Re: [PATCH-cgroup 1/4] workqueue: Add workqueue_unbound_exclude_cpumask() to exclude CPUs from wq_unbound_cpumask

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

 



Hello,

On Fri, Oct 13, 2023 at 02:11:19PM -0400, Waiman Long wrote:
> When the "isolcpus" boot command line option is used to add a set
> of isolated CPUs, those CPUs will be excluded automatically from
> wq_unbound_cpumask to avoid running work functions from unbound
> workqueues.
> 
> Recently cpuset has been extended to allow the creation of partitions
> of isolated CPUs dynamically. To make it closer to the "isolcpus"
> in functionality, the CPUs in those isolated cpuset partitions should be
> excluded from wq_unbound_cpumask as well. This can be done currently by
> explicitly writing to the workqueue's cpumask sysfs file after creating
> the isolated partitions. However, this process can be error prone.
> Ideally, the cpuset code should be allowed to request the workqueue code
> to exclude those isolated CPUs from wq_unbound_cpumask so that this
> operation can be done automatically and the isolated CPUs will be returned
> back to wq_unbound_cpumask after the destructions of the isolated
> cpuset partitions.
> 
> This patch adds a new workqueue_unbound_exclude_cpumask() to enable
> that. This new function will exclude the specified isolated CPUs
> from wq_unbound_cpumask. To be able to restore those isolated CPUs
> back after the destruction of isolated cpuset partitions, a new
> wq_user_unbound_cpumask is added to store the user provided unbound
> cpumask either from the boot command line options or from writing to
> the cpumask sysfs file. This new cpumask provides the basis for CPU
> exclusion.

The behaviors around wq_unbound_cpumask is getting pretty inconsistent:

1. Housekeeping excludes isolated CPUs on boot but allows user to override
   it to include isolated CPUs afterwards.

2. If an unbound wq's cpumask doesn't have any intersection with
   wq_unbound_cpumask we ignore the per-wq cpumask and falls back to
   wq_unbound_cpumask.

3. You're adding a masking layer on top with exclude which fails to set if
   the intersection is empty.

Can we do the followings for consistency?

1. User's requested_unbound_cpumask is stored separately (as in this patch).

2. The effect wq_unbound_cpumask is determined by requested_unbound_cpumask
   & housekeeping_cpumask & cpuset_allowed_cpumask. The operation order
   matters. When an & operation yields an cpumask, the cpumask from the
   previous step is the effective one.

3. Expose these cpumasks in sysfs so that what's happening is obvious.

> +int workqueue_unbound_exclude_cpumask(cpumask_var_t exclude_cpumask)
> +{
> +	cpumask_var_t cpumask;
> +	int ret = 0;
> +
> +	if (!zalloc_cpumask_var(&cpumask, GFP_KERNEL))
> +		return -ENOMEM;
> +
> +	/*
> +	 * The caller of this function may have called cpus_read_lock(),
> +	 * use cpus_read_trylock() to avoid potential deadlock.
> +	 */
> +	if (!cpus_read_trylock())
> +		return -EBUSY;

This means that a completely unrelated cpus_write_lock() can fail this
operation and thus cpuset config writes. Let's please not do this. Can't we
just make sure that the caller holds the lock?

> +	mutex_lock(&wq_pool_mutex);
> +
> +	if (!cpumask_andnot(cpumask, wq_user_unbound_cpumask, exclude_cpumask))
> +		ret = -EINVAL;	/* The new cpumask can't be empty */

For better or worse, the usual mode-of-failure for "no usable CPU" is just
falling back to something which works rather than failing the operation.
Let's follow that.

Thanks.

-- 
tejun



[Index of Archives]     [Linux Wireless]     [Linux Kernel]     [ATH6KL]     [Linux Bluetooth]     [Linux Netdev]     [Kernel Newbies]     [Share Photos]     [IDE]     [Security]     [Git]     [Netfilter]     [Bugtraq]     [Yosemite News]     [MIPS Linux]     [ARM Linux]     [Linux Security]     [Linux RAID]     [Linux ATA RAID]     [Samba]     [Device Mapper]

  Powered by Linux