fix(recovery): a GPU access fault no longer fails other address spaces

An address space's check of the SMMU returned the fault status while the
GPU's own faulting access was still latched, and the address space that
happened to map, unmap or attach next was taken out of use: seven processes
in one run, the render server among them. The status says nothing about
that address space's tables, so the operation now goes on and the fault is
counted. The hang check ends the fault, and also clears the SMMU's global
fault status where it is set.

A stalled job's command buffers are searched for the IB2 address the thread
stopped in, to tell an address the job's commands named from one they did
not.
This commit is contained in:
rponeawa
2026-11-12 07:41:49 +08:00
Unverified
parent a8fa18fa09
commit 031f35ccfb
4 changed files with 62 additions and 3 deletions
+9
View File
@@ -193,11 +193,20 @@ IOReturn NativeAdrenoGpuContext::clearFault(uint32_t fsr,uint32_t far[3])
if(io->write32(io->cookie,(size_t)offset+0x58,fsr) || io->barrier(io->cookie))return kIOReturnIOError;
return kIOReturnSuccess;
}
extern "C" {extern volatile uint32_t native_adreno_global_fault[4];}
IOReturn NativeAdrenoGpuContext::endFault(uint32_t out[4])
{
if(!out || !ready_)return kIOReturnBadArgument;
out[0]=out[1]=out[2]=out[3]=0;
const ArmSmmu4KIo *io=smmu_.ioHandle();
{ /* The global fault status the GPU's faulting access left (sGFSR, write one to clear): cleared with the context's, or every
* later look at the SMMU reads it. Its last value is kept. */
uint32_t global=0;
if(!io->read32(io->cookie,0x48,&global) && global) {
native_adreno_global_fault[2]=native_adreno_global_fault[2]+1;native_adreno_global_fault[3]=global;
(void)io->write32(io->cookie,0x48,global);
}
}
const uint64_t offset=smmu_.stateHandle()->context_offset;
if(!offset)return kIOReturnNotReady;
if(io->read32(io->cookie,(size_t)offset+0x58,&out[0]))return kIOReturnIOError;