hulk inclusion category: bugfix bugzilla: https://gitee.com/openeuler/kernel/issues/IBN6HC -------------------------------- Currently, when moving a task to a resource group the PQR_ASSOC MSR is updated with the new closid and rmid in an added task callback. If the task is running, the work is run as soon as possible. If the task is not running, the work is executed later in the kernel exit path when the kernel returns to the task again. Updating the PQR_ASSOC MSR as soon as possible on the CPU a moved task is running is the right thing to do. Queueing work for a task that is not running is unnecessary (the PQR_ASSOC MSR is already updated when the task is scheduled in) and causing system resource waste with the way in which it is implemented: Work to update the PQR_ASSOC register is queued every time the user writes a task id to the "tasks" file, even if the task already belongs to the resource group. This could result in multiple pending work items associated with a single task even if they are all identical and even though only a single update with most recent values is needed. Specifically, even if a task is moved between different resource groups while it is sleeping then it is only the last move that is relevant but yet a work item is queued during each move. This unnecessary queueing of work items could result in significant system resource waste, especially on tasks sleeping for a long time. For example, as demonstrated by Shakeel Butt in [1] writing the same task id to the "tasks" file can quickly consume significant memory. The same problem (wasted system resources) occurs when moving a task between different resource groups. As pointed out by Valentin Schneider in [2] there is an additional issue with the way in which the queueing of work is done in that the task_struct update is currently done after the work is queued, resulting in a race with the register update possibly done before the data needed by the update is available. To solve these issues, update the MPAM sysreg in a synchronous way right after the new closid and rmid are ready during the task movement, only if the task is running. If a moved task is not running nothing is done since the MPAM sysreg will be updated next time the task is scheduled. This is the same way used to update the register when tasks are moved as part of resource group removal. Link: https://lore.kernel.org/all/17aa2fb38fc12ce7bb710106b3e7c7b45acb9e94.1608243... Fixes: e37caef15c18 ("resctrlfs: mpam: Build basic framework for mpam") Signed-off-by: Zeng Heng <zengheng4@huawei.com> --- arch/arm64/kernel/mpam/mpam_resctrl.c | 109 ++++++++++---------------- 1 file changed, 43 insertions(+), 66 deletions(-) diff --git a/arch/arm64/kernel/mpam/mpam_resctrl.c b/arch/arm64/kernel/mpam/mpam_resctrl.c index 39ee9f6e65f2..65ef64d2bec7 100644 --- a/arch/arm64/kernel/mpam/mpam_resctrl.c +++ b/arch/arm64/kernel/mpam/mpam_resctrl.c @@ -1311,7 +1311,7 @@ void update_cpu_closid_rmid(void *info) } /* - * Update the PGR_ASSOC MSR on all cpus in @cpu_mask, + * Update the MPAM sysreg on all cpus in @cpu_mask, * * Per task closids/rmids must have been set up before calling this function. */ @@ -1326,86 +1326,63 @@ update_closid_rmid(const struct cpumask *cpu_mask, struct rdtgroup *r) put_cpu(); } -struct task_move_callback { - struct callback_head work; - struct rdtgroup *rdtgrp; -}; - -static void move_myself(struct callback_head *head) +static void _update_task_closid_rmid(void *task) { - struct task_move_callback *callback; - struct rdtgroup *rdtgrp; - - callback = container_of(head, struct task_move_callback, work); - rdtgrp = callback->rdtgrp; - /* - * If resource group was deleted before this task work callback - * was invoked, then assign the task to root group and free the - * resource group. + * If the task is still current on this CPU, update MPAM sysreg. + * Otherwise, the MSR is updated when the task is scheduled in. */ - if (atomic_dec_and_test(&rdtgrp->waitcount) && - (rdtgrp->flags & RDT_DELETED)) { - current->closid = 0; - current->rmid = 0; - rdtgroup_remove(rdtgrp); - } - - preempt_disable(); - /* update PQR_ASSOC MSR to make resource group go into effect */ - mpam_sched_in(); - preempt_enable(); + if (task == current) + mpam_sched_in(); +} - kfree(callback); +static void update_task_closid_rmid(struct task_struct *t) +{ + if (IS_ENABLED(CONFIG_SMP) && task_curr(t)) + smp_call_function_single(task_cpu(t), _update_task_closid_rmid, t, 1); + else + _update_task_closid_rmid(t); } int __resctrl_group_move_task(struct task_struct *tsk, struct rdtgroup *rdtgrp) { - struct task_move_callback *callback; - int ret; - - callback = kzalloc(sizeof(*callback), GFP_NOWAIT); - if (!callback) - return -ENOMEM; - callback->work.func = move_myself; - callback->rdtgrp = rdtgrp; - /* - * Take a refcount, so rdtgrp cannot be freed before the - * callback has been invoked. + * Set the task's closid/rmid before the MPAM sysreg can be + * updated by them. + * + * For ctrl_mon groups, move both closid and rmid. + * For monitor groups, can move the tasks only from + * their parent CTRL group. */ - atomic_inc(&rdtgrp->waitcount); - ret = task_work_add(tsk, &callback->work, true); - if (ret) { - /* - * Task is exiting. Drop the refcount and free the callback. - * No need to check the refcount as the group cannot be - * deleted before the write function unlocks resctrl_group_mutex. - */ - atomic_dec(&rdtgrp->waitcount); - kfree(callback); - rdt_last_cmd_puts("task exited\n"); - } else { - /* - * For ctrl_mon groups move both closid and rmid. - * For monitor groups, can move the tasks only from - * their parent CTRL group. - */ - if (rdtgrp->type == RDTCTRL_GROUP) { + if (rdtgrp->type == RDTCTRL_GROUP) { + tsk->closid = resctrl_navie_closid(rdtgrp->closid); + tsk->rmid = resctrl_navie_rmid(rdtgrp->mon.rmid); + } else if (rdtgrp->type == RDTMON_GROUP) { + if (rdtgrp->mon.parent->closid.intpartid == tsk->closid) { tsk->closid = resctrl_navie_closid(rdtgrp->closid); tsk->rmid = resctrl_navie_rmid(rdtgrp->mon.rmid); - } else if (rdtgrp->type == RDTMON_GROUP) { - if (rdtgrp->mon.parent->closid.intpartid == tsk->closid) { - tsk->closid = resctrl_navie_closid(rdtgrp->closid); - tsk->rmid = resctrl_navie_rmid(rdtgrp->mon.rmid); - } else { - rdt_last_cmd_puts("Can't move task to different control group\n"); - ret = -EINVAL; - } + } else { + rdt_last_cmd_puts("Can't move task to different control group\n"); + return -EINVAL; } } - return ret; + + /* + * Ensure the task's closid and rmid are written before determining if + * the task is current that will decide if it will be interrupted. + */ + barrier(); + + /* + * By now, the task's closid and rmid are set. If the task is current + * on a CPU, the MPAM sysreg needs to be updated to make the resource + * group go into effect. If the task is not current, the sysreg will be + * updated when the task is scheduled in. + */ + update_task_closid_rmid(tsk); + + return 0; } static int resctrl_group_seqfile_show(struct seq_file *m, void *arg) -- 2.25.1

hulk inclusion category: bugfix bugzilla: https://gitee.com/openeuler/kernel/issues/IBN6HC -------------------------------- When the user moves a running task to a new rdtgroup using the task's file interface or by deleting its rdtgroup, the resulting change in CLOSID/RMID must be immediately propagated to the PQR_ASSOC MSR on the task(s) CPUs. x86 allows reordering loads with prior stores, so if the task starts running between a task_curr() check that the CPU hoisted before the stores in the CLOSID/RMID update then it can start running with the old CLOSID/RMID until it is switched again because __rdtgroup_move_task() failed to determine that it needs to be interrupted to obtain the new CLOSID/RMID. Refer to the diagram below: CPU 0 CPU 1 ----- ----- __rdtgroup_move_task(): curr <- t1->cpu->rq->curr __schedule(): rq->curr <- t1 resctrl_sched_in(): t1->{closid,rmid} -> {1,1} t1->{closid,rmid} <- {2,2} if (curr == t1) // false IPI(t1->cpu) A similar race impacts rdt_move_group_tasks(), which updates tasks in a deleted rdtgroup. In both cases, use smp_mb() to order the task_struct::{closid,rmid} stores before the loads in task_curr(). In particular, in the rdt_move_group_tasks() case, simply execute an smp_mb() on every iteration with a matching task. It is possible to use a single smp_mb() in rdt_move_group_tasks(), but this would require two passes and a means of remembering which task_structs were updated in the first loop. However, benchmarking results below showed too little performance impact in the simple approach to justify implementing the two-pass approach. Times below were collected using `perf stat` to measure the time to remove a group containing a 1600-task, parallel workload. CPU: Intel(R) Xeon(R) Platinum P-8136 CPU @ 2.00GHz (112 threads) # mkdir /sys/fs/resctrl/test # echo $$ > /sys/fs/resctrl/test/tasks # perf bench sched messaging -g 40 -l 100000 task-clock time ranges collected using: # perf stat rmdir /sys/fs/resctrl/test Baseline: 1.54 - 1.60 ms smp_mb() every matching task: 1.57 - 1.67 ms Link: https://lore.kernel.org/r/20221220161123.432120-1-peternewman@google.com Fixes: 6065f5513f78 ("resctrlfs: init support resctrlfs") Signed-off-by: Zeng Heng <zengheng4@huawei.com> --- arch/arm64/kernel/mpam/mpam_resctrl.c | 4 +++- fs/resctrlfs.c | 8 ++++++++ 2 files changed, 11 insertions(+), 1 deletion(-) diff --git a/arch/arm64/kernel/mpam/mpam_resctrl.c b/arch/arm64/kernel/mpam/mpam_resctrl.c index b04c0d2b8312..ab822fd6828d 100644 --- a/arch/arm64/kernel/mpam/mpam_resctrl.c +++ b/arch/arm64/kernel/mpam/mpam_resctrl.c @@ -1371,8 +1371,10 @@ int __resctrl_group_move_task(struct task_struct *tsk, /* * Ensure the task's closid and rmid are written before determining if * the task is current that will decide if it will be interrupted. + * This pairs with the full barrier between the rq->curr update and + * resctrl_sched_in() during context switch. */ - barrier(); + smp_mb(); /* * By now, the task's closid and rmid are set. If the task is current diff --git a/fs/resctrlfs.c b/fs/resctrlfs.c index 29fe73571870..7d816b33e209 100644 --- a/fs/resctrlfs.c +++ b/fs/resctrlfs.c @@ -457,6 +457,14 @@ static void resctrl_move_group_tasks(struct resctrl_group *from, struct resctrl_ WRITE_ONCE(t->closid, resctrl_navie_closid(to->closid)); WRITE_ONCE(t->rmid, resctrl_navie_rmid(to->mon.rmid)); + /* + * Order the closid/rmid stores above before the loads + * in task_curr(). This pairs with the full barrier + * between the rq->curr update and mpam_sched_in() + * during context switch. + */ + smp_mb(); + #ifdef CONFIG_SMP /* * This is safe on x86 w/o barriers as the ordering -- 2.25.1