On 09/30, Linus Torvalds wrote: >On Thu, Sep 21, 2017 at 12:02 AM, NeilBrown <ne...@suse.com> wrote: >> >> I think it is crashing in >> static inline bool ata_is_host_link(const struct ata_link *link) >> { >> return link == &link->ap->link || link == link->ap->slave_link; >> } > >Yes. The code is > > 1a: 8b 3a mov (%edx),%edi > 1c: 8d 8f 40 16 00 00 lea 0x1640(%edi),%ecx > 22: 39 ca cmp %ecx,%edx > 24: 74 49 je 0x6f > 26: b9 01 00 00 00 mov $0x1,%ecx > 2b:* 39 97 80 24 00 00 cmp %edx,0x2480(%edi) <-- trapping instruction > 31: 74 3c je 0x6f > >and that first "je" is the test for "link == &link->ap->link" (which >only takes the address relative to "link->ap" - thus the "lea"), and >that cmp that oopses is indeed loading that actual slave_link value. > >So I agree. "link->ap" is NULL for some odd reason. > >Hmm. Absolutely nothing has changed in libata-core.c recently, >certainly not that async_port_probe() thing. > >So I suspect either it's just a timing difference, or it's some >unrelated memory corruption. > >Xiaolong, I see that you have SLUB_DEBUG and SLUB_DEBUG_ON enabled, >but wonder if you can recreate this with DEBUG_PAGEALLOC and/or >DEBUG_OBJECTS enabled too? >
Sorry for the late, just come back from vacation. I retested with with DEBUG_PAGEALLOC and DEBUG_OBJECTS enabled for both commit 47e0fb461f and commit 011067b056, and both showed the ata_dev_next crash issue, so commit 47e0fb461f should be irrelevant, sorry for the noise. Thanks, Xiaolong >Tejun, any ideas? The original report is at > > https://lkml.org/lkml/2017/9/20/939 > >in case you don't see it in your inbox from lkml. > > Linus