In DPG mode jpeg_v5_0_0_ring_reset() takes the DPG stop path, which only
clears the JPEG_PG_MODE bit and never resets the JRBC. A hung ring is not
recovered: the post-reset ring test times out and the driver falls back to
a full MODE1 reset.
Temporarily force the static power-gating path during the reset so the
stop/start sequence power-cycles the JPEG block (JMI soft reset + power
off/on), matching the jpeg_v4_0 reset which has no DPG path.
Acked-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Jesse Zhang <Jesse.Zhang@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
The functions gfx_v11_0_handle_priv_fault and
gfx_v12_0_handle_priv_fault share the same logic for searching and
triggering a scheduler fault on a ring. This patch moves the shared
ring-searching logic to a common function, amdgpu_gfx_handle_priv_fault,
in amdgpu_gfx.c. The hardware-specific decoding of ring IDs remains in
the version-specific files to maintain proper architectural separation.
Signed-off-by: Ulisses Paixao <ulissespaixao@usp.br>
Co-developed-by: Felipe Sousa <felipesousa@usp.br>
Signed-off-by: Felipe Sousa <felipesousa@usp.br>
Reviewed-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Enable IP block soft reset as a GPU recovery method for GFX7
graphics and compute rings.
This improves current user experience on all GFX7 chips:
* On Kaveri and Kabini there is currently no working
GPU recovery method so those chips currently require
the user to manually reset the computer when there
was a hang.
* On Hawaii and Bonaire, the current GPU recovery method
always clears the contents of VRAM, which means that
a buggy (hanging) app can crash the whole graphical
session, which is less than ideal.
Using GFX IP block soft reset means that we can now
have a working recovery on GFX7 APUs and we can also
move on from GFX hangs on dGPUs without crashing the
whole system.
Tested with the "hard_reset_cp_wait" test case from the
Hang Test Suite created by Natalie Vock and Konstantin Seurer.
This Vulkan testcase waits for an event that never occurs,
effectively a WAIT_REG_MEM packet that intentionally hangs.
IP block soft reset can resolve that hang and allow
the rest of the system to move on and keep functioning
without needing a full ASIC reset.
Tested on the following chips:
Bonaire (Radeon HD 7790)
Hawaii (Radeon R9 390X)
Kaveri (A10-7850K)
Signed-off-by: Timur Kristóf <timur.kristof@gmail.com>
Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@igalia.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Use basically the same implementation as GFX8,
except for the GFX7 specific MQD functions.
Reset every block using the GRBM, then proceed
to reset the GRBM and SEM blocks using the SRBM.
Remove the redundant gfx_v7_0_update_cg() function.
The soft reset now calls the clock and powergating
functions of the IP block instead.
Signed-off-by: Timur Kristóf <timur.kristof@gmail.com>
Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@igalia.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
COND_EXEC tells the CP to discard the dwords following it
when its condition is zero (false).
This is useful for GPU recovery because it can help reduce
collateral damage during GFX IP block soft reset, meaning
that it reduces the likelyhood that we fail some jobs which
are not guilty of the hang as the IP block soft reset
mechanism clears the condition before doing the reset.
Signed-off-by: Timur Kristóf <timur.kristof@gmail.com>
Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@igalia.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Clear the WPTR and RPTR at ring initialization.
Additionally clear the ring contents during reset.
After a reset, the ring contents could be "dirty"
and contain packets emitted before the reset.
and thus need to be cleared to prevent the command
processor from executing packets left over in the
ring from before the reset.
Signed-off-by: Timur Kristóf <timur.kristof@gmail.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
This packet is interpreted by the CE (constant engine).
The reason why this packet is emitted is basically to
make sure the CE can't start executing packets from the
next job submission until the current one is finished.
(Note that CE is not utilized by any maintained userspace
driver and is discontinued in new GPUs. It is now also
deprecated in the kernel.)
Implement the emit_switch_buffer() function instead of emitting
them duing emit_ib, emit_pipeline_sync and emit_vm_flush.
It isn't necessary to emit these in both emit_pipeline_sync()
and emit_vm_flush() because amdgpu_vm_flush() already calls
these when calling either of those functions.
Signed-off-by: Timur Kristóf <timur.kristof@gmail.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Return an error code instead of silently failing.
Signed-off-by: Timur Kristóf <timur.kristof@gmail.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
The gfx_v7_0_cp_compute_resume() function should only return
success when all compute rings are actually functional.
This will be especially important for soft reset which needs
this to know whether the reset was successful.
Note that the gfx_v8_0_cp_test_all_rings() function already
does this on GFX8, here we just follow the same idea.
Signed-off-by: Timur Kristóf <timur.kristof@gmail.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Call amdgpu_gfx_mqd_sw_init()/_fini() on GFX7 to initialize and
finalize MQD BOs, just like GFX8 and newer; instead of doing
an ad-hoc BO allocation. Introduce the possibility of backing
up the MQD instead of trying to reinitialize every time.
This solves an issue with GFX IP block soft reset where
all compute rings would hang after the reset.
Rename gfx_v7_0_mqd_deactivate() to gfx_v7_0_deactivate_hqd()
to more closely reflect what it does and for consistency
with the GFX8 code.
Signed-off-by: Timur Kristóf <timur.kristof@gmail.com>
Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@igalia.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
GFX7 supports KIQ, but amdgpu doesn't use it.
Change amdgpu_gfx_mqd_sw_init() to only allocate the MQD BO
for the KIQ on GFX8 and newer (that is, TOPAZ and newer).
This makes amdgpu_gfx_mqd_sw_init() usable on GFX7 without
any further changes to its functionality.
Signed-off-by: Timur Kristóf <timur.kristof@gmail.com>
Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@igalia.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
[Why]
We need to exit PSR/IPS before programming. Before calling DC for
programming in amdgpu_dm_commit_planes(), there's a
vblank_control_workqueue flush. This waits for IPS and PSR exit. (See
drm_vblank_on/off() > amdgpu_dm_crtc_set_vblank() --queue_work()->
amdgpu_dm_crtc_vblank_control_worker())
Prior to the tagged "Fixes:" change, drm_vblank_get() was called before
the workqueue flush. This ordering ensures that PSR exit occurred before
programming. After the "Fixes:" change, drm_vblank_get() is called after
the workqueue flush, leading to programming while idle optimizations are
still active. This can lead to incorrect flip_pending detection used by
vblank event delivery.
[How]
Split the vblank_get() component of `dm_arm_vblank_event()` into
`dm_arm_vblank_event_pre_programming()`, which is called before
programming. Call it before the vblank_control_workqueue flush.
Includes a drive-by cleanup of prepare_flip_isr(): the only caller is
dm_arm_vblank_event() and it's simple enough to roll-in.
v2: Fix checkpatch formatting warning on
drm_arm_vblank_event_pre_programming() arg alignment.
Fixes: f64a9be565 ("drm/amd/display: check GRPH_FLIP status before sending event")
Link: https://gitlab.freedesktop.org/drm/amd/-/work_items/4141#note_3583205
Link: https://gitlab.freedesktop.org/drm/amd/-/work_items/5527
Assisted-by: Codex:gpt-5.6-sol
Assisted-by: Claude:opus-5
Suggested-by: David Weber <weber.aulendorf@gmail.com>
Signed-off-by: Leo Li <sunpeng.li@amd.com>
Reviewed-by: Mario Limonciello (AMD) <superm1@kernel.org>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
SVM BOs are now migrated back to system memory synchronously from the TTM
eviction path (svm_range_evict_svm_bo), so the per-svm_bo eviction fence
is no longer used.
Remove the eviction fence from svm_range_bo, drop the
amdgpu_amdkfd_fence->svm_bo back pointer and the
amdgpu_amdkfd_evict_svm_bo() helper, and stop special-casing svm_bo
fences in the KFD fence enable_signaling and check_mm paths. Embed struct
amdgpu_bo directly in svm_range_bo with a dedicated svm_range_bo_destroy()
callback, and keep the owning mm via mmgrab()/mmdrop() instead of through
the fence.
Signed-off-by: Philip Yang <Philip.Yang@amd.com>
Reviewed-by: Felix Kuehling <felix.kuehling@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
APUs use firmware-owned DPM tables and do not support replacement through
pp_table. Generic callbacks can nevertheless expose the sysfs file and
accept an upload before resetting the power management stack.
Treat pp_table as unsupported on APUs. Use the same platform check in the
get and set paths to hide the file and reject uploads.
Fixes: 289921b03f ("drm/amd/powerplay: implement sysfs of pp_table for smu11 (v2)")
Signed-off-by: Yang Wang <kevinyang.wang@amd.com>
Reviewed-by: Kenneth Feng <kenneth.feng@amd.com>
Reviewed-by: Asad Kamal <asad.kamal@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
The same as the rest of the code, get_ss_info_from_atombios() uses
calc_pll_cs->ctx->logger for logging. But calc_pll_cs->ctx is
initialized only later in calc_pll_max_vco_construct(). Therefore, any
output using DC_LOG_SYNC() leads to a NULL pointer deference in
get_ss_info_from_atombios().
According to Sashiko, the very same problem exists in
dce112_get_pix_clk_dividers() and dcn3_get_pix_clk_dividers() too.
To avoid accessing the NULL context, use clk_src->base.ctx->logger
everywhere. That context in base is initialized earlier in
dce110_clk_src_construct() and dce112_clk_src_construct(). Before
get_ss_info_from_atombios() or Sashiko's get_pix_clk_dividers functions
above are actually called. This is done by redefining DC_LOGGER to
CTX->logger.
Before:
dce110_clk_src_construct() did:
-> sets clk_src->base.ctx = ctx;
-> ss_info_from_atombios_create()
-> get_ss_info_from_atombios() <- uses calc_pll_cs->ctx # BOOM
-> calc_pll_max_vco_construct() <- sets calc_pll_cs->ctx
After:
dce110_clk_src_construct() does:
-> sets clk_src->base.ctx = ctx;
-> ss_info_from_atombios_create()
-> get_ss_info_from_atombios() <- uses clk_src->base.ctx
Closes: https://bugzilla.suse.com/show_bug.cgi?id=1271175
Closes: https://lore.kernel.org/all/a9ee54e6-2413-4156-9bde-d528ae3c63a3@kernel.org/
Fixes: 1296423bf2 ("drm/amd/display: define DC_LOGGER for logger")
Reviewed-by: Bhawanpreet Lakha <bhawanpreet.lakha@amd.com>
Signed-off-by: Jiri Slaby (SUSE) <jirislaby@kernel.org>
Cc: Lakha, Bhawanpreet <Bhawanpreet.Lakha@amd.com>
Cc: Harry Wentland <harry.wentland@amd.com>
Cc: Leo Li <sunpeng.li@amd.com>
Cc: Rodrigo Siqueira <siqueira@igalia.com>
Cc: Alex Deucher <alexander.deucher@amd.com>
Cc: "Christian König" <christian.koenig@amd.com>
Cc: David Airlie <airlied@gmail.com>
Cc: Simona Vetter <simona@ffwll.ch>
Cc: amd-gfx@lists.freedesktop.org
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
svm_range_evict_svm_bo_worker() migrated an SVM BO's pages back to system
memory from a work item that took mmap_read_lock. When an mmap writer was
pending, that read lock blocked behind the writer while the thread
allocating a new migration VRAM BO waited on this BO's eviction fence - a
circular wait that hung the SVM workers.
Evict the SVM BO synchronously from the TTM eviction path
(amdgpu_ttm_bo_eviction_valuable) instead of deferring to a work item.
The BO is already reserved and the lock order is mmap_lock -> BO
reservation, so only trylock the owning process's mmap lock; on
contention return -EBUSY so TTM skips this BO. This removes the eviction
work item and the enable_signaling path, so no worker can block on
mmap_read_lock.
The SVM BO uses AMDGPU_GEM_CREATE_DISCARDABLE, so ttm_bo_evict takes the
pipeline_gutting path and skips allocating a system memory placement.
That would be wasted work, since svm_migrate_vram_to_ram allocates the
system pages and copies the data back itself.
Eviction now migrates ranges directly, so it must serialize with the
owning process: it trylocks migrate_mutex under svm_bo->list_lock before
unlinking the range, and svm_range_free() unlinks the range then waits on
migrate_mutex, so a concurrent eviction cannot free a range under it.
Drop the mm reference with mmput_async so exit_mmap() does not run under
the BO reservation.
Signed-off-by: Philip Yang <Philip.Yang@amd.com>
Reviewed-by: Felix Kuehling <felix.kuehling@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
When a mode reset happens, driver needs to notify the process on that
GPU a reset event is happening. The existing code assumes the process is
using the GPU that is getting mode reset, which is not always true. For
example, on a 8G system, the process may be only using GPU 0~4 but a
mode 2 reset is resetting the all 8 GPUs connected by XGMI. Trying to
find a process on GPU 5~7 will fail, which is fine and should skip the
event signal.
Signed-off-by: Amber Lin <amber.lin@amd.com>
Reviewed-by: David Yat Sin <david.yatsin@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
On APUs, the GTT pool is backed by system RAM, but its size is not bound
to the non-carveout memory that actually backs it. A user can end up
with GTT + VRAM exceeding total physical memory through the following
sequence:
- Have a large non-carveout memory space (~128GB) and accordingly set a
large GTT (~100GB) via the ttm module parameter.
- Lower the non-carveout memory space in BIOS by increasing the UMA
Frame Buffer Size (VRAM) to 64GB.
- The previously set GTT value (~100GB) persists, even though the new
non-carveout space (64GB) can no longer back it.
This leads to a case where kernel reports GTT (100GB) + VRAM (64GB)
despite the sum being greater than total physical memory (128GB).
Cap the GTT size to totalram_pages() on APUs. totalram_pages() already
excludes the VRAM carveout, so the resulting GTT can never exceed the
system RAM that actually backs it.
Signed-off-by: Harkirat Gill <harkirat.gill@amd.com>
Reviewed-by: David Francis <David.Francis@amd.com>
Assisted-by: Claude:claude-opus-4
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
GPU average and input power backends report a mix of whole watts,
milliwatts, Q24.8 watts and decimal-packed fractions. Q24.8 is inherited
from the legacy PowerPlay sensor format. Milliwatts are a more natural unit
for the hwmon and pm_info consumers in amdgpu_pm.c. A common decoder cannot
distinguish these formats, and converting native milliwatts through Q24.8
also loses precision.
Use milliwatts as the internal unit across all PPT and PowerPlay backends.
Decode Q24.8 only at the legacy smu7 input boundary and encode it only for
the raw amdgpu_sensors debugfs interface. This gives hwmon, pm_info and the
sensor ioctl one unambiguous unit while preserving the format used by UMR.
Fixes: 5b79d0482f ("drm/amd/pp: Remove struct pp_gpu_power")
Fixes: 01992b121f ("drm/amd/pm: fix amdgpu_pm_info power display units")
Signed-off-by: Yang Wang <kevinyang.wang@amd.com>
Reviewed-by: Kenneth Feng <kenneth.feng@amd.com>
Reported-by: Lars Nieradzik <l.nieradzik@gmail.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Runtime suspend runs GFX hw_fini and clears perfmon clock gating while
the UMD profile DPM level remains set in software. Re-apply stable
pstate after a successful runtime resume when a profile mode is active.
Signed-off-by: Candice Li <candice.li@amd.com>
Reviewed-by: Hawking Zhang <Hawking.Zhang@amd.com>
Reviewed-by: Yang Wang <kevinyang.wang@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
By anchoring the search to the DML top-level Makefile directory,
FPU CFLAGS are now correctly applied to all relevant source files.
Signed-off-by: Prike Liang <Prike.Liang@amd.com>
Reviewed-by: Leo Li <sunpeng.li@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Currently when amdgpu copies or fills a buffer,
it uses the maximum byte count supported by the
copy engine (SDMA). This is problematic when the
maximum byte count is not aligned to 256 bytes
because it then can't use all memory channels
optimally and can cause the SDMA to operate in its
slower byte mode (as opposed to the faster dword mode).
For example, when copying a 10 MiB buffer on SDMA v2.4,
we get 5 packets copying 2097151 bytes and 1 packet copying
the remaining 5 bytes. All 6 packets are misaligned and operate
in byte mode.
For this example, the optimal solution would be to have
5 packets each copying 2096896 bytes and 1 last packet to
copy the remaining 1280 bytes, in which case all 6 packets
are aligned to 256 bytes and operate in dword mode.
Let's use the following scheme from now on:
When byte count is dword-aligned and fits a single packet,
just emit a single packet.
Otherwise, align the copy packet size down to 256 bytes
for optimal use of memory channels and to ensure the HW
can use the dword mode.
This assumes that the starting addresses of BOs are always
dword aligned, which should be the case for every copy
operation in the kernel, because the kernel always copies
pages.
Signed-off-by: Timur Kristóf <timur.kristof@gmail.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
... and PCI device helpers.
The struct pci_device_id array of supported device was initialized
by list expressions. This isn't easily readable if you're not into PCI.
Use PCI_DEVICE* helper macros and named initializers which is more
explicit and thus easier to parse. Also skip explicit assignments of 0
(which the compiler then takes care of).
The secret plan is to make struct pci_device_id::driver_data an
anonymous union (similar to
https://lore.kernel.org/all/cover.1776579304.git.u.kleine-koenig@baylibre.com/)
and that requires named initializers. But it's also a nice cleanup on
its own.
This change doesn't introduce changes to the compiled pci_device_id
array. Tested on x86 and arm64.
Signed-off-by: Uwe Kleine-König (The Capable Hub) <u.kleine-koenig@baylibre.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
The debugger sets the CWSR grace period via AMDKFD_IOC_DBG_TRAP and may
request a value of 0. An earlier change clamped 0 to 1 unconditionally
because most firmware revisions locked up when given an infinite (0)
grace period.
Firmware has since been fixed on most ASICs, so gate the clamp on ASIC
type and MEC firmware version and allow 0 where the running firmware is
known to handle it. Navi3x and MI350 support 0 in every firmware
revision and need no version check. MI100 never received the firmware
fix and is kept clamped. Any unlisted or future ASIC defaults to the
safe (clamped) behaviour.
Signed-off-by: William Palacek <William.Palacek@amd.com>
Reviewed-by: Kent Russell <kent.russell@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
After S3, reloading amdgpu on MacBookPro15,1 systems with a Radeon Pro
555X or 560X fails while loading the SMU firmware. The existing SMC
register check does not request a reset because the registers do not
reliably reflect the stale SMU state on these machines.
Frederick Morlock found that forcing a VI ASIC reset allows the driver to
initialize again. This patch limits his workaround to the exact PCI
device, Apple subsystem device and revision combinations used by these
two GPUs, leaving other VI hardware unchanged.
Suggested-by: Frederick Morlock <me@freddy.us>
Signed-off-by: Andre Eikmeyer <dev@deq.rocks>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Commit a4f01bf729 ("drm/amd/display: Refactor and fix link_dpms I2C")
had also changed the "Set retimer failed" messages from DC_LOG_DEBUG()
to DC_LOG_ERROR(). This unfortunately can create log spam.
Change those back to DC_LOG_DEBUG() only.
Fixes: a4f01bf729 ("drm/amd/display: Refactor and fix link_dpms I2C")
Closes: https://gitlab.freedesktop.org/drm/amd/-/work_items/5520
Signed-off-by: Alan Swanson <reiver@improbability.net>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
All supported GPU generations now support DRM format modifiers.
Remove all code from amdgpu_dm that dealt with tiling flags.
Note that the legacy non-DC display code still relies on
tiling flags, so we can't remove them outside of DC
until we also remove the legacy display code.
Signed-off-by: Timur Kristóf <timur.kristof@gmail.com>
Tested-by: Link Mauve <linkmauve@linkmauve.fr>
Tested-by: Nikola Medić <nmedic89@gmail.com>
Tested-by: Dieter Nützel <Dieter@nuetzel-hh.de>
Reviewed-by: Marek Olšák <maraeo@gmail.com>
Reviewed-by: Daniel Stone <daniels@collabora.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Acked-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Expose displayable DRM format modifiers based on which
tiling modes are supported by the GFX block.
Technically, DCE (Display Controller Engine) could
support all possible permutations of all parameters
independently of what GFX supports, with the limitation
that it can only display LINEAR images and the
displayable micro tiling mode (MICROTILE == DISPLAY).
It doesn't make sense to expose all possible permutations
of macro tiling modes, so let's just expose what the GFX
block of the current chip supports.
The following modes will be advertised:
- 2D_TILED_THIN1 + DISPLAY + macrotile params [1]
- 1D_TILED_THIN1 + DISPLAY
- LINEAR
[1] The macro tiling parameters depend on how many
bits per pixel of the specific surface has and
how the chip is configured. There is only one set
of valid macrotile params for a given surface.
Signed-off-by: Timur Kristóf <timur.kristof@gmail.com>
Tested-by: Link Mauve <linkmauve@linkmauve.fr>
Tested-by: Nikola Medić <nmedic89@gmail.com>
Tested-by: Dieter Nützel <Dieter@nuetzel-hh.de>
Reviewed-by: Marek Olšák <maraeo@gmail.com>
Reviewed-by: Daniel Stone <daniels@collabora.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Acked-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
This is done for consistency between different GPU generations.
Tiling flags are "implicit modifiers" which are used on AMD GPUs
to let the kernel know the tiling information without modifiers.
Convert the tiling flags to modifers on GFX8 and older, so that
the DC display driver can rely on them like on newer generations.
Note that this code path will only be taken when DC actually
exposes any modifiers on GFX6-8, which is handled in a subsequent
commit after this one. This code path mainly exists for legacy
compositors which don't support explicit DRM format modifiers
and rely on the tiling flags, and for compatibility with old
Mesa that didn't support modifiers on GFX6-8.
Add amdgpu_display_verify_sizes_gfx6() to validate modifiers
on GFX6-8 and only accept the modes that the display driver
really supports, as well as validate the FB sizes.
Additionally adjust check_tiling_flags_gfx6() to allow the
LINEAR_GENERAL mode and disallow the ROTATED mode which has
never been supported by Mesa or Linux and was mistakenly
allowed previously.
Signed-off-by: Timur Kristóf <timur.kristof@gmail.com>
Tested-by: Link Mauve <linkmauve@linkmauve.fr>
Tested-by: Nikola Medić <nmedic89@gmail.com>
Tested-by: Dieter Nützel <Dieter@nuetzel-hh.de>
Reviewed-by: Marek Olšák <maraeo@gmail.com>
Reviewed-by: Daniel Stone <daniels@collabora.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Acked-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
GFX6-8 are the oldest GPUs supported by the amdgpu
kernel driver, and the last ones that didn't support
DRM format modifiers until now. These are the Southern
Islands, Sea Islands and Volcanic Islands families
of GPUs.
On GFX6-8, the GFX block can only use pre-determined tiling
modes which are programmed by the kernel according to the
tiling mode table. GFX6 uses the GB_TILE_MODE0...31 registers,
and GFX7-8 also has GB_MACROTILE_MODE0...15 registers.
DCC is also supported on GFX8, albeit not displayable.
Note that the tiling table is uAPI and userspace relies on
specific modes being present at specific indices.
How the tiling works is primarily determined by the
so-called array mode.
Use the TILE field to specify the array mode.
Pixel data is organized into micro tiles.
Each micro tile may be 8x8 / 8x8x4 / 8x8x8 pixels,
depending on the array mode.
Add the MICROTILE field to specify microtile mode.
Microtiles may be further organized into macro tiles,
which have many configurable parameters. Macro tile mode
selection depends on how many bits per pixel an image has.
Add the PIPE_CONFIG, TILE_SPLIT, BANK_WIDTH, BANK_HEIGHT,
MACRO_TILE_ASPECT, NUM_BANKS fields to specify parameters
of macro tiled modes.
Furthermore, tiling is also influenced by memory
configuration. Old RFC patches received feedback
concerning that, so I looked into it specifically:
GB_ADDR_CONFIG.ROW_SIZE needs to be considered when
calculating TILE_SPLIT, but does not need to be included
in the modifiers, and also PIPE_INTERLEAVE matters,
but it's hardcoded to the same value on all GFX6-8 GPUs
and changing it would break userspace, so let's assume
it isn't going to change. Therefore we don't need to
include that in modifiers. Mesa also reads NUM_RANKS
but actually doesn't use its value on GFX6-8.
As a side note, tiling works similarly on GFX4-5
(that is Evergreen and Northern Islands). But that
will need some additional PIPE_CONFIG enum values
as well as some extra fields not relevant to GFX6-8.
Initially, let's only expose the tiling modes that are
most relevant to sharing buffers between different
processes:
Exposed array modes (TILE field):
- 1D_TILED_THIN1: micro tiled only
- 2D_TILED_THIN1: macro tiled
Exposed micro tile modes (MICROTILE field):
- DISPLAY: supported by DCE (the display engine)
- THIN: more efficient but not displayable
Exposed macro tile modes:
All possible parameters (25088 permutations).
More modes may be exposed in the future as needed.
Technically, the amount of possible combinations
of all possible tiling parameters is in the range
of hundreds of thousands, but in practice, there are
just a handful of possible modifiers for a surface.
For example on GFX8, a surface would have these
modifiers, from best to worst performance:
- 2D_TILED_THIN1 + THIN + DCC + macrotile params [1]
- 2D_TILED_THIN1 + THIN + macrotile params [1]
- 2D_TILED_THIN1 + DISPLAY + macrotile params [1]
- 1D_TILED_THIN1 + THIN
- 1D_TILED_THIN1 + DISPLAY
- LINEAR
[1] The macro tiling parameters depend on how many
bits per pixel of the specific surface has and
how the chip is configured. There is only one set
of valid macrotile params for a given surface.
DCC is only supported by GFX8 and newer, and only
with non-displayable macrotiling modes.
When sharing buffers between different GFX6-8 GPUs,
it is very unlikely that they support the exact same
macrotiling configuration, so they will likely need
to use micro tiled modes, which are still much better
than using linear buffers. (Note that currently Mesa
always uses LINEAR when copying between two GPUs.)
Suggested-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Signed-off-by: Timur Kristóf <timur.kristof@gmail.com>
Reviewed-by: Marek Olšák <maraeo@gmail.com>
Reviewed-by: Daniel Stone <daniels@collabora.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Acked-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
There's no need to crash the kernel for these cases.
Reviewed-by: Kent Russell <kent.russell@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
There's no need to crash the kernel for these cases.
Reviewed-by: Kent Russell <kent.russell@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
There's no need to crash the kernel for these cases.
Reviewed-by: Kent Russell <kent.russell@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
There's no need to crash the kernel for these cases.
Reviewed-by: Kent Russell <kent.russell@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
There's no need to crash the kernel for these cases.
Reviewed-by: Kent Russell <kent.russell@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
There's no need to crash the kernel for these cases.
Reviewed-by: Kent Russell <kent.russell@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
There's no need to crash the kernel for these cases.
Reviewed-by: Kent Russell <kent.russell@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
There's no need to crash the kernel for these cases.
Reviewed-by: Kent Russell <kent.russell@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
There's no need to crash the kernel for these cases.
Reviewed-by: Kent Russell <kent.russell@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
There's no need to crash the kernel for this case.
Reviewed-by: Kent Russell <kent.russell@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
There's no need to crash the kernel for this case.
Reviewed-by: Kent Russell <kent.russell@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
There's no need to crash the kernel for this case.
Reviewed-by: Kent Russell <kent.russell@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
There's no need to crash the kernel for this case.
Reviewed-by: Kent Russell <kent.russell@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
There's no need to crash the kernel for these cases.
Reviewed-by: Kent Russell <kent.russell@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
There's no need to crash the kernel for these cases.
Reviewed-by: Kent Russell <kent.russell@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
There's no need to crash the kernel for these cases.
Reviewed-by: Kent Russell <kent.russell@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
There's no need to crash the kernel for this case.
Reviewed-by: Kent Russell <kent.russell@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
kfd_criu_checkpoint_events() counts the entries in p->event_idr via
kfd_get_num_events(), allocates an array sized to that count, and then
walks the same IDR to fill it. Neither the count nor the walk holds
p->event_mutex.
The CRIU checkpoint caller holds only p->mutex. Event create and destroy
(kfd_event_create()/kfd_event_destroy()) take p->event_mutex and do not
take p->mutex, so a second thread in the same process can insert or remove
events between the count and the walk. If an event is inserted, the walk
iterates more entries than were counted and writes past the end of the
ev_privs allocation; if an event is removed, the walk dereferences an
entry that is being freed.
Hold p->event_mutex across the count and the walk so both observe a
consistent view of p->event_idr. The lock is released before
copy_to_user(), which only touches the local buffer. The caller already
holds p->mutex and the create/destroy paths never take p->mutex, so the
p->mutex -> p->event_mutex order is not inverted and no deadlock is
introduced.
Fixes: 40e8a766a7 ("drm/amdkfd: CRIU checkpoint and restore events")
Signed-off-by: William Palacek <William.Palacek@amd.com>
Reviewed-by: Alysa Liu <Alysa.Liu@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Original amdgpu_cper_fini skips all CPER cleanup work for any SR-IOV VF directly.
When AMDGIM_FEATURE_RAS_CPER is enabled on VF side, CPER related buffers/workitems
are allocated during initialization, but the old code returns early without
releasing these resources, which leads to kernel memory leak.
Signed-off-by: Ce Sun <cesun102@amd.com>
Reviewed-by: Tao Zhou <tao.zhou1@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
User-space read/write syscalls interpret positive driver return values
as successful transfer sizes. Our current RAS wrappers return positive
error codes, so we must convert them to standard negative error codes.
Signed-off-by: Ce Sun <cesun102@amd.com>
Reviewed-by: Tao Zhou <tao.zhou1@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
