sched/cache: Keep nr_pref_llc_running in the runnable domain, to fix LLC mis-scheduling buggrafted

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 <zhanxusheng@xiaomi.com>
Suggested-by: Chen Yu <yu.c.chen@intel.com>
Signed-off-by: Tim Chen <tim.c.chen@linux.intel.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Reviewed-by: Kayra Cizmeci <kayracizmeci@gmail.com>
Cc: <stable@kernel.org> # v7.2.x
Link: https://patch.msgid.link/06af61afedac32e6477f57feb4d658f6c411c3af.1790035273.git.tim.c.chen@linux.intel.com

1 files changed, 431 insertions, 0 deletions

diff --git a/mm/percpu-vm.c b/mm/percpu-vm.c
new file mode 100644
index 000000000..509d89018
--- /dev/null
+++ b/mm/percpu-vm.c
@@ -0,0 +1,431 @@
+// SPDX-License-Identifier: GPL-2.0-only
+/*
+ * mm/percpu-vm.c - vmalloc area based chunk allocation
+ *
+ * Copyright (C) 2010		SUSE Linux Products GmbH
+ * Copyright (C) 2010		Tejun Heo <tj@kernel.org>
+ *
+ * Chunks are mapped into vmalloc areas and populated page by page.
+ * This is the default chunk allocator.
+ */
+#include "internal.h"
+#include "vmalloc.h"
+
+static struct page *pcpu_chunk_page(struct pcpu_chunk *chunk,
+				    unsigned int cpu, int page_idx)
+{
+	/* must not be used on pre-mapped chunk */
+	WARN_ON(chunk->immutable);
+
+	return vmalloc_to_page((void *)pcpu_chunk_addr(chunk, cpu, page_idx));
+}
+
+/**
+ * pcpu_get_pages - get temp pages array
+ * @gfp: allocation flags passed to the underlying allocator, 0 to only
+ *	 return the cached array
+ *
+ * Returns pointer to array of pointers to struct page which can be indexed
+ * with pcpu_page_idx().  Note that there is only one array and accesses
+ * should be serialized by pcpu_alloc_mutex.
+ *
+ * RETURNS:
+ * Pointer to temp pages array on success.
+ */
+static struct page **pcpu_get_pages(gfp_t gfp)
+{
+	static struct page **pages;
+	size_t pages_size = pcpu_nr_units * pcpu_unit_pages * sizeof(pages[0]);
+
+	lockdep_assert_held(&pcpu_alloc_mutex);
+
+	if (!pages && gfp)
+		pages = pcpu_mem_zalloc(pages_size, gfp);
+	return pages;
+}
+
+static struct page **pcpu_get_pages_cached(void)
+{
+	return pcpu_get_pages(0);
+}
+
+/**
+ * pcpu_free_pages - free pages which were allocated for @chunk
+ * @chunk: chunk pages were allocated for
+ * @pages: array of pages to be freed, indexed by pcpu_page_idx()
+ * @page_start: page index of the first page to be freed
+ * @page_end: page index of the last page to be freed + 1
+ *
+ * Free pages [@page_start and @page_end) in @pages for all units.
+ * The pages were allocated for @chunk.
+ */
+static void pcpu_free_pages(struct pcpu_chunk *chunk,
+			    struct page **pages, int page_start, int page_end)
+{
+	unsigned int cpu;
+	int i;
+
+	for_each_possible_cpu(cpu) {
+		for (i = page_start; i < page_end; i++) {
+			struct page *page = pages[pcpu_page_idx(cpu, i)];
+
+			if (page)
+				__free_page(page);
+		}
+	}
+}
+
+/**
+ * pcpu_alloc_pages - allocates pages for @chunk
+ * @chunk: target chunk
+ * @pages: array to put the allocated pages into, indexed by pcpu_page_idx()
+ * @page_start: page index of the first page to be allocated
+ * @page_end: page index of the last page to be allocated + 1
+ * @gfp: allocation flags passed to the underlying allocator
+ *
+ * Allocate pages [@page_start,@page_end) into @pages for all units.
+ * The allocation is for @chunk.  Percpu core doesn't care about the
+ * content of @pages and will pass it verbatim to pcpu_map_pages().
+ */
+static int pcpu_alloc_pages(struct pcpu_chunk *chunk,
+			    struct page **pages, int page_start, int page_end,
+			    gfp_t gfp)
+{
+	unsigned int cpu, tcpu;
+	int i;
+
+	gfp |= __GFP_HIGHMEM;
+
+	for_each_possible_cpu(cpu) {
+		for (i = page_start; i < page_end; i++) {
+			struct page **pagep = &pages[pcpu_page_idx(cpu, i)];
+
+			*pagep = alloc_pages_node(cpu_to_node(cpu), gfp, 0);
+			if (!*pagep)
+				goto err;
+		}
+	}
+	return 0;
+
+err:
+	while (--i >= page_start)
+		__free_page(pages[pcpu_page_idx(cpu, i)]);
+
+	for_each_possible_cpu(tcpu) {
+		if (tcpu == cpu)
+			break;
+		for (i = page_start; i < page_end; i++)
+			__free_page(pages[pcpu_page_idx(tcpu, i)]);
+	}
+	return -ENOMEM;
+}
+
+/**
+ * pcpu_pre_unmap_flush - flush cache prior to unmapping
+ * @chunk: chunk the regions to be flushed belongs to
+ * @page_start: page index of the first page to be flushed
+ * @page_end: page index of the last page to be flushed + 1
+ *
+ * Pages in [@page_start,@page_end) of @chunk are about to be
+ * unmapped.  Flush cache.  As each flushing trial can be very
+ * expensive, issue flush on the whole region at once rather than
+ * doing it for each cpu.  This could be an overkill but is more
+ * scalable.
+ */
+static void pcpu_pre_unmap_flush(struct pcpu_chunk *chunk,
+				 int page_start, int page_end)
+{
+	flush_cache_vunmap(
+		pcpu_chunk_addr(chunk, pcpu_low_unit_cpu, page_start),
+		pcpu_chunk_addr(chunk, pcpu_high_unit_cpu, page_end));
+}
+
+static void __pcpu_unmap_pages(unsigned long addr, int nr_pages)
+{
+	vunmap_range_noflush(addr, addr + (nr_pages << PAGE_SHIFT));
+}
+
+/**
+ * pcpu_unmap_pages - unmap pages out of a pcpu_chunk
+ * @chunk: chunk of interest
+ * @pages: pages array which can be used to pass information to free
+ * @page_start: page index of the first page to unmap
+ * @page_end: page index of the last page to unmap + 1
+ *
+ * For each cpu, unmap pages [@page_start,@page_end) out of @chunk.
+ * Corresponding elements in @pages were cleared by the caller and can
+ * be used to carry information to pcpu_free_pages() which will be
+ * called after all unmaps are finished.  The caller should call
+ * proper pre/post flush functions.
+ */
+static void pcpu_unmap_pages(struct pcpu_chunk *chunk,
+			     struct page **pages, int page_start, int page_end)
+{
+	unsigned int cpu;
+	int i;
+
+	for_each_possible_cpu(cpu) {
+		for (i = page_start; i < page_end; i++) {
+			struct page *page;
+
+			page = pcpu_chunk_page(chunk, cpu, i);
+			WARN_ON(!page);
+			pages[pcpu_page_idx(cpu, i)] = page;
+		}
+		__pcpu_unmap_pages(pcpu_chunk_addr(chunk, cpu, page_start),
+				   page_end - page_start);
+	}
+}
+
+/**
+ * pcpu_post_unmap_tlb_flush - flush TLB after unmapping
+ * @chunk: pcpu_chunk the regions to be flushed belong to
+ * @page_start: page index of the first page to be flushed
+ * @page_end: page index of the last page to be flushed + 1
+ *
+ * Pages [@page_start,@page_end) of @chunk have been unmapped.  Flush
+ * TLB for the regions.  This can be skipped if the area is to be
+ * returned to vmalloc as vmalloc will handle TLB flushing lazily.
+ *
+ * As with pcpu_pre_unmap_flush(), TLB flushing also is done at once
+ * for the whole region.
+ */
+static void pcpu_post_unmap_tlb_flush(struct pcpu_chunk *chunk,
+				      int page_start, int page_end)
+{
+	flush_tlb_kernel_range(
+		pcpu_chunk_addr(chunk, pcpu_low_unit_cpu, page_start),
+		pcpu_chunk_addr(chunk, pcpu_high_unit_cpu, page_end));
+}
+
+static int __pcpu_map_pages(unsigned long addr, struct page **pages,
+			    int nr_pages, gfp_t gfp)
+{
+	unsigned int flags;
+	int ret;
+
+	/*
+	 * The vmalloc page table allocation path does not pass @gfp down
+	 * explicitly.  Apply the corresponding memalloc scope so implicit
+	 * page table allocations preserve NOFS/NOIO constraints.
+	 */
+	flags = memalloc_apply_gfp_scope(gfp);
+	ret = vmap_pages_range_noflush(addr, addr + (nr_pages << PAGE_SHIFT),
+				       PAGE_KERNEL, pages, PAGE_SHIFT, gfp);
+	memalloc_restore_scope(flags);
+
+	return ret;
+}
+
+/**
+ * pcpu_map_pages - map pages into a pcpu_chunk
+ * @chunk: chunk of interest
+ * @pages: pages array containing pages to be mapped
+ * @page_start: page index of the first page to map
+ * @page_end: page index of the last page to map + 1
+ * @gfp: allocation flags passed to the underlying allocator
+ *
+ * For each cpu, map pages [@page_start,@page_end) into @chunk.  The
+ * caller is responsible for calling pcpu_post_map_flush() after all
+ * mappings are complete.
+ *
+ * This function is responsible for setting up whatever is necessary for
+ * reverse lookup (addr -> chunk).
+ */
+static int pcpu_map_pages(struct pcpu_chunk *chunk, struct page **pages,
+			  int page_start, int page_end, gfp_t gfp)
+{
+	unsigned int cpu, tcpu;
+	int i, err;
+
+	for_each_possible_cpu(cpu) {
+		err = __pcpu_map_pages(pcpu_chunk_addr(chunk, cpu, page_start),
+				       &pages[pcpu_page_idx(cpu, page_start)],
+				       page_end - page_start, gfp);
+		if (err < 0)
+			goto err;
+
+		for (i = page_start; i < page_end; i++)
+			pcpu_set_page_chunk(pages[pcpu_page_idx(cpu, i)],
+					    chunk);
+	}
+	return 0;
+err:
+	for_each_possible_cpu(tcpu) {
+		__pcpu_unmap_pages(pcpu_chunk_addr(chunk, tcpu, page_start),
+				   page_end - page_start);
+		if (tcpu == cpu)
+			break;
+	}
+	pcpu_post_unmap_tlb_flush(chunk, page_start, page_end);
+	return err;
+}
+
+/**
+ * pcpu_post_map_flush - flush cache after mapping
+ * @chunk: pcpu_chunk the regions to be flushed belong to
+ * @page_start: page index of the first page to be flushed
+ * @page_end: page index of the last page to be flushed + 1
+ *
+ * Pages [@page_start,@page_end) of @chunk have been mapped.  Flush
+ * cache.
+ *
+ * As with pcpu_pre_unmap_flush(), TLB flushing also is done at once
+ * for the whole region.
+ */
+static void pcpu_post_map_flush(struct pcpu_chunk *chunk,
+				int page_start, int page_end)
+{
+	flush_cache_vmap(
+		pcpu_chunk_addr(chunk, pcpu_low_unit_cpu, page_start),
+		pcpu_chunk_addr(chunk, pcpu_high_unit_cpu, page_end));
+}
+
+/**
+ * pcpu_populate_chunk - populate and map an area of a pcpu_chunk
+ * @chunk: chunk of interest
+ * @page_start: the start page
+ * @page_end: the end page
+ * @gfp: allocation flags passed to the underlying memory allocator
+ *
+ * For each cpu, populate and map pages [@page_start,@page_end) into
+ * @chunk.
+ *
+ * CONTEXT:
+ * pcpu_alloc_mutex, does @gfp allocation.
+ */
+static int pcpu_populate_chunk(struct pcpu_chunk *chunk,
+			       int page_start, int page_end, gfp_t gfp)
+{
+	struct page **pages;
+
+	pages = pcpu_get_pages(gfp);
+	if (!pages)
+		return -ENOMEM;
+
+	if (pcpu_alloc_pages(chunk, pages, page_start, page_end, gfp))
+		return -ENOMEM;
+
+	if (pcpu_map_pages(chunk, pages, page_start, page_end, gfp)) {
+		pcpu_free_pages(chunk, pages, page_start, page_end);
+		return -ENOMEM;
+	}
+	pcpu_post_map_flush(chunk, page_start, page_end);
+
+	return 0;
+}
+
+/**
+ * pcpu_depopulate_chunk - depopulate and unmap an area of a pcpu_chunk
+ * @chunk: chunk to depopulate
+ * @page_start: the start page
+ * @page_end: the end page
+ *
+ * For each cpu, depopulate and unmap pages [@page_start,@page_end)
+ * from @chunk.
+ *
+ * Caller is required to call pcpu_post_unmap_tlb_flush() if not returning the
+ * region back to vmalloc() which will lazily flush the tlb.
+ *
+ * CONTEXT:
+ * pcpu_alloc_mutex.
+ */
+static void pcpu_depopulate_chunk(struct pcpu_chunk *chunk,
+				  int page_start, int page_end)
+{
+	struct page **pages;
+
+	/*
+	 * If control reaches here, there must have been at least one
+	 * successful population attempt so the temp pages array must
+	 * be available now.
+	 */
+	pages = pcpu_get_pages_cached();
+	BUG_ON(!pages);
+
+	/* unmap and free */
+	pcpu_pre_unmap_flush(chunk, page_start, page_end);
+
+	pcpu_unmap_pages(chunk, pages, page_start, page_end);
+
+	pcpu_free_pages(chunk, pages, page_start, page_end);
+}
+
+static struct pcpu_chunk *pcpu_create_chunk(gfp_t gfp)
+{
+	struct pcpu_chunk *chunk;
+	struct vm_struct **vms;
+
+	chunk = pcpu_alloc_chunk(gfp);
+	if (!chunk)
+		return NULL;
+
+	vms = pcpu_get_vm_areas(pcpu_group_offsets, pcpu_group_sizes,
+				pcpu_nr_groups, pcpu_atom_size, gfp);
+	if (!vms) {
+		pcpu_free_chunk(chunk);
+		return NULL;
+	}
+
+	chunk->data = vms;
+	chunk->base_addr = vms[0]->addr - pcpu_group_offsets[0];
+
+	pcpu_stats_chunk_alloc();
+	trace_percpu_create_chunk(chunk->base_addr);
+
+	return chunk;
+}
+
+static void pcpu_destroy_chunk(struct pcpu_chunk *chunk)
+{
+	if (!chunk)
+		return;
+
+	pcpu_stats_chunk_dealloc();
+	trace_percpu_destroy_chunk(chunk->base_addr);
+
+	if (chunk->data)
+		pcpu_free_vm_areas(chunk->data, pcpu_nr_groups);
+	pcpu_free_chunk(chunk);
+}
+
+static struct page *pcpu_addr_to_page(void *addr)
+{
+	return vmalloc_to_page(addr);
+}
+
+static int __init pcpu_verify_alloc_info(const struct pcpu_alloc_info *ai)
+{
+	/* no extra restriction */
+	return 0;
+}
+
+/**
+ * pcpu_should_reclaim_chunk - determine if a chunk should go into reclaim
+ * @chunk: chunk of interest
+ *
+ * This is the entry point for percpu reclaim.  If a chunk qualifies, it is then
+ * isolated and managed in separate lists at the back of pcpu_slot: sidelined
+ * and to_depopulate respectively.  The to_depopulate list holds chunks slated
+ * for depopulation.  They no longer contribute to pcpu_nr_empty_pop_pages once
+ * they are on this list.  Once depopulated, they are moved onto the sidelined
+ * list which enables them to be pulled back in for allocation if no other chunk
+ * can suffice the allocation.
+ */
+static bool pcpu_should_reclaim_chunk(struct pcpu_chunk *chunk)
+{
+	/* do not reclaim either the first chunk or reserved chunk */
+	if (chunk == pcpu_first_chunk || chunk == pcpu_reserved_chunk)
+		return false;
+
+	/*
+	 * If it is isolated, it may be on the sidelined list so move it back to
+	 * the to_depopulate list.  If we hit at least 1/4 pages empty pages AND
+	 * there is no system-wide shortage of empty pages aside from this
+	 * chunk, move it to the to_depopulate list.
+	 */
+	return ((chunk->isolated && chunk->nr_empty_pop_pages) ||
+		(pcpu_nr_empty_pop_pages >
+		 (PCPU_EMPTY_POP_PAGES_HIGH + chunk->nr_empty_pop_pages) &&
+		 chunk->nr_empty_pop_pages >= chunk->nr_pages / 4));
+}