[PATCH v2 3/3] lib/group_cpus.c: honor housekeeping config when grouping CPUs

[Ming Lei]

On Mon, Jul 01, 2024 at 08:43:34AM +0200, Hannes Reinecke wrote:
> On 7/1/24 04:09, Ming Lei wrote:
> > On Thu, Jun 27, 2024 at 04:10:53PM +0200, Daniel Wagner wrote:
> > > group_cpus_evenly distributes all present CPUs into groups. This ignores
> > > the isolcpus configuration and assigns isolated CPUs into the groups.
> > > 
> > > Make group_cpus_evenly aware of isolcpus configuration and use the
> > > housekeeping CPU mask as base for distributing the available CPUs into
> > > groups.
> > > 
> > > Fixes: 11ea68f553e2 ("genirq, sched/isolation: Isolate from handling managed interrupts")
> > > Signed-off-by: Daniel Wagner <dwagner at suse.de>
> > > ---
> > >   lib/group_cpus.c | 75 ++++++++++++++++++++++++++++++++++++++++++++++++++++++--
> > >   1 file changed, 73 insertions(+), 2 deletions(-)
> > > 
> > > diff --git a/lib/group_cpus.c b/lib/group_cpus.c
> > > index ee272c4cefcc..19fb7186f9d4 100644
> > > --- a/lib/group_cpus.c
> > > +++ b/lib/group_cpus.c
> > > @@ -8,6 +8,7 @@
> > >   #include <linux/cpu.h>
> > >   #include <linux/sort.h>
> > >   #include <linux/group_cpus.h>
> > > +#include <linux/sched/isolation.h>
> > >   #ifdef CONFIG_SMP
> > > @@ -330,7 +331,7 @@ static int __group_cpus_evenly(unsigned int startgrp, unsigned int numgrps,
> > >   }
> > >   /**
> > > - * group_cpus_evenly - Group all CPUs evenly per NUMA/CPU locality
> > > + * group_possible_cpus_evenly - Group all CPUs evenly per NUMA/CPU locality
> > >    * @numgrps: number of groups
> > >    *
> > >    * Return: cpumask array if successful, NULL otherwise. And each element
> > > @@ -344,7 +345,7 @@ static int __group_cpus_evenly(unsigned int startgrp, unsigned int numgrps,
> > >    * We guarantee in the resulted grouping that all CPUs are covered, and
> > >    * no same CPU is assigned to multiple groups
> > >    */
> > > -struct cpumask *group_cpus_evenly(unsigned int numgrps)
> > > +static struct cpumask *group_possible_cpus_evenly(unsigned int numgrps)
> > >   {
> > >   	unsigned int curgrp = 0, nr_present = 0, nr_others = 0;
> > >   	cpumask_var_t *node_to_cpumask;
> > > @@ -423,6 +424,76 @@ struct cpumask *group_cpus_evenly(unsigned int numgrps)
> > >   	}
> > >   	return masks;
> > >   }
> > > +
> > > +/**
> > > + * group_mask_cpus_evenly - Group all CPUs evenly per NUMA/CPU locality
> > > + * @numgrps: number of groups
> > > + * @cpu_mask: CPU to consider for the grouping
> > > + *
> > > + * Return: cpumask array if successful, NULL otherwise. And each element
> > > + * includes CPUs assigned to this group.
> > > + *
> > > + * Try to put close CPUs from viewpoint of CPU and NUMA locality into
> > > + * same group. Allocate present CPUs on these groups evenly.
> > > + */
> > > +static struct cpumask *group_mask_cpus_evenly(unsigned int numgrps,
> > > +					      const struct cpumask *cpu_mask)
> > > +{
> > > +	cpumask_var_t *node_to_cpumask;
> > > +	cpumask_var_t nmsk;
> > > +	int ret = -ENOMEM;
> > > +	struct cpumask *masks = NULL;
> > > +
> > > +	if (!zalloc_cpumask_var(&nmsk, GFP_KERNEL))
> > > +		return NULL;
> > > +
> > > +	node_to_cpumask = alloc_node_to_cpumask();
> > > +	if (!node_to_cpumask)
> > > +		goto fail_nmsk;
> > > +
> > > +	masks = kcalloc(numgrps, sizeof(*masks), GFP_KERNEL);
> > > +	if (!masks)
> > > +		goto fail_node_to_cpumask;
> > > +
> > > +	build_node_to_cpumask(node_to_cpumask);
> > > +
> > > +	ret = __group_cpus_evenly(0, numgrps, node_to_cpumask, cpu_mask, nmsk,
> > > +				  masks);
> > > +
> > > +fail_node_to_cpumask:
> > > +	free_node_to_cpumask(node_to_cpumask);
> > > +
> > > +fail_nmsk:
> > > +	free_cpumask_var(nmsk);
> > > +	if (ret < 0) {
> > > +		kfree(masks);
> > > +		return NULL;
> > > +	}
> > > +	return masks;
> > > +}
> > > +
> > > +/**
> > > + * group_cpus_evenly - Group all CPUs evenly per NUMA/CPU locality
> > > + * @numgrps: number of groups
> > > + *
> > > + * Return: cpumask array if successful, NULL otherwise.
> > > + *
> > > + * group_possible_cpus_evently() is used for distributing the cpus on all
> > > + * possible cpus in absence of isolcpus command line argument.
> > > + * group_mask_cpu_evenly() is used when the isolcpus command line
> > > + * argument is used with managed_irq option. In this case only the
> > > + * housekeeping CPUs are considered.
> > > + */
> > > +struct cpumask *group_cpus_evenly(unsigned int numgrps)
> > > +{
> > > +	const struct cpumask *hk_mask;
> > > +
> > > +	hk_mask = housekeeping_cpumask(HK_TYPE_MANAGED_IRQ);
> > > +	if (!cpumask_empty(hk_mask))
> > > +		return group_mask_cpus_evenly(numgrps, hk_mask);
> > > +
> > > +	return group_possible_cpus_evenly(numgrps);
> > 
> > Since this patch, some isolated CPUs may not be covered in
> > blk-mq queue mapping.
> > 
> > Meantime people still may submit IO workload from isolated CPUs
> > such as by 'taskset -c', blk-mq may not work well for this situation,
> > for example, IO hang may be caused during cpu hotplug.
> > 
> > I did see this kind of usage in some RH Openshift workloads.
> > 
> > If blk-mq problem can be solved, I am fine with this kind of
> > change.
> > 
> That was kinda the idea of this patchset; when 'isolcpus' is active any
> in-kernel driver can only run on the housekeeping CPUs, and I/O from the
> isolcpus is impossible.
> (Otherwise they won't be isolated anymore, and the whole concepts becomes
> ever so shaky.).

Userspace may still force to run IO workload from isolated CPUs when they do
not care CPU isolation, and kernel still should complete IO from isolated CPUs,
and can't run into hang or panic meantime.

And we do support this kind of usage now, then regression is caused by
this patch.

Thanks, 
Ming