dmaengine: pxa: fix double counting of the hw descriptors
Subsystems (2)
PATCH v2 1/1 dmaengine: pxa: fix double counting of the hw descriptors
pxad_alloc_desc() was converted from
kzalloc(struct_size(sw_desc, hw_desc, nb_hw_desc), GFP_NOWAIT)
to kzalloc_flex(), which sets the __counted_by() counter sw_desc->nb_desc itself - but only where the compiler has __builtin_counted_by_ref(), so from gcc 15.1 or clang 22.1 on. The loop below it still increments nb_desc, which makes it come out doubled there and correct elsewhere.
nb_desc is what pxad_free_desc() iterates over and what set_updater_desc() indexes from, so set it explicitly and drop the increment. The error path has to lower it to the number of descriptors allocated so far, otherwise pxad_free_desc() would free entries that were never allocated.
---pxad_alloc_desc() was converted from
kzalloc(struct_size(sw_desc, hw_desc, nb_hw_desc), GFP_NOWAIT)
to kzalloc_flex(). hw_desc[] is annotated with __counted_by(nb_desc), so __alloc_flex() now initializes sw_desc->nb_desc to nb_hw_desc itself. The loop below it still increments nb_desc for every descriptor it allocates though, so nb_desc ends up being twice the number of descriptors that are actually there.
Drop the now redundant increment. The error path has to set nb_desc to
the number of descriptors allocated so far, otherwise pxad_free_desc()
would free entries that were never allocated.
---
Changes in v2:
- assign sw_desc->nb_desc manually as it is only automatically assigned
for compilers supporting __builtin_counted_by_ref()
- Link to v1: https://lore.kernel.org/r/[email protected]
--- drivers/dma/pxa_dma.c | 3 ++- 1 file changed, 2 insertions(+), 1 deletion(-)
drivers/dma/pxa_dma.c +2 -1
@@ -744,6 +744,7 @@ pxad_alloc_desc(struct pxad_chan *chan, unsigned int nb_hw_desc) sw_desc =kzalloc_flex (*sw_desc, hw_desc, nb_hw_desc,GFP_NOWAIT );if (!sw_desc)return NULL ; + sw_desc->nb_desc = nb_hw_desc; sw_desc->desc_pool = chan->desc_pool;for (i =0 ; i < nb_hw_desc; i++) {
@@ -752,10 +753,10 @@ pxad_alloc_desc(struct pxad_chan *chan, unsigned int nb_hw_desc)dev_err (&chan->vc.chan.dev->device,"%s(): Couldn't allocate the %dth hw_desc from dma_pool %p\n" , __func__, i, sw_desc->desc_pool); + sw_desc->nb_desc = i; goto err; } - sw_desc->nb_desc++; sw_desc->hw_desc[i] = desc;if (i ==0 )
---
base-commit: 8d3ae59288f1e7d58d76558a6ee96d533bc5019f
change-id: 20260817-dmaengine-pxa-64152bb34313
Best regards,
Signature · 1 line
Sascha Hauer <[email protected]>
[email protected]
AI reviewThank you for your contribution! Sashiko AI review found 1 potential issue(s) to consider:
Thank you for your contribution! Sashiko AI review found 1 potential issue(s) to consider:
Pre-existing issues:
- [High] For cyclic DMA transfers, the updater descriptor's DMA address is leaked, and the first descriptor's DMA address is double-freed in the DMA pool. --
--- Patch [1]: [PATCH v2] dmaengine: pxa: fix double counting of the hw descriptors --- commit 5e37421334e2f54f86d40ee190a04a05eb99596b
Author: Sascha Hauer <[email protected]>
dmaengine: pxa: fix double counting of the hw descriptors
This commit fixes a double counting issue with sw_desc->nb_desc in pxad_alloc_desc() caused by the transition to kzalloc_flex(). It explicitly sets nb_desc and removes the redundant increment in the allocation loop.
Quoted 4 lines
drivers/dma/pxa_dma.c +0
-0
[ ... ]
drivers/dma/pxa_dma.c pxad_alloc_desc(struct pxad_chan *chan, unsigned int nb_hw_desc) · line 747
Quoted 5 lines
drivers/dma/pxa_dma.c +1
-0@@ -744,6 +744,7 @@ pxad_alloc_desc(struct pxad_chan *chan, unsigned int nb_hw_desc) sw_desc =kzalloc_flex (*sw_desc, hw_desc, nb_hw_desc,GFP_NOWAIT );if (!sw_desc)return NULL ; + sw_desc->nb_desc = nb_hw_desc;
[Severity: High]
This isn't a bug introduced by this patch, but since this patch corrects the nb_desc tracking used by set_updater_desc() and pxad_free_desc(), is there a pre-existing double-free for cyclic transfers?
When a cyclic transfer is set up, set_updater_desc() creates a hardware loop by overwriting the ddadr of the second-to-last descriptor:
drivers/dma/pxa_dma.c:set_updater_desc() {
...
if (sw_desc->cyclic)
sw_desc->hw_desc[sw_desc->nb_desc - 2]->ddadr = sw_desc->first;
...
}Later, pxad_free_desc() relies on this overwritten field to retrieve the DMA address to free the next descriptor in the chain:
drivers/dma/pxa_dma.c:pxad_free_desc() {
...
for (i = sw_desc->nb_desc - 1; i >= 0; i--) {
if (i > 0)
dma = sw_desc->hw_desc[i - 1]->ddadr;
else
dma = sw_desc->first;
dma_pool_free(sw_desc->desc_pool,
sw_desc->hw_desc[i], dma);
}
...
}Because ddadr was overwritten with sw_desc->first, doesn't pxad_free_desc() read sw_desc->first instead of the original DMA address when freeing the updater descriptor?
Could this lead to double-freeing the first descriptor's DMA address and permanently leaking the updater descriptor's original address in the DMA pool?
Quoted 3 lines
drivers/dma/pxa_dma.c +0
-0sw_desc->desc_pool = chan->desc_pool;for (i =0 ; i < nb_hw_desc; i++) {
[ ... ]
Signature · 1 line
Sashiko AI review · https://sashiko.dev/#/patchset/[email protected]?part=1
Quoted 15 lines · Sascha Hauer
pxad_alloc_desc()was converted from
kzalloc(struct_size(sw_desc, hw_desc, nb_hw_desc), GFP_NOWAIT)to
kzalloc_flex(), which sets the__counted_by()countersw_desc->nb_descitself - but only where the compiler has__builtin_counted_by_ref(), so from gcc 15.1 or clang 22.1 on. The loop below it still incrementsnb_desc, which makes it come out doubled there and correct elsewhere.
nb_descis whatpxad_free_desc()iterates over and whatset_updater_desc()indexes from, so set it explicitly and drop the increment. The error path has to lower it to the number of descriptors allocated so far, otherwisepxad_free_desc()would free entries that were never allocated.
Quoted 4 lines
Fixes 69050f8d6d075 ("treewide: Replace kmalloc with kmalloc_obj for non-scalar types")Assisted-by Claude:claude-opus-5Signed-off-by Sascha Hauer <[email protected]>---
Quoted 55 lines
pxad_alloc_desc()was converted from
kzalloc(struct_size(sw_desc, hw_desc, nb_hw_desc), GFP_NOWAIT)to
kzalloc_flex(). hw_desc[] is annotated with__counted_by(nb_desc), so__alloc_flex()now initializessw_desc->nb_desctonb_hw_descitself. The loop below it still incrementsnb_descfor every descriptor it allocates though, sonb_descends up being twice the number of descriptors that are actually there.Drop the now redundant increment. The error path has to set
nb_descto
the number of descriptors allocated so far, otherwisepxad_free_desc()
would free entries that were never allocated.
---
Changes in v2:
- assignsw_desc->nb_descmanually as it is only automatically assigned
for compilers supporting__builtin_counted_by_ref()
- Link to v1: https://lore.kernel.org/r/[email protected]--- drivers/dma/pxa_dma.c | 3 ++- 1 file changed, 2 insertions(+), 1 deletion(-)drivers/dma/pxa_dma.c +2
-2@@ -744,6 +744,7 @@ pxad_alloc_desc(struct pxad_chan *chan, unsigned int nb_hw_desc) sw_desc =kzalloc_flex (*sw_desc, hw_desc, nb_hw_desc,GFP_NOWAIT );if (!sw_desc)return NULL ; + sw_desc->nb_desc = nb_hw_desc; sw_desc->desc_pool = chan->desc_pool;for (i =0 ; i < nb_hw_desc; i++) {@@ -752,10 +753,10 @@ pxad_alloc_desc(struct pxad_chan *chan, unsigned int nb_hw_desc)dev_err (&chan->vc.chan.dev->device,"%s(): Couldn't allocate the %dth hw_desc from dma_pool %p\n" , __func__, i, sw_desc->desc_pool); + sw_desc->nb_desc = i; goto err; } - sw_desc->nb_desc++; sw_desc->hw_desc[i] = desc;if (i ==0 ) ---base-commit:
8d3ae59288f1e7d58d76558a6ee96d533bc5019f
change-id: 20260817-dmaengine-pxa-64152bb34313Best regards,
--
Sascha Hauer <[email protected]>
Quoted 10 lines · Sascha Hauer
pxad_alloc_desc()was converted from
kzalloc(struct_size(sw_desc, hw_desc, nb_hw_desc), GFP_NOWAIT)to
kzalloc_flex(), which sets the__counted_by()countersw_desc->nb_descitself - but only where the compiler has__builtin_counted_by_ref(), so from gcc 15.1 or clang 22.1 on. The loop below it still incrementsnb_desc, which makes it come out doubled there and correct elsewhere.[...]
Applied, thanks!
[1/1] dmaengine: pxa: fix double counting of the hw descriptors
commit: f6504be006aa4bb4bd26285f410a885c17920d65
Best regards,
Signature · 1 line
~Vinod