linux-kernel.vger.kernel.org archive mirror
 help / color / mirror / Atom feed
From: Riccardo Mancini <rickyman7@gmail.com>
To: Arnaldo Carvalho de Melo <arnaldo.melo@gmail.com>
Cc: Ian Rogers <irogers@google.com>,
	Namhyung Kim <namhyung@kernel.org>,
	Peter Zijlstra <peterz@infradead.org>,
	Ingo Molnar <mingo@redhat.com>,
	Mark Rutland <mark.rutland@arm.com>, Jiri Olsa <jolsa@redhat.com>,
	linux-kernel@vger.kernel.org, linux-perf-users@vger.kernel.org,
	Alexey Bayduraev <alexey.v.bayduraev@linux.intel.com>
Subject: Re: [RFC PATCH 01/10] perf workqueue: threadpool creation and destruction
Date: Thu, 15 Jul 2021 18:31:07 +0200	[thread overview]
Message-ID: <cf0067475311633b91fc22eeb7215c5dac79031e.camel@gmail.com> (raw)
In-Reply-To: <YO7xpFg0F5Fv/7sI@kernel.org>

Hi Arnaldo,

thanks for reviewing the patch!

On Wed, 2021-07-14 at 11:16 -0300, Arnaldo Carvalho de Melo wrote:
<SNIP>
> > +
> > +enum threadpool_status {
> > +       THREADPOOL_STATUS__STOPPED,             /* no threads */
> > +       THREADPOOL_STATUS__ERROR,               /* errors */
> > +       THREADPOOL_STATUS__MAX
> > +};
> > +
> > +struct threadpool_struct {
> 
> Can this be just 'struct threadpool'? I think its descriptive enough:

I agree, but I wanted to keep the naming consistent between workqueue.c and
threadpool.c.

> 
> > +       int                     nr_threads;     /* number of threads in the
> > pool */
> > +       struct thread_struct    *threads;       /* array of threads in the
> > pool */
> > +       struct task_struct      *current_task;  /* current executing
> > function
> > */
> > +       enum threadpool_status  status;         /* current status of the
> > pool
> > */
> > +};
> > +
> > +struct thread_struct {
> > +       int                             idx;    /* idx of thread in pool-
> > > threads */
> > +       pid_t                           tid;    /* tid of thread */
> > +       struct threadpool_struct        *pool;  /* parent threadpool */
> > +       struct {
> > +               int from[2];                    /* messages from thread
> > (acks)
> > */
> > +               int to[2];                      /* messages to thread
> > (commands) */
> > +       } pipes;
> > +};
> 
> This one, since we have already a 'struct thread' in tools/perf, to
> represent a PERF_RECORD_FORK, perhaps we can call it 'struct
> threadpool_entry'?

Agreed.

> 
> > +
> > +/**
> > + * init_pipes - initialize all pipes of @thread
> > + */
> > +static void init_pipes(struct thread_struct *thread)
> > +{
> > +       thread->pipes.from[0] = -1;
> > +       thread->pipes.from[1] = -1;
> > +       thread->pipes.to[0] = -1;
> > +       thread->pipes.to[1] = -1;
> > +}
> > +
> > +/**
> > + * open_pipes - open all pipes of @thread
> > + */
> > +static int open_pipes(struct thread_struct *thread)
> 
> Here please:
> 
> threadpool_entry__open_pipes()
> 
> Its longer, but helps with ctags/cscope navigation and we can go
> directly to it via:
> 
> :ta threadpool_entry__open_p<TAB>
> 
> While 'ta: open_pipes' may bo to various places where this idiom is
> used.

Agreed.

<SNIP>
> > +/**
> > + * create_threadpool - create a fixed threadpool with @n_threads threads
> > + */
> > +struct threadpool_struct *create_threadpool(int n_threads)
> 
> 
> Is this already something the kernel has and thus we should keep the
> naming? I couldn't find it in the kernel, so please name it:
> 
> struct threadpool *threadpool__new(int nthreads)

As before, I did this to keep consistency with workqueue.
Since this threadpool+workqueue can be a standalone library, I preferred to keep
the naming consistent inside it, instead of making it consistent with perf (this
is what I was referring to in the cover letter, not just the workqueue API).
What do you think?
I also prefer perf's naming conventions, but it'd feel strange to use two
different naming conventions inside the same library.

> 
> > +{
> > +       int ret, t;
> > +       struct threadpool_struct *pool = malloc(sizeof(*pool));
> > +
> > +       if (!pool) {
> > +               pr_err("threadpool: cannot allocate pool: %s\n",
> > +                       strerror(errno));o
> 
> Humm, pr_err() at this level isn't appropriate, please make callers
> complain.

ok.

> 
> > +               return NULL;
> > +       }
> > +
> > +       if (n_threads <= 0) {
> > +               pr_err("threadpool: invalid number of threads: %d\n",
> > +                       n_threads);
> 
> pr_debug()

ok

> 
> > +               goto out_free_pool;
> > +       }
> > +
> > +       pool->nr_threads = n_threads;
> > +       pool->current_task = NULL;
> > +
> > +       pool->threads = malloc(n_threads * sizeof(*pool->threads));
> > +       if (!pool->threads) {
> > +               pr_err("threadpool: cannot allocate threads: %s\n",
> > +                       strerror(errno));
> > +               goto out_free_pool;
> > +       }
> > +
> > +       for (t = 0; t < n_threads; t++) {
> > +               pool->threads[t].idx = t;
> > +               pool->threads[t].tid = -1;
> > +               pool->threads[t].pool = pool;
> > +               init_pipes(&pool->threads[t]);
> > +       }
> > +
> > +       for (t = 0; t < n_threads; t++) {
> > +               ret = open_pipes(&pool->threads[t]);
> > +               if (ret)
> > +                       goto out_close_pipes;
> > +       }
> > +
> > +       pool->status = THREADPOOL_STATUS__STOPPED;
> > +
> > +       return pool;
> > +
> > +out_close_pipes:
> > +       for (t = 0; t < n_threads; t++)
> > +               close_pipes(&pool->threads[t]);
> > +
> > +       free(pool->threads);
> > +out_free_pool:
> > +       free(pool);
> > +       return NULL;
> 
> Here we can use ERR_PTR()/PTR_ERR() to let the caller know what was the
> problem, i.e. we can ditch all the pr_err/pr_debug(), etc and instead
> have a threadpool__strerror(struct threadpool *pool, int err) like we
> have for 'struct evsel', please take a look at evsel__open_strerror().

Thanks, I'll have a look at it.
So, what I sould do is not use pr_* higher than debug inside library code and
return meaningful errors through PR_ERR, right?

> 
> 
> > +}
> > +
> > +/**
> > + * destroy_threadpool - free the @pool and all its resources
> > + */
> > +void destroy_threadpool(struct threadpool_struct *pool)
> 
> 
> void threadpool__delete(struct threadpool *pool)
> > +{
> > +       int t;
> > +
> > +       if (!pool)
> > +               return;
> > +
> > +       WARN_ON(pool->status != THREADPOOL_STATUS__STOPPED
> > +               && pool->status != THREADPOOL_STATUS__ERROR);
> > +
> > +       for (t = 0; t < pool->nr_threads; t++)
> > +               close_pipes(&pool->threads[t]);
> 
> reset pool->threads[t] to -1

already inside close_pipes. I agree it might be confusing without the
threadpool_entry__ prefix.

> 
> > +
> > +       free(pool->threads);
> 
> zfree

In general, when should I use zfree instead of free?

> 
> > +       free(pool);
> > +}
> > +
> > +/**
> > + * threadpool_size - get number of threads in the threadpool
> > + */
> > +int threadpool_size(struct threadpool_struct *pool)
>   
> threadpool__size()

ok

Thanks,
Riccardo

> 
> > +{
> > +       return pool->nr_threads;
> > +}
> > diff --git a/tools/perf/util/workqueue/threadpool.h
> > b/tools/perf/util/workqueue/threadpool.h
> > new file mode 100644
> > index 0000000000000000..2b9388c768a0b588
> > --- /dev/null
> > +++ b/tools/perf/util/workqueue/threadpool.h
> > @@ -0,0 +1,19 @@
> > +/* SPDX-License-Identifier: GPL-2.0 */
> > +#ifndef __WORKQUEUE_THREADPOOL_H
> > +#define __WORKQUEUE_THREADPOOL_H
> > +
> > +struct threadpool_struct;
> > +struct task_struct;
> > +
> > +typedef void (*task_func_t)(int tidx, struct task_struct *task);
> > +
> > +struct task_struct {
> > +       task_func_t fn;
> > +};
> > +
> > +extern struct threadpool_struct *create_threadpool(int n_threads);
> > +extern void destroy_threadpool(struct threadpool_struct *pool);
> > +
> > +extern int threadpool_size(struct threadpool_struct *pool);
> > +
> > +#endif /* __WORKQUEUE_THREADPOOL_H */
> > -- 
> > 2.31.1
> > 
> 




  reply	other threads:[~2021-07-15 16:31 UTC|newest]

Thread overview: 33+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2021-07-13 12:11 [RFC PATCH 00/10] perf: add workqueue library and use it in synthetic-events Riccardo Mancini
2021-07-13 12:11 ` [RFC PATCH 01/10] perf workqueue: threadpool creation and destruction Riccardo Mancini
2021-07-14 14:16   ` Arnaldo Carvalho de Melo
2021-07-15 16:31     ` Riccardo Mancini [this message]
2021-07-15 20:48       ` Arnaldo Carvalho de Melo
2021-07-15 23:29     ` Namhyung Kim
2021-07-16 13:36       ` Riccardo Mancini
2021-07-19 19:39         ` Namhyung Kim
2021-07-13 12:11 ` [RFC PATCH 02/10] perf tests: add test for workqueue Riccardo Mancini
2021-07-14 15:10   ` Arnaldo Carvalho de Melo
2021-07-15 16:33     ` Riccardo Mancini
2021-07-13 12:11 ` [RFC PATCH 03/10] perf workqueue: add threadpool start and stop functions Riccardo Mancini
2021-07-14 15:15   ` Arnaldo Carvalho de Melo
2021-07-15 16:42     ` Riccardo Mancini
2021-07-15 20:43       ` Arnaldo Carvalho de Melo
2021-07-15 23:48   ` Namhyung Kim
2021-07-16 13:53     ` Riccardo Mancini
2021-07-16 16:29       ` Arnaldo Carvalho de Melo
2021-07-13 12:11 ` [RFC PATCH 04/10] perf workqueue: add threadpool execute and wait functions Riccardo Mancini
2021-07-15 23:56   ` Namhyung Kim
2021-07-16 13:55     ` Riccardo Mancini
2021-07-13 12:11 ` [RFC PATCH 05/10] perf workqueue: add sparse annotation header Riccardo Mancini
2021-07-13 12:11 ` [RFC PATCH 06/10] perf workqueue: introduce workqueue struct Riccardo Mancini
2021-07-14 15:22   ` Arnaldo Carvalho de Melo
2021-07-15 16:49     ` Riccardo Mancini
2021-07-15 20:47       ` Arnaldo Carvalho de Melo
2021-07-13 12:11 ` [RFC PATCH 07/10] perf workqueue: implement worker thread and management Riccardo Mancini
2021-07-13 12:11 ` [RFC PATCH 08/10] perf workqueue: add queue_work and flush_workqueue functions Riccardo Mancini
2021-07-13 12:11 ` [RFC PATCH 09/10] perf workqueue: add utility to execute a for loop in parallel Riccardo Mancini
2021-07-13 12:11 ` [RFC PATCH 10/10] perf synthetic-events: use workqueue parallel_for Riccardo Mancini
2021-07-13 19:14 ` [RFC PATCH 00/10] perf: add workqueue library and use it in synthetic-events Arnaldo Carvalho de Melo
2021-07-19 21:13 ` Jiri Olsa
2021-07-22 16:15   ` Riccardo Mancini

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=cf0067475311633b91fc22eeb7215c5dac79031e.camel@gmail.com \
    --to=rickyman7@gmail.com \
    --cc=alexey.v.bayduraev@linux.intel.com \
    --cc=arnaldo.melo@gmail.com \
    --cc=irogers@google.com \
    --cc=jolsa@redhat.com \
    --cc=linux-kernel@vger.kernel.org \
    --cc=linux-perf-users@vger.kernel.org \
    --cc=mark.rutland@arm.com \
    --cc=mingo@redhat.com \
    --cc=namhyung@kernel.org \
    --cc=peterz@infradead.org \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line before the message body.
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).