drm: Allow drm_atomic_helper_wait_for_vblanks while cold.

This enables us to do a mode-switch before interrupts are running.

(riastradh)

nouveau: Disable MSI on G84 devices.

Fails to switch modes with MSI, works without MSI, on T61p.

(riastradh)

rkdrm: Convert to atomic modesetting, as needed for bridges.

Author: Jared McNeill <jmcneill@invisible.ca>
Committer: Taylor R Campbell <riastradh@NetBSD.org>

(riastradh)

rkdrm: Do drm_mode_config_reset on init.

(riastradh)

drm: Use {} instead of [] for drm prints.

Using [] confuses dmesg timestamp parsing.  We should maybe fix that,
but this will serve for now.

(riastradh)

i915: For now, pretend gtt size is zero.

Something goes wrong with EXEC_OBJECT_PINNED in Mesa i965 brw bufmgr
if we expose this parameter; this is a sleazy workaround pending
diagnosis of the underlying problem.

(riastradh)

drm: Work around busted kthread_join.

(riastradh)

drm: Don't try to use sched->thread until it's initialized.

(riastradh)

drm: Fix error return for kthread_run: error pointer, not null.

(riastradh)

drm: Omit needless local brace diff.

(riastradh)

drm: Take advantage of kthread interlock to reduce diff.

(riastradh)

drm: Rework Linux `kthread' abstraction to avoid race to sleep.

Requires passing in the caller's lock and condvar to kthread_run, but
for the one user that appears not to be an onerous requirement.

(riastradh)

drm: Trigger `spurious' wakeup for kthread_stop too.

(riastradh)

drm: Mark variable as used only by trace point.

(riastradh)

drm: add drm_dp_aux_fini; use it in i915 to fix missing mutex destroy.

Author: chuq <chs@NetBSD.org>
Committer: Taylor R Campbell <riastradh@NetBSD.org>

(riastradh)

drm: Omit bogus spin locking around kthread_stop, which blocks.

Blocking is not allowed while holding a spin lock, and there's
nothing here that needs a lock either.

(riastradh)

amdgpu: Don't unmap doorbells if we didn't map them.

(riastradh)

amdgpu: Fix register offsets in amdgpu_mm_rreg8/wreg8.

We already added rmmio_base when mapping rmmioh; let's not add it
again!

(riastradh)

drm: Fix comments about wait return values.

(riastradh)

drm: Ensure DRM_*WAIT* never counts down from MAX_SCHEDULE_TIMEOUT.

This faux timeout means `wait forever'.

(riastradh)

i915: Use splx, not splraise, to restore IPL after splsoftserial.

(riastradh)

drm: Switch sense of #if __linux__ conditionals to use __NetBSD__.

Keep it consistent with the rest of drm.

(riastradh)

i915: Fix some timeout return values in gmbus and dp logic.

(riastradh)

nouveau: fix a "can return -1 to userland" problem.

nvkm_ioctl_del() for some reason converts 0 into 1 return.  this
appears to be a way to avoid having a log printed by nvkm_ioctl()
and to trigger some dead code ("hack" is always NULL afaict.)
after consuming this value convert it back to 0 for ioctl return.

Author: phone <mrg@NetBSD.org>
Committer: Taylor R Campbell <riastradh@NetBSD.org>

(riastradh)

linux: Fix rcu_barrier so it actually waits for everything.

(riastradh)

drm: Make dma fence array name strings match Linux.

(riastradh)

drm: Make scheduler mandatory to make amdgpu module simpler.

(riastradh)

drm: Use atomic_load_consume/relaxed to simplify code.

(riastradh)

drm: Implement dma fence chains.

(riastradh)

drm: Support 64-bit fence context and sequence numbers.

(riastradh)

drm: Release fence in dma_fence_chain_walk.

(riastradh)

amdgpu: New kernel module build goo.

(riastradh)

linux: In kthread_park, wake thread to re-check kthread_shouldpark.

Otherwise, the thread to pakr may be sleeping on some condvar not
noticing it has to notify kthread_park --> deadlock.

(riastradh)

amdgpu: Try turning on MSI.

(riastradh)

amdgpu: Read amdgpu_ctx_entity::sequence under the lock.

(riastradh)

drm: Make sure dma_fence_wait gets traced too.  Deduplicate.

(riastradh)

amdgpu: store process and lwp info for netbsd in amdgpu taskinfo.

Author: phone <mrg@NetBSD.org>
Committer: Taylor R Campbell <riastradh@NetBSD.org>

(riastradh)

drm: Add dtrace probes to fences.

(riastradh)

drm: Fix signalling return value if no enable_signaling op.

(riastradh)

amdgpu: Reduce unhelpful sprintf diffs.

(riastradh)

Destroy lock before freeing

Author: Maya Rashish <maya@NetBSD.org>
Committer: Taylor R Campbell <riastradh@NetBSD.org>

(riastradh)

i915: Make intel_gtt_get prototype match Linux.

(riastradh)

i915: Omit spurious compile-time assertion.

The assertion failed on i386 with our struct drm_fence_callback.
Could change drm_fence_callback to be a little smaller (use the
nullness of the previous pointer instead of a separate boolean to
detect whether it's on the list, perhaps).  But it's unclear why the
relative sizes of these members matter; they're in a union.  Upstream
commit message doesn't illuminate.

(riastradh)

i915: Suppress spurious -Wmaybe-uninitialized warnings.

(riastradh)

drm: Make resource_size return resource_size_t like Linux.

Because Linux uses resource_size_t for absolute bus addresses, it is
actually bus_addr_t rather than bus_size_t.  So using bus_size_t here
causes trouble with the i915 range_overflows macro, which verifies
that types match.

(riastradh)

amdgpu: Add a cautionary comment about non-LP64 branch.

(riastradh)

drm: Work around bus_space_read_8 on LP32.

(riastradh)

amdgpu: Fix indexing for 64-bit doorbells.

(riastradh)

drm: Must hold event_lock around wait on event_queue.

We must not have hit this path before if it never crashed!

(riastradh)

drm: Spruce up ww_mutex comments.  Audit return values.

(riastradh)

i915: Fix destroy/free order for i915 gem stub fence.

(riastradh)

linux: Fix wait_bit semantics.

- wait_on_bit is supposed to wait until the bit is cleared, not set.

- wait_on_bit_timeout is supposed to return 0 on success, -EAGAIN on
  faiure.

Omit wake_up_bit; nothing uses it and clear_and_wake_up_bit is a more
semantically coherent operation.

(riastradh)

drm: Fix comment about dma_resv_lock_* return values.

(riastradh)

drm: Timeout audit -- linux/sched.h.

(riastradh)

drm: Timeout audit -- drm_syncobj.c.

(riastradh)

linux: Timeout audit -- linux/completion.h.

(riastradh)

drm: Make sure to initialize syncobj wait point.

(riastradh)

drm: Use atomic_load_relaxed for fence->flags when not locked.

Shouldn't make a semantic difference on any machines we run on, but
will make sanitizers happier and emphasize to readers that the result
may not be reliable at these points.

(riastradh)

i915: Terminate inner loop if we hit end of view.

(riastradh)

i915: Assert output segment index.

(riastradh)

drm: Assert dma_fence_wait internal return value.

(riastradh)

i915: Tighten assertion about segment index.

(riastradh)

drm: Fix drm_symstr with pserialize.

(riastradh)

i915: Fix locking-against-self in i915_request_wait.

(riastradh)

drm/scheduler: Release job list lock arond kthread_parkme.

(riastradh)

drm: Rework timeout return code logic.

(riastradh)

drm: Rework dma_fence_wait_any_timeout.

- Yield the first index, not the most recently signalled one.
- Check again after last wakeup.
- Make sure to return positive for nontimeout.
- Handle ip=null.

(riastradh)

i915: Fix wakeref destruction.

(riastradh)

drm: Fix semantics of del_timer, del_timer_sync.

(riastradh)

drm: Paranoia: handle fencep=null, fence!=null, and 0 shared.

Not sure this is possible but I don't know the API well enough to
prove it can't happen.

(riastradh)

drm: Fix missing restart on race in dma_resv_get_fences_rcu.

(riastradh)

drm: Another missing spin_lock_destroy.

(riastradh)

amdgpu: call pmap_pv_track/untrack for the graphics aperture.

Author: phone <mrg@NetBSD.org>
Committer: Taylor R Campbell <riastradh@NetBSD.org>

(riastradh)

drm: Fix an even number of sign errors in interval_tree_generic.h.

(riastradh)

amdgpu: more destroy fixes.

Author: phone <mrg@NetBSD.org>
Committer: Taylor R Campbell <riastradh@NetBSD.org>

(riastradh)

drm: Support dma_resv_reserve_shared(..., num_fences>1).

(riastradh)

drm/scheduler: spin_lock_init/destroy audit

(riastradh)

i915: mutex_init/destroy and spin_lock_init/destroy audit

(riastradh)

drm: mutex_init/destroy and spin_lock_init/destroy audit

(riastradh)

amdgpu: mutex_init/destroy and spin_lock_init/destroy audit

(riastradh)

drm: Implement drm_resv_get_fences_rcu with null fencep.

(riastradh)

nouveau: Remove post-merge audit comments -- LGTM.

(riastradh)

nouveau: Map write-combining in instmem/nv40 like Linux does.

(riastradh)

drm: Set fence->ops for stub fences too.

(riastradh)

amdgpu: Call idr_preload(_end) around idr_alloc, outside lock.

(riastradh)

drm: add destruction.

Author: phone <mrg@NetBSD.org>
Committer: Taylor R Campbell <riastradh@NetBSD.org>

(riastradh)

amdgpu: Run module initialization.

(riastradh)

drm: Fill in fake connector->kdev pointing at the drm device.

Hope this will make progress -- hope it won't screw up too badly if
two connectors have the same kdev!

(riastradh)

drm: Make via build again.

(riastradh)

amdgpu: Disable atomic modeset like upstream does for now.

(riastradh)

drm: Make some legacy UMS crap build.

(riastradh)

drm: Avoid locking against self in scheduler.

(riastradh)

drm: Avoid potential integer overflow in pages->bytes conversion.

(riastradh)

linux: Make kthread_park/stop work at IPL_VM.

Needed in order to hold spin locks around it.

(riastradh)

amdgpu: Set a sensible maxsegsz for ih ring DMA map.

We need this to fit in one contiguous segment, so the maxsegsz had
better be at least the size we ask for, not something much smaller!

(riastradh)

amdgpu: Fix bogus assertion about number of DMA segments.

(riastradh)

drm/ttm: Fix ttm_tt pages and DMA map allocation.

(riastradh)

drm: Downgrade warnings from aprint_error to aprint_normal.

(riastradh)

i915: Backport fix for locking-against-self on attach.

commit d1b2828af0cc414356c18d7814b83ba33b472054
Author: Ville Syrj辰l辰 <ville.syrjala@linux.intel.com>
Date:  Wed Jan 22 22:43:29 2020 +0200

    drm/i915: Fix modeset locks in sanitize_watermarks()

    We've added more internal things that use modeset locks and
    thus we need to be prepared for intel_atomic_check() grabbing
    more locks than what our initial drm_modeset_lock_all_ctx()
    took. So we're missing the backoff handling here.

    Also drm_atomic_helper_duplicate_state() works against us
    by clearing state->acquire_ctx in anticipation of
    drm_atomic_helper_commit_duplicated_state() being used to
    commit the state.

    We could probably just reset acquire_ctx back, but instead
    let's just rewrite the whole thing without using either of
    those "helpers". There's also no need to add any connectors
    to the state here since we just want the new watermarks
    which don't depend on connectors.

    Cc: Chris Wilson <chris@chris-wilson.co.uk>
    Signed-off-by: Ville Syrj辰l辰 <ville.syrjala@linux.intel.com>
    Link: https://patchwork.freedesktop.org/patch/msgid/20200122204329.2477-1-ville.syrjala@linux.intel.com
    Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>

(riastradh)

sunxi/drm, rockchip/drm: Avoid messy error output.

(riastradh)

rockchip/drm: use drm_helper_mode_fill_fb_struct or say why not.

Author: phone <mrg@NetBSD.org>
Committer: Taylor R Campbell <riastradh@NetBSD.org>

(riastradh)

rockchip/drm: use an explicit task queue to avoid config_defer pitfalls.

Author: phone <mrg@NetBSD.org>
Committer: Taylor R Campbell <riastradh@NetBSD.org>

(riastradh)

sunxi/drm: Use an explicit task queue to avoid config_defer pitfalls.

Same as the other drm drivers.

(riastradh)

drm: Rework attach/detach and deferred task logic.

- Reduce the number of states the softc can be in.
- Fix races between attach and other threads.

(riastradh)

drm: Move io_mapping to very own .c file.

(riastradh)

drm: Allow multiple concurrent io_mapping_map_wc on the same space.

(This should maybe check that there aren't overlapping maps of the
same parts of the space.)

(riastradh)

linux: Downgrade warning about flushing from workqueue to debug.

(riastradh)

i915: Fix gen6 pd_vma_bind.

We only need space for the ppgtt in this vma, not space for the whole
VM.

(riastradh)

i915: Spruce up assertion in gen6_ppgtt.c pd_vma_bind.

(riastradh)

i915: Use pmap_pv_protect on the paddrs for unmap_mapping_range.

(riastradh)

i915: Fill out i915_gem_phys_pwrite.

(riastradh)

i915: Figure out the graphics stolen memory base.

(riastradh)

i915: Rearrange how mmap offsets work.

Rather than use the magic offsets in each GEM object's uvm object
itself, just create a separate uvm object for each mapping type and
start at zero within each uvm object.  This should avoid problems
with things like ubc_uiomove which don't know to adjust the starting
offset.

(riastradh)

i915: Fix partial object views -- start at the offset.

(riastradh)

drm: Factor dma_fence_put out of multiple dma_resv_do_poll branches.

(riastradh)

drm: Fix dma_resv_test_signaled_rcu and dma_resv_wait_timeout_rcu.

Make them implemented the semantics as documented like Linux does:
only test the exclusive fence if there's no shared fences.

(riastradh)

drm: Make sure to check read ticket after reading shared fences.

Not entirely sure it was a bug not to do that, but let's make it easy
to prove that we got a consistent snapshot of the whole picture.

(riastradh)

drm: Write comment for dma_resv_get_fences_rcu semantics.

(riastradh)

drm: Factor out logic to read snapshot of fences in dma_resv.

Should make auditing a little easier.

(riastradh)

i915: Mark shmem object pages dirty if needed before unwiring them.

Unlikely to fix the graphics corruption we're seeing, though --
should be relevant only if we're swapping.

(riastradh)

i915: Suggest gnats.n.o, not fdo gitlab, for bugs.

(riastradh)

i915: Omit more unnecessary local diffs.

(riastradh)

i915: Restore i915_gem.c shmem_pwrite.

This was ifdef'd out back when it was a fast path shmem_pwrite_fast
using __copy_in_user_inatomic, but the ifdef got merged anyway into a
different function that is not a fast path.

(riastradh)

linux: For uniprocessor, define smp_*mb as __insn_barrier.

Otherwise ordering might not be guaranteed in the event of
preemption.

XXX pullup

(riastradh)

linux: Eliminate read_barrier_depends, smp_read_barrier_depends.

These no longer exist in Linux; instead, Linux now uses READ_ONCE.

(riastradh)

linux: Define mb/wmb/rmb for x86.

These are _not_ SMP barriers; on Linux they specifically mean
mfence/sfence/lfence.  Although membar_sync and mb agree, and
membar_consumer and rmb agree, membar_producer and wmb _do not_ agree
on x86 -- Linux's wmb is specifically sfence, which is important only
for unusual memory types, whereas our membar_producer is limited to
ordinary memory.

(riastradh)

drm: Use drm_vblank_get_locked/unlocked.

(riastradh)

i915: No need for #ifdef __NetBSD__ to sprinkle legitimate const.

(riastradh)

i915: Remove local diffs obviated by -fwrapv.

(riastradh)

amdgpu: Factor out float build options.

(riastradh)

linux: Paranoia: Don't read kt_shouldstop unlocked.

(riastradh)

Avoid kernel assertion mod->mod_required == NULL kern_module.c:910

Author: Maya Rashish <maya@NetBSD.org>
Committer: Taylor R Campbell <riastradh@NetBSD.org>

(riastradh)

linux: ktime_get_real_seconds

(riastradh)

radeon: Fix dma_addressing_limited call.

(riastradh)

drm: Draft dma-fence-array.

(riastradh)

amdgpu: Make amdgpu_file_to_fpriv work.

(riastradh)

drm: In dma_fence_wait_timeout, allow MAX_SCHEDULE_TIMEOUT.

In the Linux API, this means wait forever.

(riastradh)

drm: Print bad timeout in dma_fence_wait_timeout kasserts.

(riastradh)

amdgpu: Stub out irq domain stuff with a finer grain.

(riastradh)

drm: linux kthread stubs

(riastradh)

linux: wait_for_completion

(riastradh)

amdgpu: Tie up some loose ends.

(riastradh)

amdgpu: Thread bus_dma segments through amdgpu_vm.c.

(riastradh)

amdgpu: make most of amdgpu_vm.c build

Not quite done yet -- need to adapt the DMA address logic to use
bus_dmamap_t.

(riastradh)

amdgpu: Delete redundant local amdgpu_trace.h.

(riastradh)

drm: Knuckle under and define sprintf in nbsd-namespace.h.

(riastradh)

linux: INIT_KFIFO

(riastradh)

amdgpu: Another pass through.

amdgpu_object.c
amdgpu_pci.c
amdgpu_pm.c
amdgpu_polaris10_smumgr.c
amdgpu_pp_psm.c
amdgpu_ppatomctrl.c
amdgpu_ppatomfwctrl.c
amdgpu_process_pptables_v1_0.o
amdgpu_psp.o
amdgpu_psp_v10_0.c
amdgpu_psp_v11_0.c
amdgpu_psp_v12_0.c
amdgpu_psp_v3_1.c
amdgpu_ras.c
amdgpu_ras_eeprom.c
amdgpu_rc_calc.c
amdgpu_rc_calc_dpi.c
amdgpu_ring.c
amdgpu_rlc.c
amdgpu_rn_clk_mgr.c
amdgpu_sdma_v3_0.c
amdgpu_sdma_v4_0.c
amdgpu_sdma_v5_0.c
amdgpu_si.c
amdgpu_si_dma.c
amdgpu_si_dpm.c
amdgpu_smu.c
amdgpu_smu10_hwmgr.c
amdgpu_smu7_hwmgr.c
amdgpu_smu8_hwmgr.c
amdgpu_smu8_smumgr.c
amdgpu_smu_v11_0.c (and a sprinkling of consts)
amdgpu_smu_v11_0_i2c.c
amdgpu_soc15.c
amdgpu_tonga_smumgr.c
amdgpu_ttm.c
amdgpu_ucode.c
amdgpu_uvd.c
amdgpu_uvd_v5_0.c
amdgpu_uvd_v6_0.c
amdgpu_uvd_v7_0.c
amdgpu_vce.c
amdgpu_vce_v2_0.c
amdgpu_vce_v4_0.c
amdgpu_vcn_v1_0.c
amdgpu_vcn_v2_0.c
amdgpu_vcn_v2_5.c
amdgpu_vega10_hwmgr.c
amdgpu_vega10_powertune.c
amdgpu_vega10_processpptables.c
amdgpu_vega10_reg_init.c
amdgpu_vega12_hwmgr.c
amdgpu_vega20_hwmgr.c
amdgpu_vega20_ppt.c
amdgpu_vega20_reg_init.c
amdgpu_vi.c
amdgpu_virt.c
amdgpu_vram_mgr.c
amdgpu_xgmi.c -- disable (for now?)
amdgpufb.c

(riastradh)

amdgpu: amdgpu_renoir_ppt.c

This abuses snprintf with SIZE_MAX as a substitute for sprintf, which
would be super-sketchy, except this code is all dead because it's
only reached via sysfs stuff that isn't wired up in NetBSD.

(riastradh)

linux/idr: Remove meaningless stray membar.

(riastradh)

linux: Allow non-VLA structs in struct_size.

Upstream sometimes uses

struct foo {
int x;
short y[1];
};

instead of  short y[]  with struct_size.

(riastradh)

linux: Side-load linux/overflow.h in linux/slab.h.

(riastradh)

linux: Move linux/overflow.h from drm2 to common.

Needed so linux/slab.h can side-load it.

(riastradh)

drm: Fence leak audit.  No functional change intended.

Sprinkle nulling out variables, add kasserts to reflect them, and
propagate their consequences to eliminate dead code.  Should make it
easier to detect similar leak bugs.

(riastradh)

drm: Fix sense of conditional to avoid null pointer dereference.

(riastradh)

drm: dma_fence_get allows fence to be null; fix assertion.

While here, update comments about semantics for dma_fence_get and
dma_fence_get_rcu (which does not allow null fence).

(riastradh)

i915: Destroy lock in fence before freeing fence itself.

Author: Maya Rashish <maya@NetBSD.org>
Committer: Taylor R Campbell <riastradh@NetBSD.org>

(riastradh)

i915: Use i915_gem_object_has_pages, not a simple null test.

The pages pointer may be ERR_PTR(-EWHATEVER) instead of null.

(riastradh)

linux: Side-load asm/io.h in linux/scatterlist.h.

(riastradh)

linux: PCI_BUS_NUM; tweak types for PCI_SLOT/PCI_FUNC.

(riastradh)

drm: Need write lock on vmobjlock for pgo_put.

(riastradh)

i915: Destroy gen8 ppgtt page directory spin lock.

(riastradh)

i915: Destroy page directory lock on free.

(riastradh)

i915: Fix priolist ordering.

(riastradh)

i915: Allow duplicates in active tree, sorted by kva.

Upstream looks like it uses insertion order, but that's not really
convenient.

(riastradh)

i915: Don't need obj->mmo.lock on destruction.

(riastradh)

linux: Allow GFP_NOWAIT too.

(riastradh)

i915: Fix typo: = is not +.

(riastradh)

linux: Make WARN_ONCE actually happen only once.

(riastradh)

i915: Destroy only spin locks we created in gen8 ppgtt.

(riastradh)