Set remove_queue_after_reset=1 (MES >= 0x5a) so MES drops its internal state instead of re-unmapping an already MMIO-reset queue, which can timeout into a GPU reset.
Suggested-by: Shaoyun Liu <[email protected]> Signed-off-by: Jesse Zhang <[email protected]> --- drivers/gpu/drm/amd/amdgpu/mes_v12_0.c | 11 +++++++++++ 1 file changed, 11 insertions(+) diff --git a/drivers/gpu/drm/amd/amdgpu/mes_v12_0.c b/drivers/gpu/drm/amd/amdgpu/mes_v12_0.c index d80a983b1b6c..20f4fd57b1da 100644 --- a/drivers/gpu/drm/amd/amdgpu/mes_v12_0.c +++ b/drivers/gpu/drm/amd/amdgpu/mes_v12_0.c @@ -749,6 +749,17 @@ static int mes_v12_0_unmap_legacy_queue(struct amdgpu_mes *mes, mes_remove_queue_pkt.unmap_legacy_queue = 1; mes_remove_queue_pkt.queue_type = convert_to_mes_queue_type(input->queue_type); + /* + * A reset-time unmap: the queue was already reset via MMIO while + * gangs are suspended and it is on the MES hung/fail list. Tell + * MES to just drop its internal state for it. Without this flag + * MES asks CP to unmap the already-reset (still wedged) queue + * again, which times out and forces a GPU reset. + */ + if (input->action == RESET_QUEUES && + (mes->sched_version & AMDGPU_MES_VERSION_MASK) >= 0x5a) + mes_remove_queue_pkt.remove_queue_after_reset = 1; + } if (mes->adev->enable_uni_mes) { -- 2.49.0
