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, 209 insertions, 0 deletions

diff --git a/drivers/misc/ibmvmc.h b/drivers/misc/ibmvmc.h
new file mode 100644
index 000000000..0e1756fff
--- /dev/null
+++ b/drivers/misc/ibmvmc.h
@@ -0,0 +1,209 @@
+/* SPDX-License-Identifier: GPL-2.0+ */
+/*
+ * linux/drivers/misc/ibmvmc.h
+ *
+ * IBM Power Systems Virtual Management Channel Support.
+ *
+ * Copyright (c) 2004, 2018 IBM Corp.
+ *   Dave Engebretsen engebret@us.ibm.com
+ *   Steven Royer seroyer@linux.vnet.ibm.com
+ *   Adam Reznechek adreznec@linux.vnet.ibm.com
+ *   Bryant G. Ly <bryantly@linux.vnet.ibm.com>
+ */
+#ifndef IBMVMC_H
+#define IBMVMC_H
+
+#include <linux/types.h>
+#include <linux/cdev.h>
+
+#include <asm/vio.h>
+
+#define IBMVMC_PROTOCOL_VERSION    0x0101
+
+#define MIN_BUF_POOL_SIZE 16
+#define MIN_HMCS          1
+#define MIN_MTU           4096
+#define MAX_BUF_POOL_SIZE 64
+#define MAX_HMCS          2
+#define MAX_MTU           (4 * 4096)
+#define DEFAULT_BUF_POOL_SIZE 32
+#define DEFAULT_HMCS          1
+#define DEFAULT_MTU           4096
+#define HMC_ID_LEN        32
+
+#define VMC_INVALID_BUFFER_ID 0xFFFF
+
+/* ioctl numbers */
+#define VMC_BASE	     0xCC
+#define VMC_IOCTL_SETHMCID   _IOW(VMC_BASE, 0x00, unsigned char *)
+#define VMC_IOCTL_QUERY      _IOR(VMC_BASE, 0x01, struct ibmvmc_query_struct)
+#define VMC_IOCTL_REQUESTVMC _IOR(VMC_BASE, 0x02, u32)
+
+#define VMC_MSG_CAP          0x01
+#define VMC_MSG_CAP_RESP     0x81
+#define VMC_MSG_OPEN         0x02
+#define VMC_MSG_OPEN_RESP    0x82
+#define VMC_MSG_CLOSE        0x03
+#define VMC_MSG_CLOSE_RESP   0x83
+#define VMC_MSG_ADD_BUF      0x04
+#define VMC_MSG_ADD_BUF_RESP 0x84
+#define VMC_MSG_REM_BUF      0x05
+#define VMC_MSG_REM_BUF_RESP 0x85
+#define VMC_MSG_SIGNAL       0x06
+
+#define VMC_MSG_SUCCESS 0
+#define VMC_MSG_INVALID_HMC_INDEX 1
+#define VMC_MSG_INVALID_BUFFER_ID 2
+#define VMC_MSG_CLOSED_HMC        3
+#define VMC_MSG_INTERFACE_FAILURE 4
+#define VMC_MSG_NO_BUFFER         5
+
+#define VMC_BUF_OWNER_ALPHA 0
+#define VMC_BUF_OWNER_HV    1
+
+enum ibmvmc_states {
+	ibmvmc_state_sched_reset  = -1,
+	ibmvmc_state_initial      = 0,
+	ibmvmc_state_crqinit      = 1,
+	ibmvmc_state_capabilities = 2,
+	ibmvmc_state_ready        = 3,
+	ibmvmc_state_failed       = 4,
+};
+
+enum ibmhmc_states {
+	/* HMC connection not established */
+	ibmhmc_state_free    = 0,
+
+	/* HMC connection established (open called) */
+	ibmhmc_state_initial = 1,
+
+	/* open msg sent to HV, due to ioctl(1) call */
+	ibmhmc_state_opening = 2,
+
+	/* HMC connection ready, open resp msg from HV */
+	ibmhmc_state_ready   = 3,
+
+	/* HMC connection failure */
+	ibmhmc_state_failed  = 4,
+};
+
+struct ibmvmc_buffer {
+	u8 valid;	/* 1 when DMA storage allocated to buffer          */
+	u8 free;	/* 1 when buffer available for the Alpha Partition */
+	u8 owner;
+	u16 id;
+	u32 size;
+	u32 msg_len;
+	dma_addr_t dma_addr_local;
+	dma_addr_t dma_addr_remote;
+	void *real_addr_local;
+};
+
+struct ibmvmc_admin_crq_msg {
+	u8 valid;	/* RPA Defined           */
+	u8 type;	/* ibmvmc msg type       */
+	u8 status;	/* Response msg status. Zero is success and on failure,
+			 * either 1 - General Failure, or 2 - Invalid Version is
+			 * returned.
+			 */
+	u8 rsvd[2];
+	u8 max_hmc;	/* Max # of independent HMC connections supported */
+	__be16 pool_size;	/* Maximum number of buffers supported per HMC
+				 * connection
+				 */
+	__be32 max_mtu;		/* Maximum message size supported (bytes) */
+	__be16 crq_size;	/* # of entries available in the CRQ for the
+				 * source partition. The target partition must
+				 * limit the number of outstanding messages to
+				 * one half or less.
+				 */
+	__be16 version;	/* Indicates the code level of the management partition
+			 * or the hypervisor with the high-order byte
+			 * indicating a major version and the low-order byte
+			 * indicating a minor version.
+			 */
+};
+
+struct ibmvmc_crq_msg {
+	u8 valid;     /* RPA Defined           */
+	u8 type;      /* ibmvmc msg type       */
+	u8 status;    /* Response msg status   */
+	union {
+		u8 rsvd;  /* Reserved              */
+		u8 owner;
+	} var1;
+	u8 hmc_session;	/* Session Identifier for the current VMC connection */
+	u8 hmc_index;	/* A unique HMC Idx would be used if multiple management
+			 * applications running concurrently were desired
+			 */
+	union {
+		__be16 rsvd;
+		__be16 buffer_id;
+	} var2;
+	__be32 rsvd;
+	union {
+		__be32 rsvd;
+		__be32 lioba;
+		__be32 msg_len;
+	} var3;
+};
+
+/* an RPA command/response transport queue */
+struct crq_queue {
+	struct ibmvmc_crq_msg *msgs;
+	int size, cur;
+	dma_addr_t msg_token;
+	spinlock_t lock;
+};
+
+/* VMC server adapter settings */
+struct crq_server_adapter {
+	struct device *dev;
+	struct crq_queue queue;
+	u32 liobn;
+	u32 riobn;
+	struct tasklet_struct work_task;
+	wait_queue_head_t reset_wait_queue;
+	struct task_struct *reset_task;
+};
+
+/* Driver wide settings */
+struct ibmvmc_struct {
+	u32 state;
+	u32 max_mtu;
+	u32 max_buffer_pool_size;
+	u32 max_hmc_index;
+	struct crq_server_adapter *adapter;
+	struct cdev cdev;
+	u32 vmc_drc_index;
+};
+
+struct ibmvmc_file_session;
+
+/* Connection specific settings */
+struct ibmvmc_hmc {
+	u8 session;
+	u8 index;
+	u32 state;
+	struct crq_server_adapter *adapter;
+	spinlock_t lock;
+	unsigned char hmc_id[HMC_ID_LEN];
+	struct ibmvmc_buffer buffer[MAX_BUF_POOL_SIZE];
+	unsigned short queue_outbound_msgs[MAX_BUF_POOL_SIZE];
+	int queue_head, queue_tail;
+	struct ibmvmc_file_session *file_session;
+};
+
+struct ibmvmc_file_session {
+	struct file *file;
+	struct ibmvmc_hmc *hmc;
+	bool valid;
+};
+
+struct ibmvmc_query_struct {
+	int have_vmc;
+	int state;
+	int vmc_drc_index;
+};
+
+#endif /* __IBMVMC_H */