pte_modify fix backport...
This patch [1] corrects the behaviour of pte_modify() on ia64 to not clear page protection flags it shouldn't touch.
Without this patch, it clears the uncacheable flag on UC memory regions, which can cause MCAs with device I/O.
It was taken into 2.6.0 a while ago, but is yet to be included in 2.4 - the patch here is rediffed against 2.4.26-rc5.
Is there chance of it being taken in?
--- [1]
include/asm-ia64/pgtable.h +3 -2
@@ -60,7 +60,8 @@#define _PAGE_PROTNONE (__IA64_UL(1) << 63)#define _PFN_MASK _PAGE_PPN_MASK -#define _PAGE_CHG_MASK (_PFN_MASK | _PAGE_A | _PAGE_D) +/* Mask of bits which may be changed by pte_modify(); the odd bits are there for _PAGE_PROTNONE */ +#define _PAGE_CHG_MASK (_PAGE_P | _PAGE_PROTNONE | _PAGE_PL_MASK | _PAGE_AR_MASK | _PAGE_ED)#define _PAGE_SIZE_4K 12#define _PAGE_SIZE_8K 13
@@ -216,7 +217,7 @@ ({ pte_t __pte; pte_val(__pte) = physpage + pgprot_val(pgprot); __pte; })#define pte_modify (_pte, newprot) \ - (__pte((pte_val(_pte) & _PAGE_CHG_MASK) | pgprot_val(newprot))) + (__pte((pte_val(_pte) & ~_PAGE_CHG_MASK) | (pgprot_val(newprot) & _PAGE_CHG_MASK)))#define page_pte_prot (page,prot) mk_pte(page, prot)#define page_pte (page) page_pte_prot(page, __pgprot(0))
Signature · 2 lines
Daniel J Blueman Software Engineer, Quadrics Ltd
Quoted 10 lines · Daniel J Blueman
This patch [1] corrects the behaviour of
pte_modify()on ia64 to not clear page protection flags it shouldn't touch.Without this patch, it clears the uncacheable flag on UC memory regions, which can cause MCAs with device I/O.
It was taken into 2.6.0 a while ago, but is yet to be included in 2.4 - the patch here is rediffed against 2.4.26-rc5.
Is there chance of it being taken in?
I put this in the 2.4 ia64 BK tree a couple days ago. Can you check
here: http://lia64.bkbits.net:8080/linux-ia64-2.4 and make sure that it works for you?
The plain patch is also available here, if that works better for you:
ftp://ftp.kernel.org/pub/linux/kernel/ports/ia64/v2.4/testing/cset/index.html
Quoted 26 lines
diff -durN
linux-2.4.26-rc5-orig/include/asm-ia64/pgtable.hlinux-2.4.26-rc5-patched/include/asm-ia64/pgtable.hinclude/asm-ia64/pgtable.h +3
-2@@ -60,7 +60,8 @@#define _PAGE_PROTNONE (__IA64_UL(1) << 63)#define _PFN_MASK _PAGE_PPN_MASK -#define _PAGE_CHG_MASK (_PFN_MASK | _PAGE_A | _PAGE_D) +/* Mask of bits which may be changed by pte_modify(); the odd bits are there for _PAGE_PROTNONE */ +#define _PAGE_CHG_MASK (_PAGE_P | _PAGE_PROTNONE | _PAGE_PL_MASK | _PAGE_AR_MASK | _PAGE_ED)#define _PAGE_SIZE_4K 12#define _PAGE_SIZE_8K 13@@ -216,7 +217,7 @@ ({ pte_t __pte; pte_val(__pte) = physpage + pgprot_val(pgprot); __pte; })#define pte_modify (_pte, newprot) \ - (__pte((pte_val(_pte) & _PAGE_CHG_MASK) | pgprot_val(newprot))) + (__pte((pte_val(_pte) & ~_PAGE_CHG_MASK) | (pgprot_val(newprot) & _PAGE_CHG_MASK)))#define page_pte_prot (page,prot) mk_pte(page, prot)#define page_pte (page) page_pte_prot(page, __pgprot(0))Signature · 3 lines
-- Daniel J Blueman Software Engineer, Quadrics Ltd
Quoted 35 lines · Bjorn Helgaas
Quoted 10 lines · Daniel J Blueman
This patch [1] corrects the behaviour of
pte_modify()on ia64 to not clear page protection flags it shouldn't touch.Without this patch, it clears the uncacheable flag on UC memory regions, which can cause MCAs with device I/O.
It was taken into 2.6.0 a while ago, but is yet to be included in 2.4 - the patch here is rediffed against 2.4.26-rc5.
Is there chance of it being taken in?
I put this in the 2.4 ia64 BK tree a couple days ago. Can you check
here: http://lia64.bkbits.net:8080/linux-ia64-2.4 and make sure that it works for you?The plain patch is also available here, if that works better for you:
ftp://ftp.kernel.org/pub/linux/kernel/ports/ia64/v2.4/testing/cset/index.htmlQuoted 16 lines
diff -durN
linux-2.4.26-rc5-orig/include/asm-ia64/pgtable.hlinux-2.4.26-rc5-patched/include/asm-ia64/pgtable.hinclude/asm-ia64/pgtable.h +2
-1@@ -60,7 +60,8 @@#define _PAGE_PROTNONE (__IA64_UL(1) << 63)#define _PFN_MASK _PAGE_PPN_MASK -#define _PAGE_CHG_MASK (_PFN_MASK | _PAGE_A | _PAGE_D) +/* Mask of bits which may be changed by pte_modify(); the odd bits are there for _PAGE_PROTNONE */ +#define _PAGE_CHG_MASK (_PAGE_P | _PAGE_PROTNONE | _PAGE_PL_MASK | _PAGE_AR_MASK | _PAGE_ED)#define _PAGE_SIZE_4K 12#define _PAGE_SIZE_8K 13@@ -216,7 +217,7 @@ ({pte_t __pte;pte_val (__pte) = physpage +pgprot_val (pgprot); __pte; })
Quoted 6 lines
include/asm-ia64/pgtable.h line 223
Quoted 6 lines
include/asm-ia64/pgtable.h +1
-1#define pte_modify (_pte, newprot) \ - (__pte((pte_val(_pte) & _PAGE_CHG_MASK) | pgprot_val(newprot))) + (__pte((pte_val(_pte) & ~_PAGE_CHG_MASK) | (pgprot_val(newprot) & _PAGE_CHG_MASK)))#define page_pte_prot (page,prot) mk_pte(page, prot)#define page_pte (page) page_pte_prot(page, __pgprot(0))
Bjorn,
The two changesets are good. Thanks!
Signature · 2 lines
Daniel J Blueman Software Engineer, Quadrics Ltd