Pul drm fixes from Dave Airlie:
"Regular weekly fixes, mostly amdgpu and xe. One nouveau fix is a
better fix for the deadlock and also helps with a sync race we were
seeing.
dma-buf:
- heaps CMA page accounting fix
virtio-gpu:
- fix segment size
xe:
- A crash fix
- A fix for an assert due to missing mem_acces ref
- Only allow a single user-fence per exec / bind.
- Some sparse warning fixes
- Two fixes for compilation failures on various odd combinations of
gcc / arch pointed out on LKML.
- Fix a fragile partial allocation pointed out on LKML.
- A sysfs ABI documentation warning fix
amdgpu:
- Fix reboot issue seen on some 7000 series dGPUs
- Fix client init order for KFD
- Misc display fixes
- USB-C fix
- DCN 3.5 fixes
- Fix issues with GPU scheduler and GPU reset
- GPU firmware loading fix
- Misc fixes
- GC 11.5 fix
- VCN 4.0.5 fix
- IH overflow fix
amdkfd:
- SVM fixes
- Trap handler fix
- Fix device permission lookup
- Properly reserve BO before validating it
nouveau:
- fence/irq lock deadlock fix (second attempt)
- gsp command size fix
* tag 'drm-fixes-2024-02-03' of git://anongit.freedesktop.org/drm/drm: (35 commits)
nouveau: offload fence uevents work to workqueue
nouveau/gsp: use correct size for registry rpc.
drm/amdgpu/pm: Use inline function for IP version check
drm/hwmon: Fix abi doc warnings
drm/xe: Make all GuC ABI shift values unsigned
drm/xe/vm: Subclass userptr vmas
drm/xe: Use LRC prefix rather than CTX prefix in lrc desc defines
drm/xe: Don't use __user error pointers
drm/xe: Annotate mcr_[un]lock()
drm/xe: Only allow 1 ufence per exec / bind IOCTL
drm/xe: Grab mem_access when disabling C6 on skip_guc_pc platforms
drm/xe: Fix crash in trace_dma_fence_init()
drm/amdgpu: Reset IH OVERFLOW_CLEAR bit
drm/amdgpu: remove asymmetrical irq disabling in vcn 4.0.5 suspend
drm/amdgpu: drm/amdgpu: remove golden setting for gfx 11.5.0
drm/amdkfd: reserve the BO before validating it
drm/amdgpu: Fix missing error code in 'gmc_v6/7/8/9_0_hw_init()'
drm/amd/display: Fix buffer overflow in 'get_host_router_total_dp_tunnel_bw()'
drm/amd/display: Add NULL check for kzalloc in 'amdgpu_dm_atomic_commit_tail()'
drm/amd: Don't init MEC2 firmware when it fails to load
...
Timur pointed this out before, and it just slipped my mind,
but this might help some things work better, around pcie power
management.
Fixes: 8d55b0a940 ("nouveau/gsp: add some basic registry entries.")
Signed-off-by: Dave Airlie <airlied@redhat.com>
Link: https://patchwork.freedesktop.org/patch/576336/
UAPI Changes:
- Only allow a single user-fence per exec / bind.
The reason for this clarification fix is a limitation in the implementation
which can be lifted moving forward, if needed.
Driver Changes:
- A crash fix
- A fix for an assert due to missing mem_acces ref
- Only allow a single user-fence per exec / bind.
- Some sparse warning fixes
- Two fixes for compilation failures on various odd
combinations of gcc / arch pointed out on LKML.
- Fix a fragile partial allocation pointed out on LKML.
Cross-driver Change:
- A sysfs ABI documentation warning fix
This also touches i915 and is acked by i915 maintainers.
Signed-off-by: Dave Airlie <airlied@redhat.com>
From: Thomas Hellstrom <thomas.hellstrom@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/ZbuCYdMDVK-kAWC5@fedora
The error message buffer overflow 'dc->links' 12 <= 12 suggests that the
code is trying to access an element of the dc->links array that is
beyond its bounds. In C, arrays are zero-indexed, so an array with 12
elements has valid indices from 0 to 11. Trying to access dc->links[12]
would be an attempt to access the 13th element of a 12-element array,
which is a buffer overflow.
To fix this, ensure that the loop does not go beyond the last valid
index when accessing dc->links[i + 1] by subtracting 1 from the loop
condition.
This would ensure that i + 1 is always a valid index in the array.
Fixes the below:
drivers/gpu/drm/amd/amdgpu/../display/dc/link/protocols/link_dp_dpia_bw.c:208 get_host_router_total_dp_tunnel_bw() error: buffer overflow 'dc->links' 12 <= 12
Fixes: 59f1622a5f ("drm/amd/display: Add dpia display mode validation logic")
Cc: PeiChen Huang <peichen.huang@amd.com>
Cc: Aric Cyr <aric.cyr@amd.com>
Cc: Rodrigo Siqueira <rodrigo.siqueira@amd.com>
Cc: Aurabindo Pillai <aurabindo.pillai@amd.com>
Cc: Meenakshikumar Somasundaram <meenakshikumar.somasundaram@amd.com>
Signed-off-by: Srinivasan Shanmugam <srinivasan.shanmugam@amd.com>
Reviewed-by: Tom Chung <chiahsuan.chung@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Fix the warning info below during mode1 reset.
[ +0.000004] Call Trace:
[ +0.000004] <TASK>
[ +0.000006] ? show_regs+0x6e/0x80
[ +0.000011] ? __flush_work.isra.0+0x2e8/0x390
[ +0.000005] ? __warn+0x91/0x150
[ +0.000009] ? __flush_work.isra.0+0x2e8/0x390
[ +0.000006] ? report_bug+0x19d/0x1b0
[ +0.000013] ? handle_bug+0x46/0x80
[ +0.000012] ? exc_invalid_op+0x1d/0x80
[ +0.000011] ? asm_exc_invalid_op+0x1f/0x30
[ +0.000014] ? __flush_work.isra.0+0x2e8/0x390
[ +0.000007] ? __flush_work.isra.0+0x208/0x390
[ +0.000007] ? _prb_read_valid+0x216/0x290
[ +0.000008] __cancel_work_timer+0x11d/0x1a0
[ +0.000007] ? try_to_grab_pending+0xe8/0x190
[ +0.000012] cancel_work_sync+0x14/0x20
[ +0.000008] amddrm_sched_stop+0x3c/0x1d0 [amd_sched]
[ +0.000032] amdgpu_device_gpu_recover+0x29a/0xe90 [amdgpu]
This warning info was printed after applying the patch
"drm/sched: Convert drm scheduler to use a work queue rather than kthread".
The root cause is that amdgpu driver tries to use the uninitialized
work_struct in the struct drm_gpu_scheduler
v2:
- Rename the function to amdgpu_ring_sched_ready and move it to
amdgpu_ring.c (Alex)
v3:
- Fix a few more checks based on Vitaly's patch (Alex)
v4:
- squash in fix noticed by Bert in
https://gitlab.freedesktop.org/drm/amd/-/issues/3139
Fixes: 11b3b9f461 ("drm/sched: Check scheduler ready before calling timeout handling")
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Vitaly Prosyak <vitaly.prosyak@amd.com>
Signed-off-by: Ma Jun <Jun.Ma2@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
[Why]
On 14us for exit latency time causes underflow for 8K monitor with HDR on.
Increasing the latency to 28us fixes the underflow.
[How]
Increase the latency to 28us. This workaround should be sufficient
before we figure out why SR exit so long.
Reviewed-by: Chaitanya Dhere <chaitanya.dhere@amd.com>
Acked-by: Tom Chung <chiahsuan.chung@amd.com>
Signed-off-by: Nicholas Susanto <nicholas.susanto@amd.com>
Tested-by: Daniel Wheeler <daniel.wheeler@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
[why]
Originally, PMFW said min FCLK is 300Mhz, but min DCFCLK can be increased
to 400Mhz because min FCLK is now 600Mhz so FCLK >= 1.5 * DCFCLK hardware
requirement will still be satisfied. Increasing min DCFCLK addresses
underflow issues (underflow occurs when phantom pipe is turned on for some
Sub-Viewport configs).
[how]
Increasing DCFCLK by raising the min_dcfclk_mhz
Reviewed-by: Chaitanya Dhere <chaitanya.dhere@amd.com>
Reviewed-by: Alvin Lee <alvin.lee2@amd.com>
Acked-by: Tom Chung <chiahsuan.chung@amd.com>
Signed-off-by: Sohaib Nadeem <sohaib.nadeem@amd.com>
Tested-by: Daniel Wheeler <daniel.wheeler@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Partial migration to system memory should use migrate.addr, not
prange->start as virtual address to allocate system memory page.
Fixes: a546a27684 ("drm/amdkfd: Use partial migrations/mapping for GPU/CPU page faults in SVM")
Signed-off-by: Philip Yang <Philip.Yang@amd.com>
Reviewed-by: Xiaogang Chen <Xiaogang.Chen@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
This patch is to eliminate interrupt warning below:
"[drm] Fence fallback timer expired on ring sdma0.0".
An early vm pt clearing job is sent to SDMA ahead of interrupt enabled.
And re-locating the drm client creation following after drm_dev_register
looks like a more proper flow.
v2: wrap the drm client creation
Fixes: 1819200166 ("drm/amdkfd: Export DMABufs from KFD using GEM handles")
Signed-off-by: Le Ma <le.ma@amd.com>
Reviewed-by: Felix Kuehling <felix.kuehling@amd.com>
Reviewed-by: Lijo Lazar <lijo.lazar@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
This reverts commit eacabb5462.
This commit causes some regressions in desktop usage, this will
reintroduce the original deadlock in DRI_PRIME situations, I've
got an idea to fix it by offloading to a workqueue in a different
spot, however this code has a race condition where we sometimes
miss interrupts so I'd like to fix that as well.
Cc: stable@vger.kernel.org
Signed-off-by: Dave Airlie <airlied@redhat.com>
The FORCE_STOP_STATE bit is unsuitable to force the DSI link into LP-11
mode. It seems the bridge internally queues DSI packets and when the
FORCE_STOP_STATE bit is cleared, they are sent in close succession
without any useful timing (this also means that the DSI lanes won't go
into LP-11 mode). The length of this gibberish varies between 1ms and
5ms. This sometimes breaks an attached bridge (TI SN65DSI84 in this
case). In our case, the bridge will fail in about 1 per 500 reboots.
The FORCE_STOP_STATE handling was introduced to have the DSI lanes in
LP-11 state during the .pre_enable phase. But as it turns out, none of
this is needed at all. Between samsung_dsim_init() and
samsung_dsim_set_display_enable() the lanes are already in LP-11 mode.
The code as it was before commit 20c827683d ("drm: bridge:
samsung-dsim: Fix init during host transfer") and 0c14d31306 ("drm:
bridge: samsung-dsim: Fix i.MX8M enable flow to meet spec") was correct
in this regard.
This patch basically reverts both commits. It was tested on an i.MX8M
SoC with an SN65DSI84 bridge. The signals were probed and the DSI
packets were decoded during initialization and link start-up. After this
patch the first DSI packet on the link is a VSYNC packet and the timing
is correct.
Command mode between .pre_enable and .enable was also briefly tested by
a quick hack. There was no DSI link partner which would have responded,
but it was made sure the DSI packet was send on the link. As a side
note, the command mode seems to just work in HS mode. I couldn't find
that the bridge will handle commands in LP mode.
Fixes: 20c827683d ("drm: bridge: samsung-dsim: Fix init during host transfer")
Fixes: 0c14d31306 ("drm: bridge: samsung-dsim: Fix i.MX8M enable flow to meet spec")
Signed-off-by: Michael Walle <mwalle@kernel.org>
Signed-off-by: Inki Dae <inki.dae@samsung.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20231113164344.1612602-1-mwalle@kernel.org
[Why]
IPS was temporary disabled due to instability.
It was fixed in dmub firmware and with:
- "drm/amd/display: Add IPS checks before dcn register access"
- "drm/amd/display: Disable ips before dc interrupt setting"
[How]
Enable IPS by default.
Disable IPS if 0x800 bit set in amdgpu.dcdebugmask module params
Signed-off-by: Roman Li <Roman.Li@amd.com>
Tested-by: Mark Broadworth <mark.broadworth@amd.com>
Reviewed-by: Mario Limonciello <mario.limonciello@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
[Why]
While in IPS2 an access to dcn registers is not allowed.
If interrupt results in dc call, we should disable IPS.
[How]
Safeguard register access in IPS2 by disabling idle optimization
before calling dc interrupt setting api.
Signed-off-by: Roman Li <Roman.Li@amd.com>
Tested-by: Mark Broadworth <mark.broadworth@amd.com>
Reviewed-by: Mario Limonciello <mario.limonciello@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
[Why]
Because ABM will wait VStart to start getting histogram data,
it will cause we can't enter IPS while full screnn video playing.
[How]
Modify the panel refresh rate to the maximun multiple of current
refresh rate.
Reviewed-by: Dennis Chan <dennis.chan@amd.com>
Acked-by: Roman Li <roman.li@amd.com>
Signed-off-by: ChunTao Tso <chuntao.tso@amd.com>
Tested-by: Daniel Wheeler <daniel.wheeler@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
