Using order 4 pages would be helpful for many IOMMUs, but it could spend
quite much time in page allocation perspective.
The order 4 allocation with __GFP_RECLAIM may spend much time in
reclaim and compation logic. __GFP_NORETRY also may affect. These cause
unpredictable delay.
To get reasonable allocation speed from dma-buf system heap, use
HIGH_ORDER_GFP for order 4 to avoid reclaim.
Signed-off-by: Jaewon Kim <[email protected]>
---
drivers/dma-buf/heaps/system_heap.c | 3 +--
1 file changed, 1 insertion(+), 2 deletions(-)
diff --git a/drivers/dma-buf/heaps/system_heap.c b/drivers/dma-buf/heaps/system_heap.c
index e8bd10e60998..5a405e99ef1e 100644
--- a/drivers/dma-buf/heaps/system_heap.c
+++ b/drivers/dma-buf/heaps/system_heap.c
@@ -42,11 +42,10 @@ struct dma_heap_attachment {
};
#define LOW_ORDER_GFP (GFP_HIGHUSER | __GFP_ZERO | __GFP_COMP)
-#define MID_ORDER_GFP (LOW_ORDER_GFP | __GFP_NOWARN)
#define HIGH_ORDER_GFP (((GFP_HIGHUSER | __GFP_ZERO | __GFP_NOWARN \
| __GFP_NORETRY) & ~__GFP_RECLAIM) \
| __GFP_COMP)
-static gfp_t order_flags[] = {HIGH_ORDER_GFP, MID_ORDER_GFP, LOW_ORDER_GFP};
+static gfp_t order_flags[] = {HIGH_ORDER_GFP, HIGH_ORDER_GFP, LOW_ORDER_GFP};
/*
* The selection of the orders used for allocation (1MB, 64K, 4K) is designed
* to match with the sizes often found in IOMMUs. Using order 4 pages instead
--
2.17.1
>On Tue, Jan 17, 2023 at 12:31 AM Jaewon Kim <[email protected]> wrote:
>> > Using order 4 pages would be helpful for many IOMMUs, but it could spend
>> > quite much time in page allocation perspective.
>> >
>> > The order 4 allocation with __GFP_RECLAIM may spend much time in
>> > reclaim and compation logic. __GFP_NORETRY also may affect. These cause
>> > unpredictable delay.
>> >
>> > To get reasonable allocation speed from dma-buf system heap, use
>> > HIGH_ORDER_GFP for order 4 to avoid reclaim.
>
>Thanks for sharing this!
>The case where the allocation gets stuck behind reclaim under pressure
>does sound undesirable, but I'd be a bit hesitant to tweak numbers
>that have been used for a long while (going back to ion) without a bit
>more data.
>
>It might be good to also better understand the tradeoff of potential
>on-going impact to performance from using low order pages when the
>buffer is used. Do you have any details like or tests that you could
>share to help ensure this won't impact other users?
>
>TJ: Do you have any additional thoughts on this?
>
>thanks
>-john
Let me share what I tested with trace events of compaction, direct_reclaim.
This log was captured within one dma-buf alloc for 1,957,888 byte size.
It shows direct reclaim spent 4.973 msec total so that it could reclaim 209 pages total.
But it spent 12.606 msec in compaction.
I might miss preempted time but I think compaction is quite time consuming job.
Sorry I don't know how much the current gfp is helpful to IOMMU.
I think the default should be set like my change to improve allocation,
and the current gfp could be set if need through other options like cmdline.
-5972 [006] .... 7971.248176: mm_compaction_begin: zone_start=0x40000 migrate_pfn=0x40000 free_pfn=0xffc00 zone_end=0x100000, mode=async
-5972 [006] .... 7971.248394: mm_compaction_end: zone_start=0x40000 migrate_pfn=0xac800 free_pfn=0xffc00 zone_end=0x100000, mode=async status=contended
-5972 [006] .... 7971.248399: mm_vmscan_direct_reclaim_begin: order=4 gfp_flags=GFP_HIGHUSER|__GFP_NOWARN|__GFP_COMP|__GFP_ZERO
-5972 [006] .... 7971.248922: mm_vmscan_direct_reclaim_end: nr_reclaimed=17
-5972 [006] .... 7971.248974: mm_compaction_begin: zone_start=0x40000 migrate_pfn=0xacc00 free_pfn=0xffc00 zone_end=0x100000, mode=async
-5972 [006] .... 7971.249063: mm_compaction_end: zone_start=0x40000 migrate_pfn=0xb5c00 free_pfn=0xffc00 zone_end=0x100000, mode=async status=contended
-5972 [006] .... 7971.249067: mm_vmscan_direct_reclaim_begin: order=4 gfp_flags=GFP_HIGHUSER|__GFP_NOWARN|__GFP_COMP|__GFP_ZERO
-5972 [006] .... 7971.249768: mm_vmscan_direct_reclaim_end: nr_reclaimed=18
-5972 [006] .... 7971.249843: mm_compaction_begin: zone_start=0x40000 migrate_pfn=0xb6000 free_pfn=0xafc00 zone_end=0x100000, mode=async
-5972 [006] .... 7971.249845: mm_compaction_end: zone_start=0x40000 migrate_pfn=0xb6000 free_pfn=0xafc00 zone_end=0x100000, mode=async status=partial_skipped
-5972 [006] .... 7971.249849: mm_vmscan_direct_reclaim_begin: order=4 gfp_flags=GFP_HIGHUSER|__GFP_NOWARN|__GFP_COMP|__GFP_ZERO
-5972 [006] .... 7971.250116: mm_vmscan_direct_reclaim_end: nr_reclaimed=22
-5972 [006] .... 7971.250152: mm_compaction_begin: zone_start=0x40000 migrate_pfn=0x40000 free_pfn=0xaf800 zone_end=0x100000, mode=async
-5972 [006] .... 7971.250305: mm_compaction_end: zone_start=0x40000 migrate_pfn=0xac400 free_pfn=0xaf800 zone_end=0x100000, mode=async status=success
-5972 [006] .... 7971.250375: mm_compaction_begin: zone_start=0x40000 migrate_pfn=0xac800 free_pfn=0xaf800 zone_end=0x100000, mode=async
-5972 [006] .... 7971.250380: mm_compaction_end: zone_start=0x40000 migrate_pfn=0xaf800 free_pfn=0xaf800 zone_end=0x100000, mode=async status=partial_skipped
-5972 [006] .... 7971.250384: mm_vmscan_direct_reclaim_begin: order=4 gfp_flags=GFP_HIGHUSER|__GFP_NOWARN|__GFP_COMP|__GFP_ZERO
-5972 [006] .... 7971.250597: mm_vmscan_direct_reclaim_end: nr_reclaimed=20
-5972 [006] .... 7971.251289: mm_compaction_begin: zone_start=0x40000 migrate_pfn=0x40000 free_pfn=0xaf000 zone_end=0x100000, mode=async
-5972 [006] .... 7971.251497: mm_compaction_end: zone_start=0x40000 migrate_pfn=0xac400 free_pfn=0xaf000 zone_end=0x100000, mode=async status=contended
-5972 [006] .... 7971.251502: mm_vmscan_direct_reclaim_begin: order=4 gfp_flags=GFP_HIGHUSER|__GFP_NOWARN|__GFP_COMP|__GFP_ZERO
-5972 [006] .... 7971.251797: mm_vmscan_direct_reclaim_end: nr_reclaimed=22
-5972 [006] .... 7971.251890: mm_compaction_begin: zone_start=0x40000 migrate_pfn=0x5b400 free_pfn=0xb0400 zone_end=0x100000, mode=sync
-5972 [006] .... 7971.252740: mm_compaction_end: zone_start=0x40000 migrate_pfn=0x63800 free_pfn=0xb0400 zone_end=0x100000, mode=sync status=success
-5972 [006] .... 7971.252774: mm_compaction_begin: zone_start=0x40000 migrate_pfn=0x63c00 free_pfn=0xb0000 zone_end=0x100000, mode=async
-5972 [006] .... 7971.253325: mm_compaction_end: zone_start=0x40000 migrate_pfn=0x74e20 free_pfn=0xade92 zone_end=0x100000, mode=async status=success
-5972 [006] .... 7971.253333: mm_compaction_begin: zone_start=0x40000 migrate_pfn=0x83400 free_pfn=0xafc00 zone_end=0x100000, mode=async
-5972 [006] .... 7971.253683: mm_compaction_end: zone_start=0x40000 migrate_pfn=0xac400 free_pfn=0xafc00 zone_end=0x100000, mode=async status=contended
-5972 [006] .... 7971.253688: mm_vmscan_direct_reclaim_begin: order=4 gfp_flags=GFP_HIGHUSER|__GFP_NOWARN|__GFP_COMP|__GFP_ZERO
-5972 [006] .... 7971.254134: mm_vmscan_direct_reclaim_end: nr_reclaimed=22
-5972 [006] .... 7971.254206: mm_compaction_begin: zone_start=0x40000 migrate_pfn=0xad000 free_pfn=0xafc00 zone_end=0x100000, mode=async
-5972 [006] .... 7971.254289: mm_compaction_end: zone_start=0x40000 migrate_pfn=0xae1c0 free_pfn=0xae000 zone_end=0x100000, mode=async status=partial_skipped
-5972 [006] .... 7971.254304: mm_compaction_begin: zone_start=0x40000 migrate_pfn=0x40000 free_pfn=0xffc00 zone_end=0x100000, mode=async
-5972 [006] .... 7971.254510: mm_compaction_end: zone_start=0x40000 migrate_pfn=0x74db0 free_pfn=0xadcc1 zone_end=0x100000, mode=async status=success
-5972 [006] .... 7971.254519: mm_compaction_begin: zone_start=0x40000 migrate_pfn=0x75000 free_pfn=0xffc00 zone_end=0x100000, mode=async
-5972 [006] .... 7971.254612: mm_compaction_end: zone_start=0x40000 migrate_pfn=0xac400 free_pfn=0xffc00 zone_end=0x100000, mode=async status=contended
-5972 [006] .... 7971.254617: mm_vmscan_direct_reclaim_begin: order=4 gfp_flags=GFP_HIGHUSER|__GFP_NOWARN|__GFP_COMP|__GFP_ZERO
-5972 [006] .... 7971.255303: mm_vmscan_direct_reclaim_end: nr_reclaimed=22
-5972 [006] .... 7971.255307: mm_compaction_begin: zone_start=0x40000 migrate_pfn=0x75000 free_pfn=0xffc00 zone_end=0x100000, mode=sync
-5972 [006] .... 7971.260198: mm_compaction_end: zone_start=0x40000 migrate_pfn=0x751e5 free_pfn=0xaa83f zone_end=0x100000, mode=sync status=success
-5972 [006] .... 7971.260236: mm_compaction_begin: zone_start=0x40000 migrate_pfn=0x40000 free_pfn=0xffc00 zone_end=0x100000, mode=async
-5972 [006] .... 7971.260883: mm_compaction_end: zone_start=0x40000 migrate_pfn=0xb5c00 free_pfn=0xffc00 zone_end=0x100000, mode=async status=contended
-5972 [006] .... 7971.260891: mm_vmscan_direct_reclaim_begin: order=4 gfp_flags=GFP_HIGHUSER|__GFP_NOWARN|__GFP_COMP|__GFP_ZERO
-5972 [006] .... 7971.261174: mm_vmscan_direct_reclaim_end: nr_reclaimed=22
-5972 [006] .... 7971.261178: mm_compaction_begin: zone_start=0x40000 migrate_pfn=0x75000 free_pfn=0xffc00 zone_end=0x100000, mode=sync
-5972 [006] .... 7971.264861: mm_compaction_end: zone_start=0x40000 migrate_pfn=0x75681 free_pfn=0xffc00 zone_end=0x100000, mode=sync status=success
-5972 [006] .... 7971.264873: mm_compaction_begin: zone_start=0x40000 migrate_pfn=0xbbc00 free_pfn=0xffc00 zone_end=0x100000, mode=async
-5972 [006] .... 7971.264889: mm_compaction_end: zone_start=0x40000 migrate_pfn=0xc1400 free_pfn=0xffc00 zone_end=0x100000, mode=async status=contended
-5972 [006] .... 7971.264891: mm_vmscan_direct_reclaim_begin: order=4 gfp_flags=GFP_HIGHUSER|__GFP_NOWARN|__GFP_COMP|__GFP_ZERO
-5972 [006] .... 7971.265141: mm_vmscan_direct_reclaim_end: nr_reclaimed=22
-5972 [002] .... 7971.265585: mm_compaction_begin: zone_start=0x40000 migrate_pfn=0xc1800 free_pfn=0xffc00 zone_end=0x100000, mode=async
-5972 [002] .... 7971.265669: mm_compaction_end: zone_start=0x40000 migrate_pfn=0xc2c00 free_pfn=0xffc00 zone_end=0x100000, mode=async status=contended
-5972 [002] .... 7971.265678: mm_vmscan_direct_reclaim_begin: order=4 gfp_flags=GFP_HIGHUSER|__GFP_NOWARN|__GFP_COMP|__GFP_ZERO
-5972 [002] .... 7971.266987: mm_vmscan_direct_reclaim_end: nr_reclaimed=22
-5972 [006] .... 7971.267466: mm_compaction_begin: zone_start=0x40000 migrate_pfn=0x40000 free_pfn=0xffc00 zone_end=0x100000, mode=sync
-5972 [006] .... 7971.267778: mm_compaction_end: zone_start=0x40000 migrate_pfn=0x5d000 free_pfn=0xffc00 zone_end=0x100000, mode=sync status=success
-5972 [006] .... 7971.267927: mm_compaction_begin: zone_start=0x40000 migrate_pfn=0x5c000 free_pfn=0xffc00 zone_end=0x100000, mode=async
-5972 [006] .... 7971.268092: mm_compaction_end: zone_start=0x40000 migrate_pfn=0x74d70 free_pfn=0xffc00 zone_end=0x100000, mode=async status=success
Thank you
Jaewon Kim