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 --- drivers/firmware/efi/memmap.c | 148 ++++++++++++++++++++++++++++++++++++++++++ 1 file changed, 148 insertions(+) create mode 100644 drivers/firmware/efi/memmap.c (limited to 'drivers/firmware/efi/memmap.c') diff --git a/drivers/firmware/efi/memmap.c b/drivers/firmware/efi/memmap.c new file mode 100644 index 000000000..f1c04d7cf --- /dev/null +++ b/drivers/firmware/efi/memmap.c @@ -0,0 +1,148 @@ +// SPDX-License-Identifier: GPL-2.0 +/* + * Common EFI memory map functions. + */ + +#define pr_fmt(fmt) "efi: " fmt + +#include +#include +#include +#include +#include +#include + +#include +#include + +/** + * __efi_memmap_init - Common code for mapping the EFI memory map + * @data: EFI memory map data + * + * This function takes care of figuring out which function to use to + * map the EFI memory map in efi.memmap based on how far into the boot + * we are. + * + * During bootup EFI_MEMMAP_LATE in data->flags should be clear since we + * only have access to the early_memremap*() functions as the vmalloc + * space isn't setup. Once the kernel is fully booted we can fallback + * to the more robust memremap*() API. + * + * Returns: zero on success, a negative error code on failure. + */ +int __init __efi_memmap_init(struct efi_memory_map_data *data) +{ + struct efi_memory_map map; + phys_addr_t phys_map; + + phys_map = data->phys_map; + + if (data->flags & EFI_MEMMAP_LATE) + map.map = memremap(phys_map, data->size, MEMREMAP_WB); + else + map.map = early_memremap(phys_map, data->size); + + if (!map.map) { + pr_err("Could not map the memory map! phys_map=%pa, size=0x%lx\n", + &phys_map, data->size); + return -ENOMEM; + } + + map.phys_map = data->phys_map; + map.nr_map = data->size / data->desc_size; + map.map_end = map.map + data->size; + + map.desc_version = data->desc_version; + map.desc_size = data->desc_size; + map.flags = data->flags; + + set_bit(EFI_MEMMAP, &efi.flags); + + efi.memmap = map; + + return 0; +} + +/** + * efi_memmap_init_early - Map the EFI memory map data structure + * @data: EFI memory map data + * + * Use early_memremap() to map the passed in EFI memory map and assign + * it to efi.memmap. + * + * Returns: zero on success, a negative error code on failure. + */ +int __init efi_memmap_init_early(struct efi_memory_map_data *data) +{ + /* Cannot go backwards */ + WARN_ON(efi.memmap.flags & EFI_MEMMAP_LATE); + + data->flags = 0; + return __efi_memmap_init(data); +} + +void __init efi_memmap_unmap(void) +{ + if (!efi_enabled(EFI_MEMMAP)) + return; + + if (!(efi.memmap.flags & EFI_MEMMAP_LATE)) { + unsigned long size; + + size = efi.memmap.desc_size * efi.memmap.nr_map; + early_memunmap(efi.memmap.map, size); + } else { + memunmap(efi.memmap.map); + } + + efi.memmap.map = NULL; + clear_bit(EFI_MEMMAP, &efi.flags); +} + +/** + * efi_memmap_init_late - Map efi.memmap with memremap() + * @addr: Physical address of the new EFI memory map + * @size: Size in bytes of the new EFI memory map + * + * Setup a mapping of the EFI memory map using ioremap_cache(). This + * function should only be called once the vmalloc space has been + * setup and is therefore not suitable for calling during early EFI + * initialise, e.g. in efi_init(). Additionally, it expects + * efi_memmap_init_early() to have already been called. + * + * The reason there are two EFI memmap initialisation + * (efi_memmap_init_early() and this late version) is because the + * early EFI memmap should be explicitly unmapped once EFI + * initialisation is complete as the fixmap space used to map the EFI + * memmap (via early_memremap()) is a scarce resource. + * + * This late mapping is intended to persist for the duration of + * runtime so that things like efi_mem_desc_lookup() and + * efi_mem_attributes() always work. + * + * Returns: zero on success, a negative error code on failure. + */ +int __init efi_memmap_init_late(phys_addr_t addr, unsigned long size) +{ + struct efi_memory_map_data data = { + .phys_map = addr, + .size = size, + .flags = EFI_MEMMAP_LATE, + }; + + /* Did we forget to unmap the early EFI memmap? */ + WARN_ON(efi.memmap.map); + + /* Were we already called? */ + WARN_ON(efi.memmap.flags & EFI_MEMMAP_LATE); + + /* + * It makes no sense to allow callers to register different + * values for the following fields. Copy them out of the + * existing early EFI memmap. + */ + data.desc_version = efi.memmap.desc_version; + data.desc_size = efi.memmap.desc_size; + + return __efi_memmap_init(&data); +} -- cgit v1.3.1