Re: [PATCH] arm64: mm: reset address tag set by kasan sw tagging
Hi Catalin, On 6/10/2020 5:06 PM, Catalin Marinas wrote: On Wed, Jun 10, 2020 at 04:39:44PM +0530, Shyam Thombre wrote: KASAN sw tagging sets a random tag of 8 bits in the top byte of the pointer returned by the memory allocating functions. So for the functions unaware of this change, the top 8 bits of the address must be reset which is done by the function arch_kasan_reset_tag(). Signed-off-by: Shyam Thombre --- arch/arm64/mm/mmu.c | 1 + 1 file changed, 1 insertion(+) diff --git a/arch/arm64/mm/mmu.c b/arch/arm64/mm/mmu.c index e7fbc62..eae7655 100644 --- a/arch/arm64/mm/mmu.c +++ b/arch/arm64/mm/mmu.c @@ -723,6 +723,7 @@ int kern_addr_valid(unsigned long addr) pmd_t *pmdp, pmd; pte_t *ptep, pte; + addr = arch_kasan_reset_tag(addr); if long)addr) >> VA_BITS) != -1UL) return 0; It would be interesting to know what fails without this patch. The only user seems to be read_kcore() and, at a quick look, I don't see how it can generate tagged addresses. This issue is seen in downstream GPU drivers. It currently doesn't look to impact any upstream users. -- The Qualcomm Innovation Center, Inc. is a member of the Code Aurora Forum, a Linux Foundation Collaborative Project
Re: [PATCH] arm64: mm: reset address tag set by kasan sw tagging
On Wed, 10 Jun 2020 16:39:44 +0530, Shyam Thombre wrote: > KASAN sw tagging sets a random tag of 8 bits in the top byte of the pointer > returned by the memory allocating functions. So for the functions unaware > of this change, the top 8 bits of the address must be reset which is done > by the function arch_kasan_reset_tag(). Applied to arm64 (for-next/fixes), thanks! [1/1] arm64: mm: reset address tag set by kasan sw tagging https://git.kernel.org/arm64/c/8dd4daa04278 Cheers, -- Will https://fixes.arm64.dev https://next.arm64.dev https://will.arm64.dev
Re: [PATCH] arm64: mm: reset address tag set by kasan sw tagging
On Wed, Jun 10, 2020 at 04:39:44PM +0530, Shyam Thombre wrote: > KASAN sw tagging sets a random tag of 8 bits in the top byte of the pointer > returned by the memory allocating functions. So for the functions unaware > of this change, the top 8 bits of the address must be reset which is done > by the function arch_kasan_reset_tag(). > > Signed-off-by: Shyam Thombre > --- > arch/arm64/mm/mmu.c | 1 + > 1 file changed, 1 insertion(+) > > diff --git a/arch/arm64/mm/mmu.c b/arch/arm64/mm/mmu.c > index e7fbc62..eae7655 100644 > --- a/arch/arm64/mm/mmu.c > +++ b/arch/arm64/mm/mmu.c > @@ -723,6 +723,7 @@ int kern_addr_valid(unsigned long addr) > pmd_t *pmdp, pmd; > pte_t *ptep, pte; > > + addr = arch_kasan_reset_tag(addr); > if long)addr) >> VA_BITS) != -1UL) > return 0; It would be interesting to know what fails without this patch. The only user seems to be read_kcore() and, at a quick look, I don't see how it can generate tagged addresses. -- Catalin