lkml.org 
[lkml]   [2005]   [Apr]   [7]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
Date
From
SubjectRe: crash in entry.S restore_all, 2.6.12-rc2, x86, PAGEALLOC
Hello.

Ingo Molnar wrote:
> now if an interrupt hits at this point, it will set up a 'same privilege
> level' stackframe, which has eip/xcs/eflags, i.e. no esp/xss.
Yes, that's something I tried to say
when talking about the interrupt gates
(sorry if I wasn't clear).

> If upon
> irq-return we then examine the stack due to your patch, it will be an
> incorrect stackframe -> kaboom.
Yes, and that's where I think my patch is
at fault, i.e. it just shouldn't do this.
Another option is to make it always possible
to access OLDSS(%esp), but I think it is just
my patch have to be fixed to not do this at all.

> your patch doesnt remove the condition, it only removes the crash,
No, that wasn't my point at all. That example
with moving "sti" was *only* to answer Linus's
question of where we have an empty stack.
And I guess I wasn't clear enough also here,
I was in a hurry :(
The real patch I meant, is this one:
http://www.uwsg.iu.edu/hypermail/linux/kernel/0504.0/1287.html
This, I am sure, fixes a real bug. But there
can be the other approaches too.

> because it adds the 2 words space that is needed - but the information
> relied on by your irq-return test is still bogus.
But as an example for demonstrating the problem,
I thought, it could do:)

> At this point i'd
> suggest to remove the ESP patch altogether.
That's probably too heavy-handed. The fix is
really simple, we can either store the right
values by hands (as you propose), or fix my
patch (as I propose).

> the correct solution is to always let the sysenter path set up a full
> and correct stackframe,
But what will this solve? If I understand you
correctly, you will push the %ss/%esp of the
user-space process that did sysenter. Then
you enable the interrupts and get pre-empted.
Now what we have: OLDSS/OLDESP are of the
user-space process, but the EFLAGS/CS/EIP
are of the kernel (where it got pre-empted
on a sysenter path). This will avoid the crash,
but the information on stack is still wrong.
Or am I missing something?

> before allowing preemption (see the attached
> patch).
Hmm, will it work also for NMIs? You move
the sti, you can't be pre-empted, but the
NMI uses the restore_all too, no?
Also, it seems that Linus wants only the
*some* values available on stack, just to
make it not to crash. I think we can simply
adjust the tss.esp0 to point 8 bytes below
the real stack top, and so we are always safe.

-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/

\
 
 \ /
  Last update: 2005-04-07 18:15    [W:0.061 / U:2.872 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site