2011-06-01 08:38:40

by Namhyung Kim

Subject: [PATCH] blktrace: add FLUSH/FUA support

Add FLUSH/FUA support to blktrace. As FLUSH precedes WRITE and/or
FUA follows WRITE, use the same 'F' flag for both cases and
distinguish them by their (relative) position. The end results
look like (other flags might be shown also):

- WRITE: W
- WRITE_FLUSH: FW
- WRITE_FUA: WF
- WRITE_FLUSH_FUA: FWF

Note that BLK_TC_FLUSH should be the last one due to MASC_TC_BIT().
Otherwise it will cause unpleasant result because __REQ_FLUSH (23)
would be greater than ilog2(BLK_TC_FLUSH) + BLK_TC_SHIFT (16) so the
negative value. __REQ_FUA (12) doesn't have this problem.

Signed-off-by: Namhyung Kim <[email protected]>
Cc: Steven Rostedt <[email protected]>
Cc: Frederic Weisbecker <[email protected]>
Cc: Ingo Molnar <[email protected]>
---
include/linux/blktrace_api.h | 5 +++--
include/trace/events/block.h | 18 +++++++++---------
kernel/trace/blktrace.c | 21 ++++++++++++++++-----
3 files changed, 28 insertions(+), 16 deletions(-)

diff --git a/include/linux/blktrace_api.h b/include/linux/blktrace_api.h
index b22fb0d3db0f..05a688648bef 100644
--- a/include/linux/blktrace_api.h
+++ b/include/linux/blktrace_api.h
@@ -14,7 +14,7 @@
enum blktrace_cat {
BLK_TC_READ = 1 << 0, /* reads */
BLK_TC_WRITE = 1 << 1, /* writes */
- BLK_TC_BARRIER = 1 << 2, /* barrier */
+ BLK_TC_FUA = 1 << 2, /* fua requests */
BLK_TC_SYNC = 1 << 3, /* sync IO */
BLK_TC_SYNCIO = BLK_TC_SYNC,
BLK_TC_QUEUE = 1 << 4, /* queueing/merging */
@@ -28,8 +28,9 @@ enum blktrace_cat {
BLK_TC_META = 1 << 12, /* metadata */
BLK_TC_DISCARD = 1 << 13, /* discard requests */
BLK_TC_DRV_DATA = 1 << 14, /* binary per-driver data */
+ BLK_TC_FLUSH = 1 << 15, /* flush requests */

- BLK_TC_END = 1 << 15, /* only 16-bits, reminder */
+ BLK_TC_END = 1 << 15, /* we've run out of bits! */
};

#define BLK_TC_SHIFT (16)
diff --git a/include/trace/events/block.h b/include/trace/events/block.h
index bf366547da25..f21fea24216d 100644
--- a/include/trace/events/block.h
+++ b/include/trace/events/block.h
@@ -19,7 +19,7 @@ DECLARE_EVENT_CLASS(block_rq_with_error,
__field( sector_t, sector )
__field( unsigned int, nr_sector )
__field( int, errors )
- __array( char, rwbs, 6 )
+ __array( char, rwbs, 8 )
__dynamic_array( char, cmd, blk_cmd_buf_len(rq) )
),

@@ -104,7 +104,7 @@ DECLARE_EVENT_CLASS(block_rq,
__field( sector_t, sector )
__field( unsigned int, nr_sector )
__field( unsigned int, bytes )
- __array( char, rwbs, 6 )
+ __array( char, rwbs, 8 )
__array( char, comm, TASK_COMM_LEN )
__dynamic_array( char, cmd, blk_cmd_buf_len(rq) )
),
@@ -183,7 +183,7 @@ TRACE_EVENT(block_bio_bounce,
__field( dev_t, dev )
__field( sector_t, sector )
__field( unsigned int, nr_sector )
- __array( char, rwbs, 6 )
+ __array( char, rwbs, 8 )
__array( char, comm, TASK_COMM_LEN )
),

@@ -222,7 +222,7 @@ TRACE_EVENT(block_bio_complete,
__field( sector_t, sector )
__field( unsigned, nr_sector )
__field( int, error )
- __array( char, rwbs, 6 )
+ __array( char, rwbs, 8 )
),

TP_fast_assign(
@@ -249,7 +249,7 @@ DECLARE_EVENT_CLASS(block_bio,
__field( dev_t, dev )
__field( sector_t, sector )
__field( unsigned int, nr_sector )
- __array( char, rwbs, 6 )
+ __array( char, rwbs, 8 )
__array( char, comm, TASK_COMM_LEN )
),

@@ -321,7 +321,7 @@ DECLARE_EVENT_CLASS(block_get_rq,
__field( dev_t, dev )
__field( sector_t, sector )
__field( unsigned int, nr_sector )
- __array( char, rwbs, 6 )
+ __array( char, rwbs, 8 )
__array( char, comm, TASK_COMM_LEN )
),

@@ -456,7 +456,7 @@ TRACE_EVENT(block_split,
__field( dev_t, dev )
__field( sector_t, sector )
__field( sector_t, new_sector )
- __array( char, rwbs, 6 )
+ __array( char, rwbs, 8 )
__array( char, comm, TASK_COMM_LEN )
),

@@ -498,7 +498,7 @@ TRACE_EVENT(block_bio_remap,
__field( unsigned int, nr_sector )
__field( dev_t, old_dev )
__field( sector_t, old_sector )
- __array( char, rwbs, 6 )
+ __array( char, rwbs, 8 )
),

TP_fast_assign(
@@ -542,7 +542,7 @@ TRACE_EVENT(block_rq_remap,
__field( unsigned int, nr_sector )
__field( dev_t, old_dev )
__field( sector_t, old_sector )
- __array( char, rwbs, 6 )
+ __array( char, rwbs, 8 )
),

TP_fast_assign(
diff --git a/kernel/trace/blktrace.c b/kernel/trace/blktrace.c
index 6957aa298dfa..f5deb6f49e76 100644
--- a/kernel/trace/blktrace.c
+++ b/kernel/trace/blktrace.c
@@ -206,6 +206,8 @@ static void __blk_add_trace(struct blk_trace *bt, sector_t sector, int bytes,
what |= MASK_TC_BIT(rw, RAHEAD);
what |= MASK_TC_BIT(rw, META);
what |= MASK_TC_BIT(rw, DISCARD);
+ what |= MASK_TC_BIT(rw, FLUSH);
+ what |= MASK_TC_BIT(rw, FUA);

pid = tsk->pid;
if (act_log_check(bt, what, sector, pid))
@@ -1054,6 +1056,9 @@ static void fill_rwbs(char *rwbs, const struct blk_io_trace *t)
goto out;
}

+ if (tc & BLK_TC_FLUSH)
+ rwbs[i++] = 'F';
+
if (tc & BLK_TC_DISCARD)
rwbs[i++] = 'D';
else if (tc & BLK_TC_WRITE)
@@ -1063,10 +1068,10 @@ static void fill_rwbs(char *rwbs, const struct blk_io_trace *t)
else
rwbs[i++] = 'N';

+ if (tc & BLK_TC_FUA)
+ rwbs[i++] = 'F';
if (tc & BLK_TC_AHEAD)
rwbs[i++] = 'A';
- if (tc & BLK_TC_BARRIER)
- rwbs[i++] = 'B';
if (tc & BLK_TC_SYNC)
rwbs[i++] = 'S';
if (tc & BLK_TC_META)
@@ -1132,7 +1137,7 @@ typedef int (blk_log_action_t) (struct trace_iterator *iter, const char *act);

static int blk_log_action_classic(struct trace_iterator *iter, const char *act)
{
- char rwbs[6];
+ char rwbs[8];
unsigned long long ts = iter->ts;
unsigned long nsec_rem = do_div(ts, NSEC_PER_SEC);
unsigned secs = (unsigned long)ts;
@@ -1148,7 +1153,7 @@ static int blk_log_action_classic(struct trace_iterator *iter, const char *act)

static int blk_log_action(struct trace_iterator *iter, const char *act)
{
- char rwbs[6];
+ char rwbs[8];
const struct blk_io_trace *t = te_blk_io_trace(iter->ent);

fill_rwbs(rwbs, t);
@@ -1561,7 +1566,7 @@ static const struct {
} mask_maps[] = {
{ BLK_TC_READ, "read" },
{ BLK_TC_WRITE, "write" },
- { BLK_TC_BARRIER, "barrier" },
+ { BLK_TC_FUA, "fua" },
{ BLK_TC_SYNC, "sync" },
{ BLK_TC_QUEUE, "queue" },
{ BLK_TC_REQUEUE, "requeue" },
@@ -1573,6 +1578,7 @@ static const struct {
{ BLK_TC_META, "meta" },
{ BLK_TC_DISCARD, "discard" },
{ BLK_TC_DRV_DATA, "drv_data" },
+ { BLK_TC_FLUSH, "flush" },
};

static int blk_trace_str2mask(const char *str)
@@ -1788,6 +1794,9 @@ void blk_fill_rwbs(char *rwbs, u32 rw, int bytes)
{
int i = 0;

+ if (rw & REQ_FLUSH)
+ rwbs[i++] = 'F';
+
if (rw & WRITE)
rwbs[i++] = 'W';
else if (rw & REQ_DISCARD)
@@ -1797,6 +1806,8 @@ void blk_fill_rwbs(char *rwbs, u32 rw, int bytes)
else
rwbs[i++] = 'N';

+ if (rw & REQ_FUA)
+ rwbs[i++] = 'F';
if (rw & REQ_RAHEAD)
rwbs[i++] = 'A';
if (rw & REQ_SYNC)
--
1.7.5.2

2011-06-07 23:20:14

by Steven Rostedt

Subject: Re: [PATCH] blktrace: add FLUSH/FUA support

On Wed, 2011-06-01 at 17:38 +0900, Namhyung Kim wrote:
> Add FLUSH/FUA support to blktrace. As FLUSH precedes WRITE and/or
> FUA follows WRITE, use the same 'F' flag for both cases and
> distinguish them by their (relative) position. The end results
> look like (other flags might be shown also):
>
> - WRITE: W
> - WRITE_FLUSH: FW
> - WRITE_FUA: WF
> - WRITE_FLUSH_FUA: FWF
>
> Note that BLK_TC_FLUSH should be the last one due to MASC_TC_BIT().
> Otherwise it will cause unpleasant result because __REQ_FLUSH (23)
> would be greater than ilog2(BLK_TC_FLUSH) + BLK_TC_SHIFT (16) so the
> negative value. __REQ_FUA (12) doesn't have this problem.
>
> Signed-off-by: Namhyung Kim <[email protected]>
> Cc: Steven Rostedt <[email protected]>
> Cc: Frederic Weisbecker <[email protected]>
> Cc: Ingo Molnar <[email protected]>
> ---
> include/linux/blktrace_api.h | 5 +++--
> include/trace/events/block.h | 18 +++++++++---------
> kernel/trace/blktrace.c | 21 ++++++++++++++++-----
> 3 files changed, 28 insertions(+), 16 deletions(-)
>
> diff --git a/include/linux/blktrace_api.h b/include/linux/blktrace_api.h
> index b22fb0d3db0f..05a688648bef 100644
> --- a/include/linux/blktrace_api.h
> +++ b/include/linux/blktrace_api.h
> @@ -14,7 +14,7 @@
> enum blktrace_cat {
> BLK_TC_READ = 1 << 0, /* reads */
> BLK_TC_WRITE = 1 << 1, /* writes */
> - BLK_TC_BARRIER = 1 << 2, /* barrier */
> + BLK_TC_FUA = 1 << 2, /* fua requests */
> BLK_TC_SYNC = 1 << 3, /* sync IO */
> BLK_TC_SYNCIO = BLK_TC_SYNC,
> BLK_TC_QUEUE = 1 << 4, /* queueing/merging */
> @@ -28,8 +28,9 @@ enum blktrace_cat {
> BLK_TC_META = 1 << 12, /* metadata */
> BLK_TC_DISCARD = 1 << 13, /* discard requests */
> BLK_TC_DRV_DATA = 1 << 14, /* binary per-driver data */
> + BLK_TC_FLUSH = 1 << 15, /* flush requests */
>
> - BLK_TC_END = 1 << 15, /* only 16-bits, reminder */
> + BLK_TC_END = 1 << 15, /* we've run out of bits! */
> };
>
> #define BLK_TC_SHIFT (16)
> diff --git a/include/trace/events/block.h b/include/trace/events/block.h
> index bf366547da25..f21fea24216d 100644
> --- a/include/trace/events/block.h
> +++ b/include/trace/events/block.h
> @@ -19,7 +19,7 @@ DECLARE_EVENT_CLASS(block_rq_with_error,
> __field( sector_t, sector )
> __field( unsigned int, nr_sector )
> __field( int, errors )
> - __array( char, rwbs, 6 )
> + __array( char, rwbs, 8 )
> __dynamic_array( char, cmd, blk_cmd_buf_len(rq) )
> ),
>
> @@ -104,7 +104,7 @@ DECLARE_EVENT_CLASS(block_rq,
> __field( sector_t, sector )
> __field( unsigned int, nr_sector )
> __field( unsigned int, bytes )
> - __array( char, rwbs, 6 )
> + __array( char, rwbs, 8 )
> __array( char, comm, TASK_COMM_LEN )
> __dynamic_array( char, cmd, blk_cmd_buf_len(rq) )
> ),
> @@ -183,7 +183,7 @@ TRACE_EVENT(block_bio_bounce,
> __field( dev_t, dev )
> __field( sector_t, sector )
> __field( unsigned int, nr_sector )
> - __array( char, rwbs, 6 )
> + __array( char, rwbs, 8 )
> __array( char, comm, TASK_COMM_LEN )
> ),
>


Wouldn't it be better to replace all these "6" and "8"'s with a macro?
Then it would be a heck of a lot easier to change in the future.

-- Steve

2011-06-08 03:11:27

by Namhyung Kim

Subject: Re: [PATCH] blktrace: add FLUSH/FUA support

2011-06-07 (화), 19:20 -0400, Steven Rostedt:
> On Wed, 2011-06-01 at 17:38 +0900, Namhyung Kim wrote:
> > diff --git a/include/trace/events/block.h b/include/trace/events/block.h
> > index bf366547da25..f21fea24216d 100644
> > --- a/include/trace/events/block.h
> > +++ b/include/trace/events/block.h
> > @@ -19,7 +19,7 @@ DECLARE_EVENT_CLASS(block_rq_with_error,
> > __field( sector_t, sector )
> > __field( unsigned int, nr_sector )
> > __field( int, errors )
> > - __array( char, rwbs, 6 )
> > + __array( char, rwbs, 8 )
> > __dynamic_array( char, cmd, blk_cmd_buf_len(rq) )
> > ),
> >
> > @@ -104,7 +104,7 @@ DECLARE_EVENT_CLASS(block_rq,
> > __field( sector_t, sector )
> > __field( unsigned int, nr_sector )
> > __field( unsigned int, bytes )
> > - __array( char, rwbs, 6 )
> > + __array( char, rwbs, 8 )
> > __array( char, comm, TASK_COMM_LEN )
> > __dynamic_array( char, cmd, blk_cmd_buf_len(rq) )
> > ),
> > @@ -183,7 +183,7 @@ TRACE_EVENT(block_bio_bounce,
> > __field( dev_t, dev )
> > __field( sector_t, sector )
> > __field( unsigned int, nr_sector )
> > - __array( char, rwbs, 6 )
> > + __array( char, rwbs, 8 )
> > __array( char, comm, TASK_COMM_LEN )
> > ),
> >
>
>
> Wouldn't it be better to replace all these "6" and "8"'s with a macro?
> Then it would be a heck of a lot easier to change in the future.
>
> -- Steve
>
>

Will fix in the next version, Thanks.


--
Regards,
Namhyung Kim