On 08/31/2018 09:52 AM, David Hildenbrand wrote:
> On 31.08.2018 08:20, Janosch Frank wrote:
>> On 30.08.2018 16:18, David Hildenbrand wrote:
>>> On 30.08.2018 16:15, Janosch Frank wrote:
>>>> We have to do down_write on the mm semaphore to set a bitfield in the
>>>> mm context.
>>>>
>>>> Signed-off-by: Janosch Frank <frankja@xxxxxxxxxxxxx>
>>>> Fixes: a4499382 ("KVM: s390: Add huge page enablement control")
>>>> ---
>>>> arch/s390/include/asm/mmu.h | 8 +++++++-
>>>> arch/s390/kvm/kvm-s390.c | 2 ++
>>>> 2 files changed, 9 insertions(+), 1 deletion(-)
>>>>
>>>> diff --git a/arch/s390/include/asm/mmu.h b/arch/s390/include/asm/mmu.h
>>>> index f31a15044c24..a8418e1379eb 100644
>>>> --- a/arch/s390/include/asm/mmu.h
>>>> +++ b/arch/s390/include/asm/mmu.h
>>>> @@ -16,7 +16,13 @@ typedef struct {
>>>> unsigned long asce;
>>>> unsigned long asce_limit;
>>>> unsigned long vdso_base;
>>>> - /* The mmu context allocates 4K page tables. */
>>>> + /*
>>>> + * The following bitfields need a down_write on the mm
>>>> + * semaphore when they are written to. As they are only
>>>> + * written once, they can be read without a lock.
>>>> + *
>>>> + * The mmu context allocates 4K page tables.
>>>> + */
>>>> unsigned int alloc_pgste:1;
>>>> /* The mmu context uses extended page tables. */
>>>> unsigned int has_pgste:1;
>>>> diff --git a/arch/s390/kvm/kvm-s390.c b/arch/s390/kvm/kvm-s390.c
>>>> index 8c039e12cb24..ac5da6b0b862 100644
>>>> --- a/arch/s390/kvm/kvm-s390.c
>>>> +++ b/arch/s390/kvm/kvm-s390.c
>>>> @@ -695,7 +695,9 @@ static int kvm_vm_ioctl_enable_cap(struct kvm *kvm, struct kvm_enable_cap *cap)
>>>> r = -EINVAL;
>>>> else {
>>>> r = 0;
>>>> + down_write(&kvm->mm->mmap_sem);
>>>> kvm->mm->context.allow_gmap_hpage_1m = 1;
>>>> + up_write(&kvm->mm->mmap_sem);
>>>> /*
>>>> * We might have to create fake 4k page
>>>> * tables. To avoid that the hardware works on
>>>>
>>>
>>> This is a theoretical scenario, right?
>>>
>>> (GMAP without vcpus will not be populated as far as I remember)
>>
>> Yes, currently highly unlikely.
>>
>> It's not the GMAP that bothers me here, it's the bitfield within the
>> context. If we are very unlucky, we could loose a write to it.
>
> Right, although I think this is also very theoretical. Never the less,
> this is certainly the right thing to do.
Yes. The compiler might even use the oi instruction (which is on all newer machines
atomic in terms of setting bits).
>
> (I am just wondering if this one is worth backporting, do you see the need?)
This also fixes a patch that was added after 4.18 so we should just add it for 4.19.
>>
>> The locking is a bit interesting. The context has a spinlock, but the
>> bitfield is protected by mm sem because for skey enablement and SIE
>> enablement we need to take it for mm operations anyhow.
>>
>>
>
>
