arm64: fpsimd: Prevent registers leaking across exec

PATCH New Dave Martin 2 participants 4 messages Aug 29, 2017 – Sep 1, 2017

Subsystems (2)

Stable branch arm64

lore

Share & tools

lore

Dave Martin

author

2017-08-29 16:49 UTC #0 lore raw

PATCH 1/1 arm64: fpsimd: Prevent registers leaking across exec

commit 096622104e14d8a1db4860bd557717067a0515d2 upstream.

There are some tricky dependencies between the different stages of flushing the FPSIMD register state during exec, and these can race with context switch in ways that can cause the old task's regs to leak across. In particular, a context switch during the memset() can cause some of the task's old FPSIMD registers to reappear.

Disabling preemption for this small window would be no big deal for
performance: preemption is already disabled for similar scenarios like updating the FPSIMD registers in sigreturn.

So, instead of rearranging things in ways that might swap existing subtle bugs for new ones, this patch just disables preemption around the FPSIMD state flushing so that races of this type can't occur here. This brings fpsimd_flush_thread() into line with other code paths.

Cc <[email protected]> # 3.16.x

Cc <[email protected]> # 4.1.x-4.9.x

Fixes 674c242c9323 ("arm64: flush FP/SIMD state correctly after execve()")

Reviewed-by Ard Biesheuvel <[email protected]>

Signed-off-by Dave Martin <[email protected]>

Signed-off-by Will Deacon <[email protected]>

---

For stable only.

3.17.x-4.0.x don't appear active, and this patch isn't sufficient to fix them (they would need 674c242c9323 also).

 arch/arm64/kernel/fpsimd.c | 2 ++
 1 file changed, 2 insertions(+)

arch/arm64/kernel/fpsimd.c +2 -0

index 394c61d..1d5890f 100644
@@ -157,9 +157,11 @@ void fpsimd_thread_switch(struct task_struct *next)
 
 void fpsimd_flush_thread(void)
 {
+	preempt_disable();
 	memset(&current->thread.fpsimd_state, 0, sizeof(struct fpsimd_state));
 	fpsimd_flush_task_state(current);
 	set_thread_flag(TIF_FOREIGN_FPSTATE);
+	preempt_enable();
 }
 
 /*

Signature · 1 line

2.1.4

Greg KH

2017-08-31 06:01 UTC #1 lore raw

PATCH 1/1 arm64: fpsimd: Prevent registers leaking across exec

Quoted 24 lines · Dave Martin

commit 096622104e14d8a1db4860bd557717067a0515d2 upstream.
There are some tricky dependencies between the different stages of flushing the FPSIMD register state during exec, and these can race with context switch in ways that can cause the old task's regs to leak across. In particular, a context switch during the memset() can cause some of the task's old FPSIMD registers to reappear.
Disabling preemption for this small window would be no big deal for
performance: preemption is already disabled for similar scenarios like updating the FPSIMD registers in sigreturn.
So, instead of rearranging things in ways that might swap existing subtle bugs for new ones, this patch just disables preemption around the FPSIMD state flushing so that races of this type can't occur here. This brings fpsimd_flush_thread() into line with other code paths.
Cc <[email protected]> # 3.16.x
Cc <[email protected]> # 4.1.x-4.9.x
Fixes 674c242c9323 ("arm64: flush FP/SIMD state correctly after execve()")
Reviewed-by Ard Biesheuvel <[email protected]>
Signed-off-by Dave Martin <[email protected]>
Signed-off-by Will Deacon <[email protected]>

Quoted 6 lines

---
For stable only.
3.17.x-4.0.x don't appear active, and this patch isn't sufficient to fix them (they would need 674c242c9323 also).

3.18 is "semi-active" in that I'm trying to keep it alive for the millions of devices that rely on it out in the world at the moment...

So I've queued up both of these patches for that tree, don't know why 3.18 didn't catch it in the first place.

thanks,

greg k-h

Dave Martin

author

2017-09-01 16:16 UTC #2 lore raw

PATCH 1/1 arm64: fpsimd: Prevent registers leaking across exec

Quoted 27 lines · Greg KH

Quoted 26 lines · Dave Martin
commit 096622104e14d8a1db4860bd557717067a0515d2 upstream.
There are some tricky dependencies between the different stages of flushing the FPSIMD register state during exec, and these can race with context switch in ways that can cause the old task's regs to leak across. In particular, a context switch during the memset() can cause some of the task's old FPSIMD registers to reappear.
Disabling preemption for this small window would be no big deal for
performance: preemption is already disabled for similar scenarios like updating the FPSIMD registers in sigreturn.
So, instead of rearranging things in ways that might swap existing subtle bugs for new ones, this patch just disables preemption around the FPSIMD state flushing so that races of this type can't occur here. This brings fpsimd_flush_thread() into line with other code paths.
Cc <[email protected]> # 3.16.x
Cc <[email protected]> # 4.1.x-4.9.x
Fixes 674c242c9323 ("arm64: flush FP/SIMD state correctly after execve()")
Reviewed-by Ard Biesheuvel <[email protected]>
Signed-off-by Dave Martin <[email protected]>
Signed-off-by Will Deacon <[email protected]>
---

Quoted 7 lines

Quoted 4 lines
For stable only.
3.17.x-4.0.x don't appear active, and this patch isn't sufficient to fix them (they would need 674c242c9323 also).
3.18 is "semi-active" in that I'm trying to keep it alive for the millions of devices that rely on it out in the world at the moment...

Fair enough, I'll bear that in mind for future fixes.

Any ideas how long it's likely to linger for?

Quoted 1 line

So I've queued up both of these patches for that tree, don't know why

Thanks.

Quoted 1 line

3.18 didn't catch it in the first place.

The commit referenced by the Fixes tag already wasn't in 3.18.x, though that doesn't answer why that patch wasn't picked to 3.18.x in the first place. Dunno.

Cheers
---Dave

Greg KH

2017-09-01 16:27 UTC #3 lore raw

PATCH 1/1 arm64: fpsimd: Prevent registers leaking across exec

Quoted 33 lines · Dave Martin

Quoted 32 lines · Greg KH
Quoted 30 lines · Dave Martin
commit 096622104e14d8a1db4860bd557717067a0515d2 upstream.
There are some tricky dependencies between the different stages of flushing the FPSIMD register state during exec, and these can race with context switch in ways that can cause the old task's regs to leak across. In particular, a context switch during the memset() can cause some of the task's old FPSIMD registers to reappear.
Disabling preemption for this small window would be no big deal for
performance: preemption is already disabled for similar scenarios like updating the FPSIMD registers in sigreturn.
So, instead of rearranging things in ways that might swap existing subtle bugs for new ones, this patch just disables preemption around the FPSIMD state flushing so that races of this type can't occur here. This brings fpsimd_flush_thread() into line with other code paths.
Cc <[email protected]> # 3.16.x
Cc <[email protected]> # 4.1.x-4.9.x
Fixes 674c242c9323 ("arm64: flush FP/SIMD state correctly after execve()")
Reviewed-by Ard Biesheuvel <[email protected]>
Signed-off-by Dave Martin <[email protected]>
Signed-off-by Will Deacon <[email protected]>
---
For stable only.
3.17.x-4.0.x don't appear active, and this patch isn't sufficient to fix them (they would need 674c242c9323 also).

Quoted 6 lines

Quoted 2 lines
3.18 is "semi-active" in that I'm trying to keep it alive for the millions of devices that rely on it out in the world at the moment...
Fair enough, I'll bear that in mind for future fixes.
Any ideas how long it's likely to linger for?

No idea, probably until I get really sick of it. If you, or any company you know, relies on it, please let me know and I'll be glad to talk about it with you...

thanks,

greg k-h