On Tue, Oct 17, 2017 at 09:50:04AM +0200, Thomas Gleixner wrote:
> On Tue, 17 Oct 2017, Joonsoo Kim wrote:
> > On Wed, Oct 11, 2017 at 12:01:20PM -0500, Josh Poimboeuf wrote:
> > > > Looking at the panic, the code in slob_free() was:
> > > >
> > > > 0: e8 8d f7 ff ff callq 0xfffffffffffff792
> > > > 5: 48 ff 05 c9 8c 91 02 incq 0x2918cc9(%rip) # 0x2918cd5
> > > > c: 85 c0 test %eax,%eax
> > > > e: 75 51 jne 0x61
> > > > 10: 49 0f bf c5 movswq %r13w,%rax
> > > > 14: 48 ff 05 c2 8c 91 02 incq 0x2918cc2(%rip) # 0x2918cdd
> > > > 1b: 48 8d 3c 43 lea (%rbx,%rax,2),%rdi
> > > > 1f: 48 39 ef cmp %rbp,%rdi
> > > > 22: 75 3d jne 0x61
> > > > 24: 48 ff 05 ba 8c 91 02 incq 0x2918cba(%rip) # 0x2918ce5
> > > > 2b:* 8b 6d 00 mov 0x0(%rbp),%ebp <-- trapping instruction
> > > > 2e: 66 85 ed test %bp,%bp
> > > > 31: 7e 09 jle 0x3c
> > > > 33: 48 ff 05 b3 8c 91 02 incq 0x2918cb3(%rip) # 0x2918ced
> > > > 3a: eb 05 jmp 0x41
> > > > 3c: bd .byte 0xbd
> > > > 3d: 01 00 add %eax,(%rax)
> > > >
> > > > The slob_free() code tried to read four bytes at ffff88001c4afffe, and
> > > > ended up reading past the page into a bad area. I think the bad address
> > > > (ffff88001c4afffe) was returned from slob_next() and it panicked trying
> > > > to read s->units in slob_units().
> >
> > Hello,
> >
> > It looks like a compiler bug. The code of slob_units() try to read two
> > bytes at ffff88001c4afffe. It's valid. But the compiler generates
> > wrong code that try to read four bytes.
> >
> > static slobidx_t slob_units(slob_t *s)
> > {
> > if (s->units > 0)
> > return s->units;
> > return 1;
> > }
> >
> > s->units is defined as two bytes in this setup.
> >
> > Wrongly generated code for this part.
> >
> > 'mov 0x0(%rbp), %ebp'
> >
> > %ebp is four bytes.
> >
> > I guess that this wrong four bytes read cross over the valid memory
> > boundary and this issue happend.
> >
> > Proper code (two bytes read) is generated if different version of gcc
> > is used.
>
> Which version fails to generate proper code and which versions work?
>
gcc 4.8 and 4.9 fails to generate proper code. gcc 5.1 and
the latest version works fine.
I guess that this problem is related to the corner case of some
optimization feature since minor code change makes the result
different. And, with -O2, proper code is generated even if gcc 4.8 is
used.
Thanks.
--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@xxxxxxxxx. For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"dont@xxxxxxxxx";> email@xxxxxxxxx </a>
