Re: [PATCH AUTOSEL 4.20 60/81] arm64: ptdump: Don't iterate kernel page tables using PTRS_PER_PXX

From: Will Deacon
Date: Thu Feb 28 2019 - 10:14:49 EST


On Thu, Feb 28, 2019 at 10:07:52AM -0500, Sasha Levin wrote:
> From: Will Deacon <will.deacon@xxxxxxx>
>
> [ Upstream commit d23c808c6fc6132e812690648e14c0d6b0cbe273 ]
>
> When 52-bit virtual addressing is enabled for userspace
> (CONFIG_ARM64_USER_VA_BITS_52=y), the kernel continues to utilise 48-bit
> virtual addressing in TTBR1. Consequently, PTRS_PER_PGD reflects the
> larger page table size for userspace and the pgd pointer for kernel page
> tables is offset before being written to TTBR1.
>
> This means that we can't use PTRS_PER_PGD to iterate over kernel page
> tables unless we apply the same offset, which is fiddly to get right and
> leads to some non-idiomatic walking code. Instead, just follow the usual
> pattern when walking page tables by using a while loop driven by
> pXd_offset() and pXd_addr_end().
>
> Reported-by: Qian Cai <cai@xxxxxx>
> Tested-by: Qian Cai <cai@xxxxxx>
> Acked-by: Steve Capper <steve.capper@xxxxxxx>
> Tested-by: Steve Capper <steve.capper@xxxxxxx>
> Signed-off-by: Will Deacon <will.deacon@xxxxxxx>
> Signed-off-by: Sasha Levin <sashal@xxxxxxxxxx>
> ---
> arch/arm64/mm/dump.c | 59 ++++++++++++++++++++++----------------------
> 1 file changed, 29 insertions(+), 30 deletions(-)

4.20 doesn't support 52-bit VAs, so this isn't needed there afaict (and is
why I didn't tag the commit for stable in the first place).

Will