Initialize the workqueue flush structures correctly.

(riastradh)

Make Linux workqueues at IPL_VM for now.

(riastradh)

Pass the wq to linux_worker as it expects, not NULL.

(riastradh)

Actually insert the delayed work into the workqueue.

(riastradh)

Helps to actually allocate and free the workqueue memory!

(riastradh)

Add flush_scheduled_work back.

(riastradh)

First (plausible) draft of Linux workqueue implementation rework.

Untested, but this looks better than what was there before, or any of
the drafts leading up to this which got torn out of the typewriter,
crumpled up, and crudely tossed in frustration toward the wastepaper
basket by my desk alongside the empty bottles of Jack Daniels that
fueled them, or something like that.

Can't use multiple CPUs per workqueue.  That requires some explicit
management of per-CPU workqueue state, since NetBSD's workqueue(9)
doesn't provide that or cancellation or flushing.  Oops.

(riastradh)

New utility getaddrinfo(1) to reflect getaddrinfo(3).

Discussed on tech-userlevel back in April:

https://mail-index.netbsd.org/tech-userlevel/2013/04/25/msg007719.html

Not hooked into the build or sets yet.

(riastradh)

Fix spurious kassert on interrupted blocking read from /dev/random.

Return EINTR in this case instead.  While here, clarify comment.

Fixes PR kern/48119, simpler than the patch attached there, per
discussion with tls@, who had this right in the earlier version
of rndpseudo.c before I broke it (oops!).

(riastradh)

First draft of genfb attachment for i915.

(riastradh)

Rework Linux `work' to use NetBSD workqueues, not callouts.

Callers expect to be able to allocate in the workers, which callouts
don't allow.

Delayed work uses callouts only to delay enqueueing work.

Linux `workqueues' are still stubs.

(riastradh)

Round of hacking on i915 framebuffers.

(riastradh)

Omit more needless #ifdef __NetBSD__.

(riastradh)

Fix DRM_WAIT_UNTIL to initialize (RET) on every exit.

(riastradh)

Miscellaneous Linux header file crud.

(riastradh)

Create Linux spin locks at IPL_VM for now.

(riastradh)

Simplify drm_ioremap and map only what is requested, exclusively.

(riastradh)

Still more error branch cleanup.

(riastradh)

Hook up some GTT logic again in i915_driver_load.

(riastradh)

Use the cached mmio bar information to map the GTT.

(riastradh)

i915_gem_cleanup_stolen need not take down mm.stolen after all.

The one caller, i915_driver_unload, already takes care of this.
Having i915_gem_cleanup_stolen do it is obviously the right thing,
but this state of affairs reduces diffs to upstream.

(riastradh)

Just use bus_space_map for io-mapping, and limit to one at a time.

Reserving the whole region interferes with other parts of the driver
which want to map it in different ways.  It also horrifically wastes
space when actually mapped, because there's no way to map a subregion
of a reservation.  And the bus_space_reservation API is x86-only at
the moment.

(riastradh)

linux_ namespace for kmap and kunmap.

(riastradh)

Fix kunmap: unlike kunmap_atomic, it takes a page, not a vaddr.

Unmapping the vaddr where the struct vm_page is stored makes for a
rather confusing state of affairs!

(riastradh)

Use the 64-bit PCI DMA tag if available.

Otherwise, it can't handle >32-bit physical addresses, which
uvm_obj_wirepages seems to have a tendency to return.

(riastradh)

Fix arithmetic in gen6_ggtt_bind_object.

(riastradh)

Sprinkle some kasserts into i915_gem_gtt.c.

(riastradh)

Specify the ring map type as _DRM_REGISTERS.

(riastradh)

Remove some more needless #ifdef __NetBSD__.

(riastradh)

Unconditionally avoid arithmetic on void * in drm_bufs.c.

(riastradh)

Unconditionally avoid arithmetic on void * in drm_buffer.c.

(riastradh)

Use Linux atomics, not spinlocks or NetBSD atomics in drm_auth.c.

Removes an #ifdef __NetBSD__.

(riastradh)

Add Linux ATOMIC_INIT, atomic_inc_return, and atomic_dec_return.

(riastradh)

Remove #ifdef __NetBSD__ from const fixes in dvo_ch7xxx.c.

(riastradh)

Check that GEM wired pages use at most 40 bits of physical addresses.

Can't express this constraint to uvm_obj_wirepages at the moment, so
at least fail early and noisily if we violate it.

(riastradh)

The bus_dmamap_load_raw got moved to i915_gem_gtt but not committed.

(riastradh)

Use dev->struct_mutex, not drm_global_mutex, for ring->irq_queue.

(riastradh)

Fix result of DRM_TIMED_WAIT_UNTIL.

This has to return a positive number of ticks left if we haven't
timed out in order to match the semantics of Linux waitqueues.

This also fixes the amount of time for timeout if we ever wait for
more than one iteration.

Now we can actually wait for results from the ring buffers!

(riastradh)

More error branch cleanup.

(riastradh)

Destroy the mutex too in destroy_completion.

(riastradh)

More little error branch cleanup.

(riastradh)

Implement destroy_completion in <linux/completion.h>.

(riastradh)

Fix a bunch of i915 error branches.

Some of these fixes are necessary only on NetBSD because we require
mutex_destroy, while Linux doesn't.

(riastradh)

Adapt intel_pm.c to NetBSD.  Drop stubs.

(riastradh)

Add missing includes to intel_pm.c.

(riastradh)

Buncha new cruft for <linux/*.h>.

(riastradh)

Use -Wno-shadow, not a cpp kludge, for drm_crtc.c's local min/max.

(riastradh)

Shift the GTT size by twenty only once.

(riastradh)

Use the _unlocked variant of unreference in drm_gem_pager_detach.

(riastradh)

Hook up the i915_gem uvm pager ops.

(riastradh)

Use pmap_enter_default, not pmap_enter.

Seems to be a problem with modules referring to weak aliases?

(riastradh)

Helps to commit all my changes to drm_cache.c.

(riastradh)

Adapt the Intel ringbuffer code to NetBSD.

(riastradh)

Adapt the i915 GEM code to NetBSD.

(riastradh)

Simplify drm_mem_util.h; fix source compatibility for drm_free_large.

(riastradh)

Add missing initialization of dev->dmat in drm_pci_attach.

(riastradh)

Avoid the PAGE_MASK mess in drm_gem_vm.c, drm_scatter.c, & drm_vm.c.

(riastradh)

intel_ringbuffer.c needs <asm/param.h> for HZ.

(riastradh)

Add missing includes to i915_gem_execbuffer.c.

(riastradh)

Add missing includes to i915_gem.c.

(riastradh)

Implement drm_cache.c, for x86 only at the moment.

(riastradh)

Rework Intel GTT abstraction to use bus_dma.

(riastradh)

Implement a bus_dmamem_wire_uvm_object hack.

This is fit only for x86 at the moment -- it parrots the x86
bus_dmamem_alloc code but with uvm_obj_wirepages instead of
uvm_pglistalloc.

XXX THIS IS A MEGA-KLUDGE THAT NEEDS TO BE DONE PROPERLY.

(riastradh)

Rename delayed_work::dw_work to delayed_work::work for Linux source.

(riastradh)

Fix definition of Linux PAGE_ALIGN in <linux/mm.h>.

Can't use uvm round_page because that depends on PAGE_MASK, whose
sense we have to invert for Linux!  Plurgh.

(riastradh)

More miscellaneous Linux header cruft.  See patch for details.

(riastradh)

Add Linuxoid non-atomic __set/clear_bit to <linux/bitops.h>.

(riastradh)

Fix Linux atomic set/clear/change_bit to work on arrays.

(riastradh)

Add ERESTARTSYS as a Linuxoid alias for ERESTART.

(riastradh)

Add trace_i915_gem_object_unbind.

(riastradh)

Add some copy_to/from_user variants to Linux <asm/uaccess.h>.

(riastradh)

Define page_to_phys in Linux <asm/io.h>.

(riastradh)

Decrement sc_opencount on drm_close.

(riastradh)

Destroy context_idr and mm.lock in i915_driver_postclose.

(riastradh)

Fix sense of atomic_cas loop condition and use kassertmsg for krefs.

(riastradh)

Restore config `drm' attribute in sys/conf/files.

Not sure why this started to be necessary after the merge, but
whatever...

(riastradh)

Initialize and destroy drm_global_mutex on module init/fini.

(riastradh)

Remove trailing whitespace in texi2dvi which confuses sed.

XXX pull up to 5, 6?

(riastradh)

Make the rpi hwrng feed data to the entropy pool in a softint.

The two-lock scheme here is excessive but will do provisionally until
rnd(9) gets disentangled.

Tested by Aymeric Vincent.

(riastradh)

Comment on possible data-dependent branch in `!res'.

(riastradh)

Update set lists after libc minor bump.

(riastradh)

Get some entropy early on bcm2835_rng attach.

(riastradh)

Bump libc minor for consttime_memequal and explicit_memset.

(riastradh)

Publish explicit_memset and consttime_memequal in userland libc.

Remove the double-underscore from the userland versions, and do the
weak alias dance instead, now that these are public parts of libc.

As discussed on tech-userlevel:

https://mail-index.netbsd.org/tech-userlevel/2013/06/24/msg007843.html
(option 3)

(riastradh)

Fix libkern's prototype for explicit_memset.

(riastradh)

Make explicit_memset match memset's return value.

(riastradh)

Fix sense of consttime_memequal and update all callers.

Now it returns true (nonzero) to mean equal and false (zero) to mean
inequal, as the name suggests.

As promised on tech-userlevel back in June:

https://mail-index.netbsd.org/tech-userlevel/2013/06/24/msg007843.html

(riastradh)

Tighten initialization of rnd softints.

- Do rnd_init_softint as early as possible in main, after configure2,
  and before networking is initialized.

- Initialize the rnd_wakeup softint in rnd_init_softint, not lazily in
  rnd_schedule_wakeup.

ok tls

(riastradh)

Back out the recent rnd stop-gap/stop-gap/stop-gap measures.

This reverts

sys/dev/rnd_private.h -> r1.1
sys/kern/init_main.c -> r1.450
sys/kern/kern_rndq.c -> r1.14
sys/kern/kern_rndsink.c -> r1.2

Parts of these changes will be added back, and the rndsource
callbacks will be fixed to avoid the lock recursion bug that
motivated the stop-gaps in the first place.

ok tls

(riastradh)

Schedule wakeups asynchronously in rnd_extract_data.

Otherwise, rndsink_request takes rndsinks_lock and calls
rnd_extract_data, which synchronously calls rndsinks_distribute,
which takes rndsinks_lock -> boom.

This is a stop-gap on a stop-gap on a stop-gap; we really ought to
back out all of these stop-gaps, make bcm2835_rng call rnd_add_data
asynchronously to work around the original symptom, and design a real
solution when we have time to sort this mess out properly.

(riastradh)

Use PCI_BAR(m), not PCI_MAPREG_START + m, which is wrong.

(riastradh)

Don't cancel rps.work or hotplug_work in i915_driver_unload.

These are cancelled in intel_modeset_cleanup.

(riastradh)

Call drm_vblank_cleanup in drm_detach.

(riastradh)

Initialize mm.retire_work in i915_gem_load stub.

(riastradh)

Use mdelay for msleep of less than one tick.

(riastradh)

Destroy locks in i915_driver_unload.

(riastradh)

Ifdef out gtt unmapping for now in i915_driver_unload.

(riastradh)

Enable drm_debug by default for now.

(riastradh)

Clean up the waitqueue and locks in drm_vblank_cleanup.

(riastradh)

Destroy mode config locks in drm_mode_config_cleanup.

(riastradh)

Rework error branches of idr_get_new_above.

Return ENOSPC if we're about to overflow int.

(riastradh)

Actually return the new id in idr_get_new_above.

(riastradh)

Don't redundantly cancel mm.retire_work or error_work.

(riastradh)

Kludge up i915 stubs and start using the real i915_irq.c.

(riastradh)

Allow kfree(NULL).

(riastradh)

Initialize dev->dev, dev->primary, and dev->control in drm_attach.

(riastradh)

Convert i915_irq.c to use DRM_IRQ_ARGS in irq handlers.

(riastradh)

Add assert_spin_locked to <linux/spinlock.h>.

(riastradh)

Stub out intel_gtt initialization until we' ready for gem stuff.

(riastradh)

Stub out the content, not definition, of intel_console_resume.

(riastradh)

Fix sense of kassert in kcalloc.

(riastradh)

Add i915_hangcheck_elapsed stub.

(riastradh)

Add intel_irq_init stub.

(riastradh)

Add intel_gt_init stub.

(riastradh)

Add i1915_setup_sysfs stub.

(riastradh)

Use a non-NULL value for bogus <linux/workqueue.h> stubs.

(riastradh)

Initialize PCI memory maps and bus space/dma tags in drm_pci_attach.

(riastradh)

Let drm_fill_in_dev initialize dev->driver after all.

(riastradh)

Little stab at making drm_attach fail more gracefully.

(riastradh)

Mega-kludge: reverse sense of PAGE_MASK in <linux/mm.h>.

(riastradh)

Offset mmio_bar by PCI_MAPREG_START in i915_dma_attach.

(riastradh)

Set num_ioctls and enable DRIVER_MODESET in i915 module init.

(riastradh)

Call drm_pci_init/exit on i915drm2 module load/unload.

(riastradh)

Move driver->bus setting to drm_pci_init.

(riastradh)

Fill in files.drm2 a little.

(riastradh)

Call dev->driver->load in drm_attach.

(riastradh)

Pass drm_config_found arguments via explicit struct drm_attach_args.

(riastradh)

Fill in drm_detach some more.

(riastradh)

Destroy the idr in drm_ctxbitmap_cleanup.

(riastradh)

Destroy mm->unused_lock in drm_mm_takedown.

(riastradh)

Manage an opencount for each drm device to block drm_detach.

(riastradh)

Initialize sc->sc_drm_dev in drm_attach.

(riastradh)

Fix up drm attachment code.

(riastradh)

Destroy gmbus_mutex in intel_teardown_gmbus.

(riastradh)

Make drm_pci_agp_init succeed for now.

(riastradh)

Initialize sc->sc_drm_dev->driver = i915_drm_driver in i915drm_attach.

(riastradh)

Constify bus member of struct drm_driver.

(riastradh)

Initialize dev->driver->bus in drm_pci_attach.

(riastradh)

Provisional NO_DRM config for testing with a module.

Too much trouble to keep this as an uncommitted change...

(riastradh)

Stub out ironlake_teardown_rc6 in intel_pm.c too.

(riastradh)

Add drm_sysfs stubs.

(riastradh)

Stub out i915_{gem,irq,sysfs}.c and intel_{fb,pm,ringbuffer}.c for now.

(riastradh)

Provisionally ifdef out some kobject_uevent stuff in i915_irq.c.

(riastradh)

i915_irq.c needs <linux/printk.h.

(riastradh)

Add pr_err as an alias for printf to drm2 <linux/printk.h>.

(riastradh)

spin_(un)lock_irq in drm2 <linux/spinlock.h>.

(riastradh)

Bogus definition of round_jiffies_up in drm2 <linux/timer.h>.

(riastradh)

drm2 module is still called `drm' for now, so fix i915drm2 depends.

(riastradh)

Add forgotten <linux/notifier.h>.

(riastradh)

Add stub <drm/intel-gtt.h>, forgotten a while ago.

(riastradh)

Kludgerously define __user to be empty in Linux uapi <drm.h>.

(riastradh)

Constify intel_tv.c.

(riastradh)

intel_tv.c needs <linux/math64.h>.

(riastradh)

Use a static const, not `= {0}', to zero-initialize in intel_sdvo.c.

(riastradh)

Fix const casts in intel_sdvo.c.

(riastradh)

intel_sdvo.c needs <linux/bitops.h> and <linux/module.h>.

(riastradh)

Provisionally add kludgerific ioread32 to intel_panel.c.

(riastradh)

Kludge around max as a local variable in intel_panel.c.

(riastradh)

intel_panel.c needs <linux/module.h> and <linux/printk.h>.

(riastradh)

Implement drm_getsarea in drm_drv.c.

(riastradh)

Implement non-`atomic' kmap/kunmap in drm2 Linux shims.

Uses uvm_km_alloc and pmap_kenter_pa.  Does not use direct mappings
on architectures that support this (e.g., amd64).

(riastradh)

Add pr_warn_once as a bogus alias for printf in <linux/printk.h>.

(riastradh)

Add bogus struct device dev member to <linux/pci.h> struct pci_dev.

(riastradh)

Add div_u64 to <linux/math64.h>.

(riastradh)

Implement bogus max_t in <linux/kernel.h>.

(riastradh)

Add ENOTSUPP as an alias for ENOTSUP (XXX huh?) in <linux/err.h>.

(riastradh)

Make udelay a static inline in <linux/delay.h>.

(riastradh)

Implement hweight16 in <linux/bitops.h>.

(riastradh)

Define acpi_os_iounmap as AcpiOsUnmapMemory in <linux/acpi_io.h>.

(riastradh)

Define memcpy_*io as memcpy in drm2 Linux <asm/io.h>.

These work only for memory-mapped I/O anyway.

(riastradh)

Implement BUILD_BUG_ON via CTASSERT in drm2 Linux <asm/bug.h>.

(riastradh)

Shim acpi_lid_notifier implementation in drm2 Linux <acpi/button.h>.

(riastradh)

Unhook intel_pm.c and intel_ringbuffer.c for now.

(riastradh)

Fix typo in intel_ringbuffer: drm_iounmap takes dev_priv->dev, not dev.

(riastradh)

Unhook intel_fb.c from the build until rewritten for wscons.

(riastradh)

Unhook i915_gem*.c from the build for now.

(riastradh)

Use acpi_os_iounmap in intel_opregion.c.

(riastradh)

Avoid void * arithmetic in intel_opregion.c.

(riastradh)

intel_opregion.c needs <linux/err.h>.

(riastradh)

Provisionally ifdef out dmi hacks in intel_lvds.c.

(riastradh)

intel_lvds.c needs <linux/err.h> and <linux/notifier.h>.

(riastradh)

Convert intel_i2c.c to linux_mutex_init.

(riastradh)

intel_i2c.c needs <linux/module.h>.

(riastradh)

Include <linux/acpi_io.h> for __acpi_iomem.

(riastradh)

Constify drm_mode_create_tv_properties.

(riastradh)

Add bogus x86-specific mmiowb to Linux <asm/io.h>.

(riastradh)

intel_dp.c needs <linux/err.h>, <linux/module.h>, and <linux/printk.h>.

(riastradh)

NetBSDify (bus_spaceify) i915_disable_vga.  (XXX Move elsewhere?)

(riastradh)

Half-arsed conversion of i915 pending flip queue to drm_waitqueues.

(riastradh)

Avoid {0} struct initializer in intel_dp_init_connector.

(riastradh)

Avoid {0} struct initializer in ironlake_set_m_n in intel_display.c.

(riastradh)

Ifdef out Linux framebuffer stuff in i915 intel_display.c.

(riastradh)

i915 intel_display.c needs <linux/err.h> and Linux <asm/bug.h>.

(riastradh)

Add kludgey dev field to struct i2c_adapter for intel_dp.

(riastradh)

Add stubs for flush_workqueue and flush_scheduled_work that warn.

(riastradh)

Add EREMOTEIO as an alias for EIO to <linux/err.h>.

(riastradh)

Add min_t and DIV_ROUND_UP to <linux/kernel.h>.

(riastradh)

Add mdelay to <linux/delay.h>.

(riastradh)

Add several operations to <linux/atomic.h>.

atomic_add_unless
atomic_clear_mask
atomic_inc_not_zero
atomic_set_mask
atomic_sub

(riastradh)

Define WARN_ONCE in Linux <asm/bug.h>.

(riastradh)