Re: [PATCH 2/3] writeback: per memcg dirty flush

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

 



[CC memcg maintainers and Tejun who has been quite active in the area
as well. Also linux-api ML added - please add this list whenever you are
suggesting user visible API]

On Wed 27-04-22 05:32:40, Xie Yongmei wrote:
> Currently, dirty writeback is under global control. We can tune it by
> parameters in /proc/sys/vm/
>   - dirty_expire_centisecs: expire interval in centiseconds
>   - dirty_writeback_centisecs: periodcal writeback interval in centiseconds
>   - dirty_background_bytes/dirty_background_ratio: async writeback
>     threshold
>   - dirty_bytes/dirty_ratio: sync writeback threshold
> 
> Sometimes, we'd like to specify special wrtiteback policy for user
> application, especially for offline application in co-location scenerio.
> 
> This patch provides dirty flush policy per memcg, user can specify them
> in memcg interface.
> 
> Actually, writeback code maintains two dimensions of dirty pages control in
> balance_dirty_pages.
>    - gdtc for global control
>    - mdtc for cgroup control
> 
> When dirty pages is under both of control, it leaves the check quickly.
> Otherwise, it computes the wb threshold (along with bg_threshold) taking
> the writeout bandwidth into consideration. And computes position ratio
> against wb_thresh for both global control and cgroup control as well.
> After that, it takes the smaller one (IOW the strict one) as the factor
> to generate task ratelimit based on wb's dirty_ratelimit.
> 
> So far, the writeback code can control the dirty limit for both global
> view and cgroup view. That means the framework works well for controlling
> cgroup's dirty limit.
> 
> This patch only provides an extra interface for memcg to tune writeback
> behavior.
> 
> Signed-off-by: Xie Yongmei <yongmeixie@xxxxxxxxxxx>
> ---
>  include/linux/memcontrol.h |  22 ++++++
>  init/Kconfig               |   7 ++
>  mm/memcontrol.c            | 136 +++++++++++++++++++++++++++++++++++++
>  mm/page-writeback.c        |  15 +++-
>  4 files changed, 178 insertions(+), 2 deletions(-)
> 
> diff --git a/include/linux/memcontrol.h b/include/linux/memcontrol.h
> index a68dce3873fc..386fc9b70c95 100644
> --- a/include/linux/memcontrol.h
> +++ b/include/linux/memcontrol.h
> @@ -344,6 +344,11 @@ struct mem_cgroup {
>  	struct deferred_split deferred_split_queue;
>  #endif
>  
> +#ifdef CONFIG_CGROUP_WRITEBACK_PARA
> +	int dirty_background_ratio;
> +	int dirty_ratio;
> +#endif
> +
>  	struct mem_cgroup_per_node *nodeinfo[];
>  };
>  
> @@ -1634,6 +1639,23 @@ static inline void mem_cgroup_flush_foreign(struct bdi_writeback *wb)
>  
>  #endif	/* CONFIG_CGROUP_WRITEBACK */
>  
> +#ifdef CONFIG_CGROUP_WRITEBACK_PARA
> +unsigned int wb_dirty_background_ratio(struct bdi_writeback *wb);
> +unsigned int wb_dirty_ratio(struct bdi_writeback *wb);
> +#else
> +static inline
> +unsigned int wb_dirty_background_ratio(struct bdi_writeback *wb)
> +{
> +	return dirty_background_ratio;
> +}
> +
> +static inline
> +unsigned int wb_dirty_ratio(struct bdi_writeback *wb)
> +{
> +	return vm_dirty_ratio;
> +}
> +#endif
> +
>  struct sock;
>  bool mem_cgroup_charge_skmem(struct mem_cgroup *memcg, unsigned int nr_pages,
>  			     gfp_t gfp_mask);
> diff --git a/init/Kconfig b/init/Kconfig
> index ddcbefe535e9..0b8152000d6e 100644
> --- a/init/Kconfig
> +++ b/init/Kconfig
> @@ -989,6 +989,13 @@ config CGROUP_WRITEBACK
>  	depends on MEMCG && BLK_CGROUP
>  	default y
>  
> +config CGROUP_WRITEBACK_PARA
> +	bool "Enable setup dirty flush parameters per memcg"
> +	depends on CGROUP_WRITEBACK
> +	default y
> +	help
> +	  This feature helps cgroup could specify its own diry wriback policy.
> +
>  menuconfig CGROUP_SCHED
>  	bool "CPU controller"
>  	default n
> diff --git a/mm/memcontrol.c b/mm/memcontrol.c
> index e8922bacfe2a..b1c1b150637a 100644
> --- a/mm/memcontrol.c
> +++ b/mm/memcontrol.c
> @@ -4822,6 +4822,112 @@ static int mem_cgroup_slab_show(struct seq_file *m, void *p)
>  }
>  #endif
>  
> +#ifdef CONFIG_CGROUP_WRITEBACK_PARA
> +unsigned int wb_dirty_background_ratio(struct bdi_writeback *wb)
> +{
> +	struct mem_cgroup *memcg;
> +
> +	if (mem_cgroup_disabled() || !wb)
> +		return dirty_background_ratio;
> +
> +	memcg = mem_cgroup_from_css(wb->memcg_css);
> +	if (memcg == root_mem_cgroup || memcg->dirty_background_ratio < 0)
> +		return dirty_background_ratio;
> +
> +	return memcg->dirty_background_ratio;
> +}
> +
> +unsigned int wb_dirty_ratio(struct bdi_writeback *wb)
> +{
> +	struct mem_cgroup *memcg;
> +
> +	if (mem_cgroup_disabled() || !wb)
> +		return vm_dirty_ratio;
> +
> +	memcg = mem_cgroup_from_css(wb->memcg_css);
> +	if (memcg == root_mem_cgroup || memcg->dirty_ratio < 0)
> +		return vm_dirty_ratio;
> +
> +	return memcg->dirty_ratio;
> +}
> +
> +static void wb_memcg_inherit_from_parent(struct mem_cgroup *parent,
> +					 struct mem_cgroup *memcg)
> +{
> +	memcg->dirty_background_ratio = parent->dirty_background_ratio;
> +	memcg->dirty_ratio = parent->dirty_ratio;
> +}
> +
> +static void wb_memcg_init(struct mem_cgroup *memcg)
> +{
> +	memcg->dirty_background_ratio = -1;
> +	memcg->dirty_ratio = -1;
> +}
> +
> +static int mem_cgroup_dirty_background_ratio_show(struct seq_file *m, void *v)
> +{
> +	struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m));
> +
> +	seq_printf(m, "%d\n", memcg->dirty_background_ratio);
> +	return 0;
> +}
> +
> +static ssize_t
> +mem_cgroup_dirty_background_ratio_write(struct kernfs_open_file *of,
> +					char *buf, size_t nbytes,
> +					loff_t off)
> +{
> +	struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of));
> +	int ret, background_ratio;
> +
> +	buf = strstrip(buf);
> +	ret = kstrtoint(buf, 0, &background_ratio);
> +	if (ret)
> +		return ret;
> +
> +	if (background_ratio < -1 || background_ratio > 100)
> +		return -EINVAL;
> +
> +	memcg->dirty_background_ratio = background_ratio;
> +	return nbytes;
> +}
> +
> +static int mem_cgroup_dirty_ratio_show(struct seq_file *m, void *v)
> +{
> +	struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m));
> +
> +	seq_printf(m, "%d\n", memcg->dirty_ratio);
> +	return 0;
> +}
> +
> +static ssize_t
> +mem_cgroup_dirty_ratio_write(struct kernfs_open_file *of,
> +			     char *buf, size_t nbytes, loff_t off)
> +{
> +	struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of));
> +	int ret, dirty_ratio;
> +
> +	buf = strstrip(buf);
> +	ret = kstrtoint(buf, 0, &dirty_ratio);
> +	if (ret)
> +		return ret;
> +
> +	if (dirty_ratio < -1 || dirty_ratio > 100)
> +		return -EINVAL;
> +
> +	memcg->dirty_ratio = dirty_ratio;
> +	return nbytes;
> +}
> +#else
> +static void wb_memcg_inherit_from_parent(struct mem_cgroup *parent,
> +					 struct mem_cgroup *memcg)
> +{
> +}
> +
> +static inline void wb_memcg_init(struct mem_cgroup *memcg)
> +{
> +}
> +#endif
>  static struct cftype mem_cgroup_legacy_files[] = {
>  	{
>  		.name = "usage_in_bytes",
> @@ -4948,6 +5054,20 @@ static struct cftype mem_cgroup_legacy_files[] = {
>  		.write = mem_cgroup_reset,
>  		.read_u64 = mem_cgroup_read_u64,
>  	},
> +#ifdef CONFIG_CGROUP_WRITEBACK_PARA
> +	{
> +		.name = "dirty_background_ratio",
> +		.flags = CFTYPE_NOT_ON_ROOT,
> +		.seq_show = mem_cgroup_dirty_background_ratio_show,
> +		.write = mem_cgroup_dirty_background_ratio_write,
> +	},
> +	{
> +		.name = "dirty_ratio",
> +		.flags = CFTYPE_NOT_ON_ROOT,
> +		.seq_show = mem_cgroup_dirty_ratio_show,
> +		.write = mem_cgroup_dirty_ratio_write,
> +	},
> +#endif
>  	{ },	/* terminate */
>  };
>  
> @@ -5151,11 +5271,13 @@ mem_cgroup_css_alloc(struct cgroup_subsys_state *parent_css)
>  		page_counter_init(&memcg->swap, &parent->swap);
>  		page_counter_init(&memcg->kmem, &parent->kmem);
>  		page_counter_init(&memcg->tcpmem, &parent->tcpmem);
> +		wb_memcg_inherit_from_parent(parent, memcg);
>  	} else {
>  		page_counter_init(&memcg->memory, NULL);
>  		page_counter_init(&memcg->swap, NULL);
>  		page_counter_init(&memcg->kmem, NULL);
>  		page_counter_init(&memcg->tcpmem, NULL);
> +		wb_memcg_init(memcg);
>  
>  		root_mem_cgroup = memcg;
>  		return &memcg->css;
> @@ -6414,6 +6536,20 @@ static struct cftype memory_files[] = {
>  		.seq_show = memory_oom_group_show,
>  		.write = memory_oom_group_write,
>  	},
> +#ifdef CONFIG_CGROUP_WRITEBACK_PARA
> +	{
> +		.name = "dirty_background_ratio",
> +		.flags = CFTYPE_NOT_ON_ROOT,
> +		.seq_show = mem_cgroup_dirty_background_ratio_show,
> +		.write = mem_cgroup_dirty_background_ratio_write,
> +	},
> +	{
> +		.name = "dirty_ratio",
> +		.flags = CFTYPE_NOT_ON_ROOT,
> +		.seq_show = mem_cgroup_dirty_ratio_show,
> +		.write = mem_cgroup_dirty_ratio_write,
> +	},
> +#endif
>  	{ }	/* terminate */
>  };
>  
> diff --git a/mm/page-writeback.c b/mm/page-writeback.c
> index 7e2da284e427..cec2ef032927 100644
> --- a/mm/page-writeback.c
> +++ b/mm/page-writeback.c
> @@ -395,12 +395,23 @@ static void domain_dirty_limits(struct dirty_throttle_control *dtc)
>  		 * per-PAGE_SIZE, they can be obtained by dividing bytes by
>  		 * number of pages.
>  		 */
> +#ifdef CONFIG_CGROUP_WRITEBACK_PARA
> +		ratio = (wb_dirty_ratio(dtc->wb) * PAGE_SIZE) / 100;
> +		bg_ratio = (wb_dirty_background_ratio(dtc->wb) * PAGE_SIZE) / 100;
> +		if (!ratio && bytes)
> +			ratio = min(DIV_ROUND_UP(bytes, global_avail),
> +				    PAGE_SIZE);
> +		if (!bg_ratio && bg_bytes)
> +			bg_ratio = min(DIV_ROUND_UP(bg_bytes, global_avail),
> +				       PAGE_SIZE);
> +#else
>  		if (bytes)
>  			ratio = min(DIV_ROUND_UP(bytes, global_avail),
>  				    PAGE_SIZE);
>  		if (bg_bytes)
>  			bg_ratio = min(DIV_ROUND_UP(bg_bytes, global_avail),
>  				       PAGE_SIZE);
> +#endif
>  		bytes = bg_bytes = 0;
>  	}
>  
> @@ -418,8 +429,8 @@ static void domain_dirty_limits(struct dirty_throttle_control *dtc)
>  		bg_thresh = thresh / 2;
>  	tsk = current;
>  	if (rt_task(tsk)) {
> -		bg_thresh += bg_thresh / 4 + global_wb_domain.dirty_limit / 32;
> -		thresh += thresh / 4 + global_wb_domain.dirty_limit / 32;
> +		bg_thresh += bg_thresh / 4 + dtc_dom(dtc)->dirty_limit / 32;
> +		thresh += thresh / 4 + dtc_dom(dtc)->dirty_limit / 32;
>  	}
>  	dtc->thresh = thresh;
>  	dtc->bg_thresh = bg_thresh;
> -- 
> 2.27.0

-- 
Michal Hocko
SUSE Labs




[Index of Archives]     [Linux ARM Kernel]     [Linux ARM]     [Linux Omap]     [Fedora ARM]     [IETF Annouce]     [Bugtraq]     [Linux OMAP]     [Linux MIPS]     [eCos]     [Asterisk Internet PBX]     [Linux API]

  Powered by Linux