Skip to contentSkip to conversation

powerpc/mm: Fix 40x and 8xx vs. _PAGE_SPECIAL

PATCH In review Benjamin Herrenschmidt 3 participants 7 messages Sep 23, 2009 – Sep 24, 2009 PowerPC

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 powerpc/mm: Fix 40x and 8xx vs. _PAGE_SPECIAL

The test to check whether we have _PAGE_SPECIAL defined is broken, since we always define it, just not always to a meaninful value :-)

That broke 8xx and 40x under some circumstances.

This fixes it by adding _PAGE_SPECIAL for both of these since they had a free PTE bit, and removing the condition around advertising it.

---

Bernhard, Rex, please let me know if that works for you.

arch/powerpc/include/asm/pte-40x.h +1 -0
index 6c3e1f4..ec0b0b0 100644
@@ -43,6 +43,7 @@
 #define	_PAGE_NO_CACHE	0x004	/* I: caching is inhibited */
 #define	_PAGE_WRITETHRU	0x008	/* W: caching is write-through */
 #define	_PAGE_USER	0x010	/* matches one of the zone permission bits */
+#define	_PAGE_SPECIAL	0x020	/* software: Special page */
 #define	_PAGE_RW	0x040	/* software: Writes permitted */
 #define	_PAGE_DIRTY	0x080	/* software: dirty page */
 #define _PAGE_HWWRITE	0x100	/* hardware: Dirty & RW, set in exception */
arch/powerpc/include/asm/pte-8xx.h +1 -0
index 94e9797..dd5ea95 100644
@@ -32,6 +32,7 @@
 #define _PAGE_FILE	0x0002	/* when !present: nonlinear file mapping */
 #define _PAGE_NO_CACHE	0x0002	/* I: cache inhibit */
 #define _PAGE_SHARED	0x0004	/* No ASID (context) compare */
+#define _PAGE_SPECIAL	0x0008	/* SW entry, forced to 0 by the TLB miss */
 
 /* These five software bits must be masked out when the entry is loaded
  * into the TLB.
arch/powerpc/include/asm/pte-common.h +0 -5
index c3b6507..f2b3701 100644
@@ -25,9 +25,6 @@
 #ifndef _PAGE_WRITETHRU
 #define _PAGE_WRITETHRU	0
 #endif
-#ifndef _PAGE_SPECIAL
-#define _PAGE_SPECIAL	0
-#endif
 #ifndef _PAGE_4K_PFN
 #define _PAGE_4K_PFN		0
 #endif
@@ -179,7 +176,5 @@ extern unsigned long bad_call_to_PMD_PAGE_SIZE(void);
 #define HAVE_PAGE_AGP
 
 /* Advertise support for _PAGE_SPECIAL */
-#ifdef _PAGE_SPECIAL
 #define __HAVE_ARCH_PTE_SPECIAL
-#endif
 
#1 lore raw

Works perfectly here. Thank you for the patch, I was not feeling confident = enough to play around with the page bits :)

Bernhard WEIRICH
Software Development
________________________________________
Von: Benjamin Herrenschmidt [[email protected]]
Gesendet: Mittwoch, 23. September 2009 06:12
An: Weirich, Bernhard; Rex Feany
Cc: linuxppc-dev list
Betreff: [PATCH] powerpc/mm: Fix 40x and 8xx vs. _PAGE_SPECIAL

The test to check whether we have _PAGE_SPECIAL defined is broken, since we always define it, just not always to a meaninful value :-)

That broke 8xx and 40x under some circumstances.

This fixes it by adding _PAGE_SPECIAL for both of these since they had a free PTE bit, and removing the condition around advertising it.

---

Bernhard, Rex, please let me know if that works for you.

arch/powerpc/include/asm/= renamed from arch/powerpc/include/asm/pte-40x.h +0 -0

pte-40x.h
index 6c3e1f4..ec0b0b0 100644

arch/powerpc/include/asm/pte-40x.h +0 -0
@@ -43,6 +43,7 @@
 #define        _PAGE_NO_CACHE  0x004   /* I: caching is inhibited */
 #define        _PAGE_WRITETHRU 0x008   /* W: caching is write-through */
 #define        _PAGE_USER      0x010   /* matches one of the zone permissi=

on bits */

+#define        _PAGE_SPECIAL   0x020   /* software: Special page */
 #define        _PAGE_RW        0x040   /* software: Writes permitted */
 #define        _PAGE_DIRTY     0x080   /* software: dirty page */
 #define _PAGE_HWWRITE  0x100   /* hardware: Dirty & RW, set in exception *=

/

arch/powerpc/include/asm/= renamed from arch/powerpc/include/asm/pte-8xx.h +0 -0

pte-8xx.h
index 94e9797..dd5ea95 100644

arch/powerpc/include/asm/pte-8xx.h +1 -0
@@ -32,6 +32,7 @@
 #define _PAGE_FILE     0x0002  /* when !present: nonlinear file mapping */
 #define _PAGE_NO_CACHE 0x0002  /* I: cache inhibit */
 #define _PAGE_SHARED   0x0004  /* No ASID (context) compare */
+#define _PAGE_SPECIAL  0x0008  /* SW entry, forced to 0 by the TLB miss */
 
 /* These five software bits must be masked out when the entry is loaded
  * into the TLB.
arch/powerpc/include/a= renamed from arch/powerpc/include/asm/pte-common.h +0 -0

sm/pte-common.h
index c3b6507..f2b3701 100644

arch/powerpc/include/asm/pte-common.h +0 -5
@@ -25,9 +25,6 @@
 #ifndef _PAGE_WRITETHRU
 #define _PAGE_WRITETHRU        0
 #endif
-#ifndef _PAGE_SPECIAL
-#define _PAGE_SPECIAL  0
-#endif
 #ifndef _PAGE_4K_PFN
 #define _PAGE_4K_PFN           0
 #endif
@@ -179,7 +176,5 @@ extern unsigned long bad_call_to_PMD_PAGE_SIZE(void);
 #define HAVE_PAGE_AGP
 
 /* Advertise support for _PAGE_SPECIAL */
-#ifdef _PAGE_SPECIAL
 #define __HAVE_ARCH_PTE_SPECIAL
-#endif
#2 lore raw
In reply to Benjamin Herrenschmidt

Thus spake Benjamin Herrenschmidt ([email protected]):

Quoted 1 line

Bernhard, Rex, please let me know if that works for you.

it doesn't work for me, it crashes differently then before though!

This patch worked for me:

arch/powerpc/include/asm/pte-common.h 2009-09-22 11:13:08.000000000 +0200 line 182

Quoted 13 lines
arch/powerpc/include/asm/pte-common.h 2009-09-22 11:13:08.000000000 +0200 renamed from arch/powerpc/include/asm/pte-common.h 2009-09-16 11:23:44.000000000 +0200 +1 -1
@@ -176,7 +176,7 @@
 #define HAVE_PAGE_AGP
 
 /* Advertise support for _PAGE_SPECIAL */
-#ifdef _PAGE_SPECIAL
+#if _PAGE_SPECIAL != 0
 #define __HAVE_ARCH_PTE_SPECIAL
 #endif
 

without either patch:

boots into userspace, starts running my code
which uses mmap to access an fpga:

BUG: Bad page map in process loadfpga  pte:ff1008d3 pmd:02103001
addr:30028000 vm_flags:400844fb anon_vma:(null) mapping:c30000e8 index:ff100
vma->vm_ops->fault: 0x0
vma->vm_file->f_op->mmap: mmap_mem+0x0/0xa8
Call Trace:
[c2107db0] [c00067d0] show_stack+0x44/0x14c (unreliable)
[c2107df0] [c0056854] print_bad_pte+0x1b4/0x1d0
[c2107e20] [c0056898] vm_normal_page+0x28/0x50
[c2107e30] [c0056c30] unmap_vmas+0x1b8/0x510
[c2107eb0] [c005b17c] unmap_region+0x94/0x124
[c2107ef0] [c005c110] do_munmap+0x264/0x2dc
[c2107f20] [c005c1c0] sys_munmap+0x38/0x5c
[c2107f40] [c000d0dc] ret_from_syscall+0x0/0x38

trunk+benh complicated patch:

crashes earlier:

BUG: Bad page map in process swapper  pte:001f88e9 pmd:020d8001
addr:7ffffff6 vm_flags:00100177 anon_vma:c20d95e0 mapping:(null) index:7ffff
Call Trace:
[c341fd20] [c00067d0] show_stack+0x44/0x14c (unreliable)
[c341fd60] [c0056860] print_bad_pte+0x1b4/0x1d0
[c341fd90] [c00568c8] vm_normal_page+0x4c/0x78
[c341fda0] [c0056988] follow_page+0x94/0x1b8
[c341fdc0] [c005854c] __get_user_pages+0x324/0x3c4
[c341fe20] [c006c12c] get_arg_page+0x40/0xac
[c341fe40] [c006c374] copy_strings+0xec/0x1d4
[c341fe80] [c006c47c] copy_strings_kernel+0x20/0x38
[c341fea0] [c006d3cc] do_execve+0x110/0x240
[c341fee0] [c00070c4] sys_execve+0x50/0x7c
[c341ff00] [c000d0dc] ret_from_syscall+0x0/0x38
[c341ffc0] [c00023ac] init_post+0x5c/0x168
[c341ffe0] [c01df22c] kernel_init+0x10c/0x124
[c341fff0] [c000cf3c] kernel_thread+0x4c/0x68

this is running on a MPC 875, and I'm happy to test any patches or add any debugging code if it helps.

thanks!
/rex

#3 lore raw
Quoted 2 lines · Rex Feany

this is running on a MPC 875, and I'm happy to test any patches or add any debugging code if it helps.

Hrm... smells like something screwy in the 8xx PTE handling.

Let me have another look.

Cheers,
Ben.

#4 lore raw
In reply to Rex Feany
Quoted 5 lines · Rex Feany

Thus spake Benjamin Herrenschmidt ([email protected]):

Quoted 1 line

Bernhard, Rex, please let me know if that works for you.

it doesn't work for me, it crashes differently then before though!

Hrm. This is really strange...

Quoted 5 lines

trunk+benh complicated patch:

crashes earlier:

BUG: Bad page map in process swapper  pte:001f88e9 pmd:020d8001

It's complaining because for some reason a PTE ends up with _PAGE_SPECIAL set which indeed shouldn't be the case for a stack page which is what is being accessed in our case. However, I have no idea how that bit ended up being set. The generic code shouldn't be setting it, and I looked several times at the TLB miss handling on 8xx and can't find how in hell it would inadvertently set that bit in the PTE...

Scott, I don't have any 8xx HW, any chance you can have a look at what's going on here ?

The other bits in there look fine. So something must be whacking _PAGE_SPECIAL in somewhere or at least whacking 0x8 in there.

Also, _PAGE_HWWRITE isn't set, but the DataTLBError code will set it when setting dirty, so the page must have been created pre-dirtied or something... all very weird I must say.

Quoted 22 lines

addr:7ffffff6 vm_flags:00100177 anon_vma:c20d95e0 mapping:(null) index:7ffff

Call Trace:
[c341fd20] [c00067d0] show_stack+0x44/0x14c (unreliable)
[c341fd60] [c0056860] print_bad_pte+0x1b4/0x1d0
[c341fd90] [c00568c8] vm_normal_page+0x4c/0x78
[c341fda0] [c0056988] follow_page+0x94/0x1b8
[c341fdc0] [c005854c] __get_user_pages+0x324/0x3c4
[c341fe20] [c006c12c] get_arg_page+0x40/0xac
[c341fe40] [c006c374] copy_strings+0xec/0x1d4
[c341fe80] [c006c47c] copy_strings_kernel+0x20/0x38
[c341fea0] [c006d3cc] do_execve+0x110/0x240
[c341fee0] [c00070c4] sys_execve+0x50/0x7c
[c341ff00] [c000d0dc] ret_from_syscall+0x0/0x38
[c341ffc0] [c00023ac] init_post+0x5c/0x168
[c341ffe0] [c01df22c] kernel_init+0x10c/0x124
[c341fff0] [c000cf3c] kernel_thread+0x4c/0x68

this is running on a MPC 875, and I'm happy to test any patches or add any debugging code if it helps.

thanks!
/rex

#5 lore raw

Thus spake Benjamin Herrenschmidt ([email protected]):

Quoted 1 line

Hrm. This is really strange...

Is this a problem? in pte-8xx.h you added:

+#define _PAGE_SPECIAL 0x0008 /* SW entry, forced to 0 by the TLB miss */

but there is already this:

#define _PAGE_EXEC 0x0008 /* software: i-cache coherency required */

take care!
/rex.

#6 lore raw
Quoted 11 lines · Rex Feany

Thus spake Benjamin Herrenschmidt ([email protected]):

Quoted 1 line

Hrm. This is really strange...

Is this a problem? in pte-8xx.h you added:

+#define _PAGE_SPECIAL 0x0008 /* SW entry, forced to 0 by the TLB miss */

but there is already this:

#define _PAGE_EXEC 0x0008 /* software: i-cache coherency required */

Ohhh... you aren't using upstream ! That explains :-)

I removed _PAGE_EXEC from 8xx (since it doesn't use it for anything) recently.

I think the best option is to go for Bernhard's patch for "stable" and my patch for upstream, so that "stable" gets the bug fix without the feature of _PAGE_SPECIAL and upstream gets it both.

Cheers,
Ben.