[PATCH] x86, PAT: honor CONFIG_STRICT_DEVMEM if pat is disable]

[PATCH] x86, PAT: honor CONFIG_STRICT_DEVMEM if pat is disable]

[Michal Hocko]

Hi,
I have just come across a strange behavior of /dev/[k]mem when PAT is
configured while STRICT_DEVMEM is disabled. 
One would expect that /dev/kmem would allow to access also the
system RAM in that configuration but that is not obviously true as pat
code defines range_is_allowed to protect from accessing that memory.

AFAICS this behavior was introduced in 0124cecf (x86, PAT: disable
/dev/mem mmap RAM with PAT) which says that it disables [k]mem with PAT
because it is safer. There is no explanation why it allows to access
that memory if CONFIG_NONPROMISC_DEVMEM (CONFIG_STRICT_DEVMEM now).

The thing is even more complicated by the fact that the access is
allowed when nopat kernel parameter is specified because
range_is_allowed just does't call devmem_is_allowed in that case.

While I do agree that the feature is not safe in general we should honor
STRICT_DEVMEM setting in some way IMO.

What do you think about the following fix? I have tried to preserve
"disabled for PAT" by default behavior.
---
>From e905f034d860c52de306b410ef148d601eae9e7f Mon Sep 17 00:00:00 2001
From: Michal Hocko <mhocko@suse.cz>
Date: Fri, 22 Jul 2011 11:01:01 +0200
Subject: [PATCH] x86, PAT: honor CONFIG_STRICT_DEVMEM if pat is disable

since 0124cecf (x86, PAT: disable /dev/mem mmap RAM with PAT) we are
disabling access to the system RAM as if STRICT_DEVMEM was enabled by
default. If we, however, disable pat by nopat kernel parameter we can
access that memory without considering CONFIG_STRICT_DEVMEM.

Let's make the code to honor STRICT_DEVMEM while we still preserve
disabled by default behavior. This means that the access to the system
memory is granted only if nopat is provided and STRICT_DEVMEM is
disabled.

Signed-off-by: Michal Hocko <mhocko@suse.cz>
---
 arch/x86/mm/pat.c |   10 ++--------
 1 files changed, 2 insertions(+), 8 deletions(-)

diff --git a/arch/x86/mm/pat.c b/arch/x86/mm/pat.c
index f6ff57b..fbaf3fc 100644
--- a/arch/x86/mm/pat.c
+++ b/arch/x86/mm/pat.c
@@ -472,13 +472,6 @@ pgprot_t phys_mem_access_prot(struct file *file, unsigned long pfn,
 	return vma_prot;
 }
 
-#ifdef CONFIG_STRICT_DEVMEM
-/* This check is done in drivers/char/mem.c in case of STRICT_DEVMEM*/
-static inline int range_is_allowed(unsigned long pfn, unsigned long size)
-{
-	return 1;
-}
-#else
 /* This check is needed to avoid cache aliasing when PAT is enabled */
 static inline int range_is_allowed(unsigned long pfn, unsigned long size)
 {
@@ -486,8 +479,10 @@ static inline int range_is_allowed(unsigned long pfn, unsigned long size)
 	u64 to = from + size;
 	u64 cursor = from;
 
+#ifndef CONFIG_STRICT_DEVMEM
 	if (!pat_enabled)
 		return 1;
+#endif
 
 	while (cursor < to) {
 		if (!devmem_is_allowed(pfn)) {
@@ -501,7 +496,6 @@ static inline int range_is_allowed(unsigned long pfn, unsigned long size)
 	}
 	return 1;
 }
-#endif /* CONFIG_STRICT_DEVMEM */
 
 int phys_mem_access_prot_allowed(struct file *file, unsigned long pfn,
 				unsigned long size, pgprot_t *vma_prot)
-- 
1.7.5.4

-- 
Michal Hocko
SUSE Labs
SUSE LINUX s.r.o.
Lihovarska 1060/12
190 00 Praha 9    
Czech Republic

Re: [PATCH] x86, PAT: honor CONFIG_STRICT_DEVMEM if pat is disable]

[H. Peter Anvin]

On 07/22/2011 02:11 AM, Michal Hocko wrote:
> Hi,
> I have just come across a strange behavior of /dev/[k]mem when PAT is
> configured while STRICT_DEVMEM is disabled. 
> One would expect that /dev/kmem would allow to access also the
> system RAM in that configuration but that is not obviously true as pat
> code defines range_is_allowed to protect from accessing that memory.
> 
> AFAICS this behavior was introduced in 0124cecf (x86, PAT: disable
> /dev/mem mmap RAM with PAT) which says that it disables [k]mem with PAT
> because it is safer. There is no explanation why it allows to access
> that memory if CONFIG_NONPROMISC_DEVMEM (CONFIG_STRICT_DEVMEM now).
> 
> The thing is even more complicated by the fact that the access is
> allowed when nopat kernel parameter is specified because
> range_is_allowed just does't call devmem_is_allowed in that case.
> 
> While I do agree that the feature is not safe in general we should honor
> STRICT_DEVMEM setting in some way IMO.
> 
> What do you think about the following fix? I have tried to preserve
> "disabled for PAT" by default behavior.

The reason it is disabled for PAT is that it is very hard to track maps
of that memory that are created by mapping /dev/[k]mem, since those maps
don't have a defined PAT type and really should be transparently
tracking the consensus caching type; this is a facility that *could* be
created but has no other user.

	-hpa

-- 
H. Peter Anvin, Intel Open Source Technology Center
I work for Intel.  I don't speak on their behalf.

Re: [PATCH] x86, PAT: honor CONFIG_STRICT_DEVMEM if pat is disable]

[Michal Hocko]

On Fri 22-07-11 09:15:08, H. Peter Anvin wrote:
> On 07/22/2011 02:11 AM, Michal Hocko wrote:
> > Hi,
> > I have just come across a strange behavior of /dev/[k]mem when PAT is
> > configured while STRICT_DEVMEM is disabled. 
> > One would expect that /dev/kmem would allow to access also the
> > system RAM in that configuration but that is not obviously true as pat
> > code defines range_is_allowed to protect from accessing that memory.
> > 
> > AFAICS this behavior was introduced in 0124cecf (x86, PAT: disable
> > /dev/mem mmap RAM with PAT) which says that it disables [k]mem with PAT
> > because it is safer. There is no explanation why it allows to access
> > that memory if CONFIG_NONPROMISC_DEVMEM (CONFIG_STRICT_DEVMEM now).
> > 
> > The thing is even more complicated by the fact that the access is
> > allowed when nopat kernel parameter is specified because
> > range_is_allowed just does't call devmem_is_allowed in that case.
> > 
> > While I do agree that the feature is not safe in general we should honor
> > STRICT_DEVMEM setting in some way IMO.
> > 
> > What do you think about the following fix? I have tried to preserve
> > "disabled for PAT" by default behavior.
> 
> The reason it is disabled for PAT is that it is very hard to track maps
> of that memory that are created by mapping /dev/[k]mem, since those maps
> don't have a defined PAT type and really should be transparently
> tracking the consensus caching type; this is a facility that *could* be
> created but has no other user.

Thanks for the clarification!
What do you think about fixing the nopat with STRICT_DEVMEM like the
patch does?

-- 
Michal Hocko
SUSE Labs
SUSE LINUX s.r.o.
Lihovarska 1060/12
190 00 Praha 9    
Czech Republic

Re: [PATCH] x86, PAT: honor CONFIG_STRICT_DEVMEM if pat is disable]

[H. Peter Anvin]

On 07/22/2011 10:31 AM, Michal Hocko wrote:
> 
> Thanks for the clarification!
> What do you think about fixing the nopat with STRICT_DEVMEM like the
> patch does?
> 

>From what I gather from your patch, the issue is that STRICT_DEVMEM
isn't being honored with nopat?  This is an obvious bug and should be fixed.

	-hpa

Re: [PATCH] x86, PAT: honor CONFIG_STRICT_DEVMEM if pat is disable]

[Michal Hocko]

On Fri 22-07-11 11:30:55, H. Peter Anvin wrote:
> On 07/22/2011 10:31 AM, Michal Hocko wrote:
> > 
> > Thanks for the clarification!
> > What do you think about fixing the nopat with STRICT_DEVMEM like the
> > patch does?
> > 
> 
> From what I gather from your patch, the issue is that STRICT_DEVMEM
> isn't being honored with nopat?

Yes.

> This is an obvious bug and should be fixed.

OK, thanks for confirmation. Should I just repost the patch without the
question about PAT vs. /dev/[k]mem or you will just pick it up from
there? Or would you preffer a different fix?

Thanks
-- 
Michal Hocko
SUSE Labs
SUSE LINUX s.r.o.
Lihovarska 1060/12
190 00 Praha 9    
Czech Republic