mm: emit tracepoint when RSS changes

joelagnel · torvalds · commit b3d1411b6726 · 2019-12-01T06:29:18.000-08:00
Useful to track how RSS is changing per TGID to detect spikes in RSS and memory hogs. Several Android teams have been using this patch in various kernel trees for half a year now. Many reported to me it is really useful so I'm posting it upstream. Initial patch developed by Tim Murray. Changes I made from original patch: o Prevent any additional space consumed by mm_struct. Regarding the fact that the RSS may change too often thus flooding the traces - note that, there is some "hysterisis" with this already. That is - We update the counter only if we receive 64 page faults due to SPLIT_RSS_ACCOUNTING. However, during zapping or copying of pte range, the RSS is updated immediately which can become noisy/flooding. In a previous discussion, we agreed that BPF or ftrace can be used to rate limit the signal if this becomes an issue. Also note that I added wrappers to trace_rss_stat to prevent compiler errors where linux/mm.h is included from tracing code, causing errors such as: CC kernel/trace/power-traces.o In file included from ./include/trace/define_trace.h:102, from ./include/trace/events/kmem.h:342, from ./include/linux/mm.h:31, from ./include/linux/ring_buffer.h:5, from ./include/linux/trace_events.h:6, from ./include/trace/events/power.h:12, from kernel/trace/power-traces.c:15: ./include/trace/trace_events.h:113:22: error: field `ent' has incomplete type struct trace_entry ent; \ Link: http://lore.kernel.org/r/20190903200905.198642-1-joel@joelfernandes.org Link: http://lkml.kernel.org/r/20191001172817.234886-1-joel@joelfernandes.org Co-developed-by: Tim Murray <timmurray@google.com> Signed-off-by: Tim Murray <timmurray@google.com> Signed-off-by: Joel Fernandes (Google) <joel@joelfernandes.org> Acked-by: Michal Hocko <mhocko@suse.com> Cc: Carmen Jackson <carmenjackson@google.com> Cc: Mayank Gupta <mayankgupta@google.com> Cc: Daniel Colascione <dancol@google.com> Cc: Steven Rostedt (VMware) <rostedt@goodmis.org> Cc: Minchan Kim <minchan@kernel.org> Cc: "Aneesh Kumar K.V" <aneesh.kumar@linux.ibm.com> Cc: Dan Williams <dan.j.williams@intel.com> Cc: Jerome Glisse <jglisse@redhat.com> Cc: Matthew Wilcox <willy@infradead.org> Cc: Ralph Campbell <rcampbell@nvidia.com> Cc: Vlastimil Babka <vbabka@suse.cz> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
diff --git a/include/linux/mm.h b/include/linux/mm.h
@@ -1643,19 +1643,27 @@ static inline unsigned long get_mm_counter(struct mm_struct *mm, int member)
 	return (unsigned long)val;
 }
 
+void mm_trace_rss_stat(int member, long count);
+
 static inline void add_mm_counter(struct mm_struct *mm, int member, long value)
 {
-	atomic_long_add(value, &mm->rss_stat.count[member]);
+	long count = atomic_long_add_return(value, &mm->rss_stat.count[member]);
+
+	mm_trace_rss_stat(member, count);
 }
 
 static inline void inc_mm_counter(struct mm_struct *mm, int member)
 {
-	atomic_long_inc(&mm->rss_stat.count[member]);
+	long count = atomic_long_inc_return(&mm->rss_stat.count[member]);
+
+	mm_trace_rss_stat(member, count);
 }
 
 static inline void dec_mm_counter(struct mm_struct *mm, int member)
 {
-	atomic_long_dec(&mm->rss_stat.count[member]);
+	long count = atomic_long_dec_return(&mm->rss_stat.count[member]);
+
+	mm_trace_rss_stat(member, count);
 }
 
 /* Optimized variant when page is already known not to be PageAnon */
diff --git a/include/trace/events/kmem.h b/include/trace/events/kmem.h
@@ -316,6 +316,27 @@ TRACE_EVENT(mm_page_alloc_extfrag,
 		__entry->change_ownership)
 );
 
+TRACE_EVENT(rss_stat,
+
+	TP_PROTO(int member,
+		long count),
+
+	TP_ARGS(member, count),
+
+	TP_STRUCT__entry(
+		__field(int, member)
+		__field(long, size)
+	),
+
+	TP_fast_assign(
+		__entry->member = member;
+		__entry->size = (count << PAGE_SHIFT);
+	),
+
+	TP_printk("member=%d size=%ldB",
+		__entry->member,
+		__entry->size)
+	);
 #endif /* _TRACE_KMEM_H */
 
 /* This part must be outside protection */
diff --git a/mm/memory.c b/mm/memory.c
@@ -72,6 +72,8 @@
 #include <linux/oom.h>
 #include <linux/numa.h>
 
+#include <trace/events/kmem.h>
+
 #include <asm/io.h>
 #include <asm/mmu_context.h>
 #include <asm/pgalloc.h>
@@ -152,6 +154,10 @@ static int __init init_zero_pfn(void)
 }
 core_initcall(init_zero_pfn);
 
+void mm_trace_rss_stat(int member, long count)
+{
+	trace_rss_stat(member, count);
+}
 
 #if defined(SPLIT_RSS_COUNTING)
 

Original file line number	Diff line number	Diff line change
`@@ -1643,19 +1643,27 @@ static inline unsigned long get_mm_counter(struct mm_struct *mm, int member)`
`1643`	`1643`	`return (unsigned long)val;`
`1644`	`1644`	`}`
`1645`	`1645`
	`1646`	`+void mm_trace_rss_stat(int member, long count);`
	`1647`	`+`
`1646`	`1648`	`static inline void add_mm_counter(struct mm_struct *mm, int member, long value)`
`1647`	`1649`	`{`
`1648`		`- atomic_long_add(value, &mm->rss_stat.count[member]);`
	`1650`	`+ long count = atomic_long_add_return(value, &mm->rss_stat.count[member]);`
	`1651`	`+`
	`1652`	`+ mm_trace_rss_stat(member, count);`
`1649`	`1653`	`}`
`1650`	`1654`
`1651`	`1655`	`static inline void inc_mm_counter(struct mm_struct *mm, int member)`
`1652`	`1656`	`{`
`1653`		`- atomic_long_inc(&mm->rss_stat.count[member]);`
	`1657`	`+ long count = atomic_long_inc_return(&mm->rss_stat.count[member]);`
	`1658`	`+`
	`1659`	`+ mm_trace_rss_stat(member, count);`
`1654`	`1660`	`}`
`1655`	`1661`
`1656`	`1662`	`static inline void dec_mm_counter(struct mm_struct *mm, int member)`
`1657`	`1663`	`{`
`1658`		`- atomic_long_dec(&mm->rss_stat.count[member]);`
	`1664`	`+ long count = atomic_long_dec_return(&mm->rss_stat.count[member]);`
	`1665`	`+`
	`1666`	`+ mm_trace_rss_stat(member, count);`
`1659`	`1667`	`}`
`1660`	`1668`
`1661`	`1669`	`/* Optimized variant when page is already known not to be PageAnon */`