mirrors
/
linux-stable
mirror of https://git.kernel.org/pub/scm/linux/kernel/git/stable/linux.git
1
0
Fork 0
Code Issues Releases Wiki Activity

genirq/affinity: Rename irq_build_affinity_masks as group_cpus_evenly 

Map irq vector into group, which allows to abstract the algorithm for
a generic use case outside of the interrupt core.

Rename irq_build_affinity_masks as group_cpus_evenly, so the API can be
reused for blk-mq to make default queue mapping even though irq vectors
aren't involved.

No functional change, just rename vector as group.

Signed-off-by: Ming Lei <ming.lei@redhat.com>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Jens Axboe <axboe@kernel.dk>                                                                                                                                                                                                    
Link: https://lore.kernel.org/r/20221227022905.352674-5-ming.lei@redhat.com
This commit is contained in:
Ming Lei 2022-12-27 10:29:03 +08:00 committed by Thomas Gleixner
parent e7bdd7f0cb
commit 523f1ea76a
1 changed files with 121 additions and 121 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

242
kernel/irq/affinity.c
View File

@ -9,13 +9,13 @@
#include <linux/cpu.h> #include <linux/cpu.h>
#include <linux/sort.h> #include <linux/sort.h>
static void irq_spread_init_one(struct cpumask *irqmsk, struct cpumask *nmsk, static void grp_spread_init_one(struct cpumask *irqmsk, struct cpumask *nmsk,
unsigned int cpus_per_vec) unsigned int cpus_per_grp)
{ {
const struct cpumask *siblmsk; const struct cpumask *siblmsk;
int cpu, sibl; int cpu, sibl;
for ( ; cpus_per_vec > 0; ) { for ( ; cpus_per_grp > 0; ) {
cpu = cpumask_first(nmsk); cpu = cpumask_first(nmsk);
/* Should not happen, but I'm too lazy to think about it */ /* Should not happen, but I'm too lazy to think about it */
@ -24,18 +24,18 @@ static void irq_spread_init_one(struct cpumask *irqmsk, struct cpumask *nmsk,
cpumask_clear_cpu(cpu, nmsk); cpumask_clear_cpu(cpu, nmsk);
cpumask_set_cpu(cpu, irqmsk); cpumask_set_cpu(cpu, irqmsk);
cpus_per_vec--; cpus_per_grp--;
/* If the cpu has siblings, use them first */ /* If the cpu has siblings, use them first */
siblmsk = topology_sibling_cpumask(cpu); siblmsk = topology_sibling_cpumask(cpu);
for (sibl = -1; cpus_per_vec > 0; ) { for (sibl = -1; cpus_per_grp > 0; ) {
sibl = cpumask_next(sibl, siblmsk); sibl = cpumask_next(sibl, siblmsk);
if (sibl >= nr_cpu_ids) if (sibl >= nr_cpu_ids)
break; break;
if (!cpumask_test_and_clear_cpu(sibl, nmsk)) if (!cpumask_test_and_clear_cpu(sibl, nmsk))
continue; continue;
cpumask_set_cpu(sibl, irqmsk); cpumask_set_cpu(sibl, irqmsk);
cpus_per_vec--; cpus_per_grp--;
} }
} }
} }
@ -95,48 +95,48 @@ static int get_nodes_in_cpumask(cpumask_var_t *node_to_cpumask,
return nodes; return nodes;
} }
struct node_vectors { struct node_groups {
unsigned id; unsigned id;
union { union {
unsigned nvectors; unsigned ngroups;
unsigned ncpus; unsigned ncpus;
}; };
}; };
static int ncpus_cmp_func(const void *l, const void *r) static int ncpus_cmp_func(const void *l, const void *r)
{ {
const struct node_vectors *ln = l; const struct node_groups *ln = l;
const struct node_vectors *rn = r; const struct node_groups *rn = r;
return ln->ncpus - rn->ncpus; return ln->ncpus - rn->ncpus;
} }
/* /*
* Allocate vector number for each node, so that for each node: * Allocate group number for each node, so that for each node:
* *
* 1) the allocated number is >= 1 * 1) the allocated number is >= 1
* *
* 2) the allocated numbver is <= active CPU number of this node * 2) the allocated number is <= active CPU number of this node
* *
* The actual allocated total vectors may be less than @numvecs when * The actual allocated total groups may be less than @numgrps when
* active total CPU number is less than @numvecs. * active total CPU number is less than @numgrps.
* *
* Active CPUs means the CPUs in '@cpu_mask AND @node_to_cpumask[]' * Active CPUs means the CPUs in '@cpu_mask AND @node_to_cpumask[]'
* for each node. * for each node.
*/ */
static void alloc_nodes_vectors(unsigned int numvecs, static void alloc_nodes_groups(unsigned int numgrps,
cpumask_var_t *node_to_cpumask, cpumask_var_t *node_to_cpumask,
const struct cpumask *cpu_mask, const struct cpumask *cpu_mask,
const nodemask_t nodemsk, const nodemask_t nodemsk,
struct cpumask *nmsk, struct cpumask *nmsk,
struct node_vectors *node_vectors) struct node_groups *node_groups)
{ {
unsigned n, remaining_ncpus = 0; unsigned n, remaining_ncpus = 0;
for (n = 0; n < nr_node_ids; n++) { for (n = 0; n < nr_node_ids; n++) {
node_vectors[n].id = n; node_groups[n].id = n;
node_vectors[n].ncpus = UINT_MAX; node_groups[n].ncpus = UINT_MAX;
} }
for_each_node_mask(n, nodemsk) { for_each_node_mask(n, nodemsk) {
@ -148,61 +148,61 @@ static void alloc_nodes_vectors(unsigned int numvecs,
if (!ncpus) if (!ncpus)
continue; continue;
remaining_ncpus += ncpus; remaining_ncpus += ncpus;
node_vectors[n].ncpus = ncpus; node_groups[n].ncpus = ncpus;
} }
numvecs = min_t(unsigned, remaining_ncpus, numvecs); numgrps = min_t(unsigned, remaining_ncpus, numgrps);
sort(node_vectors, nr_node_ids, sizeof(node_vectors[0]), sort(node_groups, nr_node_ids, sizeof(node_groups[0]),
ncpus_cmp_func, NULL); ncpus_cmp_func, NULL);
/* /*
* Allocate vectors for each node according to the ratio of this * Allocate groups for each node according to the ratio of this
* node's nr_cpus to remaining un-assigned ncpus. 'numvecs' is * node's nr_cpus to remaining un-assigned ncpus. 'numgrps' is
* bigger than number of active numa nodes. Always start the * bigger than number of active numa nodes. Always start the
* allocation from the node with minimized nr_cpus. * allocation from the node with minimized nr_cpus.
* *
* This way guarantees that each active node gets allocated at * This way guarantees that each active node gets allocated at
* least one vector, and the theory is simple: over-allocation * least one group, and the theory is simple: over-allocation
* is only done when this node is assigned by one vector, so * is only done when this node is assigned by one group, so
* other nodes will be allocated >= 1 vector, since 'numvecs' is * other nodes will be allocated >= 1 groups, since 'numgrps' is
* bigger than number of numa nodes. * bigger than number of numa nodes.
* *
* One perfect invariant is that number of allocated vectors for * One perfect invariant is that number of allocated groups for
* each node is <= CPU count of this node: * each node is <= CPU count of this node:
* *
* 1) suppose there are two nodes: A and B * 1) suppose there are two nodes: A and B
* ncpu(X) is CPU count of node X * ncpu(X) is CPU count of node X
* vecs(X) is the vector count allocated to node X via this * grps(X) is the group count allocated to node X via this
* algorithm * algorithm
* *
* ncpu(A) <= ncpu(B) * ncpu(A) <= ncpu(B)
* ncpu(A) + ncpu(B) = N * ncpu(A) + ncpu(B) = N
* vecs(A) + vecs(B) = V * grps(A) + grps(B) = G
* *
* vecs(A) = max(1, round_down(V * ncpu(A) / N)) * grps(A) = max(1, round_down(G * ncpu(A) / N))
* vecs(B) = V - vecs(A) * grps(B) = G - grps(A)
* *
* both N and V are integer, and 2 <= V <= N, suppose * both N and G are integer, and 2 <= G <= N, suppose
* V = N - delta, and 0 <= delta <= N - 2 * G = N - delta, and 0 <= delta <= N - 2
* *
* 2) obviously vecs(A) <= ncpu(A) because: * 2) obviously grps(A) <= ncpu(A) because:
* *
* if vecs(A) is 1, then vecs(A) <= ncpu(A) given * if grps(A) is 1, then grps(A) <= ncpu(A) given
* ncpu(A) >= 1 * ncpu(A) >= 1
* *
* otherwise, * otherwise,
* vecs(A) <= V * ncpu(A) / N <= ncpu(A), given V <= N * grps(A) <= G * ncpu(A) / N <= ncpu(A), given G <= N
* *
* 3) prove how vecs(B) <= ncpu(B): * 3) prove how grps(B) <= ncpu(B):
* *
* if round_down(V * ncpu(A) / N) == 0, vecs(B) won't be * if round_down(G * ncpu(A) / N) == 0, vecs(B) won't be
* over-allocated, so vecs(B) <= ncpu(B), * over-allocated, so grps(B) <= ncpu(B),
* *
* otherwise: * otherwise:
* *
* vecs(A) = * grps(A) =
* round_down(V * ncpu(A) / N) = * round_down(G * ncpu(A) / N) =
* round_down((N - delta) * ncpu(A) / N) = * round_down((N - delta) * ncpu(A) / N) =
* round_down((N * ncpu(A) - delta * ncpu(A)) / N) >= * round_down((N * ncpu(A) - delta * ncpu(A)) / N) >=
* round_down((N * ncpu(A) - delta * N) / N) = * round_down((N * ncpu(A) - delta * N) / N) =
@ -210,52 +210,50 @@ static void alloc_nodes_vectors(unsigned int numvecs,
* *
* then: * then:
* *
* vecs(A) - V >= ncpu(A) - delta - V * grps(A) - G >= ncpu(A) - delta - G
* => * =>
* V - vecs(A) <= V + delta - ncpu(A) * G - grps(A) <= G + delta - ncpu(A)
* => * =>
* vecs(B) <= N - ncpu(A) * grps(B) <= N - ncpu(A)
* => * =>
* vecs(B) <= cpu(B) * grps(B) <= cpu(B)
* *
* For nodes >= 3, it can be thought as one node and another big * For nodes >= 3, it can be thought as one node and another big
* node given that is exactly what this algorithm is implemented, * node given that is exactly what this algorithm is implemented,
* and we always re-calculate 'remaining_ncpus' & 'numvecs', and * and we always re-calculate 'remaining_ncpus' & 'numgrps', and
* finally for each node X: vecs(X) <= ncpu(X). * finally for each node X: grps(X) <= ncpu(X).
* *
*/ */
for (n = 0; n < nr_node_ids; n++) { for (n = 0; n < nr_node_ids; n++) {
unsigned nvectors, ncpus; unsigned ngroups, ncpus;
if (node_vectors[n].ncpus == UINT_MAX) if (node_groups[n].ncpus == UINT_MAX)
continue; continue;
WARN_ON_ONCE(numvecs == 0); WARN_ON_ONCE(numgrps == 0);
ncpus = node_vectors[n].ncpus; ncpus = node_groups[n].ncpus;
nvectors = max_t(unsigned, 1, ngroups = max_t(unsigned, 1,
numvecs * ncpus / remaining_ncpus); numgrps * ncpus / remaining_ncpus);
WARN_ON_ONCE(nvectors > ncpus); WARN_ON_ONCE(ngroups > ncpus);
node_vectors[n].nvectors = nvectors; node_groups[n].ngroups = ngroups;
remaining_ncpus -= ncpus; remaining_ncpus -= ncpus;
numvecs -= nvectors; numgrps -= ngroups;
} }
} }
static int __irq_build_affinity_masks(unsigned int startvec, static int __group_cpus_evenly(unsigned int startgrp, unsigned int numgrps,
unsigned int numvecs, cpumask_var_t *node_to_cpumask,
cpumask_var_t *node_to_cpumask, const struct cpumask *cpu_mask,
const struct cpumask *cpu_mask, struct cpumask *nmsk, struct cpumask *masks)
struct cpumask *nmsk,
struct cpumask *masks)
{ {
unsigned int i, n, nodes, cpus_per_vec, extra_vecs, done = 0; unsigned int i, n, nodes, cpus_per_grp, extra_grps, done = 0;
unsigned int last_affv = numvecs; unsigned int last_grp = numgrps;
unsigned int curvec = startvec; unsigned int curgrp = startgrp;
nodemask_t nodemsk = NODE_MASK_NONE; nodemask_t nodemsk = NODE_MASK_NONE;
struct node_vectors *node_vectors; struct node_groups *node_groups;
if (cpumask_empty(cpu_mask)) if (cpumask_empty(cpu_mask))
return 0; return 0;
@ -264,34 +262,33 @@ static int __irq_build_affinity_masks(unsigned int startvec,
/* /*
* If the number of nodes in the mask is greater than or equal the * If the number of nodes in the mask is greater than or equal the
* number of vectors we just spread the vectors across the nodes. * number of groups we just spread the groups across the nodes.
*/ */
if (numvecs <= nodes) { if (numgrps <= nodes) {
for_each_node_mask(n, nodemsk) { for_each_node_mask(n, nodemsk) {
/* Ensure that only CPUs which are in both masks are set */ /* Ensure that only CPUs which are in both masks are set */
cpumask_and(nmsk, cpu_mask, node_to_cpumask[n]); cpumask_and(nmsk, cpu_mask, node_to_cpumask[n]);
cpumask_or(&masks[curvec], &masks[curvec], nmsk); cpumask_or(&masks[curgrp], &masks[curgrp], nmsk);
if (++curvec == last_affv) if (++curgrp == last_grp)
curvec = 0; curgrp = 0;
} }
return numvecs; return numgrps;
} }
node_vectors = kcalloc(nr_node_ids, node_groups = kcalloc(nr_node_ids,
sizeof(struct node_vectors), sizeof(struct node_groups),
GFP_KERNEL); GFP_KERNEL);
if (!node_vectors) if (!node_groups)
return -ENOMEM; return -ENOMEM;
/* allocate vector number for each node */ /* allocate group number for each node */
alloc_nodes_vectors(numvecs, node_to_cpumask, cpu_mask, alloc_nodes_groups(numgrps, node_to_cpumask, cpu_mask,
nodemsk, nmsk, node_vectors); nodemsk, nmsk, node_groups);
for (i = 0; i < nr_node_ids; i++) { for (i = 0; i < nr_node_ids; i++) {
unsigned int ncpus, v; unsigned int ncpus, v;
struct node_vectors *nv = &node_vectors[i]; struct node_groups *nv = &node_groups[i];
if (nv->nvectors == UINT_MAX) if (nv->ngroups == UINT_MAX)
continue; continue;
/* Get the cpus on this node which are in the mask */ /* Get the cpus on this node which are in the mask */
@ -300,44 +297,47 @@ static int __irq_build_affinity_masks(unsigned int startvec,
if (!ncpus) if (!ncpus)
continue; continue;
WARN_ON_ONCE(nv->nvectors > ncpus); WARN_ON_ONCE(nv->ngroups > ncpus);
/* Account for rounding errors */ /* Account for rounding errors */
extra_vecs = ncpus - nv->nvectors * (ncpus / nv->nvectors); extra_grps = ncpus - nv->ngroups * (ncpus / nv->ngroups);
/* Spread allocated vectors on CPUs of the current node */ /* Spread allocated groups on CPUs of the current node */
for (v = 0; v < nv->nvectors; v++, curvec++) { for (v = 0; v < nv->ngroups; v++, curgrp++) {
cpus_per_vec = ncpus / nv->nvectors; cpus_per_grp = ncpus / nv->ngroups;
/* Account for extra vectors to compensate rounding errors */ /* Account for extra groups to compensate rounding errors */
if (extra_vecs) { if (extra_grps) {
cpus_per_vec++; cpus_per_grp++;
--extra_vecs; --extra_grps;
} }
/* /*
* wrapping has to be considered given 'startvec' * wrapping has to be considered given 'startgrp'
* may start anywhere * may start anywhere
*/ */
if (curvec >= last_affv) if (curgrp >= last_grp)
curvec = 0; curgrp = 0;
irq_spread_init_one(&masks[curvec], nmsk, grp_spread_init_one(&masks[curgrp], nmsk,
cpus_per_vec); cpus_per_grp);
} }
done += nv->nvectors; done += nv->ngroups;
} }
kfree(node_vectors); kfree(node_groups);
return done; return done;
} }
/* /*
* build affinity in two stages: * build affinity in two stages for each group, and try to put close CPUs
* 1) spread present CPU on these vectors * in viewpoint of CPU and NUMA locality into same group, and we run
* 2) spread other possible CPUs on these vectors * two-stage grouping:
*
* 1) allocate present CPUs on these groups evenly first
* 2) allocate other possible CPUs on these groups evenly
*/ */
static struct cpumask *irq_build_affinity_masks(unsigned int numvecs) static struct cpumask *group_cpus_evenly(unsigned int numgrps)
{ {
unsigned int curvec = 0, nr_present = 0, nr_others = 0; unsigned int curgrp = 0, nr_present = 0, nr_others = 0;
cpumask_var_t *node_to_cpumask; cpumask_var_t *node_to_cpumask;
cpumask_var_t nmsk, npresmsk; cpumask_var_t nmsk, npresmsk;
int ret = -ENOMEM; int ret = -ENOMEM;
@ -353,7 +353,7 @@ static struct cpumask *irq_build_affinity_masks(unsigned int numvecs)
if (!node_to_cpumask) if (!node_to_cpumask)
goto fail_npresmsk; goto fail_npresmsk;
masks = kcalloc(numvecs, sizeof(*masks), GFP_KERNEL); masks = kcalloc(numgrps, sizeof(*masks), GFP_KERNEL);
if (!masks) if (!masks)
goto fail_node_to_cpumask; goto fail_node_to_cpumask;
@ -361,26 +361,26 @@ static struct cpumask *irq_build_affinity_masks(unsigned int numvecs)
cpus_read_lock(); cpus_read_lock();
build_node_to_cpumask(node_to_cpumask); build_node_to_cpumask(node_to_cpumask);
/* Spread on present CPUs starting from affd->pre_vectors */ /* grouping present CPUs first */
ret = __irq_build_affinity_masks(curvec, numvecs, node_to_cpumask, ret = __group_cpus_evenly(curgrp, numgrps, node_to_cpumask,
cpu_present_mask, nmsk, masks); cpu_present_mask, nmsk, masks);
if (ret < 0) if (ret < 0)
goto fail_build_affinity; goto fail_build_affinity;
nr_present = ret; nr_present = ret;
/* /*
* Spread on non present CPUs starting from the next vector to be * Allocate non present CPUs starting from the next group to be
* handled. If the spreading of present CPUs already exhausted the * handled. If the grouping of present CPUs already exhausted the
* vector space, assign the non present CPUs to the already spread * group space, assign the non present CPUs to the already
* out vectors. * allocated out groups.
*/ */
if (nr_present >= numvecs) if (nr_present >= numgrps)
curvec = 0; curgrp = 0;
else else
curvec = nr_present; curgrp = nr_present;
cpumask_andnot(npresmsk, cpu_possible_mask, cpu_present_mask); cpumask_andnot(npresmsk, cpu_possible_mask, cpu_present_mask);
ret = __irq_build_affinity_masks(curvec, numvecs, node_to_cpumask, ret = __group_cpus_evenly(curgrp, numgrps, node_to_cpumask,
npresmsk, nmsk, masks); npresmsk, nmsk, masks);
if (ret >= 0) if (ret >= 0)
nr_others = ret; nr_others = ret;
@ -388,7 +388,7 @@ static struct cpumask *irq_build_affinity_masks(unsigned int numvecs)
cpus_read_unlock(); cpus_read_unlock();
if (ret >= 0) if (ret >= 0)
WARN_ON(nr_present + nr_others < numvecs); WARN_ON(nr_present + nr_others < numgrps);
fail_node_to_cpumask: fail_node_to_cpumask:
free_node_to_cpumask(node_to_cpumask); free_node_to_cpumask(node_to_cpumask);
@ -467,7 +467,7 @@ irq_create_affinity_masks(unsigned int nvecs, struct irq_affinity *affd)
for (i = 0, usedvecs = 0; i < affd->nr_sets; i++) { for (i = 0, usedvecs = 0; i < affd->nr_sets; i++) {
unsigned int this_vecs = affd->set_size[i]; unsigned int this_vecs = affd->set_size[i];
int j; int j;
struct cpumask *result = irq_build_affinity_masks(this_vecs); struct cpumask *result = group_cpus_evenly(this_vecs);
if (!result) { if (!result) {
kfree(masks); kfree(masks);