Skip to contentSkip to conversation

Closing on the CR2 leak bug

DISCUSSION H. Peter Anvin 2 participants 4 messages Mar 4, 2014 – Mar 5, 2014

lore
Share & tools
lore
#0 lore raw

So we need to get something into x86/urgent for the CR2 bug.

It seems like a no-brainer to do the hoisting patch, for which I prefer the version proposed by Jiri Olsa which reads %cr2 and then passes it to __do_page_fault() in a GPR:

http://lkml.kernel.org/r/[email protected]

As fart as I can tell this should fix Vince's problem as well.

Anyone who objects to this? Otherwise I will put it in tip:x86/urgent tomorrow.

Do we need any more refinements this late in the -rc cycle?

-hpa

#1 lore raw
Quoted 6 lines · On Tue, 04 Mar 2014 14:34:00 -0800 "H. Peter Anvin" <[email protected]> wrote:

So we need to get something into x86/urgent for the CR2 bug.

It seems like a no-brainer to do the hoisting patch, for which I prefer the version proposed by Jiri Olsa which reads %cr2 and then passes it to __do_page_fault() in a GPR:

Quoted 9 lines

http://lkml.kernel.org/r/[email protected]

As fart as I can tell this should fix Vince's problem as well.

Anyone who objects to this? Otherwise I will put it in tip:x86/urgent tomorrow.

Do we need any more refinements this late in the -rc cycle?

The only other issue is if perf traces from function context and traces the trace_do_page_fault() call. But other than that, sure, take Jiri's patch.

Please also add the reported/tested-by from Vince Weaver and add a
Link: to the other thread, as it got pretty detailed there too.

-- Steve

#2 lore raw
Quoted 5 lines · Steven Rostedt

The only other issue is if perf traces from function context and traces the trace_do_page_fault() call. But other than that, sure, take Jiri's patch.

Is there a known codepath on which that can happen?

-hpa

#3 lore raw
Quoted 8 lines · On Tue, 04 Mar 2014 15:44:26 -0800 "H. Peter Anvin" <[email protected]> wrote:
Quoted 5 lines · Steven Rostedt

The only other issue is if perf traces from function context and traces the trace_do_page_fault() call. But other than that, sure, take Jiri's patch.

Is there a known codepath on which that can happen?

I'm not sure. Jiri, is there a way we can trace userspace from the ftrace function event? What would be the command to try? I know you had a patch to prevent it, but is it currently possible?

-- Steve