Messages in this thread | | | Date | Thu, 9 Apr 2015 17:07:43 -0700 | Subject | Re: NULL deref around blkmq in v4.0-rc1–rc7 | From | Linus Torvalds <> |
| |
On Thu, Apr 9, 2015 at 3:29 PM, Jan Engelhardt <jengelh@inai.de> wrote: > > On Thursday 2015-04-09 23:12, Linus Torvalds wrote: >>That ->sense_buffer thing is supposed to be initialized by the >>blk_mq_ops.init_request() function, which is called - if it exists = >>when the array of requests ('->rqs[]') is initialized. >>[...] >>Does this entirely untested patch make any difference? > > Yes, this seems to solve it for me.
Can you humor me, and try that patch on top of a few different kernel versions that you had trouble with.
The fact that your bisection ended on a completely unrelated commit makes me think that some particular memory aligment (code or initial data or whatever) ends up showing or hiding the bug, so I'm a bit worried that my ->rqs[] initialization patch also happened to just hide the problem rather than really fix it.
That said, at least it's a lot more likely to be relevant than the ACPI merge commit you bisected to, so while I'm somewhat cautious about that patch (not that it would be wrong, but that it would matter that much) I at least can see it make a real difference.
Linus
| |