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/net/ipa/reg.h | 136 ++++++++++++++++++++++++++++++++++++++++++++++++++ 1 file changed, 136 insertions(+) create mode 100644 drivers/net/ipa/reg.h (limited to 'drivers/net/ipa/reg.h') diff --git a/drivers/net/ipa/reg.h b/drivers/net/ipa/reg.h new file mode 100644 index 000000000..53c16e594 --- /dev/null +++ b/drivers/net/ipa/reg.h @@ -0,0 +1,136 @@ +/* SPDX-License-Identifier: GPL-2.0 */ + +/* Copyright (C) 2022-2024 Linaro Ltd. */ + +#ifndef _REG_H_ +#define _REG_H_ + +#include +#include +#include +#include +#include + +/** + * struct reg - A register descriptor + * @offset: Register offset relative to base of register memory + * @stride: Distance between two instances, if parameterized + * @fcount: Number of entries in the @fmask array + * @fmask: Array of mask values defining position and width of fields + * @name: Upper-case name of the register + */ +struct reg { + u32 offset; + u32 stride; + u32 fcount; + const u32 *fmask; /* BIT(nr) or GENMASK(h, l) */ + const char *name; +}; + +/* Helper macro for defining "simple" (non-parameterized) registers */ +#define REG(__NAME, __reg_id, __offset) \ + REG_STRIDE(__NAME, __reg_id, __offset, 0) + +/* Helper macro for defining parameterized registers, specifying stride */ +#define REG_STRIDE(__NAME, __reg_id, __offset, __stride) \ + static const struct reg reg_ ## __reg_id = { \ + .name = #__NAME, \ + .offset = __offset, \ + .stride = __stride, \ + } + +#define REG_FIELDS(__NAME, __name, __offset) \ + REG_STRIDE_FIELDS(__NAME, __name, __offset, 0) + +#define REG_STRIDE_FIELDS(__NAME, __name, __offset, __stride) \ + static const struct reg reg_ ## __name = { \ + .name = #__NAME, \ + .offset = __offset, \ + .stride = __stride, \ + .fcount = ARRAY_SIZE(reg_ ## __name ## _fmask), \ + .fmask = reg_ ## __name ## _fmask, \ + } + +/** + * struct regs - Description of registers supported by hardware + * @reg_count: Number of registers in the @reg[] array + * @reg: Array of register descriptors + */ +struct regs { + u32 reg_count; + const struct reg **reg; +}; + +static inline const struct reg *reg(const struct regs *regs, u32 reg_id) +{ + if (WARN(reg_id >= regs->reg_count, + "reg out of range (%u > %u)\n", reg_id, regs->reg_count - 1)) + return NULL; + + return regs->reg[reg_id]; +} + +/* Return the field mask for a field in a register, or 0 on error */ +static inline u32 reg_fmask(const struct reg *reg, u32 field_id) +{ + if (!reg || WARN_ON(field_id >= reg->fcount)) + return 0; + + return reg->fmask[field_id]; +} + +/* Return the mask for a single-bit field in a register, or 0 on error */ +static inline u32 reg_bit(const struct reg *reg, u32 field_id) +{ + u32 fmask = reg_fmask(reg, field_id); + + if (WARN_ON(!is_power_of_2(fmask))) + return 0; + + return fmask; +} + +/* Return the maximum value representable by the given field; always 2^n - 1 */ +static inline u32 reg_field_max(const struct reg *reg, u32 field_id) +{ + u32 fmask = reg_fmask(reg, field_id); + + return fmask ? fmask >> __ffs(fmask) : 0; +} + +/* Encode a value into the given field of a register */ +static inline u32 reg_encode(const struct reg *reg, u32 field_id, u32 val) +{ + u32 fmask = reg_fmask(reg, field_id); + + if (!fmask) + return 0; + + val <<= __ffs(fmask); + if (WARN_ON(val & ~fmask)) + return 0; + + return val; +} + +/* Given a register value, decode (extract) the value in the given field */ +static inline u32 reg_decode(const struct reg *reg, u32 field_id, u32 val) +{ + u32 fmask = reg_fmask(reg, field_id); + + return fmask ? (val & fmask) >> __ffs(fmask) : 0; +} + +/* Returns 0 for NULL reg; warning should have already been issued */ +static inline u32 reg_offset(const struct reg *reg) +{ + return reg ? reg->offset : 0; +} + +/* Returns 0 for NULL reg; warning should have already been issued */ +static inline u32 reg_n_offset(const struct reg *reg, u32 n) +{ + return reg ? reg->offset + n * reg->stride : 0; +} + +#endif /* _REG_H_ */ -- cgit v1.3.1