I'm sorry, it seems I am mistaken. This was in reference to patches which
weren't committed.
Please see:
http://driverdev.linuxdriverproject.org/pipermail/driverdev-devel/2014-October/060203.html
for the patch with this specific mask used.
The reason __GFP_HIGHMEM may have been chosen is because in the chain of
calls
to ion_heap_shrink_count(), ion_system_heap_shrink(),
ion_page_pool_shrink(),
the mask is only ever compared against __GFP_HIGHMEM.
From drivers/staging/android/ion/ion_page_pool.c:
122 if (current_is_kswapd())
123 high = true;
124 else
125 high = !!(gfp_mask & __GFP_HIGHMEM);
Kind Regards,
Derek Yerger
On 02/22/2016 02:18 PM, Laura Abbott wrote:
On 02/20/2016 07:15 PM, Derek Yerger wrote:
From drivers/staging/android TODO file - sparse errors.
The current code attempts assignment of -1 to unsigned type gfp_t.
Assignment should be an enumerated type of GFP_KERNEL, GFP_ATOMIC,
GFP_HIGHMEM, or __GFP_HIGH.
The original 2014 patch by Gioh Kim adding debugfs support to
android/ion
used __GFP_HIGHMEM as the mask, but was removed in a subsequent revert.
This patch removes the sparse errors, and fixes the assignment as noted
above.
Please add commit references for the patches you are referring to.
At least in the upstream staging tree the GFP mask has been -1 since
it was
initially added (ea313b5f8 "gpu: ion: Also shrink memory cached in the
deferred
free list") and I can't find the commits from Gioh Kim you are
referring to.
Signed-off-by: Derek Yerger <dy@xxxxxxxxxx>
---
diff --git a/drivers/staging/android/ion/ion.c
b/drivers/staging/android/ion/ion.c
index c97e82b..a9ca46f 100644
--- a/drivers/staging/android/ion/ion.c
+++ b/drivers/staging/android/ion/ion.c
@@ -1524,7 +1524,7 @@ static int debug_shrink_set(void *data, u64 val)
struct shrink_control sc;
int objs;
- sc.gfp_mask = -1;
+ sc.gfp_mask = __GFP_HIGHMEM;
sc.nr_to_scan = val;
Using just __GFP_HIGHMEM is a bit weird for a GFP mask. I'd rather see
this
be something like GFP_HIGHUSER to represent what an actual allocation
might
be or add a comment explaining why just __GFP_HIGHMEM is okay.
Thanks,
Laura
if (!val) {
@@ -1542,7 +1542,7 @@ static int debug_shrink_get(void *data, u64 *val)
struct shrink_control sc;
int objs;
- sc.gfp_mask = -1;
+ sc.gfp_mask = __GFP_HIGHMEM;
sc.nr_to_scan = 0;
objs = heap->shrinker.count_objects(&heap->shrinker, &sc);
_______________________________________________
devel mailing list
devel@xxxxxxxxxxxxxxxxxxxxxx
http://driverdev.linuxdriverproject.org/mailman/listinfo/driverdev-devel
_______________________________________________
devel mailing list
devel@xxxxxxxxxxxxxxxxxxxxxx
http://driverdev.linuxdriverproject.org/mailman/listinfo/driverdev-devel
