Hi Catalin,
On Wed, Dec 01, 2021 at 07:37:49PM +0000, Catalin Marinas wrote:
> With MTE, even if the pte allows an access, a mismatched tag somewhere
> within a page can still cause a fault. Select ARCH_HAS_SUBPAGE_FAULTS if
> MTE is enabled and implement the probe_subpage_*() functions. Note that
> get_user() is sufficient for the writeable checks since the same tag
> mismatch fault would be triggered by a read.
>
> Signed-off-by: Catalin Marinas <catalin.marinas@xxxxxxx>
> ---
> arch/arm64/Kconfig | 1 +
> arch/arm64/include/asm/uaccess.h | 59 ++++++++++++++++++++++++++++++++
> 2 files changed, 60 insertions(+)
>
> diff --git a/arch/arm64/Kconfig b/arch/arm64/Kconfig
> index c4207cf9bb17..dff89fd0d817 100644
> --- a/arch/arm64/Kconfig
> +++ b/arch/arm64/Kconfig
> @@ -1777,6 +1777,7 @@ config ARM64_MTE
> depends on AS_HAS_LSE_ATOMICS
> # Required for tag checking in the uaccess routines
> depends on ARM64_PAN
> + select ARCH_HAS_SUBPAGE_FAULTS
> select ARCH_USES_HIGH_VMA_FLAGS
> help
> Memory Tagging (part of the ARMv8.5 Extensions) provides
> diff --git a/arch/arm64/include/asm/uaccess.h b/arch/arm64/include/asm/uaccess.h
> index 6e2e0b7031ab..bcbd24b97917 100644
> --- a/arch/arm64/include/asm/uaccess.h
> +++ b/arch/arm64/include/asm/uaccess.h
> @@ -445,4 +445,63 @@ static inline int __copy_from_user_flushcache(void *dst, const void __user *src,
> }
> #endif
>
> +#ifdef CONFIG_ARCH_HAS_SUBPAGE_FAULTS
> +
> +/*
> + * Return 0 on success, the number of bytes not accessed otherwise.
> + */
> +static inline size_t __mte_probe_user_range(const char __user *uaddr,
> + size_t size, bool skip_first)
> +{
> + const char __user *end = uaddr + size;
> + int err = 0;
> + char val;
> +
> + uaddr = PTR_ALIGN_DOWN(uaddr, MTE_GRANULE_SIZE);
> + if (skip_first)
> + uaddr += MTE_GRANULE_SIZE;
Do we need the skipping for a functional reason, or is that an optimization?
>From the comments in probe_subpage_writeable() and
probe_subpage_safe_writeable() I wasn't sure if the skipping was because we
*don't need to* check the first granule, or because we *must not* check the
first granule.
> + while (uaddr < end) {
> + /*
> + * A read is sufficient for MTE, the caller should have probed
> + * for the pte write permission if required.
> + */
> + __raw_get_user(val, uaddr, err);
> + if (err)
> + return end - uaddr;
> + uaddr += MTE_GRANULE_SIZE;
> + }
I think we may need to account for the residue from PTR_ALIGN_DOWN(), or we can
report more bytes not copied than was passed in `size` in the first place,
which I think might confused some callers.
Consider MTE_GRANULE_SIZE is 16, uaddr is 31, and size is 1 (so end is 32). We
align uaddr down to 16, and if we fail the first access we return (32 - 16),
i.e. 16.
Thanks,
Mark.
> + (void)val;
> +
> + return 0;
> +}
> +
> +static inline size_t probe_subpage_writeable(const void __user *uaddr,
> + size_t size)
> +{
> + if (!system_supports_mte())
> + return 0;
> + /* first put_user() done in the caller */
> + return __mte_probe_user_range(uaddr, size, true);
> +}
> +
> +static inline size_t probe_subpage_safe_writeable(const void __user *uaddr,
> + size_t size)
> +{
> + if (!system_supports_mte())
> + return 0;
> + /* the caller used GUP, don't skip the first granule */
> + return __mte_probe_user_range(uaddr, size, false);
> +}
> +
> +static inline size_t probe_subpage_readable(const void __user *uaddr,
> + size_t size)
> +{
> + if (!system_supports_mte())
> + return 0;
> + /* first get_user() done in the caller */
> + return __mte_probe_user_range(uaddr, size, true);
> +}
> +
> +#endif /* CONFIG_ARCH_HAS_SUBPAGE_FAULTS */
> +
> #endif /* __ASM_UACCESS_H */
