"jianchao.wang" <jianchao.w.wang@xxxxxxxxxx> writes: > On 4/9/19 1:13 PM, Stewart Smith wrote: >> cpu 0x4a: Vector: 400 (Instruction Access) at [c000203ff398b100] >> pc: c0000000021fe700 >> lr: c0000000002001d8: blk_mq_complete_request+0x34/0x138 > > Would you please figure out what is the source code here with gdb ? > > From the backtrace, it seems not the same issue, because no dm-mpath > is involved here. It turns out it's not the same issue, we just managed to hit it at the same time as the original issue. It's actually a bug in the STRICT_KERNEL_RWX=y code on ppc64le with CONFIG_HUGETLB_PAGE=n that manifests as corrupting random kernel memory and we just happen to hit it in this codepath and only once we brought in "Revert v5.0 blk_mq_request_issue_directly". More details over at https://github.com/linuxppc/issues/issues/237 and there should be a fix for STRICT_KERNEL_RWX coming at some point. -- Stewart Smith OPAL Architect, IBM.
