Re: [PATCH v10 12/22] kasan, arm64: fix up fault handling logic

[Andrey Konovalov]

On Thu, Nov 8, 2018 at 1:22 PM, Mark Rutland <mark.rutland@xxxxxxx> wrote:
> On Tue, Nov 06, 2018 at 06:30:27PM +0100, Andrey Konovalov wrote:
>> show_pte in arm64 fault handling relies on the fact that the top byte of
>> a kernel pointer is 0xff, which isn't always the case with tag-based
>> KASAN.
>
> That's for the TTBR1 check, right?
>
> i.e. for the following to work:
>
>         if (addr >= VA_START)
>
> ... we need the tag bits to be an extension of bit 55...
>
>>
>> This patch resets the top byte in show_pte.
>>
>> Reviewed-by: Andrey Ryabinin <aryabinin@xxxxxxxxxxxxx>
>> Reviewed-by: Dmitry Vyukov <dvyukov@xxxxxxxxxx>
>> Signed-off-by: Andrey Konovalov <andreyknvl@xxxxxxxxxx>
>> ---
>>  arch/arm64/mm/fault.c | 3 +++
>>  1 file changed, 3 insertions(+)
>>
>> diff --git a/arch/arm64/mm/fault.c b/arch/arm64/mm/fault.c
>> index 7d9571f4ae3d..d9a84d6f3343 100644
>> --- a/arch/arm64/mm/fault.c
>> +++ b/arch/arm64/mm/fault.c
>> @@ -32,6 +32,7 @@
>>  #include <linux/perf_event.h>
>>  #include <linux/preempt.h>
>>  #include <linux/hugetlb.h>
>> +#include <linux/kasan.h>
>>
>>  #include <asm/bug.h>
>>  #include <asm/cmpxchg.h>
>> @@ -141,6 +142,8 @@ void show_pte(unsigned long addr)
>>       pgd_t *pgdp;
>>       pgd_t pgd;
>>
>> +     addr = (unsigned long)kasan_reset_tag((void *)addr);
>
> ... but this ORs in (0xffUL << 56), which is not correct for addresses
> which aren't TTBR1 addresses to begin with, where bit 55 is clear, and
> throws away useful information.
>
> We could use untagged_addr() here, but that wouldn't be right for
> kernels which don't use TBI1, and we'd erroneously report addresses
> under the TTBR1 range as being in the TTBR1 range.
>
> I also see that the entry assembly for el{1,0}_{da,ia} clears the tag
> for EL0 addresses.
>
> So we could have:
>
> static inline bool is_ttbr0_addr(unsigned long addr)
> {
>         /* entry assembly clears tags for TTBR0 addrs */
>         return addr < TASK_SIZE_64;
> }
>
> static inline bool is_ttbr1_addr(unsigned long addr)
> {
>         /* TTBR1 addresses may have a tag if HWKASAN is in use */
>         return arch_kasan_reset_tag(addr) >= VA_START;
> }
>
> ... and use those in the conditionals, leaving the addr as-is for
> reporting purposes.

Actually it looks like 276e9327 ("arm64: entry: improve data abort
handling of tagged pointers") already takes care of both user and
kernel fault addresses and correctly removes tags from them. So I
think we need to drop this patch.