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/char/uv_mmtimer.c | 220 ++++++++++++++++++++++++++++++++++++++++++++++ 1 file changed, 220 insertions(+) create mode 100644 drivers/char/uv_mmtimer.c (limited to 'drivers/char/uv_mmtimer.c') diff --git a/drivers/char/uv_mmtimer.c b/drivers/char/uv_mmtimer.c new file mode 100644 index 000000000..956ebe208 --- /dev/null +++ b/drivers/char/uv_mmtimer.c @@ -0,0 +1,220 @@ +/* + * Timer device implementation for SGI UV platform. + * + * This file is subject to the terms and conditions of the GNU General Public + * License. See the file "COPYING" in the main directory of this archive + * for more details. + * + * Copyright (c) 2009 Silicon Graphics, Inc. All rights reserved. + * + */ + +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include + +#include +#include +#include +#include + +MODULE_AUTHOR("Dimitri Sivanich "); +MODULE_DESCRIPTION("SGI UV Memory Mapped RTC Timer"); +MODULE_LICENSE("GPL"); + +/* name of the device, usually in /dev */ +#define UV_MMTIMER_NAME "mmtimer" +#define UV_MMTIMER_DESC "SGI UV Memory Mapped RTC Timer" +#define UV_MMTIMER_VERSION "1.0" + +static long uv_mmtimer_ioctl(struct file *file, unsigned int cmd, + unsigned long arg); +static int uv_mmtimer_mmap(struct file *file, struct vm_area_struct *vma); + +/* + * Period in femtoseconds (10^-15 s) + */ +static unsigned long uv_mmtimer_femtoperiod; + +static const struct file_operations uv_mmtimer_fops = { + .owner = THIS_MODULE, + .mmap = uv_mmtimer_mmap, + .unlocked_ioctl = uv_mmtimer_ioctl, + .llseek = noop_llseek, +}; + +/** + * uv_mmtimer_ioctl - ioctl interface for /dev/uv_mmtimer + * @file: file structure for the device + * @cmd: command to execute + * @arg: optional argument to command + * + * Executes the command specified by @cmd. Returns 0 for success, < 0 for + * failure. + * + * Valid commands: + * + * %MMTIMER_GETOFFSET - Should return the offset (relative to the start + * of the page where the registers are mapped) for the counter in question. + * + * %MMTIMER_GETRES - Returns the resolution of the clock in femto (10^-15) + * seconds + * + * %MMTIMER_GETFREQ - Copies the frequency of the clock in Hz to the address + * specified by @arg + * + * %MMTIMER_GETBITS - Returns the number of bits in the clock's counter + * + * %MMTIMER_MMAPAVAIL - Returns 1 if registers can be mmap'd into userspace + * + * %MMTIMER_GETCOUNTER - Gets the current value in the counter and places it + * in the address specified by @arg. + */ +static long uv_mmtimer_ioctl(struct file *file, unsigned int cmd, + unsigned long arg) +{ + int ret = 0; + + switch (cmd) { + case MMTIMER_GETOFFSET: /* offset of the counter */ + /* + * Starting with HUB rev 2.0, the UV RTC register is + * replicated across all cachelines of it's own page. + * This allows faster simultaneous reads from a given socket. + * + * The offset returned is in 64 bit units. + */ + if (uv_get_min_hub_revision_id() == 1) + ret = 0; + else + ret = ((uv_blade_processor_id() * L1_CACHE_BYTES) % + PAGE_SIZE) / 8; + break; + + case MMTIMER_GETRES: /* resolution of the clock in 10^-15 s */ + if (copy_to_user((unsigned long __user *)arg, + &uv_mmtimer_femtoperiod, sizeof(unsigned long))) + ret = -EFAULT; + break; + + case MMTIMER_GETFREQ: /* frequency in Hz */ + if (copy_to_user((unsigned long __user *)arg, + &sn_rtc_cycles_per_second, + sizeof(unsigned long))) + ret = -EFAULT; + break; + + case MMTIMER_GETBITS: /* number of bits in the clock */ + ret = hweight64(UVH_RTC_REAL_TIME_CLOCK_MASK); + break; + + case MMTIMER_MMAPAVAIL: + ret = 1; + break; + + case MMTIMER_GETCOUNTER: + if (copy_to_user((unsigned long __user *)arg, + (unsigned long *)uv_local_mmr_address(UVH_RTC), + sizeof(unsigned long))) + ret = -EFAULT; + break; + default: + ret = -ENOTTY; + break; + } + return ret; +} + +/** + * uv_mmtimer_mmap - maps the clock's registers into userspace + * @file: file structure for the device + * @vma: VMA to map the registers into + * + * Calls remap_pfn_range() to map the clock's registers into + * the calling process' address space. + */ +static int uv_mmtimer_mmap(struct file *file, struct vm_area_struct *vma) +{ + unsigned long uv_mmtimer_addr; + + if (vma->vm_end - vma->vm_start != PAGE_SIZE) + return -EINVAL; + + if (vma->vm_flags & VM_WRITE) + return -EPERM; + + if (PAGE_SIZE > (1 << 16)) + return -ENOSYS; + + vma->vm_page_prot = pgprot_noncached(vma->vm_page_prot); + + uv_mmtimer_addr = UV_LOCAL_MMR_BASE | UVH_RTC; + uv_mmtimer_addr &= ~(PAGE_SIZE - 1); + uv_mmtimer_addr &= 0xfffffffffffffffUL; + + if (remap_pfn_range(vma, vma->vm_start, uv_mmtimer_addr >> PAGE_SHIFT, + PAGE_SIZE, vma->vm_page_prot)) { + printk(KERN_ERR "remap_pfn_range failed in uv_mmtimer_mmap\n"); + return -EAGAIN; + } + + return 0; +} + +static struct miscdevice uv_mmtimer_miscdev = { + MISC_DYNAMIC_MINOR, + UV_MMTIMER_NAME, + &uv_mmtimer_fops +}; + + +/** + * uv_mmtimer_init - device initialization routine + * + * Does initial setup for the uv_mmtimer device. + */ +static int __init uv_mmtimer_init(void) +{ + if (!is_uv_system()) { + printk(KERN_ERR "%s: Hardware unsupported\n", UV_MMTIMER_NAME); + return -1; + } + + /* + * Sanity check the cycles/sec variable + */ + if (sn_rtc_cycles_per_second < 100000) { + printk(KERN_ERR "%s: unable to determine clock frequency\n", + UV_MMTIMER_NAME); + return -1; + } + + uv_mmtimer_femtoperiod = ((unsigned long)1E15 + + sn_rtc_cycles_per_second / 2) / + sn_rtc_cycles_per_second; + + if (misc_register(&uv_mmtimer_miscdev)) { + printk(KERN_ERR "%s: failed to register device\n", + UV_MMTIMER_NAME); + return -1; + } + + printk(KERN_INFO "%s: v%s, %ld MHz\n", UV_MMTIMER_DESC, + UV_MMTIMER_VERSION, + sn_rtc_cycles_per_second/(unsigned long)1E6); + + return 0; +} + +module_init(uv_mmtimer_init); -- cgit v1.3.1