The cma_mutex which protects alloc_contig_range() was first appeared in
commit 7ee793a62fa8c ("cma: Remove potential deadlock situation"),
at that time, there is no guarantee the behavior of concurrency inside
alloc_contig_range().
After the commit 2c7452a075d4db2dc
("mm/page_isolation.c: make start_isolate_page_range() fail if already isolated")
> However, two subsystems (CMA and gigantic
> huge pages for example) could attempt operations on the same range. If
> this happens, one thread may 'undo' the work another thread is doing.
> This can result in pageblocks being incorrectly left marked as
> MIGRATE_ISOLATE and therefore not available for page allocation.
The concurrency inside alloc_contig_range() was clarified.
Now we can find that hugepage and virtio call alloc_contig_range() without
any lock, thus cma_mutex is "redundant" in cma_alloc() now.
Signed-off-by: Lecopzer Chen <[email protected]>
---
mm/cma.c | 4 +---
1 file changed, 1 insertion(+), 3 deletions(-)
diff --git a/mm/cma.c b/mm/cma.c
index 7f415d7cda9f..3692a34e2353 100644
--- a/mm/cma.c
+++ b/mm/cma.c
@@ -38,7 +38,6 @@
struct cma cma_areas[MAX_CMA_AREAS];
unsigned cma_area_count;
-static DEFINE_MUTEX(cma_mutex);
phys_addr_t cma_get_base(const struct cma *cma)
{
@@ -454,10 +453,9 @@ struct page *cma_alloc(struct cma *cma, size_t count, unsigned int align,
mutex_unlock(&cma->lock);
pfn = cma->base_pfn + (bitmap_no << cma->order_per_bit);
- mutex_lock(&cma_mutex);
ret = alloc_contig_range(pfn, pfn + count, MIGRATE_CMA,
GFP_KERNEL | (no_warn ? __GFP_NOWARN : 0));
- mutex_unlock(&cma_mutex);
+
if (ret == 0) {
page = pfn_to_page(pfn);
break;
--
2.18.0
On 20.10.20 12:22, Lecopzer Chen wrote:
> The cma_mutex which protects alloc_contig_range() was first appeared in
> commit 7ee793a62fa8c ("cma: Remove potential deadlock situation"),
> at that time, there is no guarantee the behavior of concurrency inside
> alloc_contig_range().
>
> After the commit 2c7452a075d4db2dc
> ("mm/page_isolation.c: make start_isolate_page_range() fail if already isolated")
> > However, two subsystems (CMA and gigantic
> > huge pages for example) could attempt operations on the same range. If
> > this happens, one thread may 'undo' the work another thread is doing.
> > This can result in pageblocks being incorrectly left marked as
> > MIGRATE_ISOLATE and therefore not available for page allocation.
> The concurrency inside alloc_contig_range() was clarified.
>
> Now we can find that hugepage and virtio call alloc_contig_range() without
> any lock, thus cma_mutex is "redundant" in cma_alloc() now.
>
> Signed-off-by: Lecopzer Chen <[email protected]>
> ---
> mm/cma.c | 4 +---
> 1 file changed, 1 insertion(+), 3 deletions(-)
>
> diff --git a/mm/cma.c b/mm/cma.c
> index 7f415d7cda9f..3692a34e2353 100644
> --- a/mm/cma.c
> +++ b/mm/cma.c
> @@ -38,7 +38,6 @@
>
> struct cma cma_areas[MAX_CMA_AREAS];
> unsigned cma_area_count;
> -static DEFINE_MUTEX(cma_mutex);
>
> phys_addr_t cma_get_base(const struct cma *cma)
> {
> @@ -454,10 +453,9 @@ struct page *cma_alloc(struct cma *cma, size_t count, unsigned int align,
> mutex_unlock(&cma->lock);
>
> pfn = cma->base_pfn + (bitmap_no << cma->order_per_bit);
> - mutex_lock(&cma_mutex);
> ret = alloc_contig_range(pfn, pfn + count, MIGRATE_CMA,
> GFP_KERNEL | (no_warn ? __GFP_NOWARN : 0));
> - mutex_unlock(&cma_mutex);
> +
> if (ret == 0) {
> page = pfn_to_page(pfn);
> break;
>
I guess this is fine. In case there is a race we return with -EBUSY -
which is suboptimal (as it could just be a temporary issue if the other
user backs off), but should be good enough for now.
Acked-by: David Hildenbrand <[email protected]>
--
Thanks,
David / dhildenb
On 10/20/20 1:27 PM, David Hildenbrand wrote:
> On 20.10.20 12:22, Lecopzer Chen wrote:
>> The cma_mutex which protects alloc_contig_range() was first appeared in
>> commit 7ee793a62fa8c ("cma: Remove potential deadlock situation"),
>> at that time, there is no guarantee the behavior of concurrency inside
>> alloc_contig_range().
>>
>> After the commit 2c7452a075d4db2dc
>> ("mm/page_isolation.c: make start_isolate_page_range() fail if already isolated")
>> > However, two subsystems (CMA and gigantic
>> > huge pages for example) could attempt operations on the same range. If
>> > this happens, one thread may 'undo' the work another thread is doing.
>> > This can result in pageblocks being incorrectly left marked as
>> > MIGRATE_ISOLATE and therefore not available for page allocation.
>> The concurrency inside alloc_contig_range() was clarified.
>>
>> Now we can find that hugepage and virtio call alloc_contig_range() without
>> any lock, thus cma_mutex is "redundant" in cma_alloc() now.
>>
>> Signed-off-by: Lecopzer Chen <[email protected]>
>> ---
>> mm/cma.c | 4 +---
>> 1 file changed, 1 insertion(+), 3 deletions(-)
>>
>> diff --git a/mm/cma.c b/mm/cma.c
>> index 7f415d7cda9f..3692a34e2353 100644
>> --- a/mm/cma.c
>> +++ b/mm/cma.c
>> @@ -38,7 +38,6 @@
>>
>> struct cma cma_areas[MAX_CMA_AREAS];
>> unsigned cma_area_count;
>> -static DEFINE_MUTEX(cma_mutex);
>>
>> phys_addr_t cma_get_base(const struct cma *cma)
>> {
>> @@ -454,10 +453,9 @@ struct page *cma_alloc(struct cma *cma, size_t count, unsigned int align,
>> mutex_unlock(&cma->lock);
>>
>> pfn = cma->base_pfn + (bitmap_no << cma->order_per_bit);
>> - mutex_lock(&cma_mutex);
>> ret = alloc_contig_range(pfn, pfn + count, MIGRATE_CMA,
>> GFP_KERNEL | (no_warn ? __GFP_NOWARN : 0));
>> - mutex_unlock(&cma_mutex);
>> +
>> if (ret == 0) {
>> page = pfn_to_page(pfn);
>> break;
>>
>
> I guess this is fine. In case there is a race we return with -EBUSY -
> which is suboptimal (as it could just be a temporary issue if the other
> user backs off), but should be good enough for now.
Agreed.
> Acked-by: David Hildenbrand <[email protected]>
Acked-by: Vlastimil Babka <[email protected]>