Heads up: dell-wmi bug

Heads up: dell-wmi bug

[Andy Lutomirski]

Hi-

I'm getting intermittent OOPSes when playing with dell-wmi.  I think
that find_hk_type function is buggy: it's saving a pointer to DMI
data, but you can't do that, because dmi_walk unmaps the data at the
end.  It probably works reliably on current kernels due to luck
(either a memory layout such that the unmap has no effect or deferred
vmap zapping).

If no one beats me to it, I'll fix this as part of the rebase work on
my big wmi series.

--Andy

Re: Heads up: dell-wmi bug

[Pali Rohár]

[-- Attachment #1: Type: Text/Plain, Size: 872 bytes --]

On Thursday 31 December 2015 13:31:03 Andy Lutomirski wrote:
> Hi-
> 
> I'm getting intermittent OOPSes when playing with dell-wmi.  I think
> that find_hk_type function is buggy: it's saving a pointer to DMI
> data, but you can't do that, because dmi_walk unmaps the data at the
> end.  It probably works reliably on current kernels due to luck
> (either a memory layout such that the unmap has no effect or deferred
> vmap zapping).
> 
> If no one beats me to it, I'll fix this as part of the rebase work on
> my big wmi series.
> 
> --Andy

Hi! If this is truth, then I think separate patch for fixing this 
problem should be created and backported also to stable kernels.

In my opinion WMI rework patches should not change functionality, just 
API... Other function changes should be in separate commits.

-- 
Pali Rohár
pali.rohar@gmail.com

[-- Attachment #2: This is a digitally signed message part. --]
[-- Type: application/pgp-signature, Size: 198 bytes --]

Re: Heads up: dell-wmi bug

[Andy Lutomirski]

On Dec 31, 2015 8:34 PM, "Pali Rohár" <pali.rohar@gmail.com> wrote:
>
> On Thursday 31 December 2015 13:31:03 Andy Lutomirski wrote:
> > Hi-
> >
> > I'm getting intermittent OOPSes when playing with dell-wmi.  I think
> > that find_hk_type function is buggy: it's saving a pointer to DMI
> > data, but you can't do that, because dmi_walk unmaps the data at the
> > end.  It probably works reliably on current kernels due to luck
> > (either a memory layout such that the unmap has no effect or deferred
> > vmap zapping).
> >
> > If no one beats me to it, I'll fix this as part of the rebase work on
> > my big wmi series.
> >
> > --Andy
>
> Hi! If this is truth, then I think separate patch for fixing this
> problem should be created and backported also to stable kernels.
>
> In my opinion WMI rework patches should not change functionality, just
> API... Other function changes should be in separate commits.

Agreed.  I just meant that I'd do it at the same time, not that it
would be the same patch.

>
> --
> Pali Rohár
> pali.rohar@gmail.com