| Message ID | 20190528130811.26480-1-jgross@suse.com (mailing list archive) |
|---|---|
| State | Superseded |
| Headers | show |
| Series | xen: remove on-stack cpumask from stop_machine_run() | expand |
>>> On 28.05.19 at 15:08, <jgross@suse.com> wrote: > --- a/xen/common/stop_machine.c > +++ b/xen/common/stop_machine.c > @@ -69,8 +69,8 @@ static void stopmachine_wait_state(void) > > int stop_machine_run(int (*fn)(void *), void *data, unsigned int cpu) > { > - cpumask_t allbutself; > unsigned int i, nr_cpus; > + unsigned int my_cpu = smp_processor_id(); Variables starting with my_ being commonly used in introductory examples, I'd prefer to avoid such names. Elsewhere we use "this_cpu", "me", or maybe "this" if plain "cpu" is already taken. > @@ -79,9 +79,7 @@ int stop_machine_run(int (*fn)(void *), void *data, unsigned int cpu) > if ( !get_cpu_maps() ) > return -EBUSY; > > - cpumask_andnot(&allbutself, &cpu_online_map, > - cpumask_of(smp_processor_id())); > - nr_cpus = cpumask_weight(&allbutself); > + nr_cpus = cpumask_weight(&cpu_online_map) - 1; Having looked at a lot of CPU offlining code recently, I notice this isn't strictly correct: You imply cpu_online(my_cpu) to produce "true". I think at present this will always hold, but I'd prefer if we could avoid gaining such a dependency. And it doesn't look overly difficult to avoid it. Also please don't open-code num_online_cpus(). > @@ -100,8 +98,9 @@ int stop_machine_run(int (*fn)(void *), void *data, unsigned int cpu) > > smp_wmb(); > > - for_each_cpu ( i, &allbutself ) > - tasklet_schedule_on_cpu(&per_cpu(stopmachine_tasklet, i), i); > + for_each_cpu ( i, &cpu_online_map ) Same here for for_each_online_cpu(). Jan
On 28/05/2019 16:32, Jan Beulich wrote: >>>> On 28.05.19 at 15:08, <jgross@suse.com> wrote: >> --- a/xen/common/stop_machine.c >> +++ b/xen/common/stop_machine.c >> @@ -69,8 +69,8 @@ static void stopmachine_wait_state(void) >> >> int stop_machine_run(int (*fn)(void *), void *data, unsigned int cpu) >> { >> - cpumask_t allbutself; >> unsigned int i, nr_cpus; >> + unsigned int my_cpu = smp_processor_id(); > > Variables starting with my_ being commonly used in introductory > examples, I'd prefer to avoid such names. Elsewhere we use > "this_cpu", "me", or maybe "this" if plain "cpu" is already taken. Okay. > >> @@ -79,9 +79,7 @@ int stop_machine_run(int (*fn)(void *), void *data, unsigned int cpu) >> if ( !get_cpu_maps() ) >> return -EBUSY; >> >> - cpumask_andnot(&allbutself, &cpu_online_map, >> - cpumask_of(smp_processor_id())); >> - nr_cpus = cpumask_weight(&allbutself); >> + nr_cpus = cpumask_weight(&cpu_online_map) - 1; > > Having looked at a lot of CPU offlining code recently, I notice this > isn't strictly correct: You imply cpu_online(my_cpu) to produce > "true". I think at present this will always hold, but I'd prefer if we > could avoid gaining such a dependency. And it doesn't look overly > difficult to avoid it. Yes, I have thought about it. If you like it better I test for the running cpu to be in cpu_online_map. > Also please don't open-code num_online_cpus(). Ah, of course! > >> @@ -100,8 +98,9 @@ int stop_machine_run(int (*fn)(void *), void *data, unsigned int cpu) >> >> smp_wmb(); >> >> - for_each_cpu ( i, &allbutself ) >> - tasklet_schedule_on_cpu(&per_cpu(stopmachine_tasklet, i), i); >> + for_each_cpu ( i, &cpu_online_map ) > > Same here for for_each_online_cpu(). Yes. Juergen
diff --git a/xen/common/stop_machine.c b/xen/common/stop_machine.c index ce6f5624c4..ccda2efa3e 100644 --- a/xen/common/stop_machine.c +++ b/xen/common/stop_machine.c @@ -69,8 +69,8 @@ static void stopmachine_wait_state(void) int stop_machine_run(int (*fn)(void *), void *data, unsigned int cpu) { - cpumask_t allbutself; unsigned int i, nr_cpus; + unsigned int my_cpu = smp_processor_id(); int ret; BUG_ON(!local_irq_is_enabled()); @@ -79,9 +79,7 @@ int stop_machine_run(int (*fn)(void *), void *data, unsigned int cpu) if ( !get_cpu_maps() ) return -EBUSY; - cpumask_andnot(&allbutself, &cpu_online_map, - cpumask_of(smp_processor_id())); - nr_cpus = cpumask_weight(&allbutself); + nr_cpus = cpumask_weight(&cpu_online_map) - 1; /* Must not spin here as the holder will expect us to be descheduled. */ if ( !spin_trylock(&stopmachine_lock) ) @@ -100,8 +98,9 @@ int stop_machine_run(int (*fn)(void *), void *data, unsigned int cpu) smp_wmb(); - for_each_cpu ( i, &allbutself ) - tasklet_schedule_on_cpu(&per_cpu(stopmachine_tasklet, i), i); + for_each_cpu ( i, &cpu_online_map ) + if ( i != my_cpu ) + tasklet_schedule_on_cpu(&per_cpu(stopmachine_tasklet, i), i); stopmachine_set_state(STOPMACHINE_PREPARE); stopmachine_wait_state();
The "allbutself" cpumask in stop_machine_run() is not needed. Instead of allocating it on the stack it can easily be avoided. Signed-off-by: Juergen Gross <jgross@suse.com> --- xen/common/stop_machine.c | 11 +++++------ 1 file changed, 5 insertions(+), 6 deletions(-)