| Message ID | 20190725094146.18560-1-ming.lei@redhat.com (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
| Series | [V2] blk-mq: balance mapping between present CPUs and queues | expand |
On 7/25/19 5:41 PM, Ming Lei wrote: > Spread queues among present CPUs first, then building mapping on other > non-present CPUs. > > So we can minimize count of dead queues which are mapped by un-present > CPUs only. Then bad IO performance can be avoided by unbalanced mapping > between present CPUs and queues. > > The similar policy has been applied on Managed IRQ affinity. > > Reported-by: Yi Zhang <yi.zhang@redhat.com> > Cc: Yi Zhang <yi.zhang@redhat.com> > Cc: Bob Liu <bob.liu@oracle.com> > Signed-off-by: Ming Lei <ming.lei@redhat.com> > --- > > V2: > - make sure that sequential mapping can be done > > block/blk-mq-cpumap.c | 29 ++++++++++++++++++++++------- > 1 file changed, 22 insertions(+), 7 deletions(-) > > diff --git a/block/blk-mq-cpumap.c b/block/blk-mq-cpumap.c > index f945621a0e8f..0157f2b3485a 100644 > --- a/block/blk-mq-cpumap.c > +++ b/block/blk-mq-cpumap.c > @@ -15,10 +15,10 @@ > #include "blk.h" > #include "blk-mq.h" > > -static int cpu_to_queue_index(struct blk_mq_queue_map *qmap, > - unsigned int nr_queues, const int cpu) > +static int queue_index(struct blk_mq_queue_map *qmap, > + unsigned int nr_queues, const int q) > { > - return qmap->queue_offset + (cpu % nr_queues); > + return qmap->queue_offset + (q % nr_queues); > } > > static int get_first_sibling(unsigned int cpu) > @@ -36,21 +36,36 @@ int blk_mq_map_queues(struct blk_mq_queue_map *qmap) > { > unsigned int *map = qmap->mq_map; > unsigned int nr_queues = qmap->nr_queues; > - unsigned int cpu, first_sibling; > + unsigned int cpu, first_sibling, q = 0; > + > + for_each_possible_cpu(cpu) > + map[cpu] = -1; > + > + /* > + * Spread queues among present CPUs first for minimizing > + * count of dead queues which are mapped by all un-present CPUs > + */ > + for_each_present_cpu(cpu) { > + if (q >= nr_queues) > + break; > + map[cpu] = queue_index(qmap, nr_queues, q++); > + } > > for_each_possible_cpu(cpu) { > + if (map[cpu] != -1) > + continue; > /* > * First do sequential mapping between CPUs and queues. > * In case we still have CPUs to map, and we have some number of > * threads per cores then map sibling threads to the same queue > * for performance optimizations. > */ > - if (cpu < nr_queues) { > - map[cpu] = cpu_to_queue_index(qmap, nr_queues, cpu); > + if (q < nr_queues) { > + map[cpu] = queue_index(qmap, nr_queues, q++); Yeah, that's what I was trying to say. > } else { > first_sibling = get_first_sibling(cpu); > if (first_sibling == cpu) > - map[cpu] = cpu_to_queue_index(qmap, nr_queues, cpu); > + map[cpu] = queue_index(qmap, nr_queues, q++); > else > map[cpu] = map[first_sibling]; > } > Reviewed-by: Bob Liu <bob.liu@oracle.com>
On Thu, Jul 25, 2019 at 05:41:46PM +0800, Ming Lei wrote: > Spread queues among present CPUs first, then building mapping on other > non-present CPUs. > > So we can minimize count of dead queues which are mapped by un-present > CPUs only. Then bad IO performance can be avoided by unbalanced mapping > between present CPUs and queues. > > The similar policy has been applied on Managed IRQ affinity. > > Reported-by: Yi Zhang <yi.zhang@redhat.com> > Cc: Yi Zhang <yi.zhang@redhat.com> > Cc: Bob Liu <bob.liu@oracle.com> > Signed-off-by: Ming Lei <ming.lei@redhat.com> > --- > > V2: > - make sure that sequential mapping can be done > > block/blk-mq-cpumap.c | 29 ++++++++++++++++++++++------- > 1 file changed, 22 insertions(+), 7 deletions(-) > > diff --git a/block/blk-mq-cpumap.c b/block/blk-mq-cpumap.c > index f945621a0e8f..0157f2b3485a 100644 > --- a/block/blk-mq-cpumap.c > +++ b/block/blk-mq-cpumap.c > @@ -15,10 +15,10 @@ > #include "blk.h" > #include "blk-mq.h" > > -static int cpu_to_queue_index(struct blk_mq_queue_map *qmap, > - unsigned int nr_queues, const int cpu) > +static int queue_index(struct blk_mq_queue_map *qmap, > + unsigned int nr_queues, const int q) > { > - return qmap->queue_offset + (cpu % nr_queues); > + return qmap->queue_offset + (q % nr_queues); > } > > static int get_first_sibling(unsigned int cpu) > @@ -36,21 +36,36 @@ int blk_mq_map_queues(struct blk_mq_queue_map *qmap) > { > unsigned int *map = qmap->mq_map; > unsigned int nr_queues = qmap->nr_queues; > - unsigned int cpu, first_sibling; > + unsigned int cpu, first_sibling, q = 0; > + > + for_each_possible_cpu(cpu) > + map[cpu] = -1; > + > + /* > + * Spread queues among present CPUs first for minimizing > + * count of dead queues which are mapped by all un-present CPUs > + */ > + for_each_present_cpu(cpu) { > + if (q >= nr_queues) > + break; > + map[cpu] = queue_index(qmap, nr_queues, q++); > + } > > for_each_possible_cpu(cpu) { > + if (map[cpu] != -1) > + continue; > /* > * First do sequential mapping between CPUs and queues. > * In case we still have CPUs to map, and we have some number of > * threads per cores then map sibling threads to the same queue > * for performance optimizations. > */ > - if (cpu < nr_queues) { > - map[cpu] = cpu_to_queue_index(qmap, nr_queues, cpu); > + if (q < nr_queues) { > + map[cpu] = queue_index(qmap, nr_queues, q++); > } else { > first_sibling = get_first_sibling(cpu); > if (first_sibling == cpu) > - map[cpu] = cpu_to_queue_index(qmap, nr_queues, cpu); > + map[cpu] = queue_index(qmap, nr_queues, q++); > else > map[cpu] = map[first_sibling]; > } Hi Jens, Could you consider to merge this patch to 5.4? Thanks, Ming
On 7/25/19 2:41 AM, Ming Lei wrote: > Spread queues among present CPUs first, then building mapping on other > non-present CPUs. > > So we can minimize count of dead queues which are mapped by un-present > CPUs only. Then bad IO performance can be avoided by unbalanced mapping > between present CPUs and queues. > > The similar policy has been applied on Managed IRQ affinity. LGTM, queued up for some testing.
diff --git a/block/blk-mq-cpumap.c b/block/blk-mq-cpumap.c index f945621a0e8f..0157f2b3485a 100644 --- a/block/blk-mq-cpumap.c +++ b/block/blk-mq-cpumap.c @@ -15,10 +15,10 @@ #include "blk.h" #include "blk-mq.h" -static int cpu_to_queue_index(struct blk_mq_queue_map *qmap, - unsigned int nr_queues, const int cpu) +static int queue_index(struct blk_mq_queue_map *qmap, + unsigned int nr_queues, const int q) { - return qmap->queue_offset + (cpu % nr_queues); + return qmap->queue_offset + (q % nr_queues); } static int get_first_sibling(unsigned int cpu) @@ -36,21 +36,36 @@ int blk_mq_map_queues(struct blk_mq_queue_map *qmap) { unsigned int *map = qmap->mq_map; unsigned int nr_queues = qmap->nr_queues; - unsigned int cpu, first_sibling; + unsigned int cpu, first_sibling, q = 0; + + for_each_possible_cpu(cpu) + map[cpu] = -1; + + /* + * Spread queues among present CPUs first for minimizing + * count of dead queues which are mapped by all un-present CPUs + */ + for_each_present_cpu(cpu) { + if (q >= nr_queues) + break; + map[cpu] = queue_index(qmap, nr_queues, q++); + } for_each_possible_cpu(cpu) { + if (map[cpu] != -1) + continue; /* * First do sequential mapping between CPUs and queues. * In case we still have CPUs to map, and we have some number of * threads per cores then map sibling threads to the same queue * for performance optimizations. */ - if (cpu < nr_queues) { - map[cpu] = cpu_to_queue_index(qmap, nr_queues, cpu); + if (q < nr_queues) { + map[cpu] = queue_index(qmap, nr_queues, q++); } else { first_sibling = get_first_sibling(cpu); if (first_sibling == cpu) - map[cpu] = cpu_to_queue_index(qmap, nr_queues, cpu); + map[cpu] = queue_index(qmap, nr_queues, q++); else map[cpu] = map[first_sibling]; }
Spread queues among present CPUs first, then building mapping on other non-present CPUs. So we can minimize count of dead queues which are mapped by un-present CPUs only. Then bad IO performance can be avoided by unbalanced mapping between present CPUs and queues. The similar policy has been applied on Managed IRQ affinity. Reported-by: Yi Zhang <yi.zhang@redhat.com> Cc: Yi Zhang <yi.zhang@redhat.com> Cc: Bob Liu <bob.liu@oracle.com> Signed-off-by: Ming Lei <ming.lei@redhat.com> --- V2: - make sure that sequential mapping can be done block/blk-mq-cpumap.c | 29 ++++++++++++++++++++++------- 1 file changed, 22 insertions(+), 7 deletions(-)