[PATCH v3] erofs: lazily initialize per-CPU workers and CPU hotplug hooks

[Gao Xiang]

Hi Sandeep,

On 2025/4/23 07:45, Sandeep Dhavale wrote:
> Currently, when EROFS is built with per-CPU workers, the workers are
> started and CPU hotplug hooks are registered during module initialization.
> This leads to unnecessary worker start/stop cycles during CPU hotplug
> events, particularly on Android devices that frequently suspend and resume.
> 
> This change defers the initialization of per-CPU workers and the
> registration of CPU hotplug hooks until the first EROFS mount. This
> ensures that these resources are only allocated and managed when EROFS is
> actually in use.
> 
> The tear down of per-CPU workers and unregistration of CPU hotplug hooks
> still occurs during z_erofs_exit_subsystem(), but only if they were
> initialized.
> 
> Signed-off-by: Sandeep Dhavale <dhavale at google.com>
> ---
> v2: https://lore.kernel.org/linux-erofs/20250402202728.2157627-1-dhavale@google.com/
> Changes since v2:
> - Renamed functions to use pcpu so it is clear.
> - Removed z_erofs_init_workers_once() declaration from internal.h as
>    there is no need.
> - Removed empty stubs for helpers erofs_init_percpu_workers() and
>    erofs_destroy_percpu_workers().
> - Moved erofs_percpu_workers_initialized under
>    CONFIG_EROFS_FS_PCPU_KTHREAD as further cleanup.
> 
>   fs/erofs/zdata.c | 65 ++++++++++++++++++++++++++++++++++++------------
>   1 file changed, 49 insertions(+), 16 deletions(-)
> 
> diff --git a/fs/erofs/zdata.c b/fs/erofs/zdata.c
> index 0671184d9cf1..e12df8b914b6 100644
> --- a/fs/erofs/zdata.c
> +++ b/fs/erofs/zdata.c
> @@ -291,6 +291,9 @@ static struct workqueue_struct *z_erofs_workqueue __read_mostly;
>   
>   #ifdef CONFIG_EROFS_FS_PCPU_KTHREAD
>   static struct kthread_worker __rcu **z_erofs_pcpu_workers;
> +static atomic_t erofs_percpu_workers_initialized = ATOMIC_INIT(0);
> +static int erofs_cpu_hotplug_init(void);
> +static void erofs_cpu_hotplug_destroy(void);
>   
>   static void erofs_destroy_percpu_workers(void)
>   {
> @@ -336,9 +339,44 @@ static int erofs_init_percpu_workers(void)
>   	}
>   	return 0;
>   }
> +
> +static int z_erofs_init_pcpu_workers(void)

I think you could just fold it in the caller.

> +{
> +	int err;
> +
> +	err = erofs_init_percpu_workers();
> +	if (err)
> +		goto err_init_percpu_workers;
> +
> +	err = erofs_cpu_hotplug_init();
> +	if (err < 0)
> +		goto err_cpuhp_init;
> +	return err;
> +
> +err_cpuhp_init:
> +	erofs_destroy_percpu_workers();
> +err_init_percpu_workers:
> +	atomic_set(&erofs_percpu_workers_initialized, 0);
> +	return err;
> +}
> +
> +static int z_erofs_init_workers_once(void)

and rename it as `z_erofs_init_pcpu_workers()` since
initializing once is just an internal implmentation.

> +{
> +	if (atomic_xchg(&erofs_percpu_workers_initialized, 1))
> +		return 0;
> +	return z_erofs_init_pcpu_workers();
> +}
> +

..

>   err_pcluster_pool:
> @@ -644,10 +670,17 @@ static const struct address_space_operations z_erofs_cache_aops = {
>   
>   int z_erofs_init_super(struct super_block *sb)
>   {
> -	struct inode *const inode = new_inode(sb);
> +	struct inode *inode;
> +	int err;
>   
> +	err = z_erofs_init_workers_once();
> +	if (err)
> +		return err;


Then just call
	err = z_erofs_init_pcpu_workers();
	if (err)
		return err;
here.


Otherwise it looks good to me.

Thanks,
Gao Xiang