| Message ID | 20191010172247.19033-1-crecklin@redhat.com (mailing list archive) |
|---|---|
| State | Mainlined |
| Commit | 86109a691a454e08cbe0356400268cb2a81f1997 |
| Headers | show |
| Series | [v2] arm64: fix kcore macros 52-bit va fallout | expand |
Hi Chris, On 10/10/2019 18:22, Chris von Recklinghausen wrote: > We export the entire kernel address space (i.e. the whole of the TTBR1 > address range) via /proc/kcore. The kc_vaddr_to_offset() and > kc_offset_to_vaddr() macros are intended to convert between a kernel > virtual address and its offset relative to the start of the TTBR1 > address space. > > Prior to commit: > > 14c127c957c1c607 ("arm64: mm: Flip kernel VA space") > > ... the offset was calculated relative to VA_START, which at the time > was the start of the TTBR1 address space. At this time, PAGE_OFFSET > pointed to the high half of the TTBR1 address space where arm64's > linear map lived. > > That commit swapped the position of VA_START and PAGE_OFFSET, but > failed to update kc_vaddr_to_offset() or kc_offset_to_vaddr(), so > since then the two macros behave incorrectly. > > Note that VA_START was subsequently renamed to PAGE_END in commit: > > 77ad4ce69321abbe ("arm64: memory: rename VA_START to PAGE_END") > > As the generic implementations of the two macros calculate the offset > relative to PAGE_OFFSET (which is now the start of the TTBR1 address > space), we can delete the arm64 implementation and use those. Reviewed-by: James Morse <james.morse@arm.com> I think you can pick up Mark's R-B tag too: https://lore.kernel.org/r/20191010165537.GA27584@lakrids.cambridge.arm.com Thanks, James
On Thu, Oct 10, 2019 at 01:22:47PM -0400, Chris von Recklinghausen wrote: > We export the entire kernel address space (i.e. the whole of the TTBR1 > address range) via /proc/kcore. The kc_vaddr_to_offset() and > kc_offset_to_vaddr() macros are intended to convert between a kernel > virtual address and its offset relative to the start of the TTBR1 > address space. > > Prior to commit: > > 14c127c957c1c607 ("arm64: mm: Flip kernel VA space") > > ... the offset was calculated relative to VA_START, which at the time > was the start of the TTBR1 address space. At this time, PAGE_OFFSET > pointed to the high half of the TTBR1 address space where arm64's > linear map lived. > > That commit swapped the position of VA_START and PAGE_OFFSET, but > failed to update kc_vaddr_to_offset() or kc_offset_to_vaddr(), so > since then the two macros behave incorrectly. > > Note that VA_START was subsequently renamed to PAGE_END in commit: > > 77ad4ce69321abbe ("arm64: memory: rename VA_START to PAGE_END") > > As the generic implementations of the two macros calculate the offset > relative to PAGE_OFFSET (which is now the start of the TTBR1 address > space), we can delete the arm64 implementation and use those. > > Fixes: 14c127c957c1c607 ("arm64: mm: Flip kernel VA space") > > Signed-off-by: Chris von Recklinghausen <crecklin@redhat.com> > --- > v1 -> v2 > reword commit comment to to what Mark Rutland suggested > add Signed-off-by > no code changes > > arch/arm64/include/asm/pgtable.h | 3 --- > 1 file changed, 3 deletions(-) Thanks. Queued as a fix with James and Mark's reviewed-by tags. Will
diff --git a/arch/arm64/include/asm/pgtable.h b/arch/arm64/include/asm/pgtable.h index 7576df00eb50..8330810f699e 100644 --- a/arch/arm64/include/asm/pgtable.h +++ b/arch/arm64/include/asm/pgtable.h @@ -876,9 +876,6 @@ static inline void update_mmu_cache(struct vm_area_struct *vma, #define update_mmu_cache_pmd(vma, address, pmd) do { } while (0) -#define kc_vaddr_to_offset(v) ((v) & ~PAGE_END) -#define kc_offset_to_vaddr(o) ((o) | PAGE_END) - #ifdef CONFIG_ARM64_PA_BITS_52 #define phys_to_ttbr(addr) (((addr) | ((addr) >> 46)) & TTBR_BADDR_MASK_52) #else
We export the entire kernel address space (i.e. the whole of the TTBR1 address range) via /proc/kcore. The kc_vaddr_to_offset() and kc_offset_to_vaddr() macros are intended to convert between a kernel virtual address and its offset relative to the start of the TTBR1 address space. Prior to commit: 14c127c957c1c607 ("arm64: mm: Flip kernel VA space") ... the offset was calculated relative to VA_START, which at the time was the start of the TTBR1 address space. At this time, PAGE_OFFSET pointed to the high half of the TTBR1 address space where arm64's linear map lived. That commit swapped the position of VA_START and PAGE_OFFSET, but failed to update kc_vaddr_to_offset() or kc_offset_to_vaddr(), so since then the two macros behave incorrectly. Note that VA_START was subsequently renamed to PAGE_END in commit: 77ad4ce69321abbe ("arm64: memory: rename VA_START to PAGE_END") As the generic implementations of the two macros calculate the offset relative to PAGE_OFFSET (which is now the start of the TTBR1 address space), we can delete the arm64 implementation and use those. Fixes: 14c127c957c1c607 ("arm64: mm: Flip kernel VA space") Signed-off-by: Chris von Recklinghausen <crecklin@redhat.com> --- v1 -> v2 reword commit comment to to what Mark Rutland suggested add Signed-off-by no code changes arch/arm64/include/asm/pgtable.h | 3 --- 1 file changed, 3 deletions(-)