mm/damon/stat: monitor all System RAM resources

DAMON_STAT usage document (Documentation/admin-guide/mm/damon/stat.rst) says it monitors the system's entire physical memory. But, it is monitoring only the biggest System RAM resource of the system. When there are multiple System RAM resources, this results in monitoring only an unexpectedly small fraction of the physical memory. For example, suppose the system has a 500 GiB System RAM, 10 MiB non-System RAM, and 500 GiB System RAM resources in order on the physical address space. DAMON_STAT will monitor only the first 500 GiB System RAM. This situation is particularly common on NUMA systems.

Select a physical address range that covers all System RAM areas of the system, to fix this issue and make it work as documented.

---
Changes from v1
(https://lore.kernel.org/[email protected])
- Fix wrong argument for damon_set_regions().

 mm/damon/stat.c | 51 ++++++++++++++++++++++++++++++++++++++++++++++---
 1 file changed, 48 insertions(+), 3 deletions(-)

index f9a2028483b05..c5af8ad4bcb65 100644
@@ -145,12 +145,57 @@ static int damon_stat_damon_call_fn(void *data)
 	return 0;
 }
 
+struct damon_stat_system_ram_range_walk_arg {
+	bool walked;
+	struct resource res;
+};
+
+static int damon_stat_system_ram_walk_fn(struct resource *res, void *arg)
+{
+	struct damon_stat_system_ram_range_walk_arg *a = arg;
+
+	if (!a->walked) {
+		a->walked = true;
+		a->res.start = res->start;
+	}
+	a->res.end = res->end;
+	return 0;
+}
+
+static unsigned long damon_stat_res_to_core_addr(resource_size_t ra,
+		unsigned long addr_unit)
+{
+	/*
+	 * Use div_u64() for avoiding linking errors related with __udivdi3,
+	 * __aeabi_uldivmod, or similar problems.  This should also improve the
+	 * performance optimization (read div_u64() comment for the detail).
+	 */
+	if (sizeof(ra) == 8 && sizeof(addr_unit) == 4)
+		return div_u64(ra, addr_unit);
+	return ra / addr_unit;
+}
+
+static int damon_stat_set_monitoring_region(struct damon_target *t,
+		unsigned long addr_unit, unsigned long min_region_sz)
+{
+	struct damon_addr_range addr_range;
+	struct damon_stat_system_ram_range_walk_arg arg = {};
+
+	walk_system_ram_res(0, -1, &arg, damon_stat_system_ram_walk_fn);
+	if (!arg.walked)
+		return -EINVAL;
+	addr_range.start = damon_stat_res_to_core_addr(
+			arg.res.start, addr_unit);
+	addr_range.end = damon_stat_res_to_core_addr(
+			arg.res.end + 1, addr_unit);
+	return damon_set_regions(t, &addr_range, 1, min_region_sz);
+}
+
 static struct damon_ctx *damon_stat_build_ctx(void)
 {
 	struct damon_ctx *ctx;
 	struct damon_attrs attrs;
 	struct damon_target *target;
-	unsigned long start = 0, end = 0;
 
 	ctx = damon_new_ctx();
 	if (!ctx)

@@ -180,8 +225,8 @@ static struct damon_ctx *damon_stat_build_ctx(void)
 	if (!target)
 		goto free_out;
 	damon_add_target(ctx, target);
-	if (damon_set_region_biggest_system_ram_default(target, &start, &end,
-				ctx->addr_unit, ctx->min_region_sz))
+	if (damon_stat_set_monitoring_region(target, ctx->addr_unit,
+				ctx->min_region_sz))
 		goto free_out;
 	return ctx;
 free_out:

base-commit: 209f54c1171365f1d942b89554808acce71aadb4

2.47.3

sashiko.dev adds [1] below comment:

'''
Does this single bounding box incorrectly include unpopulated address gaps between discrete System RAM resources?

On systems with non-contiguous physical memory, such as NUMA architectures, there can be massive physical address gaps between memory nodes. By coalescing all resources into a single addr_range and passing nr_ranges = 1 to damon_set_regions(), DAMON treats these unpopulated gaps as part of the monitored memory.

This appears to artificially inflate total_sz in damon_stat_set_idletime_percentiles(), where the gap could completely dominate the distribution and skew the percentiles to report that nearly 100% of memory is permanently idle.

Could this also wildly inflate estimated_memory_bandwidth in damon_stat_set_estimated_memory_bandwidth() if an adaptive DAMON region bridges valid RAM and a physical gap? The size (r->ar.end - r->ar.start) would be massively inflated and multiplied by nr_accesses.

Would it be better to collect all discrete System RAM resources into an array of struct damon_addr_range and pass them to damon_set_regions() using the actual number of ranges?
'''

My answer is no. It is an intended behavior and no negative impact is expected. I think the reason is in the patch description. If any human needs more clarifications about this, please let me know.

[1] https://sashiko.dev/#/patchset/[email protected]

Thanks,
SJ

[...]

This doesn't apply to current mainline?

Ah, you're right. Sorry for this noise.

I will rebase it to mm-hotfixes-unstable and post it as v3, by tonight. If you prefer mainline or another tree other than mm-hotfixes-unsable as the baseline of v3, please let me know.

Thanks,
SJ