linux.git/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c, branch v6.6.132

drm/amd: Set num IP blocks to 0 if discovery fails

2026-03-25T10:05:53+00:00

commit 3646ff28780b4c52c5b5081443199e7a430110e5 upstream.

If discovery has failed for any reason (such as no support for a block)
then there is no need to unwind all the IP blocks in fini. In this
condition there can actually be failures during the unwind too.

Reset num_ip_blocks to zero during failure path and skip the unnecessary
cleanup path.

Suggested-by: Lijo Lazar 
Reviewed-by: Lijo Lazar 
Signed-off-by: Mario Limonciello 
Signed-off-by: Alex Deucher 
(cherry picked from commit fae5984296b981c8cc3acca35b701c1f332a6cd8)
Cc: stable@vger.kernel.org
Signed-off-by: Greg Kroah-Hartman

drm/amd: Fix hang on amdgpu unload by using pci_dev_is_disconnected()

2026-03-25T10:05:35+00:00

[ Upstream commit f7afda7fcd169a9168695247d07ad94cf7b9798f ]

The commit 6a23e7b4332c ("drm/amd: Clean up kfd node on surprise
disconnect") introduced early KFD cleanup when drm_dev_is_unplugged()
returns true. However, this causes hangs during normal module unload
(rmmod amdgpu).

The issue occurs because drm_dev_unplug() is called in amdgpu_pci_remove()
for all removal scenarios, not just surprise disconnects. This was done
intentionally in commit 39934d3ed572 ("Revert "drm/amdgpu: TA unload
messages are not actually sent to psp when amdgpu is uninstalled"") to
fix IGT PCI software unplug test failures. As a result,
drm_dev_is_unplugged() returns true even during normal module unload,
triggering the early KFD cleanup inappropriately.

The correct check should distinguish between:
- Actual surprise disconnect (eGPU unplugged): pci_dev_is_disconnected()
  returns true
- Normal module unload (rmmod): pci_dev_is_disconnected() returns false

Replace drm_dev_is_unplugged() with pci_dev_is_disconnected() to ensure
the early cleanup only happens during true hardware disconnect events.

Cc: stable@vger.kernel.org
Reported-by: Cal Peake 
Closes: https://lore.kernel.org/all/b0c22deb-c0fa-3343-33cf-fd9a77d7db99@absolutedigital.net/
Fixes: 6a23e7b4332c ("drm/amd: Clean up kfd node on surprise disconnect")
Acked-by: Alex Deucher 
Signed-off-by: Mario Limonciello 
Signed-off-by: Alex Deucher 
Signed-off-by: Sasha Levin

drm/amdgpu: Replace Mutex with Spinlock for RLCG register access to avoid Priority Inversion in SRIOV

2026-02-06T15:48:27+00:00

[ Upstream commit dc0297f3198bd60108ccbd167ee5d9fa4af31ed0 ]

RLCG Register Access is a way for virtual functions to safely access GPU
registers in a virtualized environment., including TLB flushes and
register reads. When multiple threads or VFs try to access the same
registers simultaneously, it can lead to race conditions. By using the
RLCG interface, the driver can serialize access to the registers. This
means that only one thread can access the registers at a time,
preventing conflicts and ensuring that operations are performed
correctly. Additionally, when a low-priority task holds a mutex that a
high-priority task needs, ie., If a thread holding a spinlock tries to
acquire a mutex, it can lead to priority inversion. register access in
amdgpu_virt_rlcg_reg_rw especially in a fast code path is critical.

The call stack shows that the function amdgpu_virt_rlcg_reg_rw is being
called, which attempts to acquire the mutex. This function is invoked
from amdgpu_sriov_wreg, which in turn is called from
gmc_v11_0_flush_gpu_tlb.

The [ BUG: Invalid wait context ] indicates that a thread is trying to
acquire a mutex while it is in a context that does not allow it to sleep
(like holding a spinlock).

Fixes the below:

[  253.013423] =============================
[  253.013434] [ BUG: Invalid wait context ]
[  253.013446] 6.12.0-amdstaging-drm-next-lol-050225 #14 Tainted: G     U     OE
[  253.013464] -----------------------------
[  253.013475] kworker/0:1/10 is trying to lock:
[  253.013487] ffff9f30542e3cf8 (&adev->virt.rlcg_reg_lock){+.+.}-{3:3}, at: amdgpu_virt_rlcg_reg_rw+0xf6/0x330 [amdgpu]
[  253.013815] other info that might help us debug this:
[  253.013827] context-{4:4}
[  253.013835] 3 locks held by kworker/0:1/10:
[  253.013847]  #0: ffff9f3040050f58 ((wq_completion)events){+.+.}-{0:0}, at: process_one_work+0x3f5/0x680
[  253.013877]  #1: ffffb789c008be40 ((work_completion)(&wfc.work)){+.+.}-{0:0}, at: process_one_work+0x1d6/0x680
[  253.013905]  #2: ffff9f3054281838 (&adev->gmc.invalidate_lock){+.+.}-{2:2}, at: gmc_v11_0_flush_gpu_tlb+0x198/0x4f0 [amdgpu]
[  253.014154] stack backtrace:
[  253.014164] CPU: 0 UID: 0 PID: 10 Comm: kworker/0:1 Tainted: G     U     OE      6.12.0-amdstaging-drm-next-lol-050225 #14
[  253.014189] Tainted: [U]=USER, [O]=OOT_MODULE, [E]=UNSIGNED_MODULE
[  253.014203] Hardware name: Microsoft Corporation Virtual Machine/Virtual Machine, BIOS Hyper-V UEFI Release v4.1 11/18/2024
[  253.014224] Workqueue: events work_for_cpu_fn
[  253.014241] Call Trace:
[  253.014250]  
[  253.014260]  dump_stack_lvl+0x9b/0xf0
[  253.014275]  dump_stack+0x10/0x20
[  253.014287]  __lock_acquire+0xa47/0x2810
[  253.014303]  ? srso_alias_return_thunk+0x5/0xfbef5
[  253.014321]  lock_acquire+0xd1/0x300
[  253.014333]  ? amdgpu_virt_rlcg_reg_rw+0xf6/0x330 [amdgpu]
[  253.014562]  ? __lock_acquire+0xa6b/0x2810
[  253.014578]  __mutex_lock+0x85/0xe20
[  253.014591]  ? amdgpu_virt_rlcg_reg_rw+0xf6/0x330 [amdgpu]
[  253.014782]  ? sched_clock_noinstr+0x9/0x10
[  253.014795]  ? srso_alias_return_thunk+0x5/0xfbef5
[  253.014808]  ? local_clock_noinstr+0xe/0xc0
[  253.014822]  ? amdgpu_virt_rlcg_reg_rw+0xf6/0x330 [amdgpu]
[  253.015012]  ? srso_alias_return_thunk+0x5/0xfbef5
[  253.015029]  mutex_lock_nested+0x1b/0x30
[  253.015044]  ? mutex_lock_nested+0x1b/0x30
[  253.015057]  amdgpu_virt_rlcg_reg_rw+0xf6/0x330 [amdgpu]
[  253.015249]  amdgpu_sriov_wreg+0xc5/0xd0 [amdgpu]
[  253.015435]  gmc_v11_0_flush_gpu_tlb+0x44b/0x4f0 [amdgpu]
[  253.015667]  gfx_v11_0_hw_init+0x499/0x29c0 [amdgpu]
[  253.015901]  ? __pfx_smu_v13_0_update_pcie_parameters+0x10/0x10 [amdgpu]
[  253.016159]  ? srso_alias_return_thunk+0x5/0xfbef5
[  253.016173]  ? smu_hw_init+0x18d/0x300 [amdgpu]
[  253.016403]  amdgpu_device_init+0x29ad/0x36a0 [amdgpu]
[  253.016614]  amdgpu_driver_load_kms+0x1a/0xc0 [amdgpu]
[  253.017057]  amdgpu_pci_probe+0x1c2/0x660 [amdgpu]
[  253.017493]  local_pci_probe+0x4b/0xb0
[  253.017746]  work_for_cpu_fn+0x1a/0x30
[  253.017995]  process_one_work+0x21e/0x680
[  253.018248]  worker_thread+0x190/0x330
[  253.018500]  ? __pfx_worker_thread+0x10/0x10
[  253.018746]  kthread+0xe7/0x120
[  253.018988]  ? __pfx_kthread+0x10/0x10
[  253.019231]  ret_from_fork+0x3c/0x60
[  253.019468]  ? __pfx_kthread+0x10/0x10
[  253.019701]  ret_from_fork_asm+0x1a/0x30
[  253.019939]  

v2: s/spin_trylock/spin_lock_irqsave to be safe (Christian).

Fixes: e864180ee49b ("drm/amdgpu: Add lock around VF RLCG interface")
Cc: lin cao 
Cc: Jingwen Chen 
Cc: Victor Skvortsov 
Cc: Zhigang Luo 
Cc: Christian König 
Cc: Alex Deucher 
Signed-off-by: Srinivasan Shanmugam 
Suggested-by: Alex Deucher 
Reviewed-by: Christian König 
Signed-off-by: Alex Deucher 
[ Minor conflict resolved. ]
Signed-off-by: Li hongliang <1468888505@139.com>
Signed-off-by: Greg Kroah-Hartman

drm/amd: Clean up kfd node on surprise disconnect

2026-01-30T09:27:28+00:00

commit 28695ca09d326461f8078332aa01db516983e8a2 upstream.

When an eGPU is unplugged the KFD topology should also be destroyed
for that GPU. This never happens because the fini_sw callbacks never
get to run. Run them manually before calling amdgpu_device_ip_fini_early()
when a device has already been disconnected.

This location is intentionally chosen to make sure that the kfd locking
refcount doesn't get incremented unintentionally.

Cc: kent.russell@amd.com
Closes: https://community.frame.work/t/amd-egpu-on-linux/8691/33
Signed-off-by: Mario Limonciello (AMD) 
Reviewed-by: Kent Russell 
Signed-off-by: Alex Deucher 
(cherry picked from commit 6a23e7b4332c10f8b56c33a9c5431b52ecff9aab)
Cc: stable@vger.kernel.org
Signed-off-by: Greg Kroah-Hartman

drm/amdgpu: fix cyan_skillfish2 gpu info fw handling

2025-12-06T21:18:49+00:00

[ Upstream commit 7fa666ab07ba9e08f52f357cb8e1aad753e83ac6 ]

If the board supports IP discovery, we don't need to
parse the gpu info firmware.

Backport to 6.18.

Closes: https://gitlab.freedesktop.org/drm/amd/-/issues/4721
Fixes: fa819e3a7c1e ("drm/amdgpu: add support for cyan skillfish gpu_info")
Signed-off-by: Alex Deucher 
(cherry picked from commit 5427e32fa3a0ba9a016db83877851ed277b065fb)
Signed-off-by: Sasha Levin

drm/amd: Avoid evicting resources at S5

2025-11-24T09:29:38+00:00

[ Upstream commit 531df041f2a5296174abd8292d298eb62fe1ea97 ]

Normally resources are evicted on dGPUs at suspend or hibernate and
on APUs at hibernate.  These steps are unnecessary when using the S4
callbacks to put the system into S5.

Cc: AceLan Kao 
Cc: Kai-Heng Feng 
Cc: Mark Pearson 
Cc: Denis Benato 
Cc: Merthan Karakaş 
Tested-by: Eric Naim 
Acked-by: Alex Deucher 
Signed-off-by: Mario Limonciello (AMD) 
Signed-off-by: Alex Deucher 
Signed-off-by: Sasha Levin

drm/amdgpu: add support for cyan skillfish gpu_info

2025-11-24T09:29:34+00:00

[ Upstream commit fa819e3a7c1ee994ce014cc5a991c7fd91bc00f1 ]

Some SOCs which are part of the cyan skillfish family
rely on an explicit firmware for IP discovery.  Add support
for the gpu_info firmware.

Signed-off-by: Alex Deucher 
Signed-off-by: Sasha Levin

drm/amdgpu: switch job hw_fence to amdgpu_fence

2025-07-06T09:00:15+00:00

commit ebe43542702c3d15d1a1d95e8e13b1b54076f05a upstream.

Use the amdgpu fence container so we can store additional
data in the fence.  This also fixes the start_time handling
for MCBP since we were casting the fence to an amdgpu_fence
and it wasn't.

Fixes: 3f4c175d62d8 ("drm/amdgpu: MCBP based on DRM scheduler (v9)")
Reviewed-by: Christian König 
Signed-off-by: Alex Deucher 
(cherry picked from commit bf1cd14f9e2e1fdf981eed273ddd595863f5288c)
Cc: stable@vger.kernel.org
Signed-off-by: Greg Kroah-Hartman

drm/amdgpu: fix pm notifier handling

2025-05-22T12:12:26+00:00

commit 4aaffc85751da5722e858e4333e8cf0aa4b6c78f upstream.

Set the s3/s0ix and s4 flags in the pm notifier so that we can skip
the resource evictions properly in pm prepare based on whether
we are suspending or hibernating.  Drop the eviction as processes
are not frozen at this time, we we can end up getting stuck trying
to evict VRAM while applications continue to submit work which
causes the buffers to get pulled back into VRAM.

v2: Move suspend flags out of pm notifier (Mario)

Closes: https://gitlab.freedesktop.org/drm/amd/-/issues/4178
Fixes: 2965e6355dcd ("drm/amd: Add Suspend/Hibernate notification callback support")
Cc: Mario Limonciello 
Reviewed-by: Mario Limonciello 
Signed-off-by: Alex Deucher 
(cherry picked from commit 06f2dcc241e7e5c681f81fbc46cacdf4bfd7d6d7)
Cc: stable@vger.kernel.org
Signed-off-by: Greg Kroah-Hartman

Revert "drm/amd: Stop evicting resources on APUs in suspend"

2025-05-22T12:12:14+00:00

[ Upstream commit d0ce1aaa8531a4a4707711cab5721374751c51b0 ]

This reverts commit 3a9626c816db901def438dc2513622e281186d39.

This breaks S4 because we end up setting the s3/s0ix flags
even when we are entering s4 since prepare is used by both
flows.  The causes both the S3/s0ix and s4 flags to be set
which breaks several checks in the driver which assume they
are mutually exclusive.

Closes: https://gitlab.freedesktop.org/drm/amd/-/issues/3634
Cc: Mario Limonciello 
Reviewed-by: Mario Limonciello 
Signed-off-by: Alex Deucher 
(cherry picked from commit ce8f7d95899c2869b47ea6ce0b3e5bf304b2fff4)
Cc: stable@vger.kernel.org
Signed-off-by: Sasha Levin