[PATCH 5/6] powerpc/mmu: drop mmap_sem now that locked_vm is atomic

[Christophe Leroy]

Le 02/04/2019 à 22:41, Daniel Jordan a écrit :
> With locked_vm now an atomic, there is no need to take mmap_sem as
> writer.  Delete and refactor accordingly.

Could you please detail the change ? It looks like this is not the only 
change. I'm wondering what the consequences are.

Before we did:
- lock
- calculate future value
- check the future value is acceptable
- update value if future value acceptable
- return error if future value non acceptable
- unlock

Now we do:
- atomic update with future (possibly too high) value
- check the new value is acceptable
- atomic update back with older value if new value not acceptable and 
return error

So if a concurrent action wants to increase locked_vm with an acceptable 
step while another one has temporarily set it too high, it will now fail.

I think we should keep the previous approach and do a cmpxchg after 
validating the new value.

Christophe

> 
> Signed-off-by: Daniel Jordan <daniel.m.jordan at oracle.com>
> Cc: Alexey Kardashevskiy <aik at ozlabs.ru>
> Cc: Andrew Morton <akpm at linux-foundation.org>
> Cc: Benjamin Herrenschmidt <benh at kernel.crashing.org>
> Cc: Christoph Lameter <cl at linux.com>
> Cc: Davidlohr Bueso <dave at stgolabs.net>
> Cc: Michael Ellerman <mpe at ellerman.id.au>
> Cc: Paul Mackerras <paulus at samba.org>
> Cc: <linux-mm at kvack.org>
> Cc: <linuxppc-dev at lists.ozlabs.org>
> Cc: <linux-kernel at vger.kernel.org>
> ---
>   arch/powerpc/mm/mmu_context_iommu.c | 27 +++++++++++----------------
>   1 file changed, 11 insertions(+), 16 deletions(-)
> 
> diff --git a/arch/powerpc/mm/mmu_context_iommu.c b/arch/powerpc/mm/mmu_context_iommu.c
> index 8038ac24a312..a4ef22b67c07 100644
> --- a/arch/powerpc/mm/mmu_context_iommu.c
> +++ b/arch/powerpc/mm/mmu_context_iommu.c
> @@ -54,34 +54,29 @@ struct mm_iommu_table_group_mem_t {
>   static long mm_iommu_adjust_locked_vm(struct mm_struct *mm,
>   		unsigned long npages, bool incr)
>   {
> -	long ret = 0, locked, lock_limit;
> +	long ret = 0;
> +	unsigned long lock_limit;
>   	s64 locked_vm;
>   
>   	if (!npages)
>   		return 0;
>   
> -	down_write(&mm->mmap_sem);
> -	locked_vm = atomic64_read(&mm->locked_vm);
>   	if (incr) {
> -		locked = locked_vm + npages;
>   		lock_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT;
> -		if (locked > lock_limit && !capable(CAP_IPC_LOCK))
> +		locked_vm = atomic64_add_return(npages, &mm->locked_vm);
> +		if (locked_vm > lock_limit && !capable(CAP_IPC_LOCK)) {
>   			ret = -ENOMEM;
> -		else
> -			atomic64_add(npages, &mm->locked_vm);
> +			atomic64_sub(npages, &mm->locked_vm);
> +		}
>   	} else {
> -		if (WARN_ON_ONCE(npages > locked_vm))
> -			npages = locked_vm;
> -		atomic64_sub(npages, &mm->locked_vm);
> +		locked_vm = atomic64_sub_return(npages, &mm->locked_vm);
> +		WARN_ON_ONCE(locked_vm < 0);
>   	}
>   
> -	pr_debug("[%d] RLIMIT_MEMLOCK HASH64 %c%ld %ld/%ld\n",
> -			current ? current->pid : 0,
> -			incr ? '+' : '-',
> -			npages << PAGE_SHIFT,
> -			atomic64_read(&mm->locked_vm) << PAGE_SHIFT,
> +	pr_debug("[%d] RLIMIT_MEMLOCK HASH64 %c%lu %lld/%lu\n",
> +			current ? current->pid : 0, incr ? '+' : '-',
> +			npages << PAGE_SHIFT, locked_vm << PAGE_SHIFT,
>   			rlimit(RLIMIT_MEMLOCK));
> -	up_write(&mm->mmap_sem);
>   
>   	return ret;
>   }
>