apply page shift to PFN instead of VA in pfn_to_virt
Subsystems (4)
cover apply page shift to PFN instead of VA in pfn_to_virt
This is a tiny fix to pfn_to_virt() for some platforms.
The original implementaion of pfn_to_virt() takes PFN instead of PA as the input to macro __va, with PAGE_SHIFT applying to the converted VA, which is not right under most conditions, especially when there's an offset in __va.
Yan Zhao (4):
asm-generic/page.h: apply page shift to PFN instead of VA in
pfn_to_virt
csky: apply page shift to PFN instead of VA in pfn_to_virt
Hexagon: apply page shift to PFN instead of VA in pfn_to_virt
openrisc: apply page shift to PFN instead of VA in pfn_to_virt
arch/csky/include/asm/page.h | 2 +- arch/hexagon/include/asm/page.h | 2 +- arch/openrisc/include/asm/page.h | 2 +- include/asm-generic/page.h | 2 +- 4 files changed, 4 insertions(+), 4 deletions(-)
base-commit: 41bccc98fb7931d63d03f326a746ac4d429c1dd3
Signature · 1 line
2.17.1
PATCH 1/4 asm-generic/page.h: apply page shift to PFN instead of VA in pfn_to_virt
Apply the page shift to PFN to get physical address for final VA. The macro __va should take physical address instead of PFN as input.
--- include/asm-generic/page.h | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)
include/asm-generic/page.h +1 -1
@@ -81,7 +81,7 @@ static inline unsigned long virt_to_pfn(const void *kaddr)#define virt_to_pfn virt_to_pfnstatic inline void *pfn_to_virt (unsigned long pfn) { -return __va (pfn) <<PAGE_SHIFT ; +return __va (pfn <<PAGE_SHIFT ); }#define pfn_to_virt pfn_to_virt
Signature · 1 line
2.17.1
Re: [PATCH 1/4] asm-generic/page.h: apply page shift to PFN instead of VA in pfn_to_virt
Quoted 15 lines · Yan Zhao
Apply the page shift to PFN to get physical address for final VA. The macro
__vashould take physical address instead of PFN as input.Fixes 2d78057f0dd4 ("asm-generic/page.h: Make pfn accessors static inlines")Signed-off-by Yan Zhao <[email protected]>--- include/asm-generic/page.h | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)include/asm-generic/page.h +0
-0
include/asm-generic/page.h static inline unsigned long virt_to_pfn(const void *kaddr) · line 84
Quoted 5 lines
include/asm-generic/page.h +1
-1@@ -81,7 +81,7 @@ static inline unsigned long virt_to_pfn(const void *kaddr)#define virt_to_pfn virt_to_pfnstatic inline void *pfn_to_virt (unsigned long pfn) { -return __va (pfn) <<PAGE_SHIFT ; +return __va (pfn <<PAGE_SHIFT );
Oh, that's a terrible bug; Thx for fixing it.
Quoted 6 lines
include/asm-generic/page.h +0
-1}#define pfn_to_virt pfn_to_virt --2.17.1
Signature · 2 lines
Best Regards Guo Ren
Re: [PATCH 1/4] asm-generic/page.h: apply page shift to PFN instead of VA in pfn_to_virt
In reply to Yan ZhaoQuoted 5 lines · Yan Zhao
Apply the page shift to PFN to get physical address for final VA. The macro
__vashould take physical address instead of PFN as input.Fixes 2d78057f0dd4 ("asm-generic/page.h: Make pfn accessors static inlines")Signed-off-by Yan Zhao <[email protected]>
My bug, obviously. :(
I thought this was already applied with the other fixes, but maybe it was missed?
Yours,
Linus Walleij
Re: [PATCH 1/4] asm-generic/page.h: apply page shift to PFN instead of VA in pfn_to_virt
Quoted 5 lines · Linus Walleij
Quoted 3 lines · Yan Zhao
Apply the page shift to PFN to get physical address for final VA. The macro
__vashould take physical address instead of PFN as input.
Quoted 9 lines
Quoted 2 lines
Fixes 2d78057f0dd4 ("asm-generic/page.h: Make pfn accessors static inlines")Signed-off-by Yan Zhao <[email protected]>My bug, obviously. :(
Reviewed-by Linus Walleij <[email protected]>I thought this was already applied with the other fixes, but maybe it was missed?
Hi Linus,
The other 3 in csky/hexagon/openrisc should have been applied in
https://lore.kernel.org/all/[email protected]/.
This one in asm-generic is not, because Arnd said he is going to remove the header as a whole soon.
I explained it in the change log :)
"The pfn_to_virt() in asm-generic/page.h is not touched in this patch as it's referenced by page_to_virt() in that header while the whole header is going to be removed as a whole due to no one including it."
Thanks
Yan
PATCH 2/4 csky: apply page shift to PFN instead of VA in pfn_to_virt
In reply to Yan ZhaoApply the page shift to PFN to get physical address for final VA. The macro __va should take physical address instead of PFN as input.
--- arch/csky/include/asm/page.h | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)
arch/csky/include/asm/page.h +1 -1
@@ -84,7 +84,7 @@ static inline unsigned long virt_to_pfn(const void *kaddr)static inline void *pfn_to_virt (unsigned long pfn) { -return (void *)((unsigned long )__va (pfn) <<PAGE_SHIFT ); +return __va (pfn <<PAGE_SHIFT ); }#define MAP_NR (x) PFN_DOWN((unsigned long)(x) - PAGE_OFFSET - \
Signature · 1 line
2.17.1
Re: [PATCH 2/4] csky: apply page shift to PFN instead of VA in pfn_to_virt
Quoted 15 lines · Yan Zhao
Apply the page shift to PFN to get physical address for final VA. The macro
__vashould take physical address instead of PFN as input.Fixes c1884e1e1164 ("csky: Make pfn accessors static inlines")Signed-off-by Yan Zhao <[email protected]>--- arch/csky/include/asm/page.h | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)arch/csky/include/asm/page.h +0
-0
arch/csky/include/asm/page.h static inline unsigned long virt_to_pfn(const void *kaddr) · line 88
Quoted 6 lines
arch/csky/include/asm/page.h +1
-1@@ -84,7 +84,7 @@ static inline unsigned long virt_to_pfn(const void *kaddr)static inline void *pfn_to_virt (unsigned long pfn) { -return (void *)((unsigned long )__va (pfn) <<PAGE_SHIFT ); +return __va (pfn <<PAGE_SHIFT ); }
Quoted 5 lines
arch/csky/include/asm/page.h +0
-1#define MAP_NR (x) PFN_DOWN((unsigned long)(x) - PAGE_OFFSET - \ --2.17.1
Signature · 2 lines
Best Regards Guo Ren
PATCH 3/4 Hexagon: apply page shift to PFN instead of VA in pfn_to_virt
In reply to Yan ZhaoApply the page shift to PFN to get physical address for final VA. The macro __va should take physical address instead of PFN as input.
--- arch/hexagon/include/asm/page.h | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)
arch/hexagon/include/asm/page.h +1 -1
@@ -135,7 +135,7 @@ static inline unsigned long virt_to_pfn(const void *kaddr)static inline void *pfn_to_virt (unsigned long pfn) { -return (void *)((unsigned long )__va (pfn) <<PAGE_SHIFT ); +return __va (pfn <<PAGE_SHIFT ); }
Signature · 1 line
2.17.1
PATCH 4/4 openrisc: apply page shift to PFN instead of VA in pfn_to_virt
In reply to Yan ZhaoApply the page shift to PFN to get physical address for final VA. The macro __va should take physical address instead of PFN as input.
--- arch/openrisc/include/asm/page.h | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)
arch/openrisc/include/asm/page.h +1 -1
@@ -79,7 +79,7 @@ static inline unsigned long virt_to_pfn(const void *kaddr)static inline void *pfn_to_virt (unsigned long pfn) { -return (void *)((unsigned long )__va (pfn) <<PAGE_SHIFT ); +return __va (pfn <<PAGE_SHIFT ); }#define virt_to_page (addr) \
Signature · 1 line
2.17.1
Quoted 6 lines · Yan Zhao
This is a tiny fix to
pfn_to_virt()for some platforms.The original implementaion of
pfn_to_virt()takes PFN instead of PA as the input to macro__va, withPAGE_SHIFTapplying to the converted VA, which is not right under most conditions, especially when there's an offset in__va.
Ooops that's right, I wonder why I got it wrong.
Arithmetic made it not regress :/
Thank you so much for fixing this Yan!
Arnd: I think you can take most of them through the arch tree.
Yours,
Linus Walleij
Arnd Bergmann
maintainerQuoted 8 lines · Yan Zhao
This is a tiny fix to
pfn_to_virt()for some platforms.The original implementaion of
pfn_to_virt()takes PFN instead of PA as the input to macro__va, withPAGE_SHIFTapplying to the converted VA, which is not right under most conditions, especially when there's an offset in__va.
Quoted 6 lines
Yan Zhao (4):
asm-generic/page.h: apply page shift to PFN instead of VA in
pfn_to_virt
csky: apply page shift to PFN instead of VA inpfn_to_virt
Hexagon: apply page shift to PFN instead of VA inpfn_to_virt
openrisc: apply page shift to PFN instead of VA inpfn_to_virt
Nice catch, this is clearly a correct fix, and I can take the series through the asm-generic tree if we want to take this approach.
I made a couple of interesting observations looking at this patch though:
- this function is only used in architecture specific code on m68k, riscv and s390, though a couple of other architectures have the same definition.
- There is another function that does the same thing called pfn_to_kaddr(), which is defined on arm, arm64, csky, loongarch, mips, nios2, powerpc, s390, sh, sparc and x86, as well as yet another pfn_va() on parisc.
- the asm-generic/page.h file used to be included by h8300, c6x and blackfin, all of which are now gone. It has no users left and can just as well get removed, unless we find a new use for it.
Since it looks like the four broken functions you fix don't have a single caller, maybe it would be better to just remove them all?
How exactly did you notice the function being wrong, did you try to add a user somewhere, or just read through the code?
Arnd
Quoted 37 lines · Arnd Bergmann
Quoted 8 lines · Yan Zhao
This is a tiny fix to
pfn_to_virt()for some platforms.The original implementaion of
pfn_to_virt()takes PFN instead of PA as the input to macro__va, withPAGE_SHIFTapplying to the converted VA, which is not right under most conditions, especially when there's an offset in__va.Quoted 6 lines
Yan Zhao (4):
asm-generic/page.h: apply page shift to PFN instead of VA in
pfn_to_virt
csky: apply page shift to PFN instead of VA inpfn_to_virt
Hexagon: apply page shift to PFN instead of VA inpfn_to_virt
openrisc: apply page shift to PFN instead of VA inpfn_to_virtNice catch, this is clearly a correct fix, and I can take the series through the asm-generic tree if we want to take this approach.
I made a couple of interesting observations looking at this patch though:
- this function is only used in architecture specific code on m68k, riscv and s390, though a couple of other architectures have the same definition.
- There is another function that does the same thing called
pfn_to_kaddr(), which is defined on arm, arm64, csky, loongarch, mips, nios2, powerpc, s390, sh, sparc and x86, as well as yet anotherpfn_va()on parisc.- the
asm-generic/page.hfile used to be included by h8300, c6x and blackfin, all of which are now gone. It has no users left and can just as well get removed, unless we find a new use for it.
Quoted 7 lines
Since it looks like the four broken functions you fix don't have a single caller, maybe it would be better to just remove them all?
How exactly did you notice the function being wrong, did you try to add a user somewhere, or just read through the code?
I came across them when I was debugging an unexpected kernel page fault on x86, and I was not sure whether page_to_virt() was compiled in asm-generic/page.h or linux/mm.h.
Though finally, it turned out that the one in linux/mm.h was used, which yielded the right result and the unexpected kernel page fault in my case was not related to page_to_virt(), it did lead me to noticing that the pfn_to_virt() in asm-generic/page.h and other 3 archs did not look right.
Yes, unlike virt_to_pfn() which still has a caller in openrisc (among csky, Hexagon, openrisc), pfn_to_virt() now does not have a caller in the 3 archs. Though both virt_to_pfn() and pfn_to_virt() are referenced in asm-generic/page.h, I also not sure if we need to remove the asm-generic/page.h which may serve as a template to future archs ?
So, either way looks good to me :)
Arnd Bergmann
maintainerQuoted 14 lines · Yan Zhao
Quoted 5 lines · Arnd Bergmann
On Wed, Jan 31, 2024, at 06:51, Yan Zhao wrote:
How exactly did you notice the function being wrong, did you try to add a user somewhere, or just read through the code?
I came across them when I was debugging an unexpected kernel page fault on x86, and I was not sure whether
page_to_virt()was compiled inasm-generic/page.horlinux/mm.h.
Though finally, it turned out that the one inlinux/mm.hwas used, which yielded the right result and the unexpected kernel page fault in my case was not related topage_to_virt(), it did lead me to noticing that thepfn_to_virt()inasm-generic/page.hand other 3 archs did not look right.
Quoted 7 lines
Yes, unlike
virt_to_pfn()which still has a caller in openrisc (among csky, Hexagon, openrisc),pfn_to_virt()now does not have a caller in the 3 archs. Though bothvirt_to_pfn()andpfn_to_virt()are referenced inasm-generic/page.h, I also not sure if we need to remove theasm-generic/page.hwhich may serve as a template to future archs ?So, either way looks good to me :)
I think it's fair to assume we won't need asm-generic/page.h any more, as we likely won't be adding new NOMMU architectures. I can have a look myself at removing any such unused headers in include/asm-generic/, it's probably not the only one.
Can you just send a patch to remove the unused pfn_to_virt() functions?
Arnd
Hi Arnd,
Quoted 2 lines · Arnd Bergmann
I think it's fair to assume we won't need
asm-generic/page.hany more, as we likely won't be adding new NOMMU architectures.
So you think riscv-nommu (k210) was the last one we will ever see?
Gr{oetje,eeting}s,
Geert
Signature · 5 lines
Geert Uytterhoeven -- There's lots of Linux beyond ia32 -- [email protected] In personal conversations with technical people, I call myself a hacker. But when I'm talking to journalists I just say "programmer" or something like that. -- Linus Torvalds
Arnd Bergmann
maintainerQuoted 7 lines · Geert Uytterhoeven
Hi Arnd,
Quoted 2 lines · Arnd Bergmann
I think it's fair to assume we won't need
asm-generic/page.hany more, as we likely won't be adding new NOMMU architectures.So you think riscv-nommu (k210) was the last one we will ever see?
Yes. We've already removed half of the nommu architectures (blackfin, avr32, h8300, m32r, microblaze-nommu) over the past couple of years, and the remaining ones are pretty much only there to support existing users.
The only platform one that I see getting real work is esp32 [1], but that is not a new architecture.
Arnd
[1] https://github.com/jcmvbkbc/linux-xtensa/tree/xtensa-6.8-rc2-esp32
Quoted 23 lines · Arnd Bergmann
Quoted 14 lines · Yan Zhao
Quoted 5 lines · Arnd Bergmann
On Wed, Jan 31, 2024, at 06:51, Yan Zhao wrote:
How exactly did you notice the function being wrong, did you try to add a user somewhere, or just read through the code?
I came across them when I was debugging an unexpected kernel page fault on x86, and I was not sure whether
page_to_virt()was compiled inasm-generic/page.horlinux/mm.h.
Though finally, it turned out that the one inlinux/mm.hwas used, which yielded the right result and the unexpected kernel page fault in my case was not related topage_to_virt(), it did lead me to noticing that thepfn_to_virt()inasm-generic/page.hand other 3 archs did not look right.Quoted 7 lines
Yes, unlike
virt_to_pfn()which still has a caller in openrisc (among csky, Hexagon, openrisc),pfn_to_virt()now does not have a caller in the 3 archs. Though bothvirt_to_pfn()andpfn_to_virt()are referenced inasm-generic/page.h, I also not sure if we need to remove theasm-generic/page.hwhich may serve as a template to future archs ?So, either way looks good to me :)
Quoted 7 lines
I think it's fair to assume we won't need
asm-generic/page.hany more, as we likely won't be adding new NOMMU architectures. I can have a look myself at removing any such unused headers ininclude/asm-generic/, it's probably not the only one.Can you just send a patch to remove the unused
pfn_to_virt()functions?
Ok. I'll do it!
BTW: do you think it's also good to keep this fixing series though we'll remove the unused function later?
So if people want to revert the removal some day, they can get right one.
Or maybe I'm just paranoid, and explanation about the fix in the commit message of patch for function removal is enough.
What's your preference? :)
Arnd Bergmann
maintainerQuoted 14 lines · Yan Zhao
Quoted 8 lines · Arnd Bergmann
I think it's fair to assume we won't need
asm-generic/page.hany more, as we likely won't be adding new NOMMU architectures. I can have a look myself at removing any such unused headers ininclude/asm-generic/, it's probably not the only one.Can you just send a patch to remove the unused
pfn_to_virt()functions?Ok. I'll do it!
BTW: do you think it's also good to keep this fixing series though we'll remove the unused function later?
So if people want to revert the removal some day, they can get right one.
Quoted 4 lines
Or maybe I'm just paranoid, and explanation about the fix in the commit message of patch for function removal is enough.
What's your preference? :)
I would just remove it, there is no point in having both pfn_to_kaddr() and pfn_to_virt() when they do the exact same thing aside from this bug.
Just do a single patch for all architectures, no need to have three or four identical ones when I'm going to merge them all through the same tree anyway.
Just make sure you explain in the changelog what the bug was and how you noticed it, in case anyone is ever tempted to bring the function back.
Arnd
Quoted 24 lines · Arnd Bergmann
Quoted 14 lines · Yan Zhao
Quoted 8 lines · Arnd Bergmann
I think it's fair to assume we won't need
asm-generic/page.hany more, as we likely won't be adding new NOMMU architectures. I can have a look myself at removing any such unused headers ininclude/asm-generic/, it's probably not the only one.Can you just send a patch to remove the unused
pfn_to_virt()functions?Ok. I'll do it!
BTW: do you think it's also good to keep this fixing series though we'll remove the unused function later?
So if people want to revert the removal some day, they can get right one.Quoted 4 lines
Or maybe I'm just paranoid, and explanation about the fix in the commit message of patch for function removal is enough.
What's your preference? :)
I would just remove it, there is no point in having both
pfn_to_kaddr()andpfn_to_virt()when they do the exact same thing aside from this bug.
Quoted 7 lines
Just do a single patch for all architectures, no need to have three or four identical ones when I'm going to merge them all through the same tree anyway.
Just make sure you explain in the changelog what the bug was and how you noticed it, in case anyone is ever tempted to bring the function back.
Done.
https://lore.kernel.org/all/[email protected]
Thanks for you guidance :)