On Mon, Jul 17, 2023 at 06:07:59PM +0200, Thomas Monjalon wrote:
> 17/07/2023 16:09, Bruce Richardson:
> > On Mon, Jul 03, 2023 at 01:49:55PM +0200, Thomas Monjalon wrote:
> > > 03/07/2023 13:38, David Marchand:
> > > > On Mon, Jul 3, 2023 at 1:24 PM Thomas Monjalon <tho...@monjalon.net>
> > > > wrote:
> > > > >
> > > > > When looking at threads in a system, it can be confusing
> > > > > to find some unknown threads without a clue it is started by DPDK.
> > > > >
> > > > > Let's start all thread names with "dpdk-"
> > > > > and the driver name if it comes from a driver.
> > > > >
> > > > > One more constraint: the thread names are generally limited
> > > > > to 16 characters, including NUL character.
> > > >
> > > > Yes, and no way to discover this limit.
> > > >
> > > >
> > > > >
> > > > > Signed-off-by: Thomas Monjalon <tho...@monjalon.net>
> > > > > ---
> > > > > Please review carefully, I may have missed things.
> > > > >
> > > > > Note: we should use rte_ctrl_thread_create() as much as possible.
> > > > >
> > > >
> > > > As for this comment, we should advertise rte_thread_create_control()
> > > > instead, as it is thread library agnostic.
> > > > Which leads me to suggest looking at rte_thread_create_control() calls
> > > > too.
> > >
> > > You're right, I'll do a second pass.
> > >
> > > > > diff --git a/lib/eal/freebsd/eal.c b/lib/eal/freebsd/eal.c
> > > > > index 7008303e11..e201acd1c2 100644
> > > > > --- a/lib/eal/freebsd/eal.c
> > > > > +++ b/lib/eal/freebsd/eal.c
> > > > > @@ -843,7 +843,7 @@ rte_eal_init(int argc, char **argv)
> > > > >
> > > > > /* Set thread_name for aid in debugging. */
> > > > > snprintf(thread_name, sizeof(thread_name),
> > > > > - "rte-worker-%d", i);
> > > > > + "dpdk-worker-%d", i);
> > > > > rte_thread_set_name(lcore_config[i].thread_id,
> > > > > thread_name);
> > > > >
> > > > > ret =
> > > > > rte_thread_set_affinity_by_id(lcore_config[i].thread_id,
> > > >
> > > > We changed this not so long ago with 8ae946970ed3 ("eal: fix thread
> > > > name for high order lcores") which offered up to 4 digits for lcores.
> > > > With your proposal, we are back to only 3 digits, which is probably
> > > > enough, but I preferred to raise it.
> > >
> > > In general no need of hyphen before a number.
> > > I will change to dpdk-worker%d.
> > >
> > While I know that "dpdk-" is probably the clearer prefix, "rte-" is also
> > well-know as the standard API prefix for DPDK. Give we have a 16-char
> > limit, I would suggest keeping using the "rte" prefix on the basis of
> > brevity!
>
> It's one character less :)
Yes, but when you are limited to 16 characters, each char counts! "dpdk-"
uses 31% of the available chars, vs 25% for the shorter prefix.
> "rte" is used for API.
> "dpdk-" is used for file prefixes.
> When running an application, users may understand what is a "dpdk-xyz" thread,
> while "rte-xyz" is more cryptic if they are not involved with DPDK at all.
>
> > A third alternative is to just use "dp-" as prefix...
>
> A third alternative looks like more confusion.
> The idea of the thread prefix is to indicate,
> to anyone monitoring a system, where the threads come from.
>
Sure. No massive objections either way, just keen to see the names being as
meaningful as possible.
