On 08/14/2017 09:22 AM, Tycho Andersen wrote:
> On Sat, Aug 12, 2017 at 12:17:34PM +0100, Mark Rutland wrote:
>> Hi,
>>
>> On Fri, Aug 11, 2017 at 03:13:02PM -0600, Tycho Andersen wrote:
>>> On Fri, Aug 11, 2017 at 10:25:14AM -0700, Laura Abbott wrote:
>>>> On 08/09/2017 01:07 PM, Tycho Andersen wrote:
>>>>> @@ -190,7 +202,7 @@ static void init_pmd(pud_t *pud, unsigned long addr,
>>>>> unsigned long end,
>>>>> next = pmd_addr_end(addr, end);
>>>>>
>>>>> /* try section mapping first */
>>>>> - if (((addr | next | phys) & ~SECTION_MASK) == 0 &&
>>>>> + if (use_section_mapping(addr, next, phys) &&
>>>>> (flags & NO_BLOCK_MAPPINGS) == 0) {
>>>>> pmd_set_huge(pmd, phys, prot);
>>>>>
>>>>>
>>>>
>>>> There is already similar logic to disable section mappings for
>>>> debug_pagealloc at the start of map_mem, can you take advantage
>>>> of that?
>>>
>>> You're suggesting something like this instead? Seems to work fine.
>>>
>>> diff --git a/arch/arm64/mm/mmu.c b/arch/arm64/mm/mmu.c
>>> index 38026b3ccb46..3b2c17bbbf12 100644
>>> --- a/arch/arm64/mm/mmu.c
>>> +++ b/arch/arm64/mm/mmu.c
>>> @@ -434,6 +434,8 @@ static void __init map_mem(pgd_t *pgd)
>>>
>>> if (debug_pagealloc_enabled())
>>> flags = NO_BLOCK_MAPPINGS | NO_CONT_MAPPINGS;
>>> + if (IS_ENABLED(CONFIG_XPFO))
>>> + flags |= NO_BLOCK_MAPPINGS;
>>>
>>
>> IIUC, XPFO carves out individual pages just like DEBUG_PAGEALLOC, so you'll
>> also need NO_CONT_MAPPINGS.
>
> Yes, thanks!
>
> Tycho
>
Setting NO_CONT_MAPPINGS fixes the TLB conflict aborts I was seeing
on my machine.
Thanks,
Laura
