Re: [PATCH v7 04/15] mm/damon: Implement region based sampling

[Jonathan Cameron <Jonathan.Cameron@Huawei.com>]

On Wed, 18 Mar 2020 12:27:11 +0100
SeongJae Park <sjpark@amazon.com> wrote:

> From: SeongJae Park <sjpark@amazon.de>
> 
> This commit implements DAMON's basic access check and region based
> sampling mechanisms.  This change would seems make no sense, mainly
> because it is only a part of the DAMON's logics.  Following two commits
> will make more sense.
> 
> Basic Access Check
> ------------------
> 
> DAMON basically reports what pages are how frequently accessed.  Note
> that the frequency is not an absolute number of accesses, but a relative
> frequency among the pages of the target workloads.
> 
> Users can control the resolution of the reports by setting two time
> intervals, ``sampling interval`` and ``aggregation interval``.  In
> detail, DAMON checks access to each page per ``sampling interval``,
> aggregates the results (counts the number of the accesses to each page),
> and reports the aggregated results per ``aggregation interval``.  For
> the access check of each page, DAMON uses the Accessed bits of PTEs.
> 
> This is thus similar to common periodic access checks based access
> tracking mechanisms, which overhead is increasing as the size of the
> target process grows.
> 
> Region Based Sampling
> ---------------------
> 
> To avoid the unbounded increase of the overhead, DAMON groups a number
> of adjacent pages that assumed to have same access frequencies into a
> region.  As long as the assumption (pages in a region have same access
> frequencies) is kept, only one page in the region is required to be
> checked.  Thus, for each ``sampling interval``, DAMON randomly picks one
> page in each region and clears its Accessed bit.  After one more
> ``sampling interval``, DAMON reads the Accessed bit of the page and
> increases the access frequency of the region if the bit has set
> meanwhile.  Therefore, the monitoring overhead is controllable by
> setting the number of regions.
> 
> Nonetheless, this scheme cannot preserve the quality of the output if
> the assumption is not kept.  Following commit will introduce how we can
> make the guarantee with best effort.
> 
> Signed-off-by: SeongJae Park <sjpark@amazon.de>

Hi.

A few comments inline.

I've still not replicated your benchmarks so may well have some more
feedback once I've managed that on one of our servers.

Thanks,

Jonathan

> ---
>  include/linux/damon.h |  24 ++
>  mm/damon.c            | 553 ++++++++++++++++++++++++++++++++++++++++++
>  2 files changed, 577 insertions(+)
> 
> diff --git a/include/linux/damon.h b/include/linux/damon.h
> index 7117bb7e7544..f1945df6e6b4 100644
> --- a/include/linux/damon.h
> +++ b/include/linux/damon.h
> @@ -11,6 +11,8 @@
>  #define _DAMON_H_
>  
>  #include <linux/random.h>
> +#include <linux/mutex.h>
> +#include <linux/time64.h>
>  #include <linux/types.h>
>  
>  /* Represents a monitoring target region on the virtual address space */
> @@ -29,10 +31,32 @@ struct damon_task {
>  	struct list_head list;
>  };
>  
> +/*
> + * For each 'sample_interval', DAMON checks whether each region is accessed or
> + * not.  It aggregates and keeps the access information (number of accesses to
> + * each region) for each 'aggr_interval' time.
> + *
> + * All time intervals are in micro-seconds.
> + */
>  struct damon_ctx {
> +	unsigned long sample_interval;
> +	unsigned long aggr_interval;
> +	unsigned long min_nr_regions;
> +
> +	struct timespec64 last_aggregation;
> +
> +	struct task_struct *kdamond;
> +	struct mutex kdamond_lock;
> +
>  	struct rnd_state rndseed;
>  
>  	struct list_head tasks_list;	/* 'damon_task' objects */
>  };
>  
> +int damon_set_pids(struct damon_ctx *ctx, unsigned long *pids, ssize_t nr_pids);
> +int damon_set_attrs(struct damon_ctx *ctx, unsigned long sample_int,
> +		unsigned long aggr_int, unsigned long min_nr_reg);
> +int damon_start(struct damon_ctx *ctx);
> +int damon_stop(struct damon_ctx *ctx);
> +
>  #endif
> diff --git a/mm/damon.c b/mm/damon.c
> index d7e6226ab7f1..018016793555 100644
> --- a/mm/damon.c
> +++ b/mm/damon.c
> @@ -10,8 +10,14 @@
>  #define pr_fmt(fmt) "damon: " fmt
>  
>  #include <linux/damon.h>
> +#include <linux/delay.h>
> +#include <linux/kthread.h>
>  #include <linux/mm.h>
>  #include <linux/module.h>
> +#include <linux/page_idle.h>
> +#include <linux/random.h>
> +#include <linux/sched/mm.h>
> +#include <linux/sched/task.h>
>  #include <linux/slab.h>
>  
>  #define damon_get_task_struct(t) \
> @@ -171,6 +177,553 @@ static unsigned int nr_damon_regions(struct damon_task *t)
>  	return nr_regions;
>  }
>  
> +/*
> + * Get the mm_struct of the given task
> + *
> + * Caller should put the mm_struct after use, unless it is NULL.
> + *
> + * Returns the mm_struct of the task on success, NULL on failure
> + */
> +static struct mm_struct *damon_get_mm(struct damon_task *t)
> +{
> +	struct task_struct *task;
> +	struct mm_struct *mm;
> +
> +	task = damon_get_task_struct(t);
> +	if (!task)
> +		return NULL;
> +
> +	mm = get_task_mm(task);
> +	put_task_struct(task);
> +	return mm;
> +}
> +
> +/*
> + * Size-evenly split a region into 'nr_pieces' small regions
> + *
> + * Returns 0 on success, or negative error code otherwise.
> + */
> +static int damon_split_region_evenly(struct damon_ctx *ctx,
> +		struct damon_region *r, unsigned int nr_pieces)
> +{
> +	unsigned long sz_orig, sz_piece, orig_end;
> +	struct damon_region *piece = NULL, *next;
> +	unsigned long start;
> +
> +	if (!r || !nr_pieces)
> +		return -EINVAL;
> +
> +	orig_end = r->vm_end;
> +	sz_orig = r->vm_end - r->vm_start;
> +	sz_piece = sz_orig / nr_pieces;
> +
> +	if (!sz_piece)
> +		return -EINVAL;
> +
> +	r->vm_end = r->vm_start + sz_piece;
> +	next = damon_next_region(r);
> +	for (start = r->vm_end; start + sz_piece <= orig_end;
> +			start += sz_piece) {
> +		piece = damon_new_region(ctx, start, start + sz_piece);
piece may be n
> +		damon_insert_region(piece, r, next);
> +		r = piece;
> +	}
> +	/* complement last region for possible rounding error */
> +	if (piece)
> +		piece->vm_end = orig_end;

Update the sampling address to ensure it's in the region?

> +
> +	return 0;
> +}
> +
> +struct region {
> +	unsigned long start;
> +	unsigned long end;
> +};
> +
> +static unsigned long sz_region(struct region *r)
> +{
> +	return r->end - r->start;
> +}
> +
> +static void swap_regions(struct region *r1, struct region *r2)
> +{
> +	struct region tmp;
> +
> +	tmp = *r1;
> +	*r1 = *r2;
> +	*r2 = tmp;
> +}
> +
> +/*
> + * Find the three regions in an address space
> + *
> + * vma		the head vma of the target address space
> + * regions	an array of three 'struct region's that results will be saved
> + *
> + * This function receives an address space and finds three regions in it which
> + * separated by the two biggest unmapped regions in the space.  Please refer to
> + * below comments of 'damon_init_regions_of()' function to know why this is
> + * necessary.
> + *
> + * Returns 0 if success, or negative error code otherwise.
> + */
> +static int damon_three_regions_in_vmas(struct vm_area_struct *vma,
> +		struct region regions[3])
> +{
> +	struct region gap = {0,}, first_gap = {0,}, second_gap = {0,};
> +	struct vm_area_struct *last_vma = NULL;
> +	unsigned long start = 0;
> +
> +	/* Find two biggest gaps so that first_gap > second_gap > others */
> +	for (; vma; vma = vma->vm_next) {
> +		if (!last_vma) {
> +			start = vma->vm_start;
> +			last_vma = vma;
> +			continue;
> +		}
> +		gap.start = last_vma->vm_end;
> +		gap.end = vma->vm_start;
> +		if (sz_region(&gap) > sz_region(&second_gap)) {
> +			swap_regions(&gap, &second_gap);
> +			if (sz_region(&second_gap) > sz_region(&first_gap))
> +				swap_regions(&second_gap, &first_gap);
> +		}
> +		last_vma = vma;
> +	}
> +
> +	if (!sz_region(&second_gap) || !sz_region(&first_gap))
> +		return -EINVAL;
> +
> +	/* Sort the two biggest gaps by address */
> +	if (first_gap.start > second_gap.start)
> +		swap_regions(&first_gap, &second_gap);
> +
> +	/* Store the result */
> +	regions[0].start = start;
> +	regions[0].end = first_gap.start;
> +	regions[1].start = first_gap.end;
> +	regions[1].end = second_gap.start;
> +	regions[2].start = second_gap.end;
> +	regions[2].end = last_vma->vm_end;
> +
> +	return 0;
> +}
> +
> +/*
> + * Get the three regions in the given task
> + *
> + * Returns 0 on success, negative error code otherwise.
> + */
> +static int damon_three_regions_of(struct damon_task *t,
> +				struct region regions[3])
> +{
> +	struct mm_struct *mm;
> +	int rc;
> +
> +	mm = damon_get_mm(t);
> +	if (!mm)
> +		return -EINVAL;
> +
> +	down_read(&mm->mmap_sem);
> +	rc = damon_three_regions_in_vmas(mm->mmap, regions);
> +	up_read(&mm->mmap_sem);
> +
> +	mmput(mm);
> +	return rc;
> +}
> +
> +/*
> + * Initialize the monitoring target regions for the given task
> + *
> + * t	the given target task
> + *
> + * Because only a number of small portions of the entire address space
> + * is acutally mapped to the memory and accessed, monitoring the unmapped
> + * regions is wasteful.  That said, because we can deal with small noises,
> + * tracking every mapping is not strictly required but could even incur a high
> + * overhead if the mapping frequently changes or the number of mappings is
> + * high.  Nonetheless, this may seems very weird.  DAMON's dynamic regions
> + * adjustment mechanism, which will be implemented with following commit will
> + * make this more sense.
> + *
> + * For the reason, we convert the complex mappings to three distinct regions
> + * that cover every mapped areas of the address space.  Also the two gaps
> + * between the three regions are the two biggest unmapped areas in the given
> + * address space.  In detail, this function first identifies the start and the
> + * end of the mappings and the two biggest unmapped areas of the address space.
> + * Then, it constructs the three regions as below:
> + *
> + *     [mappings[0]->start, big_two_unmapped_areas[0]->start)
> + *     [big_two_unmapped_areas[0]->end, big_two_unmapped_areas[1]->start)
> + *     [big_two_unmapped_areas[1]->end, mappings[nr_mappings - 1]->end)
> + *
> + * As usual memory map of processes is as below, the gap between the heap and
> + * the uppermost mmap()-ed region, and the gap between the lowermost mmap()-ed
> + * region and the stack will be two biggest unmapped regions.  Because these
> + * gaps are exceptionally huge areas in usual address space, excluding these
> + * two biggest unmapped regions will be sufficient to make a trade-off.
> + *
> + *   <heap>
> + *   <BIG UNMAPPED REGION 1>
> + *   <uppermost mmap()-ed region>
> + *   (other mmap()-ed regions and small unmapped regions)
> + *   <lowermost mmap()-ed region>
> + *   <BIG UNMAPPED REGION 2>
> + *   <stack>
> + */
> +static void damon_init_regions_of(struct damon_ctx *c, struct damon_task *t)
> +{
> +	struct damon_region *r;
> +	struct region regions[3];
> +	int i;
> +
> +	if (damon_three_regions_of(t, regions)) {
> +		pr_err("Failed to get three regions of task %lu\n", t->pid);
> +		return;
> +	}
> +
> +	/* Set the initial three regions of the task */
> +	for (i = 0; i < 3; i++) {
> +		r = damon_new_region(c, regions[i].start, regions[i].end);
> +		damon_add_region(r, t);
> +	}
> +
> +	/* Split the middle region into 'min_nr_regions - 2' regions */
> +	r = damon_nth_region_of(t, 1);
> +	if (damon_split_region_evenly(c, r, c->min_nr_regions - 2))
> +		pr_warn("Init middle region failed to be split\n");
> +}
> +
> +/* Initialize '->regions_list' of every task */
> +static void kdamond_init_regions(struct damon_ctx *ctx)
> +{
> +	struct damon_task *t;
> +
> +	damon_for_each_task(ctx, t)
> +		damon_init_regions_of(ctx, t);
> +}
> +
> +static bool damon_pte_pmd_young(pte_t *pte, pmd_t *pmd)
> +{
> +	if (pte && pte_young(*pte))
> +		return true;
> +#ifdef CONFIG_TRANSPARENT_HUGEPAGE
> +	if (pmd && pmd_young(*pmd))
> +		return true;
> +#endif /* CONFIG_TRANSPARENT_HUGEPAGE */
> +	return false;
> +}
> +
> +static void damon_pte_pmd_mkold(pte_t *pte, pmd_t *pmd)
> +{
> +	if (pte) {
> +		if (pte_young(*pte)) {
> +			clear_page_idle(pte_page(*pte));
> +			set_page_young(pte_page(*pte));
> +		}
> +		*pte = pte_mkold(*pte);
> +		return;
> +	}
> +#ifdef CONFIG_TRANSPARENT_HUGEPAGE
> +	if (pmd) {
> +		if (pmd_young(*pmd)) {
> +			clear_page_idle(pmd_page(*pmd));
> +			set_page_young(pmd_page(*pmd));
> +		}
> +		*pmd = pmd_mkold(*pmd);
> +	}
> +#endif /* CONFIG_TRANSPARENT_HUGEPAGE */

No need to flush the TLBs?

> +}
> +
> +/*
> + * Check whether the region accessed and prepare for next check
> + *
> + * mm	'mm_struct' for the given virtual address space
> + * r	the region to be checked
> + */
> +static void kdamond_check_access(struct damon_ctx *ctx,
> +			struct mm_struct *mm, struct damon_region *r)
> +{
> +	static struct mm_struct *last_mm;
> +	static unsigned long last_addr;
> +	static int last_page_sz = PAGE_SIZE;
> +	static bool last_accessed;
> +
> +	pte_t *pte = NULL;
> +	pmd_t *pmd = NULL;
> +	spinlock_t *ptl;
> +
> +	/* If the region is in the last checked page, reuse the result */
> +	if (mm == last_mm && (ALIGN_DOWN(last_addr, last_page_sz) ==
> +				ALIGN_DOWN(r->sampling_addr, last_page_sz))) {
> +		if (last_accessed)
> +			r->nr_accesses++;
> +		return;
> +	}
> +
> +	if (follow_pte_pmd(mm, r->sampling_addr, NULL, &pte, &pmd, &ptl))
> +		goto prepare_next_check;
> +
> +	/* Read the page table access bit of the page */
> +	if (damon_pte_pmd_young(pte, pmd)) {
> +		last_accessed = true;
> +		r->nr_accesses++;
> +	}
> +	spin_unlock(ptl);
> +
> +prepare_next_check:
> +	last_mm = mm;
> +	last_addr = r->sampling_addr;
> +#ifdef CONFIG_TRANSPARENT_HUGEPAGE
> +	last_page_sz = pte ? PAGE_SIZE : ((1UL) << HPAGE_PMD_SHIFT);
> +#endif
> +
> +	r->sampling_addr = damon_rand(ctx, r->vm_start, r->vm_end);
> +	pte = NULL, pmd = NULL;
> +	if (follow_pte_pmd(mm, r->sampling_addr, NULL, &pte, &pmd, &ptl))
> +		return;
> +
> +	damon_pte_pmd_mkold(pte, pmd);
> +	spin_unlock(ptl);
> +}
> +
> +/*
> + * damon_check_reset_time_interval() - Check if a time interval is elapsed.
> + * @baseline:	the time to check whether the interval has elapsed since
> + * @interval:	the time interval (microseconds)
> + *
> + * See whether the given time interval has passed since the given baseline
> + * time.  If so, it also updates the baseline to current time for next check.
> + *
> + * Return:	true if the time interval has passed, or false otherwise.
> + */
> +static bool damon_check_reset_time_interval(struct timespec64 *baseline,
> +		unsigned long interval)
> +{
> +	struct timespec64 now;
> +
> +	ktime_get_coarse_ts64(&now);
> +	if ((timespec64_to_ns(&now) - timespec64_to_ns(baseline)) <
> +			interval * 1000)
> +		return false;
> +	*baseline = now;
> +	return true;
> +}
> +
> +/*
> + * Check whether it is time to flush the aggregated information
> + */
> +static bool kdamond_aggregate_interval_passed(struct damon_ctx *ctx)
> +{
> +	return damon_check_reset_time_interval(&ctx->last_aggregation,
> +			ctx->aggr_interval);
> +}
> +
> +/*
> + * Reset the aggregated monitoring results
> + */
> +static void kdamond_reset_aggregated(struct damon_ctx *c)
> +{
> +	struct damon_task *t;
> +	struct damon_region *r;
> +
> +	damon_for_each_task(c, t) {
> +		damon_for_each_region(r, t)
> +			r->nr_accesses = 0;
> +	}
> +}
> +
> +/*
> + * Check whether current monitoring should be stopped
> + *
> + * If users asked to stop, need stop.  Even though no user has asked to stop,
> + * need stop if every target task has dead.
> + *
> + * Returns true if need to stop current monitoring.
> + */
> +static bool kdamond_need_stop(struct damon_ctx *ctx)
> +{
> +	struct damon_task *t;
> +	struct task_struct *task;
> +	bool stop;
> +
> +	stop = kthread_should_stop();
> +	if (stop)
> +		return true;
> +
> +	damon_for_each_task(ctx, t) {
> +		task = damon_get_task_struct(t);
> +		if (task) {
> +			put_task_struct(task);
> +			return false;
> +		}
> +	}
> +
> +	return true;
> +}
> +
> +/*
> + * The monitoring daemon that runs as a kernel thread
> + */
> +static int kdamond_fn(void *data)
> +{
> +	struct damon_ctx *ctx = data;
> +	struct damon_task *t;
> +	struct damon_region *r, *next;
> +	struct mm_struct *mm;
> +
> +	pr_info("kdamond (%d) starts\n", ctx->kdamond->pid);
> +	kdamond_init_regions(ctx);

We haven't called mkold on the initial regions so first check will
get us fairly random state.

> +	while (!kdamond_need_stop(ctx)) {
> +		damon_for_each_task(ctx, t) {
> +			mm = damon_get_mm(t);
> +			if (!mm)
> +				continue;
> +			damon_for_each_region(r, t)
> +				kdamond_check_access(ctx, mm, r);
> +			mmput(mm);
> +		}
> +
> +		if (kdamond_aggregate_interval_passed(ctx))
> +			kdamond_reset_aggregated(ctx);
> +
> +		usleep_range(ctx->sample_interval, ctx->sample_interval + 1);
> +	}
> +	damon_for_each_task(ctx, t) {
> +		damon_for_each_region_safe(r, next, t)
> +			damon_destroy_region(r);
> +	}
> +	pr_debug("kdamond (%d) finishes\n", ctx->kdamond->pid);
> +	mutex_lock(&ctx->kdamond_lock);
> +	ctx->kdamond = NULL;
> +	mutex_unlock(&ctx->kdamond_lock);
> +
> +	return 0;
> +}
> +
> +/*
> + * Controller functions
> + */
> +
> +static bool damon_kdamond_running(struct damon_ctx *ctx)
> +{
> +	bool running;
> +
> +	mutex_lock(&ctx->kdamond_lock);
> +	running = ctx->kdamond != NULL;
> +	mutex_unlock(&ctx->kdamond_lock);
> +
> +	return running;
> +}
> +
> +/*
> + * Start or stop the kdamond
> + *
> + * Returns 0 if success, negative error code otherwise.
> + */
> +static int damon_turn_kdamond(struct damon_ctx *ctx, bool on)
> +{
> +	int err = -EBUSY;
> +
> +	mutex_lock(&ctx->kdamond_lock);
> +	if (!ctx->kdamond && on) {

Given there is very little shared code between on and off, I would
suggest just splitting it into two functions.

> +		err = 0;
> +		ctx->kdamond = kthread_run(kdamond_fn, ctx, "kdamond");
> +		if (IS_ERR(ctx->kdamond))
> +			err = PTR_ERR(ctx->kdamond);
> +	} else if (ctx->kdamond && !on) {
> +		mutex_unlock(&ctx->kdamond_lock);
> +		kthread_stop(ctx->kdamond);
> +		while (damon_kdamond_running(ctx))
> +			usleep_range(ctx->sample_interval,
> +					ctx->sample_interval * 2);
> +		return 0;
> +	}
> +	mutex_unlock(&ctx->kdamond_lock);
> +
> +	return err;
> +}
> +
> +/*
> + * damon_start() - Starts monitoring with given context.
> + * @ctx:	monitoring context
> + *
> + * Return: 0 on success, negative error code otherwise.
> + */
> +int damon_start(struct damon_ctx *ctx)
> +{
> +	return damon_turn_kdamond(ctx, true);
> +}
> +
> +/*
> + * damon_stop() - Stops monitoring of given context.
> + * @ctx:	monitoring context
> + *
> + * Return: 0 on success, negative error code otherwise.
> + */
> +int damon_stop(struct damon_ctx *ctx)
> +{
> +	return damon_turn_kdamond(ctx, false);
> +}
> +
> +/*
> + * damon_set_pids() - Set monitoring target processes.
> + * @ctx:	monitoring context
> + * @pids:	array of target processes pids
> + * @nr_pids:	number of entries in @pids
> + *
> + * This function should not be called while the kdamond is running.
> + *
> + * Return: 0 on usccess, negative error code otherwise.
> + */
> +int damon_set_pids(struct damon_ctx *ctx, unsigned long *pids, ssize_t nr_pids)
> +{
> +	ssize_t i;
> +	struct damon_task *t, *next;
> +
> +	damon_for_each_task_safe(ctx, t, next)
> +		damon_destroy_task(t);
> +
> +	for (i = 0; i < nr_pids; i++) {
> +		t = damon_new_task(pids[i]);
> +		if (!t) {
> +			pr_err("Failed to alloc damon_task\n");
> +			return -ENOMEM;
> +		}
> +		damon_add_task(ctx, t);
> +	}
> +
> +	return 0;
> +}
> +
> +/*

Why not make these actual kernel-doc?  That way you can use the
kernel-doc scripts to sanity check them.

/**

> + * damon_set_attrs() - Set attributes for the monitoring.
> + * @ctx:		monitoring context
> + * @sample_int:		time interval between samplings
> + * @aggr_int:		time interval between aggregations
> + * @min_nr_reg:		minimal number of regions
> + *
> + * This function should not be called while the kdamond is running.
> + * Every time interval is in micro-seconds.
> + *
> + * Return: 0 on success, negative error code otherwise.
> + */
> +int damon_set_attrs(struct damon_ctx *ctx, unsigned long sample_int,
> +		unsigned long aggr_int, unsigned long min_nr_reg)
> +{
> +	if (min_nr_reg < 3) {
> +		pr_err("min_nr_regions (%lu) should be bigger than 2\n",
> +				min_nr_reg);
> +		return -EINVAL;
> +	}
> +
> +	ctx->sample_interval = sample_int;
> +	ctx->aggr_interval = aggr_int;
> +	ctx->min_nr_regions = min_nr_reg;
> +
> +	return 0;
> +}
> +
>  static int __init damon_init(void)
>  {
>  	return 0;