[RFC PATCH 0/2] DAMON: Improve User Space Interface

[RFC PATCH 0/2] DAMON: Improve User Space Interface

[SeongJae Park]

From: SeongJae Park <sjpark@amazon.de>

NOTE: This is only an RFC for future features of DAMON patchset[1], which is
not merged in the mainline yet.  The aim of this RFC is to show how DAMON would
be evolved once it is merged in.  So, if you have some interest in this RFC,
please consider reviewing the DAMON patchset, either.

After posting DAMON patchset[1], we received a number of comments.  Based on
those, we listed and shared future works for DAMON in the kernel summit
2020[2] and had a poll for the priorities of the works.  As a result, the user
space interface improvement received a second highest priority[3].  For the
reason, this patchset is came out.

The 1st patch puts more information in the monitoring thread name so that user
space could charge the DAMON's CPU usage on them by themselves, in fine
granularity.  The 2nd patch makes multiple monitoring contexts available using
the debugfs interface.

[1] https://lore.kernel.org/linux-mm/20200817105137.19296-1-sjpark@amazon.com/
[2] https://linuxplumbersconf.org/event/7/contributions/659/
[3] https://lore.kernel.org/linux-mm/20200831112235.2675-1-sjpark@amazon.com/

Baseline and Complete Git Trees
===============================

The patches are based on the v5.8 plus DAMON v20 patchset[1], RFC v14 of DAMOS
patchset, RFC v8 of physical address space support patchset, and some more
trivial fixes (s/snprintf/scnprintf).  You can also clone the complete git
tree:

    $ git clone git://github.com/sjp38/linux -b damon-usi/rfc/v1

The web is also available:
https://github.com/sjp38/linux/releases/tag/damon-usi/rfc/v1

[1] https://lore.kernel.org/linux-mm/20200817105137.19296-1-sjpark@amazon.com/
[2] https://lore.kernel.org/linux-mm/20200804142430.15384-1-sjpark@amazon.com/
[3] https://lore.kernel.org/linux-mm/20200831104730.28970-1-sjpark@amazon.com/

SeongJae Park (2):
  mm/damon: Put more information in kdamond thread name
  mm/damon/debugfs: Support multiple contexts

 include/linux/damon.h |   2 +
 mm/damon-test.h       |  34 ++--
 mm/damon.c            | 361 +++++++++++++++++++++++++++++++++---------
 3 files changed, 308 insertions(+), 89 deletions(-)

-- 
2.17.1

[RFC PATCH 1/2] mm/damon: Put more information in kdamond thread name

[SeongJae Park]

From: SeongJae Park <sjpark@amazon.de>

While multiple kdamond threads can run, it is hard to distinguish each
thread as all kdamond threads has same name, 'kdamond'.  This commit
puts more information including if it is spawned from the user context
(via the debugfs interface) and for what context it is running.  The
format of the name is:

    kdamond.<current pid>.<# of context>

If it is not spawned from the user context, <current pid> becomes 0.
Using this information, for example, the user space could easily account
CPU usage of DAMON that they requested in fine granularity.

Signed-off-by: SeongJae Park <sjpark@amazon.de>
---
 mm/damon.c | 5 ++++-
 1 file changed, 4 insertions(+), 1 deletion(-)

diff --git a/mm/damon.c b/mm/damon.c
index c92e4c76aa53..abb8c07a4c7d 100644
--- a/mm/damon.c
+++ b/mm/damon.c
@@ -1552,6 +1552,8 @@ static bool damon_kdamond_running(struct damon_ctx *ctx)
  * __damon_start() - Starts monitoring with given context.
  * @ctx:	monitoring context
  *
+ * This function should be called while damon_lock is hold.
+ *
  * Return: 0 on success, negative error code otherwise.
  */
 static int __damon_start(struct damon_ctx *ctx)
@@ -1562,7 +1564,8 @@ static int __damon_start(struct damon_ctx *ctx)
 	if (!ctx->kdamond) {
 		err = 0;
 		ctx->kdamond_stop = false;
-		ctx->kdamond = kthread_create(kdamond_fn, ctx, "kdamond");
+		ctx->kdamond = kthread_create(kdamond_fn, ctx, "kdamond.%d.%d",
+				current ? current->pid : 0, nr_running_ctxs);
 		if (IS_ERR(ctx->kdamond))
 			err = PTR_ERR(ctx->kdamond);
 		else
-- 
2.17.1

[RFC PATCH 2/2] mm/damon/debugfs: Support multiple contexts

[SeongJae Park]

From: SeongJae Park <sjpark@amazon.de>

DAMON allows the programming interface users to run monitoring with
multiple contexts.  This could be useful in some cases.  For example, if
someone want to do highly accurate monitoring and lots of CPUs are
available, splitting the monitoring target regions into multiple small
regions and allocating context (monitoring thread) to each small region
could be helpful.  Or, someone could need to monitor different types of
address spaces simultaneously.

However, it's impossible from the user space because the DAMON debugfs
interface supports only single context.  This commit makes it available
by implementing 'nr_contexts' debugfs file.

Users can pass the number (N) of contexts they want to use to the file.
Then, N folders having name of 'ctx<1-(N-1)>' are created in the DAMON
debugfs dir.  Each of the directory is associated with the contexts and
contains the the files for context setting (attrs, init_regions, record,
schemes, and target_ids).  The first context related files are still in
the DAMON debugfs root directory, though.

Signed-off-by: SeongJae Park <sjpark@amazon.de>
---
 include/linux/damon.h |   2 +
 mm/damon-test.h       |  34 ++--
 mm/damon.c            | 356 +++++++++++++++++++++++++++++++++---------
 3 files changed, 304 insertions(+), 88 deletions(-)

diff --git a/include/linux/damon.h b/include/linux/damon.h
index 306fa221ceae..be391e7df9cf 100644
--- a/include/linux/damon.h
+++ b/include/linux/damon.h
@@ -259,5 +259,7 @@ int damon_set_recording(struct damon_ctx *ctx,
 				unsigned int rbuf_len, char *rfile_path);
 int damon_start(struct damon_ctx *ctxs, int nr_ctxs);
 int damon_stop(struct damon_ctx *ctxs, int nr_ctxs);
+int damon_start_ctx_ptrs(struct damon_ctx **ctxs, int nr_ctxs);
+int damon_stop_ctx_ptrs(struct damon_ctx **ctxs, int nr_ctxs);
 
 #endif
diff --git a/mm/damon-test.h b/mm/damon-test.h
index e67e8fb17eca..681adead0339 100644
--- a/mm/damon-test.h
+++ b/mm/damon-test.h
@@ -102,14 +102,14 @@ static void damon_test_regions(struct kunit *test)
 
 static void damon_test_target(struct kunit *test)
 {
-	struct damon_ctx *c = &damon_user_ctx;
+	struct damon_ctx *c = debugfs_ctxs[0];
 	struct damon_target *t;
 
 	t = damon_new_target(42);
 	KUNIT_EXPECT_EQ(test, 42ul, t->id);
 	KUNIT_EXPECT_EQ(test, 0u, nr_damon_targets(c));
 
-	damon_add_target(&damon_user_ctx, t);
+	damon_add_target(c, t);
 	KUNIT_EXPECT_EQ(test, 1u, nr_damon_targets(c));
 
 	damon_destroy_target(t);
@@ -118,7 +118,7 @@ static void damon_test_target(struct kunit *test)
 
 static void damon_test_set_targets(struct kunit *test)
 {
-	struct damon_ctx *ctx = &damon_user_ctx;
+	struct damon_ctx *ctx = debugfs_ctxs[0];
 	unsigned long ids[] = {1, 2, 3};
 	char buf[64];
 
@@ -148,7 +148,7 @@ static void damon_test_set_targets(struct kunit *test)
 
 static void damon_test_set_recording(struct kunit *test)
 {
-	struct damon_ctx *ctx = &damon_user_ctx;
+	struct damon_ctx *ctx = debugfs_ctxs[0];
 	int err;
 
 	err = damon_set_recording(ctx, 42, "foo");
@@ -163,7 +163,7 @@ static void damon_test_set_recording(struct kunit *test)
 
 static void damon_test_set_init_regions(struct kunit *test)
 {
-	struct damon_ctx *ctx = &damon_user_ctx;
+	struct damon_ctx *ctx = debugfs_ctxs[0];
 	unsigned long ids[] = {1, 2, 3};
 	/* Each line represents one region in ``<target id> <start> <end>`` */
 	char * const valid_inputs[] = {"2 10 20\n 2   20 30\n2 35 45",
@@ -299,10 +299,10 @@ static void damon_cleanup_global_state(void)
 {
 	struct damon_target *t, *next;
 
-	damon_for_each_target_safe(t, next, &damon_user_ctx)
+	damon_for_each_target_safe(t, next, debugfs_ctxs[0])
 		damon_destroy_target(t);
 
-	damon_user_ctx.rbuf_offset = 0;
+	debugfs_ctxs[0]->rbuf_offset = 0;
 }
 
 /*
@@ -317,7 +317,7 @@ static void damon_cleanup_global_state(void)
  */
 static void damon_test_aggregate(struct kunit *test)
 {
-	struct damon_ctx *ctx = &damon_user_ctx;
+	struct damon_ctx *ctx = debugfs_ctxs[0];
 	unsigned long target_ids[] = {1, 2, 3};
 	unsigned long saddr[][3] = {{10, 20, 30}, {5, 42, 49}, {13, 33, 55} };
 	unsigned long eaddr[][3] = {{15, 27, 40}, {31, 45, 55}, {23, 44, 66} };
@@ -367,10 +367,10 @@ static void damon_test_aggregate(struct kunit *test)
 
 static void damon_test_write_rbuf(struct kunit *test)
 {
-	struct damon_ctx *ctx = &damon_user_ctx;
+	struct damon_ctx *ctx = debugfs_ctxs[0];
 	char *data;
 
-	damon_set_recording(&damon_user_ctx, 4242, "damon.data");
+	damon_set_recording(debugfs_ctxs[0], 4242, "damon.data");
 
 	data = "hello";
 	damon_write_rbuf(ctx, data, strnlen(data, 256));
@@ -380,7 +380,7 @@ static void damon_test_write_rbuf(struct kunit *test)
 	KUNIT_EXPECT_EQ(test, ctx->rbuf_offset, 5u);
 
 	KUNIT_EXPECT_STREQ(test, (char *)ctx->rbuf, data);
-	damon_set_recording(&damon_user_ctx, 0, "damon.data");
+	damon_set_recording(debugfs_ctxs[0], 0, "damon.data");
 }
 
 static struct damon_region *__nth_region_of(struct damon_target *t, int idx)
@@ -432,9 +432,9 @@ static void damon_do_test_apply_three_regions(struct kunit *test,
 		r = damon_new_region(regions[i * 2], regions[i * 2 + 1]);
 		damon_add_region(r, t);
 	}
-	damon_add_target(&damon_user_ctx, t);
+	damon_add_target(debugfs_ctxs[0], t);
 
-	damon_apply_three_regions(&damon_user_ctx, t, three_regions);
+	damon_apply_three_regions(debugfs_ctxs[0], t, three_regions);
 
 	for (i = 0; i < nr_expected / 2; i++) {
 		r = __nth_region_of(t, i);
@@ -541,7 +541,7 @@ static void damon_test_apply_three_regions4(struct kunit *test)
 
 static void damon_test_split_evenly(struct kunit *test)
 {
-	struct damon_ctx *c = &damon_user_ctx;
+	struct damon_ctx *c = debugfs_ctxs[0];
 	struct damon_target *t;
 	struct damon_region *r;
 	unsigned long i;
@@ -601,7 +601,7 @@ static void damon_test_split_at(struct kunit *test)
 	t = damon_new_target(42);
 	r = damon_new_region(0, 100);
 	damon_add_region(r, t);
-	damon_split_region_at(&damon_user_ctx, r, 25);
+	damon_split_region_at(debugfs_ctxs[0], r, 25);
 	KUNIT_EXPECT_EQ(test, r->ar.start, 0ul);
 	KUNIT_EXPECT_EQ(test, r->ar.end, 25ul);
 
@@ -679,14 +679,14 @@ static void damon_test_split_regions_of(struct kunit *test)
 	t = damon_new_target(42);
 	r = damon_new_region(0, 22);
 	damon_add_region(r, t);
-	damon_split_regions_of(&damon_user_ctx, t, 2);
+	damon_split_regions_of(debugfs_ctxs[0], t, 2);
 	KUNIT_EXPECT_EQ(test, nr_damon_regions(t), 2u);
 	damon_free_target(t);
 
 	t = damon_new_target(42);
 	r = damon_new_region(0, 220);
 	damon_add_region(r, t);
-	damon_split_regions_of(&damon_user_ctx, t, 4);
+	damon_split_regions_of(debugfs_ctxs[0], t, 4);
 	KUNIT_EXPECT_EQ(test, nr_damon_regions(t), 4u);
 	damon_free_target(t);
 }
diff --git a/mm/damon.c b/mm/damon.c
index abb8c07a4c7d..053a4d160b83 100644
--- a/mm/damon.c
+++ b/mm/damon.c
@@ -86,22 +86,6 @@
 static DEFINE_MUTEX(damon_lock);
 static int nr_running_ctxs;
 
-/* A monitoring context for debugfs interface users. */
-static struct damon_ctx damon_user_ctx = {
-	.sample_interval = 5 * 1000,
-	.aggr_interval = 100 * 1000,
-	.regions_update_interval = 1000 * 1000,
-	.min_nr_regions = 10,
-	.max_nr_regions = 1000,
-
-	.init_target_regions = kdamond_init_vm_regions,
-	.update_target_regions = kdamond_update_vm_regions,
-	.prepare_access_checks = kdamond_prepare_vm_access_checks,
-	.check_accesses = kdamond_check_vm_accesses,
-	.target_valid = kdamond_vm_target_valid,
-	.cleanup = kdamond_vm_cleanup,
-};
-
 /*
  * Construct a damon_region struct
  *
@@ -247,6 +231,72 @@ static void damon_destroy_scheme(struct damos *s)
 	damon_free_scheme(s);
 }
 
+static void damon_set_vaddr_primitives(struct damon_ctx *ctx)
+{
+	ctx->init_target_regions = kdamond_init_vm_regions;
+	ctx->update_target_regions = kdamond_update_vm_regions;
+	ctx->prepare_access_checks = kdamond_prepare_vm_access_checks;
+	ctx->check_accesses = kdamond_check_vm_accesses;
+	ctx->target_valid = kdamond_vm_target_valid;
+	ctx->cleanup = kdamond_vm_cleanup;
+}
+
+static void damon_set_paddr_primitives(struct damon_ctx *ctx)
+{
+	ctx->init_target_regions = kdamond_init_phys_regions;
+	ctx->update_target_regions = kdamond_update_phys_regions;
+	ctx->prepare_access_checks = kdamond_prepare_phys_access_checks;
+	ctx->check_accesses = kdamond_check_phys_accesses;
+	ctx->target_valid = NULL;
+	ctx->cleanup = NULL;
+}
+
+static struct damon_ctx *damon_new_ctx(void)
+{
+	struct damon_ctx *ctx;
+
+	ctx = kzalloc(sizeof(*ctx), GFP_KERNEL);
+	if (!ctx)
+		return NULL;
+
+	ctx->sample_interval = 5 * 1000;
+	ctx->aggr_interval = 100 * 1000;
+	ctx->regions_update_interval = 1000 * 1000;
+	ctx->min_nr_regions = 10;
+	ctx->max_nr_regions = 1000;
+
+	damon_set_vaddr_primitives(ctx);
+
+	ktime_get_coarse_ts64(&ctx->last_aggregation);
+	ctx->last_regions_update = ctx->last_aggregation;
+
+	if (damon_set_recording(ctx, 0, "none")) {
+		kfree(ctx);
+		return NULL;
+	}
+
+	mutex_init(&ctx->kdamond_lock);
+
+	INIT_LIST_HEAD(&ctx->targets_list);
+	INIT_LIST_HEAD(&ctx->schemes_list);
+
+	return ctx;
+}
+
+static void damon_destroy_ctx(struct damon_ctx *ctx)
+{
+	struct damon_target *t, *next_t;
+	struct damos *s, *next_s;
+
+	damon_for_each_target_safe(t, next_t, ctx)
+		damon_destroy_target(t);
+
+	damon_for_each_scheme_safe(s, next_s, ctx)
+		damon_destroy_scheme(s);
+
+	kfree(ctx);
+}
+
 static unsigned int nr_damon_targets(struct damon_ctx *ctx)
 {
 	struct damon_target *t;
@@ -1611,6 +1661,28 @@ int damon_start(struct damon_ctx *ctxs, int nr_ctxs)
 	return err;
 }
 
+int damon_start_ctx_ptrs(struct damon_ctx **ctxs, int nr_ctxs)
+{
+	int i;
+	int err = 0;
+
+	mutex_lock(&damon_lock);
+	if (nr_running_ctxs) {
+		mutex_unlock(&damon_lock);
+		return -EBUSY;
+	}
+
+	for (i = 0; i < nr_ctxs; i++) {
+		err = __damon_start(ctxs[i]);
+		if (err)
+			break;
+		nr_running_ctxs++;
+	}
+	mutex_unlock(&damon_lock);
+
+	return err;
+}
+
 /*
  * __damon_stop() - Stops monitoring of given context.
  * @ctx:	monitoring context
@@ -1654,6 +1726,20 @@ int damon_stop(struct damon_ctx *ctxs, int nr_ctxs)
 	return err;
 }
 
+int damon_stop_ctx_ptrs(struct damon_ctx **ctxs, int nr_ctxs)
+{
+	int i, err = 0;
+
+	for (i = 0; i < nr_ctxs; i++) {
+		/* nr_running_ctxs is decremented in kdamond_fn */
+		err = __damon_stop(ctxs[i]);
+		if (err)
+			return err;
+	}
+
+	return err;
+}
+
 /**
  * damon_set_schemes() - Set data access monitoring based operation schemes.
  * @ctx:	monitoring context
@@ -1799,15 +1885,21 @@ int damon_set_attrs(struct damon_ctx *ctx, unsigned long sample_int,
  * Functions for the DAMON debugfs interface
  */
 
+/* Monitoring contexts for debugfs interface users. */
+static struct damon_ctx **debugfs_ctxs;
+static int debugfs_nr_ctxs = 1;
+
 static ssize_t debugfs_monitor_on_read(struct file *file,
 		char __user *buf, size_t count, loff_t *ppos)
 {
-	struct damon_ctx *ctx = &damon_user_ctx;
 	char monitor_on_buf[5];
 	bool monitor_on;
 	int len;
 
-	monitor_on = damon_kdamond_running(ctx);
+	mutex_lock(&damon_lock);
+	monitor_on = nr_running_ctxs != 0;
+	mutex_unlock(&damon_lock);
+
 	len = scnprintf(monitor_on_buf, 5, monitor_on ? "on\n" : "off\n");
 
 	return simple_read_from_buffer(buf, count, ppos, monitor_on_buf, len);
@@ -1842,7 +1934,6 @@ static char *user_input_str(const char __user *buf, size_t count, loff_t *ppos)
 static ssize_t debugfs_monitor_on_write(struct file *file,
 		const char __user *buf, size_t count, loff_t *ppos)
 {
-	struct damon_ctx *ctx = &damon_user_ctx;
 	ssize_t ret = count;
 	char *kbuf;
 	int err;
@@ -1855,9 +1946,9 @@ static ssize_t debugfs_monitor_on_write(struct file *file,
 	if (sscanf(kbuf, "%s", kbuf) != 1)
 		return -EINVAL;
 	if (!strncmp(kbuf, "on", count))
-		err = damon_start(ctx, 1);
+		err = damon_start_ctx_ptrs(debugfs_ctxs, debugfs_nr_ctxs);
 	else if (!strncmp(kbuf, "off", count))
-		err = damon_stop(ctx, 1);
+		err = damon_stop_ctx_ptrs(debugfs_ctxs, debugfs_nr_ctxs);
 	else
 		return -EINVAL;
 
@@ -1890,7 +1981,7 @@ static ssize_t sprint_schemes(struct damon_ctx *c, char *buf, ssize_t len)
 static ssize_t debugfs_schemes_read(struct file *file, char __user *buf,
 		size_t count, loff_t *ppos)
 {
-	struct damon_ctx *ctx = &damon_user_ctx;
+	struct damon_ctx *ctx = file->private_data;
 	char *kbuf;
 	ssize_t len;
 
@@ -1985,7 +2076,7 @@ static struct damos **str_to_schemes(const char *str, ssize_t len,
 static ssize_t debugfs_schemes_write(struct file *file, const char __user *buf,
 		size_t count, loff_t *ppos)
 {
-	struct damon_ctx *ctx = &damon_user_ctx;
+	struct damon_ctx *ctx = file->private_data;
 	char *kbuf;
 	struct damos **schemes;
 	ssize_t nr_schemes = 0, ret = count;
@@ -2050,7 +2141,7 @@ static ssize_t sprint_target_ids(struct damon_ctx *ctx, char *buf, ssize_t len)
 static ssize_t debugfs_target_ids_read(struct file *file,
 		char __user *buf, size_t count, loff_t *ppos)
 {
-	struct damon_ctx *ctx = &damon_user_ctx;
+	struct damon_ctx *ctx = file->private_data;
 	ssize_t len;
 	char ids_buf[320];
 
@@ -2116,7 +2207,7 @@ static struct pid *damon_get_pidfd_pid(unsigned int pidfd)
 static ssize_t debugfs_target_ids_write(struct file *file,
 		const char __user *buf, size_t count, loff_t *ppos)
 {
-	struct damon_ctx *ctx = &damon_user_ctx;
+	struct damon_ctx *ctx = file->private_data;
 	char *kbuf, *nrs;
 	bool received_pidfds = false;
 	unsigned long *targets;
@@ -2132,23 +2223,12 @@ static ssize_t debugfs_target_ids_write(struct file *file,
 	nrs = kbuf;
 	if (!strncmp(kbuf, "paddr\n", count)) {
 		/* Configure the context for physical memory monitoring */
-		ctx->init_target_regions = kdamond_init_phys_regions;
-		ctx->update_target_regions = kdamond_update_phys_regions;
-		ctx->prepare_access_checks = kdamond_prepare_phys_access_checks;
-		ctx->check_accesses = kdamond_check_phys_accesses;
-		ctx->target_valid = NULL;
-		ctx->cleanup = NULL;
-
+		damon_set_paddr_primitives(ctx);
 		/* target id is meaningless here, but we set it just for fun */
 		scnprintf(kbuf, count, "42    ");
 	} else {
 		/* Configure the context for virtual memory monitoring */
-		ctx->init_target_regions = kdamond_init_vm_regions;
-		ctx->update_target_regions = kdamond_update_vm_regions;
-		ctx->prepare_access_checks = kdamond_prepare_vm_access_checks;
-		ctx->check_accesses = kdamond_check_vm_accesses;
-		ctx->target_valid = kdamond_vm_target_valid;
-		ctx->cleanup = kdamond_vm_cleanup;
+		damon_set_vaddr_primitives(ctx);
 		if (!strncmp(kbuf, "pidfd ", 6)) {
 			received_pidfds = true;
 			nrs = &kbuf[6];
@@ -2191,7 +2271,7 @@ static ssize_t debugfs_target_ids_write(struct file *file,
 static ssize_t debugfs_record_read(struct file *file,
 		char __user *buf, size_t count, loff_t *ppos)
 {
-	struct damon_ctx *ctx = &damon_user_ctx;
+	struct damon_ctx *ctx = file->private_data;
 	char record_buf[20 + MAX_RFILE_PATH_LEN];
 	int ret;
 
@@ -2205,7 +2285,7 @@ static ssize_t debugfs_record_read(struct file *file,
 static ssize_t debugfs_record_write(struct file *file,
 		const char __user *buf, size_t count, loff_t *ppos)
 {
-	struct damon_ctx *ctx = &damon_user_ctx;
+	struct damon_ctx *ctx = file->private_data;
 	char *kbuf;
 	unsigned int rbuf_len;
 	char rfile_path[MAX_RFILE_PATH_LEN];
@@ -2261,7 +2341,7 @@ static ssize_t sprint_init_regions(struct damon_ctx *c, char *buf, ssize_t len)
 static ssize_t debugfs_init_regions_read(struct file *file, char __user *buf,
 		size_t count, loff_t *ppos)
 {
-	struct damon_ctx *ctx = &damon_user_ctx;
+	struct damon_ctx *ctx = file->private_data;
 	char *kbuf;
 	ssize_t len;
 
@@ -2354,7 +2434,7 @@ static ssize_t debugfs_init_regions_write(struct file *file,
 					  const char __user *buf, size_t count,
 					  loff_t *ppos)
 {
-	struct damon_ctx *ctx = &damon_user_ctx;
+	struct damon_ctx *ctx = file->private_data;
 	char *kbuf;
 	ssize_t ret = count;
 	int err;
@@ -2382,7 +2462,7 @@ static ssize_t debugfs_init_regions_write(struct file *file,
 static ssize_t debugfs_attrs_read(struct file *file,
 		char __user *buf, size_t count, loff_t *ppos)
 {
-	struct damon_ctx *ctx = &damon_user_ctx;
+	struct damon_ctx *ctx = file->private_data;
 	char kbuf[128];
 	int ret;
 
@@ -2399,7 +2479,7 @@ static ssize_t debugfs_attrs_read(struct file *file,
 static ssize_t debugfs_attrs_write(struct file *file,
 		const char __user *buf, size_t count, loff_t *ppos)
 {
-	struct damon_ctx *ctx = &damon_user_ctx;
+	struct damon_ctx *ctx = file->private_data;
 	unsigned long s, a, r, minr, maxr;
 	char *kbuf;
 	ssize_t ret = count;
@@ -2431,6 +2511,128 @@ static ssize_t debugfs_attrs_write(struct file *file,
 	return ret;
 }
 
+static ssize_t debugfs_nr_contexts_read(struct file *file,
+		char __user *buf, size_t count, loff_t *ppos)
+{
+	char kbuf[32];
+	int ret;
+
+	mutex_lock(&damon_lock);
+	ret = scnprintf(kbuf, ARRAY_SIZE(kbuf), "%d\n", debugfs_nr_ctxs);
+	mutex_unlock(&damon_lock);
+
+	return simple_read_from_buffer(buf, count, ppos, kbuf, ret);
+}
+
+static struct dentry **debugfs_dirs;
+
+static int debugfs_fill_ctx_dir(struct dentry *dir, struct damon_ctx *ctx);
+
+static ssize_t debugfs_nr_contexts_write(struct file *file,
+		const char __user *buf, size_t count, loff_t *ppos)
+{
+	char *kbuf;
+	ssize_t ret = count;
+	int nr_contexts, i;
+	char dirname[32];
+	struct dentry *root;
+	struct dentry **new_dirs;
+	struct damon_ctx **new_ctxs;
+
+	kbuf = user_input_str(buf, count, ppos);
+	if (IS_ERR(kbuf))
+		return PTR_ERR(kbuf);
+
+	if (sscanf(kbuf, "%d", &nr_contexts) != 1) {
+		ret = -EINVAL;
+		goto out;
+	}
+	if (nr_contexts < 1) {
+		pr_err("nr_contexts should be >=1\n");
+		ret = -EINVAL;
+		goto out;
+	}
+	if (nr_contexts == debugfs_nr_ctxs)
+		goto out;
+
+	mutex_lock(&damon_lock);
+	if (nr_running_ctxs) {
+		ret = -EBUSY;
+		goto unlock_out;
+	}
+
+	for (i = nr_contexts; i < debugfs_nr_ctxs; i++) {
+		debugfs_remove(debugfs_dirs[i]);
+		damon_destroy_ctx(debugfs_ctxs[i]);
+	}
+
+	new_dirs = kmalloc_array(nr_contexts, sizeof(*new_dirs), GFP_KERNEL);
+	if (!new_dirs) {
+		ret = -ENOMEM;
+		goto unlock_out;
+	}
+
+	new_ctxs = kmalloc_array(nr_contexts, sizeof(*debugfs_ctxs),
+			GFP_KERNEL);
+	if (!new_ctxs) {
+		ret = -ENOMEM;
+		goto unlock_out;
+	}
+
+	for (i = 0; i < debugfs_nr_ctxs && i < nr_contexts; i++) {
+		new_dirs[i] = debugfs_dirs[i];
+		new_ctxs[i] = debugfs_ctxs[i];
+	}
+	kfree(debugfs_dirs);
+	debugfs_dirs = new_dirs;
+	kfree(debugfs_ctxs);
+	debugfs_ctxs = new_ctxs;
+
+	root = debugfs_dirs[0];
+	if (!root) {
+		ret = -ENOENT;
+		goto unlock_out;
+	}
+
+	for (i = debugfs_nr_ctxs; i < nr_contexts; i++) {
+		scnprintf(dirname, sizeof(dirname), "ctx%d", i);
+		debugfs_dirs[i] = debugfs_create_dir(dirname, root);
+		if (!debugfs_dirs[i]) {
+			pr_err("dir %s creation failed\n", dirname);
+			ret = -ENOMEM;
+			break;
+		}
+
+		debugfs_ctxs[i] = damon_new_ctx();
+		if (!debugfs_ctxs[i]) {
+			pr_err("ctx for %s creation failed\n", dirname);
+			ret = -ENOMEM;
+			break;
+		}
+
+		if (debugfs_fill_ctx_dir(debugfs_dirs[i], debugfs_ctxs[i])) {
+			ret = -ENOMEM;
+			break;
+		}
+	}
+
+	debugfs_nr_ctxs = i;
+
+unlock_out:
+	mutex_unlock(&damon_lock);
+
+out:
+	kfree(kbuf);
+	return ret;
+}
+
+static int damon_debugfs_open(struct inode *inode, struct file *file)
+{
+	file->private_data = inode->i_private;
+
+	return nonseekable_open(inode, file);
+}
+
 static const struct file_operations monitor_on_fops = {
 	.owner = THIS_MODULE,
 	.read = debugfs_monitor_on_read,
@@ -2439,43 +2641,71 @@ static const struct file_operations monitor_on_fops = {
 
 static const struct file_operations target_ids_fops = {
 	.owner = THIS_MODULE,
+	.open = damon_debugfs_open,
 	.read = debugfs_target_ids_read,
 	.write = debugfs_target_ids_write,
 };
 
 static const struct file_operations schemes_fops = {
 	.owner = THIS_MODULE,
+	.open = damon_debugfs_open,
 	.read = debugfs_schemes_read,
 	.write = debugfs_schemes_write,
 };
 
 static const struct file_operations record_fops = {
 	.owner = THIS_MODULE,
+	.open = damon_debugfs_open,
 	.read = debugfs_record_read,
 	.write = debugfs_record_write,
 };
 
 static const struct file_operations init_regions_fops = {
 	.owner = THIS_MODULE,
+	.open = damon_debugfs_open,
 	.read = debugfs_init_regions_read,
 	.write = debugfs_init_regions_write,
 };
 
 static const struct file_operations attrs_fops = {
 	.owner = THIS_MODULE,
+	.open = damon_debugfs_open,
 	.read = debugfs_attrs_read,
 	.write = debugfs_attrs_write,
 };
 
-static struct dentry *debugfs_root;
+static const struct file_operations nr_contexts_fops = {
+	.owner = THIS_MODULE,
+	.read = debugfs_nr_contexts_read,
+	.write = debugfs_nr_contexts_write,
+};
 
-static int __init damon_debugfs_init(void)
+static int debugfs_fill_ctx_dir(struct dentry *dir, struct damon_ctx *ctx)
 {
 	const char * const file_names[] = {"attrs", "init_regions", "record",
-		"schemes", "target_ids", "monitor_on"};
+		"schemes", "target_ids"};
 	const struct file_operations *fops[] = {&attrs_fops,
 		&init_regions_fops, &record_fops, &schemes_fops,
-		&target_ids_fops, &monitor_on_fops};
+		&target_ids_fops};
+	int i;
+
+	for (i = 0; i < ARRAY_SIZE(file_names); i++) {
+		if (!debugfs_create_file(file_names[i], 0600, dir,
+					ctx, fops[i])) {
+			pr_err("failed to create %s file\n", file_names[i]);
+			return -ENOMEM;
+		}
+	}
+
+	return 0;
+}
+
+static int __init damon_debugfs_init(void)
+{
+	struct dentry *debugfs_root;
+	const char * const file_names[] = {"nr_contexts", "monitor_on"};
+	const struct file_operations *fops[] = {&nr_contexts_fops,
+		&monitor_on_fops};
 	int i;
 
 	debugfs_root = debugfs_create_dir("damon", NULL);
@@ -2491,27 +2721,10 @@ static int __init damon_debugfs_init(void)
 			return -ENOMEM;
 		}
 	}
+	debugfs_fill_ctx_dir(debugfs_root, debugfs_ctxs[0]);
 
-	return 0;
-}
-
-static int __init damon_init_user_ctx(void)
-{
-	int rc;
-
-	struct damon_ctx *ctx = &damon_user_ctx;
-
-	ktime_get_coarse_ts64(&ctx->last_aggregation);
-	ctx->last_regions_update = ctx->last_aggregation;
-
-	rc = damon_set_recording(ctx, 1024 * 1024, "/damon.data");
-	if (rc)
-		return rc;
-
-	mutex_init(&ctx->kdamond_lock);
-
-	INIT_LIST_HEAD(&ctx->targets_list);
-	INIT_LIST_HEAD(&ctx->schemes_list);
+	debugfs_dirs = kmalloc_array(1, sizeof(debugfs_root), GFP_KERNEL);
+	debugfs_dirs[0] = debugfs_root;
 
 	return 0;
 }
@@ -2524,9 +2737,10 @@ static int __init damon_init(void)
 {
 	int rc;
 
-	rc = damon_init_user_ctx();
-	if (rc)
-		return rc;
+	debugfs_ctxs = kmalloc(sizeof(*debugfs_ctxs), GFP_KERNEL);
+	debugfs_ctxs[0] = damon_new_ctx();
+	if (!debugfs_ctxs[0])
+		return -ENOMEM;
 
 	rc = damon_debugfs_init();
 	if (rc)
-- 
2.17.1

[RFC PATCH 0/2] DAMON: Improve User Space Interface

[SeongJae Park]

From: SeongJae Park <sjpark@amazon.de>

Seems lore.kernel.org missed original posting of this mail.  Resending.

NOTE: This is only an RFC for future features of DAMON patchset[1], which is
not merged in the mainline yet.  The aim of this RFC is to show how DAMON would
be evolved once it is merged in.  So, if you have some interest in this RFC,
please consider reviewing the DAMON patchset, either.

After posting DAMON patchset[1], we received a number of comments.  Based on
those, we listed and shared future works for DAMON in the kernel summit
2020[2] and had a poll for the priorities of the works.  As a result, the user
space interface improvement received a second highest priority[3].  For the
reason, this patchset is came out.

The 1st patch puts more information in the monitoring thread name so that user
space could charge the DAMON's CPU usage on them by themselves, in fine
granularity.  The 2nd patch makes multiple monitoring contexts available using
the debugfs interface.

[1] https://lore.kernel.org/linux-mm/20200817105137.19296-1-sjpark@amazon.com/
[2] https://linuxplumbersconf.org/event/7/contributions/659/
[3] https://lore.kernel.org/linux-mm/20200831112235.2675-1-sjpark@amazon.com/

Baseline and Complete Git Trees
===============================

The patches are based on the v5.8 plus DAMON v20 patchset[1], RFC v14 of DAMOS
patchset, RFC v8 of physical address space support patchset, and some more
trivial fixes (s/snprintf/scnprintf).  You can also clone the complete git
tree:

    $ git clone git://github.com/sjp38/linux -b damon-usi/rfc/v1

The web is also available:
https://github.com/sjp38/linux/releases/tag/damon-usi/rfc/v1

[1] https://lore.kernel.org/linux-mm/20200817105137.19296-1-sjpark@amazon.com/
[2] https://lore.kernel.org/linux-mm/20200804142430.15384-1-sjpark@amazon.com/
[3] https://lore.kernel.org/linux-mm/20200831104730.28970-1-sjpark@amazon.com/

SeongJae Park (2):
  mm/damon: Put more information in kdamond thread name
  mm/damon/debugfs: Support multiple contexts

 include/linux/damon.h |   2 +
 mm/damon-test.h       |  34 ++--
 mm/damon.c            | 361 +++++++++++++++++++++++++++++++++---------
 3 files changed, 308 insertions(+), 89 deletions(-)

-- 
2.17.1

Re: [RFC PATCH 2/2] mm/damon/debugfs: Support multiple contexts

[Shakeel Butt]

On Tue, Sep 15, 2020 at 11:03 AM SeongJae Park <sjpark@amazon.com> wrote:
>
> From: SeongJae Park <sjpark@amazon.de>
>
> DAMON allows the programming interface users to run monitoring with
> multiple contexts.  This could be useful in some cases.  For example, if
> someone want to do highly accurate monitoring and lots of CPUs are
> available, splitting the monitoring target regions into multiple small
> regions and allocating context (monitoring thread) to each small region
> could be helpful.  Or, someone could need to monitor different types of
> address spaces simultaneously.
>
> However, it's impossible from the user space because the DAMON debugfs
> interface supports only single context.  This commit makes it available
> by implementing 'nr_contexts' debugfs file.
>
> Users can pass the number (N) of contexts they want to use to the file.

Why not just mkdir which will create a new context?

> Then, N folders having name of 'ctx<1-(N-1)>' are created in the DAMON
> debugfs dir.  Each of the directory is associated with the contexts and
> contains the the files for context setting (attrs, init_regions, record,
> schemes, and target_ids).

Also instead of naming the kthread with context number why not give
the kthread pids through attrs (or new interface) and the admin can
move those kthreads to the cgroup they want to charge against?

> The first context related files are still in
> the DAMON debugfs root directory, though.
>
> Signed-off-by: SeongJae Park <sjpark@amazon.de>
> ---

Re: [RFC PATCH 2/2] mm/damon/debugfs: Support multiple contexts

[Shakeel Butt]

On Tue, Sep 15, 2020 at 11:03 AM SeongJae Park <sjpark@amazon.com> wrote:
>
> From: SeongJae Park <sjpark@amazon.de>
>
> DAMON allows the programming interface users to run monitoring with
> multiple contexts.  This could be useful in some cases.  For example, if
> someone want to do highly accurate monitoring and lots of CPUs are
> available, splitting the monitoring target regions into multiple small
> regions and allocating context (monitoring thread) to each small region
> could be helpful.  Or, someone could need to monitor different types of
> address spaces simultaneously.
>
> However, it's impossible from the user space because the DAMON debugfs
> interface supports only single context.  This commit makes it available
> by implementing 'nr_contexts' debugfs file.
>
> Users can pass the number (N) of contexts they want to use to the file.

Why not just mkdir which will create a new context?

> Then, N folders having name of 'ctx<1-(N-1)>' are created in the DAMON
> debugfs dir.  Each of the directory is associated with the contexts and
> contains the the files for context setting (attrs, init_regions, record,
> schemes, and target_ids).

Also instead of naming the kthread with context number why not give
the kthread pids through attrs (or new interface) and the admin can
move those kthreads to the cgroup they want to charge against?

> The first context related files are still in
> the DAMON debugfs root directory, though.
>
> Signed-off-by: SeongJae Park <sjpark@amazon.de>
> ---

Re: [RFC PATCH 2/2] mm/damon/debugfs: Support multiple contexts

[SeongJae Park]

On Wed, 16 Sep 2020 09:11:18 -0700 Shakeel Butt <shakeelb@google.com> wrote:

> On Tue, Sep 15, 2020 at 11:03 AM SeongJae Park <sjpark@amazon.com> wrote:
> >
> > From: SeongJae Park <sjpark@amazon.de>
> >
> > DAMON allows the programming interface users to run monitoring with
> > multiple contexts.  This could be useful in some cases.  For example, if
> > someone want to do highly accurate monitoring and lots of CPUs are
> > available, splitting the monitoring target regions into multiple small
> > regions and allocating context (monitoring thread) to each small region
> > could be helpful.  Or, someone could need to monitor different types of
> > address spaces simultaneously.
> >
> > However, it's impossible from the user space because the DAMON debugfs
> > interface supports only single context.  This commit makes it available
> > by implementing 'nr_contexts' debugfs file.
> >
> > Users can pass the number (N) of contexts they want to use to the file.
> 
> Why not just mkdir which will create a new context?

Because I referenced the naming rule of zram and because I just wanted to make
this as simple as possible.  I find no special functional difference between
current way and you suggested way.

That said, I also think you suggested way is much more flexible.  I will try to
make this in that way.

> 
> > Then, N folders having name of 'ctx<1-(N-1)>' are created in the DAMON
> > debugfs dir.  Each of the directory is associated with the contexts and
> > contains the the files for context setting (attrs, init_regions, record,
> > schemes, and target_ids).
> 
> Also instead of naming the kthread with context number why not give
> the kthread pids through attrs (or new interface) and the admin can
> move those kthreads to the cgroup they want to charge against?

Again, I just wanted to make this as simple as possible, and I also referenced
blkback thread naming.  And I find no special functional difference here,
either.

And yes, I think you suggested way is much more flexible and easy to be used.
I will make this in that way.

---

Just a note (not for only Shakeel, but anyone who interested in DAMON).  The
DAMON debugfs interface is not highly coupled with DAMON.  Rather than that, it
is implemented as an application using DAMON as a framework via the DAMON's
kernel space interface.  This means that anyone who want different user space
interface can implement their own on top of the framework.  I will try to make
the changes as soon as possible, but if you cannot wait until then, you could
implement your own.

Similarly, DAMON lacks many functionality such as support of special address
spaces.  I already have a TODO list.  And I will implement those one by one.
So, please give me the requests.  However, I believe the kernel space interface
is already almost complete.  Thus, almost every such future features could be
implemented on top of current DAMON interface.  Therefore, if you cannot wait
until I finish the work, you can implement your own on DAMON's kernel space
interface.

Of course, if you send the patches for your own implementations, that will make
not only I, but the community happier.

And this is the reason why the patchset containing only the framework part and
minimal applications / addons is titled as official 'PATCH', while the
patchsets for the future features are titled 'RFC'.  So, I am asking you to
consider giving more attention to the framework part (the first 4 patches of
the patchset[1]), as all future features will depend on it.

[1] https://lore.kernel.org/linux-mm/20200817105137.19296-1-sjpark@amazon.com/


Thanks,
SeongJae Park