Commit 8f1de51f49be ("drm/amdgpu: prevent immediate PASID reuse case")
converted the global PASID allocator from IDA to IDR with a spinlock
for cyclic allocation. This introduced two locking bugs:
1) idr_alloc_cyclic() is called with GFP_KERNEL under spin_lock(),
which can sleep.
2) amdgpu_pasid_free() can be called from hardirq context via the
fence signal path (amdgpu_pasid_free_cb), but the lock is taken
with plain spin_lock() in process context, creating a potential
deadlock:
CPU0
----
spin_lock(&amdgpu_pasid_idr_lock) // process context, IRQs on
<Interrupt>
spin_lock(&amdgpu_pasid_idr_lock) // deadlock
The hardirq call chain is:
sdma_v6_0_process_trap_irq
-> amdgpu_fence_process
-> dma_fence_signal
-> drm_sched_job_done
-> dma_fence_signal
-> amdgpu_pasid_free_cb
-> amdgpu_pasid_free
This was observed on an RX 7900 XTX when exiting a Vulkan game
running under Proton/Wine, which triggers the fence callback path
during VM teardown.
Patch 1 fixes the sleeping-under-spinlock by using idr_preload() with
GFP_KERNEL before taking the lock, then GFP_NOWAIT for the actual
allocation.
Patch 2 converts all three spin_lock/spin_unlock call sites to
spin_lock_irqsave/spin_unlock_irqrestore.
Tested on ASUS ROG STRIX B650E-I / Ryzen 9 7950X / RX 7900 XTX with
CONFIG_PROVE_LOCKING=y. The lockdep warning is no longer triggered
after applying both patches.
Mikhail Gavrilov (2):
drm/amdgpu: fix sleeping allocation under spinlock in PASID IDR
drm/amdgpu: use spin_lock_irqsave for PASID IDR lock
drivers/gpu/drm/amd/amdgpu/amdgpu_ids.c | 20 +++++++++++++-------
1 file changed, 13 insertions(+), 7 deletions(-)
--
2.53.0
