[tip:locking/rwsem] x86/rwsem: Save and restore all callee-clobbered regs in 32-bit ____down_write()

[tip-bot for Borislav Petkov]

Commit-ID:  9da6e1cf7f044569a7e8a607ecdea6f5192c6bce
Gitweb:     http://git.kernel.org/tip/9da6e1cf7f044569a7e8a607ecdea6f5192c6bce
Author:     Borislav Petkov <bp@xxxxxxx>
AuthorDate: Thu, 12 May 2016 19:29:38 +0200
Committer:  Ingo Molnar <mingo@xxxxxxxxxx>
CommitDate: Fri, 13 May 2016 11:15:19 +0200

x86/rwsem: Save and restore all callee-clobbered regs in 32-bit ____down_write()

____down_write() calls a function to handle the slow path when the lock
is contended. But in order to be able to call a C function, one has to
stash all callee-clobbered registers. The 32-bit path saves only %ecx
for a reason unknown to me. However, after

  71c01930b42e ("locking/rwsem, x86: Clean up ____down_write()")

the useless dependency on edx was removed and this caused the following
splat:

  BUG: unable to handle kernel NULL pointer dereference at 00000015
  IP: [<c185e094>] down_write+0x24/0x30
  *pde = 00000000
  Oops: 0002 [#1] PREEMPT SMP
  Modules linked in:
  CPU: 0 PID: 1 Comm: swapper/0 Tainted: G S      W       4.6.0-rc7-next-20160511-yocto-standard #1
  Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS 1.7.5-20140531_083030-gandalf 04/01/2014
  task: f4d00000 ti: f4d08000 task.ti: f4d08000
  EIP: 0060:[<c185e094>] EFLAGS: 00210282 CPU: 0
  EIP is at down_write+0x24/0x30
  EAX: f4d00000 EBX: f4f6d600 ECX: ffff0001 EDX: 00000001
  ESI: 00000168 EDI: c1c2eb68 EBP: f4d09ef4 ESP: f4d09eec
   DS: 007b ES: 007b FS: 00d8 GS: 0000 SS: 0068
  CR0: 80050033 CR2: 00000015 CR3: 01ccb000 CR4: 000406d0

This happens because gcc decided to stash the pointer to @sem in edx (it
is not used in the inline asm anymore, thus free):

  movl    %eax, %edx      # sem, sem

  lock;   xadd      %ecx,(%eax)   # tmp91, sem
  test ...

  call call_rwsem_down_write_failed

  mov    %eax,0x14(%edx)

*before* the slow path happens and if we hit it on 32-bit, it can
clobber edx and we're staring at garbage value at deref time.

The simple fix is to save/restore edx too, around the slow path. We
don't need to stash eax because it is used in the slow path as the @sem
arg.

Reported-by: Guenter Roeck <linux@xxxxxxxxxxxx>
Tested-by: Guenter Roeck <linux@xxxxxxxxxxxx>
Signed-off-by: Borislav Petkov <bp@xxxxxxx>
Cc: Alexander Shishkin <alexander.shishkin@xxxxxxxxxxxxxxx>
Cc: Andrew Morton <akpm@xxxxxxxxxxxxxxxxxxxx>
Cc: Andy Lutomirski <luto@xxxxxxxxxxxxxx>
Cc: Arnaldo Carvalho de Melo <acme@xxxxxxxxxx>
Cc: Borislav Petkov <bp@xxxxxxxxx>
Cc: Brian Gerst <brgerst@xxxxxxxxx>
Cc: Denys Vlasenko <dvlasenk@xxxxxxxxxx>
Cc: H. Peter Anvin <hpa@xxxxxxxxx>
Cc: Jiri Olsa <jolsa@xxxxxxxxxx>
Cc: Linus Torvalds <torvalds@xxxxxxxxxxxxxxxxxxxx>
Cc: Paul E. McKenney <paulmck@xxxxxxxxxxxxxxxxxx>
Cc: Peter Zijlstra <peterz@xxxxxxxxxxxxx>
Cc: Stephane Eranian <eranian@xxxxxxxxxx>
Cc: Thomas Gleixner <tglx@xxxxxxxxxxxxx>
Cc: Vince Weaver <vincent.weaver@xxxxxxxxx>
Cc: linux-next@xxxxxxxxxxxxxxx
Link: http://lkml.kernel.org/r/20160512172938.GB14245@xxxxxxx
Signed-off-by: Ingo Molnar <mingo@xxxxxxxxxx>
---
 arch/x86/lib/rwsem.S | 8 +++++---
 1 file changed, 5 insertions(+), 3 deletions(-)

diff --git a/arch/x86/lib/rwsem.S b/arch/x86/lib/rwsem.S
index a37462a..0224080 100644
--- a/arch/x86/lib/rwsem.S
+++ b/arch/x86/lib/rwsem.S
@@ -33,10 +33,12 @@
  * value or just clobbered..
  */
 
-#define save_common_regs \
-	pushl %ecx
+#define save_common_regs	\
+	pushl %ecx;		\
+	pushl %edx
 
-#define restore_common_regs \
+#define restore_common_regs	\
+	popl %edx;		\
 	popl %ecx
 
 	/* Avoid uglifying the argument copying x86-64 needs to do. */