | 2186 | * pmap_pdp_free: free a PDP |
2201 | */ | | 2187 | */ |
2202 | | | 2188 | |
2203 | static void | | 2189 | static void |
2204 | pmap_pdp_free(struct pool *pp, void *v) | | 2190 | pmap_pdp_free(struct pool *pp, void *v) |
2205 | { | | 2191 | { |
2206 | uvm_km_free(kernel_map, (vaddr_t)v, PAGE_SIZE * PDP_SIZE, | | 2192 | uvm_km_free(kernel_map, (vaddr_t)v, PAGE_SIZE * PDP_SIZE, |
2207 | UVM_KMF_WIRED); | | 2193 | UVM_KMF_WIRED); |
2208 | } | | 2194 | } |
2209 | #endif /* PAE */ | | 2195 | #endif /* PAE */ |
2210 | | | 2196 | |
2211 | /* | | 2197 | /* |
2212 | * pmap_create: create a pmap | | 2198 | * pmap_create: create a pmap |
2213 | * | | 2199 | * |
2214 | * => note: old pmap interface took a "size" args which allowed for | | 2200 | * => note: old pmap interface took a "size" args which allowed for |
2215 | * the creation of "software only" pmaps (not in bsd). | | 2201 | * the creation of "software only" pmaps (not in bsd). |
2216 | */ | | 2202 | */ |
2217 | | | 2203 | |
2218 | struct pmap * | | 2204 | struct pmap * |
2219 | pmap_create(void) | | 2205 | pmap_create(void) |
2220 | { | | 2206 | { |
2221 | struct pmap *pmap; | | 2207 | struct pmap *pmap; |
2222 | int i; | | 2208 | int i; |
2223 | | | 2209 | |
2224 | pmap = pool_cache_get(&pmap_cache, PR_WAITOK); | | 2210 | pmap = pool_cache_get(&pmap_cache, PR_WAITOK); |
2225 | | | 2211 | |
2226 | /* init uvm_object */ | | 2212 | /* init uvm_object */ |
2227 | for (i = 0; i < PTP_LEVELS - 1; i++) { | | 2213 | for (i = 0; i < PTP_LEVELS - 1; i++) { |
2228 | mutex_init(&pmap->pm_obj_lock[i], MUTEX_DEFAULT, IPL_NONE); | | 2214 | mutex_init(&pmap->pm_obj_lock[i], MUTEX_DEFAULT, IPL_NONE); |
2229 | uvm_obj_init(&pmap->pm_obj[i], NULL, false, 1); | | 2215 | uvm_obj_init(&pmap->pm_obj[i], NULL, false, 1); |
2230 | uvm_obj_setlock(&pmap->pm_obj[i], &pmap->pm_obj_lock[i]); | | 2216 | uvm_obj_setlock(&pmap->pm_obj[i], &pmap->pm_obj_lock[i]); |
2231 | pmap->pm_ptphint[i] = NULL; | | 2217 | pmap->pm_ptphint[i] = NULL; |
2232 | } | | 2218 | } |
2233 | pmap->pm_stats.wired_count = 0; | | 2219 | pmap->pm_stats.wired_count = 0; |
2234 | /* count the PDP allocd below */ | | 2220 | /* count the PDP allocd below */ |
2235 | pmap->pm_stats.resident_count = PDP_SIZE; | | 2221 | pmap->pm_stats.resident_count = PDP_SIZE; |
2236 | #if !defined(__x86_64__) | | 2222 | #if !defined(__x86_64__) |
2237 | pmap->pm_hiexec = 0; | | 2223 | pmap->pm_hiexec = 0; |
2238 | #endif /* !defined(__x86_64__) */ | | 2224 | #endif /* !defined(__x86_64__) */ |
2239 | pmap->pm_flags = 0; | | 2225 | pmap->pm_flags = 0; |
2240 | pmap->pm_cpus = 0; | | 2226 | pmap->pm_cpus = 0; |
2241 | pmap->pm_kernel_cpus = 0; | | 2227 | pmap->pm_kernel_cpus = 0; |
2242 | pmap->pm_gc_ptp = NULL; | | 2228 | pmap->pm_gc_ptp = NULL; |
2243 | | | 2229 | |
2244 | /* init the LDT */ | | 2230 | /* init the LDT */ |
2245 | pmap->pm_ldt = NULL; | | 2231 | pmap->pm_ldt = NULL; |
2246 | pmap->pm_ldt_len = 0; | | 2232 | pmap->pm_ldt_len = 0; |
2247 | pmap->pm_ldt_sel = GSYSSEL(GLDT_SEL, SEL_KPL); | | 2233 | pmap->pm_ldt_sel = GSYSSEL(GLDT_SEL, SEL_KPL); |
2248 | | | 2234 | |
2249 | /* allocate PDP */ | | 2235 | /* allocate PDP */ |
2250 | try_again: | | 2236 | try_again: |
2251 | pmap->pm_pdir = pool_cache_get(&pmap_pdp_cache, PR_WAITOK); | | 2237 | pmap->pm_pdir = pool_cache_get(&pmap_pdp_cache, PR_WAITOK); |
2252 | | | 2238 | |
2253 | mutex_enter(&pmaps_lock); | | 2239 | mutex_enter(&pmaps_lock); |
2254 | | | 2240 | |
2255 | if (pmap->pm_pdir[PDIR_SLOT_KERN + nkptp[PTP_LEVELS - 1] - 1] == 0) { | | 2241 | if (pmap->pm_pdir[PDIR_SLOT_KERN + nkptp[PTP_LEVELS - 1] - 1] == 0) { |
2256 | mutex_exit(&pmaps_lock); | | 2242 | mutex_exit(&pmaps_lock); |
2257 | pool_cache_destruct_object(&pmap_pdp_cache, pmap->pm_pdir); | | 2243 | pool_cache_destruct_object(&pmap_pdp_cache, pmap->pm_pdir); |
2258 | goto try_again; | | 2244 | goto try_again; |
2259 | } | | 2245 | } |
2260 | | | 2246 | |
2261 | for (i = 0; i < PDP_SIZE; i++) | | 2247 | for (i = 0; i < PDP_SIZE; i++) |
2262 | pmap->pm_pdirpa[i] = | | 2248 | pmap->pm_pdirpa[i] = |
2263 | pmap_pte2pa(pmap->pm_pdir[PDIR_SLOT_PTE + i]); | | 2249 | pmap_pte2pa(pmap->pm_pdir[PDIR_SLOT_PTE + i]); |
2264 | | | 2250 | |
2265 | LIST_INSERT_HEAD(&pmaps, pmap, pm_list); | | 2251 | LIST_INSERT_HEAD(&pmaps, pmap, pm_list); |
2266 | | | 2252 | |
2267 | mutex_exit(&pmaps_lock); | | 2253 | mutex_exit(&pmaps_lock); |
2268 | | | 2254 | |
2269 | return (pmap); | | 2255 | return (pmap); |
2270 | } | | 2256 | } |
2271 | | | 2257 | |
2272 | /* | | 2258 | /* |
2273 | * pmap_free_ptps: put a list of ptps back to the freelist. | | 2259 | * pmap_free_ptps: put a list of ptps back to the freelist. |
2274 | */ | | 2260 | */ |
2275 | | | 2261 | |
2276 | static void | | 2262 | static void |
2277 | pmap_free_ptps(struct vm_page *empty_ptps) | | 2263 | pmap_free_ptps(struct vm_page *empty_ptps) |
2278 | { | | 2264 | { |
2279 | struct vm_page *ptp; | | 2265 | struct vm_page *ptp; |
2280 | struct pmap_page *pp; | | 2266 | struct pmap_page *pp; |
2281 | | | 2267 | |
2282 | while ((ptp = empty_ptps) != NULL) { | | 2268 | while ((ptp = empty_ptps) != NULL) { |
2283 | pp = VM_PAGE_TO_PP(ptp); | | 2269 | pp = VM_PAGE_TO_PP(ptp); |
2284 | empty_ptps = pp->pp_link; | | 2270 | empty_ptps = pp->pp_link; |
2285 | LIST_INIT(&pp->pp_head.pvh_list); | | 2271 | LIST_INIT(&pp->pp_head.pvh_list); |
2286 | uvm_pagefree(ptp); | | 2272 | uvm_pagefree(ptp); |
2287 | } | | 2273 | } |
2288 | } | | 2274 | } |
2289 | | | 2275 | |
2290 | /* | | 2276 | /* |
2291 | * pmap_destroy: drop reference count on pmap. free pmap if | | 2277 | * pmap_destroy: drop reference count on pmap. free pmap if |
2292 | * reference count goes to zero. | | 2278 | * reference count goes to zero. |
2293 | */ | | 2279 | */ |
2294 | | | 2280 | |
2295 | void | | 2281 | void |
2296 | pmap_destroy(struct pmap *pmap) | | 2282 | pmap_destroy(struct pmap *pmap) |
2297 | { | | 2283 | { |
2298 | int i; | | 2284 | int i; |
2299 | #ifdef DIAGNOSTIC | | 2285 | #ifdef DIAGNOSTIC |
2300 | struct cpu_info *ci; | | 2286 | struct cpu_info *ci; |
2301 | CPU_INFO_ITERATOR cii; | | 2287 | CPU_INFO_ITERATOR cii; |
2302 | #endif /* DIAGNOSTIC */ | | 2288 | #endif /* DIAGNOSTIC */ |
2303 | lwp_t *l; | | 2289 | lwp_t *l; |
2304 | | | 2290 | |
2305 | /* | | 2291 | /* |
2306 | * If we have torn down this pmap, process deferred frees and | | 2292 | * If we have torn down this pmap, process deferred frees and |
2307 | * invalidations. Free now if the system is low on memory. | | 2293 | * invalidations. Free now if the system is low on memory. |
2308 | * Otherwise, free when the pmap is destroyed thus avoiding a | | 2294 | * Otherwise, free when the pmap is destroyed thus avoiding a |
2309 | * TLB shootdown. | | 2295 | * TLB shootdown. |
2310 | */ | | 2296 | */ |
2311 | l = curlwp; | | 2297 | l = curlwp; |
2312 | if (__predict_false(l->l_md.md_gc_pmap == pmap)) { | | 2298 | if (__predict_false(l->l_md.md_gc_pmap == pmap)) { |
2313 | if (uvmexp.free < uvmexp.freetarg) { | | 2299 | if (uvmexp.free < uvmexp.freetarg) { |
2314 | pmap_update(pmap); | | 2300 | pmap_update(pmap); |
2315 | } else { | | 2301 | } else { |
2316 | KASSERT(pmap->pm_gc_ptp == NULL); | | 2302 | KASSERT(pmap->pm_gc_ptp == NULL); |
2317 | pmap->pm_gc_ptp = l->l_md.md_gc_ptp; | | 2303 | pmap->pm_gc_ptp = l->l_md.md_gc_ptp; |
2318 | l->l_md.md_gc_ptp = NULL; | | 2304 | l->l_md.md_gc_ptp = NULL; |
2319 | l->l_md.md_gc_pmap = NULL; | | 2305 | l->l_md.md_gc_pmap = NULL; |
2320 | } | | 2306 | } |
2321 | } | | 2307 | } |
2322 | | | 2308 | |
2323 | /* | | 2309 | /* |
2324 | * drop reference count | | 2310 | * drop reference count |
2325 | */ | | 2311 | */ |
2326 | | | 2312 | |
2327 | if (atomic_dec_uint_nv(&pmap->pm_obj[0].uo_refs) > 0) { | | 2313 | if (atomic_dec_uint_nv(&pmap->pm_obj[0].uo_refs) > 0) { |
2328 | return; | | 2314 | return; |
2329 | } | | 2315 | } |
2330 | | | 2316 | |
2331 | #ifdef DIAGNOSTIC | | 2317 | #ifdef DIAGNOSTIC |
2332 | for (CPU_INFO_FOREACH(cii, ci)) | | 2318 | for (CPU_INFO_FOREACH(cii, ci)) |
2333 | if (ci->ci_pmap == pmap) | | 2319 | if (ci->ci_pmap == pmap) |
2334 | panic("destroying pmap being used"); | | 2320 | panic("destroying pmap being used"); |
2335 | #endif /* DIAGNOSTIC */ | | 2321 | #endif /* DIAGNOSTIC */ |
2336 | | | 2322 | |
2337 | /* | | 2323 | /* |
2338 | * reference count is zero, free pmap resources and then free pmap. | | 2324 | * reference count is zero, free pmap resources and then free pmap. |
2339 | */ | | 2325 | */ |
2340 | | | 2326 | |
2341 | /* | | 2327 | /* |
2342 | * remove it from global list of pmaps | | 2328 | * remove it from global list of pmaps |
2343 | */ | | 2329 | */ |
2344 | | | 2330 | |
2345 | mutex_enter(&pmaps_lock); | | 2331 | mutex_enter(&pmaps_lock); |
2346 | LIST_REMOVE(pmap, pm_list); | | 2332 | LIST_REMOVE(pmap, pm_list); |
2347 | mutex_exit(&pmaps_lock); | | 2333 | mutex_exit(&pmaps_lock); |
2348 | | | 2334 | |
2349 | /* | | 2335 | /* |
2350 | * Process deferred PTP frees. No TLB shootdown required, as the | | 2336 | * Process deferred PTP frees. No TLB shootdown required, as the |
2351 | * PTP pages are no longer visible to any CPU. | | 2337 | * PTP pages are no longer visible to any CPU. |
2352 | */ | | 2338 | */ |
2353 | | | 2339 | |
2354 | pmap_free_ptps(pmap->pm_gc_ptp); | | 2340 | pmap_free_ptps(pmap->pm_gc_ptp); |
2355 | | | 2341 | |
2356 | /* | | 2342 | /* |
2357 | * destroyed pmap shouldn't have remaining PTPs | | 2343 | * destroyed pmap shouldn't have remaining PTPs |
2358 | */ | | 2344 | */ |
2359 | | | 2345 | |
2360 | for (i = 0; i < PTP_LEVELS - 1; i++) { | | 2346 | for (i = 0; i < PTP_LEVELS - 1; i++) { |
2361 | KASSERT(pmap->pm_obj[i].uo_npages == 0); | | 2347 | KASSERT(pmap->pm_obj[i].uo_npages == 0); |
2362 | KASSERT(TAILQ_EMPTY(&pmap->pm_obj[i].memq)); | | 2348 | KASSERT(TAILQ_EMPTY(&pmap->pm_obj[i].memq)); |
2363 | } | | 2349 | } |
2364 | | | 2350 | |
2365 | /* | | 2351 | /* |
2366 | * MULTIPROCESSOR -- no need to flush out of other processors' | | 2352 | * MULTIPROCESSOR -- no need to flush out of other processors' |
2367 | * APTE space because we do that in pmap_unmap_ptes(). | | 2353 | * APTE space because we do that in pmap_unmap_ptes(). |
2368 | */ | | 2354 | */ |
2369 | pool_cache_put(&pmap_pdp_cache, pmap->pm_pdir); | | 2355 | pool_cache_put(&pmap_pdp_cache, pmap->pm_pdir); |
2370 | | | 2356 | |
2371 | #ifdef USER_LDT | | 2357 | #ifdef USER_LDT |
2372 | if (pmap->pm_ldt != NULL) { | | 2358 | if (pmap->pm_ldt != NULL) { |
2373 | /* | | 2359 | /* |
2374 | * no need to switch the LDT; this address space is gone, | | 2360 | * no need to switch the LDT; this address space is gone, |
2375 | * nothing is using it. | | 2361 | * nothing is using it. |
2376 | * | | 2362 | * |
2377 | * No need to lock the pmap for ldt_free (or anything else), | | 2363 | * No need to lock the pmap for ldt_free (or anything else), |
2378 | * we're the last one to use it. | | 2364 | * we're the last one to use it. |
2379 | */ | | 2365 | */ |
2380 | mutex_enter(&cpu_lock); | | 2366 | mutex_enter(&cpu_lock); |
2381 | ldt_free(pmap->pm_ldt_sel); | | 2367 | ldt_free(pmap->pm_ldt_sel); |
2382 | mutex_exit(&cpu_lock); | | 2368 | mutex_exit(&cpu_lock); |
2383 | uvm_km_free(kernel_map, (vaddr_t)pmap->pm_ldt, | | 2369 | uvm_km_free(kernel_map, (vaddr_t)pmap->pm_ldt, |
2384 | pmap->pm_ldt_len, UVM_KMF_WIRED); | | 2370 | pmap->pm_ldt_len, UVM_KMF_WIRED); |
2385 | } | | 2371 | } |
2386 | #endif | | 2372 | #endif |
2387 | | | 2373 | |
2388 | for (i = 0; i < PTP_LEVELS - 1; i++) { | | 2374 | for (i = 0; i < PTP_LEVELS - 1; i++) { |
2389 | uvm_obj_destroy(&pmap->pm_obj[i], false); | | 2375 | uvm_obj_destroy(&pmap->pm_obj[i], false); |
2390 | mutex_destroy(&pmap->pm_obj_lock[i]); | | 2376 | mutex_destroy(&pmap->pm_obj_lock[i]); |
2391 | } | | 2377 | } |
2392 | pool_cache_put(&pmap_cache, pmap); | | 2378 | pool_cache_put(&pmap_cache, pmap); |
2393 | } | | 2379 | } |
2394 | | | 2380 | |
2395 | /* | | 2381 | /* |
2396 | * pmap_remove_all: pmap is being torn down by the current thread. | | 2382 | * pmap_remove_all: pmap is being torn down by the current thread. |
2397 | * avoid unnecessary invalidations. | | 2383 | * avoid unnecessary invalidations. |
2398 | */ | | 2384 | */ |
2399 | | | 2385 | |
2400 | void | | 2386 | void |
2401 | pmap_remove_all(struct pmap *pmap) | | 2387 | pmap_remove_all(struct pmap *pmap) |
2402 | { | | 2388 | { |
2403 | lwp_t *l = curlwp; | | 2389 | lwp_t *l = curlwp; |
2404 | | | 2390 | |
2405 | KASSERT(l->l_md.md_gc_pmap == NULL); | | 2391 | KASSERT(l->l_md.md_gc_pmap == NULL); |
2406 | | | 2392 | |
2407 | l->l_md.md_gc_pmap = pmap; | | 2393 | l->l_md.md_gc_pmap = pmap; |
2408 | } | | 2394 | } |
2409 | | | 2395 | |
2410 | #if defined(PMAP_FORK) | | 2396 | #if defined(PMAP_FORK) |
2411 | /* | | 2397 | /* |
2412 | * pmap_fork: perform any necessary data structure manipulation when | | 2398 | * pmap_fork: perform any necessary data structure manipulation when |
2413 | * a VM space is forked. | | 2399 | * a VM space is forked. |
2414 | */ | | 2400 | */ |
2415 | | | 2401 | |
2416 | void | | 2402 | void |
2417 | pmap_fork(struct pmap *pmap1, struct pmap *pmap2) | | 2403 | pmap_fork(struct pmap *pmap1, struct pmap *pmap2) |
2418 | { | | 2404 | { |
2419 | #ifdef USER_LDT | | 2405 | #ifdef USER_LDT |
2420 | union descriptor *new_ldt; | | 2406 | union descriptor *new_ldt; |
2421 | size_t len; | | 2407 | size_t len; |
2422 | int sel; | | 2408 | int sel; |
2423 | | | 2409 | |
2424 | if (__predict_true(pmap1->pm_ldt == NULL)) { | | 2410 | if (__predict_true(pmap1->pm_ldt == NULL)) { |
2425 | return; | | 2411 | return; |
2426 | } | | 2412 | } |
2427 | | | 2413 | |
2428 | retry: | | 2414 | retry: |
2429 | if (pmap1->pm_ldt != NULL) { | | 2415 | if (pmap1->pm_ldt != NULL) { |
2430 | len = pmap1->pm_ldt_len; | | 2416 | len = pmap1->pm_ldt_len; |
2431 | new_ldt = (union descriptor *)uvm_km_alloc(kernel_map, len, 0, | | 2417 | new_ldt = (union descriptor *)uvm_km_alloc(kernel_map, len, 0, |
2432 | UVM_KMF_WIRED); | | 2418 | UVM_KMF_WIRED); |
2433 | mutex_enter(&cpu_lock); | | 2419 | mutex_enter(&cpu_lock); |
2434 | sel = ldt_alloc(new_ldt, len); | | 2420 | sel = ldt_alloc(new_ldt, len); |
2435 | if (sel == -1) { | | 2421 | if (sel == -1) { |
2436 | mutex_exit(&cpu_lock); | | 2422 | mutex_exit(&cpu_lock); |
2437 | uvm_km_free(kernel_map, (vaddr_t)new_ldt, len, | | 2423 | uvm_km_free(kernel_map, (vaddr_t)new_ldt, len, |
2438 | UVM_KMF_WIRED); | | 2424 | UVM_KMF_WIRED); |
2439 | printf("WARNING: pmap_fork: unable to allocate LDT\n"); | | 2425 | printf("WARNING: pmap_fork: unable to allocate LDT\n"); |
2440 | return; | | 2426 | return; |
2441 | } | | 2427 | } |
2442 | } else { | | 2428 | } else { |
2443 | len = -1; | | 2429 | len = -1; |
2444 | new_ldt = NULL; | | 2430 | new_ldt = NULL; |
2445 | sel = -1; | | 2431 | sel = -1; |
2446 | mutex_enter(&cpu_lock); | | 2432 | mutex_enter(&cpu_lock); |
2447 | } | | 2433 | } |
2448 | | | 2434 | |
2449 | /* Copy the LDT, if necessary. */ | | 2435 | /* Copy the LDT, if necessary. */ |
2450 | if (pmap1->pm_ldt != NULL) { | | 2436 | if (pmap1->pm_ldt != NULL) { |
2451 | if (len != pmap1->pm_ldt_len) { | | 2437 | if (len != pmap1->pm_ldt_len) { |
2452 | if (len != -1) { | | 2438 | if (len != -1) { |
2453 | ldt_free(sel); | | 2439 | ldt_free(sel); |
2454 | uvm_km_free(kernel_map, (vaddr_t)new_ldt, | | 2440 | uvm_km_free(kernel_map, (vaddr_t)new_ldt, |
2455 | len, UVM_KMF_WIRED); | | 2441 | len, UVM_KMF_WIRED); |
2456 | } | | 2442 | } |
2457 | mutex_exit(&cpu_lock); | | 2443 | mutex_exit(&cpu_lock); |
2458 | goto retry; | | 2444 | goto retry; |
2459 | } | | 2445 | } |
2460 | | | 2446 | |
2461 | memcpy(new_ldt, pmap1->pm_ldt, len); | | 2447 | memcpy(new_ldt, pmap1->pm_ldt, len); |
2462 | pmap2->pm_ldt = new_ldt; | | 2448 | pmap2->pm_ldt = new_ldt; |
2463 | pmap2->pm_ldt_len = pmap1->pm_ldt_len; | | 2449 | pmap2->pm_ldt_len = pmap1->pm_ldt_len; |
2464 | pmap2->pm_ldt_sel = sel; | | 2450 | pmap2->pm_ldt_sel = sel; |
2465 | len = -1; | | 2451 | len = -1; |
2466 | } | | 2452 | } |
2467 | | | 2453 | |
2468 | if (len != -1) { | | 2454 | if (len != -1) { |
2469 | ldt_free(sel); | | 2455 | ldt_free(sel); |
2470 | uvm_km_free(kernel_map, (vaddr_t)new_ldt, len, | | 2456 | uvm_km_free(kernel_map, (vaddr_t)new_ldt, len, |
2471 | UVM_KMF_WIRED); | | 2457 | UVM_KMF_WIRED); |
2472 | } | | 2458 | } |
2473 | mutex_exit(&cpu_lock); | | 2459 | mutex_exit(&cpu_lock); |
2474 | #endif /* USER_LDT */ | | 2460 | #endif /* USER_LDT */ |
2475 | } | | 2461 | } |
2476 | #endif /* PMAP_FORK */ | | 2462 | #endif /* PMAP_FORK */ |
2477 | | | 2463 | |
2478 | #ifdef USER_LDT | | 2464 | #ifdef USER_LDT |
2479 | | | 2465 | |
2480 | /* | | 2466 | /* |
2481 | * pmap_ldt_xcall: cross call used by pmap_ldt_sync. if the named pmap | | 2467 | * pmap_ldt_xcall: cross call used by pmap_ldt_sync. if the named pmap |
2482 | * is active, reload LDTR. | | 2468 | * is active, reload LDTR. |
2483 | */ | | 2469 | */ |
2484 | static void | | 2470 | static void |
2485 | pmap_ldt_xcall(void *arg1, void *arg2) | | 2471 | pmap_ldt_xcall(void *arg1, void *arg2) |
2486 | { | | 2472 | { |
2487 | struct pmap *pm; | | 2473 | struct pmap *pm; |
2488 | | | 2474 | |
2489 | kpreempt_disable(); | | 2475 | kpreempt_disable(); |
2490 | pm = arg1; | | 2476 | pm = arg1; |
2491 | if (curcpu()->ci_pmap == pm) { | | 2477 | if (curcpu()->ci_pmap == pm) { |
2492 | lldt(pm->pm_ldt_sel); | | 2478 | lldt(pm->pm_ldt_sel); |
2493 | } | | 2479 | } |
2494 | kpreempt_enable(); | | 2480 | kpreempt_enable(); |
2495 | } | | 2481 | } |
2496 | | | 2482 | |
2497 | /* | | 2483 | /* |
2498 | * pmap_ldt_sync: LDT selector for the named pmap is changing. swap | | 2484 | * pmap_ldt_sync: LDT selector for the named pmap is changing. swap |
2499 | * in the new selector on all CPUs. | | 2485 | * in the new selector on all CPUs. |
2500 | */ | | 2486 | */ |
2501 | void | | 2487 | void |
2502 | pmap_ldt_sync(struct pmap *pm) | | 2488 | pmap_ldt_sync(struct pmap *pm) |
2503 | { | | 2489 | { |
2504 | uint64_t where; | | 2490 | uint64_t where; |
2505 | | | 2491 | |
2506 | KASSERT(mutex_owned(&cpu_lock)); | | 2492 | KASSERT(mutex_owned(&cpu_lock)); |
2507 | | | 2493 | |
2508 | pmap_ldt_evcnt.ev_count++; | | 2494 | pmap_ldt_evcnt.ev_count++; |
2509 | where = xc_broadcast(0, pmap_ldt_xcall, pm, NULL); | | 2495 | where = xc_broadcast(0, pmap_ldt_xcall, pm, NULL); |
2510 | xc_wait(where); | | 2496 | xc_wait(where); |
2511 | } | | 2497 | } |
2512 | | | 2498 | |
2513 | /* | | 2499 | /* |
2514 | * pmap_ldt_cleanup: if the pmap has a local LDT, deallocate it, and | | 2500 | * pmap_ldt_cleanup: if the pmap has a local LDT, deallocate it, and |
2515 | * restore the default. | | 2501 | * restore the default. |
2516 | */ | | 2502 | */ |
2517 | | | 2503 | |
2518 | void | | 2504 | void |
2519 | pmap_ldt_cleanup(struct lwp *l) | | 2505 | pmap_ldt_cleanup(struct lwp *l) |
2520 | { | | 2506 | { |
2521 | pmap_t pmap = l->l_proc->p_vmspace->vm_map.pmap; | | 2507 | pmap_t pmap = l->l_proc->p_vmspace->vm_map.pmap; |
2522 | union descriptor *dp = NULL; | | 2508 | union descriptor *dp = NULL; |
2523 | size_t len = 0; | | 2509 | size_t len = 0; |
2524 | int sel = -1; | | 2510 | int sel = -1; |
2525 | | | 2511 | |
2526 | if (__predict_true(pmap->pm_ldt == NULL)) { | | 2512 | if (__predict_true(pmap->pm_ldt == NULL)) { |
2527 | return; | | 2513 | return; |
2528 | } | | 2514 | } |
2529 | | | 2515 | |
2530 | mutex_enter(&cpu_lock); | | 2516 | mutex_enter(&cpu_lock); |
2531 | if (pmap->pm_ldt != NULL) { | | 2517 | if (pmap->pm_ldt != NULL) { |
2532 | sel = pmap->pm_ldt_sel; | | 2518 | sel = pmap->pm_ldt_sel; |
2533 | dp = pmap->pm_ldt; | | 2519 | dp = pmap->pm_ldt; |
2534 | len = pmap->pm_ldt_len; | | 2520 | len = pmap->pm_ldt_len; |
2535 | pmap->pm_ldt_sel = GSYSSEL(GLDT_SEL, SEL_KPL); | | 2521 | pmap->pm_ldt_sel = GSYSSEL(GLDT_SEL, SEL_KPL); |
2536 | pmap->pm_ldt = NULL; | | 2522 | pmap->pm_ldt = NULL; |
2537 | pmap->pm_ldt_len = 0; | | 2523 | pmap->pm_ldt_len = 0; |
2538 | pmap_ldt_sync(pmap); | | 2524 | pmap_ldt_sync(pmap); |
2539 | ldt_free(sel); | | 2525 | ldt_free(sel); |
2540 | uvm_km_free(kernel_map, (vaddr_t)dp, len, UVM_KMF_WIRED); | | 2526 | uvm_km_free(kernel_map, (vaddr_t)dp, len, UVM_KMF_WIRED); |
2541 | } | | 2527 | } |
2542 | mutex_exit(&cpu_lock); | | 2528 | mutex_exit(&cpu_lock); |
2543 | } | | 2529 | } |
2544 | #endif /* USER_LDT */ | | 2530 | #endif /* USER_LDT */ |
2545 | | | 2531 | |
2546 | /* | | 2532 | /* |
2547 | * pmap_activate: activate a process' pmap | | 2533 | * pmap_activate: activate a process' pmap |
2548 | * | | 2534 | * |
2549 | * => must be called with kernel preemption disabled | | 2535 | * => must be called with kernel preemption disabled |
2550 | * => if lwp is the curlwp, then set ci_want_pmapload so that | | 2536 | * => if lwp is the curlwp, then set ci_want_pmapload so that |
2551 | * actual MMU context switch will be done by pmap_load() later | | 2537 | * actual MMU context switch will be done by pmap_load() later |
2552 | */ | | 2538 | */ |
2553 | | | 2539 | |
2554 | void | | 2540 | void |
2555 | pmap_activate(struct lwp *l) | | 2541 | pmap_activate(struct lwp *l) |
2556 | { | | 2542 | { |
2557 | struct cpu_info *ci; | | 2543 | struct cpu_info *ci; |
2558 | struct pmap *pmap = vm_map_pmap(&l->l_proc->p_vmspace->vm_map); | | 2544 | struct pmap *pmap = vm_map_pmap(&l->l_proc->p_vmspace->vm_map); |
2559 | | | 2545 | |
2560 | KASSERT(kpreempt_disabled()); | | 2546 | KASSERT(kpreempt_disabled()); |
2561 | | | 2547 | |
2562 | ci = curcpu(); | | 2548 | ci = curcpu(); |
2563 | | | 2549 | |
2564 | if (l == ci->ci_curlwp) { | | 2550 | if (l == ci->ci_curlwp) { |
2565 | KASSERT(ci->ci_want_pmapload == 0); | | 2551 | KASSERT(ci->ci_want_pmapload == 0); |
2566 | KASSERT(ci->ci_tlbstate != TLBSTATE_VALID); | | 2552 | KASSERT(ci->ci_tlbstate != TLBSTATE_VALID); |
2567 | #ifdef KSTACK_CHECK_DR0 | | 2553 | #ifdef KSTACK_CHECK_DR0 |
2568 | /* | | 2554 | /* |
2569 | * setup breakpoint on the top of stack | | 2555 | * setup breakpoint on the top of stack |
2570 | */ | | 2556 | */ |
2571 | if (l == &lwp0) | | 2557 | if (l == &lwp0) |
2572 | dr0(0, 0, 0, 0); | | 2558 | dr0(0, 0, 0, 0); |
2573 | else | | 2559 | else |
2574 | dr0(KSTACK_LOWEST_ADDR(l), 1, 3, 1); | | 2560 | dr0(KSTACK_LOWEST_ADDR(l), 1, 3, 1); |
2575 | #endif | | 2561 | #endif |
2576 | | | 2562 | |
2577 | /* | | 2563 | /* |
2578 | * no need to switch to kernel vmspace because | | 2564 | * no need to switch to kernel vmspace because |
2579 | * it's a subset of any vmspace. | | 2565 | * it's a subset of any vmspace. |
2580 | */ | | 2566 | */ |
2581 | | | 2567 | |
2582 | if (pmap == pmap_kernel()) { | | 2568 | if (pmap == pmap_kernel()) { |
2583 | ci->ci_want_pmapload = 0; | | 2569 | ci->ci_want_pmapload = 0; |
2584 | return; | | 2570 | return; |
2585 | } | | 2571 | } |
2586 | | | 2572 | |
2587 | ci->ci_want_pmapload = 1; | | 2573 | ci->ci_want_pmapload = 1; |
2588 | } | | 2574 | } |
2589 | } | | 2575 | } |
2590 | | | 2576 | |
2591 | /* | | 2577 | /* |
2592 | * pmap_reactivate: try to regain reference to the pmap. | | 2578 | * pmap_reactivate: try to regain reference to the pmap. |
2593 | * | | 2579 | * |
2594 | * => must be called with kernel preemption disabled | | 2580 | * => must be called with kernel preemption disabled |
2595 | */ | | 2581 | */ |
2596 | | | 2582 | |
2597 | static bool | | 2583 | static bool |
2598 | pmap_reactivate(struct pmap *pmap) | | 2584 | pmap_reactivate(struct pmap *pmap) |
2599 | { | | 2585 | { |
2600 | struct cpu_info *ci; | | 2586 | struct cpu_info *ci; |
2601 | uint32_t cpumask; | | 2587 | uint32_t cpumask; |
2602 | bool result; | | 2588 | bool result; |
2603 | uint32_t oldcpus; | | 2589 | uint32_t oldcpus; |
2604 | | | 2590 | |
2605 | ci = curcpu(); | | 2591 | ci = curcpu(); |
2606 | cpumask = ci->ci_cpumask; | | 2592 | cpumask = ci->ci_cpumask; |
2607 | | | 2593 | |
2608 | KASSERT(kpreempt_disabled()); | | 2594 | KASSERT(kpreempt_disabled()); |
2609 | #if defined(XEN) && defined(__x86_64__) | | 2595 | #if defined(XEN) && defined(__x86_64__) |
2610 | KASSERT(pmap_pdirpa(pmap, 0) == ci->ci_xen_current_user_pgd); | | 2596 | KASSERT(pmap_pdirpa(pmap, 0) == ci->ci_xen_current_user_pgd); |
2611 | #elif defined(PAE) | | 2597 | #elif defined(PAE) |
2612 | KASSERT(pmap_pdirpa(pmap, 0) == pmap_pte2pa(ci->ci_pae_l3_pdir[0])); | | 2598 | KASSERT(pmap_pdirpa(pmap, 0) == pmap_pte2pa(ci->ci_pae_l3_pdir[0])); |
2613 | #elif !defined(XEN) | | 2599 | #elif !defined(XEN) |
2614 | KASSERT(pmap_pdirpa(pmap, 0) == pmap_pte2pa(rcr3())); | | 2600 | KASSERT(pmap_pdirpa(pmap, 0) == pmap_pte2pa(rcr3())); |
2615 | #endif | | 2601 | #endif |
2616 | | | 2602 | |
2617 | /* | | 2603 | /* |
2618 | * if we still have a lazy reference to this pmap, | | 2604 | * if we still have a lazy reference to this pmap, |
2619 | * we can assume that there was no tlb shootdown | | 2605 | * we can assume that there was no tlb shootdown |
2620 | * for this pmap in the meantime. | | 2606 | * for this pmap in the meantime. |
2621 | * | | 2607 | * |
2622 | * the order of events here is important as we must | | 2608 | * the order of events here is important as we must |
2623 | * synchronize with TLB shootdown interrupts. declare | | 2609 | * synchronize with TLB shootdown interrupts. declare |
2624 | * interest in invalidations (TLBSTATE_VALID) and then | | 2610 | * interest in invalidations (TLBSTATE_VALID) and then |
2625 | * check the cpumask, which the IPIs can change only | | 2611 | * check the cpumask, which the IPIs can change only |
2626 | * when the state is TLBSTATE_LAZY. | | 2612 | * when the state is TLBSTATE_LAZY. |
2627 | */ | | 2613 | */ |
2628 | | | 2614 | |
2629 | ci->ci_tlbstate = TLBSTATE_VALID; | | 2615 | ci->ci_tlbstate = TLBSTATE_VALID; |
2630 | oldcpus = pmap->pm_cpus; | | 2616 | oldcpus = pmap->pm_cpus; |
2631 | KASSERT((pmap->pm_kernel_cpus & cpumask) != 0); | | 2617 | KASSERT((pmap->pm_kernel_cpus & cpumask) != 0); |
2632 | if (oldcpus & cpumask) { | | 2618 | if (oldcpus & cpumask) { |
2633 | /* got it */ | | 2619 | /* got it */ |
2634 | result = true; | | 2620 | result = true; |
2635 | } else { | | 2621 | } else { |
2636 | /* must reload */ | | 2622 | /* must reload */ |
2637 | atomic_or_32(&pmap->pm_cpus, cpumask); | | 2623 | atomic_or_32(&pmap->pm_cpus, cpumask); |
2638 | result = false; | | 2624 | result = false; |
2639 | } | | 2625 | } |
2640 | | | 2626 | |
2641 | return result; | | 2627 | return result; |
2642 | } | | 2628 | } |
2643 | | | 2629 | |
2644 | /* | | 2630 | /* |
2645 | * pmap_load: actually switch pmap. (fill in %cr3 and LDT info) | | 2631 | * pmap_load: actually switch pmap. (fill in %cr3 and LDT info) |
2646 | * | | 2632 | * |
2647 | * ensures that the current process' pmap is loaded on the current cpu's MMU | | 2633 | * ensures that the current process' pmap is loaded on the current cpu's MMU |
2648 | * and there's no stale TLB entries. | | 2634 | * and there's no stale TLB entries. |
2649 | * | | 2635 | * |
2650 | * the caller should disable preemption or do check-and-retry to prevent | | 2636 | * the caller should disable preemption or do check-and-retry to prevent |
2651 | * a preemption from undoing our efforts. | | 2637 | * a preemption from undoing our efforts. |
2652 | * | | 2638 | * |
2653 | * this function can block. | | 2639 | * this function can block. |
2654 | */ | | 2640 | */ |
2655 | | | 2641 | |
2656 | void | | 2642 | void |
2657 | pmap_load(void) | | 2643 | pmap_load(void) |
2658 | { | | 2644 | { |
2659 | struct cpu_info *ci; | | 2645 | struct cpu_info *ci; |
2660 | uint32_t cpumask; | | 2646 | uint32_t cpumask; |
2661 | struct pmap *pmap; | | 2647 | struct pmap *pmap; |
2662 | struct pmap *oldpmap; | | 2648 | struct pmap *oldpmap; |
2663 | struct lwp *l; | | 2649 | struct lwp *l; |
2664 | struct pcb *pcb; | | 2650 | struct pcb *pcb; |
2665 | uint64_t ncsw; | | 2651 | uint64_t ncsw; |
2666 | | | 2652 | |
2667 | kpreempt_disable(); | | 2653 | kpreempt_disable(); |
2668 | retry: | | 2654 | retry: |
2669 | ci = curcpu(); | | 2655 | ci = curcpu(); |
2670 | if (!ci->ci_want_pmapload) { | | 2656 | if (!ci->ci_want_pmapload) { |
2671 | kpreempt_enable(); | | 2657 | kpreempt_enable(); |
2672 | return; | | 2658 | return; |
2673 | } | | 2659 | } |
2674 | cpumask = ci->ci_cpumask; | | 2660 | cpumask = ci->ci_cpumask; |
2675 | l = ci->ci_curlwp; | | 2661 | l = ci->ci_curlwp; |
2676 | ncsw = l->l_ncsw; | | 2662 | ncsw = l->l_ncsw; |
2677 | | | 2663 | |
2678 | /* should be able to take ipis. */ | | 2664 | /* should be able to take ipis. */ |
2679 | KASSERT(ci->ci_ilevel < IPL_HIGH); | | 2665 | KASSERT(ci->ci_ilevel < IPL_HIGH); |
2680 | #ifdef XEN | | 2666 | #ifdef XEN |
2681 | /* Check to see if interrupts are enabled (ie; no events are masked) */ | | 2667 | /* Check to see if interrupts are enabled (ie; no events are masked) */ |
2682 | KASSERT(x86_read_psl() == 0); | | 2668 | KASSERT(x86_read_psl() == 0); |
2683 | #else | | 2669 | #else |
2684 | KASSERT((x86_read_psl() & PSL_I) != 0); | | 2670 | KASSERT((x86_read_psl() & PSL_I) != 0); |
2685 | #endif | | 2671 | #endif |
2686 | | | 2672 | |
2687 | KASSERT(l != NULL); | | 2673 | KASSERT(l != NULL); |
2688 | pmap = vm_map_pmap(&l->l_proc->p_vmspace->vm_map); | | 2674 | pmap = vm_map_pmap(&l->l_proc->p_vmspace->vm_map); |
2689 | KASSERT(pmap != pmap_kernel()); | | 2675 | KASSERT(pmap != pmap_kernel()); |
2690 | oldpmap = ci->ci_pmap; | | 2676 | oldpmap = ci->ci_pmap; |
2691 | pcb = lwp_getpcb(l); | | 2677 | pcb = lwp_getpcb(l); |
2692 | | | 2678 | |
2693 | if (pmap == oldpmap) { | | 2679 | if (pmap == oldpmap) { |
2694 | if (!pmap_reactivate(pmap)) { | | 2680 | if (!pmap_reactivate(pmap)) { |
2695 | u_int gen = uvm_emap_gen_return(); | | 2681 | u_int gen = uvm_emap_gen_return(); |
2696 | | | 2682 | |
2697 | /* | | 2683 | /* |
2698 | * pmap has been changed during deactivated. | | 2684 | * pmap has been changed during deactivated. |
2699 | * our tlb may be stale. | | 2685 | * our tlb may be stale. |
2700 | */ | | 2686 | */ |
2701 | | | 2687 | |
2702 | tlbflush(); | | 2688 | tlbflush(); |
2703 | uvm_emap_update(gen); | | 2689 | uvm_emap_update(gen); |
2704 | } | | 2690 | } |
2705 | | | 2691 | |
2706 | ci->ci_want_pmapload = 0; | | 2692 | ci->ci_want_pmapload = 0; |
2707 | kpreempt_enable(); | | 2693 | kpreempt_enable(); |
2708 | return; | | 2694 | return; |
2709 | } | | 2695 | } |
2710 | | | 2696 | |
2711 | /* | | 2697 | /* |
2712 | * grab a reference to the new pmap. | | 2698 | * grab a reference to the new pmap. |
2713 | */ | | 2699 | */ |
2714 | | | 2700 | |
2715 | pmap_reference(pmap); | | 2701 | pmap_reference(pmap); |
2716 | | | 2702 | |
2717 | /* | | 2703 | /* |
2718 | * actually switch pmap. | | 2704 | * actually switch pmap. |
2719 | */ | | 2705 | */ |
2720 | | | 2706 | |
2721 | atomic_and_32(&oldpmap->pm_cpus, ~cpumask); | | 2707 | atomic_and_32(&oldpmap->pm_cpus, ~cpumask); |
2722 | atomic_and_32(&oldpmap->pm_kernel_cpus, ~cpumask); | | 2708 | atomic_and_32(&oldpmap->pm_kernel_cpus, ~cpumask); |
2723 | | | 2709 | |
2724 | #if defined(XEN) && defined(__x86_64__) | | 2710 | #if defined(XEN) && defined(__x86_64__) |
2725 | KASSERT(pmap_pdirpa(oldpmap, 0) == ci->ci_xen_current_user_pgd || | | 2711 | KASSERT(pmap_pdirpa(oldpmap, 0) == ci->ci_xen_current_user_pgd || |
2726 | oldpmap == pmap_kernel()); | | 2712 | oldpmap == pmap_kernel()); |
2727 | #elif defined(PAE) | | 2713 | #elif defined(PAE) |
2728 | KASSERT(pmap_pdirpa(oldpmap, 0) == pmap_pte2pa(ci->ci_pae_l3_pdir[0])); | | 2714 | KASSERT(pmap_pdirpa(oldpmap, 0) == pmap_pte2pa(ci->ci_pae_l3_pdir[0])); |
2729 | #elif !defined(XEN) | | 2715 | #elif !defined(XEN) |
2730 | KASSERT(pmap_pdirpa(oldpmap, 0) == pmap_pte2pa(rcr3())); | | 2716 | KASSERT(pmap_pdirpa(oldpmap, 0) == pmap_pte2pa(rcr3())); |
2731 | #endif | | 2717 | #endif |
2732 | KASSERT((pmap->pm_cpus & cpumask) == 0); | | 2718 | KASSERT((pmap->pm_cpus & cpumask) == 0); |
2733 | KASSERT((pmap->pm_kernel_cpus & cpumask) == 0); | | 2719 | KASSERT((pmap->pm_kernel_cpus & cpumask) == 0); |
2734 | | | 2720 | |
2735 | /* | | 2721 | /* |
2736 | * mark the pmap in use by this processor. again we must | | 2722 | * mark the pmap in use by this processor. again we must |
2737 | * synchronize with TLB shootdown interrupts, so set the | | 2723 | * synchronize with TLB shootdown interrupts, so set the |
2738 | * state VALID first, then register us for shootdown events | | 2724 | * state VALID first, then register us for shootdown events |
2739 | * on this pmap. | | 2725 | * on this pmap. |
2740 | */ | | 2726 | */ |
2741 | | | 2727 | |
2742 | ci->ci_tlbstate = TLBSTATE_VALID; | | 2728 | ci->ci_tlbstate = TLBSTATE_VALID; |
2743 | atomic_or_32(&pmap->pm_cpus, cpumask); | | 2729 | atomic_or_32(&pmap->pm_cpus, cpumask); |
2744 | atomic_or_32(&pmap->pm_kernel_cpus, cpumask); | | 2730 | atomic_or_32(&pmap->pm_kernel_cpus, cpumask); |
2745 | ci->ci_pmap = pmap; | | 2731 | ci->ci_pmap = pmap; |
2746 | | | 2732 | |
2747 | /* | | 2733 | /* |
2748 | * update tss. now that we have registered for invalidations | | 2734 | * update tss. now that we have registered for invalidations |
2749 | * from other CPUs, we're good to load the page tables. | | 2735 | * from other CPUs, we're good to load the page tables. |
2750 | */ | | 2736 | */ |
2751 | #ifdef PAE | | 2737 | #ifdef PAE |
2752 | pcb->pcb_cr3 = ci->ci_pae_l3_pdirpa; | | 2738 | pcb->pcb_cr3 = ci->ci_pae_l3_pdirpa; |
2753 | #else | | 2739 | #else |
2754 | pcb->pcb_cr3 = pmap_pdirpa(pmap, 0); | | 2740 | pcb->pcb_cr3 = pmap_pdirpa(pmap, 0); |
2755 | #endif | | 2741 | #endif |
2756 | | | 2742 | |
2757 | #ifdef i386 | | 2743 | #ifdef i386 |
2758 | #ifndef XEN | | 2744 | #ifndef XEN |
2759 | ci->ci_tss.tss_ldt = pmap->pm_ldt_sel; | | 2745 | ci->ci_tss.tss_ldt = pmap->pm_ldt_sel; |
2760 | ci->ci_tss.tss_cr3 = pcb->pcb_cr3; | | 2746 | ci->ci_tss.tss_cr3 = pcb->pcb_cr3; |
2761 | #endif /* !XEN */ | | 2747 | #endif /* !XEN */ |
2762 | #endif /* i386 */ | | 2748 | #endif /* i386 */ |
2763 | | | 2749 | |
2764 | lldt(pmap->pm_ldt_sel); | | 2750 | lldt(pmap->pm_ldt_sel); |
2765 | | | 2751 | |
2766 | u_int gen = uvm_emap_gen_return(); | | 2752 | u_int gen = uvm_emap_gen_return(); |
2767 | cpu_load_pmap(pmap); | | 2753 | cpu_load_pmap(pmap); |
2768 | uvm_emap_update(gen); | | 2754 | uvm_emap_update(gen); |
2769 | | | 2755 | |
2770 | ci->ci_want_pmapload = 0; | | 2756 | ci->ci_want_pmapload = 0; |
2771 | | | 2757 | |
2772 | /* | | 2758 | /* |
2773 | * we're now running with the new pmap. drop the reference | | 2759 | * we're now running with the new pmap. drop the reference |
2774 | * to the old pmap. if we block, we need to go around again. | | 2760 | * to the old pmap. if we block, we need to go around again. |
2775 | */ | | 2761 | */ |
2776 | | | 2762 | |
2777 | pmap_destroy(oldpmap); | | 2763 | pmap_destroy(oldpmap); |
2778 | if (l->l_ncsw != ncsw) { | | 2764 | if (l->l_ncsw != ncsw) { |
2779 | goto retry; | | 2765 | goto retry; |
2780 | } | | 2766 | } |
2781 | | | 2767 | |
2782 | kpreempt_enable(); | | 2768 | kpreempt_enable(); |
2783 | } | | 2769 | } |
2784 | | | 2770 | |
2785 | /* | | 2771 | /* |
2786 | * pmap_deactivate: deactivate a process' pmap. | | 2772 | * pmap_deactivate: deactivate a process' pmap. |
2787 | * | | 2773 | * |
2788 | * => Must be called with kernel preemption disabled (high IPL is enough). | | 2774 | * => Must be called with kernel preemption disabled (high IPL is enough). |
2789 | */ | | 2775 | */ |
2790 | void | | 2776 | void |
2791 | pmap_deactivate(struct lwp *l) | | 2777 | pmap_deactivate(struct lwp *l) |
2792 | { | | 2778 | { |
2793 | struct pmap *pmap; | | 2779 | struct pmap *pmap; |
2794 | struct cpu_info *ci; | | 2780 | struct cpu_info *ci; |
2795 | | | 2781 | |
2796 | KASSERT(kpreempt_disabled()); | | 2782 | KASSERT(kpreempt_disabled()); |
2797 | | | 2783 | |
2798 | if (l != curlwp) { | | 2784 | if (l != curlwp) { |
2799 | return; | | 2785 | return; |
2800 | } | | 2786 | } |
2801 | | | 2787 | |
2802 | /* | | 2788 | /* |
2803 | * Wait for pending TLB shootdowns to complete. Necessary because | | 2789 | * Wait for pending TLB shootdowns to complete. Necessary because |
2804 | * TLB shootdown state is per-CPU, and the LWP may be coming off | | 2790 | * TLB shootdown state is per-CPU, and the LWP may be coming off |
2805 | * the CPU before it has a chance to call pmap_update(), e.g. due | | 2791 | * the CPU before it has a chance to call pmap_update(), e.g. due |
2806 | * to kernel preemption or blocking routine in between. | | 2792 | * to kernel preemption or blocking routine in between. |
2807 | */ | | 2793 | */ |
2808 | pmap_tlb_shootnow(); | | 2794 | pmap_tlb_shootnow(); |
2809 | | | 2795 | |
2810 | ci = curcpu(); | | 2796 | ci = curcpu(); |
2811 | | | 2797 | |
2812 | if (ci->ci_want_pmapload) { | | 2798 | if (ci->ci_want_pmapload) { |
2813 | /* | | 2799 | /* |
2814 | * ci_want_pmapload means that our pmap is not loaded on | | 2800 | * ci_want_pmapload means that our pmap is not loaded on |
2815 | * the CPU or TLB might be stale. note that pmap_kernel() | | 2801 | * the CPU or TLB might be stale. note that pmap_kernel() |
2816 | * is always considered loaded. | | 2802 | * is always considered loaded. |
2817 | */ | | 2803 | */ |
2818 | KASSERT(vm_map_pmap(&l->l_proc->p_vmspace->vm_map) | | 2804 | KASSERT(vm_map_pmap(&l->l_proc->p_vmspace->vm_map) |
2819 | != pmap_kernel()); | | 2805 | != pmap_kernel()); |
2820 | KASSERT(vm_map_pmap(&l->l_proc->p_vmspace->vm_map) | | 2806 | KASSERT(vm_map_pmap(&l->l_proc->p_vmspace->vm_map) |
2821 | != ci->ci_pmap || ci->ci_tlbstate != TLBSTATE_VALID); | | 2807 | != ci->ci_pmap || ci->ci_tlbstate != TLBSTATE_VALID); |
2822 | | | 2808 | |
2823 | /* | | 2809 | /* |
2824 | * userspace has not been touched. | | 2810 | * userspace has not been touched. |
2825 | * nothing to do here. | | 2811 | * nothing to do here. |
2826 | */ | | 2812 | */ |
2827 | | | 2813 | |
2828 | ci->ci_want_pmapload = 0; | | 2814 | ci->ci_want_pmapload = 0; |
2829 | return; | | 2815 | return; |
2830 | } | | 2816 | } |
2831 | | | 2817 | |
2832 | pmap = vm_map_pmap(&l->l_proc->p_vmspace->vm_map); | | 2818 | pmap = vm_map_pmap(&l->l_proc->p_vmspace->vm_map); |
2833 | | | 2819 | |
2834 | if (pmap == pmap_kernel()) { | | 2820 | if (pmap == pmap_kernel()) { |
2835 | return; | | 2821 | return; |
2836 | } | | 2822 | } |
2837 | | | 2823 | |
2838 | #if defined(XEN) && defined(__x86_64__) | | 2824 | #if defined(XEN) && defined(__x86_64__) |
2839 | KASSERT(pmap_pdirpa(pmap, 0) == ci->ci_xen_current_user_pgd); | | 2825 | KASSERT(pmap_pdirpa(pmap, 0) == ci->ci_xen_current_user_pgd); |
2840 | #elif defined(PAE) | | 2826 | #elif defined(PAE) |
2841 | KASSERT(pmap_pdirpa(pmap, 0) == pmap_pte2pa(ci->ci_pae_l3_pdir[0])); | | 2827 | KASSERT(pmap_pdirpa(pmap, 0) == pmap_pte2pa(ci->ci_pae_l3_pdir[0])); |
2842 | #elif !defined(XEN) | | 2828 | #elif !defined(XEN) |
2843 | KASSERT(pmap_pdirpa(pmap, 0) == pmap_pte2pa(rcr3())); | | 2829 | KASSERT(pmap_pdirpa(pmap, 0) == pmap_pte2pa(rcr3())); |
2844 | #endif | | 2830 | #endif |
2845 | KASSERT(ci->ci_pmap == pmap); | | 2831 | KASSERT(ci->ci_pmap == pmap); |
2846 | | | 2832 | |
2847 | /* | | 2833 | /* |
2848 | * we aren't interested in TLB invalidations for this pmap, | | 2834 | * we aren't interested in TLB invalidations for this pmap, |
2849 | * at least for the time being. | | 2835 | * at least for the time being. |
2850 | */ | | 2836 | */ |
2851 | | | 2837 | |
2852 | KASSERT(ci->ci_tlbstate == TLBSTATE_VALID); | | 2838 | KASSERT(ci->ci_tlbstate == TLBSTATE_VALID); |
2853 | ci->ci_tlbstate = TLBSTATE_LAZY; | | 2839 | ci->ci_tlbstate = TLBSTATE_LAZY; |
2854 | } | | 2840 | } |
2855 | | | 2841 | |
2856 | /* | | 2842 | /* |
2857 | * end of lifecycle functions | | 2843 | * end of lifecycle functions |
2858 | */ | | 2844 | */ |
2859 | | | 2845 | |
2860 | /* | | 2846 | /* |
2861 | * some misc. functions | | 2847 | * some misc. functions |
2862 | */ | | 2848 | */ |
2863 | | | 2849 | |
2864 | int | | 2850 | int |
2865 | pmap_pdes_invalid(vaddr_t va, pd_entry_t * const *pdes, pd_entry_t *lastpde) | | 2851 | pmap_pdes_invalid(vaddr_t va, pd_entry_t * const *pdes, pd_entry_t *lastpde) |
2866 | { | | 2852 | { |
2867 | int i; | | 2853 | int i; |
2868 | unsigned long index; | | 2854 | unsigned long index; |
2869 | pd_entry_t pde; | | 2855 | pd_entry_t pde; |
2870 | | | 2856 | |
2871 | for (i = PTP_LEVELS; i > 1; i--) { | | 2857 | for (i = PTP_LEVELS; i > 1; i--) { |
2872 | index = pl_i(va, i); | | 2858 | index = pl_i(va, i); |
2873 | pde = pdes[i - 2][index]; | | 2859 | pde = pdes[i - 2][index]; |
2874 | if ((pde & PG_V) == 0) | | 2860 | if ((pde & PG_V) == 0) |
2875 | return i; | | 2861 | return i; |
2876 | } | | 2862 | } |
2877 | if (lastpde != NULL) | | 2863 | if (lastpde != NULL) |
2878 | *lastpde = pde; | | 2864 | *lastpde = pde; |
2879 | return 0; | | 2865 | return 0; |
2880 | } | | 2866 | } |
2881 | | | 2867 | |
2882 | /* | | 2868 | /* |
2883 | * pmap_extract: extract a PA for the given VA | | 2869 | * pmap_extract: extract a PA for the given VA |
2884 | */ | | 2870 | */ |
2885 | | | 2871 | |
2886 | bool | | 2872 | bool |
2887 | pmap_extract(struct pmap *pmap, vaddr_t va, paddr_t *pap) | | 2873 | pmap_extract(struct pmap *pmap, vaddr_t va, paddr_t *pap) |
2888 | { | | 2874 | { |
2889 | pt_entry_t *ptes, pte; | | 2875 | pt_entry_t *ptes, pte; |
2890 | pd_entry_t pde; | | 2876 | pd_entry_t pde; |
2891 | pd_entry_t * const *pdes; | | 2877 | pd_entry_t * const *pdes; |
2892 | struct pmap *pmap2; | | 2878 | struct pmap *pmap2; |
2893 | struct cpu_info *ci; | | 2879 | struct cpu_info *ci; |
2894 | paddr_t pa; | | 2880 | paddr_t pa; |
2895 | lwp_t *l; | | 2881 | lwp_t *l; |
2896 | bool hard, rv; | | 2882 | bool hard, rv; |
2897 | | | 2883 | |
2898 | #ifdef __HAVE_DIRECT_MAP | | 2884 | #ifdef __HAVE_DIRECT_MAP |
2899 | if (va >= PMAP_DIRECT_BASE && va < PMAP_DIRECT_END) { | | 2885 | if (va >= PMAP_DIRECT_BASE && va < PMAP_DIRECT_END) { |
2900 | if (pap != NULL) { | | 2886 | if (pap != NULL) { |
2901 | *pap = va - PMAP_DIRECT_BASE; | | 2887 | *pap = va - PMAP_DIRECT_BASE; |
2902 | } | | 2888 | } |
2903 | return true; | | 2889 | return true; |
2904 | } | | 2890 | } |
2905 | #endif | | 2891 | #endif |
2906 | | | 2892 | |
2907 | rv = false; | | 2893 | rv = false; |
2908 | pa = 0; | | 2894 | pa = 0; |
2909 | l = curlwp; | | 2895 | l = curlwp; |
2910 | | | 2896 | |
2911 | KPREEMPT_DISABLE(l); | | 2897 | KPREEMPT_DISABLE(l); |
2912 | ci = l->l_cpu; | | 2898 | ci = l->l_cpu; |