Re: [GIT PULL] x86/topology changes for v5.3

[Jiri Kosina]

On Wed, 10 Jul 2019, Peter Zijlstra wrote:

> > > BUG: unable to handle page fault for address: ffffffff9edc1598
> > > #PF: supervisor write access in kernel mode
> > > #PF: error_code(0x0003) - permissions violation
> > > PGD 1a20c067 P4D 1a20c067 PUD 1a20d063 PMD 8000000019e000e1 
> > > Oops: 0003 [#1] SMP PTI
> > > 2 PID: 151 Comm: systemd-udevd Not tainted 5.2.0+ #54
> > > Hardware name: LENOVO 20175/INVALID, BIOS 66CN54WW 01/21/2013
> > > RIP: 0010:static_key_set_mod.isra.0+0x10/0x30
> > > Code: 48 8b 37 83 e6 03 48 09 c6 48 89 37 c3 66 66 2e 0f 1f 84 00 00 00 00 00 66 90 48 89 f0 a8 03 75 0d 48 8b 37 83 e6 03 48 09 c6 <48> 89 37 c3 0f 0b 48 8b 37 83 e6 03 48 09 c6 48 89 37 c3 66 66 2e
> > > RSP: 0000:ffffa606c032bc98 EFLAGS: 00010286
> > > RAX: ffff9981ddce30a0 RBX: ffffffff9edc1590 RCX: 0000000000000000
> > > RDX: 0000000000000020 RSI: ffff9981ddce30a0 RDI: ffffffff9edc1598
> > > RBP: ffffffffc06f4000 R08: ffff9981e6003980 R09: ffff9981ddce30a0
> > > R10: 0000000000000000 R11: 0000000000028b56 R12: ffffffffc06f8880
> > > R13: ffff9981ddce3080 R14: ffffffffc06f4008 R15: ffffffffc06f6dc0
> > > FS:  00007f992dd9a680(0000) GS:ffff9981e7080000(0000) knlGS:0000000000000000
> > > CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
> > > CR2: ffffffff9edc1598 CR3: 00000002233aa001 CR4: 00000000001606e0
> > > Call Trace:
> > >   jump_label_module_notify+0x1e7/0x2b0
> > >   notifier_call_chain+0x44/0x70
> > >   blocking_notifier_call_chain+0x43/0x60
> > >   load_module+0x1bcb/0x2490
> > >   ? vfs_read+0x11f/0x150
> > >   ? __do_sys_finit_module+0xbf/0xe0
> > >   __do_sys_finit_module+0xbf/0xe0
> > >   do_syscall_64+0x43/0x110
> > >   entry_SYSCALL_64_after_hwframe+0x44/0xa9
> > > 
> > > Josh, didn't you mention that yesterday or so?
> > 
> > That's what Tony yesterday indicated on IRC that his system is suffering 
> > from as well.
> > 
> > Adding Daniel to check whether this couldn't be some fallout of jumplabel 
> > batching.
> 
> AFAICT this is _before_ we get to patching. The function that explodes
> is static_key_set_mod(), as called from jump_label_add_module().
> 
> What that function does is for all patch sites in the module, find the
> corresponding key; if that key is not also in that module, allocate a
> static_key_mod structure and link the module entries to the key. Such
> that we can find all instances from a given key.
> 
> I don't think anything here has changed in a while.

Hm, and it seems to explode on dereferencing the static_key* in %rsi

  21:   48 8b 37                mov    (%rdi),%rsi
  24:   83 e6 03                and    $0x3,%esi
  27:   48 09 c6                or     %rax,%rsi
  2a:*  48 89 37                mov    %rsi,(%rdi)              <-- trapping instruction

which looks odd, as it derefenced it successfully just 3 instructions ago.

-- 
Jiri Kosina
SUSE Labs