From 0d6526f82c3cdefcca47f73f5fc08dc6f335eac6 Mon Sep 17 00:00:00 2001 From: Tim Chen Date: Mon, 21 Sep 2026 17:37:22 -0700 Subject: sched/cache: Keep nr_pref_llc_running in the runnable domain, to fix LLC mis-scheduling bug alb_break_llc() decides whether to break LLC preference during active load balance. It does so by testing that every runnable fair task on the source rq prefers its LLC: env->src_rq->nr_pref_llc_running == env->src_rq->cfs.h_nr_runnable But the two counters cover different sets. nr_pref_llc_running is updated in account_llc_enqueue()/account_llc_dequeue(), next to cfs_rq->nr_queued, so it follows queued tasks. h_nr_runnable is updated in set_delayed()/ clear_delayed() and drops delay-dequeued tasks. So under DELAY_DEQUEUE, a preferring task that goes to sleep stays counted in nr_pref_llc_running while h_nr_runnable falls. The equality then breaks, alb_break_llc() returns false, and active balance is free to pull a task off its preferred LLC. Active balance only moves runnable tasks, and this is the only LLC check it consults: once the stopper runs, LBF_ACTIVE_LB skips the per-task test in can_migrate_task(). The runnable set is the one we want. Fix it on the counter side. A task should be counted in nr_pref_llc_running exactly while it is both queued on its preferred LLC (pref_llc_queued) and runnable (!sched_delayed). Define that membership once in task_pref_llc_runnable(), and adjust the counter only through pref_llc_running_inc()/pref_llc_running_dec() from the four sites that change either input: account_llc_enqueue(), account_llc_dequeue(), set_delayed() and clear_delayed(). Gating every update on the same predicate keeps the delay, wake and dequeue paths from double-counting or underflowing; see the comments at those sites for the ordering. nr_llc_running and sd->llc_counts are not touched and stay on queued semantics. Fixes: 714059f79ff0 ("sched/cache: Handle moving single tasks to/from their preferred LLC") Closes: https://lore.kernel.org/lkml/20260827135000.735138-1-zhanxusheng@xiaomi.com/ Reported-by: Zhan Xusheng Suggested-by: Chen Yu Signed-off-by: Tim Chen Signed-off-by: Peter Zijlstra (Intel) Signed-off-by: Ingo Molnar Reviewed-by: Kayra Cizmeci Cc: # v7.2.x Link: https://patch.msgid.link/06af61afedac32e6477f57feb4d658f6c411c3af.1790035273.git.tim.c.chen@linux.intel.com --- mm/page_reporting.c | 429 ++++++++++++++++++++++++++++++++++++++++++++++++++++ 1 file changed, 429 insertions(+) create mode 100644 mm/page_reporting.c (limited to 'mm/page_reporting.c') diff --git a/mm/page_reporting.c b/mm/page_reporting.c new file mode 100644 index 000000000..de587be17 --- /dev/null +++ b/mm/page_reporting.c @@ -0,0 +1,429 @@ +// SPDX-License-Identifier: GPL-2.0 +#include +#include +#include +#include +#include +#include +#include +#include + +#include "page_alloc.h" +#include "page_reporting.h" +#include "internal.h" + +/* Initialize to an unsupported value */ +unsigned int page_reporting_order = PAGE_REPORTING_ORDER_UNSPECIFIED; + +static int page_order_update_notify(const char *val, const struct kernel_param *kp) +{ + /* + * If param is set beyond this limit, order is set to default + * pageblock_order value + */ + return param_set_uint_minmax(val, kp, 0, MAX_PAGE_ORDER); +} + +static const struct kernel_param_ops page_reporting_param_ops = { + .set = &page_order_update_notify, + /* + * For the get op, use param_get_int instead of param_get_uint. + * This is to make sure that when unset the initialized value of + * -1 is shown correctly + */ + .get = ¶m_get_int, +}; + +module_param_cb(page_reporting_order, &page_reporting_param_ops, + &page_reporting_order, 0644); +MODULE_PARM_DESC(page_reporting_order, "Set page reporting order"); + +/* + * This symbol is also a kernel parameter. Export the page_reporting_order + * symbol so that other drivers can access it to control order values without + * having to introduce another configurable parameter. Only one driver can + * register with the page_reporting driver for the service, so we have just + * one control parameter for the use case(which can be accessed in both + * drivers) + */ +EXPORT_SYMBOL_GPL(page_reporting_order); + +static unsigned int page_reporting_delay_ms = 2 * MSEC_PER_SEC; +module_param(page_reporting_delay_ms, uint, 0644); +MODULE_PARM_DESC(page_reporting_delay_ms, + "Set page reporting delay in milliseconds"); + +static struct page_reporting_dev_info __rcu *pr_dev_info __read_mostly; + +enum { + PAGE_REPORTING_IDLE = 0, + PAGE_REPORTING_REQUESTED, + PAGE_REPORTING_ACTIVE +}; + +/* schedule work for page reporting */ +static void page_reporting_schedule_work(struct page_reporting_dev_info *prdev) +{ + queue_delayed_work(system_freezable_wq, &prdev->work, + msecs_to_jiffies(page_reporting_delay_ms)); +} + +/* request page reporting */ +static void +__page_reporting_request(struct page_reporting_dev_info *prdev) +{ + unsigned int state; + + /* Check to see if we are in desired state */ + state = atomic_read(&prdev->state); + if (state == PAGE_REPORTING_REQUESTED) + return; + + /* + * If reporting is already active there is nothing we need to do. + * Test against 0 as that represents PAGE_REPORTING_IDLE. + */ + state = atomic_xchg(&prdev->state, PAGE_REPORTING_REQUESTED); + if (state != PAGE_REPORTING_IDLE) + return; + + /* + * Delay the start of work to allow a sizable queue to build. + * We limit this based on page_reporting_delay_ms. + */ + page_reporting_schedule_work(prdev); +} + +/* notify prdev of free page reporting request */ +void __page_reporting_notify(void) +{ + struct page_reporting_dev_info *prdev; + + /* + * We use RCU to protect the pr_dev_info pointer. In almost all + * cases this should be present, however in the unlikely case of + * a shutdown this will be NULL and we should exit. + */ + rcu_read_lock(); + prdev = rcu_dereference(pr_dev_info); + if (likely(prdev)) + __page_reporting_request(prdev); + + rcu_read_unlock(); +} + +static void +page_reporting_drain(struct page_reporting_dev_info *prdev, + struct scatterlist *sgl, unsigned int nents, bool reported) +{ + struct scatterlist *sg = sgl; + + /* + * Drain the now reported pages back into their respective + * free lists/areas. We assume at least one page is populated. + */ + do { + struct page *page = sg_page(sg); + int mt = get_pageblock_migratetype(page); + unsigned int order = get_order(sg->length); + + __putback_isolated_page(page, order, mt); + + /* If the pages were not reported due to error skip flagging */ + if (!reported) + continue; + + /* + * If page was not commingled with another page we can + * consider the result to be "reported" since the page + * hasn't been modified, otherwise we will need to + * report on the new larger page when we make our way + * up to that higher order. + */ + if (PageBuddy(page) && buddy_order(page) == order) + __SetPageReported(page); + } while ((sg = sg_next(sg))); + + /* reinitialize scatterlist now that it is empty */ + sg_init_table(sgl, nents); +} + +/* + * The page reporting cycle consists of 4 stages, fill, report, drain, and + * idle. We will cycle through the first 3 stages until we cannot obtain a + * full scatterlist of pages, in that case we will switch to idle. + */ +static int +page_reporting_cycle(struct page_reporting_dev_info *prdev, struct zone *zone, + unsigned int order, unsigned int mt, + struct scatterlist *sgl, unsigned int *offset) +{ + struct free_area *area = &zone->free_area[order]; + struct list_head *list = &area->free_list[mt]; + unsigned int page_len = PAGE_SIZE << order; + struct page *page, *next; + long budget; + int err = 0; + + /* + * Perform early check, if free area is empty there is + * nothing to process so we can skip this free_list. + */ + if (list_empty(list)) + return err; + + spin_lock_irq(&zone->lock); + + /* + * Limit how many calls we will be making to the page reporting + * device for this list. By doing this we avoid processing any + * given list for too long. + * + * The current value used allows us enough calls to process over a + * sixteenth of the current list plus one additional call to handle + * any pages that may have already been present from the previous + * list processed. This should result in us reporting all pages on + * an idle system in about 30 seconds. + */ + budget = DIV_ROUND_UP(area->nr_free, prdev->capacity * 16); + + /* loop through free list adding unreported pages to sg list */ + list_for_each_entry_safe(page, next, list, lru) { + /* We are going to skip over the reported pages. */ + if (PageReported(page)) + continue; + + /* + * If we fully consumed our budget then update our + * state to indicate that we are requesting additional + * processing and exit this list. + */ + if (budget < 0) { + atomic_set(&prdev->state, PAGE_REPORTING_REQUESTED); + next = page; + break; + } + + /* Attempt to pull page from list and place in scatterlist */ + if (*offset) { + if (!__isolate_free_page(page, order)) { + next = page; + break; + } + + /* Add page to scatter list */ + --(*offset); + sg_set_page(&sgl[*offset], page, page_len, 0); + + continue; + } + + /* + * Make the first non-reported page in the free list + * the new head of the free list before we release the + * zone lock. + */ + if (!list_is_first(&page->lru, list)) + list_rotate_to_front(&page->lru, list); + + /* release lock before waiting on report processing */ + spin_unlock_irq(&zone->lock); + + /* begin processing pages in local list */ + err = prdev->report(prdev, sgl, prdev->capacity); + + /* reset offset since the full list was reported */ + *offset = prdev->capacity; + + /* update budget to reflect call to report function */ + budget--; + + /* reacquire zone lock and resume processing */ + spin_lock_irq(&zone->lock); + + /* flush reported pages from the sg list */ + page_reporting_drain(prdev, sgl, prdev->capacity, !err); + + /* + * Reset next to first entry, the old next isn't valid + * since we dropped the lock to report the pages + */ + next = list_first_entry(list, struct page, lru); + + /* exit on error */ + if (err) + break; + } + + /* Rotate any leftover pages to the head of the freelist */ + if (!list_entry_is_head(next, list, lru) && !list_is_first(&next->lru, list)) + list_rotate_to_front(&next->lru, list); + + spin_unlock_irq(&zone->lock); + + return err; +} + +static int +page_reporting_process_zone(struct page_reporting_dev_info *prdev, + struct scatterlist *sgl, struct zone *zone) +{ + unsigned int order, mt, leftover, offset = prdev->capacity; + unsigned long watermark; + int err = 0; + + /* Generate minimum watermark to be able to guarantee progress */ + watermark = low_wmark_pages(zone) + + (prdev->capacity << page_reporting_order); + + /* + * Cancel request if insufficient free memory or if we failed + * to allocate page reporting statistics for the zone. + */ + if (!zone_watermark_ok(zone, 0, watermark, 0, ALLOC_CMA)) + return err; + + /* Process each free list starting from lowest order/mt */ + for (order = page_reporting_order; order < NR_PAGE_ORDERS; order++) { + for (mt = 0; mt < MIGRATE_TYPES; mt++) { + /* We do not pull pages from the isolate free list */ + if (is_migrate_isolate(mt)) + continue; + + err = page_reporting_cycle(prdev, zone, order, mt, + sgl, &offset); + if (err) + return err; + } + } + + /* report the leftover pages before going idle */ + leftover = prdev->capacity - offset; + if (leftover) { + sgl = &sgl[offset]; + err = prdev->report(prdev, sgl, leftover); + + /* flush any remaining pages out from the last report */ + spin_lock_irq(&zone->lock); + page_reporting_drain(prdev, sgl, leftover, !err); + spin_unlock_irq(&zone->lock); + } + + return err; +} + +static void page_reporting_process(struct work_struct *work) +{ + struct delayed_work *d_work = to_delayed_work(work); + struct page_reporting_dev_info *prdev = + container_of(d_work, struct page_reporting_dev_info, work); + int err = 0, state = PAGE_REPORTING_ACTIVE; + struct scatterlist *sgl; + struct zone *zone; + + /* + * Change the state to "Active" so that we can track if there is + * anyone requests page reporting after we complete our pass. If + * the state is not altered by the end of the pass we will switch + * to idle and quit scheduling reporting runs. + */ + atomic_set(&prdev->state, state); + + /* allocate scatterlist to store pages being reported on */ + sgl = kmalloc_objs(*sgl, prdev->capacity); + if (!sgl) + goto err_out; + + sg_init_table(sgl, prdev->capacity); + + for_each_zone(zone) { + err = page_reporting_process_zone(prdev, sgl, zone); + if (err) + break; + } + + kfree(sgl); +err_out: + /* + * If the state has reverted back to requested then there may be + * additional pages to be processed. We will defer by + * page_reporting_delay_ms to allow more pages to accumulate. + */ + state = atomic_cmpxchg(&prdev->state, state, PAGE_REPORTING_IDLE); + if (state == PAGE_REPORTING_REQUESTED) + page_reporting_schedule_work(prdev); +} + +static DEFINE_MUTEX(page_reporting_mutex); +DEFINE_STATIC_KEY_FALSE(page_reporting_enabled); + +int page_reporting_register(struct page_reporting_dev_info *prdev) +{ + int err = 0; + + mutex_lock(&page_reporting_mutex); + + /* nothing to do if already in use */ + if (rcu_dereference_protected(pr_dev_info, + lockdep_is_held(&page_reporting_mutex))) { + err = -EBUSY; + goto err_out; + } + + /* + * If the page_reporting_order value is not set, we check if + * an order is provided from the driver that is performing the + * registration. If that is not provided either, we default to + * pageblock_order. + */ + + if (page_reporting_order == PAGE_REPORTING_ORDER_UNSPECIFIED) { + if (prdev->order != PAGE_REPORTING_ORDER_UNSPECIFIED && + prdev->order <= MAX_PAGE_ORDER) + page_reporting_order = prdev->order; + else + page_reporting_order = pageblock_order; + } + + if (!prdev->capacity || prdev->capacity > PAGE_REPORTING_CAPACITY) + prdev->capacity = PAGE_REPORTING_CAPACITY; + + /* initialize state and work structures */ + atomic_set(&prdev->state, PAGE_REPORTING_IDLE); + INIT_DELAYED_WORK(&prdev->work, &page_reporting_process); + + /* Begin initial flush of zones */ + __page_reporting_request(prdev); + + /* Assign device to allow notifications */ + rcu_assign_pointer(pr_dev_info, prdev); + + /* enable page reporting notification */ + if (!static_key_enabled(&page_reporting_enabled)) { + static_branch_enable(&page_reporting_enabled); + pr_info("Free page reporting enabled\n"); + } +err_out: + mutex_unlock(&page_reporting_mutex); + + return err; +} +EXPORT_SYMBOL_GPL(page_reporting_register); + +void page_reporting_unregister(struct page_reporting_dev_info *prdev) +{ + mutex_lock(&page_reporting_mutex); + + if (prdev == rcu_dereference_protected(pr_dev_info, + lockdep_is_held(&page_reporting_mutex))) { + /* Disable page reporting notification */ + RCU_INIT_POINTER(pr_dev_info, NULL); + synchronize_rcu(); + + /* Flush any existing work, and lock it out */ + cancel_delayed_work_sync(&prdev->work); + } + + mutex_unlock(&page_reporting_mutex); +} +EXPORT_SYMBOL_GPL(page_reporting_unregister); -- cgit v1.3.1