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, 221 insertions, 0 deletions

diff --git a/arch/powerpc/mm/cacheflush.c b/arch/powerpc/mm/cacheflush.c
new file mode 100644
index 000000000..7186516ec
--- /dev/null
+++ b/arch/powerpc/mm/cacheflush.c
@@ -0,0 +1,221 @@
+// SPDX-License-Identifier: GPL-2.0-or-later
+
+#include <linux/highmem.h>
+#include <linux/kprobes.h>
+
+/**
+ * flush_coherent_icache() - if a CPU has a coherent icache, flush it
+ * Return true if the cache was flushed, false otherwise
+ */
+static inline bool flush_coherent_icache(void)
+{
+	/*
+	 * For a snooping icache, we still need a dummy icbi to purge all the
+	 * prefetched instructions from the ifetch buffers. We also need a sync
+	 * before the icbi to order the actual stores to memory that might
+	 * have modified instructions with the icbi.
+	 */
+	if (cpu_has_feature(CPU_FTR_COHERENT_ICACHE)) {
+		mb(); /* sync */
+		icbi((void *)PAGE_OFFSET);
+		mb(); /* sync */
+		isync();
+		return true;
+	}
+
+	return false;
+}
+
+/**
+ * invalidate_icache_range() - Flush the icache by issuing icbi across an address range
+ * @start: the start address
+ * @stop: the stop address (exclusive)
+ */
+static void invalidate_icache_range(unsigned long start, unsigned long stop)
+{
+	unsigned long shift = l1_icache_shift();
+	unsigned long bytes = l1_icache_bytes();
+	char *addr = (char *)(start & ~(bytes - 1));
+	unsigned long size = stop - (unsigned long)addr + (bytes - 1);
+	unsigned long i;
+
+	for (i = 0; i < size >> shift; i++, addr += bytes)
+		icbi(addr);
+
+	mb(); /* sync */
+	isync();
+}
+
+/**
+ * flush_icache_range: Write any modified data cache blocks out to memory
+ * and invalidate the corresponding blocks in the instruction cache
+ *
+ * Generic code will call this after writing memory, before executing from it.
+ *
+ * @start: the start address
+ * @stop: the stop address (exclusive)
+ */
+void flush_icache_range(unsigned long start, unsigned long stop)
+{
+	if (flush_coherent_icache())
+		return;
+
+	clean_dcache_range(start, stop);
+
+	if (IS_ENABLED(CONFIG_44x)) {
+		/*
+		 * Flash invalidate on 44x because we are passed kmapped
+		 * addresses and this doesn't work for userspace pages due to
+		 * the virtually tagged icache.
+		 */
+		iccci((void *)start);
+		mb(); /* sync */
+		isync();
+	} else
+		invalidate_icache_range(start, stop);
+}
+EXPORT_SYMBOL(flush_icache_range);
+
+#ifdef CONFIG_HIGHMEM
+/**
+ * flush_dcache_icache_phys() - Flush a page by its physical address
+ * @physaddr: the physical address of the page
+ */
+static void flush_dcache_icache_phys(unsigned long physaddr)
+{
+	unsigned long bytes = l1_dcache_bytes();
+	unsigned long nb = PAGE_SIZE / bytes;
+	unsigned long addr = physaddr & PAGE_MASK;
+	unsigned long msr, msr0;
+	unsigned long loop1 = addr, loop2 = addr;
+
+	msr0 = mfmsr();
+	msr = msr0 & ~MSR_DR;
+	/*
+	 * This must remain as ASM to prevent potential memory accesses
+	 * while the data MMU is disabled
+	 */
+	asm volatile(
+		"   mtctr %2;\n"
+		"   mtmsr %3;\n"
+		"   isync;\n"
+		"0: dcbst   0, %0;\n"
+		"   addi    %0, %0, %4;\n"
+		"   bdnz    0b;\n"
+		"   sync;\n"
+		"   mtctr %2;\n"
+		"1: icbi    0, %1;\n"
+		"   addi    %1, %1, %4;\n"
+		"   bdnz    1b;\n"
+		"   sync;\n"
+		"   mtmsr %5;\n"
+		"   isync;\n"
+		: "+&r" (loop1), "+&r" (loop2)
+		: "r" (nb), "r" (msr), "i" (bytes), "r" (msr0)
+		: "ctr", "memory");
+}
+NOKPROBE_SYMBOL(flush_dcache_icache_phys)
+#else
+static void flush_dcache_icache_phys(unsigned long physaddr)
+{
+}
+#endif
+
+/**
+ * __flush_dcache_icache(): Flush a particular page from the data cache to RAM.
+ * Note: this is necessary because the instruction cache does *not*
+ * snoop from the data cache.
+ *
+ * @p: the address of the page to flush
+ */
+static void __flush_dcache_icache(void *p)
+{
+	unsigned long addr = (unsigned long)p & PAGE_MASK;
+
+	clean_dcache_range(addr, addr + PAGE_SIZE);
+
+	/*
+	 * We don't flush the icache on 44x. Those have a virtual icache and we
+	 * don't have access to the virtual address here (it's not the page
+	 * vaddr but where it's mapped in user space). The flushing of the
+	 * icache on these is handled elsewhere, when a change in the address
+	 * space occurs, before returning to user space.
+	 */
+
+	if (mmu_has_feature(MMU_FTR_TYPE_44x))
+		return;
+
+	invalidate_icache_range(addr, addr + PAGE_SIZE);
+}
+
+void flush_dcache_icache_folio(struct folio *folio)
+{
+	unsigned int i, nr = folio_nr_pages(folio);
+
+	if (flush_coherent_icache())
+		return;
+
+	if (!folio_test_highmem(folio)) {
+		void *addr = folio_address(folio);
+		for (i = 0; i < nr; i++)
+			__flush_dcache_icache(addr + i * PAGE_SIZE);
+	} else if (IS_ENABLED(CONFIG_BOOKE) || sizeof(phys_addr_t) > sizeof(void *)) {
+		for (i = 0; i < nr; i++) {
+			void *start = kmap_local_folio(folio, i * PAGE_SIZE);
+
+			__flush_dcache_icache(start);
+			kunmap_local(start);
+		}
+	} else {
+		unsigned long pfn = folio_pfn(folio);
+		for (i = 0; i < nr; i++)
+			flush_dcache_icache_phys((pfn + i) * PAGE_SIZE);
+	}
+}
+EXPORT_SYMBOL(flush_dcache_icache_folio);
+
+void clear_user_page(void *page, unsigned long vaddr, struct page *pg)
+{
+	clear_page(page);
+
+	/*
+	 * We shouldn't have to do this, but some versions of glibc
+	 * require it (ld.so assumes zero filled pages are icache clean)
+	 * - Anton
+	 */
+	flush_dcache_page(pg);
+}
+EXPORT_SYMBOL(clear_user_page);
+
+void copy_user_page(void *vto, void *vfrom, unsigned long vaddr,
+		    struct page *pg)
+{
+	copy_page(vto, vfrom);
+
+	/*
+	 * We should be able to use the following optimisation, however
+	 * there are two problems.
+	 * Firstly a bug in some versions of binutils meant PLT sections
+	 * were not marked executable.
+	 * Secondly the first word in the GOT section is blrl, used
+	 * to establish the GOT address. Until recently the GOT was
+	 * not marked executable.
+	 * - Anton
+	 */
+#if 0
+	if (!vma->vm_file && ((vma->vm_flags & VM_EXEC) == 0))
+		return;
+#endif
+
+	flush_dcache_page(pg);
+}
+
+void flush_icache_user_page(struct vm_area_struct *vma, struct page *page,
+			     unsigned long addr, int len)
+{
+	void *maddr;
+
+	maddr = kmap_local_page(page) + (addr & ~PAGE_MASK);
+	flush_icache_range((unsigned long)maddr, (unsigned long)maddr + len);
+	kunmap_local(maddr);
+}