Re: [PATCH 1/2] io_uring/sqpoll: switch away from getrusage() for CPU accounting

[Jens Axboe <axboe@kernel.dk>]

On 10/21/25 5:35 PM, Gabriel Krisman Bertazi wrote:
> Jens Axboe <axboe@kernel.dk> writes:
> 
>> getrusage() does a lot more than what the SQPOLL accounting needs, the
>> latter only cares about (and uses) the stime. Rather than do a full
>> RUSAGE_SELF summation, just query the used stime instead.
>>
>> Cc: stable@vger.kernel.org
>> Fixes: 3fcb9d17206e ("io_uring/sqpoll: statistics of the true utilization of sq threads")
>> Signed-off-by: Jens Axboe <axboe@kernel.dk>
>> ---
>>  io_uring/fdinfo.c |  9 +++++----
>>  io_uring/sqpoll.c | 34 ++++++++++++++++++++--------------
>>  io_uring/sqpoll.h |  1 +
>>  3 files changed, 26 insertions(+), 18 deletions(-)
>>
>> diff --git a/io_uring/fdinfo.c b/io_uring/fdinfo.c
>> index ff3364531c77..966e06b078f6 100644
>> --- a/io_uring/fdinfo.c
>> +++ b/io_uring/fdinfo.c
>> @@ -59,7 +59,6 @@ static void __io_uring_show_fdinfo(struct io_ring_ctx *ctx, struct seq_file *m)
>>  {
>>  	struct io_overflow_cqe *ocqe;
>>  	struct io_rings *r = ctx->rings;
>> -	struct rusage sq_usage;
>>  	unsigned int sq_mask = ctx->sq_entries - 1, cq_mask = ctx->cq_entries - 1;
>>  	unsigned int sq_head = READ_ONCE(r->sq.head);
>>  	unsigned int sq_tail = READ_ONCE(r->sq.tail);
>> @@ -152,14 +151,16 @@ static void __io_uring_show_fdinfo(struct io_ring_ctx *ctx, struct seq_file *m)
>>  		 * thread termination.
>>  		 */
>>  		if (tsk) {
>> +			struct timespec64 ts;
>> +
>>  			get_task_struct(tsk);
>>  			rcu_read_unlock();
>> -			getrusage(tsk, RUSAGE_SELF, &sq_usage);
>> +			ts = io_sq_cpu_time(tsk);
>>  			put_task_struct(tsk);
>>  			sq_pid = sq->task_pid;
>>  			sq_cpu = sq->sq_cpu;
>> -			sq_total_time = (sq_usage.ru_stime.tv_sec * 1000000
>> -					 + sq_usage.ru_stime.tv_usec);
>> +			sq_total_time = (ts.tv_sec * 1000000
>> +					 + ts.tv_nsec / 1000);
>>  			sq_work_time = sq->work_time;
>>  		} else {
>>  			rcu_read_unlock();
>> diff --git a/io_uring/sqpoll.c b/io_uring/sqpoll.c
>> index a3f11349ce06..8705b0aa82e0 100644
>> --- a/io_uring/sqpoll.c
>> +++ b/io_uring/sqpoll.c
>> @@ -11,6 +11,7 @@
>>  #include <linux/audit.h>
>>  #include <linux/security.h>
>>  #include <linux/cpuset.h>
>> +#include <linux/sched/cputime.h>
>>  #include <linux/io_uring.h>
>>  
>>  #include <uapi/linux/io_uring.h>
>> @@ -169,6 +170,22 @@ static inline bool io_sqd_events_pending(struct io_sq_data *sqd)
>>  	return READ_ONCE(sqd->state);
>>  }
>>  
>> +struct timespec64 io_sq_cpu_time(struct task_struct *tsk)
>> +{
>> +	u64 utime, stime;
>> +
>> +	task_cputime_adjusted(tsk, &utime, &stime);
>> +	return ns_to_timespec64(stime);
>> +}
>> +
>> +static void io_sq_update_worktime(struct io_sq_data *sqd, struct timespec64 start)
>> +{
>> +	struct timespec64 ts;
>> +
>> +	ts = timespec64_sub(io_sq_cpu_time(current), start);
>> +	sqd->work_time += ts.tv_sec * 1000000 + ts.tv_nsec / 1000;
>> +}
> 
> Hi Jens,
> 
> Patch looks good. I'd just mention you are converting ns to timespec64,
> just to convert it back to ms when writing to sqd->work_time and
> sq_total_time.  I think wraparound is not a concern for
> task_cputime_adjusted since this is the actual system cputime of a
> single thread inside a u64.  So io_sq_cpu_time could just return ms
> directly and io_sq_update_worktime would be trivial:
> 
>   sqd->work_time = io_sq_pu_time(current) - start.

That's a good point - I'll update both patches, folding and incremental
like the below in. Thanks!

-- 
Jens Axboe