* [PATCHv2] tracing/events: Add bounce tracing to swiotbl
@ 2013-09-04 20:11 Zoltan Kiss
0 siblings, 0 replies; 6+ messages in thread
From: Zoltan Kiss @ 2013-09-04 20:11 UTC (permalink / raw)
To: Konrad Rzeszutek Wilk, Jeremy Fitzhardinge, Steven Rostedt,
Frederic Weisbecker, Ingo Molnar, linux-kernel, xen-devel,
virtualization
Cc: Zoltan Kiss
Ftrace is currently not able to detect when SWIOTLB has to do double buffering.
Under Xen you can only see it indirectly in function_graph, when
xen_swiotlb_map_page() doesn't stop after range_straddles_page_boundary(), but
calls spinlock functions, memcpy() and xen_phys_to_bus() as well. This patch
introduces the swiotlb:swiotlb_bounced event, which also prints out the
following informations to help you find out why bouncing happened:
dev_name: 0000:08:00.0 dma_mask=ffffffffffffffff dev_addr=9149f000 size=32768
swiotlb_force=0
If you use Xen, and (dev_addr + size + 1) > dma_mask, the buffer is out of the
device's DMA range. If swiotlb_force == 1, you should really change the kernel
parameters. Otherwise, the buffer is not contiguous in mfn space.
Signed-off-by: Zoltan Kiss <zoltan.kiss@citrix.com>
---
drivers/xen/swiotlb-xen.c | 5 +++++
include/trace/events/swiotlb.h | 46 ++++++++++++++++++++++++++++++++++++++++
lib/swiotlb.c | 4 ++++
3 files changed, 55 insertions(+)
diff --git a/drivers/xen/swiotlb-xen.c b/drivers/xen/swiotlb-xen.c
index aadffcf..67a4b77 100644
--- a/drivers/xen/swiotlb-xen.c
+++ b/drivers/xen/swiotlb-xen.c
@@ -42,6 +42,9 @@
#include <xen/page.h>
#include <xen/xen-ops.h>
#include <xen/hvc-console.h>
+
+#define CREATE_TRACE_POINTS
+#include <trace/events/swiotlb.h>
/*
* Used to do a quick range check in swiotlb_tbl_unmap_single and
* swiotlb_tbl_sync_single_*, to see if the memory was in fact allocated by this
@@ -358,6 +361,8 @@ dma_addr_t xen_swiotlb_map_page(struct device *dev, struct page *page,
/*
* Oh well, have to allocate and map a bounce buffer.
*/
+ trace_swiotlb_bounced(dev, dev_addr, size, swiotlb_force);
+
map = swiotlb_tbl_map_single(dev, start_dma_addr, phys, size, dir);
if (map == SWIOTLB_MAP_ERROR)
return DMA_ERROR_CODE;
diff --git a/include/trace/events/swiotlb.h b/include/trace/events/swiotlb.h
new file mode 100644
index 0000000..6d21410
--- /dev/null
+++ b/include/trace/events/swiotlb.h
@@ -0,0 +1,46 @@
+#undef TRACE_SYSTEM
+#define TRACE_SYSTEM swiotlb
+
+#if !defined(_TRACE_SWIOTLB_H) || defined(TRACE_HEADER_MULTI_READ)
+#define _TRACE_SWIOTLB_H
+
+#include <linux/tracepoint.h>
+
+TRACE_EVENT(swiotlb_bounced,
+
+ TP_PROTO(struct device *dev,
+ dma_addr_t dev_addr,
+ size_t size,
+ int swiotlb_force),
+
+ TP_ARGS(dev, dev_addr, size, swiotlb_force),
+
+ TP_STRUCT__entry(
+ __string( dev_name, dev_name(dev) )
+ __field( u64, dma_mask )
+ __field( dma_addr_t, dev_addr )
+ __field( size_t, size )
+ __field( int, swiotlb_force )
+ ),
+
+ TP_fast_assign(
+ __assign_str(dev_name, dev_name(dev));
+ __entry->dma_mask = (dev->dma_mask ? *dev->dma_mask : 0);
+ __entry->dev_addr = dev_addr;
+ __entry->size = size;
+ __entry->swiotlb_force = swiotlb_force;
+ ),
+
+ TP_printk("dev_name: %s dma_mask=%llx dev_addr=%llx "
+ "size=%zu swiotlb_force=%x",
+ __get_str(dev_name),
+ __entry->dma_mask,
+ (unsigned long long)__entry->dev_addr,
+ __entry->size,
+ __entry->swiotlb_force)
+);
+
+#endif /* _TRACE_SWIOTLB_H */
+
+/* This part must be outside protection */
+#include <trace/define_trace.h>
diff --git a/lib/swiotlb.c b/lib/swiotlb.c
index d23762e..d69ecbe 100644
--- a/lib/swiotlb.c
+++ b/lib/swiotlb.c
@@ -38,6 +38,8 @@
#include <linux/bootmem.h>
#include <linux/iommu-helper.h>
+#include <trace/events/swiotlb.h>
+
#define OFFSET(val,align) ((unsigned long) \
( (val) & ( (align) - 1)))
@@ -726,6 +728,8 @@ dma_addr_t swiotlb_map_page(struct device *dev, struct page *page,
if (dma_capable(dev, dev_addr, size) && !swiotlb_force)
return dev_addr;
+ trace_swiotlb_bounced(dev, dev_addr, size, swiotlb_force);
+
/* Oh well, have to allocate and map a bounce buffer. */
map = map_single(dev, phys, size, dir);
if (map == SWIOTLB_MAP_ERROR) {
^ permalink raw reply related [flat|nested] 6+ messages in thread
* Re: [PATCHv2] tracing/events: Add bounce tracing to swiotbl
[not found] <1378325465-10384-1-git-send-email-zoltan.kiss@citrix.com>
@ 2013-09-18 16:04 ` Zoltan Kiss
2013-09-25 17:56 ` Konrad Rzeszutek Wilk
0 siblings, 1 reply; 6+ messages in thread
From: Zoltan Kiss @ 2013-09-18 16:04 UTC (permalink / raw)
To: Zoltan Kiss
Cc: Jeremy Fitzhardinge, xen-devel, Konrad Rzeszutek Wilk,
Frederic Weisbecker, linux-kernel, Steven Rostedt, virtualization,
Ingo Molnar
Hi,
I haven't got a reply in the past 2 weeks, so I would like to bump the
patch, just to make sure it haven't fell off the radar.
Zoli
On 04/09/13 21:11, Zoltan Kiss wrote:
> Ftrace is currently not able to detect when SWIOTLB has to do double buffering.
> Under Xen you can only see it indirectly in function_graph, when
> xen_swiotlb_map_page() doesn't stop after range_straddles_page_boundary(), but
> calls spinlock functions, memcpy() and xen_phys_to_bus() as well. This patch
> introduces the swiotlb:swiotlb_bounced event, which also prints out the
> following informations to help you find out why bouncing happened:
>
> dev_name: 0000:08:00.0 dma_mask=ffffffffffffffff dev_addr=9149f000 size=32768
> swiotlb_force=0
>
> If you use Xen, and (dev_addr + size + 1) > dma_mask, the buffer is out of the
> device's DMA range. If swiotlb_force == 1, you should really change the kernel
> parameters. Otherwise, the buffer is not contiguous in mfn space.
>
> Signed-off-by: Zoltan Kiss <zoltan.kiss@citrix.com>
> ---
> drivers/xen/swiotlb-xen.c | 5 +++++
> include/trace/events/swiotlb.h | 46 ++++++++++++++++++++++++++++++++++++++++
> lib/swiotlb.c | 4 ++++
> 3 files changed, 55 insertions(+)
>
> diff --git a/drivers/xen/swiotlb-xen.c b/drivers/xen/swiotlb-xen.c
> index aadffcf..67a4b77 100644
> --- a/drivers/xen/swiotlb-xen.c
> +++ b/drivers/xen/swiotlb-xen.c
> @@ -42,6 +42,9 @@
> #include <xen/page.h>
> #include <xen/xen-ops.h>
> #include <xen/hvc-console.h>
> +
> +#define CREATE_TRACE_POINTS
> +#include <trace/events/swiotlb.h>
> /*
> * Used to do a quick range check in swiotlb_tbl_unmap_single and
> * swiotlb_tbl_sync_single_*, to see if the memory was in fact allocated by this
> @@ -358,6 +361,8 @@ dma_addr_t xen_swiotlb_map_page(struct device *dev, struct page *page,
> /*
> * Oh well, have to allocate and map a bounce buffer.
> */
> + trace_swiotlb_bounced(dev, dev_addr, size, swiotlb_force);
> +
> map = swiotlb_tbl_map_single(dev, start_dma_addr, phys, size, dir);
> if (map == SWIOTLB_MAP_ERROR)
> return DMA_ERROR_CODE;
> diff --git a/include/trace/events/swiotlb.h b/include/trace/events/swiotlb.h
> new file mode 100644
> index 0000000..6d21410
> --- /dev/null
> +++ b/include/trace/events/swiotlb.h
> @@ -0,0 +1,46 @@
> +#undef TRACE_SYSTEM
> +#define TRACE_SYSTEM swiotlb
> +
> +#if !defined(_TRACE_SWIOTLB_H) || defined(TRACE_HEADER_MULTI_READ)
> +#define _TRACE_SWIOTLB_H
> +
> +#include <linux/tracepoint.h>
> +
> +TRACE_EVENT(swiotlb_bounced,
> +
> + TP_PROTO(struct device *dev,
> + dma_addr_t dev_addr,
> + size_t size,
> + int swiotlb_force),
> +
> + TP_ARGS(dev, dev_addr, size, swiotlb_force),
> +
> + TP_STRUCT__entry(
> + __string( dev_name, dev_name(dev) )
> + __field( u64, dma_mask )
> + __field( dma_addr_t, dev_addr )
> + __field( size_t, size )
> + __field( int, swiotlb_force )
> + ),
> +
> + TP_fast_assign(
> + __assign_str(dev_name, dev_name(dev));
> + __entry->dma_mask = (dev->dma_mask ? *dev->dma_mask : 0);
> + __entry->dev_addr = dev_addr;
> + __entry->size = size;
> + __entry->swiotlb_force = swiotlb_force;
> + ),
> +
> + TP_printk("dev_name: %s dma_mask=%llx dev_addr=%llx "
> + "size=%zu swiotlb_force=%x",
> + __get_str(dev_name),
> + __entry->dma_mask,
> + (unsigned long long)__entry->dev_addr,
> + __entry->size,
> + __entry->swiotlb_force)
> +);
> +
> +#endif /* _TRACE_SWIOTLB_H */
> +
> +/* This part must be outside protection */
> +#include <trace/define_trace.h>
> diff --git a/lib/swiotlb.c b/lib/swiotlb.c
> index d23762e..d69ecbe 100644
> --- a/lib/swiotlb.c
> +++ b/lib/swiotlb.c
> @@ -38,6 +38,8 @@
> #include <linux/bootmem.h>
> #include <linux/iommu-helper.h>
>
> +#include <trace/events/swiotlb.h>
> +
> #define OFFSET(val,align) ((unsigned long) \
> ( (val) & ( (align) - 1)))
>
> @@ -726,6 +728,8 @@ dma_addr_t swiotlb_map_page(struct device *dev, struct page *page,
> if (dma_capable(dev, dev_addr, size) && !swiotlb_force)
> return dev_addr;
>
> + trace_swiotlb_bounced(dev, dev_addr, size, swiotlb_force);
> +
> /* Oh well, have to allocate and map a bounce buffer. */
> map = map_single(dev, phys, size, dir);
> if (map == SWIOTLB_MAP_ERROR) {
>
^ permalink raw reply [flat|nested] 6+ messages in thread
* Re: [PATCHv2] tracing/events: Add bounce tracing to swiotbl
2013-09-18 16:04 ` [PATCHv2] tracing/events: Add bounce tracing to swiotbl Zoltan Kiss
@ 2013-09-25 17:56 ` Konrad Rzeszutek Wilk
2013-09-25 18:26 ` Steven Rostedt
2013-09-26 16:58 ` Zoltan Kiss
0 siblings, 2 replies; 6+ messages in thread
From: Konrad Rzeszutek Wilk @ 2013-09-25 17:56 UTC (permalink / raw)
To: Zoltan Kiss
Cc: Jeremy Fitzhardinge, xen-devel, Frederic Weisbecker, linux-kernel,
Steven Rostedt, virtualization, Ingo Molnar
On Wed, Sep 18, 2013 at 05:04:17PM +0100, Zoltan Kiss wrote:
> Hi,
>
> I haven't got a reply in the past 2 weeks, so I would like to bump
> the patch, just to make sure it haven't fell off the radar.
Hey,
I have this in my queue to put on 3.13 as it is past the merge window.
.. with that in mind:
.. snip..
> >+ TP_printk("dev_name: %s dma_mask=%llx dev_addr=%llx "
> >+ "size=%zu swiotlb_force=%x",
> >+ __get_str(dev_name),
> >+ __entry->dma_mask,
> >+ (unsigned long long)__entry->dev_addr,
> >+ __entry->size,
> >+ __entry->swiotlb_force)
Would it make sense to do something like this:
__entry->swiotlb_force ? "swiotlb_force" : "")
^ permalink raw reply [flat|nested] 6+ messages in thread
* Re: [PATCHv2] tracing/events: Add bounce tracing to swiotbl
2013-09-25 17:56 ` Konrad Rzeszutek Wilk
@ 2013-09-25 18:26 ` Steven Rostedt
2013-09-26 16:58 ` Zoltan Kiss
1 sibling, 0 replies; 6+ messages in thread
From: Steven Rostedt @ 2013-09-25 18:26 UTC (permalink / raw)
To: Konrad Rzeszutek Wilk
Cc: Jeremy Fitzhardinge, xen-devel, Frederic Weisbecker, linux-kernel,
virtualization, Ingo Molnar, Zoltan Kiss
On Wed, 25 Sep 2013 13:56:49 -0400
Konrad Rzeszutek Wilk <konrad.wilk@oracle.com> wrote:
> .. snip..
> > >+ TP_printk("dev_name: %s dma_mask=%llx dev_addr=%llx "
> > >+ "size=%zu swiotlb_force=%x",
> > >+ __get_str(dev_name),
> > >+ __entry->dma_mask,
> > >+ (unsigned long long)__entry->dev_addr,
> > >+ __entry->size,
> > >+ __entry->swiotlb_force)
>
> Would it make sense to do something like this:
>
> __entry->swiotlb_force ? "swiotlb_force" : "")
I think that's fine. I do believe that the libtraceevents can parse the
"?:" syntax.
-- Steve
^ permalink raw reply [flat|nested] 6+ messages in thread
* Re: [PATCHv2] tracing/events: Add bounce tracing to swiotbl
2013-09-25 17:56 ` Konrad Rzeszutek Wilk
2013-09-25 18:26 ` Steven Rostedt
@ 2013-09-26 16:58 ` Zoltan Kiss
2013-09-26 18:03 ` Konrad Rzeszutek Wilk
1 sibling, 1 reply; 6+ messages in thread
From: Zoltan Kiss @ 2013-09-26 16:58 UTC (permalink / raw)
To: Konrad Rzeszutek Wilk
Cc: Jeremy Fitzhardinge, xen-devel, Frederic Weisbecker, linux-kernel,
Steven Rostedt, virtualization, Ingo Molnar
On 25/09/13 18:56, Konrad Rzeszutek Wilk wrote:
> On Wed, Sep 18, 2013 at 05:04:17PM +0100, Zoltan Kiss wrote:
>> Hi,
>>
>> I haven't got a reply in the past 2 weeks, so I would like to bump
>> the patch, just to make sure it haven't fell off the radar.
>
> Hey,
>
> I have this in my queue to put on 3.13 as it is past the merge window.
> .. with that in mind:
>
>
> .. snip..
>>> + TP_printk("dev_name: %s dma_mask=%llx dev_addr=%llx "
>>> + "size=%zu swiotlb_force=%x",
>>> + __get_str(dev_name),
>>> + __entry->dma_mask,
>>> + (unsigned long long)__entry->dev_addr,
>>> + __entry->size,
>>> + __entry->swiotlb_force)
>
> Would it make sense to do something like this:
>
> __entry->swiotlb_force ? "swiotlb_force" : "")
>
I would then rather do:
+ TP_printk("dev_name: %s dma_mask=%llx dev_addr=%llx "
+ "size=%zu swiotlb_force=",
+ __entry->swiotlb_force ? " yes" : "no",
+ __get_str(dev_name),
Or do you mean?:
+ TP_printk("dev_name: %s dma_mask=%llx dev_addr=%llx "
+ "size=%zu",
+ __entry->swiotlb_force ? " swiotlb_force" : "",
+ __get_str(dev_name),
This one doesn't tell you explicitly if swiotlb_force is NOT set, maybe
that's not so good? And adds a bit of complexity to your grep regexp?
Either way is fine with me, but I think "swiotlb_force=0|1" is also
pretty straightforward to understand, and I guess it makes printk
slightly faster (I assume the conditional operator gives a little bit of
overhead)
Regards,
Zoli
^ permalink raw reply [flat|nested] 6+ messages in thread
* Re: [PATCHv2] tracing/events: Add bounce tracing to swiotbl
2013-09-26 16:58 ` Zoltan Kiss
@ 2013-09-26 18:03 ` Konrad Rzeszutek Wilk
0 siblings, 0 replies; 6+ messages in thread
From: Konrad Rzeszutek Wilk @ 2013-09-26 18:03 UTC (permalink / raw)
To: Zoltan Kiss
Cc: Jeremy Fitzhardinge, xen-devel, Frederic Weisbecker, linux-kernel,
Steven Rostedt, virtualization, Ingo Molnar
Zoltan Kiss <zoltan.kiss@citrix.com> wrote:
>On 25/09/13 18:56, Konrad Rzeszutek Wilk wrote:
>> On Wed, Sep 18, 2013 at 05:04:17PM +0100, Zoltan Kiss wrote:
>>> Hi,
>>>
>>> I haven't got a reply in the past 2 weeks, so I would like to bump
>>> the patch, just to make sure it haven't fell off the radar.
>>
>> Hey,
>>
>> I have this in my queue to put on 3.13 as it is past the merge
>window.
>> .. with that in mind:
>>
>>
>> .. snip..
>>>> + TP_printk("dev_name: %s dma_mask=%llx dev_addr=%llx "
>>>> + "size=%zu swiotlb_force=%x",
>>>> + __get_str(dev_name),
>>>> + __entry->dma_mask,
>>>> + (unsigned long long)__entry->dev_addr,
>>>> + __entry->size,
>>>> + __entry->swiotlb_force)
>>
>> Would it make sense to do something like this:
>>
>> __entry->swiotlb_force ? "swiotlb_force" : "")
>>
>
>I would then rather do:
>
>+ TP_printk("dev_name: %s dma_mask=%llx dev_addr=%llx "
>+ "size=%zu swiotlb_force=",
>+ __entry->swiotlb_force ? " yes" : "no",
>+ __get_str(dev_name),
>
>Or do you mean?:
>
>+ TP_printk("dev_name: %s dma_mask=%llx dev_addr=%llx "
>+ "size=%zu",
>+ __entry->swiotlb_force ? " swiotlb_force" : "",
>+ __get_str(dev_name),
>
>This one doesn't tell you explicitly if swiotlb_force is NOT set, maybe
>
>that's not so good? And adds a bit of complexity to your grep regexp?
>Either way is fine with me, but I think "swiotlb_force=0|1" is also
>pretty straightforward to understand, and I guess it makes printk
>slightly faster (I assume the conditional operator gives a little bit
>of
>overhead)
>
>Regards,
>
>Zoli
I was thinking it would be good to print the swiotlb-force only when it is enabled. So your second one would be it.
^ permalink raw reply [flat|nested] 6+ messages in thread
end of thread, other threads:[~2013-09-26 18:03 UTC | newest]
Thread overview: 6+ messages (download: mbox.gz follow: Atom feed
-- links below jump to the message on this page --
[not found] <1378325465-10384-1-git-send-email-zoltan.kiss@citrix.com>
2013-09-18 16:04 ` [PATCHv2] tracing/events: Add bounce tracing to swiotbl Zoltan Kiss
2013-09-25 17:56 ` Konrad Rzeszutek Wilk
2013-09-25 18:26 ` Steven Rostedt
2013-09-26 16:58 ` Zoltan Kiss
2013-09-26 18:03 ` Konrad Rzeszutek Wilk
2013-09-04 20:11 Zoltan Kiss
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).