It can be surprising to the user if DMA functions are only traced on
success. On failure, it can be unclear what the source of the problem
is. Fix this by tracing all functions even when they fail. Cases where
we BUG/WARN are skipped, since those should be sufficiently noisy
already.
Signed-off-by: Sean Anderson <sean.anderson@linux.dev>
---
include/trace/events/dma.h | 41 ++++++++++++++++++++++++++++++++++++++
kernel/dma/mapping.c | 27 +++++++++++++++++--------
2 files changed, 60 insertions(+), 8 deletions(-)
diff --git a/include/trace/events/dma.h b/include/trace/events/dma.h
index 9bc647f9ad4d..321cce327404 100644
--- a/include/trace/events/dma.h
+++ b/include/trace/events/dma.h
@@ -161,6 +161,12 @@ DEFINE_EVENT(_dma_alloc, dma_alloc_pages,
unsigned long attrs),
TP_ARGS(dev, virt_addr, dma_addr, size, dir, flags, attrs));
+DEFINE_EVENT(_dma_alloc, dma_alloc_sgt_err,
+ TP_PROTO(struct device *dev, void *virt_addr, dma_addr_t dma_addr,
+ size_t size, enum dma_data_direction dir, gfp_t flags,
+ unsigned long attrs),
+ TP_ARGS(dev, virt_addr, dma_addr, size, dir, flags, attrs));
+
TRACE_EVENT(dma_alloc_sgt,
TP_PROTO(struct device *dev, struct sg_table *sgt, size_t size,
enum dma_data_direction dir, gfp_t flags, unsigned long attrs),
@@ -325,6 +331,41 @@ TRACE_EVENT(dma_map_sg,
decode_dma_attrs(__entry->attrs))
);
+TRACE_EVENT(dma_map_sg_err,
+ TP_PROTO(struct device *dev, struct scatterlist *sgl, int nents,
+ int err, enum dma_data_direction dir, unsigned long attrs),
+ TP_ARGS(dev, sgl, nents, err, dir, attrs),
+
+ TP_STRUCT__entry(
+ __string(device, dev_name(dev))
+ __dynamic_array(u64, phys_addrs, nents)
+ __field(int, err)
+ __field(enum dma_data_direction, dir)
+ __field(unsigned long, attrs)
+ ),
+
+ TP_fast_assign(
+ struct scatterlist *sg;
+ int i;
+
+ __assign_str(device);
+ for_each_sg(sgl, sg, nents, i)
+ ((u64 *)__get_dynamic_array(phys_addrs))[i] = sg_phys(sg);
+ __entry->err = err;
+ __entry->dir = dir;
+ __entry->attrs = attrs;
+ ),
+
+ TP_printk("%s dir=%s dma_addrs=%s err=%d attrs=%s",
+ __get_str(device),
+ decode_dma_data_direction(__entry->dir),
+ __print_array(__get_dynamic_array(phys_addrs),
+ __get_dynamic_array_len(phys_addrs) /
+ sizeof(u64), sizeof(u64)),
+ __entry->err,
+ decode_dma_attrs(__entry->attrs))
+);
+
TRACE_EVENT(dma_unmap_sg,
TP_PROTO(struct device *dev, struct scatterlist *sgl, int nents,
enum dma_data_direction dir, unsigned long attrs),
diff --git a/kernel/dma/mapping.c b/kernel/dma/mapping.c
index b8a6bc492fae..636dbb0629a4 100644
--- a/kernel/dma/mapping.c
+++ b/kernel/dma/mapping.c
@@ -223,6 +223,7 @@ static int __dma_map_sg_attrs(struct device *dev, struct scatterlist *sg,
debug_dma_map_sg(dev, sg, nents, ents, dir, attrs);
} else if (WARN_ON_ONCE(ents != -EINVAL && ents != -ENOMEM &&
ents != -EIO && ents != -EREMOTEIO)) {
+ trace_dma_map_sg_err(dev, sg, nents, ents, dir, attrs);
return -EIO;
}
@@ -604,20 +605,26 @@ void *dma_alloc_attrs(struct device *dev, size_t size, dma_addr_t *dma_handle,
if (WARN_ON_ONCE(flag & __GFP_COMP))
return NULL;
- if (dma_alloc_from_dev_coherent(dev, size, dma_handle, &cpu_addr))
+ if (dma_alloc_from_dev_coherent(dev, size, dma_handle, &cpu_addr)) {
+ trace_dma_alloc(dev, cpu_addr, *dma_handle, size,
+ DMA_BIDIRECTIONAL, flag, attrs);
return cpu_addr;
+ }
/* let the implementation decide on the zone to allocate from: */
flag &= ~(__GFP_DMA | __GFP_DMA32 | __GFP_HIGHMEM);
- if (dma_alloc_direct(dev, ops))
+ if (dma_alloc_direct(dev, ops)) {
cpu_addr = dma_direct_alloc(dev, size, dma_handle, flag, attrs);
- else if (use_dma_iommu(dev))
+ } else if (use_dma_iommu(dev)) {
cpu_addr = iommu_dma_alloc(dev, size, dma_handle, flag, attrs);
- else if (ops->alloc)
+ } else if (ops->alloc) {
cpu_addr = ops->alloc(dev, size, dma_handle, flag, attrs);
- else
+ } else {
+ trace_dma_alloc(dev, NULL, 0, size, DMA_BIDIRECTIONAL, flag,
+ attrs);
return NULL;
+ }
trace_dma_alloc(dev, cpu_addr, *dma_handle, size, DMA_BIDIRECTIONAL,
flag, attrs);
@@ -642,11 +649,11 @@ void dma_free_attrs(struct device *dev, size_t size, void *cpu_addr,
*/
WARN_ON(irqs_disabled());
- if (!cpu_addr)
- return;
-
trace_dma_free(dev, cpu_addr, dma_handle, size, DMA_BIDIRECTIONAL,
attrs);
+ if (!cpu_addr)
+ return;
+
debug_dma_free_coherent(dev, size, cpu_addr, dma_handle);
if (dma_alloc_direct(dev, ops))
dma_direct_free(dev, size, cpu_addr, dma_handle, attrs);
@@ -688,6 +695,8 @@ struct page *dma_alloc_pages(struct device *dev, size_t size,
trace_dma_alloc_pages(dev, page_to_virt(page), *dma_handle,
size, dir, gfp, 0);
debug_dma_map_page(dev, page, 0, size, dir, *dma_handle, 0);
+ } else {
+ trace_dma_alloc_pages(dev, NULL, 0, size, dir, gfp, 0);
}
return page;
}
@@ -772,6 +781,8 @@ struct sg_table *dma_alloc_noncontiguous(struct device *dev, size_t size,
sgt->nents = 1;
trace_dma_alloc_sgt(dev, sgt, size, dir, gfp, attrs);
debug_dma_map_sg(dev, sgt->sgl, sgt->orig_nents, 1, dir, attrs);
+ } else {
+ trace_dma_alloc_sgt_err(dev, NULL, 0, size, gfp, dir, attrs);
}
return sgt;
}
--
2.35.1.1320.gc452695387.dirtyOn 2024-10-17 7:13 pm, Sean Anderson wrote:
> It can be surprising to the user if DMA functions are only traced on
> success. On failure, it can be unclear what the source of the problem
> is. Fix this by tracing all functions even when they fail. Cases where
> we BUG/WARN are skipped, since those should be sufficiently noisy
> already.
>
> Signed-off-by: Sean Anderson <sean.anderson@linux.dev>
> ---
>
> include/trace/events/dma.h | 41 ++++++++++++++++++++++++++++++++++++++
> kernel/dma/mapping.c | 27 +++++++++++++++++--------
> 2 files changed, 60 insertions(+), 8 deletions(-)
>
> diff --git a/include/trace/events/dma.h b/include/trace/events/dma.h
> index 9bc647f9ad4d..321cce327404 100644
> --- a/include/trace/events/dma.h
> +++ b/include/trace/events/dma.h
> @@ -161,6 +161,12 @@ DEFINE_EVENT(_dma_alloc, dma_alloc_pages,
> unsigned long attrs),
> TP_ARGS(dev, virt_addr, dma_addr, size, dir, flags, attrs));
>
> +DEFINE_EVENT(_dma_alloc, dma_alloc_sgt_err,
> + TP_PROTO(struct device *dev, void *virt_addr, dma_addr_t dma_addr,
> + size_t size, enum dma_data_direction dir, gfp_t flags,
> + unsigned long attrs),
> + TP_ARGS(dev, virt_addr, dma_addr, size, dir, flags, attrs));
> +
> TRACE_EVENT(dma_alloc_sgt,
> TP_PROTO(struct device *dev, struct sg_table *sgt, size_t size,
> enum dma_data_direction dir, gfp_t flags, unsigned long attrs),
> @@ -325,6 +331,41 @@ TRACE_EVENT(dma_map_sg,
> decode_dma_attrs(__entry->attrs))
> );
>
> +TRACE_EVENT(dma_map_sg_err,
> + TP_PROTO(struct device *dev, struct scatterlist *sgl, int nents,
> + int err, enum dma_data_direction dir, unsigned long attrs),
> + TP_ARGS(dev, sgl, nents, err, dir, attrs),
> +
> + TP_STRUCT__entry(
> + __string(device, dev_name(dev))
> + __dynamic_array(u64, phys_addrs, nents)
> + __field(int, err)
> + __field(enum dma_data_direction, dir)
> + __field(unsigned long, attrs)
> + ),
> +
> + TP_fast_assign(
> + struct scatterlist *sg;
> + int i;
> +
> + __assign_str(device);
> + for_each_sg(sgl, sg, nents, i)
> + ((u64 *)__get_dynamic_array(phys_addrs))[i] = sg_phys(sg);
> + __entry->err = err;
> + __entry->dir = dir;
> + __entry->attrs = attrs;
> + ),
> +
> + TP_printk("%s dir=%s dma_addrs=%s err=%d attrs=%s",
> + __get_str(device),
> + decode_dma_data_direction(__entry->dir),
> + __print_array(__get_dynamic_array(phys_addrs),
> + __get_dynamic_array_len(phys_addrs) /
> + sizeof(u64), sizeof(u64)),
> + __entry->err,
> + decode_dma_attrs(__entry->attrs))
> +);
> +
> TRACE_EVENT(dma_unmap_sg,
> TP_PROTO(struct device *dev, struct scatterlist *sgl, int nents,
> enum dma_data_direction dir, unsigned long attrs),
> diff --git a/kernel/dma/mapping.c b/kernel/dma/mapping.c
> index b8a6bc492fae..636dbb0629a4 100644
> --- a/kernel/dma/mapping.c
> +++ b/kernel/dma/mapping.c
> @@ -223,6 +223,7 @@ static int __dma_map_sg_attrs(struct device *dev, struct scatterlist *sg,
> debug_dma_map_sg(dev, sg, nents, ents, dir, attrs);
> } else if (WARN_ON_ONCE(ents != -EINVAL && ents != -ENOMEM &&
> ents != -EIO && ents != -EREMOTEIO)) {
> + trace_dma_map_sg_err(dev, sg, nents, ents, dir, attrs);
Isn't this just a case of moving the existing tracepoint up outside the
"if (ents > 0)" condition?
> return -EIO;
> }
>
> @@ -604,20 +605,26 @@ void *dma_alloc_attrs(struct device *dev, size_t size, dma_addr_t *dma_handle,
> if (WARN_ON_ONCE(flag & __GFP_COMP))
> return NULL;
>
> - if (dma_alloc_from_dev_coherent(dev, size, dma_handle, &cpu_addr))
> + if (dma_alloc_from_dev_coherent(dev, size, dma_handle, &cpu_addr)) {
> + trace_dma_alloc(dev, cpu_addr, *dma_handle, size,
> + DMA_BIDIRECTIONAL, flag, attrs);
> return cpu_addr;
> + }
>
> /* let the implementation decide on the zone to allocate from: */
> flag &= ~(__GFP_DMA | __GFP_DMA32 | __GFP_HIGHMEM);
>
> - if (dma_alloc_direct(dev, ops))
> + if (dma_alloc_direct(dev, ops)) {
> cpu_addr = dma_direct_alloc(dev, size, dma_handle, flag, attrs);
> - else if (use_dma_iommu(dev))
> + } else if (use_dma_iommu(dev)) {
> cpu_addr = iommu_dma_alloc(dev, size, dma_handle, flag, attrs);
> - else if (ops->alloc)
> + } else if (ops->alloc) {
> cpu_addr = ops->alloc(dev, size, dma_handle, flag, attrs);
> - else
> + } else {
> + trace_dma_alloc(dev, NULL, 0, size, DMA_BIDIRECTIONAL, flag,
> + attrs);
> return NULL;
Similarly just move this return down past the tracepoint, same as the
hunk below?
> + }
>
> trace_dma_alloc(dev, cpu_addr, *dma_handle, size, DMA_BIDIRECTIONAL,
> flag, attrs);
> @@ -642,11 +649,11 @@ void dma_free_attrs(struct device *dev, size_t size, void *cpu_addr,
> */
> WARN_ON(irqs_disabled());
>
> - if (!cpu_addr)
> - return;
> -
> trace_dma_free(dev, cpu_addr, dma_handle, size, DMA_BIDIRECTIONAL,
> attrs);
> + if (!cpu_addr)
> + return;
> +
> debug_dma_free_coherent(dev, size, cpu_addr, dma_handle);
> if (dma_alloc_direct(dev, ops))
> dma_direct_free(dev, size, cpu_addr, dma_handle, attrs);
> @@ -688,6 +695,8 @@ struct page *dma_alloc_pages(struct device *dev, size_t size,
> trace_dma_alloc_pages(dev, page_to_virt(page), *dma_handle,
> size, dir, gfp, 0);
> debug_dma_map_page(dev, page, 0, size, dir, *dma_handle, 0);
> + } else {
> + trace_dma_alloc_pages(dev, NULL, 0, size, dir, gfp, 0);
Could we move the page_to_virt() into the event definiton and let that
handle NULL, then similarly hoist the tracepoint out of the condition?
> }
> return page;
> }
> @@ -772,6 +781,8 @@ struct sg_table *dma_alloc_noncontiguous(struct device *dev, size_t size,
> sgt->nents = 1;
> trace_dma_alloc_sgt(dev, sgt, size, dir, gfp, attrs);
> debug_dma_map_sg(dev, sgt->sgl, sgt->orig_nents, 1, dir, attrs);
> + } else {
> + trace_dma_alloc_sgt_err(dev, NULL, 0, size, gfp, dir, attrs);
And again similarly here - if I'm interested in calls to
dma_alloc_contiguous(), I'd rather have a "dma_alloc_contiguous"
tracepoint which can tell me both the arguments and the result at a
glance, than have to remember to trace two distinct other things based
on internal details.
Thanks,
Robin.
> }
> return sgt;
> }
On 10/21/24 06:38, Robin Murphy wrote:
> On 2024-10-17 7:13 pm, Sean Anderson wrote:
>> It can be surprising to the user if DMA functions are only traced on
>> success. On failure, it can be unclear what the source of the problem
>> is. Fix this by tracing all functions even when they fail. Cases where
>> we BUG/WARN are skipped, since those should be sufficiently noisy
>> already.
>>
>> Signed-off-by: Sean Anderson <sean.anderson@linux.dev>
>> ---
>>
>> include/trace/events/dma.h | 41 ++++++++++++++++++++++++++++++++++++++
>> kernel/dma/mapping.c | 27 +++++++++++++++++--------
>> 2 files changed, 60 insertions(+), 8 deletions(-)
>>
>> diff --git a/include/trace/events/dma.h b/include/trace/events/dma.h
>> index 9bc647f9ad4d..321cce327404 100644
>> --- a/include/trace/events/dma.h
>> +++ b/include/trace/events/dma.h
>> @@ -161,6 +161,12 @@ DEFINE_EVENT(_dma_alloc, dma_alloc_pages,
>> unsigned long attrs),
>> TP_ARGS(dev, virt_addr, dma_addr, size, dir, flags, attrs));
>> +DEFINE_EVENT(_dma_alloc, dma_alloc_sgt_err,
>> + TP_PROTO(struct device *dev, void *virt_addr, dma_addr_t dma_addr,
>> + size_t size, enum dma_data_direction dir, gfp_t flags,
>> + unsigned long attrs),
>> + TP_ARGS(dev, virt_addr, dma_addr, size, dir, flags, attrs));
>> +
>> TRACE_EVENT(dma_alloc_sgt,
>> TP_PROTO(struct device *dev, struct sg_table *sgt, size_t size,
>> enum dma_data_direction dir, gfp_t flags, unsigned long attrs),
>> @@ -325,6 +331,41 @@ TRACE_EVENT(dma_map_sg,
>> decode_dma_attrs(__entry->attrs))
>> );
>> +TRACE_EVENT(dma_map_sg_err,
>> + TP_PROTO(struct device *dev, struct scatterlist *sgl, int nents,
>> + int err, enum dma_data_direction dir, unsigned long attrs),
>> + TP_ARGS(dev, sgl, nents, err, dir, attrs),
>> +
>> + TP_STRUCT__entry(
>> + __string(device, dev_name(dev))
>> + __dynamic_array(u64, phys_addrs, nents)
>> + __field(int, err)
>> + __field(enum dma_data_direction, dir)
>> + __field(unsigned long, attrs)
>> + ),
>> +
>> + TP_fast_assign(
>> + struct scatterlist *sg;
>> + int i;
>> +
>> + __assign_str(device);
>> + for_each_sg(sgl, sg, nents, i)
>> + ((u64 *)__get_dynamic_array(phys_addrs))[i] = sg_phys(sg);
>> + __entry->err = err;
>> + __entry->dir = dir;
>> + __entry->attrs = attrs;
>> + ),
>> +
>> + TP_printk("%s dir=%s dma_addrs=%s err=%d attrs=%s",
>> + __get_str(device),
>> + decode_dma_data_direction(__entry->dir),
>> + __print_array(__get_dynamic_array(phys_addrs),
>> + __get_dynamic_array_len(phys_addrs) /
>> + sizeof(u64), sizeof(u64)),
>> + __entry->err,
>> + decode_dma_attrs(__entry->attrs))
>> +);
>> +
>> TRACE_EVENT(dma_unmap_sg,
>> TP_PROTO(struct device *dev, struct scatterlist *sgl, int nents,
>> enum dma_data_direction dir, unsigned long attrs),
>> diff --git a/kernel/dma/mapping.c b/kernel/dma/mapping.c
>> index b8a6bc492fae..636dbb0629a4 100644
>> --- a/kernel/dma/mapping.c
>> +++ b/kernel/dma/mapping.c
>> @@ -223,6 +223,7 @@ static int __dma_map_sg_attrs(struct device *dev, struct scatterlist *sg,
>> debug_dma_map_sg(dev, sg, nents, ents, dir, attrs);
>> } else if (WARN_ON_ONCE(ents != -EINVAL && ents != -ENOMEM &&
>> ents != -EIO && ents != -EREMOTEIO)) {
>> + trace_dma_map_sg_err(dev, sg, nents, ents, dir, attrs);
>
> Isn't this just a case of moving the existing tracepoint up outside the "if (ents > 0)" condition?
This is a separate tracepoint (_err) to simplify collecting/printing the
addresses.
>
>> return -EIO;
>> }
>> @@ -604,20 +605,26 @@ void *dma_alloc_attrs(struct device *dev, size_t size, dma_addr_t *dma_handle,
>> if (WARN_ON_ONCE(flag & __GFP_COMP))
>> return NULL;
>> - if (dma_alloc_from_dev_coherent(dev, size, dma_handle, &cpu_addr))
>> + if (dma_alloc_from_dev_coherent(dev, size, dma_handle, &cpu_addr)) {
>> + trace_dma_alloc(dev, cpu_addr, *dma_handle, size,
>> + DMA_BIDIRECTIONAL, flag, attrs);
>> return cpu_addr;
>> + }
>> /* let the implementation decide on the zone to allocate from: */
>> flag &= ~(__GFP_DMA | __GFP_DMA32 | __GFP_HIGHMEM);
>> - if (dma_alloc_direct(dev, ops))
>> + if (dma_alloc_direct(dev, ops)) {
>> cpu_addr = dma_direct_alloc(dev, size, dma_handle, flag, attrs);
>> - else if (use_dma_iommu(dev))
>> + } else if (use_dma_iommu(dev)) {
>> cpu_addr = iommu_dma_alloc(dev, size, dma_handle, flag, attrs);
>> - else if (ops->alloc)
>> + } else if (ops->alloc) {
>> cpu_addr = ops->alloc(dev, size, dma_handle, flag, attrs);
>> - else
>> + } else {
>> + trace_dma_alloc(dev, NULL, 0, size, DMA_BIDIRECTIONAL, flag,
>> + attrs);
>> return NULL;
>
> Similarly just move this return down past the tracepoint, same as the hunk below?
OK
>> + }
>> trace_dma_alloc(dev, cpu_addr, *dma_handle, size, DMA_BIDIRECTIONAL,
>> flag, attrs);
>> @@ -642,11 +649,11 @@ void dma_free_attrs(struct device *dev, size_t size, void *cpu_addr,
>> */
>> WARN_ON(irqs_disabled());
>> - if (!cpu_addr)
>> - return;
>> -
>> trace_dma_free(dev, cpu_addr, dma_handle, size, DMA_BIDIRECTIONAL,
>> attrs);
>> + if (!cpu_addr)
>> + return;
>> +
>> debug_dma_free_coherent(dev, size, cpu_addr, dma_handle);
>> if (dma_alloc_direct(dev, ops))
>> dma_direct_free(dev, size, cpu_addr, dma_handle, attrs);
>> @@ -688,6 +695,8 @@ struct page *dma_alloc_pages(struct device *dev, size_t size,
>> trace_dma_alloc_pages(dev, page_to_virt(page), *dma_handle,
>> size, dir, gfp, 0);
>> debug_dma_map_page(dev, page, 0, size, dir, *dma_handle, 0);
>> + } else {
>> + trace_dma_alloc_pages(dev, NULL, 0, size, dir, gfp, 0);
>
> Could we move the page_to_virt() into the event definiton and let that handle NULL, then similarly hoist the tracepoint out of the condition?
Then we cannot reuse the event class for trace_dma_alloc, which does not
have a struct page available.
>> }
>> return page;
>> }
>> @@ -772,6 +781,8 @@ struct sg_table *dma_alloc_noncontiguous(struct device *dev, size_t size,
>> sgt->nents = 1;
>> trace_dma_alloc_sgt(dev, sgt, size, dir, gfp, attrs);
>> debug_dma_map_sg(dev, sgt->sgl, sgt->orig_nents, 1, dir, attrs);
>> + } else {
>> + trace_dma_alloc_sgt_err(dev, NULL, 0, size, gfp, dir, attrs);
>
> And again similarly here - if I'm interested in calls to dma_alloc_contiguous(), I'd rather have a "dma_alloc_contiguous" tracepoint which can tell me both the arguments and the result at a glance, than have to remember to trace two distinct other things based on internal details.
Isn't it a case of just doing `perf record -e 'dma:dma_alloc_sgt*'`?
Fundamentally, these events have different information available, and
dealing with this in a generic manner is inconvenient.
--Sean
© 2016 - 2024 Red Hat, Inc.