In encode_message() the per-transaction lower-bound check compares
trans_hdr->len against sizeof(trans_hdr), i.e. the size of the pointer,
instead of sizeof(*trans_hdr), the size of struct qaic_manage_trans_hdr.
Every other length check in this file (encode_message() at the loop
guard, decode_message(), etc.) correctly uses sizeof(*trans_hdr), so
this is an inconsistency. On 64-bit builds the pointer and the struct
are both 8 bytes, so the check is correct by coincidence and there is
no behavioural change. On 32-bit builds the pointer is 4 bytes, which
weakens the minimum-length check below the 8-byte header size.
Use sizeof(*trans_hdr) so the check validates against the actual
transaction header size on all builds.
Fixes: ea33cb6fc2 ("accel/qaic: tighten bounds checking in encode_message()")
Signed-off-by: Muhammad Bilal <meatuni001@gmail.com>
Reviewed-by: Jeff Hugo <jeff.hugo@oss.qualcomm.com>
Signed-off-by: Jeff Hugo <jeff.hugo@oss.qualcomm.com>
Link: https://patch.msgid.link/20260617212520.59801-1-meatuni001@gmail.com
iface_fw_to_cpu_addr() only checks that the firmware-provided MCU virtual
address points inside the shared section. The returned pointer is later
used as a full firmware interface structure, so accepting an address near
the end of the shared section can still lead to out-of-bounds accesses.
Pass the expected object size to iface_fw_to_cpu_addr() and reject ranges
that do not fit entirely in the shared section.
Fixes: 2718d91816 ("drm/panthor: Add the FW logical block")
Cc: stable@vger.kernel.org
Signed-off-by: Osama Abdelkader <osama.abdelkader@gmail.com>
Reviewed-by: Steven Price <steven.price@arm.com>
Signed-off-by: Steven Price <steven.price@arm.com>
Link: https://patch.msgid.link/20260720114918.15973-1-osama.abdelkader@gmail.com
The OVL adaptor registers both an aggregate driver for its child devices
and a component for the main DRM aggregate. Probe currently ignores an
error from registering the child aggregate and leaves that aggregate
registered if registering the DRM component fails. The remove callback
also leaves the DRM component registered.
These imbalances can leave component framework entries referring to a
device whose probe failed or whose driver has been detached. The aggregate
unbind callback also fails to undo component_bind_all(), leaving its child
components marked as bound when the aggregate is removed.
Check the aggregate registration result, unwind it when the component
registration fails, and unregister the component before the aggregate on
remove. Keep runtime PM enabled until both framework registrations have
been removed, and unbind all child components from the aggregate unbind
callback.
Fixes: 453c336463 ("drm/mediatek: Add ovl_adaptor support for MT8195")
Cc: stable@vger.kernel.org # 6.4+
Co-developed-by: Ijae Kim <ae878000@gmail.com>
Signed-off-by: Ijae Kim <ae878000@gmail.com>
Signed-off-by: Myeonghun Pak <mhun512@gmail.com>
Reviewed-by: CK Hu <ck.hu@mediatek.com>
Link: https://patchwork.kernel.org/project/linux-mediatek/patch/20260721152242.47138-1-mhun512@gmail.com/
Signed-off-by: Chun-Kuang Hu <chunkuang.hu@kernel.org>
Commit 76255024ca ("drm/mediatek: mtk_dsi: enable hs clock
during pre-enable") rightfully moves the HS clock enablement to
before atomic_enable(), but it's moving it to mtk_dsi_poweron(),
which is not only called in the .atomic_pre_enable() callback
for the DRM bridge, but also in the MediaTek DRM's .ddp_start()
callback, which happens way before the bridge ones.
The HS clock enablement should be done at just the right time,
otherwise some bridge chips (or some Display Driver ICs) may
not work correctly: this is seen at least with a Parade DSI to
eDP bridge (PS8640) on the MT8173 Elm Chromebook.
This resolves a regression that was seen on the aforementioned
machine, which was happening only after a suspend-resume cycle.
Cc: <stable@vger.kernel.org> # 7.1.x
Fixes: 76255024ca ("drm/mediatek: mtk_dsi: enable hs clock during pre-enable")
Signed-off-by: AngeloGioacchino Del Regno <angelogioacchino.delregno@collabora.com>
Tested-by: Adam Thiede <me@adamthiede.com>
Reviewed-by: CK Hu <ck.hu@mediatek.com>
Link: https://patchwork.kernel.org/project/linux-mediatek/patch/20260728153145.578404-1-angelogioacchino.delregno@collabora.com/
Signed-off-by: Chun-Kuang Hu <chunkuang.hu@kernel.org>
The PCON max FRL bandwidth field lives in byte 2 of the DFP Detailed
Capability Info (DPCD 0x82 for the first DFP).
The DP standard defines the meaning of descriptor bytes 1-3 strictly
per DFP type, and for a DisplayPort type DFP all of them are
reserved, with "read all 0s" semantics (DP v2.0, section 2.12.3,
Table 2-183).
The FRL bandwidth field is an HDMI DFP extension added by the VESA
DP-to-HDMI PCON specification.
drm_dp_get_pcon_max_frl_bw() however parses the byte without checking
the DFP type, the branch presence or DETAILED_CAP_INFO_AVAILABLE.
Without the latter the port descriptors are one byte wide and
port_cap[2] is not even the right register.
All neighbouring helpers parsing the same descriptor are scoped by
the DFP type already, see for instance drm_dp_downstream_max_bpc()
reading the same byte and returning 0 for a DP type DFP.
amdgpu's DC parses the field only for HDMI(/DP++) detailed types as
well.
This is not theoretical.
A Synaptics VMM7100 based USB-C to HDMI adapter with a macOS targeted
firmware advertises a DisplayPort type DFP with the type byte
replicated across the whole descriptor (08 08 08 08).
i915 decodes that as "PCON limited to 18 Gbps FRL" and prunes every
mode above ~750 MHz dotclock, including all the 4k@100/120 modes the
sink EDID offers, while macOS drives 4k@120 through the same adapter
just fine via DP DSC (and amdgpu's type-scoped parser would ignore
the bogus field as well).
Only parse the field for an HDMI DFP behind a DPCD 1.1+ branch
device that reports detailed cap info, matching the type-scoped
field layout of the spec and the rest of the helpers.
Fixes: ce32a6239d ("drm/dp_helper: Add Helpers for FRL Link Training support for DP-HDMI2.1 PCON")
Cc: Ankit Nautiyal <ankit.k.nautiyal@intel.com>
Cc: Uma Shankar <uma.shankar@intel.com> (v2)
Cc: Jani Nikula <jani.nikula@intel.com>
Cc: Maarten Lankhorst <maarten.lankhorst@linux.intel.com>
Cc: dri-devel@lists.freedesktop.org
Cc: <stable@vger.kernel.org> # v5.12+
Signed-off-by: Alexander Kaplan <alexander.kaplan@sms-medipool.de>
Reviewed-by: Ankit Nautiyal <ankit.k.nautiyal@intel.com>
Signed-off-by: Ankit Nautiyal <ankit.k.nautiyal@intel.com>
Link: https://patch.msgid.link/20260610193825.2933-1-alexander.kaplan@sms-medipool.de
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>
(cherry picked from commit 6f16fcbb0c)
Cc: stable@vger.kernel.org
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>
(cherry picked from commit 5dafdd6492)
Cc: stable@vger.kernel.org
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>
(cherry picked from commit 757ba0790b)
Cc: stable@vger.kernel.org
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>
(cherry picked from commit 138531c885)
Cc: stable@vger.kernel.org
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>
(cherry picked from commit ff57e223ab)
Cc: stable@vger.kernel.org
If a plane reaches calculate_mcache_setting with a zero-area viewport,
calculate_mcache_setting exits early with num_mcaches == 0 and
mvmpg_width/height == 0. This will cause a divide-by-zero panic and can
also cause an underflow on num_mcaches.
Fix this by changing calculate_mcache_setting to bool and adding guards
after each calculate_mcache_row_bytes call. If num_mcaches or
mvmpg_width/height is zero, return a false. Callers will propagate the
failure as a rejected mode, which prevents the panic.
Closes: https://gitlab.freedesktop.org/drm/amd/-/issues/5302
Reviewed-by: Sun peng (Leo) Li <sunpeng.li@amd.com>
Reviewed-by: Dillon Varone <dillon.varone@amd.com>
Signed-off-by: George Zhang <george.zhang@amd.com>
Signed-off-by: Fangzhi Zuo <jerry.zuo@amd.com>
Tested-by: Dan Wheeler <daniel.wheeler@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
(cherry picked from commit 29c0f7c655)
Cc: stable@vger.kernel.org
amdgpu_dpm_get_gpu_metrics() returns a pointer to the shared metrics cache
after dropping adev->pm.mutex. The sysfs path then copies from that pointer.
Another reader can refresh the cache in place during the copy and return a
snapshot containing data from two generations.
Pass caller-provided storage through the DPM interface and copy the metrics
while the mutex is held. This keeps the cache pointer private and makes each
sysfs read observe one complete sample.
Fixes: 25c933b1c4 ("drm/amd/powerplay: add new sysfs interface for retrieving gpu metrics(V2)")
Signed-off-by: Yang Wang <kevinyang.wang@amd.com>
Reviewed-by: Kenneth Feng <kenneth.feng@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
(cherry picked from commit 862333bb48)
Cc: stable@vger.kernel.org
amdgpu_dpm_get_pp_table() returns a pointer to a driver-owned power table
after dropping adev->pm.mutex. The sysfs path then copies from that pointer.
A concurrent pp_table write can replace and free the allocation during the
copy, causing a use-after-free.
Change the DPM interface to copy into caller-provided storage while the mutex
is held. Keep the size-only query for attribute discovery without exposing
the driver-owned pointer.
Fixes: 1684d3ba48 ("drm/amd/amdgpu: change pptable output format from ASCII to binary")
Signed-off-by: Yang Wang <kevinyang.wang@amd.com>
Reviewed-by: Kenneth Feng <kenneth.feng@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
(cherry picked from commit f6eed7acfd)
Cc: stable@vger.kernel.org
eop_ring_buffer_size in struct queue_properties is a u32. In
kfd_queue_acquire_buffers() the expected EOP buffer size is computed as
ALIGN(eop_ring_buffer_size, PAGE_SIZE); ALIGN uses typeof(x), so the
addition is done in 32-bit. A user-supplied size of 0xFFFFF001 wraps to
0, causing kfd_queue_buffer_get() to skip its exact-size check (gated on
size != 0) and accept any BO mapped at the address. On GFX8/GFX9 the MQD
cp_hqd_eop_control is then programmed for an 8KB EOP ring backed by a 4KB
BO, so CP EOP writes can land past the buffer and fault the GPU.
Cast the operand to u64 so the alignment is computed in 64-bit; the size
check in kfd_queue_buffer_get() then rejects the oversized request.
Fixes: 42ea9cf2f1 ("drm/amdkfd: Relax size checking during queue buffer get")
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>
(cherry picked from commit ae443117b7)
Cc: stable@vger.kernel.org
Prevent unauthorized termination of active GPU debug sessions.
Previously, users with /dev/kfd access could terminate another process's
debug session without proper ownership or ptrace authorization.
Signed-off-by: Gang Ba <Gang.Ba@amd.com>
Reviewed-by: Kent Russell <kent.russell@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
(cherry picked from commit 4db4c5ffd5)
Cc: stable@vger.kernel.org
The commit referenced below restarts the CS if the validation is
still in progress. When debug_vm is enabled, all BOs from the CS
are invalidated so we will hit an infinite loop.
To avoid that, defer BO invalidation to amdgpu_cs_parser_fini.
Fixes: 59720bfd8c ("drm/amdgpu: restart the CS if some parts of the VM are still invalidated")
Signed-off-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Reviewed-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
(cherry picked from commit 8c990ee9da)
Cc: stable@vger.kernel.org
In kfd_criu_restore_event, there was no handling for
the event priv data having an invalid event type. The priv
data here is untrusted and can be invalid.
In that case, fail with EINVAL.
Signed-off-by: David Francis <David.Francis@amd.com>
Reviewed-by: Kent Russell <kent.russell@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
(cherry picked from commit 2e8e9963cd)
Cc: stable@vger.kernel.org
Updates PPSMC_MSGs and set/get functions for gathering GFXOFF logs
on Van Gogh. Logs are now gathered live rather than starting then
stopping logging and reading an average value afterwards. This is
in accordance to changes made in PMFW.
In regards to messageID 0x52, the old interface uses a start/stop
parameter, and the new one doesn't. The firmware is checked to
determine which method to use.
v2: added firmware guard to new interface, old interface kept as
fallback
Signed-off-by: Fares Soliman <Fares.Soliman@amd.com>
Acked-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
(cherry picked from commit 482e2cadea)
Building for ARCH=um with W=1 C=1 makes the "amd_sriov_msg_vf2pf_info
must be 1 KB" static assertion in amdgv_sriovmsg.h fail under sparse,
exposed after UML builds were enabled.
Sparse does not honor #pragma pack(push, 1) for the nested ucode_info
struct, so it sizes each element as 8 bytes instead of 5 and computes
the surrounding structure as larger than 1 KB. The compilers get this
right via the enclosing pragma, but the annotation should be explicit.
Fixes: af3f2f5db2 ("drm/amdgpu: Remove UML build exclusion from Kconfig")
Reported-by: kernel test robot <lkp@intel.com>
Closes: https://lore.kernel.org/oe-kbuild-all/202607091659.SHEscT0c-lkp@intel.com/
Cc: Harry Wentland <harry.wentland@amd.com>
Assisted-by: Copilot:Claude-Opus-4.8
Signed-off-by: Alex Hung <alex.hung@amd.com>
Reviewed-by: Harry Wentland <harry.wentland@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
(cherry picked from commit 1d8cfeb69d)
Building for ARCH=um with W=1 C=1 makes sparse report "incompatible
types in comparison expression (different address spaces)" warnings in
the KFD code, exposed after UML builds were enabled:
- amdgpu_amdkfd_fence.c compares the __rcu-annotated dma_fence.ops
pointer directly in to_amdgpu_amdkfd_fence().
- amdgpu_amdkfd_gpuvm.c compares the __rcu eviction fence pointer
directly in amdgpu_amdkfd_gpuvm_restore_process_bos().
Fixes: af3f2f5db2 ("drm/amdgpu: Remove UML build exclusion from Kconfig")
Reported-by: kernel test robot <lkp@intel.com>
Closes: https://lore.kernel.org/oe-kbuild-all/202607091659.SHEscT0c-lkp@intel.com/
Cc: Harry Wentland <harry.wentland@amd.com>
Assisted-by: Copilot:Claude-Opus-4.8
Signed-off-by: Alex Hung <alex.hung@amd.com>
Reviewed-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
(cherry picked from commit 764f241ad2)
When MES is enabled and amdgpu_amdkfd_alloc_kernel_mem() fails during
the first queue creation for a process, pqm_create_queue() returns
early via 'return retval' without going through the err_create_queue
cleanup label.
This means clear_bit(*qid, pqm->queue_slot_bitmap) is never called,
leaving the reserved QID bit permanently set in queue_slot_bitmap.
Over time this leaks QID slots, potentially exhausting all available
queue slots.
Fix this by replacing 'return retval' with 'goto err_allocate_pqn'
so that clear_bit() is always called on the error path without
touching the uninitialized pqn pointer.
AILIKFD-813
Reported-by: Deucher, Alexander <alexander.deucher@amd.com>
Signed-off-by: Vladimir Marioukhine <Vladimir.Marioukhine@amd.com>
Reviewed-by: Kent Russell <kent.russell@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
(cherry picked from commit a107f74c38)
Cc: stable@vger.kernel.org
Some HDMI sinks need additional GCP packets to properly process the
mute state before the timing generator is disabled, especially after
link re-establishment with HDMI 2.0 scrambling enabled. Waiting for
only 2 frames is insufficient for certain monitor firmware, resulting
in garbled display output on resume from suspend.
Increase the AV mute wait in dcn30_set_avmute() from 2 to 3 frames
to ensure the sink receives enough GCP packets.
Closes: https://gitlab.freedesktop.org/drm/amd/-/work_items/5167
Assisted-by: Cursor:Claude-Opus-4.6
Reviewed-by: Wayne Lin <wayne.lin@amd.com>
Signed-off-by: Ray Wu <ray.wu@amd.com>
Tested-by: Dan Wheeler <daniel.wheeler@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
(cherry picked from commit 0c0d5174b0)
Cc: stable@vger.kernel.org
amdgpu_vm_clear_freed() allocates an amdgpu_sync object and walks the VM
reservation fences via amdgpu_sync_resv() before checking whether vm->freed
has anything to clear. Return early when the list is empty to skip this
overhead on a hot path (every GEM close and command submission).
Signed-off-by: Bob Zhou <bobzhou2@amd.com>
Reviewed-by: Felix Kuehling <felix.kuehling@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
(cherry picked from commit 8ba869e852)
The CRAT parser validates that the subtype header fits within the image,
but does not verify that the advertised subtype length fits. A malformed
CRAT table with an oversized length field causes out-of-bounds reads when
kfd_parse_subtype() casts the header to specific subtype structures.
Add validation that sub_type_hdr + length does not exceed the image
boundary before parsing the subtype contents.
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>
(cherry picked from commit 48e1d1e6e8)
Cc: stable@vger.kernel.org
The binner BO is a single 16MB buffer split into 512KB slots that are
handed out to jobs at submission time and recycled as jobs complete,
without ever being cleared. Each slot holds the job's Tile State Data
Array (TSDA) at its start, followed by the tile allocation pool.
While the tile allocation pool is only walked by the render thread
through branches the binner generated during the current job, the
TSDA is the PTB's own per-tile bookkeeping and is consumed by the
hardware itself. Although the kernel sets the "Auto-initialise Tile
State Data Array" flag in the tile binning mode configuration, the
PTB demonstrably still acts on stale tile state left by the slot's
previous user: the binner ends up creating invalid command streams
with invalid primitive streams and branches, which can cause GPU hangs
as observed in [1][2].
Zero the TSDA when the job's binning slot is configured. This clears
48 bytes per tile (~24KB for a 1080p frame) in the submission path, and
guarantees the PTB never sees another job's tile state.
The tile count is only checked for being non-zero today, so the 8-bit
fields it comes from can describe a tile state array almost six times
larger than the slot it has to live in. Bound it before the slot is
handed out, since such size decides how much of the slot is left for
the tile alloc pool.
Link: https://github.com/raspberrypi/linux/issues/3221 [1]
Link: https://github.com/raspberrypi/linux/issues/5780 [2]
Fixes: 553c942f8b ("drm/vc4: Allow using more than 256MB of CMA memory.")
Cc: stable@vger.kernel.org
Reviewed-by: Iago Toral Quiroga <itoral@igalia.com>
Link: https://patch.msgid.link/20260727-vc4-bin-oom-fixes-v2-2-0d8a5eddc7c9@igalia.com
Signed-off-by: Maíra Canal <mcanal@igalia.com>
vc4_overflow_mem_work() points BPOA at a 512KB slot inside the 16MB
binner BO, but writes the size of the whole BO to BPOS. On every binner
out-of-memory event the PTB is therefore authorized to write tile lists
across all the other slots (which may hold the tile state, tile alloc and
overflow memory of in-flight jobs) and, for any slot but the first, past
the end of the binner BO into unrelated CMA memory.
Since CMA pages are recycled into page cache and user allocations, this
is arbitrary memory corruption by GPU DMA. In practice it shows up as GPU
hangs with corrupted control list pointers, userspace heap corruption, a
GPU that stays permanently wedged after the first hang, and occasional
full system crashes, whenever a job overflows the initial binner slot.
The bug dates back to the conversion from a dedicated overflow BO (where
writing the full BO size was correct) to the slotted binner BO.
Fixes: 553c942f8b ("drm/vc4: Allow using more than 256MB of CMA memory.")
Cc: stable@vger.kernel.org
Assisted-by: Claude:claude-opus-4.8
Signed-off-by: Jose Maria Casanova Crespo <jmcasanova@igalia.com>
Reviewed-by: Maíra Canal <mcanal@igalia.com>
Reviewed-by: Iago Toral Quiroga <itoral@igalia.com>
Link: https://patch.msgid.link/20260727-vc4-bin-oom-fixes-v2-1-0d8a5eddc7c9@igalia.com
Signed-off-by: Maíra Canal <mcanal@igalia.com>
vmw_external_bo_copy() trusts caller-supplied offsets, strides, and
heights and operates on imported dma-buf vmaps:
- The equal-stride memcpy() bound was clamped after subtracting the
offsets from dst_size and src_size; an offset larger than the BO
size wraps the unsigned subtraction to a huge value and the
resulting memcpy() runs off the end of the vmap. dst_stride *
height is also a u32 multiplication that can overflow.
- The non-equal-stride row-by-row path had no bound at all. The
loop touches bytes through offset + (height - 1) * stride +
width_in_bytes, with only a WARN_ON(dst_stride < width_in_bytes),
and could likewise step past the end of either mapping.
The offsets and strides are derived from STDU/SOU plane state, so a
configured CRTC submitting a crafted atomic commit on an imported
framebuffer can reach this path.
Validate the exact row-copy endpoint against each BO's size up front
using check_mul_overflow() and check_add_overflow(). Use the bulk
memcpy() path only when width_in_bytes covers the whole stride;
otherwise copy one row at a time so partial-row updates near the bottom
of a framebuffer remain valid. Also reject zero strides and stride <
width_in_bytes, both of which the row-by-row path cannot represent
safely.
Fixes: 50f1199250 ("drm/vmwgfx: Fix prime with external buffers")
Cc: stable@vger.kernel.org
Assisted-by: Claude:claude-opus-4.7
Signed-off-by: Zack Rusin <zack.rusin@broadcom.com>
Reviewed-by: Ian Forbes <ian.forbes@broadcom.com>
Link: https://patch.msgid.link/20260505222728.519626-13-zack.rusin@broadcom.com
vmw_shader_define() validates the user-supplied shader window against
its backing buffer with
(u64)buffer->tbo.base.size < (u64)size + (u64)offset
drm_vmw_shader_create_arg::offset is __u64 in the uapi; when it is
near U64_MAX the unsigned addition wraps and the resulting tiny value
passes the check. The unbounded offset is then stored in
res->guest_memory_offset and forwarded to host SVGA shader-create
commands.
Use check_add_overflow() to detect the wrap and compare the resulting
endpoint against the buffer size.
Fixes: 668b206601 ("drm/vmwgfx: Stop using raw ttm_buffer_object's")
Cc: stable@vger.kernel.org
Assisted-by: Claude:claude-opus-4.7
Signed-off-by: Zack Rusin <zack.rusin@broadcom.com>
Reviewed-by: Ian Forbes <ian.forbes@broadcom.com>
Link: https://patch.msgid.link/20260505222728.519626-12-zack.rusin@broadcom.com
vmw_validation_add_resource() calls hash_add_rcu() only when
ctx->sw_context is non-NULL, but the doomed-resource error path calls
hash_del_rcu() unconditionally. The validation contexts declared with
DECLARE_VAL_CONTEXT(_, NULL, 0) in vmwgfx_kms.c, vmwgfx_scrn.c,
vmwgfx_stdu.c and vmwgfx_execbuf.c consequently reach a delete for a
node that was never added to any hash chain.
That is harmless today, but only incidentally so. hash_del_rcu() is
hlist_del_init_rcu(), which is guarded by hlist_unhashed(), and
vmw_validation_mem_alloc() hands out memory from __GFP_ZERO pages that
are never recycled within a context's lifetime, so
node->hash.head.pprev is always NULL and the delete does nothing.
Neither property is apparent at the call site, and the asymmetry with
the add side invites a real bug the first time either one changes.
Mirror the condition from the add side so the node is only unlinked
when it was actually linked. No functional change.
Assisted-by: Claude:claude-opus-4.7
Signed-off-by: Zack Rusin <zack.rusin@broadcom.com>
Reviewed-by: Ian Forbes <ian.forbes@broadcom.com>
Link: https://patch.msgid.link/20260505222728.519626-11-zack.rusin@broadcom.com
vmw_cursor_plane_atomic_check() bounds cursor width and height only
on the legacy update path; the SVGA_CAP2_CURSOR_MOB path -- the
default on modern hosts -- accepts any size. When the requested size
exceeds SVGA_REG_CURSOR_MAX_DIMENSION or SVGA_REG_MOB_MAX_SIZE,
vmw_cursor_mob_get() returns -EINVAL and leaves vps->cursor.mob NULL.
Its return value is then discarded in vmw_cursor_plane_prepare_fb(),
so the subsequent vmw_cursor_update_mob() calls
vmw_bo_map_and_cache(NULL) and oopses inside
vmw_bo_map_and_cache_size() on the tbo.base.size load.
Reachable from any DRM master via DRM_IOCTL_MODE_CURSOR2 with a
sufficiently large width or height (e.g. cursor_max_dim + 1).
Reject oversized cursors in atomic_check for both MOB-backed cursor
update types. The MOB byte-size limit only applies to the
SVGA_CAP2_CURSOR_MOB path (vmw_cursor_mob_size() returns 0 for
GB_ONLY); compute the required MOB size in 64-bit to avoid overflow
when very large dimensions are requested.
In prepare_fb only call vmw_cursor_mob_get()/_map() for
VMW_CURSOR_UPDATE_MOB -- the GB_ONLY path uses bo->map.virtual
directly and would otherwise be silently downgraded to NONE on hosts
without SVGA_CAP2_CURSOR_MOB (where vmw_cursor_mob_get() always
returns -EINVAL). Degrade the update to NONE if vmw_cursor_mob_get()
or vmw_cursor_mob_map() fails so the update path does not run with a
NULL backing MOB.
Fixes: 965544150d ("drm/vmwgfx: Refactor cursor handling")
Cc: stable@vger.kernel.org
Assisted-by: Claude:claude-opus-4.7
Signed-off-by: Zack Rusin <zack.rusin@broadcom.com>
Reviewed-by: Ian Forbes <ian.forbes@broadcom.com>
Link: https://patch.msgid.link/20260505222728.519626-10-zack.rusin@broadcom.com
Two paths through vmw_vkms_init() can leave vmw->crc_workq NULL while
still leaving the rest of the driver in a state that calls
vmw_vkms_cleanup() at module unload:
1. vmw_host_get_guestinfo(GUESTINFO_VBLANK, ...) failing or
returning an oversized buffer -- the common case on hosts
without a VBLANK guestinfo entry -- early-returned before the
workqueue allocation.
2. alloc_ordered_workqueue() returning NULL on memory pressure.
vmw_vkms_cleanup() then calls destroy_workqueue(NULL), which
dereferences wq->name and panics.
Fix the first case by removing the early return: vmw->vkms_enabled
is already false on the rpci-failure path so no work will ever be
queued, and allocating the workqueue unconditionally keeps the
control flow simple. Fix the second case by guarding the cleanup
with a NULL check, since alloc_ordered_workqueue() can still fail
under low memory.
Fixes: 7b0062036c ("drm/vmwgfx: Implement virtual crc generation")
Cc: stable@vger.kernel.org
Assisted-by: Claude:claude-opus-4.7
Signed-off-by: Zack Rusin <zack.rusin@broadcom.com>
Reviewed-by: Ian Forbes <ian.forbes@broadcom.com>
Link: https://patch.msgid.link/20260505222728.519626-9-zack.rusin@broadcom.com
vmw_cmd_dma() locates the DMA suffix at
(unsigned long) &cmd->body + header->size - sizeof(*suffix)
without checking that header->size is large enough to contain both
cmd->body and the suffix. An undersized header makes the suffix
pointer underflow back into the previous command in the bounce
buffer. The verifier later writes suffix->maximumOffset, clobbering
verified fields of an already-relocated earlier command -- a TOCTOU
on the device-visible command stream that lets one command rewrite
another's GMR id, surface id, or other authenticated fields.
Reject the command if the body is too small for the suffix to fit.
Fixes: 4e4ddd4777 ("drm/vmwgfx: Fix queries if no dma buffer thrashing is occuring.")
Cc: stable@vger.kernel.org
Assisted-by: Claude:claude-opus-4.7
Signed-off-by: Zack Rusin <zack.rusin@broadcom.com>
Reviewed-by: Ian Forbes <ian.forbes@broadcom.com>
Link: https://patch.msgid.link/20260505222728.519626-8-zack.rusin@broadcom.com
vmw_cmd_draw() computes
maxnum = (header->size - sizeof(cmd->body)) / sizeof(*decl);
where header->size is u32 and is taken straight from the user-supplied
command stream. When header->size is less than sizeof(cmd->body) the
unsigned subtraction wraps to nearly 4 GiB, producing a huge maxnum.
Any user-controlled cmd->body.numVertexDecls then passes the bound and
the loop dereferences decl[i] far past the end of the kernel command
bounce buffer, producing an out-of-bounds read of kernel memory.
Reject undersized headers up front.
Fixes: 7a73ba7469 ("drm/vmwgfx: Use TTM handles instead of SIDs as user-space surface handles.")
Cc: stable@vger.kernel.org
Assisted-by: Claude:claude-opus-4.7
Signed-off-by: Zack Rusin <zack.rusin@broadcom.com>
Reviewed-by: Ian Forbes <ian.forbes@broadcom.com>
Link: https://patch.msgid.link/20260505222728.519626-7-zack.rusin@broadcom.com
ttm_prime_fd_to_handle() returns -ENOSYS when the imported fd's
dma_buf->ops do not match the ttm_object_device's ops, but does so
without releasing the reference acquired by dma_buf_get(). Any
unprivileged renderD client passing a non-vmwgfx prime fd through the
DRM_VMW_GB_SURFACE_REF{,_EXT} path leaks one dma_buf reference per
call and indefinitely pins the foreign exporter's GEM resources.
Funnel the error path through the existing dma_buf_put() so the
reference is always dropped.
Fixes: 65981f7681 ("drm/ttm: Add a minimal prime implementation for ttm base objects")
Cc: stable@vger.kernel.org
Assisted-by: Claude:claude-opus-4.7
Signed-off-by: Zack Rusin <zack.rusin@broadcom.com>
Reviewed-by: Ian Forbes <ian.forbes@broadcom.com>
Link: https://patch.msgid.link/20260505222728.519626-6-zack.rusin@broadcom.com
vmw_fence_fifo_down() drops fman->lock to wait on a fence and, on
timeout, mutates fman->fence_list via list_del_init() and signals
the fence without re-acquiring the lock. __vmw_fences_update() walks
and removes entries from the same list under fman->lock from any
other waiter, the fence-IRQ thread, or vmw_fences_update(), so the
unlocked list_del_init() can corrupt the list head.
Re-take fman->lock before manipulating fence->head and use
dma_fence_signal_locked(). Wrap the locked signalling in
dma_fence_begin_signalling() / dma_fence_end_signalling() so the
lockdep annotation that dma_fence_signal() previously provided is
preserved (the same pattern as __vmw_fences_update()).
dma_fence_put() is moved outside the lock to avoid a recursive
acquire from vmw_fence_obj_destroy(), which also takes fman->lock.
Fixes: ae2a104058 ("vmwgfx: Implement fence objects")
Cc: stable@vger.kernel.org
Assisted-by: Claude:claude-opus-4.7
Signed-off-by: Zack Rusin <zack.rusin@broadcom.com>
Reviewed-by: Ian Forbes <ian.forbes@broadcom.com>
Link: https://patch.msgid.link/20260505222728.519626-5-zack.rusin@broadcom.com
vmw_bo_dirty_transfer_to_res() and vmw_bo_dirty_clear() compute the
intersection of a resource's page range with the BO's tracked dirty
range, but clamp res_end against dirty->end with max() instead of
min(). When dirty->end exceeds the resource end, the loop walks past
the resource's pages, calls vmw_resource_dirty_update() for ranges
owned by other resources sharing the same backing MOB and clears
their pending dirty bits via bitmap_clear(). The result is silent
loss of writeback for unrelated resources whenever two resources
share a MOB.
Use min() in both functions so the loop is bounded to the
intersection of the resource and dirty ranges.
Fixes: b7468b15d2 ("drm/vmwgfx: Implement an infrastructure for write-coherent resources")
Fixes: 965544150d ("drm/vmwgfx: Refactor cursor handling")
Cc: stable@vger.kernel.org
Assisted-by: Claude:claude-opus-4.7
Signed-off-by: Zack Rusin <zack.rusin@broadcom.com>
Reviewed-by: Ian Forbes <ian.forbes@broadcom.com>
Link: https://patch.msgid.link/20260505222728.519626-4-zack.rusin@broadcom.com
