linux-next.git/drivers/gpu/drm/amd/ras, branch master

drm/amd/ras: take a log batch id only once there is something to log

2026-09-02T18:50:45+00:00

The id is claimed before the banks are read, so a poll that ends up logging nothing still burns one and leaves a gap that every reader of the log then has to walk over. Claim it when the first record of the batch is about to go in. Signed-off-by: Xiang Liu Reviewed-by: Hawking Zhang Signed-off-by: Alex Deucher

drm/amd/ras: report how far a CPER record query got

2026-09-02T18:50:42+00:00

The caller walks the log with cper_start_id and resumes at cper_start_id + real_cper_num, but the reply only counts the ids that held a record. A batch id that holds nothing therefore never moves the caller forward, and every such id makes it read the next populated batch one more time. Count the ids covered instead. The rewind to the last populated batch when the query starts at the end has the same effect once the caller has drained the log, and the latest id is already reported by the snapshot query, so drop it. Signed-off-by: Xiang Liu Reviewed-by: Hawking Zhang Signed-off-by: Alex Deucher

drm/amd/ras: report the banks the scrubber logged

2026-09-02T18:50:30+00:00

A bank carrying the scrub bit was found by the background scrubber rather than by an access, which is worth telling apart from the rest when reading a log. Signed-off-by: Xiang Liu Reviewed-by: Hawking Zhang Signed-off-by: Alex Deucher

drm/amd/ras: drop the ras_ prefix from the ACA v5 file names

2026-09-02T18:50:27+00:00

The prefix marks the files that back a hardware IP block. ACA is a bank format rather than an IP, and the rest of its files are named after it alone, so the v5 pair is the odd one out. Signed-off-by: Xiang Liu Reviewed-by: Hawking Zhang Signed-off-by: Alex Deucher

drm/amd/ras: skip poison banks read through the UE channel

2026-09-02T18:50:15+00:00

Outside of UMC a bank with the poison bit set describes the consumption of an error that was already reported, not a new uncorrectable one. Injecting an uncorrectable error while a workload is running therefore buries the real bank under a dump of GFX poison banks collected from every other device of the hive. Signed-off-by: Xiang Liu Reviewed-by: Hawking Zhang Signed-off-by: Alex Deucher

drm/amd/ras: add supported aca blocks mapping

2026-08-25T22:46:12+00:00

Add dacee_be, ucie_pcs, lsdma aca blocks mapping, add pcs_xgmi aca block info. Signed-off-by: Stanley.Yang Reviewed-by: Hawking Zhang Signed-off-by: Alex Deucher

drm/amd/ras: record the fatal state on every device of the hive

2026-08-25T22:41:11+00:00

The fatal error interrupt is broadcast to every device of the hive and they all race for amdgpu_ras_global_ras_isr(), which hands -EBUSY to everyone but the winner. Treating that as a failure returns before the device is marked, so seven devices out of eight are left without their fatal and SRAM ECC state, the one that actually logged the error among them. KFD then tells the processes on those devices that the reset was a plain hang. -EBUSY only means the reset has already been asked for. Record the state anyway and leave the request to the winner. Signed-off-by: Xiang Liu Reviewed-by: Hawking Zhang Signed-off-by: Alex Deucher

drm/amd/ras: tell KFD the reset came from an ECC error

2026-08-25T22:41:09+00:00

kfd_signal_reset_event() picks between KFD_HW_EXCEPTION_ECC and KFD_HW_EXCEPTION_GPU_HANG from the SRAM ECC flag, and only delivers the memory exception event for the former. Nothing raises that flag on the RAS module paths, so a reset caused by an uncorrectable or a consumed poison error is reported to every process on the device as a plain hang and the runtime carries on instead of tearing the workload down. Raise it the way the per IP callbacks used to. Signed-off-by: Xiang Liu Reviewed-by: Hawking Zhang Signed-off-by: Alex Deucher

drm/amd/ras: drop the RAS module state when its sw init fails

2026-08-25T22:41:07+00:00

The failure paths free the RAS manager but keep the flags that advertise it as the owner of RAS and of the RAS firmware load. Every entry point gated on those hits the NULL manager, hw_init turns that into -EINVAL and the whole probe fails. Clear both once the manager is gone, and route the allocation failure through the same place. Signed-off-by: Xiang Liu Reviewed-by: Tao Zhou Signed-off-by: Alex Deucher

drm/amd/ras: tolerate an unsupported MCA debug mode control

2026-08-25T22:41:05+00:00

ras_mp1_set_debug_mode() reports -EOPNOTSUPP where the control does not exist, an SR-IOV guest among them. Treating that as fatal fails ras_core_sw_init() and takes RAS down on every VF. Nothing else in the MP1 block depends on it, so carry on. Signed-off-by: Xiang Liu Reviewed-by: Tao Zhou Signed-off-by: Alex Deucher