Skip to contentSkip to conversation

dma-direct: do not allocate a single page from CMA area

RFC PATCH Superseded Nicolin Chen 3 participants 10 messages Oct 31, 2018 – Nov 21, 2018 DMA mapping helpers

v1 of 2 Superseded Latest is v2 →

lore
Share & tools
lore
#0 lore raw

RFC PATCH 1/1 dma-direct: do not allocate a single page from CMA area

The addresses within a single page are always contiguous, so it's not so necessary to allocate one single page from CMA area. Since the CMA area has a limited predefined size of space, it might run out of space in some heavy use case, where there might be quite a lot CMA pages being allocated for single pages.

This patch tries to skip CMA allocations of single pages and lets them go through normal page allocations. This would save resource in the CMA area for further more CMA allocations.

---
 kernel/dma/direct.c | 8 ++++++--
 1 file changed, 6 insertions(+), 2 deletions(-)
kernel/dma/direct.c +6 -2
index 22a12ab5a5e9..14c5d49eded2 100644
@@ -120,8 +120,12 @@ void *dma_direct_alloc_pages(struct device *dev, size_t size,
 	gfp |= __dma_direct_optimal_gfp_mask(dev, dev->coherent_dma_mask,
 			&phys_mask);
 again:
-	/* CMA can be used only in the context which permits sleeping */
-	if (gfpflags_allow_blocking(gfp)) {
+	/*
+	 * CMA can be used only in the context which permits sleeping.
+	 * Since addresses within one PAGE are always contiguous, skip
+	 * CMA allocation for a single page to save CMA reserved space
+	 */
+	if (gfpflags_allow_blocking(gfp) && count > 1) {
 		page = dma_alloc_from_contiguous(dev, count, page_order,
 						 gfp & __GFP_NOWARN);
 		if (page && !dma_coherent_ok(dev, page_to_phys(page), size)) {
Signature · 1 line
2.17.1
reviewer
#1 lore raw
Quoted 9 lines · Nicolin Chen

The addresses within a single page are always contiguous, so it's not so necessary to allocate one single page from CMA area. Since the CMA area has a limited predefined size of space, it might run out of space in some heavy use case, where there might be quite a lot CMA pages being allocated for single pages.

This patch tries to skip CMA allocations of single pages and lets them go through normal page allocations. This would save resource in the CMA area for further more CMA allocations.

In general, this seems to make sense to me. It does represent a theoretical change in behaviour for devices which have their own CMA area somewhere other than kernel memory, and only ever make non-atomic allocations, but I'm not sure whether that's a realistic or common enough case to really worry about.

Robin.

Quoted 25 lines
---
  kernel/dma/direct.c | 8 ++++++--
  1 file changed, 6 insertions(+), 2 deletions(-)
kernel/dma/direct.c +6 -2
index 22a12ab5a5e9..14c5d49eded2 100644
@@ -120,8 +120,12 @@ void *dma_direct_alloc_pages(struct device *dev, size_t size,
  	gfp |= __dma_direct_optimal_gfp_mask(dev, dev->coherent_dma_mask,
  			&phys_mask);
  again:
-	/* CMA can be used only in the context which permits sleeping */
-	if (gfpflags_allow_blocking(gfp)) {
+	/*
+	 * CMA can be used only in the context which permits sleeping.
+	 * Since addresses within one PAGE are always contiguous, skip
+	 * CMA allocation for a single page to save CMA reserved space
+	 */
+	if (gfpflags_allow_blocking(gfp) && count > 1) {
  		page = dma_alloc_from_contiguous(dev, count, page_order,
  						 gfp & __GFP_NOWARN);
  		if (page && !dma_coherent_ok(dev, page_to_phys(page), size)) {
#2 lore raw

Hi Robin,

Thanks for the comments.

Quoted 11 lines · Robin Murphy
Quoted 9 lines · Nicolin Chen

The addresses within a single page are always contiguous, so it's not so necessary to allocate one single page from CMA area. Since the CMA area has a limited predefined size of space, it might run out of space in some heavy use case, where there might be quite a lot CMA pages being allocated for single pages.

This patch tries to skip CMA allocations of single pages and lets them go through normal page allocations. This would save resource in the CMA area for further more CMA allocations.

Quoted 5 lines

In general, this seems to make sense to me. It does represent a theoretical change in behaviour for devices which have their own CMA area somewhere other than kernel memory, and only ever make non-atomic allocations, but I'm not sure whether that's a realistic or common enough case to really worry about.

Hmm..I don't quite understand the part of worrying its realisticness. Would you mind elaborating a bit? As I tested this change on Tegra186 board, and saw some single-page allocations have been directed to the normal allocation; and the "CmaFree" size reported from /proc/meminfo is also increased. Does this mean it's realistic?

Thank you
Nicolin

-----

Quoted 28 lines

Robin.

Quoted 25 lines
---
  kernel/dma/direct.c | 8 ++++++--
  1 file changed, 6 insertions(+), 2 deletions(-)
kernel/dma/direct.c +6 -2
index 22a12ab5a5e9..14c5d49eded2 100644
@@ -120,8 +120,12 @@ void *dma_direct_alloc_pages(struct device *dev, size_t size,
  	gfp |= __dma_direct_optimal_gfp_mask(dev, dev->coherent_dma_mask,
  			&phys_mask);
  again:
-	/* CMA can be used only in the context which permits sleeping */
-	if (gfpflags_allow_blocking(gfp)) {
+	/*
+	 * CMA can be used only in the context which permits sleeping.
+	 * Since addresses within one PAGE are always contiguous, skip
+	 * CMA allocation for a single page to save CMA reserved space
+	 */
+	if (gfpflags_allow_blocking(gfp) && count > 1) {
  		page = dma_alloc_from_contiguous(dev, count, page_order,
  						 gfp & __GFP_NOWARN);
  		if (page && !dma_coherent_ok(dev, page_to_phys(page), size)) {
reviewer
#3 lore raw
Quoted 16 lines · Nicolin Chen

Hi Robin,

Thanks for the comments.

Quoted 11 lines · Robin Murphy
Quoted 9 lines · Nicolin Chen

The addresses within a single page are always contiguous, so it's not so necessary to allocate one single page from CMA area. Since the CMA area has a limited predefined size of space, it might run out of space in some heavy use case, where there might be quite a lot CMA pages being allocated for single pages.

This patch tries to skip CMA allocations of single pages and lets them go through normal page allocations. This would save resource in the CMA area for further more CMA allocations.

Quoted 8 lines
Quoted 5 lines

In general, this seems to make sense to me. It does represent a theoretical change in behaviour for devices which have their own CMA area somewhere other than kernel memory, and only ever make non-atomic allocations, but I'm not sure whether that's a realistic or common enough case to really worry about.

Hmm..I don't quite understand the part of worrying its realisticness. Would you mind elaborating a bit?

I only mean the case where a driver previously happened to get single pages allocated from a per-device CMA area, would now always get them fulfilled from regular kernel memory instead, and actually cares about the difference. As I say, that's a contrived case that I doubt is honestly a significant concern, but it's not *entirely* inconceivable. I've just been bitten before by drivers relying on specific DMA API implementation behaviour which was never guaranteed or even necessarily correct by the terms of the API itself, so I'm naturally wary of the corner cases ;)

On second thought, however, I suppose we could always key this off DMA_ATTR_FORCE_CONTIGUOUS as well if we really want - technically it has a more general meaning than "only ever allocate from CMA", but in practice if that's the behaviour a driver wants, then that flag is already the only way it can even hope to get dma_alloc_coherent() to comply anywhere near reliably.

Quoted 4 lines

As I tested this change on Tegra186
board, and saw some single-page allocations have been directed to the normal allocation; and the "CmaFree" size reported from /proc/meminfo is also increased. Does this mean it's realistic?

Indeed - I happen to have CMA debug enabled for no good reason in my current development config, and on my relatively unexciting Juno board single-page allocations turn out to be the majority by number, even if not by total consumption:

[    0.519663] cma: cma_alloc(cma (____ptrval____), count 64, align 6)
[    0.527508] cma: cma_alloc(): returned (____ptrval____)
[    3.768066] cma: cma_alloc(cma (____ptrval____), count 1, align 0)
[    3.774566] cma: cma_alloc(): returned (____ptrval____)
[    3.860097] cma: cma_alloc(cma (____ptrval____), count 1875, align 8)
[    3.867150] cma: cma_alloc(): returned (____ptrval____)
[    3.920796] cma: cma_alloc(cma (____ptrval____), count 31, align 5)
[    3.927093] cma: cma_alloc(): returned (____ptrval____)
[    3.932326] cma: cma_alloc(cma (____ptrval____), count 31, align 5)
[    3.938643] cma: cma_alloc(): returned (____ptrval____)
[    4.022188] cma: cma_alloc(cma (____ptrval____), count 1, align 0)
[    4.028415] cma: cma_alloc(): returned (____ptrval____)
[    4.033600] cma: cma_alloc(cma (____ptrval____), count 1, align 0)
[    4.039786] cma: cma_alloc(): returned (____ptrval____)
[    4.044968] cma: cma_alloc(cma (____ptrval____), count 1, align 0)
[    4.051150] cma: cma_alloc(): returned (____ptrval____)
[    4.113556] cma: cma_alloc(cma (____ptrval____), count 1, align 0)
[    4.119785] cma: cma_alloc(): returned (____ptrval____)
[    5.012654] cma: cma_alloc(cma (____ptrval____), count 1, align 0)
[    5.019047] cma: cma_alloc(): returned (____ptrval____)
[   11.485179] cma: cma_alloc(cma 000000009dd074ee, count 1, align 0)
[   11.492096] cma: cma_alloc(): returned 000000009264a86c
[   12.269355] cma: cma_alloc(cma 000000009dd074ee, count 1875, align 8)
[   12.277535] cma: cma_alloc(): returned 00000000d7bb9ae5
[   12.286110] cma: cma_alloc(cma 000000009dd074ee, count 4, align 2)
[   12.292507] cma: cma_alloc(): returned 0000000007ba7a39

I don't have any exciting peripherals to really exercise the coherent allocator, but I imagine that fragmentation is probably just as good a reason as total CMA usage for avoiding trivial allocations by default.

Robin.

#4 lore raw
Quoted 18 lines · Robin Murphy
Quoted 18 lines · Robin Murphy
Quoted 16 lines · Nicolin Chen
Quoted 9 lines

The addresses within a single page are always contiguous, so it's not so necessary to allocate one single page from CMA area. Since the CMA area has a limited predefined size of space, it might run out of space in some heavy use case, where there might be quite a lot CMA pages being allocated for single pages.

This patch tries to skip CMA allocations of single pages and lets them go through normal page allocations. This would save resource in the CMA area for further more CMA allocations.

In general, this seems to make sense to me. It does represent a theoretical change in behaviour for devices which have their own CMA area somewhere other than kernel memory, and only ever make non-atomic allocations, but I'm not sure whether that's a realistic or common enough case to really worry about.

Quoted 6 lines
Quoted 2 lines

Hmm..I don't quite understand the part of worrying its realisticness. Would you mind elaborating a bit?

I only mean the case where a driver previously happened to get single pages allocated from a per-device CMA area, would now always get them fulfilled from regular kernel memory instead, and actually cares about the difference.

I see. I think that's a good question.

Quoted 5 lines

As I say, that's a contrived case that I doubt is honestly a significant concern, but it's not *entirely* inconceivable. I've just been bitten before by drivers relying on specific DMA API implementation behaviour which was never guaranteed or even necessarily correct by the terms of the API itself, so I'm naturally wary of the corner cases ;)

I also have a vague concern that CMA pages might turn out to be special so this change would make some differences in stability or performance for those who actually rely on actual CMA pages, though I ain't sure if it can be true or realistic as you said.

Quoted 6 lines

On second thought, however, I suppose we could always key this off DMA_ATTR_FORCE_CONTIGUOUS as well if we really want - technically it has a more general meaning than "only ever allocate from CMA", but in practice if that's the behaviour a driver wants, then that flag is already the only way it can even hope to get dma_alloc_coherent() to comply anywhere near reliably.

That is a good input. Would you prefer to have it in condition check now with this patch?

Quoted 37 lines
Quoted 4 lines

As I tested this change on Tegra186
board, and saw some single-page allocations have been directed to the normal allocation; and the "CmaFree" size reported from /proc/meminfo is also increased. Does this mean it's realistic?

Indeed - I happen to have CMA debug enabled for no good reason in my current development config, and on my relatively unexciting Juno board single-page allocations turn out to be the majority by number, even if not by total consumption:

[    0.519663] cma: cma_alloc(cma (____ptrval____), count 64, align 6)
[    0.527508] cma: cma_alloc(): returned (____ptrval____)
[    3.768066] cma: cma_alloc(cma (____ptrval____), count 1, align 0)
[    3.774566] cma: cma_alloc(): returned (____ptrval____)
[    3.860097] cma: cma_alloc(cma (____ptrval____), count 1875, align 8)
[    3.867150] cma: cma_alloc(): returned (____ptrval____)
[    3.920796] cma: cma_alloc(cma (____ptrval____), count 31, align 5)
[    3.927093] cma: cma_alloc(): returned (____ptrval____)
[    3.932326] cma: cma_alloc(cma (____ptrval____), count 31, align 5)
[    3.938643] cma: cma_alloc(): returned (____ptrval____)
[    4.022188] cma: cma_alloc(cma (____ptrval____), count 1, align 0)
[    4.028415] cma: cma_alloc(): returned (____ptrval____)
[    4.033600] cma: cma_alloc(cma (____ptrval____), count 1, align 0)
[    4.039786] cma: cma_alloc(): returned (____ptrval____)
[    4.044968] cma: cma_alloc(cma (____ptrval____), count 1, align 0)
[    4.051150] cma: cma_alloc(): returned (____ptrval____)
[    4.113556] cma: cma_alloc(cma (____ptrval____), count 1, align 0)
[    4.119785] cma: cma_alloc(): returned (____ptrval____)
[    5.012654] cma: cma_alloc(cma (____ptrval____), count 1, align 0)
[    5.019047] cma: cma_alloc(): returned (____ptrval____)
[   11.485179] cma: cma_alloc(cma 000000009dd074ee, count 1, align 0)
[   11.492096] cma: cma_alloc(): returned 000000009264a86c
[   12.269355] cma: cma_alloc(cma 000000009dd074ee, count 1875, align 8)
[   12.277535] cma: cma_alloc(): returned 00000000d7bb9ae5
[   12.286110] cma: cma_alloc(cma 000000009dd074ee, count 4, align 2)
[   12.292507] cma: cma_alloc(): returned 0000000007ba7a39
Quoted 3 lines

I don't have any exciting peripherals to really exercise the coherent allocator, but I imagine that fragmentation is probably just as good a reason as total CMA usage for avoiding trivial allocations by default.

I will also add to the commit message fragmentation reduction.

Thanks
Nicolin

#5 lore raw
In reply to Robin Murphy
Quoted 11 lines · Robin Murphy
Quoted 9 lines · Nicolin Chen

The addresses within a single page are always contiguous, so it's not so necessary to allocate one single page from CMA area. Since the CMA area has a limited predefined size of space, it might run out of space in some heavy use case, where there might be quite a lot CMA pages being allocated for single pages.

This patch tries to skip CMA allocations of single pages and lets them go through normal page allocations. This would save resource in the CMA area for further more CMA allocations.

Quoted 5 lines

In general, this seems to make sense to me. It does represent a theoretical change in behaviour for devices which have their own CMA area somewhere other than kernel memory, and only ever make non-atomic allocations, but I'm not sure whether that's a realistic or common enough case to really worry about.

Yes, I think we should make the decision in dma_alloc_from_contiguous based on having a per-dev CMA area or not. There is a lot of cruft in this area that should be cleaned up while we're at it, like always falling back to the normal page allocator if there is no CMA area or nothing suitable found in dma_alloc_from_contiguous instead of having to duplicate all that in the caller.

#6 lore raw
Quoted 12 lines · Christoph Hellwig
Quoted 11 lines · Robin Murphy
Quoted 9 lines · Nicolin Chen

The addresses within a single page are always contiguous, so it's not so necessary to allocate one single page from CMA area. Since the CMA area has a limited predefined size of space, it might run out of space in some heavy use case, where there might be quite a lot CMA pages being allocated for single pages.

This patch tries to skip CMA allocations of single pages and lets them go through normal page allocations. This would save resource in the CMA area for further more CMA allocations.

Quoted 8 lines
Quoted 5 lines

In general, this seems to make sense to me. It does represent a theoretical change in behaviour for devices which have their own CMA area somewhere other than kernel memory, and only ever make non-atomic allocations, but I'm not sure whether that's a realistic or common enough case to really worry about.

Yes, I think we should make the decision in dma_alloc_from_contiguous based on having a per-dev CMA area or not. There is a lot of cruft in

It seems that cma_alloc() already has a CMA area check? Would it be duplicated to have a similar one in dma_alloc_from_contiguous?

Quoted 4 lines

this area that should be cleaned up while we're at it, like always falling back to the normal page allocator if there is no CMA area or nothing suitable found in dma_alloc_from_contiguous instead of having to duplicate all that in the caller.

Am I supposed to clean up things that's mentioned above by moving the fallback allocator into dma_alloc_from_contiguous, or to just move my change (the count check) into dma_alloc_from_contiguous?

I understand that'd be great to have a cleanup, yet feel it could be done separately as this patch isn't really a cleanup change.

Thanks
Nicolin

#7 lore raw

Robin? Christ?

Quoted 39 lines · Nicolin Chen
Quoted 12 lines · Christoph Hellwig
Quoted 11 lines · Robin Murphy
Quoted 9 lines · Nicolin Chen

The addresses within a single page are always contiguous, so it's not so necessary to allocate one single page from CMA area. Since the CMA area has a limited predefined size of space, it might run out of space in some heavy use case, where there might be quite a lot CMA pages being allocated for single pages.

This patch tries to skip CMA allocations of single pages and lets them go through normal page allocations. This would save resource in the CMA area for further more CMA allocations.

Quoted 8 lines
Quoted 5 lines

In general, this seems to make sense to me. It does represent a theoretical change in behaviour for devices which have their own CMA area somewhere other than kernel memory, and only ever make non-atomic allocations, but I'm not sure whether that's a realistic or common enough case to really worry about.

Yes, I think we should make the decision in dma_alloc_from_contiguous based on having a per-dev CMA area or not. There is a lot of cruft in

It seems that cma_alloc() already has a CMA area check? Would it be duplicated to have a similar one in dma_alloc_from_contiguous?

Quoted 4 lines

this area that should be cleaned up while we're at it, like always falling back to the normal page allocator if there is no CMA area or nothing suitable found in dma_alloc_from_contiguous instead of having to duplicate all that in the caller.

Am I supposed to clean up things that's mentioned above by moving the fallback allocator into dma_alloc_from_contiguous, or to just move my change (the count check) into dma_alloc_from_contiguous?

I understand that'd be great to have a cleanup, yet feel it could be done separately as this patch isn't really a cleanup change.

Thanks
Nicolin

#8 lore raw
In reply to Nicolin Chen
Quoted 11 lines · Nicolin Chen
Quoted 8 lines
Quoted 5 lines

In general, this seems to make sense to me. It does represent a theoretical change in behaviour for devices which have their own CMA area somewhere other than kernel memory, and only ever make non-atomic allocations, but I'm not sure whether that's a realistic or common enough case to really worry about.

Yes, I think we should make the decision in dma_alloc_from_contiguous based on having a per-dev CMA area or not. There is a lot of cruft in

It seems that cma_alloc() already has a CMA area check? Would it be duplicated to have a similar one in dma_alloc_from_contiguous?

It isn't duplicate if it serves a different purpose.

Quoted 11 lines
Quoted 4 lines

this area that should be cleaned up while we're at it, like always falling back to the normal page allocator if there is no CMA area or nothing suitable found in dma_alloc_from_contiguous instead of having to duplicate all that in the caller.

Am I supposed to clean up things that's mentioned above by moving the fallback allocator into dma_alloc_from_contiguous, or to just move my change (the count check) into dma_alloc_from_contiguous?

I understand that'd be great to have a cleanup, yet feel it could be done separately as this patch isn't really a cleanup change.

I can take care of any cleanups. I've been trying to dust up that area anyway.

#9 lore raw
Quoted 24 lines · Christoph Hellwig
Quoted 11 lines · Nicolin Chen
Quoted 8 lines
Quoted 5 lines

In general, this seems to make sense to me. It does represent a theoretical change in behaviour for devices which have their own CMA area somewhere other than kernel memory, and only ever make non-atomic allocations, but I'm not sure whether that's a realistic or common enough case to really worry about.

Yes, I think we should make the decision in dma_alloc_from_contiguous based on having a per-dev CMA area or not. There is a lot of cruft in

It seems that cma_alloc() already has a CMA area check? Would it be duplicated to have a similar one in dma_alloc_from_contiguous?

It isn't duplicate if it serves a different purpose.

Quoted 9 lines
Quoted 4 lines

this area that should be cleaned up while we're at it, like always falling back to the normal page allocator if there is no CMA area or nothing suitable found in dma_alloc_from_contiguous instead of having to duplicate all that in the caller.

Am I supposed to clean up things that's mentioned above by moving the fallback allocator into dma_alloc_from_contiguous, or to just move my change (the count check) into dma_alloc_from_contiguous?

Quoted 5 lines
Quoted 2 lines

I understand that'd be great to have a cleanup, yet feel it could be done separately as this patch isn't really a cleanup change.

I can take care of any cleanups. I've been trying to dust up that area anyway.

Thanks for the reply. It looks like it'd be better for me to wait for the cleanup being done? I feel odd merely adding a size check in the dma_alloc_from_contiguous().