Re: [PATCH v3] erofs: replace erofs_unzipd workqueue with per-cpu threads

[Gao Xiang <hsiangkao@linux.alibaba.com>]

Hi Sandeep,

On 2023/1/6 13:41, Sandeep Dhavale wrote:
> Using per-cpu thread pool we can reduce the scheduling latency compared
> to workqueue implementation. With this patch scheduling latency and
> variation is reduced as per-cpu threads are high priority kthread_workers.
> 
> The results were evaluated on arm64 Android devices running 5.10 kernel.
> 
> The table below shows resulting improvements of total scheduling latency
> for the same app launch benchmark runs with 50 iterations. Scheduling
> latency is the latency between when the task (workqueue kworker vs
> kthread_worker) became eligible to run to when it actually started
> running.
> +-------------------------+-----------+----------------+---------+
> |                         | workqueue | kthread_worker |  diff   |
> +-------------------------+-----------+----------------+---------+
> | Average (us)            |     15253 |           2914 | -80.89% |
> | Median (us)             |     14001 |           2912 | -79.20% |
> | Minimum (us)            |      3117 |           1027 | -67.05% |
> | Maximum (us)            |     30170 |           3805 | -87.39% |
> | Standard deviation (us) |      7166 |            359 |         |
> +-------------------------+-----------+----------------+---------+
> 
> Signed-off-by: Sandeep Dhavale <dhavale@google.com>
> ---
> V2 -> V3
> Fix a warning Reported-by: kernel test robot <lkp@intel.com>
> 
> V1 -> V2
> * Changed name of kthread_workers from z_erofs to erofs_worker
> * Added kernel configuration to run kthread_workers at normal or
>    high priority
> * Added cpu hotplug support
> * Added wrapped kthread_workers under worker_pool
> * Added one unbound thread in a pool to handle a context where
>    we already stopped per-cpu kthread worker
> * Updated commit message
> 
> Background: Boot times and cold app launch benchmarks are very
> important to android ecosystem as they directly translate to
> responsiveness from user point of view. While erofs provides
> lot of important features like space savings, we saw some
> performance penalty in cold app launch benchmarks in few scenarios.
> Analysis showed that the significant variance was coming from the
> scheduling cost while decompression cost was more or less the same.
> 
> Having per-cpu thread pool we can see from above table that this
> variation is reduced by ~80% on average. Link to LPC 2022 slides and
> talk at [1]
> 
> [1] https://lpc.events/event/16/contributions/1338/
> ---
>   fs/erofs/Kconfig |  11 +++
>   fs/erofs/zdata.c | 201 +++++++++++++++++++++++++++++++++++++++++------
>   fs/erofs/zdata.h |   4 +-
>   3 files changed, 192 insertions(+), 24 deletions(-)
> 
> diff --git a/fs/erofs/Kconfig b/fs/erofs/Kconfig
> index 85490370e0ca..879f493c6641 100644
> --- a/fs/erofs/Kconfig
> +++ b/fs/erofs/Kconfig
> @@ -108,3 +108,14 @@ config EROFS_FS_ONDEMAND
>   	  read support.
>   
>   	  If unsure, say N.
> +
> +config EROFS_FS_KTHREAD_HIPRI
> +	bool "EROFS high priority percpu kthread workers"
> +	depends on EROFS_FS
> +	default n
> +	help
> +	  EROFS uses per cpu kthread workers pool to carry out async work.
> +	  This permits EROFS to configure per cpu kthread workers to run
> +	  at higher priority.
> +
> +	  If unsure, say N.
> diff --git a/fs/erofs/zdata.c b/fs/erofs/zdata.c
> index ccf7c55d477f..b4bf2da72d1a 100644
> --- a/fs/erofs/zdata.c
> +++ b/fs/erofs/zdata.c
> @@ -8,6 +8,8 @@
>   #include "compress.h"
>   #include <linux/prefetch.h>
>   #include <linux/psi.h>
> +#include <linux/slab.h>
> +#include <linux/cpuhotplug.h>
>   
>   #include <trace/events/erofs.h>
>   
> @@ -184,26 +186,152 @@ typedef tagptr1_t compressed_page_t;
>   #define tag_compressed_page_justfound(page) \
>   	tagptr_fold(compressed_page_t, page, 1)
>   
> -static struct workqueue_struct *z_erofs_workqueue __read_mostly;
> +struct erofs_kthread_worker_pool {
> +	struct kthread_worker __rcu **workers;
> +	struct kthread_worker *unbound_worker;

Sorry for the late reply.

The commit overall looks good to me, but consider it's mainly to
resolve workqueue issue on some ARM64 processors, I wonder if we
could leave the original workqueue for unbounded fallback as well.

> +};
>   
> -void z_erofs_exit_zip_subsystem(void)
> +static struct erofs_kthread_worker_pool worker_pool;
> +DEFINE_SPINLOCK(worker_pool_lock);
> +
> +static void erofs_destroy_worker_pool(void)
>   {
> -	destroy_workqueue(z_erofs_workqueue);
> -	z_erofs_destroy_pcluster_pool();
> +	unsigned int cpu;
> +	struct kthread_worker *worker;
> +
> +	for_each_possible_cpu(cpu) {
> +		worker = rcu_dereference_protected(
> +				worker_pool.workers[cpu],
> +				1);
> +		rcu_assign_pointer(worker_pool.workers[cpu], NULL);
> +
> +		if (worker)
> +			kthread_destroy_worker(worker);
> +	}
> +
> +	if (worker_pool.unbound_worker)
> +		kthread_destroy_worker(worker_pool.unbound_worker);
> +
> +	kfree(worker_pool.workers);
>   }
>   
> -static inline int z_erofs_init_workqueue(void)
> +static inline void erofs_set_worker_priority(struct kthread_worker *worker)
>   {
> -	const unsigned int onlinecpus = num_possible_cpus();
> +#ifdef CONFIG_EROFS_FS_KTHREAD_HIPRI
> +	sched_set_fifo_low(worker->task);
> +#else
> +	sched_set_normal(worker->task, 0);
> +#endif
> +}
>   
> -	/*
> -	 * no need to spawn too many threads, limiting threads could minimum
> -	 * scheduling overhead, perhaps per-CPU threads should be better?
> -	 */
> -	z_erofs_workqueue = alloc_workqueue("erofs_unzipd",
> -					    WQ_UNBOUND | WQ_HIGHPRI,
> -					    onlinecpus + onlinecpus / 4);
> -	return z_erofs_workqueue ? 0 : -ENOMEM;
> +static int erofs_create_kthread_workers(void)
> +{
> +	unsigned int cpu;
> +	struct kthread_worker *worker;
> +
> +	for_each_online_cpu(cpu) {
> +		worker = kthread_create_worker_on_cpu(cpu, 0, "erofs_worker/%u", cpu);
> +		if (IS_ERR(worker)) {
> +			erofs_destroy_worker_pool();
> +			return -ENOMEM;
> +		}
> +		erofs_set_worker_priority(worker);
> +		rcu_assign_pointer(worker_pool.workers[cpu], worker);
> +	}
> +
> +	worker = kthread_create_worker(0, "erofs_unbound_worker");
> +	if (IS_ERR(worker)) {
> +		erofs_destroy_worker_pool();
> +		return PTR_ERR(worker);
> +	}
> +	erofs_set_worker_priority(worker);
> +	worker_pool.unbound_worker = worker;
> +
> +	return 0;
> +}
> +
> +static int erofs_init_worker_pool(void)
> +{
> +	int err;
> +
> +	worker_pool.workers = kcalloc(num_possible_cpus(),
> +			sizeof(struct kthread_worker *), GFP_ATOMIC);
> +	if (!worker_pool.workers)
> +		return -ENOMEM;
> +	err = erofs_create_kthread_workers();
> +
> +	return err;
> +}
> +
> +#ifdef CONFIG_HOTPLUG_CPU
> +static enum cpuhp_state erofs_cpuhp_state;
> +static int erofs_cpu_online(unsigned int cpu)
> +{
> +	struct kthread_worker *worker;
> +
> +	worker = kthread_create_worker_on_cpu(cpu, 0, "erofs_worker/%u", cpu);
> +	if (IS_ERR(worker))
> +		return -ENOMEM;
> +
> +	erofs_set_worker_priority(worker);
> +
> +	spin_lock(&worker_pool_lock);
> +	rcu_assign_pointer(worker_pool.workers[cpu], worker);
> +	spin_unlock(&worker_pool_lock);
> +
> +	synchronize_rcu();
> +
> +	return 0;
> +}
> +
> +static int erofs_cpu_offline(unsigned int cpu)
> +{
> +	struct kthread_worker *worker;
> +
> +	spin_lock(&worker_pool_lock);
> +	worker = rcu_dereference_protected(worker_pool.workers[cpu],
> +			lockdep_is_held(&worker_pool_lock));
> +	rcu_assign_pointer(worker_pool.workers[cpu], NULL);
> +	spin_unlock(&worker_pool_lock);
> +
> +	synchronize_rcu();
> +
> +	if (worker)
> +		kthread_destroy_worker(worker);
> +
> +	return 0;
> +}
> +
> +static int erofs_cpu_hotplug_init(void)
> +{
> +	int state;
> +
> +	state = cpuhp_setup_state_nocalls(CPUHP_AP_ONLINE_DYN,
> +			"fs/erofs:online",
> +			erofs_cpu_online, erofs_cpu_offline);
> +	if (state < 0)
> +		return state;
> +
> +	erofs_cpuhp_state = state;
> +
> +	return 0;
> +}
> +
> +static void erofs_cpu_hotplug_destroy(void)
> +{
> +	if (erofs_cpuhp_state)
> +		cpuhp_remove_state_nocalls(erofs_cpuhp_state);
> +}
> +#else /* !CONFIG_HOTPLUG_CPU */
> +static inline int erofs_cpu_hotplug_init(void) { return 0; }
> +static inline void erofs_cpu_hotplug_destroy(void) {}
> +#endif
> +
> +void z_erofs_exit_zip_subsystem(void)
> +{
> +	erofs_cpu_hotplug_destroy();
> +	erofs_destroy_worker_pool();
> +	z_erofs_destroy_pcluster_pool();
>   }
>   
>   int __init z_erofs_init_zip_subsystem(void)
> @@ -211,10 +339,23 @@ int __init z_erofs_init_zip_subsystem(void)
>   	int err = z_erofs_create_pcluster_pool();
>   
>   	if (err)
> -		return err;
> -	err = z_erofs_init_workqueue();
> +		goto out_error_pcluster_pool;
> +
> +	err = erofs_init_worker_pool();
>   	if (err)
> -		z_erofs_destroy_pcluster_pool();
> +		goto out_error_worker_pool;
> +
> +	err = erofs_cpu_hotplug_init();
> +	if (err < 0)
> +		goto out_error_cpuhp_init;
> +
> +	return err;
> +
> +out_error_cpuhp_init:
> +	erofs_destroy_worker_pool();
> +out_error_worker_pool:
> +	z_erofs_destroy_pcluster_pool();
> +out_error_pcluster_pool:
>   	return err;
>   }
>   
> @@ -1143,7 +1284,7 @@ static void z_erofs_decompress_queue(const struct z_erofs_decompressqueue *io,
>   	}
>   }
>   
> -static void z_erofs_decompressqueue_work(struct work_struct *work)
> +static void z_erofs_decompressqueue_kthread_work(struct kthread_work *work)
>   {
>   	struct z_erofs_decompressqueue *bgq =
>   		container_of(work, struct z_erofs_decompressqueue, u.work);
> @@ -1156,6 +1297,20 @@ static void z_erofs_decompressqueue_work(struct work_struct *work)
>   	kvfree(bgq);
>   }
>   
> +static void erofs_schedule_kthread_work(struct kthread_work *work)
> +{
> +	struct kthread_worker *worker;
> +	unsigned int cpu = raw_smp_processor_id();
> +
> +	rcu_read_lock();
> +	worker = rcu_dereference(worker_pool.workers[cpu]);
> +	if (!worker)
> +		worker = worker_pool.unbound_worker;

so that we have two modes,
1) per-cpu kworker mode;
2) unbounded workqueue (as a fallback / just use this if per-cpu kworker is disabled)

and 1) can be turned off by Kconfig as well if users don't need to use it,
but important to resolve the current Android scheduling issue.


Thanks,
Gao Xiang