Unbound workqueue CPU affinity is going to receive an overhaul and the NUMA specific knobs won't make sense anymore. Remove them. Also, the pool_ids knob was used for debugging and not really meaningful given that there is no visibility into the pools associated with those IDs. Remove it too. A future patch will improve overall visibility.
Unified · Split
| @@ -6992,15 +6992,6 @@ · 6991 unmodified lines | |||
| 6992 | threshold repeatedly. They are likely good | 6992 | threshold repeatedly. They are likely good |
| 6993 | candidates for using WQ_UNBOUND workqueues instead. | 6993 | candidates for using WQ_UNBOUND workqueues instead. |
| 6994 | 6994 | ||
| 6995 | workqueue.disable_numa | ||
| 6996 | By default, all work items queued to unbound | ||
| 6997 | workqueues are affine to the NUMA nodes they're | ||
| 6998 | issued on, which results in better behavior in | ||
| 6999 | general. If NUMA affinity needs to be disabled for | ||
| 7000 | whatever reason, this option can be used. Note | ||
| 7001 | that this also can be controlled per-workqueue for | ||
| 7002 | workqueues visible under /sys/bus/workqueue/. | ||
| 7003 | |||
| 7004 | workqueue.power_efficient | 6995 | workqueue.power_efficient |
| 7005 | Per-cpu workqueues are generally preferred because | 6996 | Per-cpu workqueues are generally preferred because |
| 7006 | they show better performance thanks to cache | 6997 | they show better performance thanks to cache |
| @@ -340,9 +340,6 @@ static cpumask_var_t *wq_numa_possible_cpumask; · 339 unmodified lines | |||
| 340 | static unsigned long wq_cpu_intensive_thresh_us = ULONG_MAX; | 340 | static unsigned long wq_cpu_intensive_thresh_us = ULONG_MAX; |
| 341 | module_param_named(cpu_intensive_thresh_us, wq_cpu_intensive_thresh_us, ulong, 0644); | 341 | module_param_named(cpu_intensive_thresh_us, wq_cpu_intensive_thresh_us, ulong, 0644); |
| 342 | 342 | ||
| 343 | static bool wq_disable_numa; | ||
| 344 | module_param_named(disable_numa, wq_disable_numa, bool, 0444); | ||
| 345 | |||
| 346 | /* see the comment above the definition of WQ_POWER_EFFICIENT */ | 343 | /* see the comment above the definition of WQ_POWER_EFFICIENT */ |
| 347 | static bool wq_power_efficient = IS_ENABLED(CONFIG_WQ_POWER_EFFICIENT_DEFAULT); | 344 | static bool wq_power_efficient = IS_ENABLED(CONFIG_WQ_POWER_EFFICIENT_DEFAULT); |
| 348 | module_param_named(power_efficient, wq_power_efficient, bool, 0444); | 345 | module_param_named(power_efficient, wq_power_efficient, bool, 0444); |
| @@ -5794,10 +5791,8 @@ out_unlock: · 5445 unmodified lines | |||
| 5794 | * | 5791 | * |
| 5795 | * Unbound workqueues have the following extra attributes. | 5792 | * Unbound workqueues have the following extra attributes. |
| 5796 | * | 5793 | * |
| 5797 | * pool_ids RO int : the associated pool IDs for each node | ||
| 5798 | * nice RW int : nice value of the workers | 5794 | * nice RW int : nice value of the workers |
| 5799 | * cpumask RW mask : bitmask of allowed CPUs for the workers | 5795 | * cpumask RW mask : bitmask of allowed CPUs for the workers |
| 5800 | * numa RW bool : whether enable NUMA affinity | ||
| 5801 | */ | 5796 | */ |
| 5802 | struct wq_device { | 5797 | struct wq_device { |
| 5803 | struct workqueue_struct *wq; | 5798 | struct workqueue_struct *wq; |
| @@ -5850,28 +5845,6 @@ static struct attribute *wq_sysfs_attrs[] = { · 46 unmodified lines | |||
| 5850 | }; | 5845 | }; |
| 5851 | ATTRIBUTE_GROUPS(wq_sysfs); | 5846 | ATTRIBUTE_GROUPS(wq_sysfs); |
| 5852 | 5847 | ||
| 5853 | static ssize_t wq_pool_ids_show(struct device *dev, | ||
| 5854 | struct device_attribute *attr, char *buf) | ||
| 5855 | { | ||
| 5856 | struct workqueue_struct *wq = dev_to_wq(dev); | ||
| 5857 | const char *delim = ""; | ||
| 5858 | int node, written = 0; | ||
| 5859 | |||
| 5860 | cpus_read_lock(); | ||
| 5861 | rcu_read_lock(); | ||
| 5862 | for_each_node(node) { | ||
| 5863 | written += scnprintf(buf + written, PAGE_SIZE - written, | ||
| 5864 | "%s%d:%d", delim, node, | ||
| 5865 | unbound_pwq_by_node(wq, node)->pool->id); | ||
| 5866 | delim = " "; | ||
| 5867 | } | ||
| 5868 | written += scnprintf(buf + written, PAGE_SIZE - written, "\n"); | ||
| 5869 | rcu_read_unlock(); | ||
| 5870 | cpus_read_unlock(); | ||
| 5871 | |||
| 5872 | return written; | ||
| 5873 | } | ||
| 5874 | |||
| 5875 | static ssize_t wq_nice_show(struct device *dev, struct device_attribute *attr, | 5848 | static ssize_t wq_nice_show(struct device *dev, struct device_attribute *attr, |
| 5876 | char *buf) | 5849 | char *buf) |
| 5877 | { | 5850 | { |
| @@ -5962,50 +5935,9 @@ out_unlock: · 84 unmodified lines | |||
| 5962 | return ret ?: count; | 5935 | return ret ?: count; |
| 5963 | } | 5936 | } |
| 5964 | 5937 | ||
| 5965 | static ssize_t wq_numa_show(struct device *dev, struct device_attribute *attr, | ||
| 5966 | char *buf) | ||
| 5967 | { | ||
| 5968 | struct workqueue_struct *wq = dev_to_wq(dev); | ||
| 5969 | int written; | ||
| 5970 | |||
| 5971 | mutex_lock(&wq->mutex); | ||
| 5972 | written = scnprintf(buf, PAGE_SIZE, "%d\n", | ||
| 5973 | !wq->unbound_attrs->no_numa); | ||
| 5974 | mutex_unlock(&wq->mutex); | ||
| 5975 | |||
| 5976 | return written; | ||
| 5977 | } | ||
| 5978 | |||
| 5979 | static ssize_t wq_numa_store(struct device *dev, struct device_attribute *attr, | ||
| 5980 | const char *buf, size_t count) | ||
| 5981 | { | ||
| 5982 | struct workqueue_struct *wq = dev_to_wq(dev); | ||
| 5983 | struct workqueue_attrs *attrs; | ||
| 5984 | int v, ret = -ENOMEM; | ||
| 5985 | |||
| 5986 | apply_wqattrs_lock(); | ||
| 5987 | |||
| 5988 | attrs = wq_sysfs_prep_attrs(wq); | ||
| 5989 | if (!attrs) | ||
| 5990 | goto out_unlock; | ||
| 5991 | |||
| 5992 | ret = -EINVAL; | ||
| 5993 | if (sscanf(buf, "%d", &v) == 1) { | ||
| 5994 | attrs->no_numa = !v; | ||
| 5995 | ret = apply_workqueue_attrs_locked(wq, attrs); | ||
| 5996 | } | ||
| 5997 | |||
| 5998 | out_unlock: | ||
| 5999 | apply_wqattrs_unlock(); | ||
| 6000 | free_workqueue_attrs(attrs); | ||
| 6001 | return ret ?: count; | ||
| 6002 | } | ||
| 6003 | |||
| 6004 | static struct device_attribute wq_sysfs_unbound_attrs[] = { | 5938 | static struct device_attribute wq_sysfs_unbound_attrs[] = { |
| 6005 | __ATTR(pool_ids, 0444, wq_pool_ids_show, NULL), | ||
| 6006 | __ATTR(nice, 0644, wq_nice_show, wq_nice_store), | 5939 | __ATTR(nice, 0644, wq_nice_show, wq_nice_store), |
| 6007 | __ATTR(cpumask, 0644, wq_cpumask_show, wq_cpumask_store), | 5940 | __ATTR(cpumask, 0644, wq_cpumask_show, wq_cpumask_store), |
| 6008 | __ATTR(numa, 0644, wq_numa_show, wq_numa_store), | ||
| 6009 | __ATTR_NULL, | 5941 | __ATTR_NULL, |
| 6010 | }; | 5942 | }; |
| 6011 | 5943 | ||
| @@ -6379,11 +6311,6 @@ static void __init wq_numa_init(void) · 367 unmodified lines | |||
| 6379 | if (num_possible_nodes() <= 1) | 6311 | if (num_possible_nodes() <= 1) |
| 6380 | return; | 6312 | return; |
| 6381 | 6313 | ||
| 6382 | if (wq_disable_numa) { | ||
| 6383 | pr_info("workqueue: NUMA affinity support disabled\n"); | ||
| 6384 | return; | ||
| 6385 | } | ||
| 6386 | |||
| 6387 | for_each_possible_cpu(cpu) { | 6314 | for_each_possible_cpu(cpu) { |
| 6388 | if (WARN_ON(cpu_to_node(cpu) == NUMA_NO_NODE)) { | 6315 | if (WARN_ON(cpu_to_node(cpu) == NUMA_NO_NODE)) { |
| 6389 | pr_warn("workqueue: NUMA node mapping not available for cpu%d, disabling NUMA support\n", cpu); | 6316 | pr_warn("workqueue: NUMA node mapping not available for cpu%d, disabling NUMA support\n", cpu); |