Re: [PATCH RESEND] block: Add ioprio to block_rq tracepoint

[Steven Rostedt]

On Mon, 15 Apr 2024 14:07:10 +0800
Dongliang Cui <dongliang.cui@xxxxxxxxxx> wrote:

> Sometimes we need to track the processing order of requests with
> ioprio set. So the ioprio of request can be useful information.
> 
> Example：
> 
> block_rq_insert: 8,0 WS 4096 () 16573296 + 8 rt,4 [highpool[1]]
> block_rq_issue: 8,0 WS 4096 () 16573296 + 8 rt,4 [kworker/7:0H]
> block_rq_complete: 8,0 WS () 16573296 + 8 rt,4 [0]

Note, it's up to the subsystem to take trace event patches.

> 
> Signed-off-by: Dongliang Cui <dongliang.cui@xxxxxxxxxx>
> ---
>  include/linux/blktrace_api.h |  2 ++
>  include/trace/events/block.h | 63 ++++++++++++++++++++++--------------
>  kernel/trace/blktrace.c      | 11 +++++++
>  3 files changed, 51 insertions(+), 25 deletions(-)
> 
> diff --git a/include/linux/blktrace_api.h b/include/linux/blktrace_api.h
> index 122c62e561fc..adb0333efbdb 100644
> --- a/include/linux/blktrace_api.h
> +++ b/include/linux/blktrace_api.h
> @@ -112,6 +112,8 @@ struct compat_blk_user_trace_setup {
>  
>  void blk_fill_rwbs(char *rwbs, blk_opf_t opf);
>  
> +void blk_fill_ioprio(u32 ioprio, char *ioprio_class, u32 *ioprio_value);
> +
>  static inline sector_t blk_rq_trace_sector(struct request *rq)
>  {
>  	/*
> diff --git a/include/trace/events/block.h b/include/trace/events/block.h
> index 0e128ad51460..1d41fade160a 100644
> --- a/include/trace/events/block.h
> +++ b/include/trace/events/block.h
> @@ -10,7 +10,8 @@
>  #include <linux/buffer_head.h>
>  #include <linux/tracepoint.h>
>  
> -#define RWBS_LEN	8
> +#define RWBS_LEN		8
> +#define IOPRIO_CLASS_LEN	8
>  
>  #ifdef CONFIG_BUFFER_HEAD
>  DECLARE_EVENT_CLASS(block_buffer,
> @@ -79,11 +80,13 @@ TRACE_EVENT(block_rq_requeue,
>  	TP_ARGS(rq),
>  
>  	TP_STRUCT__entry(
> -		__field(  dev_t,	dev			)
> -		__field(  sector_t,	sector			)
> -		__field(  unsigned int,	nr_sector		)
> -		__array(  char,		rwbs,	RWBS_LEN	)
> -		__dynamic_array( char,	cmd,	1		)
> +		__field(  dev_t,	dev				)
> +		__field(  sector_t,	sector				)
> +		__field(  unsigned int,	nr_sector			)
> +		__array(  char,		rwbs,	RWBS_LEN		)
> +		__array(  char,		ioprio_class, IOPRIO_CLASS_LEN	)
> +		__field(  unsigned int, ioprio_value			)
> +		__dynamic_array( char,	cmd,	1			)
>  	),
>  
>  	TP_fast_assign(
> @@ -92,14 +95,16 @@ TRACE_EVENT(block_rq_requeue,
>  		__entry->nr_sector = blk_rq_trace_nr_sectors(rq);
>  
>  		blk_fill_rwbs(__entry->rwbs, rq->cmd_flags);
> +		blk_fill_ioprio(rq->ioprio, __entry->ioprio_class, &__entry->ioprio_value);
>  		__get_str(cmd)[0] = '\0';
>  	),
>  
> -	TP_printk("%d,%d %s (%s) %llu + %u [%d]",
> +	TP_printk("%d,%d %s (%s) %llu + %u %s,%u [%d]",
>  		  MAJOR(__entry->dev), MINOR(__entry->dev),
>  		  __entry->rwbs, __get_str(cmd),
>  		  (unsigned long long)__entry->sector,
> -		  __entry->nr_sector, 0)
> +		  __entry->nr_sector, __entry->ioprio_class,
> +		  __entry->ioprio_value, 0)
>  );


> diff --git a/kernel/trace/blktrace.c b/kernel/trace/blktrace.c
> index d5d94510afd3..e55aa49f94db 100644
> --- a/kernel/trace/blktrace.c
> +++ b/kernel/trace/blktrace.c
> @@ -19,6 +19,7 @@
>  #include <linux/uaccess.h>
>  #include <linux/list.h>
>  #include <linux/blk-cgroup.h>
> +#include <linux/ioprio.h>
>  
>  #include "../../block/blk.h"
>  
> @@ -26,6 +27,9 @@
>  
>  #include "trace_output.h"
>  
> +/* Type of ioprio */
> +static char *classes[] = {"none", "rt", "be", "idle"};
> +
>  #ifdef CONFIG_BLK_DEV_IO_TRACE
>  
>  static unsigned int blktrace_seq __read_mostly = 1;
> @@ -1914,5 +1918,12 @@ void blk_fill_rwbs(char *rwbs, blk_opf_t opf)
>  }
>  EXPORT_SYMBOL_GPL(blk_fill_rwbs);
>  
> +void blk_fill_ioprio(u32 ioprio, char *ioprio_class, u32 *ioprio_value)
> +{
> +	memcpy(ioprio_class, classes[(ioprio >> IOPRIO_CLASS_SHIFT) & 0x3], IOPRIO_CLASS_LEN);
> +	*ioprio_value = ioprio & 0xff;
> +}
> +EXPORT_SYMBOL_GPL(blk_fill_ioprio);
> +
>  #endif /* CONFIG_EVENT_TRACING */
>  

Instead of doing:

		blk_fill_ioprio(rq->ioprio, __entry->ioprio_class, &__entry->ioprio_value);


Why not do:

	__field(	int,	ioprio_class	)

[..]

	__entry->ioprio_class = rq->ioprio >> IOPRIO_CLASS_SHIFT & 0x3;

[..]

	TP_printk("%d,%d %s (%s) %llu + %u %s,%u [%d]",
		  MAJOR(__entry->dev), MINOR(__entry->dev),
		  __entry->rwbs, __get_str(cmd),
		  (unsigned long long)__entry->sector,
		  __entry->nr_sector,
		  __print_symbolic(ioclass,
			{ 0, "none" }, { 1, "rt" }, { 2, "bt" }, { 3, "idle" }),
		  __entry->ioprio_value, 0)
);

 ?

Then you don't need to save a string into the buffer, and just print
the mapping on reading of the buffer.

-- Steve