| @@ -1,1041 +1,1043 @@ | | | @@ -1,1041 +1,1043 @@ |
1 | /* $NetBSD: e500_tlb.c,v 1.20 2020/07/06 09:34:16 rin Exp $ */ | | 1 | /* $NetBSD: e500_tlb.c,v 1.21 2020/07/06 10:12:04 rin Exp $ */ |
2 | /*- | | 2 | /*- |
3 | * Copyright (c) 2010, 2011 The NetBSD Foundation, Inc. | | 3 | * Copyright (c) 2010, 2011 The NetBSD Foundation, Inc. |
4 | * All rights reserved. | | 4 | * All rights reserved. |
5 | * | | 5 | * |
6 | * This code is derived from software contributed to The NetBSD Foundation | | 6 | * This code is derived from software contributed to The NetBSD Foundation |
7 | * by Raytheon BBN Technologies Corp and Defense Advanced Research Projects | | 7 | * by Raytheon BBN Technologies Corp and Defense Advanced Research Projects |
8 | * Agency and which was developed by Matt Thomas of 3am Software Foundry. | | 8 | * Agency and which was developed by Matt Thomas of 3am Software Foundry. |
9 | * | | 9 | * |
10 | * This material is based upon work supported by the Defense Advanced Research | | 10 | * This material is based upon work supported by the Defense Advanced Research |
11 | * Projects Agency and Space and Naval Warfare Systems Center, Pacific, under | | 11 | * Projects Agency and Space and Naval Warfare Systems Center, Pacific, under |
12 | * Contract No. N66001-09-C-2073. | | 12 | * Contract No. N66001-09-C-2073. |
13 | * Approved for Public Release, Distribution Unlimited | | 13 | * Approved for Public Release, Distribution Unlimited |
14 | * | | 14 | * |
15 | * Redistribution and use in source and binary forms, with or without | | 15 | * Redistribution and use in source and binary forms, with or without |
16 | * modification, are permitted provided that the following conditions | | 16 | * modification, are permitted provided that the following conditions |
17 | * are met: | | 17 | * are met: |
18 | * 1. Redistributions of source code must retain the above copyright | | 18 | * 1. Redistributions of source code must retain the above copyright |
19 | * notice, this list of conditions and the following disclaimer. | | 19 | * notice, this list of conditions and the following disclaimer. |
20 | * 2. Redistributions in binary form must reproduce the above copyright | | 20 | * 2. Redistributions in binary form must reproduce the above copyright |
21 | * notice, this list of conditions and the following disclaimer in the | | 21 | * notice, this list of conditions and the following disclaimer in the |
22 | * documentation and/or other materials provided with the distribution. | | 22 | * documentation and/or other materials provided with the distribution. |
23 | * | | 23 | * |
24 | * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS | | 24 | * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS |
25 | * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED | | 25 | * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED |
26 | * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR | | 26 | * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR |
27 | * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS | | 27 | * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS |
28 | * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR | | 28 | * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR |
29 | * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF | | 29 | * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF |
30 | * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS | | 30 | * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS |
31 | * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN | | 31 | * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN |
32 | * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) | | 32 | * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) |
33 | * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE | | 33 | * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE |
34 | * POSSIBILITY OF SUCH DAMAGE. | | 34 | * POSSIBILITY OF SUCH DAMAGE. |
35 | */ | | 35 | */ |
36 | | | 36 | |
37 | #define __PMAP_PRIVATE | | 37 | #define __PMAP_PRIVATE |
38 | | | 38 | |
39 | #include <sys/cdefs.h> | | 39 | #include <sys/cdefs.h> |
40 | __KERNEL_RCSID(0, "$NetBSD: e500_tlb.c,v 1.20 2020/07/06 09:34:16 rin Exp $"); | | 40 | __KERNEL_RCSID(0, "$NetBSD: e500_tlb.c,v 1.21 2020/07/06 10:12:04 rin Exp $"); |
41 | | | 41 | |
42 | #ifdef _KERNEL_OPT | | 42 | #ifdef _KERNEL_OPT |
| | | 43 | #include "opt_multiprocessor.h" |
| | | 44 | #include "opt_pmap.h" |
43 | #include "opt_ppcparam.h" | | 45 | #include "opt_ppcparam.h" |
44 | #endif | | 46 | #endif |
45 | | | 47 | |
46 | #include <sys/param.h> | | 48 | #include <sys/param.h> |
47 | | | 49 | |
48 | #include <uvm/uvm_extern.h> | | 50 | #include <uvm/uvm_extern.h> |
49 | | | 51 | |
50 | #include <powerpc/spr.h> | | 52 | #include <powerpc/spr.h> |
51 | #include <powerpc/booke/spr.h> | | 53 | #include <powerpc/booke/spr.h> |
52 | #include <powerpc/booke/cpuvar.h> | | 54 | #include <powerpc/booke/cpuvar.h> |
53 | #include <powerpc/booke/e500reg.h> | | 55 | #include <powerpc/booke/e500reg.h> |
54 | #include <powerpc/booke/e500var.h> | | 56 | #include <powerpc/booke/e500var.h> |
55 | #include <powerpc/booke/pmap.h> | | 57 | #include <powerpc/booke/pmap.h> |
56 | | | 58 | |
57 | struct e500_tlb { | | 59 | struct e500_tlb { |
58 | vaddr_t tlb_va; | | 60 | vaddr_t tlb_va; |
59 | uint32_t tlb_pte; | | 61 | uint32_t tlb_pte; |
60 | uint32_t tlb_asid; | | 62 | uint32_t tlb_asid; |
61 | vsize_t tlb_size; | | 63 | vsize_t tlb_size; |
62 | }; | | 64 | }; |
63 | | | 65 | |
64 | struct e500_hwtlb { | | 66 | struct e500_hwtlb { |
65 | uint32_t hwtlb_mas0; | | 67 | uint32_t hwtlb_mas0; |
66 | uint32_t hwtlb_mas1; | | 68 | uint32_t hwtlb_mas1; |
67 | uint32_t hwtlb_mas2; | | 69 | uint32_t hwtlb_mas2; |
68 | uint32_t hwtlb_mas3; | | 70 | uint32_t hwtlb_mas3; |
69 | }; | | 71 | }; |
70 | | | 72 | |
71 | struct e500_xtlb { | | 73 | struct e500_xtlb { |
72 | struct e500_tlb e_tlb; | | 74 | struct e500_tlb e_tlb; |
73 | struct e500_hwtlb e_hwtlb; | | 75 | struct e500_hwtlb e_hwtlb; |
74 | u_long e_refcnt; | | 76 | u_long e_refcnt; |
75 | }; | | 77 | }; |
76 | | | 78 | |
77 | static struct e500_tlb1 { | | 79 | static struct e500_tlb1 { |
78 | uint32_t tlb1_maxsize; | | 80 | uint32_t tlb1_maxsize; |
79 | uint32_t tlb1_minsize; | | 81 | uint32_t tlb1_minsize; |
80 | u_int tlb1_numentries; | | 82 | u_int tlb1_numentries; |
81 | u_int tlb1_numfree; | | 83 | u_int tlb1_numfree; |
82 | u_int tlb1_freelist[32]; | | 84 | u_int tlb1_freelist[32]; |
83 | struct e500_xtlb tlb1_entries[32]; | | 85 | struct e500_xtlb tlb1_entries[32]; |
84 | } e500_tlb1; | | 86 | } e500_tlb1; |
85 | | | 87 | |
86 | static inline register_t mftlb0cfg(void) __pure; | | 88 | static inline register_t mftlb0cfg(void) __pure; |
87 | static inline register_t mftlb1cfg(void) __pure; | | 89 | static inline register_t mftlb1cfg(void) __pure; |
88 | | | 90 | |
89 | static inline register_t | | 91 | static inline register_t |
90 | mftlb0cfg(void) | | 92 | mftlb0cfg(void) |
91 | { | | 93 | { |
92 | register_t tlb0cfg; | | 94 | register_t tlb0cfg; |
93 | __asm("mfspr %0, %1" : "=r"(tlb0cfg) : "n"(SPR_TLB0CFG)); | | 95 | __asm("mfspr %0, %1" : "=r"(tlb0cfg) : "n"(SPR_TLB0CFG)); |
94 | return tlb0cfg; | | 96 | return tlb0cfg; |
95 | } | | 97 | } |
96 | | | 98 | |
97 | static inline register_t | | 99 | static inline register_t |
98 | mftlb1cfg(void) | | 100 | mftlb1cfg(void) |
99 | { | | 101 | { |
100 | register_t tlb1cfg; | | 102 | register_t tlb1cfg; |
101 | __asm("mfspr %0, %1" : "=r"(tlb1cfg) : "n"(SPR_TLB1CFG)); | | 103 | __asm("mfspr %0, %1" : "=r"(tlb1cfg) : "n"(SPR_TLB1CFG)); |
102 | return tlb1cfg; | | 104 | return tlb1cfg; |
103 | } | | 105 | } |
104 | | | 106 | |
105 | static struct e500_tlb | | 107 | static struct e500_tlb |
106 | hwtlb_to_tlb(const struct e500_hwtlb hwtlb) | | 108 | hwtlb_to_tlb(const struct e500_hwtlb hwtlb) |
107 | { | | 109 | { |
108 | struct e500_tlb tlb; | | 110 | struct e500_tlb tlb; |
109 | register_t prot_mask; | | 111 | register_t prot_mask; |
110 | u_int prot_shift; | | 112 | u_int prot_shift; |
111 | | | 113 | |
112 | tlb.tlb_va = MAS2_EPN & hwtlb.hwtlb_mas2; | | 114 | tlb.tlb_va = MAS2_EPN & hwtlb.hwtlb_mas2; |
113 | tlb.tlb_size = 1024 << (2 * MASX_TSIZE_GET(hwtlb.hwtlb_mas1)); | | 115 | tlb.tlb_size = 1024 << (2 * MASX_TSIZE_GET(hwtlb.hwtlb_mas1)); |
114 | tlb.tlb_asid = MASX_TID_GET(hwtlb.hwtlb_mas1); | | 116 | tlb.tlb_asid = MASX_TID_GET(hwtlb.hwtlb_mas1); |
115 | tlb.tlb_pte = (hwtlb.hwtlb_mas2 & MAS2_WIMGE) | | 117 | tlb.tlb_pte = (hwtlb.hwtlb_mas2 & MAS2_WIMGE) |
116 | | (hwtlb.hwtlb_mas3 & MAS3_RPN); | | 118 | | (hwtlb.hwtlb_mas3 & MAS3_RPN); |
117 | if (hwtlb.hwtlb_mas1 & MAS1_TS) { | | 119 | if (hwtlb.hwtlb_mas1 & MAS1_TS) { |
118 | prot_mask = MAS3_UX|MAS3_UW|MAS3_UR; | | 120 | prot_mask = MAS3_UX|MAS3_UW|MAS3_UR; |
119 | prot_shift = PTE_RWX_SHIFT - 1; | | 121 | prot_shift = PTE_RWX_SHIFT - 1; |
120 | } else { | | 122 | } else { |
121 | prot_mask = MAS3_SX|MAS3_SW|MAS3_SR; | | 123 | prot_mask = MAS3_SX|MAS3_SW|MAS3_SR; |
122 | prot_shift = PTE_RWX_SHIFT; | | 124 | prot_shift = PTE_RWX_SHIFT; |
123 | } | | 125 | } |
124 | tlb.tlb_pte |= (prot_mask & hwtlb.hwtlb_mas3) << prot_shift; | | 126 | tlb.tlb_pte |= (prot_mask & hwtlb.hwtlb_mas3) << prot_shift; |
125 | return tlb; | | 127 | return tlb; |
126 | } | | 128 | } |
127 | | | 129 | |
128 | static inline struct e500_hwtlb | | 130 | static inline struct e500_hwtlb |
129 | hwtlb_read(uint32_t mas0, u_int slot) | | 131 | hwtlb_read(uint32_t mas0, u_int slot) |
130 | { | | 132 | { |
131 | struct e500_hwtlb hwtlb; | | 133 | struct e500_hwtlb hwtlb; |
132 | register_t tlbcfg; | | 134 | register_t tlbcfg; |
133 | | | 135 | |
134 | if (__predict_true(mas0 == MAS0_TLBSEL_TLB0)) { | | 136 | if (__predict_true(mas0 == MAS0_TLBSEL_TLB0)) { |
135 | tlbcfg = mftlb0cfg(); | | 137 | tlbcfg = mftlb0cfg(); |
136 | } else if (mas0 == MAS0_TLBSEL_TLB1) { | | 138 | } else if (mas0 == MAS0_TLBSEL_TLB1) { |
137 | tlbcfg = mftlb1cfg(); | | 139 | tlbcfg = mftlb1cfg(); |
138 | } else { | | 140 | } else { |
139 | panic("%s:%d: unexpected MAS0 %#" PRIx32, | | 141 | panic("%s:%d: unexpected MAS0 %#" PRIx32, |
140 | __func__, __LINE__, mas0); | | 142 | __func__, __LINE__, mas0); |
141 | } | | 143 | } |
142 | | | 144 | |
143 | /* | | 145 | /* |
144 | * ESEL is the way we want to look up. | | 146 | * ESEL is the way we want to look up. |
145 | * If tlbassoc is the same as tlbentries (like in TLB1) then the TLB is | | 147 | * If tlbassoc is the same as tlbentries (like in TLB1) then the TLB is |
146 | * fully associative, the entire slot is placed into ESEL. If tlbassoc | | 148 | * fully associative, the entire slot is placed into ESEL. If tlbassoc |
147 | * is less than the number of tlb entries, the slot is split in two | | 149 | * is less than the number of tlb entries, the slot is split in two |
148 | * fields. Since the TLB is M rows by N ways, the lowers bits are for | | 150 | * fields. Since the TLB is M rows by N ways, the lowers bits are for |
149 | * row (MAS2[EPN]) and the upper for the way (MAS1[ESEL]). | | 151 | * row (MAS2[EPN]) and the upper for the way (MAS1[ESEL]). |
150 | */ | | 152 | */ |
151 | const u_int tlbassoc = TLBCFG_ASSOC(tlbcfg); | | 153 | const u_int tlbassoc = TLBCFG_ASSOC(tlbcfg); |
152 | const u_int tlbentries = TLBCFG_NENTRY(tlbcfg); | | 154 | const u_int tlbentries = TLBCFG_NENTRY(tlbcfg); |
153 | const u_int esel_shift = | | 155 | const u_int esel_shift = |
154 | __builtin_clz(tlbassoc) - __builtin_clz(tlbentries); | | 156 | __builtin_clz(tlbassoc) - __builtin_clz(tlbentries); |
155 | | | 157 | |
156 | /* | | 158 | /* |
157 | * Disable interrupts since we don't want anyone else mucking with | | 159 | * Disable interrupts since we don't want anyone else mucking with |
158 | * the MMU Assist registers | | 160 | * the MMU Assist registers |
159 | */ | | 161 | */ |
160 | const register_t msr = wrtee(0); | | 162 | const register_t msr = wrtee(0); |
161 | const register_t saved_mas0 = mfspr(SPR_MAS0); | | 163 | const register_t saved_mas0 = mfspr(SPR_MAS0); |
162 | mtspr(SPR_MAS0, mas0 | MAS0_ESEL_MAKE(slot >> esel_shift)); | | 164 | mtspr(SPR_MAS0, mas0 | MAS0_ESEL_MAKE(slot >> esel_shift)); |
163 | | | 165 | |
164 | if (__predict_true(tlbassoc > tlbentries)) | | 166 | if (__predict_true(tlbassoc > tlbentries)) |
165 | mtspr(SPR_MAS2, slot << PAGE_SHIFT); | | 167 | mtspr(SPR_MAS2, slot << PAGE_SHIFT); |
166 | | | 168 | |
167 | /* | | 169 | /* |
168 | * Now select the entry and grab its contents. | | 170 | * Now select the entry and grab its contents. |
169 | */ | | 171 | */ |
170 | __asm volatile("tlbre"); | | 172 | __asm volatile("tlbre"); |
171 | | | 173 | |
172 | hwtlb.hwtlb_mas0 = mfspr(SPR_MAS0); | | 174 | hwtlb.hwtlb_mas0 = mfspr(SPR_MAS0); |
173 | hwtlb.hwtlb_mas1 = mfspr(SPR_MAS1); | | 175 | hwtlb.hwtlb_mas1 = mfspr(SPR_MAS1); |
174 | hwtlb.hwtlb_mas2 = mfspr(SPR_MAS2); | | 176 | hwtlb.hwtlb_mas2 = mfspr(SPR_MAS2); |
175 | hwtlb.hwtlb_mas3 = mfspr(SPR_MAS3); | | 177 | hwtlb.hwtlb_mas3 = mfspr(SPR_MAS3); |
176 | | | 178 | |
177 | mtspr(SPR_MAS0, saved_mas0); | | 179 | mtspr(SPR_MAS0, saved_mas0); |
178 | wrtee(msr); /* restore interrupts */ | | 180 | wrtee(msr); /* restore interrupts */ |
179 | | | 181 | |
180 | return hwtlb; | | 182 | return hwtlb; |
181 | } | | 183 | } |
182 | | | 184 | |
183 | static inline void | | 185 | static inline void |
184 | hwtlb_write(const struct e500_hwtlb hwtlb, bool needs_sync) | | 186 | hwtlb_write(const struct e500_hwtlb hwtlb, bool needs_sync) |
185 | { | | 187 | { |
186 | const register_t msr = wrtee(0); | | 188 | const register_t msr = wrtee(0); |
187 | const uint32_t saved_mas0 = mfspr(SPR_MAS0); | | 189 | const uint32_t saved_mas0 = mfspr(SPR_MAS0); |
188 | | | 190 | |
189 | /* | | 191 | /* |
190 | * Need to always write MAS0 and MAS1 | | 192 | * Need to always write MAS0 and MAS1 |
191 | */ | | 193 | */ |
192 | mtspr(SPR_MAS0, hwtlb.hwtlb_mas0); | | 194 | mtspr(SPR_MAS0, hwtlb.hwtlb_mas0); |
193 | mtspr(SPR_MAS1, hwtlb.hwtlb_mas1); | | 195 | mtspr(SPR_MAS1, hwtlb.hwtlb_mas1); |
194 | | | 196 | |
195 | /* | | 197 | /* |
196 | * Only write the VPN/WIMGE if this is in TLB0 or if a valid mapping. | | 198 | * Only write the VPN/WIMGE if this is in TLB0 or if a valid mapping. |
197 | */ | | 199 | */ |
198 | if ((hwtlb.hwtlb_mas0 & MAS0_TLBSEL) == MAS0_TLBSEL_TLB0 | | 200 | if ((hwtlb.hwtlb_mas0 & MAS0_TLBSEL) == MAS0_TLBSEL_TLB0 |
199 | || (hwtlb.hwtlb_mas1 & MAS1_V)) { | | 201 | || (hwtlb.hwtlb_mas1 & MAS1_V)) { |
200 | mtspr(SPR_MAS2, hwtlb.hwtlb_mas2); | | 202 | mtspr(SPR_MAS2, hwtlb.hwtlb_mas2); |
201 | } | | 203 | } |
202 | /* | | 204 | /* |
203 | * Only need to write the RPN/prot if we are dealing with a valid | | 205 | * Only need to write the RPN/prot if we are dealing with a valid |
204 | * mapping. | | 206 | * mapping. |
205 | */ | | 207 | */ |
206 | if (hwtlb.hwtlb_mas1 & MAS1_V) { | | 208 | if (hwtlb.hwtlb_mas1 & MAS1_V) { |
207 | mtspr(SPR_MAS3, hwtlb.hwtlb_mas3); | | 209 | mtspr(SPR_MAS3, hwtlb.hwtlb_mas3); |
208 | //mtspr(SPR_MAS7, 0); | | 210 | //mtspr(SPR_MAS7, 0); |
209 | } | | 211 | } |
210 | | | 212 | |
211 | #if 0 | | 213 | #if 0 |
212 | printf("%s->[%x,%x,%x,%x]\n", | | 214 | printf("%s->[%x,%x,%x,%x]\n", |
213 | __func__, | | 215 | __func__, |
214 | hwtlb.hwtlb_mas0, hwtlb.hwtlb_mas1, | | 216 | hwtlb.hwtlb_mas0, hwtlb.hwtlb_mas1, |
215 | hwtlb.hwtlb_mas2, hwtlb.hwtlb_mas3); | | 217 | hwtlb.hwtlb_mas2, hwtlb.hwtlb_mas3); |
216 | #endif | | 218 | #endif |
217 | __asm volatile("tlbwe"); | | 219 | __asm volatile("tlbwe"); |
218 | if (needs_sync) { | | 220 | if (needs_sync) { |
219 | __asm volatile("tlbsync\n\tisync\n\tsync"); | | 221 | __asm volatile("tlbsync\n\tisync\n\tsync"); |
220 | } | | 222 | } |
221 | | | 223 | |
222 | mtspr(SPR_MAS0, saved_mas0); | | 224 | mtspr(SPR_MAS0, saved_mas0); |
223 | wrtee(msr); | | 225 | wrtee(msr); |
224 | } | | 226 | } |
225 | | | 227 | |
226 | static struct e500_hwtlb | | 228 | static struct e500_hwtlb |
227 | tlb_to_hwtlb(const struct e500_tlb tlb) | | 229 | tlb_to_hwtlb(const struct e500_tlb tlb) |
228 | { | | 230 | { |
229 | struct e500_hwtlb hwtlb; | | 231 | struct e500_hwtlb hwtlb; |
230 | | | 232 | |
231 | KASSERT(trunc_page(tlb.tlb_va) == tlb.tlb_va); | | 233 | KASSERT(trunc_page(tlb.tlb_va) == tlb.tlb_va); |
232 | KASSERT(tlb.tlb_size != 0); | | 234 | KASSERT(tlb.tlb_size != 0); |
233 | KASSERT((tlb.tlb_size & (tlb.tlb_size - 1)) == 0); | | 235 | KASSERT((tlb.tlb_size & (tlb.tlb_size - 1)) == 0); |
234 | const uint32_t prot_mask = tlb.tlb_pte & PTE_RWX_MASK; | | 236 | const uint32_t prot_mask = tlb.tlb_pte & PTE_RWX_MASK; |
235 | if (__predict_true(tlb.tlb_size == PAGE_SIZE)) { | | 237 | if (__predict_true(tlb.tlb_size == PAGE_SIZE)) { |
236 | hwtlb.hwtlb_mas0 = 0; | | 238 | hwtlb.hwtlb_mas0 = 0; |
237 | hwtlb.hwtlb_mas1 = MAS1_V | MASX_TSIZE_MAKE(1); | | 239 | hwtlb.hwtlb_mas1 = MAS1_V | MASX_TSIZE_MAKE(1); |
238 | /* | | 240 | /* |
239 | * A non-zero ASID means this is a user page so mark it as | | 241 | * A non-zero ASID means this is a user page so mark it as |
240 | * being in the user's address space. | | 242 | * being in the user's address space. |
241 | */ | | 243 | */ |
242 | if (tlb.tlb_asid) { | | 244 | if (tlb.tlb_asid) { |
243 | hwtlb.hwtlb_mas1 |= MAS1_TS | | 245 | hwtlb.hwtlb_mas1 |= MAS1_TS |
244 | | MASX_TID_MAKE(tlb.tlb_asid); | | 246 | | MASX_TID_MAKE(tlb.tlb_asid); |
245 | hwtlb.hwtlb_mas3 = (prot_mask >> (PTE_RWX_SHIFT - 1)) | | 247 | hwtlb.hwtlb_mas3 = (prot_mask >> (PTE_RWX_SHIFT - 1)) |
246 | | ((prot_mask & ~PTE_xX) >> PTE_RWX_SHIFT); | | 248 | | ((prot_mask & ~PTE_xX) >> PTE_RWX_SHIFT); |
247 | KASSERT(prot_mask & PTE_xR); | | 249 | KASSERT(prot_mask & PTE_xR); |
248 | KASSERT(hwtlb.hwtlb_mas3 & MAS3_UR); | | 250 | KASSERT(hwtlb.hwtlb_mas3 & MAS3_UR); |
249 | CTASSERT(MAS3_UR == (PTE_xR >> (PTE_RWX_SHIFT - 1))); | | 251 | CTASSERT(MAS3_UR == (PTE_xR >> (PTE_RWX_SHIFT - 1))); |
250 | CTASSERT(MAS3_SR == (PTE_xR >> PTE_RWX_SHIFT)); | | 252 | CTASSERT(MAS3_SR == (PTE_xR >> PTE_RWX_SHIFT)); |
251 | } else { | | 253 | } else { |
252 | hwtlb.hwtlb_mas3 = prot_mask >> PTE_RWX_SHIFT; | | 254 | hwtlb.hwtlb_mas3 = prot_mask >> PTE_RWX_SHIFT; |
253 | } | | 255 | } |
254 | if (tlb.tlb_pte & PTE_UNMODIFIED) | | 256 | if (tlb.tlb_pte & PTE_UNMODIFIED) |
255 | hwtlb.hwtlb_mas3 &= ~(MAS3_UW|MAS3_SW); | | 257 | hwtlb.hwtlb_mas3 &= ~(MAS3_UW|MAS3_SW); |
256 | if (tlb.tlb_pte & PTE_UNSYNCED) | | 258 | if (tlb.tlb_pte & PTE_UNSYNCED) |
257 | hwtlb.hwtlb_mas3 &= ~(MAS3_UX|MAS3_SX); | | 259 | hwtlb.hwtlb_mas3 &= ~(MAS3_UX|MAS3_SX); |
258 | } else { | | 260 | } else { |
259 | KASSERT(tlb.tlb_asid == 0); | | 261 | KASSERT(tlb.tlb_asid == 0); |
260 | KASSERT((tlb.tlb_size & 0xaaaaa7ff) == 0); | | 262 | KASSERT((tlb.tlb_size & 0xaaaaa7ff) == 0); |
261 | u_int cntlz = __builtin_clz(tlb.tlb_size); | | 263 | u_int cntlz = __builtin_clz(tlb.tlb_size); |
262 | KASSERT(cntlz & 1); | | 264 | KASSERT(cntlz & 1); |
263 | KASSERT(cntlz <= 19); | | 265 | KASSERT(cntlz <= 19); |
264 | hwtlb.hwtlb_mas0 = MAS0_TLBSEL_TLB1; | | 266 | hwtlb.hwtlb_mas0 = MAS0_TLBSEL_TLB1; |
265 | /* | | 267 | /* |
266 | * TSIZE is defined (4^TSIZE) Kbytes except a TSIZE of 0 is not | | 268 | * TSIZE is defined (4^TSIZE) Kbytes except a TSIZE of 0 is not |
267 | * allowed. So 1K would be 0x00000400 giving 21 leading zero | | 269 | * allowed. So 1K would be 0x00000400 giving 21 leading zero |
268 | * bits. Subtracting the leading number of zero bits from 21 | | 270 | * bits. Subtracting the leading number of zero bits from 21 |
269 | * and dividing by 2 gives us the number that the MMU wants. | | 271 | * and dividing by 2 gives us the number that the MMU wants. |
270 | */ | | 272 | */ |
271 | hwtlb.hwtlb_mas1 = MASX_TSIZE_MAKE(((31 - 10) - cntlz) / 2) | | 273 | hwtlb.hwtlb_mas1 = MASX_TSIZE_MAKE(((31 - 10) - cntlz) / 2) |
272 | | MAS1_IPROT | MAS1_V; | | 274 | | MAS1_IPROT | MAS1_V; |
273 | hwtlb.hwtlb_mas3 = prot_mask >> PTE_RWX_SHIFT; | | 275 | hwtlb.hwtlb_mas3 = prot_mask >> PTE_RWX_SHIFT; |
274 | } | | 276 | } |
275 | /* We are done with MAS1, on to MAS2 ... */ | | 277 | /* We are done with MAS1, on to MAS2 ... */ |
276 | hwtlb.hwtlb_mas2 = tlb.tlb_va | (tlb.tlb_pte & PTE_WIMGE_MASK); | | 278 | hwtlb.hwtlb_mas2 = tlb.tlb_va | (tlb.tlb_pte & PTE_WIMGE_MASK); |
277 | hwtlb.hwtlb_mas3 |= tlb.tlb_pte & PTE_RPN_MASK; | | 279 | hwtlb.hwtlb_mas3 |= tlb.tlb_pte & PTE_RPN_MASK; |
278 | | | 280 | |
279 | return hwtlb; | | 281 | return hwtlb; |
280 | } | | 282 | } |
281 | | | 283 | |
282 | void * | | 284 | void * |
283 | e500_tlb1_fetch(size_t slot) | | 285 | e500_tlb1_fetch(size_t slot) |
284 | { | | 286 | { |
285 | struct e500_tlb1 * const tlb1 = &e500_tlb1; | | 287 | struct e500_tlb1 * const tlb1 = &e500_tlb1; |
286 | | | 288 | |
287 | return &tlb1->tlb1_entries[slot].e_hwtlb; | | 289 | return &tlb1->tlb1_entries[slot].e_hwtlb; |
288 | } | | 290 | } |
289 | | | 291 | |
290 | void | | 292 | void |
291 | e500_tlb1_sync(void) | | 293 | e500_tlb1_sync(void) |
292 | { | | 294 | { |
293 | struct e500_tlb1 * const tlb1 = &e500_tlb1; | | 295 | struct e500_tlb1 * const tlb1 = &e500_tlb1; |
294 | for (u_int slot = 1; slot < tlb1->tlb1_numentries; slot++) { | | 296 | for (u_int slot = 1; slot < tlb1->tlb1_numentries; slot++) { |
295 | const struct e500_hwtlb * const new_hwtlb = | | 297 | const struct e500_hwtlb * const new_hwtlb = |
296 | &tlb1->tlb1_entries[slot].e_hwtlb; | | 298 | &tlb1->tlb1_entries[slot].e_hwtlb; |
297 | const struct e500_hwtlb old_hwtlb = | | 299 | const struct e500_hwtlb old_hwtlb = |
298 | hwtlb_read(MAS0_TLBSEL_TLB1, slot); | | 300 | hwtlb_read(MAS0_TLBSEL_TLB1, slot); |
299 | #define CHANGED(n,o,f) ((n)->f != (o).f) | | 301 | #define CHANGED(n,o,f) ((n)->f != (o).f) |
300 | bool mas1_changed_p = CHANGED(new_hwtlb, old_hwtlb, hwtlb_mas1); | | 302 | bool mas1_changed_p = CHANGED(new_hwtlb, old_hwtlb, hwtlb_mas1); |
301 | bool mas2_changed_p = CHANGED(new_hwtlb, old_hwtlb, hwtlb_mas2); | | 303 | bool mas2_changed_p = CHANGED(new_hwtlb, old_hwtlb, hwtlb_mas2); |
302 | bool mas3_changed_p = CHANGED(new_hwtlb, old_hwtlb, hwtlb_mas3); | | 304 | bool mas3_changed_p = CHANGED(new_hwtlb, old_hwtlb, hwtlb_mas3); |
303 | #undef CHANGED | | 305 | #undef CHANGED |
304 | bool new_valid_p = (new_hwtlb->hwtlb_mas1 & MAS1_V) != 0; | | 306 | bool new_valid_p = (new_hwtlb->hwtlb_mas1 & MAS1_V) != 0; |
305 | bool old_valid_p = (old_hwtlb.hwtlb_mas1 & MAS1_V) != 0; | | 307 | bool old_valid_p = (old_hwtlb.hwtlb_mas1 & MAS1_V) != 0; |
306 | if ((new_valid_p || old_valid_p) | | 308 | if ((new_valid_p || old_valid_p) |
307 | && (mas1_changed_p | | 309 | && (mas1_changed_p |
308 | || (new_valid_p | | 310 | || (new_valid_p |
309 | && (mas2_changed_p || mas3_changed_p)))) | | 311 | && (mas2_changed_p || mas3_changed_p)))) |
310 | hwtlb_write(*new_hwtlb, true); | | 312 | hwtlb_write(*new_hwtlb, true); |
311 | } | | 313 | } |
312 | } | | 314 | } |
313 | | | 315 | |
314 | static int | | 316 | static int |
315 | e500_alloc_tlb1_entry(void) | | 317 | e500_alloc_tlb1_entry(void) |
316 | { | | 318 | { |
317 | struct e500_tlb1 * const tlb1 = &e500_tlb1; | | 319 | struct e500_tlb1 * const tlb1 = &e500_tlb1; |
318 | | | 320 | |
319 | if (tlb1->tlb1_numfree == 0) | | 321 | if (tlb1->tlb1_numfree == 0) |
320 | return -1; | | 322 | return -1; |
321 | const u_int slot = tlb1->tlb1_freelist[--tlb1->tlb1_numfree]; | | 323 | const u_int slot = tlb1->tlb1_freelist[--tlb1->tlb1_numfree]; |
322 | KASSERT((tlb1->tlb1_entries[slot].e_hwtlb.hwtlb_mas1 & MAS1_V) == 0); | | 324 | KASSERT((tlb1->tlb1_entries[slot].e_hwtlb.hwtlb_mas1 & MAS1_V) == 0); |
323 | tlb1->tlb1_entries[slot].e_hwtlb.hwtlb_mas0 = | | 325 | tlb1->tlb1_entries[slot].e_hwtlb.hwtlb_mas0 = |
324 | MAS0_TLBSEL_TLB1 | __SHIFTIN(slot, MAS0_ESEL); | | 326 | MAS0_TLBSEL_TLB1 | __SHIFTIN(slot, MAS0_ESEL); |
325 | return (int)slot; | | 327 | return (int)slot; |
326 | } | | 328 | } |
327 | | | 329 | |
328 | static void | | 330 | static void |
329 | e500_free_tlb1_entry(struct e500_xtlb *xtlb, u_int slot, bool needs_sync) | | 331 | e500_free_tlb1_entry(struct e500_xtlb *xtlb, u_int slot, bool needs_sync) |
330 | { | | 332 | { |
331 | struct e500_tlb1 * const tlb1 = &e500_tlb1; | | 333 | struct e500_tlb1 * const tlb1 = &e500_tlb1; |
332 | KASSERT(slot < tlb1->tlb1_numentries); | | 334 | KASSERT(slot < tlb1->tlb1_numentries); |
333 | KASSERT(&tlb1->tlb1_entries[slot] == xtlb); | | 335 | KASSERT(&tlb1->tlb1_entries[slot] == xtlb); |
334 | | | 336 | |
335 | KASSERT(xtlb->e_hwtlb.hwtlb_mas0 == (MAS0_TLBSEL_TLB1|__SHIFTIN(slot, MAS0_ESEL))); | | 337 | KASSERT(xtlb->e_hwtlb.hwtlb_mas0 == (MAS0_TLBSEL_TLB1|__SHIFTIN(slot, MAS0_ESEL))); |
336 | xtlb->e_hwtlb.hwtlb_mas1 &= ~(MAS1_V|MAS1_IPROT); | | 338 | xtlb->e_hwtlb.hwtlb_mas1 &= ~(MAS1_V|MAS1_IPROT); |
337 | hwtlb_write(xtlb->e_hwtlb, needs_sync); | | 339 | hwtlb_write(xtlb->e_hwtlb, needs_sync); |
338 | | | 340 | |
339 | const register_t msr = wrtee(0); | | 341 | const register_t msr = wrtee(0); |
340 | tlb1->tlb1_freelist[tlb1->tlb1_numfree++] = slot; | | 342 | tlb1->tlb1_freelist[tlb1->tlb1_numfree++] = slot; |
341 | wrtee(msr); | | 343 | wrtee(msr); |
342 | } | | 344 | } |
343 | | | 345 | |
344 | static tlb_asid_t | | 346 | static tlb_asid_t |
345 | e500_tlb_get_asid(void) | | 347 | e500_tlb_get_asid(void) |
346 | { | | 348 | { |
347 | return mfspr(SPR_PID0); | | 349 | return mfspr(SPR_PID0); |
348 | } | | 350 | } |
349 | | | 351 | |
350 | static void | | 352 | static void |
351 | e500_tlb_set_asid(tlb_asid_t asid) | | 353 | e500_tlb_set_asid(tlb_asid_t asid) |
352 | { | | 354 | { |
353 | mtspr(SPR_PID0, asid); | | 355 | mtspr(SPR_PID0, asid); |
354 | } | | 356 | } |
355 | | | 357 | |
356 | static void | | 358 | static void |
357 | e500_tlb_invalidate_all(void) | | 359 | e500_tlb_invalidate_all(void) |
358 | { | | 360 | { |
359 | /* | | 361 | /* |
360 | * This does a flash invalidate of all entries in TLB0. | | 362 | * This does a flash invalidate of all entries in TLB0. |
361 | * We don't touch TLB1 since we don't expect those to be volatile. | | 363 | * We don't touch TLB1 since we don't expect those to be volatile. |
362 | */ | | 364 | */ |
363 | #if 1 | | 365 | #if 1 |
364 | __asm volatile("tlbivax\t0, %0" :: "b"(4)); /* INV_ALL */ | | 366 | __asm volatile("tlbivax\t0, %0" :: "b"(4)); /* INV_ALL */ |
365 | __asm volatile("tlbsync\n\tisync\n\tsync"); | | 367 | __asm volatile("tlbsync\n\tisync\n\tsync"); |
366 | #else | | 368 | #else |
367 | mtspr(SPR_MMUCSR0, MMUCSR0_TLB0_FI); | | 369 | mtspr(SPR_MMUCSR0, MMUCSR0_TLB0_FI); |
368 | while (mfspr(SPR_MMUCSR0) != 0) | | 370 | while (mfspr(SPR_MMUCSR0) != 0) |
369 | ; | | 371 | ; |
370 | #endif | | 372 | #endif |
371 | } | | 373 | } |
372 | | | 374 | |
373 | static void | | 375 | static void |
374 | e500_tlb_invalidate_globals(void) | | 376 | e500_tlb_invalidate_globals(void) |
375 | { | | 377 | { |
376 | #if defined(MULTIPROCESSOR) | | 378 | #if defined(MULTIPROCESSOR) |
377 | e500_tlb_invalidate_all(); | | 379 | e500_tlb_invalidate_all(); |
378 | #else /* !MULTIPROCESSOR */ | | 380 | #else /* !MULTIPROCESSOR */ |
379 | const size_t tlbassoc = TLBCFG_ASSOC(mftlb0cfg()); | | 381 | const size_t tlbassoc = TLBCFG_ASSOC(mftlb0cfg()); |
380 | const size_t tlbentries = TLBCFG_NENTRY(mftlb0cfg()); | | 382 | const size_t tlbentries = TLBCFG_NENTRY(mftlb0cfg()); |
381 | const size_t max_epn = (tlbentries / tlbassoc) << PAGE_SHIFT; | | 383 | const size_t max_epn = (tlbentries / tlbassoc) << PAGE_SHIFT; |
382 | const vaddr_t kstack_lo = (uintptr_t)curlwp->l_addr; | | 384 | const vaddr_t kstack_lo = (uintptr_t)curlwp->l_addr; |
383 | const vaddr_t kstack_hi = kstack_lo + USPACE - 1; | | 385 | const vaddr_t kstack_hi = kstack_lo + USPACE - 1; |
384 | const vaddr_t epn_kstack_lo = kstack_lo & (max_epn - 1); | | 386 | const vaddr_t epn_kstack_lo = kstack_lo & (max_epn - 1); |
385 | const vaddr_t epn_kstack_hi = kstack_hi & (max_epn - 1); | | 387 | const vaddr_t epn_kstack_hi = kstack_hi & (max_epn - 1); |
386 | | | 388 | |
387 | const register_t msr = wrtee(0); | | 389 | const register_t msr = wrtee(0); |
388 | for (size_t assoc = 0; assoc < tlbassoc; assoc++) { | | 390 | for (size_t assoc = 0; assoc < tlbassoc; assoc++) { |
389 | mtspr(SPR_MAS0, MAS0_ESEL_MAKE(assoc) | MAS0_TLBSEL_TLB0); | | 391 | mtspr(SPR_MAS0, MAS0_ESEL_MAKE(assoc) | MAS0_TLBSEL_TLB0); |
390 | for (size_t epn = 0; epn < max_epn; epn += PAGE_SIZE) { | | 392 | for (size_t epn = 0; epn < max_epn; epn += PAGE_SIZE) { |
391 | mtspr(SPR_MAS2, epn); | | 393 | mtspr(SPR_MAS2, epn); |
392 | __asm volatile("tlbre"); | | 394 | __asm volatile("tlbre"); |
393 | uint32_t mas1 = mfspr(SPR_MAS1); | | 395 | uint32_t mas1 = mfspr(SPR_MAS1); |
394 | | | 396 | |
395 | /* | | 397 | /* |
396 | * Make sure this is a valid kernel entry first. | | 398 | * Make sure this is a valid kernel entry first. |
397 | */ | | 399 | */ |
398 | if ((mas1 & (MAS1_V|MAS1_TID|MAS1_TS)) != MAS1_V) | | 400 | if ((mas1 & (MAS1_V|MAS1_TID|MAS1_TS)) != MAS1_V) |
399 | continue; | | 401 | continue; |
400 | | | 402 | |
401 | /* | | 403 | /* |
402 | * We have a valid kernel TLB entry. But if it matches | | 404 | * We have a valid kernel TLB entry. But if it matches |
403 | * the stack we are currently running on, it would | | 405 | * the stack we are currently running on, it would |
404 | * unwise to invalidate it. First see if the epn | | 406 | * unwise to invalidate it. First see if the epn |
405 | * overlaps the stack. If it does then get the | | 407 | * overlaps the stack. If it does then get the |
406 | * VA and see if it really is part of the stack. | | 408 | * VA and see if it really is part of the stack. |
407 | */ | | 409 | */ |
408 | if (epn_kstack_lo < epn_kstack_hi | | 410 | if (epn_kstack_lo < epn_kstack_hi |
409 | ? (epn_kstack_lo <= epn && epn <= epn_kstack_hi) | | 411 | ? (epn_kstack_lo <= epn && epn <= epn_kstack_hi) |
410 | : (epn <= epn_kstack_hi || epn_kstack_lo <= epn)) { | | 412 | : (epn <= epn_kstack_hi || epn_kstack_lo <= epn)) { |
411 | const uint32_t mas2_epn = | | 413 | const uint32_t mas2_epn = |
412 | mfspr(SPR_MAS2) & MAS2_EPN; | | 414 | mfspr(SPR_MAS2) & MAS2_EPN; |
413 | if (kstack_lo <= mas2_epn | | 415 | if (kstack_lo <= mas2_epn |
414 | && mas2_epn <= kstack_hi) | | 416 | && mas2_epn <= kstack_hi) |
415 | continue; | | 417 | continue; |
416 | } | | 418 | } |
417 | mtspr(SPR_MAS1, mas1 ^ MAS1_V); | | 419 | mtspr(SPR_MAS1, mas1 ^ MAS1_V); |
418 | __asm volatile("tlbwe"); | | 420 | __asm volatile("tlbwe"); |
419 | } | | 421 | } |
420 | } | | 422 | } |
421 | __asm volatile("isync\n\tsync"); | | 423 | __asm volatile("isync\n\tsync"); |
422 | wrtee(msr); | | 424 | wrtee(msr); |
423 | #endif /* MULTIPROCESSOR */ | | 425 | #endif /* MULTIPROCESSOR */ |
424 | } | | 426 | } |
425 | | | 427 | |
426 | static void | | 428 | static void |
427 | e500_tlb_invalidate_asids(tlb_asid_t asid_lo, tlb_asid_t asid_hi) | | 429 | e500_tlb_invalidate_asids(tlb_asid_t asid_lo, tlb_asid_t asid_hi) |
428 | { | | 430 | { |
429 | #if defined(MULTIPROCESSOR) | | 431 | #if defined(MULTIPROCESSOR) |
430 | e500_tlb_invalidate_all(); | | 432 | e500_tlb_invalidate_all(); |
431 | #else /* !MULTIPROCESSOR */ | | 433 | #else /* !MULTIPROCESSOR */ |
432 | const size_t tlbassoc = TLBCFG_ASSOC(mftlb0cfg()); | | 434 | const size_t tlbassoc = TLBCFG_ASSOC(mftlb0cfg()); |
433 | const size_t tlbentries = TLBCFG_NENTRY(mftlb0cfg()); | | 435 | const size_t tlbentries = TLBCFG_NENTRY(mftlb0cfg()); |
434 | const size_t max_epn = (tlbentries / tlbassoc) << PAGE_SHIFT; | | 436 | const size_t max_epn = (tlbentries / tlbassoc) << PAGE_SHIFT; |
435 | | | 437 | |
436 | asid_lo = __SHIFTIN(asid_lo, MAS1_TID); | | 438 | asid_lo = __SHIFTIN(asid_lo, MAS1_TID); |
437 | asid_hi = __SHIFTIN(asid_hi, MAS1_TID); | | 439 | asid_hi = __SHIFTIN(asid_hi, MAS1_TID); |
438 | | | 440 | |
439 | const register_t msr = wrtee(0); | | 441 | const register_t msr = wrtee(0); |
440 | for (size_t assoc = 0; assoc < tlbassoc; assoc++) { | | 442 | for (size_t assoc = 0; assoc < tlbassoc; assoc++) { |
441 | mtspr(SPR_MAS0, MAS0_ESEL_MAKE(assoc) | MAS0_TLBSEL_TLB0); | | 443 | mtspr(SPR_MAS0, MAS0_ESEL_MAKE(assoc) | MAS0_TLBSEL_TLB0); |
442 | for (size_t epn = 0; epn < max_epn; epn += PAGE_SIZE) { | | 444 | for (size_t epn = 0; epn < max_epn; epn += PAGE_SIZE) { |
443 | mtspr(SPR_MAS2, epn); | | 445 | mtspr(SPR_MAS2, epn); |
444 | __asm volatile("tlbre"); | | 446 | __asm volatile("tlbre"); |
445 | const uint32_t mas1 = mfspr(SPR_MAS1); | | 447 | const uint32_t mas1 = mfspr(SPR_MAS1); |
446 | /* | | 448 | /* |
447 | * If this is a valid entry for AS space 1 and | | 449 | * If this is a valid entry for AS space 1 and |
448 | * its asid matches the constraints of the caller, | | 450 | * its asid matches the constraints of the caller, |
449 | * clear its valid bit. | | 451 | * clear its valid bit. |
450 | */ | | 452 | */ |
451 | if ((mas1 & (MAS1_V|MAS1_TS)) == (MAS1_V|MAS1_TS) | | 453 | if ((mas1 & (MAS1_V|MAS1_TS)) == (MAS1_V|MAS1_TS) |
452 | && asid_lo <= (mas1 & MAS1_TID) | | 454 | && asid_lo <= (mas1 & MAS1_TID) |
453 | && (mas1 & MAS1_TID) <= asid_hi) { | | 455 | && (mas1 & MAS1_TID) <= asid_hi) { |
454 | mtspr(SPR_MAS1, mas1 ^ MAS1_V); | | 456 | mtspr(SPR_MAS1, mas1 ^ MAS1_V); |
455 | #if 0 | | 457 | #if 0 |
456 | printf("%s[%zu,%zu]->[%x]\n", | | 458 | printf("%s[%zu,%zu]->[%x]\n", |
457 | __func__, assoc, epn, mas1); | | 459 | __func__, assoc, epn, mas1); |
458 | #endif | | 460 | #endif |
459 | __asm volatile("tlbwe"); | | 461 | __asm volatile("tlbwe"); |
460 | } | | 462 | } |
461 | } | | 463 | } |
462 | } | | 464 | } |
463 | __asm volatile("isync\n\tsync"); | | 465 | __asm volatile("isync\n\tsync"); |
464 | wrtee(msr); | | 466 | wrtee(msr); |
465 | #endif /* MULTIPROCESSOR */ | | 467 | #endif /* MULTIPROCESSOR */ |
466 | } | | 468 | } |
467 | | | 469 | |
468 | static u_int | | 470 | static u_int |
469 | e500_tlb_record_asids(u_long *bitmap, tlb_asid_t asid_max) | | 471 | e500_tlb_record_asids(u_long *bitmap, tlb_asid_t asid_max) |
470 | { | | 472 | { |
471 | const size_t tlbassoc = TLBCFG_ASSOC(mftlb0cfg()); | | 473 | const size_t tlbassoc = TLBCFG_ASSOC(mftlb0cfg()); |
472 | const size_t tlbentries = TLBCFG_NENTRY(mftlb0cfg()); | | 474 | const size_t tlbentries = TLBCFG_NENTRY(mftlb0cfg()); |
473 | const size_t max_epn = (tlbentries / tlbassoc) << PAGE_SHIFT; | | 475 | const size_t max_epn = (tlbentries / tlbassoc) << PAGE_SHIFT; |
474 | const size_t nbits = 8 * sizeof(bitmap[0]); | | 476 | const size_t nbits = 8 * sizeof(bitmap[0]); |
475 | u_int found = 0; | | 477 | u_int found = 0; |
476 | | | 478 | |
477 | const register_t msr = wrtee(0); | | 479 | const register_t msr = wrtee(0); |
478 | for (size_t assoc = 0; assoc < tlbassoc; assoc++) { | | 480 | for (size_t assoc = 0; assoc < tlbassoc; assoc++) { |
479 | mtspr(SPR_MAS0, MAS0_ESEL_MAKE(assoc) | MAS0_TLBSEL_TLB0); | | 481 | mtspr(SPR_MAS0, MAS0_ESEL_MAKE(assoc) | MAS0_TLBSEL_TLB0); |
480 | for (size_t epn = 0; epn < max_epn; epn += PAGE_SIZE) { | | 482 | for (size_t epn = 0; epn < max_epn; epn += PAGE_SIZE) { |
481 | mtspr(SPR_MAS2, epn); | | 483 | mtspr(SPR_MAS2, epn); |
482 | __asm volatile("tlbre"); | | 484 | __asm volatile("tlbre"); |
483 | const uint32_t mas1 = mfspr(SPR_MAS1); | | 485 | const uint32_t mas1 = mfspr(SPR_MAS1); |
484 | /* | | 486 | /* |
485 | * If this is a valid entry for AS space 1 and | | 487 | * If this is a valid entry for AS space 1 and |
486 | * its asid matches the constraints of the caller, | | 488 | * its asid matches the constraints of the caller, |
487 | * clear its valid bit. | | 489 | * clear its valid bit. |
488 | */ | | 490 | */ |
489 | if ((mas1 & (MAS1_V|MAS1_TS)) == (MAS1_V|MAS1_TS)) { | | 491 | if ((mas1 & (MAS1_V|MAS1_TS)) == (MAS1_V|MAS1_TS)) { |
490 | const uint32_t asid = MASX_TID_GET(mas1); | | 492 | const uint32_t asid = MASX_TID_GET(mas1); |
491 | const u_int i = asid / nbits; | | 493 | const u_int i = asid / nbits; |
492 | const u_long mask = 1UL << (asid & (nbits - 1)); | | 494 | const u_long mask = 1UL << (asid & (nbits - 1)); |
493 | if ((bitmap[i] & mask) == 0) { | | 495 | if ((bitmap[i] & mask) == 0) { |
494 | bitmap[i] |= mask; | | 496 | bitmap[i] |= mask; |
495 | found++; | | 497 | found++; |
496 | } | | 498 | } |
497 | } | | 499 | } |
498 | } | | 500 | } |
499 | } | | 501 | } |
500 | wrtee(msr); | | 502 | wrtee(msr); |
501 | | | 503 | |
502 | return found; | | 504 | return found; |
503 | } | | 505 | } |
504 | | | 506 | |
505 | static void | | 507 | static void |
506 | e500_tlb_invalidate_addr(vaddr_t va, tlb_asid_t asid) | | 508 | e500_tlb_invalidate_addr(vaddr_t va, tlb_asid_t asid) |
507 | { | | 509 | { |
508 | KASSERT((va & PAGE_MASK) == 0); | | 510 | KASSERT((va & PAGE_MASK) == 0); |
509 | /* | | 511 | /* |
510 | * Bits 60 & 61 have meaning | | 512 | * Bits 60 & 61 have meaning |
511 | */ | | 513 | */ |
512 | if (asid == KERNEL_PID) { | | 514 | if (asid == KERNEL_PID) { |
513 | /* | | 515 | /* |
514 | * For data accesses, the context-synchronizing instruction | | 516 | * For data accesses, the context-synchronizing instruction |
515 | * before tlbwe or tlbivax ensures that all memory accesses | | 517 | * before tlbwe or tlbivax ensures that all memory accesses |
516 | * due to preceding instructions have completed to a point | | 518 | * due to preceding instructions have completed to a point |
517 | * at which they have reported all exceptions they will cause. | | 519 | * at which they have reported all exceptions they will cause. |
518 | */ | | 520 | */ |
519 | __asm volatile("isync"); | | 521 | __asm volatile("isync"); |
520 | } | | 522 | } |
521 | __asm volatile("tlbivax\t0, %0" :: "b"(va)); | | 523 | __asm volatile("tlbivax\t0, %0" :: "b"(va)); |
522 | __asm volatile("tlbsync"); | | 524 | __asm volatile("tlbsync"); |
523 | __asm volatile("tlbsync"); /* Why? */ | | 525 | __asm volatile("tlbsync"); /* Why? */ |
524 | if (asid == KERNEL_PID) { | | 526 | if (asid == KERNEL_PID) { |
525 | /* | | 527 | /* |
526 | * The context-synchronizing instruction after tlbwe or tlbivax | | 528 | * The context-synchronizing instruction after tlbwe or tlbivax |
527 | * ensures that subsequent accesses (data and instruction) use | | 529 | * ensures that subsequent accesses (data and instruction) use |
528 | * the updated value in any TLB entries affected. | | 530 | * the updated value in any TLB entries affected. |
529 | */ | | 531 | */ |
530 | __asm volatile("isync\n\tsync"); | | 532 | __asm volatile("isync\n\tsync"); |
531 | } | | 533 | } |
532 | } | | 534 | } |
533 | | | 535 | |
534 | static bool | | 536 | static bool |
535 | e500_tlb_update_addr(vaddr_t va, tlb_asid_t asid, pt_entry_t pte, bool insert) | | 537 | e500_tlb_update_addr(vaddr_t va, tlb_asid_t asid, pt_entry_t pte, bool insert) |
536 | { | | 538 | { |
537 | #if defined(MULTIPROCESSOR) | | 539 | #if defined(MULTIPROCESSOR) |
538 | e500_tlb_invalidate_addr(va, asid); | | 540 | e500_tlb_invalidate_addr(va, asid); |
539 | return true; | | 541 | return true; |
540 | #else /* !MULTIPROCESSOR */ | | 542 | #else /* !MULTIPROCESSOR */ |
541 | struct e500_hwtlb hwtlb = tlb_to_hwtlb( | | 543 | struct e500_hwtlb hwtlb = tlb_to_hwtlb( |
542 | (struct e500_tlb){ .tlb_va = va, .tlb_asid = asid, | | 544 | (struct e500_tlb){ .tlb_va = va, .tlb_asid = asid, |
543 | .tlb_size = PAGE_SIZE, .tlb_pte = pte,}); | | 545 | .tlb_size = PAGE_SIZE, .tlb_pte = pte,}); |
544 | | | 546 | |
545 | register_t msr = wrtee(0); | | 547 | register_t msr = wrtee(0); |
546 | mtspr(SPR_MAS6, asid ? __SHIFTIN(asid, MAS6_SPID0) | MAS6_SAS : 0); | | 548 | mtspr(SPR_MAS6, asid ? __SHIFTIN(asid, MAS6_SPID0) | MAS6_SAS : 0); |
547 | __asm volatile("tlbsx 0, %0" :: "b"(va)); | | 549 | __asm volatile("tlbsx 0, %0" :: "b"(va)); |
548 | register_t mas1 = mfspr(SPR_MAS1); | | 550 | register_t mas1 = mfspr(SPR_MAS1); |
549 | if ((mas1 & MAS1_V) == 0) { | | 551 | if ((mas1 & MAS1_V) == 0) { |
550 | if (!insert) { | | 552 | if (!insert) { |
551 | wrtee(msr); | | 553 | wrtee(msr); |
552 | #if 0 | | 554 | #if 0 |
553 | printf("%s(%#lx,%#x,%#x,%x)<no update>\n", | | 555 | printf("%s(%#lx,%#x,%#x,%x)<no update>\n", |
554 | __func__, va, asid, pte, insert); | | 556 | __func__, va, asid, pte, insert); |
555 | #endif | | 557 | #endif |
556 | return false; | | 558 | return false; |
557 | } | | 559 | } |
558 | mas1 = hwtlb.hwtlb_mas1 | MAS1_V; | | 560 | mas1 = hwtlb.hwtlb_mas1 | MAS1_V; |
559 | mtspr(SPR_MAS1, mas1); | | 561 | mtspr(SPR_MAS1, mas1); |
560 | } | | 562 | } |
561 | mtspr(SPR_MAS2, hwtlb.hwtlb_mas2); | | 563 | mtspr(SPR_MAS2, hwtlb.hwtlb_mas2); |
562 | mtspr(SPR_MAS3, hwtlb.hwtlb_mas3); | | 564 | mtspr(SPR_MAS3, hwtlb.hwtlb_mas3); |
563 | //mtspr(SPR_MAS7, 0); | | 565 | //mtspr(SPR_MAS7, 0); |
564 | __asm volatile("tlbwe"); | | 566 | __asm volatile("tlbwe"); |
565 | if (asid == KERNEL_PID) | | 567 | if (asid == KERNEL_PID) |
566 | __asm volatile("isync\n\tsync"); | | 568 | __asm volatile("isync\n\tsync"); |
567 | wrtee(msr); | | 569 | wrtee(msr); |
568 | #if 0 | | 570 | #if 0 |
569 | if (asid) | | 571 | if (asid) |
570 | printf("%s(%#lx,%#x,%#x,%x)->[%x,%x,%x]\n", | | 572 | printf("%s(%#lx,%#x,%#x,%x)->[%x,%x,%x]\n", |
571 | __func__, va, asid, pte, insert, | | 573 | __func__, va, asid, pte, insert, |
572 | hwtlb.hwtlb_mas1, hwtlb.hwtlb_mas2, hwtlb.hwtlb_mas3); | | 574 | hwtlb.hwtlb_mas1, hwtlb.hwtlb_mas2, hwtlb.hwtlb_mas3); |
573 | #endif | | 575 | #endif |
574 | return (mas1 & MAS1_V) != 0; | | 576 | return (mas1 & MAS1_V) != 0; |
575 | #endif /* MULTIPROCESSOR */ | | 577 | #endif /* MULTIPROCESSOR */ |
576 | } | | 578 | } |
577 | | | 579 | |
578 | static void | | 580 | static void |
579 | e500_tlb_write_entry(size_t index, const struct tlbmask *tlb) | | 581 | e500_tlb_write_entry(size_t index, const struct tlbmask *tlb) |
580 | { | | 582 | { |
581 | } | | 583 | } |
582 | | | 584 | |
583 | static void | | 585 | static void |
584 | e500_tlb_read_entry(size_t index, struct tlbmask *tlb) | | 586 | e500_tlb_read_entry(size_t index, struct tlbmask *tlb) |
585 | { | | 587 | { |
586 | } | | 588 | } |
587 | | | 589 | |
588 | static void | | 590 | static void |
589 | e500_tlb_dump(void (*pr)(const char *, ...)) | | 591 | e500_tlb_dump(void (*pr)(const char *, ...)) |
590 | { | | 592 | { |
591 | const size_t tlbassoc = TLBCFG_ASSOC(mftlb0cfg()); | | 593 | const size_t tlbassoc = TLBCFG_ASSOC(mftlb0cfg()); |
592 | const size_t tlbentries = TLBCFG_NENTRY(mftlb0cfg()); | | 594 | const size_t tlbentries = TLBCFG_NENTRY(mftlb0cfg()); |
593 | const size_t max_epn = (tlbentries / tlbassoc) << PAGE_SHIFT; | | 595 | const size_t max_epn = (tlbentries / tlbassoc) << PAGE_SHIFT; |
594 | const uint32_t saved_mas0 = mfspr(SPR_MAS0); | | 596 | const uint32_t saved_mas0 = mfspr(SPR_MAS0); |
595 | size_t valid = 0; | | 597 | size_t valid = 0; |
596 | | | 598 | |
597 | if (pr == NULL) | | 599 | if (pr == NULL) |
598 | pr = printf; | | 600 | pr = printf; |
599 | | | 601 | |
600 | const register_t msr = wrtee(0); | | 602 | const register_t msr = wrtee(0); |
601 | for (size_t assoc = 0; assoc < tlbassoc; assoc++) { | | 603 | for (size_t assoc = 0; assoc < tlbassoc; assoc++) { |
602 | struct e500_hwtlb hwtlb; | | 604 | struct e500_hwtlb hwtlb; |
603 | hwtlb.hwtlb_mas0 = MAS0_ESEL_MAKE(assoc) | MAS0_TLBSEL_TLB0; | | 605 | hwtlb.hwtlb_mas0 = MAS0_ESEL_MAKE(assoc) | MAS0_TLBSEL_TLB0; |
604 | mtspr(SPR_MAS0, hwtlb.hwtlb_mas0); | | 606 | mtspr(SPR_MAS0, hwtlb.hwtlb_mas0); |
605 | for (size_t epn = 0; epn < max_epn; epn += PAGE_SIZE) { | | 607 | for (size_t epn = 0; epn < max_epn; epn += PAGE_SIZE) { |
606 | mtspr(SPR_MAS2, epn); | | 608 | mtspr(SPR_MAS2, epn); |
607 | __asm volatile("tlbre"); | | 609 | __asm volatile("tlbre"); |
608 | hwtlb.hwtlb_mas1 = mfspr(SPR_MAS1); | | 610 | hwtlb.hwtlb_mas1 = mfspr(SPR_MAS1); |
609 | /* | | 611 | /* |
610 | * If this is a valid entry for AS space 1 and | | 612 | * If this is a valid entry for AS space 1 and |
611 | * its asid matches the constraints of the caller, | | 613 | * its asid matches the constraints of the caller, |
612 | * clear its valid bit. | | 614 | * clear its valid bit. |
613 | */ | | 615 | */ |
614 | if (hwtlb.hwtlb_mas1 & MAS1_V) { | | 616 | if (hwtlb.hwtlb_mas1 & MAS1_V) { |
615 | hwtlb.hwtlb_mas2 = mfspr(SPR_MAS2); | | 617 | hwtlb.hwtlb_mas2 = mfspr(SPR_MAS2); |
616 | hwtlb.hwtlb_mas3 = mfspr(SPR_MAS3); | | 618 | hwtlb.hwtlb_mas3 = mfspr(SPR_MAS3); |
617 | struct e500_tlb tlb = hwtlb_to_tlb(hwtlb); | | 619 | struct e500_tlb tlb = hwtlb_to_tlb(hwtlb); |
618 | (*pr)("[%zu,%zu]->[%x,%x,%x]", | | 620 | (*pr)("[%zu,%zu]->[%x,%x,%x]", |
619 | assoc, atop(epn), | | 621 | assoc, atop(epn), |
620 | hwtlb.hwtlb_mas1, | | 622 | hwtlb.hwtlb_mas1, |
621 | hwtlb.hwtlb_mas2, | | 623 | hwtlb.hwtlb_mas2, |
622 | hwtlb.hwtlb_mas3); | | 624 | hwtlb.hwtlb_mas3); |
623 | (*pr)(": VA=%#lx size=4KB asid=%u pte=%x", | | 625 | (*pr)(": VA=%#lx size=4KB asid=%u pte=%x", |
624 | tlb.tlb_va, tlb.tlb_asid, tlb.tlb_pte); | | 626 | tlb.tlb_va, tlb.tlb_asid, tlb.tlb_pte); |
625 | (*pr)(" (RPN=%#x,%s%s%s%s%s,%s%s%s%s%s)\n", | | 627 | (*pr)(" (RPN=%#x,%s%s%s%s%s,%s%s%s%s%s)\n", |
626 | tlb.tlb_pte & PTE_RPN_MASK, | | 628 | tlb.tlb_pte & PTE_RPN_MASK, |
627 | tlb.tlb_pte & PTE_xR ? "R" : "", | | 629 | tlb.tlb_pte & PTE_xR ? "R" : "", |
628 | tlb.tlb_pte & PTE_xW ? "W" : "", | | 630 | tlb.tlb_pte & PTE_xW ? "W" : "", |
629 | tlb.tlb_pte & PTE_UNMODIFIED ? "*" : "", | | 631 | tlb.tlb_pte & PTE_UNMODIFIED ? "*" : "", |
630 | tlb.tlb_pte & PTE_xX ? "X" : "", | | 632 | tlb.tlb_pte & PTE_xX ? "X" : "", |
631 | tlb.tlb_pte & PTE_UNSYNCED ? "*" : "", | | 633 | tlb.tlb_pte & PTE_UNSYNCED ? "*" : "", |
632 | tlb.tlb_pte & PTE_W ? "W" : "", | | 634 | tlb.tlb_pte & PTE_W ? "W" : "", |
633 | tlb.tlb_pte & PTE_I ? "I" : "", | | 635 | tlb.tlb_pte & PTE_I ? "I" : "", |
634 | tlb.tlb_pte & PTE_M ? "M" : "", | | 636 | tlb.tlb_pte & PTE_M ? "M" : "", |
635 | tlb.tlb_pte & PTE_G ? "G" : "", | | 637 | tlb.tlb_pte & PTE_G ? "G" : "", |
636 | tlb.tlb_pte & PTE_E ? "E" : ""); | | 638 | tlb.tlb_pte & PTE_E ? "E" : ""); |
637 | valid++; | | 639 | valid++; |
638 | } | | 640 | } |
639 | } | | 641 | } |
640 | } | | 642 | } |
641 | mtspr(SPR_MAS0, saved_mas0); | | 643 | mtspr(SPR_MAS0, saved_mas0); |
642 | wrtee(msr); | | 644 | wrtee(msr); |
643 | (*pr)("%s: %zu valid entries\n", __func__, valid); | | 645 | (*pr)("%s: %zu valid entries\n", __func__, valid); |
644 | } | | 646 | } |
645 | | | 647 | |
646 | static void | | 648 | static void |
647 | e500_tlb_walk(void *ctx, bool (*func)(void *, vaddr_t, uint32_t, uint32_t)) | | 649 | e500_tlb_walk(void *ctx, bool (*func)(void *, vaddr_t, uint32_t, uint32_t)) |
648 | { | | 650 | { |
649 | const size_t tlbassoc = TLBCFG_ASSOC(mftlb0cfg()); | | 651 | const size_t tlbassoc = TLBCFG_ASSOC(mftlb0cfg()); |
650 | const size_t tlbentries = TLBCFG_NENTRY(mftlb0cfg()); | | 652 | const size_t tlbentries = TLBCFG_NENTRY(mftlb0cfg()); |
651 | const size_t max_epn = (tlbentries / tlbassoc) << PAGE_SHIFT; | | 653 | const size_t max_epn = (tlbentries / tlbassoc) << PAGE_SHIFT; |
652 | const uint32_t saved_mas0 = mfspr(SPR_MAS0); | | 654 | const uint32_t saved_mas0 = mfspr(SPR_MAS0); |
653 | | | 655 | |
654 | const register_t msr = wrtee(0); | | 656 | const register_t msr = wrtee(0); |
655 | for (size_t assoc = 0; assoc < tlbassoc; assoc++) { | | 657 | for (size_t assoc = 0; assoc < tlbassoc; assoc++) { |
656 | struct e500_hwtlb hwtlb; | | 658 | struct e500_hwtlb hwtlb; |
657 | hwtlb.hwtlb_mas0 = MAS0_ESEL_MAKE(assoc) | MAS0_TLBSEL_TLB0; | | 659 | hwtlb.hwtlb_mas0 = MAS0_ESEL_MAKE(assoc) | MAS0_TLBSEL_TLB0; |
658 | mtspr(SPR_MAS0, hwtlb.hwtlb_mas0); | | 660 | mtspr(SPR_MAS0, hwtlb.hwtlb_mas0); |
659 | for (size_t epn = 0; epn < max_epn; epn += PAGE_SIZE) { | | 661 | for (size_t epn = 0; epn < max_epn; epn += PAGE_SIZE) { |
660 | mtspr(SPR_MAS2, epn); | | 662 | mtspr(SPR_MAS2, epn); |
661 | __asm volatile("tlbre"); | | 663 | __asm volatile("tlbre"); |
662 | hwtlb.hwtlb_mas1 = mfspr(SPR_MAS1); | | 664 | hwtlb.hwtlb_mas1 = mfspr(SPR_MAS1); |
663 | if (hwtlb.hwtlb_mas1 & MAS1_V) { | | 665 | if (hwtlb.hwtlb_mas1 & MAS1_V) { |
664 | hwtlb.hwtlb_mas2 = mfspr(SPR_MAS2); | | 666 | hwtlb.hwtlb_mas2 = mfspr(SPR_MAS2); |
665 | hwtlb.hwtlb_mas3 = mfspr(SPR_MAS3); | | 667 | hwtlb.hwtlb_mas3 = mfspr(SPR_MAS3); |
666 | struct e500_tlb tlb = hwtlb_to_tlb(hwtlb); | | 668 | struct e500_tlb tlb = hwtlb_to_tlb(hwtlb); |
667 | if (!(*func)(ctx, tlb.tlb_va, tlb.tlb_asid, | | 669 | if (!(*func)(ctx, tlb.tlb_va, tlb.tlb_asid, |
668 | tlb.tlb_pte)) | | 670 | tlb.tlb_pte)) |
669 | break; | | 671 | break; |
670 | } | | 672 | } |
671 | } | | 673 | } |
672 | } | | 674 | } |
673 | mtspr(SPR_MAS0, saved_mas0); | | 675 | mtspr(SPR_MAS0, saved_mas0); |
674 | wrtee(msr); | | 676 | wrtee(msr); |
675 | } | | 677 | } |
676 | | | 678 | |
677 | static struct e500_xtlb * | | 679 | static struct e500_xtlb * |
678 | e500_tlb_lookup_xtlb_pa(vaddr_t pa, u_int *slotp) | | 680 | e500_tlb_lookup_xtlb_pa(vaddr_t pa, u_int *slotp) |
679 | { | | 681 | { |
680 | struct e500_tlb1 * const tlb1 = &e500_tlb1; | | 682 | struct e500_tlb1 * const tlb1 = &e500_tlb1; |
681 | struct e500_xtlb *xtlb = tlb1->tlb1_entries; | | 683 | struct e500_xtlb *xtlb = tlb1->tlb1_entries; |
682 | | | 684 | |
683 | /* | | 685 | /* |
684 | * See if we have a TLB entry for the pa. | | 686 | * See if we have a TLB entry for the pa. |
685 | */ | | 687 | */ |
686 | for (u_int i = 0; i < tlb1->tlb1_numentries; i++, xtlb++) { | | 688 | for (u_int i = 0; i < tlb1->tlb1_numentries; i++, xtlb++) { |
687 | psize_t mask = ~(xtlb->e_tlb.tlb_size - 1); | | 689 | psize_t mask = ~(xtlb->e_tlb.tlb_size - 1); |
688 | if ((xtlb->e_hwtlb.hwtlb_mas1 & MAS1_V) | | 690 | if ((xtlb->e_hwtlb.hwtlb_mas1 & MAS1_V) |
689 | && ((pa ^ xtlb->e_tlb.tlb_pte) & mask) == 0) { | | 691 | && ((pa ^ xtlb->e_tlb.tlb_pte) & mask) == 0) { |
690 | if (slotp != NULL) | | 692 | if (slotp != NULL) |
691 | *slotp = i; | | 693 | *slotp = i; |
692 | return xtlb; | | 694 | return xtlb; |
693 | } | | 695 | } |
694 | } | | 696 | } |
695 | | | 697 | |
696 | return NULL; | | 698 | return NULL; |
697 | } | | 699 | } |
698 | | | 700 | |
699 | struct e500_xtlb * | | 701 | struct e500_xtlb * |
700 | e500_tlb_lookup_xtlb(vaddr_t va, u_int *slotp) | | 702 | e500_tlb_lookup_xtlb(vaddr_t va, u_int *slotp) |
701 | { | | 703 | { |
702 | struct e500_tlb1 * const tlb1 = &e500_tlb1; | | 704 | struct e500_tlb1 * const tlb1 = &e500_tlb1; |
703 | struct e500_xtlb *xtlb = tlb1->tlb1_entries; | | 705 | struct e500_xtlb *xtlb = tlb1->tlb1_entries; |
704 | | | 706 | |
705 | /* | | 707 | /* |
706 | * See if we have a TLB entry for the va. | | 708 | * See if we have a TLB entry for the va. |
707 | */ | | 709 | */ |
708 | for (u_int i = 0; i < tlb1->tlb1_numentries; i++, xtlb++) { | | 710 | for (u_int i = 0; i < tlb1->tlb1_numentries; i++, xtlb++) { |
709 | vsize_t mask = ~(xtlb->e_tlb.tlb_size - 1); | | 711 | vsize_t mask = ~(xtlb->e_tlb.tlb_size - 1); |
710 | if ((xtlb->e_hwtlb.hwtlb_mas1 & MAS1_V) | | 712 | if ((xtlb->e_hwtlb.hwtlb_mas1 & MAS1_V) |
711 | && ((va ^ xtlb->e_tlb.tlb_va) & mask) == 0) { | | 713 | && ((va ^ xtlb->e_tlb.tlb_va) & mask) == 0) { |
712 | if (slotp != NULL) | | 714 | if (slotp != NULL) |
713 | *slotp = i; | | 715 | *slotp = i; |
714 | return xtlb; | | 716 | return xtlb; |
715 | } | | 717 | } |
716 | } | | 718 | } |
717 | | | 719 | |
718 | return NULL; | | 720 | return NULL; |
719 | } | | 721 | } |
720 | | | 722 | |
721 | static struct e500_xtlb * | | 723 | static struct e500_xtlb * |
722 | e500_tlb_lookup_xtlb2(vaddr_t va, vsize_t len) | | 724 | e500_tlb_lookup_xtlb2(vaddr_t va, vsize_t len) |
723 | { | | 725 | { |
724 | struct e500_tlb1 * const tlb1 = &e500_tlb1; | | 726 | struct e500_tlb1 * const tlb1 = &e500_tlb1; |
725 | struct e500_xtlb *xtlb = tlb1->tlb1_entries; | | 727 | struct e500_xtlb *xtlb = tlb1->tlb1_entries; |
726 | | | 728 | |
727 | /* | | 729 | /* |
728 | * See if we have a TLB entry for the pa. | | 730 | * See if we have a TLB entry for the pa. |
729 | */ | | 731 | */ |
730 | for (u_int i = 0; i < tlb1->tlb1_numentries; i++, xtlb++) { | | 732 | for (u_int i = 0; i < tlb1->tlb1_numentries; i++, xtlb++) { |
731 | vsize_t mask = ~(xtlb->e_tlb.tlb_size - 1); | | 733 | vsize_t mask = ~(xtlb->e_tlb.tlb_size - 1); |
732 | if ((xtlb->e_hwtlb.hwtlb_mas1 & MAS1_V) | | 734 | if ((xtlb->e_hwtlb.hwtlb_mas1 & MAS1_V) |
733 | && ((va ^ xtlb->e_tlb.tlb_va) & mask) == 0 | | 735 | && ((va ^ xtlb->e_tlb.tlb_va) & mask) == 0 |
734 | && (((va + len - 1) ^ va) & mask) == 0) { | | 736 | && (((va + len - 1) ^ va) & mask) == 0) { |
735 | return xtlb; | | 737 | return xtlb; |
736 | } | | 738 | } |
737 | } | | 739 | } |
738 | | | 740 | |
739 | return NULL; | | 741 | return NULL; |
740 | } | | 742 | } |
741 | | | 743 | |
742 | static void * | | 744 | static void * |
743 | e500_tlb_mapiodev(paddr_t pa, psize_t len, bool prefetchable) | | 745 | e500_tlb_mapiodev(paddr_t pa, psize_t len, bool prefetchable) |
744 | { | | 746 | { |
745 | struct e500_xtlb * const xtlb = e500_tlb_lookup_xtlb_pa(pa, NULL); | | 747 | struct e500_xtlb * const xtlb = e500_tlb_lookup_xtlb_pa(pa, NULL); |
746 | | | 748 | |
747 | /* | | 749 | /* |
748 | * See if we have a TLB entry for the pa. If completely falls within | | 750 | * See if we have a TLB entry for the pa. If completely falls within |
749 | * mark the reference and return the pa. But only if the tlb entry | | 751 | * mark the reference and return the pa. But only if the tlb entry |
750 | * is not cacheable. | | 752 | * is not cacheable. |
751 | */ | | 753 | */ |
752 | if (xtlb | | 754 | if (xtlb |
753 | && (prefetchable | | 755 | && (prefetchable |
754 | || (xtlb->e_tlb.tlb_pte & PTE_WIG) == (PTE_I|PTE_G))) { | | 756 | || (xtlb->e_tlb.tlb_pte & PTE_WIG) == (PTE_I|PTE_G))) { |
755 | xtlb->e_refcnt++; | | 757 | xtlb->e_refcnt++; |
756 | return (void *) (xtlb->e_tlb.tlb_va | | 758 | return (void *) (xtlb->e_tlb.tlb_va |
757 | + pa - (xtlb->e_tlb.tlb_pte & PTE_RPN_MASK)); | | 759 | + pa - (xtlb->e_tlb.tlb_pte & PTE_RPN_MASK)); |
758 | } | | 760 | } |
759 | return NULL; | | 761 | return NULL; |
760 | } | | 762 | } |
761 | | | 763 | |
762 | static void | | 764 | static void |
763 | e500_tlb_unmapiodev(vaddr_t va, vsize_t len) | | 765 | e500_tlb_unmapiodev(vaddr_t va, vsize_t len) |
764 | { | | 766 | { |
765 | if (va < VM_MIN_KERNEL_ADDRESS || VM_MAX_KERNEL_ADDRESS <= va) { | | 767 | if (va < VM_MIN_KERNEL_ADDRESS || VM_MAX_KERNEL_ADDRESS <= va) { |
766 | struct e500_xtlb * const xtlb = e500_tlb_lookup_xtlb(va, NULL); | | 768 | struct e500_xtlb * const xtlb = e500_tlb_lookup_xtlb(va, NULL); |
767 | if (xtlb) | | 769 | if (xtlb) |
768 | xtlb->e_refcnt--; | | 770 | xtlb->e_refcnt--; |
769 | } | | 771 | } |
770 | } | | 772 | } |
771 | | | 773 | |
772 | static int | | 774 | static int |
773 | e500_tlb_ioreserve(vaddr_t va, vsize_t len, pt_entry_t pte) | | 775 | e500_tlb_ioreserve(vaddr_t va, vsize_t len, pt_entry_t pte) |
774 | { | | 776 | { |
775 | struct e500_tlb1 * const tlb1 = &e500_tlb1; | | 777 | struct e500_tlb1 * const tlb1 = &e500_tlb1; |
776 | struct e500_xtlb *xtlb; | | 778 | struct e500_xtlb *xtlb; |
777 | | | 779 | |
778 | KASSERT(len & 0x55555000); | | 780 | KASSERT(len & 0x55555000); |
779 | KASSERT((len & ~0x55555000) == 0); | | 781 | KASSERT((len & ~0x55555000) == 0); |
780 | KASSERT(len >= PAGE_SIZE); | | 782 | KASSERT(len >= PAGE_SIZE); |
781 | KASSERT((len & (len - 1)) == 0); | | 783 | KASSERT((len & (len - 1)) == 0); |
782 | KASSERT((va & (len - 1)) == 0); | | 784 | KASSERT((va & (len - 1)) == 0); |
783 | KASSERT(((pte & PTE_RPN_MASK) & (len - 1)) == 0); | | 785 | KASSERT(((pte & PTE_RPN_MASK) & (len - 1)) == 0); |
784 | | | 786 | |
785 | if ((xtlb = e500_tlb_lookup_xtlb2(va, len)) != NULL) { | | 787 | if ((xtlb = e500_tlb_lookup_xtlb2(va, len)) != NULL) { |
786 | psize_t mask __diagused = ~(xtlb->e_tlb.tlb_size - 1); | | 788 | psize_t mask __diagused = ~(xtlb->e_tlb.tlb_size - 1); |
787 | KASSERT(len <= xtlb->e_tlb.tlb_size); | | 789 | KASSERT(len <= xtlb->e_tlb.tlb_size); |
788 | KASSERT((pte & mask) == (xtlb->e_tlb.tlb_pte & mask)); | | 790 | KASSERT((pte & mask) == (xtlb->e_tlb.tlb_pte & mask)); |
789 | xtlb->e_refcnt++; | | 791 | xtlb->e_refcnt++; |
790 | return 0; | | 792 | return 0; |
791 | } | | 793 | } |
792 | | | 794 | |
793 | const int slot = e500_alloc_tlb1_entry(); | | 795 | const int slot = e500_alloc_tlb1_entry(); |
794 | if (slot < 0) | | 796 | if (slot < 0) |
795 | return ENOMEM; | | 797 | return ENOMEM; |
796 | | | 798 | |
797 | xtlb = &tlb1->tlb1_entries[slot]; | | 799 | xtlb = &tlb1->tlb1_entries[slot]; |
798 | xtlb->e_tlb.tlb_va = va; | | 800 | xtlb->e_tlb.tlb_va = va; |
799 | xtlb->e_tlb.tlb_size = len; | | 801 | xtlb->e_tlb.tlb_size = len; |
800 | xtlb->e_tlb.tlb_pte = pte; | | 802 | xtlb->e_tlb.tlb_pte = pte; |
801 | xtlb->e_tlb.tlb_asid = KERNEL_PID; | | 803 | xtlb->e_tlb.tlb_asid = KERNEL_PID; |
802 | | | 804 | |
803 | xtlb->e_hwtlb = tlb_to_hwtlb(xtlb->e_tlb); | | 805 | xtlb->e_hwtlb = tlb_to_hwtlb(xtlb->e_tlb); |
804 | xtlb->e_hwtlb.hwtlb_mas0 |= __SHIFTIN(slot, MAS0_ESEL); | | 806 | xtlb->e_hwtlb.hwtlb_mas0 |= __SHIFTIN(slot, MAS0_ESEL); |
805 | hwtlb_write(xtlb->e_hwtlb, true); | | 807 | hwtlb_write(xtlb->e_hwtlb, true); |
806 | | | 808 | |
807 | #if defined(MULTIPROCESSOR) | | 809 | #if defined(MULTIPROCESSOR) |
808 | cpu_send_ipi(IPI_DST_NOTME, IPI_TLB1SYNC); | | 810 | cpu_send_ipi(IPI_DST_NOTME, IPI_TLB1SYNC); |
809 | #endif | | 811 | #endif |
810 | | | 812 | |
811 | return 0; | | 813 | return 0; |
812 | } | | 814 | } |
813 | | | 815 | |
814 | static int | | 816 | static int |
815 | e500_tlb_iorelease(vaddr_t va) | | 817 | e500_tlb_iorelease(vaddr_t va) |
816 | { | | 818 | { |
817 | u_int slot; | | 819 | u_int slot; |
818 | struct e500_xtlb * const xtlb = e500_tlb_lookup_xtlb(va, &slot); | | 820 | struct e500_xtlb * const xtlb = e500_tlb_lookup_xtlb(va, &slot); |
819 | | | 821 | |
820 | if (xtlb == NULL) | | 822 | if (xtlb == NULL) |
821 | return ENOENT; | | 823 | return ENOENT; |
822 | | | 824 | |
823 | if (xtlb->e_refcnt) | | 825 | if (xtlb->e_refcnt) |
824 | return EBUSY; | | 826 | return EBUSY; |
825 | | | 827 | |
826 | e500_free_tlb1_entry(xtlb, slot, true); | | 828 | e500_free_tlb1_entry(xtlb, slot, true); |
827 | | | 829 | |
828 | #if defined(MULTIPROCESSOR) | | 830 | #if defined(MULTIPROCESSOR) |
829 | cpu_send_ipi(IPI_DST_NOTME, IPI_TLB1SYNC); | | 831 | cpu_send_ipi(IPI_DST_NOTME, IPI_TLB1SYNC); |
830 | #endif | | 832 | #endif |
831 | | | 833 | |
832 | return 0; | | 834 | return 0; |
833 | } | | 835 | } |
834 | | | 836 | |
835 | static u_int | | 837 | static u_int |
836 | e500_tlbmemmap(paddr_t memstart, psize_t memsize, struct e500_tlb1 *tlb1) | | 838 | e500_tlbmemmap(paddr_t memstart, psize_t memsize, struct e500_tlb1 *tlb1) |
837 | { | | 839 | { |
838 | u_int slotmask = 0; | | 840 | u_int slotmask = 0; |
839 | u_int slots = 0, nextslot = 0; | | 841 | u_int slots = 0, nextslot = 0; |
840 | KASSERT(tlb1->tlb1_numfree > 1); | | 842 | KASSERT(tlb1->tlb1_numfree > 1); |
841 | KASSERT(((memstart + memsize - 1) & -memsize) == memstart); | | 843 | KASSERT(((memstart + memsize - 1) & -memsize) == memstart); |
842 | for (paddr_t lastaddr = memstart; 0 < memsize; ) { | | 844 | for (paddr_t lastaddr = memstart; 0 < memsize; ) { |
843 | u_int cnt = __builtin_clz(memsize); | | 845 | u_int cnt = __builtin_clz(memsize); |
844 | psize_t size = uimin(1UL << (31 - (cnt | 1)), tlb1->tlb1_maxsize); | | 846 | psize_t size = uimin(1UL << (31 - (cnt | 1)), tlb1->tlb1_maxsize); |
845 | slots += memsize / size; | | 847 | slots += memsize / size; |
846 | if (slots > 4) | | 848 | if (slots > 4) |
847 | panic("%s: %d: can't map memory (%#lx) into TLB1: %s", | | 849 | panic("%s: %d: can't map memory (%#lx) into TLB1: %s", |
848 | __func__, __LINE__, memsize, "too fragmented"); | | 850 | __func__, __LINE__, memsize, "too fragmented"); |
849 | if (slots > tlb1->tlb1_numfree - 1) | | 851 | if (slots > tlb1->tlb1_numfree - 1) |
850 | panic("%s: %d: can't map memory (%#lx) into TLB1: %s", | | 852 | panic("%s: %d: can't map memory (%#lx) into TLB1: %s", |
851 | __func__, __LINE__, memsize, | | 853 | __func__, __LINE__, memsize, |
852 | "insufficent TLB entries"); | | 854 | "insufficent TLB entries"); |
853 | for (; nextslot < slots; nextslot++) { | | 855 | for (; nextslot < slots; nextslot++) { |
854 | const u_int freeslot = e500_alloc_tlb1_entry(); | | 856 | const u_int freeslot = e500_alloc_tlb1_entry(); |
855 | struct e500_xtlb * const xtlb = | | 857 | struct e500_xtlb * const xtlb = |
856 | &tlb1->tlb1_entries[freeslot]; | | 858 | &tlb1->tlb1_entries[freeslot]; |
857 | xtlb->e_tlb.tlb_asid = KERNEL_PID; | | 859 | xtlb->e_tlb.tlb_asid = KERNEL_PID; |
858 | xtlb->e_tlb.tlb_size = size; | | 860 | xtlb->e_tlb.tlb_size = size; |
859 | xtlb->e_tlb.tlb_va = lastaddr; | | 861 | xtlb->e_tlb.tlb_va = lastaddr; |
860 | xtlb->e_tlb.tlb_pte = lastaddr | | 862 | xtlb->e_tlb.tlb_pte = lastaddr |
861 | | PTE_M | PTE_xX | PTE_xW | PTE_xR; | | 863 | | PTE_M | PTE_xX | PTE_xW | PTE_xR; |
862 | lastaddr += size; | | 864 | lastaddr += size; |
863 | memsize -= size; | | 865 | memsize -= size; |
864 | slotmask |= 1 << (31 - freeslot); /* clz friendly */ | | 866 | slotmask |= 1 << (31 - freeslot); /* clz friendly */ |
865 | } | | 867 | } |
866 | } | | 868 | } |
867 | | | 869 | |
868 | #if defined(MULTIPROCESSOR) | | 870 | #if defined(MULTIPROCESSOR) |
869 | cpu_send_ipi(IPI_DST_NOTME, IPI_TLB1SYNC); | | 871 | cpu_send_ipi(IPI_DST_NOTME, IPI_TLB1SYNC); |
870 | #endif | | 872 | #endif |
871 | | | 873 | |
872 | return nextslot; | | 874 | return nextslot; |
873 | } | | 875 | } |
874 | | | 876 | |
875 | static const struct tlb_md_ops e500_tlb_ops = { | | 877 | static const struct tlb_md_ops e500_tlb_ops = { |
876 | .md_tlb_get_asid = e500_tlb_get_asid, | | 878 | .md_tlb_get_asid = e500_tlb_get_asid, |
877 | .md_tlb_set_asid = e500_tlb_set_asid, | | 879 | .md_tlb_set_asid = e500_tlb_set_asid, |
878 | .md_tlb_invalidate_all = e500_tlb_invalidate_all, | | 880 | .md_tlb_invalidate_all = e500_tlb_invalidate_all, |
879 | .md_tlb_invalidate_globals = e500_tlb_invalidate_globals, | | 881 | .md_tlb_invalidate_globals = e500_tlb_invalidate_globals, |
880 | .md_tlb_invalidate_asids = e500_tlb_invalidate_asids, | | 882 | .md_tlb_invalidate_asids = e500_tlb_invalidate_asids, |
881 | .md_tlb_invalidate_addr = e500_tlb_invalidate_addr, | | 883 | .md_tlb_invalidate_addr = e500_tlb_invalidate_addr, |
882 | .md_tlb_update_addr = e500_tlb_update_addr, | | 884 | .md_tlb_update_addr = e500_tlb_update_addr, |
883 | .md_tlb_record_asids = e500_tlb_record_asids, | | 885 | .md_tlb_record_asids = e500_tlb_record_asids, |
884 | .md_tlb_write_entry = e500_tlb_write_entry, | | 886 | .md_tlb_write_entry = e500_tlb_write_entry, |
885 | .md_tlb_read_entry = e500_tlb_read_entry, | | 887 | .md_tlb_read_entry = e500_tlb_read_entry, |
886 | .md_tlb_dump = e500_tlb_dump, | | 888 | .md_tlb_dump = e500_tlb_dump, |
887 | .md_tlb_walk = e500_tlb_walk, | | 889 | .md_tlb_walk = e500_tlb_walk, |
888 | }; | | 890 | }; |
889 | | | 891 | |
890 | static const struct tlb_md_io_ops e500_tlb_io_ops = { | | 892 | static const struct tlb_md_io_ops e500_tlb_io_ops = { |
891 | .md_tlb_mapiodev = e500_tlb_mapiodev, | | 893 | .md_tlb_mapiodev = e500_tlb_mapiodev, |
892 | .md_tlb_unmapiodev = e500_tlb_unmapiodev, | | 894 | .md_tlb_unmapiodev = e500_tlb_unmapiodev, |
893 | .md_tlb_ioreserve = e500_tlb_ioreserve, | | 895 | .md_tlb_ioreserve = e500_tlb_ioreserve, |
894 | .md_tlb_iorelease = e500_tlb_iorelease, | | 896 | .md_tlb_iorelease = e500_tlb_iorelease, |
895 | }; | | 897 | }; |
896 | | | 898 | |
897 | void | | 899 | void |
898 | e500_tlb_init(vaddr_t endkernel, psize_t memsize) | | 900 | e500_tlb_init(vaddr_t endkernel, psize_t memsize) |
899 | { | | 901 | { |
900 | struct e500_tlb1 * const tlb1 = &e500_tlb1; | | 902 | struct e500_tlb1 * const tlb1 = &e500_tlb1; |
901 | | | 903 | |
902 | #if 0 | | 904 | #if 0 |
903 | register_t mmucfg = mfspr(SPR_MMUCFG); | | 905 | register_t mmucfg = mfspr(SPR_MMUCFG); |
904 | register_t mas4 = mfspr(SPR_MAS4); | | 906 | register_t mas4 = mfspr(SPR_MAS4); |
905 | #endif | | 907 | #endif |
906 | | | 908 | |
907 | const uint32_t tlb1cfg = mftlb1cfg(); | | 909 | const uint32_t tlb1cfg = mftlb1cfg(); |
908 | tlb1->tlb1_numentries = TLBCFG_NENTRY(tlb1cfg); | | 910 | tlb1->tlb1_numentries = TLBCFG_NENTRY(tlb1cfg); |
909 | KASSERT(tlb1->tlb1_numentries <= __arraycount(tlb1->tlb1_entries)); | | 911 | KASSERT(tlb1->tlb1_numentries <= __arraycount(tlb1->tlb1_entries)); |
910 | /* | | 912 | /* |
911 | * Limit maxsize to 1G since 4G isn't really useful to us. | | 913 | * Limit maxsize to 1G since 4G isn't really useful to us. |
912 | */ | | 914 | */ |
913 | tlb1->tlb1_minsize = 1024 << (2 * TLBCFG_MINSIZE(tlb1cfg)); | | 915 | tlb1->tlb1_minsize = 1024 << (2 * TLBCFG_MINSIZE(tlb1cfg)); |
914 | tlb1->tlb1_maxsize = 1024 << (2 * uimin(10, TLBCFG_MAXSIZE(tlb1cfg))); | | 916 | tlb1->tlb1_maxsize = 1024 << (2 * uimin(10, TLBCFG_MAXSIZE(tlb1cfg))); |
915 | | | 917 | |
916 | #ifdef VERBOSE_INITPPC | | 918 | #ifdef VERBOSE_INITPPC |
917 | printf(" tlb1cfg=%#x numentries=%u minsize=%#xKB maxsize=%#xKB", | | 919 | printf(" tlb1cfg=%#x numentries=%u minsize=%#xKB maxsize=%#xKB", |
918 | tlb1cfg, tlb1->tlb1_numentries, tlb1->tlb1_minsize >> 10, | | 920 | tlb1cfg, tlb1->tlb1_numentries, tlb1->tlb1_minsize >> 10, |
919 | tlb1->tlb1_maxsize >> 10); | | 921 | tlb1->tlb1_maxsize >> 10); |
920 | #endif | | 922 | #endif |
921 | | | 923 | |
922 | /* | | 924 | /* |
923 | * Let's see what's in TLB1 and we need to invalidate any entry that | | 925 | * Let's see what's in TLB1 and we need to invalidate any entry that |
924 | * would fit within the kernel's mapped address space. | | 926 | * would fit within the kernel's mapped address space. |
925 | */ | | 927 | */ |
926 | psize_t memmapped = 0; | | 928 | psize_t memmapped = 0; |
927 | for (u_int i = 0; i < tlb1->tlb1_numentries; i++) { | | 929 | for (u_int i = 0; i < tlb1->tlb1_numentries; i++) { |
928 | struct e500_xtlb * const xtlb = &tlb1->tlb1_entries[i]; | | 930 | struct e500_xtlb * const xtlb = &tlb1->tlb1_entries[i]; |
929 | | | 931 | |
930 | xtlb->e_hwtlb = hwtlb_read(MAS0_TLBSEL_TLB1, i); | | 932 | xtlb->e_hwtlb = hwtlb_read(MAS0_TLBSEL_TLB1, i); |
931 | | | 933 | |
932 | if ((xtlb->e_hwtlb.hwtlb_mas1 & MAS1_V) == 0) { | | 934 | if ((xtlb->e_hwtlb.hwtlb_mas1 & MAS1_V) == 0) { |
933 | tlb1->tlb1_freelist[tlb1->tlb1_numfree++] = i; | | 935 | tlb1->tlb1_freelist[tlb1->tlb1_numfree++] = i; |
934 | #ifdef VERBOSE_INITPPC | | 936 | #ifdef VERBOSE_INITPPC |
935 | printf(" TLB1[%u]=<unused>", i); | | 937 | printf(" TLB1[%u]=<unused>", i); |
936 | #endif | | 938 | #endif |
937 | continue; | | 939 | continue; |
938 | } | | 940 | } |
939 | | | 941 | |
940 | xtlb->e_tlb = hwtlb_to_tlb(xtlb->e_hwtlb); | | 942 | xtlb->e_tlb = hwtlb_to_tlb(xtlb->e_hwtlb); |
941 | #ifdef VERBOSE_INITPPC | | 943 | #ifdef VERBOSE_INITPPC |
942 | printf(" TLB1[%u]=<%#lx,%#lx,%#x,%#x>", | | 944 | printf(" TLB1[%u]=<%#lx,%#lx,%#x,%#x>", |
943 | i, xtlb->e_tlb.tlb_va, xtlb->e_tlb.tlb_size, | | 945 | i, xtlb->e_tlb.tlb_va, xtlb->e_tlb.tlb_size, |
944 | xtlb->e_tlb.tlb_asid, xtlb->e_tlb.tlb_pte); | | 946 | xtlb->e_tlb.tlb_asid, xtlb->e_tlb.tlb_pte); |
945 | #endif | | 947 | #endif |
946 | if ((VM_MIN_KERNEL_ADDRESS <= xtlb->e_tlb.tlb_va | | 948 | if ((VM_MIN_KERNEL_ADDRESS <= xtlb->e_tlb.tlb_va |
947 | && xtlb->e_tlb.tlb_va < VM_MAX_KERNEL_ADDRESS) | | 949 | && xtlb->e_tlb.tlb_va < VM_MAX_KERNEL_ADDRESS) |
948 | || (xtlb->e_tlb.tlb_va < VM_MIN_KERNEL_ADDRESS | | 950 | || (xtlb->e_tlb.tlb_va < VM_MIN_KERNEL_ADDRESS |
949 | && VM_MIN_KERNEL_ADDRESS < | | 951 | && VM_MIN_KERNEL_ADDRESS < |
950 | xtlb->e_tlb.tlb_va + xtlb->e_tlb.tlb_size)) { | | 952 | xtlb->e_tlb.tlb_va + xtlb->e_tlb.tlb_size)) { |
951 | #ifdef VERBOSE_INITPPC | | 953 | #ifdef VERBOSE_INITPPC |
952 | printf("free"); | | 954 | printf("free"); |
953 | #endif | | 955 | #endif |
954 | e500_free_tlb1_entry(xtlb, i, false); | | 956 | e500_free_tlb1_entry(xtlb, i, false); |
955 | #ifdef VERBOSE_INITPPC | | 957 | #ifdef VERBOSE_INITPPC |
956 | printf("d"); | | 958 | printf("d"); |
957 | #endif | | 959 | #endif |
958 | continue; | | 960 | continue; |
959 | } | | 961 | } |
960 | if ((xtlb->e_hwtlb.hwtlb_mas1 & MAS1_IPROT) == 0) { | | 962 | if ((xtlb->e_hwtlb.hwtlb_mas1 & MAS1_IPROT) == 0) { |
961 | xtlb->e_hwtlb.hwtlb_mas1 |= MAS1_IPROT; | | 963 | xtlb->e_hwtlb.hwtlb_mas1 |= MAS1_IPROT; |
962 | hwtlb_write(xtlb->e_hwtlb, false); | | 964 | hwtlb_write(xtlb->e_hwtlb, false); |
963 | #ifdef VERBOSE_INITPPC | | 965 | #ifdef VERBOSE_INITPPC |
964 | printf("+iprot"); | | 966 | printf("+iprot"); |
965 | #endif | | 967 | #endif |
966 | } | | 968 | } |
967 | if (xtlb->e_tlb.tlb_pte & PTE_I) | | 969 | if (xtlb->e_tlb.tlb_pte & PTE_I) |
968 | continue; | | 970 | continue; |
969 | | | 971 | |
970 | if (xtlb->e_tlb.tlb_va == 0 | | 972 | if (xtlb->e_tlb.tlb_va == 0 |
971 | || xtlb->e_tlb.tlb_va + xtlb->e_tlb.tlb_size <= memsize) { | | 973 | || xtlb->e_tlb.tlb_va + xtlb->e_tlb.tlb_size <= memsize) { |
972 | memmapped += xtlb->e_tlb.tlb_size; | | 974 | memmapped += xtlb->e_tlb.tlb_size; |
973 | /* | | 975 | /* |
974 | * Let make sure main memory is setup so it's memory | | 976 | * Let make sure main memory is setup so it's memory |
975 | * coherent. For some reason u-boot doesn't set it up | | 977 | * coherent. For some reason u-boot doesn't set it up |
976 | * that way. | | 978 | * that way. |
977 | */ | | 979 | */ |
978 | if ((xtlb->e_hwtlb.hwtlb_mas2 & MAS2_M) == 0) { | | 980 | if ((xtlb->e_hwtlb.hwtlb_mas2 & MAS2_M) == 0) { |
979 | xtlb->e_hwtlb.hwtlb_mas2 |= MAS2_M; | | 981 | xtlb->e_hwtlb.hwtlb_mas2 |= MAS2_M; |
980 | hwtlb_write(xtlb->e_hwtlb, true); | | 982 | hwtlb_write(xtlb->e_hwtlb, true); |
981 | } | | 983 | } |
982 | } | | 984 | } |
983 | } | | 985 | } |
984 | | | 986 | |
985 | cpu_md_ops.md_tlb_ops = &e500_tlb_ops; | | 987 | cpu_md_ops.md_tlb_ops = &e500_tlb_ops; |
986 | cpu_md_ops.md_tlb_io_ops = &e500_tlb_io_ops; | | 988 | cpu_md_ops.md_tlb_io_ops = &e500_tlb_io_ops; |
987 | | | 989 | |
988 | if (__predict_false(memmapped < memsize)) { | | 990 | if (__predict_false(memmapped < memsize)) { |
989 | /* | | 991 | /* |
990 | * Let's see how many TLB entries are needed to map memory. | | 992 | * Let's see how many TLB entries are needed to map memory. |
991 | */ | | 993 | */ |
992 | u_int slotmask = e500_tlbmemmap(0, memsize, tlb1); | | 994 | u_int slotmask = e500_tlbmemmap(0, memsize, tlb1); |
993 | | | 995 | |
994 | /* | | 996 | /* |
995 | * To map main memory into the TLB, we need to flush any | | 997 | * To map main memory into the TLB, we need to flush any |
996 | * existing entries from the TLB that overlap the virtual | | 998 | * existing entries from the TLB that overlap the virtual |
997 | * address space needed to map physical memory. That may | | 999 | * address space needed to map physical memory. That may |
998 | * include the entries for the pages currently used by the | | 1000 | * include the entries for the pages currently used by the |
999 | * stack or that we are executing. So to avoid problems, we | | 1001 | * stack or that we are executing. So to avoid problems, we |
1000 | * are going to temporarily map the kernel and stack into AS 1, | | 1002 | * are going to temporarily map the kernel and stack into AS 1, |
1001 | * switch to it, and clear out the TLB entries from AS 0, | | 1003 | * switch to it, and clear out the TLB entries from AS 0, |
1002 | * install the new TLB entries to map memory, and then switch | | 1004 | * install the new TLB entries to map memory, and then switch |
1003 | * back to AS 0 and free the temp entry used for AS1. | | 1005 | * back to AS 0 and free the temp entry used for AS1. |
1004 | */ | | 1006 | */ |
1005 | u_int b = __builtin_clz(endkernel); | | 1007 | u_int b = __builtin_clz(endkernel); |
1006 | | | 1008 | |
1007 | /* | | 1009 | /* |
1008 | * If the kernel doesn't end on a clean power of 2, we need | | 1010 | * If the kernel doesn't end on a clean power of 2, we need |
1009 | * to round the size up (by decrementing the number of leading | | 1011 | * to round the size up (by decrementing the number of leading |
1010 | * zero bits). If the size isn't a power of 4KB, decrement | | 1012 | * zero bits). If the size isn't a power of 4KB, decrement |
1011 | * again to make it one. | | 1013 | * again to make it one. |
1012 | */ | | 1014 | */ |
1013 | if (endkernel & (endkernel - 1)) | | 1015 | if (endkernel & (endkernel - 1)) |
1014 | b--; | | 1016 | b--; |
1015 | if ((b & 1) == 0) | | 1017 | if ((b & 1) == 0) |
1016 | b--; | | 1018 | b--; |
1017 | | | 1019 | |
1018 | /* | | 1020 | /* |
1019 | * Create a TLB1 mapping for the kernel in AS1. | | 1021 | * Create a TLB1 mapping for the kernel in AS1. |
1020 | */ | | 1022 | */ |
1021 | const u_int kslot = e500_alloc_tlb1_entry(); | | 1023 | const u_int kslot = e500_alloc_tlb1_entry(); |
1022 | struct e500_xtlb * const kxtlb = &tlb1->tlb1_entries[kslot]; | | 1024 | struct e500_xtlb * const kxtlb = &tlb1->tlb1_entries[kslot]; |
1023 | kxtlb->e_tlb.tlb_va = 0; | | 1025 | kxtlb->e_tlb.tlb_va = 0; |
1024 | kxtlb->e_tlb.tlb_size = 1UL << (31 - b); | | 1026 | kxtlb->e_tlb.tlb_size = 1UL << (31 - b); |
1025 | kxtlb->e_tlb.tlb_pte = PTE_M|PTE_xR|PTE_xW|PTE_xX; | | 1027 | kxtlb->e_tlb.tlb_pte = PTE_M|PTE_xR|PTE_xW|PTE_xX; |
1026 | kxtlb->e_tlb.tlb_asid = KERNEL_PID; | | 1028 | kxtlb->e_tlb.tlb_asid = KERNEL_PID; |
1027 | | | 1029 | |
1028 | kxtlb->e_hwtlb = tlb_to_hwtlb(kxtlb->e_tlb); | | 1030 | kxtlb->e_hwtlb = tlb_to_hwtlb(kxtlb->e_tlb); |
1029 | kxtlb->e_hwtlb.hwtlb_mas0 |= __SHIFTIN(kslot, MAS0_ESEL); | | 1031 | kxtlb->e_hwtlb.hwtlb_mas0 |= __SHIFTIN(kslot, MAS0_ESEL); |
1030 | kxtlb->e_hwtlb.hwtlb_mas1 |= MAS1_TS; | | 1032 | kxtlb->e_hwtlb.hwtlb_mas1 |= MAS1_TS; |
1031 | hwtlb_write(kxtlb->e_hwtlb, true); | | 1033 | hwtlb_write(kxtlb->e_hwtlb, true); |
1032 | | | 1034 | |
1033 | /* | | 1035 | /* |
1034 | * Now that we have a TLB mapping in AS1 for the kernel and its | | 1036 | * Now that we have a TLB mapping in AS1 for the kernel and its |
1035 | * stack, we switch to AS1 to cleanup the TLB mappings for TLB0. | | 1037 | * stack, we switch to AS1 to cleanup the TLB mappings for TLB0. |
1036 | */ | | 1038 | */ |
1037 | const register_t saved_msr = mfmsr(); | | 1039 | const register_t saved_msr = mfmsr(); |
1038 | mtmsr(saved_msr | PSL_DS | PSL_IS); | | 1040 | mtmsr(saved_msr | PSL_DS | PSL_IS); |
1039 | __asm volatile("isync"); | | 1041 | __asm volatile("isync"); |
1040 | | | 1042 | |
1041 | /* | | 1043 | /* |