diff options
Diffstat (limited to 'mm/damon/core.c')
| -rw-r--r-- | mm/damon/core.c | 1063 |
1 files changed, 981 insertions, 82 deletions
diff --git a/mm/damon/core.c b/mm/damon/core.c index 6d503c1c125e..339116ea30e3 100644 --- a/mm/damon/core.c +++ b/mm/damon/core.c @@ -14,6 +14,7 @@ #include <linux/psi.h> #include <linux/slab.h> #include <linux/string.h> +#include <linux/string_choices.h> #define CREATE_TRACE_POINTS #include <trace/events/damon.h> @@ -75,14 +76,13 @@ int damon_register_ops(struct damon_operations *ops) if (ops->id >= NR_DAMON_OPS) return -EINVAL; + mutex_lock(&damon_ops_lock); /* Fail for already registered ops */ - if (__damon_is_registered_ops(ops->id)) { + if (__damon_is_registered_ops(ops->id)) err = -EINVAL; - goto out; - } - damon_registered_ops[ops->id] = *ops; -out: + else + damon_registered_ops[ops->id] = *ops; mutex_unlock(&damon_ops_lock); return err; } @@ -266,7 +266,7 @@ int damon_set_regions(struct damon_target *t, struct damon_addr_range *ranges, } struct damos_filter *damos_new_filter(enum damos_filter_type type, - bool matching) + bool matching, bool allow) { struct damos_filter *filter; @@ -275,13 +275,36 @@ struct damos_filter *damos_new_filter(enum damos_filter_type type, return NULL; filter->type = type; filter->matching = matching; + filter->allow = allow; INIT_LIST_HEAD(&filter->list); return filter; } +/** + * damos_filter_for_ops() - Return if the filter is ops-hndled one. + * @type: type of the filter. + * + * Return: true if the filter of @type needs to be handled by ops layer, false + * otherwise. + */ +bool damos_filter_for_ops(enum damos_filter_type type) +{ + switch (type) { + case DAMOS_FILTER_TYPE_ADDR: + case DAMOS_FILTER_TYPE_TARGET: + return false; + default: + break; + } + return true; +} + void damos_add_filter(struct damos *s, struct damos_filter *f) { - list_add_tail(&f->list, &s->filters); + if (damos_filter_for_ops(f->type)) + list_add_tail(&f->list, &s->ops_filters); + else + list_add_tail(&f->list, &s->filters); } static void damos_del_filter(struct damos_filter *f) @@ -346,6 +369,7 @@ static struct damos_quota *damos_quota_init(struct damos_quota *quota) quota->charged_from = 0; quota->charge_target_from = NULL; quota->charge_addr_from = 0; + quota->esz_bp = 0; return quota; } @@ -353,7 +377,8 @@ struct damos *damon_new_scheme(struct damos_access_pattern *pattern, enum damos_action action, unsigned long apply_interval_us, struct damos_quota *quota, - struct damos_watermarks *wmarks) + struct damos_watermarks *wmarks, + int target_nid) { struct damos *scheme; @@ -369,7 +394,9 @@ struct damos *damon_new_scheme(struct damos_access_pattern *pattern, * or damon_attrs are updated. */ scheme->next_apply_sis = 0; + scheme->walk_completed = false; INIT_LIST_HEAD(&scheme->filters); + INIT_LIST_HEAD(&scheme->ops_filters); scheme->stat = (struct damos_stat){}; INIT_LIST_HEAD(&scheme->list); @@ -380,6 +407,8 @@ struct damos *damon_new_scheme(struct damos_access_pattern *pattern, scheme->wmarks = *wmarks; scheme->wmarks.activated = true; + scheme->target_nid = target_nid; + return scheme; } @@ -495,11 +524,13 @@ struct damon_ctx *damon_new_ctx(void) ctx->attrs.ops_update_interval = 60 * 1000 * 1000; ctx->passed_sample_intervals = 0; - /* These will be set from kdamond_init_intervals_sis() */ + /* These will be set from kdamond_init_ctx() */ ctx->next_aggregation_sis = 0; ctx->next_ops_update_sis = 0; mutex_init(&ctx->kdamond_lock); + mutex_init(&ctx->call_control_lock); + mutex_init(&ctx->walk_control_lock); ctx->attrs.min_nr_regions = 10; ctx->attrs.max_nr_regions = 1000; @@ -548,7 +579,13 @@ static unsigned int damon_accesses_bp_to_nr_accesses( return accesses_bp * damon_max_nr_accesses(attrs) / 10000; } -/* convert nr_accesses to access ratio in bp (per 10,000) */ +/* + * Convert nr_accesses to access ratio in bp (per 10,000). + * + * Callers should ensure attrs.aggr_interval is not zero, like + * damon_update_monitoring_results() does . Otherwise, divide-by-zero would + * happen. + */ static unsigned int damon_nr_accesses_to_accesses_bp( unsigned int nr_accesses, struct damon_attrs *attrs) { @@ -565,11 +602,25 @@ static unsigned int damon_nr_accesses_for_new_attrs(unsigned int nr_accesses, } static void damon_update_monitoring_result(struct damon_region *r, - struct damon_attrs *old_attrs, struct damon_attrs *new_attrs) -{ - r->nr_accesses = damon_nr_accesses_for_new_attrs(r->nr_accesses, - old_attrs, new_attrs); - r->nr_accesses_bp = r->nr_accesses * 10000; + struct damon_attrs *old_attrs, struct damon_attrs *new_attrs, + bool aggregating) +{ + if (!aggregating) { + r->nr_accesses = damon_nr_accesses_for_new_attrs( + r->nr_accesses, old_attrs, new_attrs); + r->nr_accesses_bp = r->nr_accesses * 10000; + } else { + /* + * if this is called in the middle of the aggregation, reset + * the aggregations we made so far for this aggregation + * interval. In other words, make the status like + * kdamond_reset_aggregated() is called. + */ + r->last_nr_accesses = damon_nr_accesses_for_new_attrs( + r->last_nr_accesses, old_attrs, new_attrs); + r->nr_accesses_bp = r->last_nr_accesses * 10000; + r->nr_accesses = 0; + } r->age = damon_age_for_new_attrs(r->age, old_attrs, new_attrs); } @@ -582,7 +633,7 @@ static void damon_update_monitoring_result(struct damon_region *r, * ->nr_accesses and ->age of given damon_ctx's regions for new damon_attrs. */ static void damon_update_monitoring_results(struct damon_ctx *ctx, - struct damon_attrs *new_attrs) + struct damon_attrs *new_attrs, bool aggregating) { struct damon_attrs *old_attrs = &ctx->attrs; struct damon_target *t; @@ -597,7 +648,26 @@ static void damon_update_monitoring_results(struct damon_ctx *ctx, damon_for_each_target(t, ctx) damon_for_each_region(r, t) damon_update_monitoring_result( - r, old_attrs, new_attrs); + r, old_attrs, new_attrs, aggregating); +} + +/* + * damon_valid_intervals_goal() - return if the intervals goal of @attrs is + * valid. + */ +static bool damon_valid_intervals_goal(struct damon_attrs *attrs) +{ + struct damon_intervals_goal *goal = &attrs->intervals_goal; + + /* tuning is disabled */ + if (!goal->aggrs) + return true; + if (goal->min_sample_us > goal->max_sample_us) + return false; + if (attrs->sample_interval < goal->min_sample_us || + goal->max_sample_us < attrs->sample_interval) + return false; + return true; } /** @@ -605,10 +675,10 @@ static void damon_update_monitoring_results(struct damon_ctx *ctx, * @ctx: monitoring context * @attrs: monitoring attributes * - * This function should be called while the kdamond is not running, or an - * access check results aggregation is not ongoing (e.g., from - * &struct damon_callback->after_aggregation or - * &struct damon_callback->after_wmarks_check callbacks). + * This function should be called while the kdamond is not running, an access + * check results aggregation is not ongoing (e.g., from &struct + * damon_callback->after_aggregation or &struct + * damon_callback->after_wmarks_check callbacks), or from damon_call(). * * Every time interval is in micro-seconds. * @@ -619,6 +689,11 @@ int damon_set_attrs(struct damon_ctx *ctx, struct damon_attrs *attrs) unsigned long sample_interval = attrs->sample_interval ? attrs->sample_interval : 1; struct damos *s; + bool aggregating = ctx->passed_sample_intervals < + ctx->next_aggregation_sis; + + if (!damon_valid_intervals_goal(attrs)) + return -EINVAL; if (attrs->min_nr_regions < 3) return -EINVAL; @@ -627,12 +702,16 @@ int damon_set_attrs(struct damon_ctx *ctx, struct damon_attrs *attrs) if (attrs->sample_interval > attrs->aggr_interval) return -EINVAL; + /* calls from core-external doesn't set this. */ + if (!attrs->aggr_samples) + attrs->aggr_samples = attrs->aggr_interval / sample_interval; + ctx->next_aggregation_sis = ctx->passed_sample_intervals + attrs->aggr_interval / sample_interval; ctx->next_ops_update_sis = ctx->passed_sample_intervals + attrs->ops_update_interval / sample_interval; - damon_update_monitoring_results(ctx, attrs); + damon_update_monitoring_results(ctx, attrs, aggregating); ctx->attrs = *attrs; damon_for_each_scheme(s, ctx) @@ -662,6 +741,442 @@ void damon_set_schemes(struct damon_ctx *ctx, struct damos **schemes, damon_add_scheme(ctx, schemes[i]); } +static struct damos_quota_goal *damos_nth_quota_goal( + int n, struct damos_quota *q) +{ + struct damos_quota_goal *goal; + int i = 0; + + damos_for_each_quota_goal(goal, q) { + if (i++ == n) + return goal; + } + return NULL; +} + +static void damos_commit_quota_goal_union( + struct damos_quota_goal *dst, struct damos_quota_goal *src) +{ + switch (dst->metric) { + case DAMOS_QUOTA_NODE_MEM_USED_BP: + case DAMOS_QUOTA_NODE_MEM_FREE_BP: + dst->nid = src->nid; + break; + default: + break; + } +} + +static void damos_commit_quota_goal( + struct damos_quota_goal *dst, struct damos_quota_goal *src) +{ + dst->metric = src->metric; + dst->target_value = src->target_value; + if (dst->metric == DAMOS_QUOTA_USER_INPUT) + dst->current_value = src->current_value; + /* keep last_psi_total as is, since it will be updated in next cycle */ + damos_commit_quota_goal_union(dst, src); +} + +/** + * damos_commit_quota_goals() - Commit DAMOS quota goals to another quota. + * @dst: The commit destination DAMOS quota. + * @src: The commit source DAMOS quota. + * + * Copies user-specified parameters for quota goals from @src to @dst. Users + * should use this function for quota goals-level parameters update of running + * DAMON contexts, instead of manual in-place updates. + * + * This function should be called from parameters-update safe context, like + * DAMON callbacks. + */ +int damos_commit_quota_goals(struct damos_quota *dst, struct damos_quota *src) +{ + struct damos_quota_goal *dst_goal, *next, *src_goal, *new_goal; + int i = 0, j = 0; + + damos_for_each_quota_goal_safe(dst_goal, next, dst) { + src_goal = damos_nth_quota_goal(i++, src); + if (src_goal) + damos_commit_quota_goal(dst_goal, src_goal); + else + damos_destroy_quota_goal(dst_goal); + } + damos_for_each_quota_goal_safe(src_goal, next, src) { + if (j++ < i) + continue; + new_goal = damos_new_quota_goal( + src_goal->metric, src_goal->target_value); + if (!new_goal) + return -ENOMEM; + damos_commit_quota_goal_union(new_goal, src_goal); + damos_add_quota_goal(dst, new_goal); + } + return 0; +} + +static int damos_commit_quota(struct damos_quota *dst, struct damos_quota *src) +{ + int err; + + dst->reset_interval = src->reset_interval; + dst->ms = src->ms; + dst->sz = src->sz; + err = damos_commit_quota_goals(dst, src); + if (err) + return err; + dst->weight_sz = src->weight_sz; + dst->weight_nr_accesses = src->weight_nr_accesses; + dst->weight_age = src->weight_age; + return 0; +} + +static struct damos_filter *damos_nth_filter(int n, struct damos *s) +{ + struct damos_filter *filter; + int i = 0; + + damos_for_each_filter(filter, s) { + if (i++ == n) + return filter; + } + return NULL; +} + +static void damos_commit_filter_arg( + struct damos_filter *dst, struct damos_filter *src) +{ + switch (dst->type) { + case DAMOS_FILTER_TYPE_MEMCG: + dst->memcg_id = src->memcg_id; + break; + case DAMOS_FILTER_TYPE_ADDR: + dst->addr_range = src->addr_range; + break; + case DAMOS_FILTER_TYPE_TARGET: + dst->target_idx = src->target_idx; + break; + case DAMOS_FILTER_TYPE_HUGEPAGE_SIZE: + dst->sz_range = src->sz_range; + break; + default: + break; + } +} + +static void damos_commit_filter( + struct damos_filter *dst, struct damos_filter *src) +{ + dst->type = src->type; + dst->matching = src->matching; + damos_commit_filter_arg(dst, src); +} + +static int damos_commit_core_filters(struct damos *dst, struct damos *src) +{ + struct damos_filter *dst_filter, *next, *src_filter, *new_filter; + int i = 0, j = 0; + + damos_for_each_filter_safe(dst_filter, next, dst) { + src_filter = damos_nth_filter(i++, src); + if (src_filter) + damos_commit_filter(dst_filter, src_filter); + else + damos_destroy_filter(dst_filter); + } + + damos_for_each_filter_safe(src_filter, next, src) { + if (j++ < i) + continue; + + new_filter = damos_new_filter( + src_filter->type, src_filter->matching, + src_filter->allow); + if (!new_filter) + return -ENOMEM; + damos_commit_filter_arg(new_filter, src_filter); + damos_add_filter(dst, new_filter); + } + return 0; +} + +static int damos_commit_ops_filters(struct damos *dst, struct damos *src) +{ + struct damos_filter *dst_filter, *next, *src_filter, *new_filter; + int i = 0, j = 0; + + damos_for_each_ops_filter_safe(dst_filter, next, dst) { + src_filter = damos_nth_filter(i++, src); + if (src_filter) + damos_commit_filter(dst_filter, src_filter); + else + damos_destroy_filter(dst_filter); + } + + damos_for_each_ops_filter_safe(src_filter, next, src) { + if (j++ < i) + continue; + + new_filter = damos_new_filter( + src_filter->type, src_filter->matching, + src_filter->allow); + if (!new_filter) + return -ENOMEM; + damos_commit_filter_arg(new_filter, src_filter); + damos_add_filter(dst, new_filter); + } + return 0; +} + +/** + * damos_filters_default_reject() - decide whether to reject memory that didn't + * match with any given filter. + * @filters: Given DAMOS filters of a group. + */ +static bool damos_filters_default_reject(struct list_head *filters) +{ + struct damos_filter *last_filter; + + if (list_empty(filters)) + return false; + last_filter = list_last_entry(filters, struct damos_filter, list); + return last_filter->allow; +} + +static void damos_set_filters_default_reject(struct damos *s) +{ + if (!list_empty(&s->ops_filters)) + s->core_filters_default_reject = false; + else + s->core_filters_default_reject = + damos_filters_default_reject(&s->filters); + s->ops_filters_default_reject = + damos_filters_default_reject(&s->ops_filters); +} + +static int damos_commit_filters(struct damos *dst, struct damos *src) +{ + int err; + + err = damos_commit_core_filters(dst, src); + if (err) + return err; + err = damos_commit_ops_filters(dst, src); + if (err) + return err; + damos_set_filters_default_reject(dst); + return 0; +} + +static struct damos *damon_nth_scheme(int n, struct damon_ctx *ctx) +{ + struct damos *s; + int i = 0; + + damon_for_each_scheme(s, ctx) { + if (i++ == n) + return s; + } + return NULL; +} + +static int damos_commit(struct damos *dst, struct damos *src) +{ + int err; + + dst->pattern = src->pattern; + dst->action = src->action; + dst->apply_interval_us = src->apply_interval_us; + + err = damos_commit_quota(&dst->quota, &src->quota); + if (err) + return err; + + dst->wmarks = src->wmarks; + + err = damos_commit_filters(dst, src); + return err; +} + +static int damon_commit_schemes(struct damon_ctx *dst, struct damon_ctx *src) +{ + struct damos *dst_scheme, *next, *src_scheme, *new_scheme; + int i = 0, j = 0, err; + + damon_for_each_scheme_safe(dst_scheme, next, dst) { + src_scheme = damon_nth_scheme(i++, src); + if (src_scheme) { + err = damos_commit(dst_scheme, src_scheme); + if (err) + return err; + } else { + damon_destroy_scheme(dst_scheme); + } + } + + damon_for_each_scheme_safe(src_scheme, next, src) { + if (j++ < i) + continue; + new_scheme = damon_new_scheme(&src_scheme->pattern, + src_scheme->action, + src_scheme->apply_interval_us, + &src_scheme->quota, &src_scheme->wmarks, + NUMA_NO_NODE); + if (!new_scheme) + return -ENOMEM; + err = damos_commit(new_scheme, src_scheme); + if (err) { + damon_destroy_scheme(new_scheme); + return err; + } + damon_add_scheme(dst, new_scheme); + } + return 0; +} + +static struct damon_target *damon_nth_target(int n, struct damon_ctx *ctx) +{ + struct damon_target *t; + int i = 0; + + damon_for_each_target(t, ctx) { + if (i++ == n) + return t; + } + return NULL; +} + +/* + * The caller should ensure the regions of @src are + * 1. valid (end >= src) and + * 2. sorted by starting address. + * + * If @src has no region, @dst keeps current regions. + */ +static int damon_commit_target_regions( + struct damon_target *dst, struct damon_target *src) +{ + struct damon_region *src_region; + struct damon_addr_range *ranges; + int i = 0, err; + + damon_for_each_region(src_region, src) + i++; + if (!i) + return 0; + + ranges = kmalloc_array(i, sizeof(*ranges), GFP_KERNEL | __GFP_NOWARN); + if (!ranges) + return -ENOMEM; + i = 0; + damon_for_each_region(src_region, src) + ranges[i++] = src_region->ar; + err = damon_set_regions(dst, ranges, i); + kfree(ranges); + return err; +} + +static int damon_commit_target( + struct damon_target *dst, bool dst_has_pid, + struct damon_target *src, bool src_has_pid) +{ + int err; + + err = damon_commit_target_regions(dst, src); + if (err) + return err; + if (dst_has_pid) + put_pid(dst->pid); + if (src_has_pid) + get_pid(src->pid); + dst->pid = src->pid; + return 0; +} + +static int damon_commit_targets( + struct damon_ctx *dst, struct damon_ctx *src) +{ + struct damon_target *dst_target, *next, *src_target, *new_target; + int i = 0, j = 0, err; + + damon_for_each_target_safe(dst_target, next, dst) { + src_target = damon_nth_target(i++, src); + if (src_target) { + err = damon_commit_target( + dst_target, damon_target_has_pid(dst), + src_target, damon_target_has_pid(src)); + if (err) + return err; + } else { + struct damos *s; + + if (damon_target_has_pid(dst)) + put_pid(dst_target->pid); + damon_destroy_target(dst_target); + damon_for_each_scheme(s, dst) { + if (s->quota.charge_target_from == dst_target) { + s->quota.charge_target_from = NULL; + s->quota.charge_addr_from = 0; + } + } + } + } + + damon_for_each_target_safe(src_target, next, src) { + if (j++ < i) + continue; + new_target = damon_new_target(); + if (!new_target) + return -ENOMEM; + err = damon_commit_target(new_target, false, + src_target, damon_target_has_pid(src)); + if (err) { + damon_destroy_target(new_target); + return err; + } + damon_add_target(dst, new_target); + } + return 0; +} + +/** + * damon_commit_ctx() - Commit parameters of a DAMON context to another. + * @dst: The commit destination DAMON context. + * @src: The commit source DAMON context. + * + * This function copies user-specified parameters from @src to @dst and update + * the internal status and results accordingly. Users should use this function + * for context-level parameters update of running context, instead of manual + * in-place updates. + * + * This function should be called from parameters-update safe context, like + * DAMON callbacks. + */ +int damon_commit_ctx(struct damon_ctx *dst, struct damon_ctx *src) +{ + int err; + + err = damon_commit_schemes(dst, src); + if (err) + return err; + err = damon_commit_targets(dst, src); + if (err) + return err; + /* + * schemes and targets should be updated first, since + * 1. damon_set_attrs() updates monitoring results of targets and + * next_apply_sis of schemes, and + * 2. ops update should be done after pid handling is done (target + * committing require putting pids). + */ + err = damon_set_attrs(dst, &src->attrs); + if (err) + return err; + dst->ops = src->ops; + + return 0; +} + /** * damon_nr_running_ctxs() - Return number of currently running contexts. */ @@ -811,6 +1326,107 @@ int damon_stop(struct damon_ctx **ctxs, int nr_ctxs) return err; } +static bool damon_is_running(struct damon_ctx *ctx) +{ + bool running; + + mutex_lock(&ctx->kdamond_lock); + running = ctx->kdamond != NULL; + mutex_unlock(&ctx->kdamond_lock); + return running; +} + +/** + * damon_call() - Invoke a given function on DAMON worker thread (kdamond). + * @ctx: DAMON context to call the function for. + * @control: Control variable of the call request. + * + * Ask DAMON worker thread (kdamond) of @ctx to call a function with an + * argument data that respectively passed via &damon_call_control->fn and + * &damon_call_control->data of @control, and wait until the kdamond finishes + * handling of the request. + * + * The kdamond executes the function with the argument in the main loop, just + * after a sampling of the iteration is finished. The function can hence + * safely access the internal data of the &struct damon_ctx without additional + * synchronization. The return value of the function will be saved in + * &damon_call_control->return_code. + * + * Return: 0 on success, negative error code otherwise. + */ +int damon_call(struct damon_ctx *ctx, struct damon_call_control *control) +{ + init_completion(&control->completion); + control->canceled = false; + + mutex_lock(&ctx->call_control_lock); + if (ctx->call_control) { + mutex_unlock(&ctx->call_control_lock); + return -EBUSY; + } + ctx->call_control = control; + mutex_unlock(&ctx->call_control_lock); + if (!damon_is_running(ctx)) + return -EINVAL; + wait_for_completion(&control->completion); + if (control->canceled) + return -ECANCELED; + return 0; +} + +/** + * damos_walk() - Invoke a given functions while DAMOS walk regions. + * @ctx: DAMON context to call the functions for. + * @control: Control variable of the walk request. + * + * Ask DAMON worker thread (kdamond) of @ctx to call a function for each region + * that the kdamond will apply DAMOS action to, and wait until the kdamond + * finishes handling of the request. + * + * The kdamond executes the given function in the main loop, for each region + * just after it applied any DAMOS actions of @ctx to it. The invocation is + * made only within one &damos->apply_interval_us since damos_walk() + * invocation, for each scheme. The given callback function can hence safely + * access the internal data of &struct damon_ctx and &struct damon_region that + * each of the scheme will apply the action for next interval, without + * additional synchronizations against the kdamond. If every scheme of @ctx + * passed at least one &damos->apply_interval_us, kdamond marks the request as + * completed so that damos_walk() can wakeup and return. + * + * Return: 0 on success, negative error code otherwise. + */ +int damos_walk(struct damon_ctx *ctx, struct damos_walk_control *control) +{ + init_completion(&control->completion); + control->canceled = false; + mutex_lock(&ctx->walk_control_lock); + if (ctx->walk_control) { + mutex_unlock(&ctx->walk_control_lock); + return -EBUSY; + } + ctx->walk_control = control; + mutex_unlock(&ctx->walk_control_lock); + if (!damon_is_running(ctx)) + return -EINVAL; + wait_for_completion(&control->completion); + if (control->canceled) + return -ECANCELED; + return 0; +} + +/* + * Warn and fix corrupted ->nr_accesses[_bp] for investigations and preventing + * the problem being propagated. + */ +static void damon_warn_fix_nr_accesses_corruption(struct damon_region *r) +{ + if (r->nr_accesses_bp == r->nr_accesses * 10000) + return; + WARN_ONCE(true, "invalid nr_accesses_bp at reset: %u %u\n", + r->nr_accesses_bp, r->nr_accesses); + r->nr_accesses_bp = r->nr_accesses * 10000; +} + /* * Reset the aggregated monitoring results ('nr_accesses' of each region). */ @@ -824,6 +1440,7 @@ static void kdamond_reset_aggregated(struct damon_ctx *c) damon_for_each_region(r, t) { trace_damon_aggregated(ti, r, damon_nr_regions(t)); + damon_warn_fix_nr_accesses_corruption(r); r->last_nr_accesses = r->nr_accesses; r->nr_accesses = 0; } @@ -831,6 +1448,66 @@ static void kdamond_reset_aggregated(struct damon_ctx *c) } } +static unsigned long damon_get_intervals_score(struct damon_ctx *c) +{ + struct damon_target *t; + struct damon_region *r; + unsigned long sz_region, max_access_events = 0, access_events = 0; + unsigned long target_access_events; + unsigned long goal_bp = c->attrs.intervals_goal.access_bp; + + damon_for_each_target(t, c) { + damon_for_each_region(r, t) { + sz_region = damon_sz_region(r); + max_access_events += sz_region * c->attrs.aggr_samples; + access_events += sz_region * r->nr_accesses; + } + } + target_access_events = max_access_events * goal_bp / 10000; + target_access_events = target_access_events ? : 1; + return access_events * 10000 / target_access_events; +} + +static unsigned long damon_feed_loop_next_input(unsigned long last_input, + unsigned long score); + +static unsigned long damon_get_intervals_adaptation_bp(struct damon_ctx *c) +{ + unsigned long score_bp, adaptation_bp; + + score_bp = damon_get_intervals_score(c); + adaptation_bp = damon_feed_loop_next_input(100000000, score_bp) / + 10000; + /* + * adaptaion_bp ranges from 1 to 20,000. Avoid too rapid reduction of + * the intervals by rescaling [1,10,000] to [5000, 10,000]. + */ + if (adaptation_bp <= 10000) + adaptation_bp = 5000 + adaptation_bp / 2; + return adaptation_bp; +} + +static void kdamond_tune_intervals(struct damon_ctx *c) +{ + unsigned long adaptation_bp; + struct damon_attrs new_attrs; + struct damon_intervals_goal *goal; + + adaptation_bp = damon_get_intervals_adaptation_bp(c); + if (adaptation_bp == 10000) + return; + + new_attrs = c->attrs; + goal = &c->attrs.intervals_goal; + new_attrs.sample_interval = min(goal->max_sample_us, + c->attrs.sample_interval * adaptation_bp / 10000); + new_attrs.sample_interval = max(goal->min_sample_us, + new_attrs.sample_interval); + new_attrs.aggr_interval = new_attrs.sample_interval * + c->attrs.aggr_samples; + damon_set_attrs(c, &new_attrs); +} + static void damon_split_region_at(struct damon_target *t, struct damon_region *r, unsigned long sz_r); @@ -921,16 +1598,18 @@ static bool damos_skip_charged_region(struct damon_target *t, } static void damos_update_stat(struct damos *s, - unsigned long sz_tried, unsigned long sz_applied) + unsigned long sz_tried, unsigned long sz_applied, + unsigned long sz_ops_filter_passed) { s->stat.nr_tried++; s->stat.sz_tried += sz_tried; if (sz_applied) s->stat.nr_applied++; s->stat.sz_applied += sz_applied; + s->stat.sz_ops_filter_passed += sz_ops_filter_passed; } -static bool __damos_filter_out(struct damon_ctx *ctx, struct damon_target *t, +static bool damos_filter_match(struct damon_ctx *ctx, struct damon_target *t, struct damon_region *r, struct damos_filter *filter) { bool matched = false; @@ -983,11 +1662,102 @@ static bool damos_filter_out(struct damon_ctx *ctx, struct damon_target *t, { struct damos_filter *filter; + s->core_filters_allowed = false; damos_for_each_filter(filter, s) { - if (__damos_filter_out(ctx, t, r, filter)) - return true; + if (damos_filter_match(ctx, t, r, filter)) { + if (filter->allow) + s->core_filters_allowed = true; + return !filter->allow; + } } - return false; + return s->core_filters_default_reject; +} + +/* + * damos_walk_call_walk() - Call &damos_walk_control->walk_fn. + * @ctx: The context of &damon_ctx->walk_control. + * @t: The monitoring target of @r that @s will be applied. + * @r: The region of @t that @s will be applied. + * @s: The scheme of @ctx that will be applied to @r. + * + * This function is called from kdamond whenever it asked the operation set to + * apply a DAMOS scheme action to a region. If a DAMOS walk request is + * installed by damos_walk() and not yet uninstalled, invoke it. + */ +static void damos_walk_call_walk(struct damon_ctx *ctx, struct damon_target *t, + struct damon_region *r, struct damos *s, + unsigned long sz_filter_passed) +{ + struct damos_walk_control *control; + + if (s->walk_completed) + return; + + control = ctx->walk_control; + if (!control) + return; + + control->walk_fn(control->data, ctx, t, r, s, sz_filter_passed); +} + +/* + * damos_walk_complete() - Complete DAMOS walk request if all walks are done. + * @ctx: The context of &damon_ctx->walk_control. + * @s: A scheme of @ctx that all walks are now done. + * + * This function is called when kdamond finished applying the action of a DAMOS + * scheme to all regions that eligible for the given &damos->apply_interval_us. + * If every scheme of @ctx including @s now finished walking for at least one + * &damos->apply_interval_us, this function makrs the handling of the given + * DAMOS walk request is done, so that damos_walk() can wake up and return. + */ +static void damos_walk_complete(struct damon_ctx *ctx, struct damos *s) +{ + struct damos *siter; + struct damos_walk_control *control; + + control = ctx->walk_control; + if (!control) + return; + + s->walk_completed = true; + /* if all schemes completed, signal completion to walker */ + damon_for_each_scheme(siter, ctx) { + if (!siter->walk_completed) + return; + } + damon_for_each_scheme(siter, ctx) + siter->walk_completed = false; + + complete(&control->completion); + ctx->walk_control = NULL; +} + +/* + * damos_walk_cancel() - Cancel the current DAMOS walk request. + * @ctx: The context of &damon_ctx->walk_control. + * + * This function is called when @ctx is deactivated by DAMOS watermarks, DAMOS + * walk is requested but there is no DAMOS scheme to walk for, or the kdamond + * is already out of the main loop and therefore gonna be terminated, and hence + * cannot continue the walks. This function therefore marks the walk request + * as canceled, so that damos_walk() can wake up and return. + */ +static void damos_walk_cancel(struct damon_ctx *ctx) +{ + struct damos_walk_control *control; + + mutex_lock(&ctx->walk_control_lock); + control = ctx->walk_control; + mutex_unlock(&ctx->walk_control_lock); + + if (!control) + return; + control->canceled = true; + complete(&control->completion); + mutex_lock(&ctx->walk_control_lock); + ctx->walk_control = NULL; + mutex_unlock(&ctx->walk_control_lock); } static void damos_apply_scheme(struct damon_ctx *c, struct damon_target *t, @@ -997,7 +1767,7 @@ static void damos_apply_scheme(struct damon_ctx *c, struct damon_target *t, unsigned long sz = damon_sz_region(r); struct timespec64 begin, end; unsigned long sz_applied = 0; - int err = 0; + unsigned long sz_ops_filter_passed = 0; /* * We plan to support multiple context per kdamond, as DAMON sysfs * implies with 'nr_contexts' file. Nevertheless, only single context @@ -1037,13 +1807,11 @@ static void damos_apply_scheme(struct damon_ctx *c, struct damon_target *t, if (damos_filter_out(c, t, r, s)) return; ktime_get_coarse_ts64(&begin); - if (c->callback.before_damos_apply) - err = c->callback.before_damos_apply(c, t, r, s); - if (!err) { - trace_damos_before_apply(cidx, sidx, tidx, r, - damon_nr_regions(t), do_trace); - sz_applied = c->ops.apply_scheme(c, t, r, s); - } + trace_damos_before_apply(cidx, sidx, tidx, r, + damon_nr_regions(t), do_trace); + sz_applied = c->ops.apply_scheme(c, t, r, s, + &sz_ops_filter_passed); + damos_walk_call_walk(c, t, r, s, sz_ops_filter_passed); ktime_get_coarse_ts64(&end); quota->total_charged_ns += timespec64_to_ns(&end) - timespec64_to_ns(&begin); @@ -1057,7 +1825,7 @@ static void damos_apply_scheme(struct damon_ctx *c, struct damon_target *t, r->age = 0; update_stat: - damos_update_stat(s, sz, sz_applied); + damos_update_stat(s, sz, sz_applied, sz_ops_filter_passed); } static void damon_do_apply_schemes(struct damon_ctx *c, @@ -1069,7 +1837,7 @@ static void damon_do_apply_schemes(struct damon_ctx *c, damon_for_each_scheme(s, c) { struct damos_quota *quota = &s->quota; - if (c->passed_sample_intervals != s->next_apply_sis) + if (c->passed_sample_intervals < s->next_apply_sis) continue; if (!s->wmarks.activated) @@ -1113,17 +1881,31 @@ static unsigned long damon_feed_loop_next_input(unsigned long last_input, unsigned long score) { const unsigned long goal = 10000; - unsigned long score_goal_diff = max(goal, score) - min(goal, score); - unsigned long score_goal_diff_bp = score_goal_diff * 10000 / goal; - unsigned long compensation = last_input * score_goal_diff_bp / 10000; /* Set minimum input as 10000 to avoid compensation be zero */ const unsigned long min_input = 10000; + unsigned long score_goal_diff, compensation; + bool over_achieving = score > goal; + + if (score == goal) + return last_input; + if (score >= goal * 2) + return min_input; - if (goal > score) + if (over_achieving) + score_goal_diff = score - goal; + else + score_goal_diff = goal - score; + + if (last_input < ULONG_MAX / score_goal_diff) + compensation = last_input * score_goal_diff / goal; + else + compensation = last_input / goal * score_goal_diff; + + if (over_achieving) + return max(last_input - compensation, min_input); + if (last_input < ULONG_MAX - compensation) return last_input + compensation; - if (last_input > compensation + min_input) - return last_input - compensation; - return min_input; + return ULONG_MAX; } #ifdef CONFIG_PSI @@ -1145,6 +1927,29 @@ static inline u64 damos_get_some_mem_psi_total(void) #endif /* CONFIG_PSI */ +#ifdef CONFIG_NUMA +static __kernel_ulong_t damos_get_node_mem_bp( + struct damos_quota_goal *goal) +{ + struct sysinfo i; + __kernel_ulong_t numerator; + + si_meminfo_node(&i, goal->nid); + if (goal->metric == DAMOS_QUOTA_NODE_MEM_USED_BP) + numerator = i.totalram - i.freeram; + else /* DAMOS_QUOTA_NODE_MEM_FREE_BP */ + numerator = i.freeram; + return numerator * 10000 / i.totalram; +} +#else +static __kernel_ulong_t damos_get_node_mem_bp( + struct damos_quota_goal *goal) +{ + return 0; +} +#endif + + static void damos_set_quota_goal_current_value(struct damos_quota_goal *goal) { u64 now_psi_total; @@ -1158,6 +1963,10 @@ static void damos_set_quota_goal_current_value(struct damos_quota_goal *goal) goal->current_value = now_psi_total - goal->last_psi_total; goal->last_psi_total = now_psi_total; break; + case DAMOS_QUOTA_NODE_MEM_USED_BP: + case DAMOS_QUOTA_NODE_MEM_FREE_BP: + goal->current_value = damos_get_node_mem_bp(goal); + break; default: break; } @@ -1185,7 +1994,7 @@ static unsigned long damos_quota_score(struct damos_quota *quota) static void damos_set_effective_quota(struct damos_quota *quota) { unsigned long throughput; - unsigned long esz; + unsigned long esz = ULONG_MAX; if (!quota->ms && list_empty("a->goals)) { quota->esz = quota->sz; @@ -1207,10 +2016,7 @@ static void damos_set_effective_quota(struct damos_quota *quota) quota->total_charged_ns; else throughput = PAGE_SIZE * 1024; - if (!list_empty("a->goals)) - esz = min(throughput * quota->ms, esz); - else - esz = throughput * quota->ms; + esz = min(throughput * quota->ms, esz); } if (quota->sz && quota->sz < esz) @@ -1245,13 +2051,16 @@ static void damos_adjust_quota(struct damon_ctx *c, struct damos *s) return; /* Fill up the score histogram */ - memset(quota->histogram, 0, sizeof(quota->histogram)); + memset(c->regions_score_histogram, 0, + sizeof(*c->regions_score_histogram) * + (DAMOS_MAX_SCORE + 1)); damon_for_each_target(t, c) { damon_for_each_region(r, t) { if (!__damos_valid_target(r, s)) continue; score = c->ops.get_scheme_score(c, t, r, s); - quota->histogram[score] += damon_sz_region(r); + c->regions_score_histogram[score] += + damon_sz_region(r); if (score > max_score) max_score = score; } @@ -1259,7 +2068,7 @@ static void damos_adjust_quota(struct damon_ctx *c, struct damos *s) /* Set the min score limit */ for (cumulated_sz = 0, score = max_score; ; score--) { - cumulated_sz += quota->histogram[score]; + cumulated_sz += c->regions_score_histogram[score]; if (cumulated_sz >= quota->esz || !score) break; } @@ -1276,7 +2085,7 @@ static void kdamond_apply_schemes(struct damon_ctx *c) bool has_schemes_to_apply = false; damon_for_each_scheme(s, c) { - if (c->passed_sample_intervals != s->next_apply_sis) + if (c->passed_sample_intervals < s->next_apply_sis) continue; if (!s->wmarks.activated) @@ -1290,18 +2099,22 @@ static void kdamond_apply_schemes(struct damon_ctx *c) if (!has_schemes_to_apply) return; + mutex_lock(&c->walk_control_lock); damon_for_each_target(t, c) { damon_for_each_region_safe(r, next_r, t) damon_do_apply_schemes(c, t, r); } damon_for_each_scheme(s, c) { - if (c->passed_sample_intervals != s->next_apply_sis) + if (c->passed_sample_intervals < s->next_apply_sis) continue; - s->next_apply_sis += + damos_walk_complete(c, s); + s->next_apply_sis = c->passed_sample_intervals + (s->apply_interval_us ? s->apply_interval_us : c->attrs.aggr_interval) / sample_interval; + s->last_applied = NULL; } + mutex_unlock(&c->walk_control_lock); } /* @@ -1357,14 +2170,31 @@ static void damon_merge_regions_of(struct damon_target *t, unsigned int thres, * access frequencies are similar. This is for minimizing the monitoring * overhead under the dynamically changeable access pattern. If a merge was * unnecessarily made, later 'kdamond_split_regions()' will revert it. + * + * The total number of regions could be higher than the user-defined limit, + * max_nr_regions for some cases. For example, the user can update + * max_nr_regions to a number that lower than the current number of regions + * while DAMON is running. For such a case, repeat merging until the limit is + * met while increasing @threshold up to possible maximum level. */ static void kdamond_merge_regions(struct damon_ctx *c, unsigned int threshold, unsigned long sz_limit) { struct damon_target *t; - - damon_for_each_target(t, c) - damon_merge_regions_of(t, threshold, sz_limit); + unsigned int nr_regions; + unsigned int max_thres; + + max_thres = c->attrs.aggr_interval / + (c->attrs.sample_interval ? c->attrs.sample_interval : 1); + do { + nr_regions = 0; + damon_for_each_target(t, c) { + damon_merge_regions_of(t, threshold, sz_limit); + nr_regions += damon_nr_regions(t); + } + threshold = max(1, threshold * 2); + } while (nr_regions > c->attrs.max_nr_regions && + threshold / 2 < max_thres); } /* @@ -1480,12 +2310,14 @@ static bool kdamond_need_stop(struct damon_ctx *ctx) return true; } -static unsigned long damos_wmark_metric_value(enum damos_wmark_metric metric) +static int damos_get_wmark_metric_value(enum damos_wmark_metric metric, + unsigned long *metric_value) { switch (metric) { case DAMOS_WMARK_FREE_MEM_RATE: - return global_zone_page_state(NR_FREE_PAGES) * 1000 / + *metric_value = global_zone_page_state(NR_FREE_PAGES) * 1000 / totalram_pages(); + return 0; default: break; } @@ -1500,17 +2332,15 @@ static unsigned long damos_wmark_wait_us(struct damos *scheme) { unsigned long metric; - if (scheme->wmarks.metric == DAMOS_WMARK_NONE) + if (damos_get_wmark_metric_value(scheme->wmarks.metric, &metric)) return 0; - metric = damos_wmark_metric_value(scheme->wmarks.metric); /* higher than high watermark or lower than low watermark */ if (metric > scheme->wmarks.high || scheme->wmarks.low > metric) { if (scheme->wmarks.activated) pr_debug("deactivate a scheme (%d) for %s wmark\n", - scheme->action, - metric > scheme->wmarks.high ? - "high" : "low"); + scheme->action, + str_high_low(metric > scheme->wmarks.high)); scheme->wmarks.activated = false; return scheme->wmarks.interval; } @@ -1528,11 +2358,42 @@ static unsigned long damos_wmark_wait_us(struct damos *scheme) static void kdamond_usleep(unsigned long usecs) { - /* See Documentation/timers/timers-howto.rst for the thresholds */ - if (usecs > 20 * USEC_PER_MSEC) + if (usecs >= USLEEP_RANGE_UPPER_BOUND) schedule_timeout_idle(usecs_to_jiffies(usecs)); else - usleep_idle_range(usecs, usecs + 1); + usleep_range_idle(usecs, usecs + 1); +} + +/* + * kdamond_call() - handle damon_call_control. + * @ctx: The &struct damon_ctx of the kdamond. + * @cancel: Whether to cancel the invocation of the function. + * + * If there is a &struct damon_call_control request that registered via + * &damon_call() on @ctx, do or cancel the invocation of the function depending + * on @cancel. @cancel is set when the kdamond is already out of the main loop + * and therefore will be terminated. + */ +static void kdamond_call(struct damon_ctx *ctx, bool cancel) +{ + struct damon_call_control *control; + int ret = 0; + + mutex_lock(&ctx->call_control_lock); + control = ctx->call_control; + mutex_unlock(&ctx->call_control_lock); + if (!control) + return; + if (cancel) { + control->canceled = true; + } else { + ret = control->fn(control->data); + control->return_code = ret; + } + complete(&control->completion); + mutex_lock(&ctx->call_control_lock); + ctx->call_control = NULL; + mutex_unlock(&ctx->call_control_lock); } /* Returns negative error code if it's not activated but should return */ @@ -1559,11 +2420,13 @@ static int kdamond_wait_activation(struct damon_ctx *ctx) if (ctx->callback.after_wmarks_check && ctx->callback.after_wmarks_check(ctx)) break; + kdamond_call(ctx, false); + damos_walk_cancel(ctx); } return -EBUSY; } -static void kdamond_init_intervals_sis(struct damon_ctx *ctx) +static void kdamond_init_ctx(struct damon_ctx *ctx) { unsigned long sample_interval = ctx->attrs.sample_interval ? ctx->attrs.sample_interval : 1; @@ -1574,11 +2437,14 @@ static void kdamond_init_intervals_sis(struct damon_ctx *ctx) ctx->next_aggregation_sis = ctx->attrs.aggr_interval / sample_interval; ctx->next_ops_update_sis = ctx->attrs.ops_update_interval / sample_interval; + ctx->next_intervals_tune_sis = ctx->next_aggregation_sis * + ctx->attrs.intervals_goal.aggrs; damon_for_each_scheme(scheme, ctx) { apply_interval = scheme->apply_interval_us ? scheme->apply_interval_us : ctx->attrs.aggr_interval; scheme->next_apply_sis = apply_interval / sample_interval; + damos_set_filters_default_reject(scheme); } } @@ -1596,11 +2462,13 @@ static int kdamond_fn(void *data) pr_debug("kdamond (%d) starts\n", current->pid); complete(&ctx->kdamond_started); - kdamond_init_intervals_sis(ctx); + kdamond_init_ctx(ctx); if (ctx->ops.init) ctx->ops.init(ctx); - if (ctx->callback.before_start && ctx->callback.before_start(ctx)) + ctx->regions_score_histogram = kmalloc_array(DAMOS_MAX_SCORE + 1, + sizeof(*ctx->regions_score_histogram), GFP_KERNEL); + if (!ctx->regions_score_histogram) goto done; sz_limit = damon_region_sz_limit(ctx); @@ -1622,9 +2490,6 @@ static int kdamond_fn(void *data) if (ctx->ops.prepare_access_checks) ctx->ops.prepare_access_checks(ctx); - if (ctx->callback.after_sampling && - ctx->callback.after_sampling(ctx)) - break; kdamond_usleep(sample_interval); ctx->passed_sample_intervals++; @@ -1632,7 +2497,7 @@ static int kdamond_fn(void *data) if (ctx->ops.check_accesses) max_nr_accesses = ctx->ops.check_accesses(ctx); - if (ctx->passed_sample_intervals == next_aggregation_sis) { + if (ctx->passed_sample_intervals >= next_aggregation_sis) { kdamond_merge_regions(ctx, max_nr_accesses / 10, sz_limit); @@ -1642,25 +2507,55 @@ static int kdamond_fn(void *data) } /* - * do kdamond_apply_schemes() after kdamond_merge_regions() if - * possible, to reduce overhead + * do kdamond_call() and kdamond_apply_schemes() after + * kdamond_merge_regions() if possible, to reduce overhead */ + kdamond_call(ctx, false); if (!list_empty(&ctx->schemes)) kdamond_apply_schemes(ctx); + else + damos_walk_cancel(ctx); sample_interval = ctx->attrs.sample_interval ? ctx->attrs.sample_interval : 1; - if (ctx->passed_sample_intervals == next_aggregation_sis) { + if (ctx->passed_sample_intervals >= next_aggregation_sis) { + if (ctx->attrs.intervals_goal.aggrs && + ctx->passed_sample_intervals >= + ctx->next_intervals_tune_sis) { + /* + * ctx->next_aggregation_sis might be updated + * from kdamond_call(). In the case, + * damon_set_attrs() which will be called from + * kdamond_tune_interval() may wrongly think + * this is in the middle of the current + * aggregation, and make aggregation + * information reset for all regions. Then, + * following kdamond_reset_aggregated() call + * will make the region information invalid, + * particularly for ->nr_accesses_bp. + * + * Reset ->next_aggregation_sis to avoid that. + * It will anyway correctly updated after this + * if caluse. + */ + ctx->next_aggregation_sis = + next_aggregation_sis; + ctx->next_intervals_tune_sis += + ctx->attrs.aggr_samples * + ctx->attrs.intervals_goal.aggrs; + kdamond_tune_intervals(ctx); + sample_interval = ctx->attrs.sample_interval ? + ctx->attrs.sample_interval : 1; + + } ctx->next_aggregation_sis = next_aggregation_sis + ctx->attrs.aggr_interval / sample_interval; kdamond_reset_aggregated(ctx); kdamond_split_regions(ctx); - if (ctx->ops.reset_aggregated) - ctx->ops.reset_aggregated(ctx); } - if (ctx->passed_sample_intervals == next_ops_update_sis) { + if (ctx->passed_sample_intervals >= next_ops_update_sis) { ctx->next_ops_update_sis = next_ops_update_sis + ctx->attrs.ops_update_interval / sample_interval; @@ -1679,12 +2574,16 @@ done: ctx->callback.before_terminate(ctx); if (ctx->ops.cleanup) ctx->ops.cleanup(ctx); + kfree(ctx->regions_score_histogram); pr_debug("kdamond (%d) finishes\n", current->pid); mutex_lock(&ctx->kdamond_lock); ctx->kdamond = NULL; mutex_unlock(&ctx->kdamond_lock); + kdamond_call(ctx, true); + damos_walk_cancel(ctx); + mutex_lock(&damon_lock); nr_running_ctxs--; if (!nr_running_ctxs && running_exclusive_ctxs) @@ -1850,4 +2749,4 @@ static int __init damon_init(void) subsys_initcall(damon_init); -#include "core-test.h" +#include "tests/core-kunit.h" |