lkml.org 
[lkml]   [2015]   [Dec]   [3]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
Date
From
SubjectRe: Domain faults when CONFIG_CPU_SW_DOMAIN_PAN is enabled
On Thu, Dec 03, 2015 at 12:08:11PM +0000, Peter Rosin wrote:
> Russell King wrote:
> > On Thu, Dec 03, 2015 at 11:38:20AM +0000, Peter Rosin wrote:
> > > Russell King wrote:
> > > > On Thu, Dec 03, 2015 at 08:33:13AM +0000, Peter Rosin wrote:
> > > > > I wrote:
> > > > > > If I enable CONFIG_CPU_SW_DOMAIN_PAN, I sometimes (but not always) get the
> > > > > > following (or very similar) on boot.
> > > > >
> > > > > I should have said "if I don't disable", as the option is "default y".
> > > > >
> > > > > Also, if it survives on boot, below is an example of later trouble (after
> > > > > 30+ minutes on this occasion).
> > > >
> > > > Please apply this patch so we (might) get a slightly better oops dump,
> > > > and then try to reproduce.
> > >
> > > Sure thing, but it's still "DAC: 00000051"...
> >
> > Thanks, that confirms that something in the uaccess-with-memcpy code
> > is clearing the DACR back to 0x51.
> >
> > This has come up several times before, and I really can't spot the
> > problem in this code, so I've always said to disable the
> > uaccess-with-memcpy code. Personally, I'd like to see the back
> > of that code...
>
> Ok, but it's not me doing crazy things if that's what you are implying
> (not saying that you are), because the sama5_defconfig has
>
> CONFIG_UACCESS_WITH_MEMCPY=y
>
> So, something needs to happen or sama5 remains default-broken.

I have no solution for this, other than saying that uaccess-with-memcpy
seems to be (for some unknown reason) incompatible with SW PAN.

Out of the two features, I'd go for SW PAN over uaccess-with-memcpy,
but others will have a different opinion. The real solution is to
track down what's going on and why, but I don't think anyone has the
motivation to do that.

I've looked into this problem, and I've been unable to identify what's
going on here. I can see no reason for the DACR to be set to 0x51 in
this code.

The entry path into this code is via __copy_to_user(), which saves and
sets the DACR to 0x55 before calling arm_copy_to_user(), which for
uaccess-with-memcpy() is in arch/arm/lib/uaccess_with_memcpy.c. That
tail-calls into __copy_to_user_memcpy(), which should run with the
DACR set to 0x55.

The only place that the DACR is changed is inside __put_user(), which
saves the DACR before also setting it to 0x55, and then restores the
old DACR value, which, because the old value was 0x55, will have no
effect.

So, I can see no way that the DACR should ever be 0x51 inside
__copy_to_user_memcpy(), but you are seeing such a scenario. I've no
idea how you could ever get a value of 0x51 out of the DACR within this
code.

--
FTTC broadband for 0.8mile line: currently at 9.6Mbps down 400kbps up
according to speedtest.net.


\
 
 \ /
  Last update: 2015-12-03 15:01    [W:0.103 / U:2.668 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site