REGRESSION: 2.6.26-rc2-git4: X server failed start on X61s laptop

REGRESSION: 2.6.26-rc2-git4: X server failed start on X61s laptop

[Theodore Ts'o]

The X server failed to come up using the Intel driver, and forced a
fallback to the 800x600 VESA server.  This was on my Lenovo X61s laptop
with an Intel chipset, running Ubuntu Gutsy.  When it failed, there were
a large number of these errors in the /var/log/messages:

mtrr: type mismatch for e0760000,10000 old: write-back new: write-combining
mtrr: type mismatch for e0740000,20000 old: write-back new: write-combining
mtrr: type mismatch for e0700000,40000 old: write-back new: write-combining
mtrr: type mismatch for e0600000,100000 old: write-back new: write-combining
mtrr: type mismatch for e0400000,200000 old: write-back new: write-combining
mtrr: type mismatch for e0000000,400000 old: write-back new: write-combining
      ...

On kernels where this error did not occur during the bisect, there would
still be one of these messages:

mtrr: type mismatch for e0000000,10000000 old: write-back new: write-combining

... but the X server would correctly start.

The git bisect identified this commit as the guilty one:

1c12c4cf9411eb130b245fa8d0fbbaf989477c7b is first bad commit
commit 1c12c4cf9411eb130b245fa8d0fbbaf989477c7b
Author: Venki Pallipadi <venkatesh.pallipadi@intel.com>
Date:   Wed May 14 16:05:51 2008 -0700

    mprotect: prevent alteration of the PAT bits
    
    There is a defect in mprotect, which lets the user change the page cache
    type bits by-passing the kernel reserve_memtype and free_memtype
    wrappers.  Fix the problem by not letting mprotect change the PAT bits.
    
    Signed-off-by: Venkatesh Pallipadi <venkatesh.pallipadi@intel.com>
    Signed-off-by: Suresh Siddha <suresh.b.siddha@intel.com>
    Signed-off-by: Ingo Molnar <mingo@elte.hu>
    Signed-off-by: Hugh Dickins <hugh@veritas.com>
    Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
    Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

:040000 040000 db42a4a31f711306205648690e97cbead50183ac 3f835886fcb56fc6e732ee6f985304c4c04b93d9 M      include
:040000 040000 b3c046a2f1cdfa6a3ec4b46f89cbee6f2ab5018d 07a87010634f455313e1e7f7e3f7d3a94b9bfed3 M      mm

And the bisect log follows here:

git-bisect start
# bad: [079d04dfa5198e3d43c6db2dcfa400f7fbbd516f] ext4: fiemap implementation Here is ext4_fiemap() itself.  This still needs a bit of testing & work, but is correct for most file layouts.... I still hit occasional problems with interesting mappings such as sparse/preallocated/etc.
git-bisect bad 079d04dfa5198e3d43c6db2dcfa400f7fbbd516f
# good: [e7f379d5cabb2790ecce5d623382fa6085e7686d] alim15x3: remove WDC_ALI15X3 config option
git-bisect good e7f379d5cabb2790ecce5d623382fa6085e7686d
# good: [9dcbf35afb7359466efdf7fb81ee32f3ae2d56a3] V4L/DVB (7887): cx18: fix Compro H900 analog support.
git-bisect good 9dcbf35afb7359466efdf7fb81ee32f3ae2d56a3
# bad: [8f40f672e6bb071812f61bfbd30efc3fc1263ad1] Merge branch 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/ericvh/v9fs
git-bisect bad 8f40f672e6bb071812f61bfbd30efc3fc1263ad1
# good: [90898709dfca860d9550c85f0924007f4c0467ea] atmel_lcdfb: fix initialization of a pre-allocated framebuffer
git-bisect good 90898709dfca860d9550c85f0924007f4c0467ea
# bad: [9ffee4cbc51907755809d98613d9e7133612803a] tty_check_change(): avoid taking tasklist_lock while holding tty->ctrl_lock
git-bisect bad 9ffee4cbc51907755809d98613d9e7133612803a
# good: [122a881c776b7c155bf3f379928cc27aab435288] video/logo: add support for Blackfin/Linux logo for framebuffer console
git-bisect good 122a881c776b7c155bf3f379928cc27aab435288
# bad: [1c12c4cf9411eb130b245fa8d0fbbaf989477c7b] mprotect: prevent alteration of the PAT bits
git-bisect bad 1c12c4cf9411eb130b245fa8d0fbbaf989477c7b
# good: [fd8a4221ad76df700ff34875c9fbc42302aa4ba3] memory_hotplug: check for walk_memory_resource() failure in online_pages()
git-bisect good fd8a4221ad76df700ff34875c9fbc42302aa4ba3
# good: [44c81433e8b05dbc85985d939046f10f95901184] per_cpu: fix DEFINE_PER_CPU_SHARED_ALIGNED for modules
git-bisect good 44c81433e8b05dbc85985d939046f10f95901184

Re: REGRESSION: 2.6.26-rc2-git4: X server failed start on X61s laptop

[Sitsofe Wheeler]

On Sat, 17 May 2008 03:32:05 -0400, Theodore Ts'o wrote:

> The X server failed to come up using the Intel driver, and forced a
> fallback to the 800x600 VESA server.  This was on my Lenovo X61s laptop
>
> The git bisect identified this commit as the guilty one:
> 
> 1c12c4cf9411eb130b245fa8d0fbbaf989477c7b is first bad commit commit
> 1c12c4cf9411eb130b245fa8d0fbbaf989477c7b Author: Venki Pallipadi
> <venkatesh.pallipadi@intel.com> Date:   Wed May 14 16:05:51 2008 -0700
> 
>     mprotect: prevent alteration of the PAT bits

This sounds related to
http://groups.google.com/group/linux.kernel/browse_frm/thread/1dba81049ce4ad16/546aff3e91055ab7#546aff3e91055ab7 ...

-- 
Sitsofe | http://sucs.org/~sits/

RE: REGRESSION: 2.6.26-rc2-git4: X server failed start on X61s laptop

[Pallipadi, Venkatesh]

 

>-----Original Message-----
>From: Theodore Ts'o [mailto:tytso@mit.edu] 
>Sent: Saturday, May 17, 2008 12:32 AM
>To: linux-kernel@vger.kernel.org
>Cc: Pallipadi, Venkatesh
>Subject: REGRESSION: 2.6.26-rc2-git4: X server failed start on 
>X61s laptop
>
>
>The X server failed to come up using the Intel driver, and forced a
>fallback to the 800x600 VESA server.  This was on my Lenovo X61s laptop
>with an Intel chipset, running Ubuntu Gutsy.  When it failed, 
>there were
>a large number of these errors in the /var/log/messages:
>
>mtrr: type mismatch for e0760000,10000 old: write-back new: 
>write-combining
>mtrr: type mismatch for e0740000,20000 old: write-back new: 
>write-combining
>mtrr: type mismatch for e0700000,40000 old: write-back new: 
>write-combining
>mtrr: type mismatch for e0600000,100000 old: write-back new: 
>write-combining
>mtrr: type mismatch for e0400000,200000 old: write-back new: 
>write-combining
>mtrr: type mismatch for e0000000,400000 old: write-back new: 
>write-combining
>      ...
>
>On kernels where this error did not occur during the bisect, 
>there would
>still be one of these messages:
>
>mtrr: type mismatch for e0000000,10000000 old: write-back new: 
>write-combining
>
>... but the X server would correctly start.
>
>The git bisect identified this commit as the guilty one:
>
>1c12c4cf9411eb130b245fa8d0fbbaf989477c7b is first bad commit
>commit 1c12c4cf9411eb130b245fa8d0fbbaf989477c7b
>Author: Venki Pallipadi <venkatesh.pallipadi@intel.com>
>Date:   Wed May 14 16:05:51 2008 -0700
>
>    mprotect: prevent alteration of the PAT bits
>    
>    There is a defect in mprotect, which lets the user change 
>the page cache
>    type bits by-passing the kernel reserve_memtype and free_memtype
>    wrappers.  Fix the problem by not letting mprotect change 
>the PAT bits.
>    


Can you please send the complete dmesg after X failure, with patch here
and debugpat boot option.
http://www.ussg.iu.edu/hypermail/linux/kernel/0805.0/2657.html

Thanks,
Venki

[Bug 10732] REGRESSION: 2.6.26-rc2-git4: X server failed start on X61s laptop

[Theodore Tso]

On Sat, May 17, 2008 at 06:21:18AM -0700, Pallipadi, Venkatesh wrote:
> 
> Can you please send the complete dmesg after X failure, with patch here
> and debugpat boot option.
> http://www.ussg.iu.edu/hypermail/linux/kernel/0805.0/2657.html

With debugpat enabled, the resulting dump that landed in
/var/log/messages was something like 1.6 megabytes, so it totally
overflowed the dmesg buffer.

I've created a Bugzilla attachment of the bzip'ed messages excerpt here:

http://bugzilla.kernel.org/attachment.cgi?id=16174&action=view

The patch above was only available in HTML-only mode (and I couldn't
find Ingo's patch on lkml.org, so I had to reconstruct it, and then
forward part it to 2.6.26-rc2-git5).  What I actually used can be
found here:

http://bugzilla.kernel.org/attachment.cgi?id=16173&action=view

I'm not sure this will be helpful for you, but this should be
completely trivial for you to reproduce; just use Ubuntu Gutsy on
Intel Video based laptop, with a 2.6.26-rc2-git4 or more recent
kernel, and watch it fail.  

For me, I'll just locally revert commit 1c12c4cf since otherwise X
becomes more-or-less unsuable.  (Getting used to 1024x768 after using
a 1600x1200 display is one thing; 800x600 is quite another, and the
performance using the VESA driver is quite abysmal.)

						- Ted

RE: [Bug 10732] REGRESSION: 2.6.26-rc2-git4: X server failed start onX61s laptop

[Pallipadi, Venkatesh]

 

>-----Original Message-----
>From: Theodore Tso [mailto:tytso@mit.edu] 
>Sent: Saturday, May 17, 2008 8:42 AM
>To: Pallipadi, Venkatesh
>Cc: linux-kernel@vger.kernel.org; Ingo Molnar; Siddha, Suresh 
>B; bugme-daemon@bugzilla.kernel.org
>Subject: [Bug 10732] REGRESSION: 2.6.26-rc2-git4: X server 
>failed start onX61s laptop
>
>
>On Sat, May 17, 2008 at 06:21:18AM -0700, Pallipadi, Venkatesh wrote:
>> 
>> Can you please send the complete dmesg after X failure, with 
>patch here
>> and debugpat boot option.
>> http://www.ussg.iu.edu/hypermail/linux/kernel/0805.0/2657.html
>
>With debugpat enabled, the resulting dump that landed in
>/var/log/messages was something like 1.6 megabytes, so it totally
>overflowed the dmesg buffer.
>
>I've created a Bugzilla attachment of the bzip'ed messages 
>excerpt here:
>
>http://bugzilla.kernel.org/attachment.cgi?id=16174&action=view
>
>The patch above was only available in HTML-only mode (and I couldn't
>find Ingo's patch on lkml.org, so I had to reconstruct it, and then
>forward part it to 2.6.26-rc2-git5).  What I actually used can be
>found here:
>
>http://bugzilla.kernel.org/attachment.cgi?id=16173&action=view
>
>I'm not sure this will be helpful for you, but this should be
>completely trivial for you to reproduce; just use Ubuntu Gutsy on
>Intel Video based laptop, with a 2.6.26-rc2-git4 or more recent
>kernel, and watch it fail.  
>
>For me, I'll just locally revert commit 1c12c4cf since otherwise X
>becomes more-or-less unsuable.  (Getting used to 1024x768 after using
>a 1600x1200 display is one thing; 800x600 is quite another, and the
>performance using the VESA driver is quite abysmal.)
>

Thanks for the log Ted. From first looks, I don't see any PAT related
failures
in the log. And as the revert of the commit 1c12c4cf is fixing the
problem here,
adding the X guys to the thread.

I am thinking that may be X is depending on mprotect changes somehow and
failing when
it cannot change PAT attributes with mprotect call. Keith earlier
mentioned that X
will not depend on this. May be something to do with earlier X version.
Keith?

Thanks,
Venki

Re: [Bug 10732] REGRESSION: 2.6.26-rc2-git4: X server failed start on X61s laptop

[Arjan van de Ven]

On Sat, 17 May 2008 11:41:31 -0400

> The patch above was only available in HTML-only mode

as a totally unrelated sidenote, you do know about the "dehtmldiff"
command in patchutils, right?

Re: [Bug 10732] REGRESSION: 2.6.26-rc2-git4: X server failed start onX61s laptop

[Theodore Tso]

On Sat, May 17, 2008 at 09:02:32AM -0700, Pallipadi, Venkatesh wrote:
> Thanks for the log Ted. From first looks, I don't see any PAT
> related failures in the log. 

I forgot to add that I had tried booting with the "nopat" option, and
that didn't change anything, so yeah, it doesn't seem likely to be
related.

						- Ted

RE: [Bug 10732] REGRESSION: 2.6.26-rc2-git4: X server failed start onX61s laptop

[Keith Packard]

[-- Attachment #1: Type: text/plain, Size: 734 bytes --]

On Sat, 2008-05-17 at 09:02 -0700, Pallipadi, Venkatesh wrote:

> I am thinking that may be X is depending on mprotect changes somehow and
> failing when
> it cannot change PAT attributes with mprotect call. Keith earlier
> mentioned that X
> will not depend on this. May be something to do with earlier X version.
> Keith?

It looks like we can turn off read/write access, but we can't turn it
back on. So,

        mprotect (map->memory, map->size, PROT_NONE);
        mprotect (map->memory, map->size, PROT_READ|PROT_WRITE);
 
leaves the memory with no access.

Eric saw this last week; I don't know whether he debugged into it
further.

Which repository is commit 1c12c4cf in?

-- 
keith.packard@intel.com

[-- Attachment #2: This is a digitally signed message part --]
[-- Type: application/pgp-signature, Size: 189 bytes --]

Re: [Bug 10732] REGRESSION: 2.6.26-rc2-git4: X server failed start onX61s laptop

[Gabriel C]

Keith Packard wrote:

> On Sat, 2008-05-17 at 09:02 -0700, Pallipadi, Venkatesh wrote:
> 
>> I am thinking that may be X is depending on mprotect changes somehow and
>> failing when
>> it cannot change PAT attributes with mprotect call. Keith earlier
>> mentioned that X
>> will not depend on this. May be something to do with earlier X version.
>> Keith?
> 
> It looks like we can turn off read/write access, but we can't turn it
> back on. So,
> 
>         mprotect (map->memory, map->size, PROT_NONE);
>         mprotect (map->memory, map->size, PROT_READ|PROT_WRITE);
>  
> leaves the memory with no access.
> 
> Eric saw this last week; I don't know whether he debugged into it
> further.
> 
> Which repository is commit 1c12c4cf in?


It is in linus-git tree , http://git.kernel.org/?p=linux/kernel/git/torvalds/linux-2.6.git;a=commit;h=1c12c4cf9411eb130b245fa8d0fbbaf989477c7b

Re: [Bug 10732] REGRESSION: 2.6.26-rc2-git4: X server failed start onX61s laptop

[Theodore Tso]

On Sat, May 17, 2008 at 08:32:38PM +0200, Gabriel C wrote:
> > 
> > Which repository is commit 1c12c4cf in?
> 
> 
> It is in linus-git tree , http://git.kernel.org/?p=linux/kernel/git/torvalds/linux-2.6.git;a=commit;h=1c12c4cf9411eb130b245fa8d0fbbaf989477c7b
> 

...post 2.6.26-rc2-git3.  For Linux kernels 2.6.26-rc2-git4 and
newer, you need to revert this commit or the X Server shipped with
Ubuntu Gutsy will die horribly when run on an X61s laptop with the
Intel 965GM video chipset.

So the question is what's the right fix to keep the kernel compatible
with the X server, besides just reverting the commit entirely?

					- Ted

[Bug #10732] REGRESSION: 2.6.26-rc2-git4: X server failed start on X61s laptop

[Rafael J. Wysocki]

This message has been generated automatically as a part of a report
of recent regressions.

The following bug entry is on the current list of known regressions
from 2.6.25.  Please verify if it still should be listed.


Bug-Entry	: http://bugzilla.kernel.org/show_bug.cgi?id=10732
Subject		: REGRESSION: 2.6.26-rc2-git4: X server failed start on X61s laptop
Submitter	: Theodore Ts'o <tytso@mit.edu>
Date		: 2008-05-17 7:32 (2 days old)
References	: http://marc.info/?l=linux-kernel&amp;m=121100994722524&amp;w=4

Re: [Bug 10732] REGRESSION: 2.6.26-rc2-git4: X server failed start onX61s laptop

[Hugh Dickins]

On Sat, 17 May 2008, Theodore Tso wrote:
> On Sat, May 17, 2008 at 08:32:38PM +0200, Gabriel C wrote:
> > > 
> > > Which repository is commit 1c12c4cf in?
> > 
> > It is in linus-git tree , http://git.kernel.org/?p=linux/kernel/git/torvalds/linux-2.6.git;a=commit;h=1c12c4cf9411eb130b245fa8d0fbbaf989477c7b
> 
> ...post 2.6.26-rc2-git3.  For Linux kernels 2.6.26-rc2-git4 and
> newer, you need to revert this commit or the X Server shipped with
> Ubuntu Gutsy will die horribly when run on an X61s laptop with the
> Intel 965GM video chipset.
> 
> So the question is what's the right fix to keep the kernel compatible
> with the X server, besides just reverting the commit entirely?

[PATCH] x86: fix mprotect's NX handling on PAE

2.6.26-rc3 with CONFIG_X86_PAE may leave the NX bit set when PROT_EXEC
is intending to clear it, causing irqbalance and others to segfault,
and X startup to fail.

This comes from an assumption in 1c12c4cf9411eb130b245fa8d0fbbaf989477c7b
mprotect: prevent alteration of the PAT bits, that PTE_MASK is what it's
supposed to be: whereas it's been wrong forever with PAE, staying 32-bit
where 64-bit is needed.

Jeremy Fitzhardinge already has patches to fix that in Ingo's tree;
but if they're not considered 2.6.26 material, or people want a quick
two-liner to get working, here's a shorter hack to fix up pte_modify.

I apologize to those we've broken, and to Venki Pallipadi: it was I who
persuaded him to change his working patch to make that false assumption.

Signed-off-by: Hugh Dickins <hugh@veritas.com>
---

 include/asm-x86/pgtable.h |    4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

--- 2.6.26-rc3/include/asm-x86/pgtable.h	2008-05-19 11:19:03.000000000 +0100
+++ linux/include/asm-x86/pgtable.h	2008-05-19 21:51:08.000000000 +0100
@@ -57,8 +57,8 @@
 #define _KERNPG_TABLE	(_PAGE_PRESENT | _PAGE_RW | _PAGE_ACCESSED |	\
 			 _PAGE_DIRTY)
 
-#define _PAGE_CHG_MASK	(PTE_MASK | _PAGE_PCD | _PAGE_PWT |		\
-			 _PAGE_ACCESSED | _PAGE_DIRTY)
+#define _PAGE_CHG_MASK	(((pteval_t) PTE_MASK & ~_PAGE_NX) | 		\
+			 _PAGE_PCD | _PAGE_PWT | _PAGE_ACCESSED | _PAGE_DIRTY)
 
 #define _PAGE_CACHE_MASK	(_PAGE_PCD | _PAGE_PWT)
 #define _PAGE_CACHE_WB		(0)

Re: [Bug 10732] REGRESSION: 2.6.26-rc2-git4: X server failed start onX61s laptop

[Jeremy Fitzhardinge]

Hugh Dickins wrote:
> On Sat, 17 May 2008, Theodore Tso wrote:
>   
>> On Sat, May 17, 2008 at 08:32:38PM +0200, Gabriel C wrote:
>>     
>>>> Which repository is commit 1c12c4cf in?
>>>>         
>>> It is in linus-git tree , http://git.kernel.org/?p=linux/kernel/git/torvalds/linux-2.6.git;a=commit;h=1c12c4cf9411eb130b245fa8d0fbbaf989477c7b
>>>       
>> ...post 2.6.26-rc2-git3.  For Linux kernels 2.6.26-rc2-git4 and
>> newer, you need to revert this commit or the X Server shipped with
>> Ubuntu Gutsy will die horribly when run on an X61s laptop with the
>> Intel 965GM video chipset.
>>
>> So the question is what's the right fix to keep the kernel compatible
>> with the X server, besides just reverting the commit entirely?
>>     
>
> [PATCH] x86: fix mprotect's NX handling on PAE
>
> 2.6.26-rc3 with CONFIG_X86_PAE may leave the NX bit set when PROT_EXEC
> is intending to clear it, causing irqbalance and others to segfault,
> and X startup to fail.
>
> This comes from an assumption in 1c12c4cf9411eb130b245fa8d0fbbaf989477c7b
> mprotect: prevent alteration of the PAT bits, that PTE_MASK is what it's
> supposed to be: whereas it's been wrong forever with PAE, staying 32-bit
> where 64-bit is needed.
>
> Jeremy Fitzhardinge already has patches to fix that in Ingo's tree;
> but if they're not considered 2.6.26 material, or people want a quick
> two-liner to get working, here's a shorter hack to fix up pte_modify.
>   

 From what I can tell, my patches haven't caused any regressions, and so 
if they fix a real bug we could consider them for .26.

    J

Re: [Bug 10732] REGRESSION: 2.6.26-rc2-git4: X server failed start onX61s laptop

[Linus Torvalds]

On Mon, 19 May 2008, Hugh Dickins wrote:
> 
> This comes from an assumption in 1c12c4cf9411eb130b245fa8d0fbbaf989477c7b
> mprotect: prevent alteration of the PAT bits, that PTE_MASK is what it's
> supposed to be: whereas it's been wrong forever with PAE, staying 32-bit
> where 64-bit is needed.

Can we *please* just fix PTE_MASK?

And can we agree to never EVER use that PAGE_MASK thing (which was only 
ever meant to work on *addresses*) for any pte operations (including the 
definition of PTE_MASK)? Because PAGE_MASK is very much the word-size, and 
in 32-bit PAE, the page table entry is bigger.

IOE, PTE_MASK should be a "pteval_t". And it should have absolutely 
*nothing* to do with PAGE_MASK. EVER.

IOW, maybe something like this?

And no, I haven't tested this at all. But it should make PTE_MASK have
 (a) the right type ("pteval_t", not "long" - the latter is pure and utter 
     crap)
 (b) the right value (proper mask, not a sign-extended long - again, the 
     latter is pure and utter crap)

but for all I know there might be some broken code that depends on the 
current incorrect and totally broken #defines, so this needs testing and 
thinking about.

It also causes these warnings on 32-bit PAE:

	  AS      arch/x86/kernel/head_32.o
	arch/x86/kernel/head_32.S: Assembler messages:
	arch/x86/kernel/head_32.S:225: Warning: left operand is a bignum; integer 0 assumed
	arch/x86/kernel/head_32.S:609: Warning: left operand is a bignum; integer 0 assumed

and I do not see why (the end result seems to be identical).

Ingo, comments?

Oh, and those #define's should be moved from <asm/page.h> to 
<asm/pgtable.h>, I think. They have nothing to do with pages (despite the 
name of "physical_page_mask", and really are meaningful only in the 
context of some kind of page table entry.

		Linus

---
 include/asm-x86/page.h |    5 +++--
 1 files changed, 3 insertions(+), 2 deletions(-)

diff --git a/include/asm-x86/page.h b/include/asm-x86/page.h
index b381f4a..34b4845 100644
--- a/include/asm-x86/page.h
+++ b/include/asm-x86/page.h
@@ -10,8 +10,8 @@
 
 #ifdef __KERNEL__
 
-#define PHYSICAL_PAGE_MASK	(PAGE_MASK & __PHYSICAL_MASK)
-#define PTE_MASK		(_AT(long, PHYSICAL_PAGE_MASK))
+#define PHYSICAL_PAGE_MASK	(__PHYSICAL_MASK & ~__PHYSICAL_LOW_BITS)
+#define PTE_MASK		(_AT(pteval_t, PHYSICAL_PAGE_MASK))
 
 #define PMD_PAGE_SIZE		(_AC(1, UL) << PMD_SHIFT)
 #define PMD_PAGE_MASK		(~(PMD_PAGE_SIZE-1))
@@ -24,6 +24,7 @@
 /* to align the pointer to the (next) page boundary */
 #define PAGE_ALIGN(addr)	(((addr)+PAGE_SIZE-1)&PAGE_MASK)
 
+#define __PHYSICAL_LOW_BITS	_AT(phys_addr_t, (PAGE_SIZE-1))
 #define __PHYSICAL_MASK		_AT(phys_addr_t, (_AC(1,ULL) << __PHYSICAL_MASK_SHIFT) - 1)
 #define __VIRTUAL_MASK		((_AC(1,UL) << __VIRTUAL_MASK_SHIFT) - 1)
 

Re: [Bug 10732] REGRESSION: 2.6.26-rc2-git4: X server failed start onX61s laptop

[Linus Torvalds]

On Mon, 19 May 2008, Linus Torvalds wrote:
> 
> IOE, PTE_MASK should be a "pteval_t". And it should have absolutely 
> *nothing* to do with PAGE_MASK. EVER.
> 
> IOW, maybe something like this?

.. and in addition to that, perhaps something like this too?

Making the actual _PAGE_XYZ bits be of the proper type makes using bitops 
and negation on them automatically DTRT, so that we don't need those 
insane casts in pte_mkclean() and friends.

Again, totally untested, but we really should just get the types right, 
instead of getting them wrogn and then messing around with various silly 
and incorrect work-arounds for the fact that we got them wrong.

And while it's untested - if this cleanup is buggy (or even generates 
different code), then we're doing something really wrong somewhere.

		Linus

---
 include/asm-x86/pgtable.h |   49 +++++++++++++++++++++++----------------------
 1 files changed, 25 insertions(+), 24 deletions(-)

diff --git a/include/asm-x86/pgtable.h b/include/asm-x86/pgtable.h
index 55c3a0e..b38b540 100644
--- a/include/asm-x86/pgtable.h
+++ b/include/asm-x86/pgtable.h
@@ -21,27 +21,28 @@
 #define _PAGE_BIT_NX           63       /* No execute: only valid after cpuid check */
 
 /*
- * Note: we use _AC(1, L) instead of _AC(1, UL) so that we get a
- * sign-extended value on 32-bit with all 1's in the upper word,
- * which preserves the upper pte values on 64-bit ptes:
+ * Note: we use _AT(pteval_t,1) so that we get the right type
+ * even when we're using PAE and everything is 64-bit with a
+ " 32-bit word-size. That's important for various bit masks.
  */
-#define _PAGE_PRESENT	(_AC(1, L)<<_PAGE_BIT_PRESENT)
-#define _PAGE_RW	(_AC(1, L)<<_PAGE_BIT_RW)
-#define _PAGE_USER	(_AC(1, L)<<_PAGE_BIT_USER)
-#define _PAGE_PWT	(_AC(1, L)<<_PAGE_BIT_PWT)
-#define _PAGE_PCD	(_AC(1, L)<<_PAGE_BIT_PCD)
-#define _PAGE_ACCESSED	(_AC(1, L)<<_PAGE_BIT_ACCESSED)
-#define _PAGE_DIRTY	(_AC(1, L)<<_PAGE_BIT_DIRTY)
-#define _PAGE_PSE	(_AC(1, L)<<_PAGE_BIT_PSE)	/* 2MB page */
-#define _PAGE_GLOBAL	(_AC(1, L)<<_PAGE_BIT_GLOBAL)	/* Global TLB entry */
-#define _PAGE_UNUSED1	(_AC(1, L)<<_PAGE_BIT_UNUSED1)
-#define _PAGE_UNUSED2	(_AC(1, L)<<_PAGE_BIT_UNUSED2)
-#define _PAGE_UNUSED3	(_AC(1, L)<<_PAGE_BIT_UNUSED3)
-#define _PAGE_PAT	(_AC(1, L)<<_PAGE_BIT_PAT)
-#define _PAGE_PAT_LARGE (_AC(1, L)<<_PAGE_BIT_PAT_LARGE)
+#define _PAGE_BIT(x)	(_AT(pteval_t,1) << (x))
+#define _PAGE_PRESENT	_PAGE_BIT(_PAGE_BIT_PRESENT)
+#define _PAGE_RW	_PAGE_BIT(_PAGE_BIT_RW)
+#define _PAGE_USER	_PAGE_BIT(_PAGE_BIT_USER)
+#define _PAGE_PWT	_PAGE_BIT(_PAGE_BIT_PWT)
+#define _PAGE_PCD	_PAGE_BIT(_PAGE_BIT_PCD)
+#define _PAGE_ACCESSED	_PAGE_BIT(_PAGE_BIT_ACCESSED)
+#define _PAGE_DIRTY	_PAGE_BIT(_PAGE_BIT_DIRTY)
+#define _PAGE_PSE	_PAGE_BIT(_PAGE_BIT_PSE)	/* 2MB page */
+#define _PAGE_GLOBAL	_PAGE_BIT(_PAGE_BIT_GLOBAL)	/* Global TLB entry */
+#define _PAGE_UNUSED1	_PAGE_BIT(_PAGE_BIT_UNUSED1)
+#define _PAGE_UNUSED2	_PAGE_BIT(_PAGE_BIT_UNUSED2)
+#define _PAGE_UNUSED3	_PAGE_BIT(_PAGE_BIT_UNUSED3)
+#define _PAGE_PAT	_PAGE_BIT(_PAGE_BIT_PAT)
+#define _PAGE_PAT_LARGE _PAGE_BIT(_PAGE_BIT_PAT_LARGE)
 
 #if defined(CONFIG_X86_64) || defined(CONFIG_X86_PAE)
-#define _PAGE_NX	(_AC(1, ULL) << _PAGE_BIT_NX)
+#define _PAGE_NX	_PAGE_BIT(_PAGE_BIT_NX)
 #else
 #define _PAGE_NX	0
 #endif
@@ -209,22 +210,22 @@ static inline int pmd_large(pmd_t pte)
 
 static inline pte_t pte_mkclean(pte_t pte)
 {
-	return __pte(pte_val(pte) & ~(pteval_t)_PAGE_DIRTY);
+	return __pte(pte_val(pte) & ~_PAGE_DIRTY);
 }
 
 static inline pte_t pte_mkold(pte_t pte)
 {
-	return __pte(pte_val(pte) & ~(pteval_t)_PAGE_ACCESSED);
+	return __pte(pte_val(pte) & ~_PAGE_ACCESSED);
 }
 
 static inline pte_t pte_wrprotect(pte_t pte)
 {
-	return __pte(pte_val(pte) & ~(pteval_t)_PAGE_RW);
+	return __pte(pte_val(pte) & ~_PAGE_RW);
 }
 
 static inline pte_t pte_mkexec(pte_t pte)
 {
-	return __pte(pte_val(pte) & ~(pteval_t)_PAGE_NX);
+	return __pte(pte_val(pte) & ~_PAGE_NX);
 }
 
 static inline pte_t pte_mkdirty(pte_t pte)
@@ -249,7 +250,7 @@ static inline pte_t pte_mkhuge(pte_t pte)
 
 static inline pte_t pte_clrhuge(pte_t pte)
 {
-	return __pte(pte_val(pte) & ~(pteval_t)_PAGE_PSE);
+	return __pte(pte_val(pte) & ~_PAGE_PSE);
 }
 
 static inline pte_t pte_mkglobal(pte_t pte)
@@ -259,7 +260,7 @@ static inline pte_t pte_mkglobal(pte_t pte)
 
 static inline pte_t pte_clrglobal(pte_t pte)
 {
-	return __pte(pte_val(pte) & ~(pteval_t)_PAGE_GLOBAL);
+	return __pte(pte_val(pte) & ~_PAGE_GLOBAL);
 }
 
 static inline pte_t pte_mkspecial(pte_t pte)

Re: [Bug 10732] REGRESSION: 2.6.26-rc2-git4: X server failed start onX61s laptop

[Hugh Dickins]

On Mon, 19 May 2008, Linus Torvalds wrote:
> On Mon, 19 May 2008, Hugh Dickins wrote:
> > 
> > This comes from an assumption in 1c12c4cf9411eb130b245fa8d0fbbaf989477c7b
> > mprotect: prevent alteration of the PAT bits, that PTE_MASK is what it's
> > supposed to be: whereas it's been wrong forever with PAE, staying 32-bit
> > where 64-bit is needed.
> 
> Can we *please* just fix PTE_MASK?

That's very much what I'd prefer too.  Jeremy has patches in Ingo's
tree to do that, which have been tested - though perhaps not in
combination with the PAT pte_modify changes.  I did check that they're
not incompatible in theory, but I sure better try them out later today.

> And can we agree to never EVER use that PAGE_MASK thing (which was only 
> ever meant to work on *addresses*) for any pte operations (including the 
> definition of PTE_MASK)? Because PAGE_MASK is very much the word-size, and 
> in 32-bit PAE, the page table entry is bigger.
> 
> IOE, PTE_MASK should be a "pteval_t". And it should have absolutely 
> *nothing* to do with PAGE_MASK. EVER.

Yes, Jeremy makes it a pteval_t.  (My builds and Ingo's builds succeed,
but I've not worked out how that goes down in assembly: there was an
_AT macro in there before, which you've kept too - Jeremy?)

> IOW, maybe something like this?
> 
> And no, I haven't tested this at all. But it should make PTE_MASK have
>  (a) the right type ("pteval_t", not "long" - the latter is pure and utter 
>      crap)
>  (b) the right value (proper mask, not a sign-extended long - again, the 
>      latter is pure and utter crap)
> 
> but for all I know there might be some broken code that depends on the 
> current incorrect and totally broken #defines, so this needs testing and 
> thinking about.

Yes, I'm highly resistant to taking untested patches here.  The two-liner
I sent last night was about my fifth attempt to get it working, and I did
start off from a small PTE_MASK correction which didn't work at all.  It
looked rather like yours, I guess I missed the  __PHYSICAL_LOW_BITS part.
Jeremy's goes a lot further, he'll know the gotchas better.

> It also causes these warnings on 32-bit PAE:
> 
> 	  AS      arch/x86/kernel/head_32.o
> 	arch/x86/kernel/head_32.S: Assembler messages:
> 	arch/x86/kernel/head_32.S:225: Warning: left operand is a bignum; integer 0 assumed
> 	arch/x86/kernel/head_32.S:609: Warning: left operand is a bignum; integer 0 assumed
> 
> and I do not see why (the end result seems to be identical).
> 
> Ingo, comments?
> 
> Oh, and those #define's should be moved from <asm/page.h> to 
> <asm/pgtable.h>, I think. They have nothing to do with pages (despite the 
> name of "physical_page_mask", and really are meaningful only in the 
> context of some kind of page table entry.

Jeremy still has them in asm/page.h.  Like your subsequent pte bit
cleanups, that can be added later: the important thing is to get X
working again on 32-bit NX systems.

Hugh

Re: [Bug 10732] REGRESSION: 2.6.26-rc2-git4: X server failed start onX61s laptop

[Jeremy Fitzhardinge]

Linus Torvalds wrote:
> On Mon, 19 May 2008, Hugh Dickins wrote:
>   
>> This comes from an assumption in 1c12c4cf9411eb130b245fa8d0fbbaf989477c7b
>> mprotect: prevent alteration of the PAT bits, that PTE_MASK is what it's
>> supposed to be: whereas it's been wrong forever with PAE, staying 32-bit
>> where 64-bit is needed.
>>     
>
> Can we *please* just fix PTE_MASK?
>
> And can we agree to never EVER use that PAGE_MASK thing (which was only 
> ever meant to work on *addresses*) for any pte operations (including the 
> definition of PTE_MASK)? Because PAGE_MASK is very much the word-size, and 
> in 32-bit PAE, the page table entry is bigger.
>
> IOE, PTE_MASK should be a "pteval_t". And it should have absolutely 
> *nothing* to do with PAGE_MASK. EVER.
>
> IOW, maybe something like this?
>   

That's pretty close to the core of my patches (just reposted), which 
have been cooking in x86.git for a week or so.

One thing I'd take from your patch is something like your 
__PHYSICAL_LOW_BITS definition, since its a bit clearer than what I 
did.  (I haven't updated my patch before posting just because I wanted 
to post exactly as tested.)

> And no, I haven't tested this at all. But it should make PTE_MASK have
>  (a) the right type ("pteval_t", not "long" - the latter is pure and utter 
>      crap)
>  (b) the right value (proper mask, not a sign-extended long - again, the 
>      latter is pure and utter crap)
>
> but for all I know there might be some broken code that depends on the 
> current incorrect and totally broken #defines, so this needs testing and 
> thinking about.
>
> It also causes these warnings on 32-bit PAE:
>
> 	  AS      arch/x86/kernel/head_32.o
> 	arch/x86/kernel/head_32.S: Assembler messages:
> 	arch/x86/kernel/head_32.S:225: Warning: left operand is a bignum; integer 0 assumed
> 	arch/x86/kernel/head_32.S:609: Warning: left operand is a bignum; integer 0 assumed
>
> and I do not see why (the end result seems to be identical).
>
> Ingo, comments?
>
> Oh, and those #define's should be moved from <asm/page.h> to 
> <asm/pgtable.h>, I think. They have nothing to do with pages (despite the 
> name of "physical_page_mask", and really are meaningful only in the 
> context of some kind of page table entry.
>
> 		Linus
>
> ---
>  include/asm-x86/page.h |    5 +++--
>  1 files changed, 3 insertions(+), 2 deletions(-)
>
> diff --git a/include/asm-x86/page.h b/include/asm-x86/page.h
> index b381f4a..34b4845 100644
> --- a/include/asm-x86/page.h
> +++ b/include/asm-x86/page.h
> @@ -10,8 +10,8 @@
>  
>  #ifdef __KERNEL__
>  
> -#define PHYSICAL_PAGE_MASK	(PAGE_MASK & __PHYSICAL_MASK)
> -#define PTE_MASK		(_AT(long, PHYSICAL_PAGE_MASK))
> +#define PHYSICAL_PAGE_MASK	(__PHYSICAL_MASK & ~__PHYSICAL_LOW_BITS)
> +#define PTE_MASK		(_AT(pteval_t, PHYSICAL_PAGE_MASK))
>  
>  #define PMD_PAGE_SIZE		(_AC(1, UL) << PMD_SHIFT)
>  #define PMD_PAGE_MASK		(~(PMD_PAGE_SIZE-1))
> @@ -24,6 +24,7 @@
>  /* to align the pointer to the (next) page boundary */
>  #define PAGE_ALIGN(addr)	(((addr)+PAGE_SIZE-1)&PAGE_MASK)
>  
> +#define __PHYSICAL_LOW_BITS	_AT(phys_addr_t, (PAGE_SIZE-1))
>  #define __PHYSICAL_MASK		_AT(phys_addr_t, (_AC(1,ULL) << __PHYSICAL_MASK_SHIFT) - 1)
>  #define __VIRTUAL_MASK		((_AC(1,UL) << __VIRTUAL_MASK_SHIFT) - 1)
>  
>   

    J

Re: [Bug 10732] REGRESSION: 2.6.26-rc2-git4: X server failed start onX61s laptop

[Jeremy Fitzhardinge]

Hugh Dickins wrote:
>> And can we agree to never EVER use that PAGE_MASK thing (which was only 
>> ever meant to work on *addresses*) for any pte operations (including the 
>> definition of PTE_MASK)? Because PAGE_MASK is very much the word-size, and 
>> in 32-bit PAE, the page table entry is bigger.
>>
>> IOE, PTE_MASK should be a "pteval_t". And it should have absolutely 
>> *nothing* to do with PAGE_MASK. EVER.
>>     
>
> Yes, Jeremy makes it a pteval_t.  (My builds and Ingo's builds succeed,
> but I've not worked out how that goes down in assembly: there was an
> _AT macro in there before, which you've kept too - Jeremy?)
>   

I got rid of a bunch of _AT() uses because the constants aren't used in 
.S files anywhere.  Also, I couldn't see how to represent a 64-bit 
constant in assembler, so I wasn't sure of their correctness (the as 
manual is irritatingly vague on the matter).

> Yes, I'm highly resistant to taking untested patches here.  The two-liner
> I sent last night was about my fifth attempt to get it working, and I did
> start off from a small PTE_MASK correction which didn't work at all.  It
> looked rather like yours, I guess I missed the  __PHYSICAL_LOW_BITS part.
> Jeremy's goes a lot further, he'll know the gotchas better.
>   

__PHYSICAL_LOW_BITS is a bit more elegant than what I did there (the 
problem is getting a physaddr_t-width PAGE_MASK).  But the formulation 
in my patch certainly works.

    J

[Bug #10732] REGRESSION: 2.6.26-rc2-git4: X server failed start on X61s laptop

[Rafael J. Wysocki]

This message has been generated automatically as a part of a report
of recent regressions.

The following bug entry is on the current list of known regressions
from 2.6.25.  Please verify if it still should be listed.


Bug-Entry	: http://bugzilla.kernel.org/show_bug.cgi?id=10732
Subject		: REGRESSION: 2.6.26-rc2-git4: X server failed start on X61s laptop
Submitter	: Theodore Ts'o <tytso@mit.edu>
Date		: 2008-05-17 7:32 (8 days old)
References	: http://marc.info/?l=linux-kernel&amp;m=121100994722524&amp;w=4

Re: [Bug #10732] REGRESSION: 2.6.26-rc2-git4: X server failed start on X61s laptop

[Hugh Dickins]

On Sat, 24 May 2008, Rafael J. Wysocki wrote:
> 
> The following bug entry is on the current list of known regressions
> from 2.6.25.  Please verify if it still should be listed.
> 
> Bug-Entry	: http://bugzilla.kernel.org/show_bug.cgi?id=10732
> Subject		: REGRESSION: 2.6.26-rc2-git4: X server failed start on X61s laptop
> Submitter	: Theodore Ts'o <tytso@mit.edu>
> Date		: 2008-05-17 7:32 (8 days old)
> References	: http://marc.info/?l=linux-kernel&amp;m=121100994722524&amp;w=4

I believe this was fixed by Jeremy's PTE_MASK fixes in 2.6.26-rc3-git2,
but can only say that they fixed my own X with PAE since PAT mods problem.

Ted, when you have a chance, please try a recent kernel and let us
know if this regression can now be removed from the list - thanks.

Hugh

Re: [Bug #10732] REGRESSION: 2.6.26-rc2-git4: X server failed start on X61s laptop

[Theodore Tso]

On Mon, May 26, 2008 at 11:24:02AM +0100, Hugh Dickins wrote:
> On Sat, 24 May 2008, Rafael J. Wysocki wrote:
> > 
> > The following bug entry is on the current list of known regressions
> > from 2.6.25.  Please verify if it still should be listed.
> > 
> > Bug-Entry	: http://bugzilla.kernel.org/show_bug.cgi?id=10732
> > Subject		: REGRESSION: 2.6.26-rc2-git4: X server failed start on X61s laptop
> > Submitter	: Theodore Ts'o <tytso@mit.edu>
> > Date		: 2008-05-17 7:32 (8 days old)
> > References	: http://marc.info/?l=linux-kernel&amp;m=121100994722524&amp;w=4
> 
> I believe this was fixed by Jeremy's PTE_MASK fixes in 2.6.26-rc3-git2,
> but can only say that they fixed my own X with PAE since PAT mods problem.
> 
> Ted, when you have a chance, please try a recent kernel and let us
> know if this regression can now be removed from the list - thanks.

Yes, I just tried it, with a freshly pulled mainline, and the problem
looks like it has been fixed.  So we can close out the bug and remove
it from the regression list.

						- Ted

Re: [Bug #10732] REGRESSION: 2.6.26-rc2-git4: X server failed start on X61s laptop

[Theodore Tso]

Bug #10732 is now closed.  BTW, it seems d*mned strange that the
submitter of the bug doesn't have the power to close a bug without
first grabbing the bug as the bug-fixer, changing its state to
ASSIGNED, and *then* being allowed to close the bug.  Grumble,
bugzilla's !#@#?@ process straightjacket....

							- Ted

Re: [Bug #10732] REGRESSION: 2.6.26-rc2-git4: X server failed start on X61s laptop

[Rafael J. Wysocki]

On Tuesday, 27 of May 2008, Theodore Tso wrote:
> Bug #10732 is now closed.  BTW, it seems d*mned strange that the
> submitter of the bug doesn't have the power to close a bug without
> first grabbing the bug as the bug-fixer, changing its state to
> ASSIGNED, and *then* being allowed to close the bug.  Grumble,
> bugzilla's !#@#?@ process straightjacket....

That depends on your account's access rights within the Bugzilla, which always
have been a mystery to me.

Thanks,
Rafael

Re: [Bug #10732] REGRESSION: 2.6.26-rc2-git4: X server failed start on X61s laptop

[Randy.Dunlap]

On Thu, 29 May 2008, Rafael J. Wysocki wrote:

> On Tuesday, 27 of May 2008, Theodore Tso wrote:
> > Bug #10732 is now closed.  BTW, it seems d*mned strange that the
> > submitter of the bug doesn't have the power to close a bug without
> > first grabbing the bug as the bug-fixer, changing its state to
> > ASSIGNED, and *then* being allowed to close the bug.  Grumble,
> > bugzilla's !#@#?@ process straightjacket....
> 
> That depends on your account's access rights within the Bugzilla, which always
> have been a mystery to me.

Martin Bligh should be able to clear that mystery up.

-- 
~Randy

Re: [Bug #10732] REGRESSION: 2.6.26-rc2-git4: X server failed start on X61s laptop

[Jon Tollefson]

Martin Bligh wrote:
> On Thu, May 29, 2008 at 2:15 PM, Randy.Dunlap <rdunlap@xenotime.net> wrote:
>   
>> On Thu, 29 May 2008, Rafael J. Wysocki wrote:
>>
>>     
>>> On Tuesday, 27 of May 2008, Theodore Tso wrote:
>>>       
>>>> Bug #10732 is now closed.  BTW, it seems d*mned strange that the
>>>> submitter of the bug doesn't have the power to close a bug without
>>>> first grabbing the bug as the bug-fixer, changing its state to
>>>> ASSIGNED, and *then* being allowed to close the bug.  Grumble,
>>>> bugzilla's !#@#?@ process straightjacket....
>>>>         
>>> That depends on your account's access rights within the Bugzilla, which always
>>> have been a mystery to me.
>>>       
>> Martin Bligh should be able to clear that mystery up.
>>     
>
> Jon ... is that something we can configure? Seems ... wrong.
> Submitter and owner should both have power to edit the bug.
>   
Submitter and owner should be able to edit the bug as far I as can see.

If the bug is not already in the resolved state then you can use the
Close check box at the end of the 'Resolve' option line to close a bug
with one submit.  You do not have to change it to the assign state in
order to close it either - if it is the submitter or owner that is
closing it.  These statements do not apply to bugzilla in general but do
apply to the bugzilla.kernel.org one.

Jon

Fix for asm warning in head_32.S

[Joe Korty]

On Mon, May 19, 2008 at 04:10:02PM -0700, Linus Torvalds wrote:
> It also causes these warnings on 32-bit PAE:
> 
> 	  AS      arch/x86/kernel/head_32.o
> 	arch/x86/kernel/head_32.S: Assembler messages:
> 	arch/x86/kernel/head_32.S:225: Warning: left operand is a bignum; integer 0 assumed
> 	arch/x86/kernel/head_32.S:609: Warning: left operand is a bignum; integer 0 assumed
> 
> and I do not see why (the end result seems to be identical).



Fix head_32.S gcc bignum warnings when CONFIG_PAE=y.

    arch/x86/kernel/head_32.S: Assembler messages:
    arch/x86/kernel/head_32.S:225: Warning: left operand is a bignum; integer 0 assumed
    arch/x86/kernel/head_32.S:609: Warning: left operand is a bignum; integer 0 assumed

The assembler was stumbling over the 64-bit constant 0x100000000 in the
KPMDS #define.

Testing: a cmp(1) on head_32.o before and after shows the binary is unchanged.

Signed-off-by: Joe Korty <joe.korty@ccur.com

Index: a/arch/x86/kernel/head_32.S
===================================================================
--- a.orig/arch/x86/kernel/head_32.S	2008-06-02 16:51:20.000000000 -0400
+++ a/arch/x86/kernel/head_32.S	2008-06-02 16:54:05.000000000 -0400
@@ -189,7 +189,7 @@
 	 * this stage.
 	 */
 
-#define KPMDS ((0x100000000-__PAGE_OFFSET) >> 30) /* Number of kernel PMDs */
+#define KPMDS (((-__PAGE_OFFSET) >> 30) & 3) /* Number of kernel PMDs */
 
 	xorl %ebx,%ebx				/* %ebx is kept at zero */