vrbagal1 <vrbagal1@xxxxxxxxxxxxxxxxxx> writes: > On 2018-07-10 13:37, Nicholas Piggin wrote: >> On Tue, 10 Jul 2018 11:58:40 +0530 >> vrbagal1 <vrbagal1@xxxxxxxxxxxxxxxxxx> wrote: >> >>> Hi, >>> >>> Observing kernel oops on Power9(ZZ) box, running on PowerVM, while >>> running stress-ng. >>> >>> >>> Kernel: 4.18.0-rc4 >>> Machine: Power9 ZZ (PowerVM) >>> Test: Stress-ng >>> >>> Attached is .config file >>> >>> Traces: >>> >>> [12251.245209] Oops: Kernel access of bad area, sig: 11 [#1] >> >> Can you post the lines above this? Otherwise we don't know what address >> it tried to access (without decoding the instructions and >> reconstructing >> it from registers at least, which the XFS devs wouldn't be inclined to >> do). >> > > ah my bad. > > [12251.245179] Unable to handle kernel paging request for data at address 0x6000000060000000 > [12251.245199] Faulting instruction address: 0xc000000000319e2c Which matches the regs & disassembly: r4 = 6000000060000000 r9 = 0 ldx r9,r4,r9 <- pop So object was 0x6000000060000000. That looks like two nops, ie. we got some code? And there's only one caller of prefetch_freepointer() in slab_alloc_node(): prefetch_freepointer(s, next_object); So slab corruption is looking likely. Do you have slub_debug=FZP on the kernel command line? cheers
