Re: [PATCH V5] Tracepoint Tapset for Memory Subsystem

[Rajasekhar Duddu <rajduddu at linux dot vnet dot ibm dot com>]

Hi all,
	I have update the patch according to Franks previous
comments.
Please review it and if it is fine, please somebody commit this
version of patch on behalf of me..thanks in advance.

Changelog 1:
        Removed the Hardcoded constants in converting GFPFLAGS.
        Added a kprobe based fallback probe to kfree.

Changelog 2:
        Removed the unreliable code which was used in fetching the
call_site in kprobe based probe for
kfree.(__builtin_return_address(0))

Changelog 3:
        Defined two macros for converting the GFP_FLAGS into string
formats.
        Added k(ret)probe based fallback probes for all the
Functions.

Changelog 4:
        Added an example script and a corresponding meta file
for kmem_cache_alloc probe point at
testsuite/systemtap.examples/memory/vm.tracepoints.(stp/meta)

Changelog 5:
	Updated the example script.
	Added description of the tapset in NEWS.
	Confirmed that the tapset reference documentation
template is configured to extract the @@doc stuff from the tapset .


Sample O/P:

stap -v testsuite/systemtap.examples/memory/vm.tracepoints.stp
Pass 1: parsed user script and 63 library script(s) in
170usr/10sys/189real ms.
Pass 2: analyzed script: 2 probe(s), 2 function(s), 3 embed(s), 1
global(s) in 11480usr/3730sys/14373real ms.
Pass 3: translated to C into
"/tmp/stapxnPdcI/stap_2436291999f620c5f051efe7320a1268_3415.c" in
0usr/0sys/1real ms.
Pass 4: compiled C into
"stap_2436291999f620c5f051efe7320a1268_3415.ko" in
3600usr/610sys/4176real ms.
Pass 5: starting run.

Process:kjournald
Slab_size:256   Count:109
Slab_size:112   Count:109
Slab_size:104   Count:109
Slab_size:336   Count:2

Process:swapper
Slab_size:256   Count:1

Process:sendmail
Slab_size:192   Count:2
Slab_size:64    Count:2
Slab_size:640   Count:2
Slab_size:16    Count:2
Slab_size:32    Count:2
Slab_size:256   Count:2
Slab_size:128   Count:2
Slab_size:72    Count:2
Slab_size:4096  Count:4

-------------------------------------------------------


Process:sshd
Slab_size:256   Count:1

Process:pdflush
Slab_size:256   Count:729

Process:sshd
Slab_size:512   Count:1


Patch:


Signed-off-by: Rajasekhar Duddu <rajduddu@linux.vnet.ibm.com>
diff -rupaN a/NEWS b/NEWS
--- a/NEWS	2009-11-25 01:18:29.000000000 -0500
+++ b/NEWS	2009-12-04 06:24:33.000000000 -0500
@@ -1,3 +1,5 @@
+New tapset for Memory subsystem based on Kernel Tracepoints.
+
 * What's new in verson 1.0
 
 - process().mark() probes now use an enabling semaphore to reduce the
diff -rupaN a/tapset/memory.stp b/tapset/memory.stp
--- a/tapset/memory.stp	2009-11-25 01:18:29.000000000 -0500
+++ b/tapset/memory.stp	2009-12-04 06:42:12.000000000 -0500
@@ -195,3 +195,269 @@ probe vm.brk = kernel.function("do_brk")
 probe vm.oom_kill = kernel.function("__oom_kill_task") {
     task = $p
 }
+
+function __gfp_flag_str:string(gfp_flag:long) %{
+	long gfp_flag = THIS->gfp_flag;
+	THIS->__retvalue[0] = '\0';
+
+
+/* Macro for GFP Bitmasks. */
+/* The resulted GFP_FLAGS may be either single or concatenation of the multiple bitmasks. */
+
+
+#define __GFP_BITMASKS(FLAG)  if(gfp_flag & FLAG) { if(THIS->__retvalue[0] != '\0') \
+                strlcat(THIS->__retvalue, " | "#FLAG, MAXSTRINGLEN); \
+                else strlcat(THIS->__retvalue, #FLAG, MAXSTRINGLEN); }
+
+
+/* Macro for Composite Flags. */
+/* Each Composite GFP_FLAG is the combination of multiple bitmasks. */
+
+
+#define __GFP_COMPOSITE_FLAG(FLAG)  if(gfp_flag == FLAG) { \
+                strlcat(THIS->__retvalue, #FLAG, MAXSTRINGLEN); return; }
+
+
+/* Composite GFP FLAGS of the BitMasks. */
+
+	__GFP_COMPOSITE_FLAG(GFP_ZONEMASK)
+	__GFP_COMPOSITE_FLAG(GFP_ATOMIC)
+	__GFP_COMPOSITE_FLAG(GFP_NOIO)
+	__GFP_COMPOSITE_FLAG(GFP_NOFS)
+	__GFP_COMPOSITE_FLAG(GFP_KERNEL)
+	__GFP_COMPOSITE_FLAG(GFP_TEMPORARY)
+	__GFP_COMPOSITE_FLAG(GFP_USER)
+	__GFP_COMPOSITE_FLAG(GFP_HIGHUSER)
+	__GFP_COMPOSITE_FLAG(GFP_HIGHUSER_MOVABLE)
+	__GFP_COMPOSITE_FLAG(GFP_THISNODE)
+	__GFP_COMPOSITE_FLAG(GFP_DMA)
+	__GFP_COMPOSITE_FLAG(GFP_DMA32)
+
+/* GFP BitMasks */
+
+	__GFP_BITMASKS(__GFP_DMA)
+	__GFP_BITMASKS(__GFP_HIGHMEM)
+	__GFP_BITMASKS(__GFP_MOVABLE)
+	__GFP_BITMASKS(__GFP_WAIT)
+	__GFP_BITMASKS(__GFP_HIGH)
+	__GFP_BITMASKS(__GFP_IO)
+	__GFP_BITMASKS(__GFP_FS)
+	__GFP_BITMASKS(__GFP_COLD)
+	__GFP_BITMASKS(__GFP_NOWARN)
+	__GFP_BITMASKS(__GFP_REPEAT)
+	__GFP_BITMASKS(__GFP_NOFAIL)
+	__GFP_BITMASKS(__GFP_COMP)
+	__GFP_BITMASKS(__GFP_ZERO)
+	__GFP_BITMASKS(__GFP_NOMEMALLOC)
+	__GFP_BITMASKS(__GFP_HARDWALL)
+	__GFP_BITMASKS(__GFP_THISNODE)
+	__GFP_BITMASKS(__GFP_RECLAIMABLE)
+	__GFP_BITMASKS(__GFP_NOTRACK)
+
+
+#undef __GFP_BITMASKS
+#undef __GFP_COMPOSITE_FLAG
+%}
+
+/* The Formal Parameters will be displayed if available, otherwise \
+		 "0" or "unknown" will be displayed */
+
+probe __vm.kmalloc.tp = kernel.trace("kmalloc") {
+	name = "kmalloc"
+	call_site = $call_site
+	caller_function = symname(call_site)
+	bytes_req = $bytes_req
+	bytes_alloc = $bytes_alloc
+	gfp_flags = $gfp_flags
+	gfp_flag_name = __gfp_flag_str($gfp_flags)
+	ptr = $ptr
+}
+
+/* It is unsafe to invoke __builtin_return_address() \
+presently(to get call_site for kporbe based probes) \
+and that it can be improved later when fix for bugs bz#6961 and bz#6580 is available. */
+
+probe __vm.kmalloc.kp = kernel.function("kmalloc").return {
+	name = "kmalloc"
+	call_site = 0
+	caller_function = "unknown"
+	bytes_req = $size
+	bytes_alloc = "unknown"
+	gfp_flags = $gfp_flags
+	gfp_flag_name = __gfp_flag_str($gfp_flags)
+	ptr = $return
+}
+
+/**
+ * probe vm.kmalloc - Fires when kmalloc is requested.
+ * @call_site: Address of the kmemory function.
+ * @caller_function: Name of the caller function.
+ * @bytes_req: Requested Bytes
+ * @bytes_alloc: Allocated Bytes
+ * @gfp_flags: type of kmemory to allocate
+ * @gfp_flag_name: type of kmemory to allocate (in String format)
+ * @ptr: Pointer to the kmemory allocated
+ */
+probe vm.kmalloc = __vm.kmalloc.tp !,
+			__vm.kmalloc.kp
+{}
+
+
+probe __vm.kmem_cache_alloc.tp = kernel.trace("kmem_cache_alloc") {
+	name = "kmem_cache_alloc"
+	call_site = $call_site
+	caller_function = symname(call_site)
+	bytes_req = $bytes_req
+	bytes_alloc = $bytes_alloc
+	gfp_flags = $gfp_flags
+	gfp_flag_name = __gfp_flag_str($gfp_flags)
+	ptr = $ptr
+}
+
+probe __vm.kmem_cache_alloc.kp = kernel.function("kmem_cache_alloc").return {
+	name = "kmem_cache_alloc"
+	call_site = 0
+	caller_function = "unknown"
+	bytes_req = $size
+	bytes_alloc = "unknown"
+	gfp_flags = $gfp_flags
+	gfp_flag_name = __gfp_flag_str($gfp_flags)
+	ptr = $return
+}
+
+/**
+ * probe vm.kmem_cache_alloc - Fires when \
+ *              kmem_cache_alloc is requested.
+ * @call_site: Address of the function calling this kmemory function.
+ * @caller_function: Name of the caller function.
+ * @bytes_req: Requested Bytes
+ * @bytes_alloc: Allocated Bytes
+ * @gfp_flags: type of kmemory to allocate
+ * @gfp_flag_name: Type of kmemory to allocate(in string format)
+ * @ptr: Pointer to the kmemory allocated
+ */
+
+probe vm.kmem_cache_alloc = __vm.kmem_cache_alloc.tp !,
+				__vm.kmem_cache_alloc.kp
+{}
+
+probe __vm.kmalloc_node.tp = kernel.trace("kmalloc_node")? {
+	name = "kmalloc_node"
+	call_site = $call_site
+	caller_function = symname(call_site)
+	bytes_req = $bytes_req
+	bytes_alloc = $bytes_alloc
+	gfp_flags = $gfp_flags
+	gfp_flag_name = __gfp_flag_str($gfp_flags)
+	ptr = $ptr
+}
+
+probe __vm.kmalloc_node.kp = kernel.function("kmalloc_node").return? {
+	name = "kmalloc_node"
+	call_site = 0
+	caller_function = "unknown"
+	bytes_req = $size
+	bytes_alloc = "unknown"
+	gfp_flags = $gfp_flags
+	gfp_flag_name = __gfp_flag_str($gfp_flags)
+	ptr = $return
+}
+
+/**
+ * probe vm.kmalloc_node - Fires when kmalloc_node is requested.
+ * @call_site: Address of the function caling this  kmemory function.
+ * @caller_function: Name of the caller function.
+ * @bytes_req: Requested Bytes
+ * @bytes_alloc: Allocated Bytes
+ * @gfp_flags: type of kmemory to allocate
+ * @gfp_flag_name: Type of kmemory to allocate(in string format)
+ * @ptr: Pointer to the kmemory allocated
+ */
+probe vm.kmalloc_node = __vm.kmalloc_node.tp !,
+			__vm.kmalloc_node.kp
+{}
+
+probe __vm.kmem_cache_alloc_node.tp = kernel.trace("kmem_cache_alloc_node")? {
+	name = "kmem_cache_alloc_node"
+	call_site = $call_site
+	caller_function = symname(call_site)
+	bytes_req = $bytes_req
+	bytes_alloc = $bytes_alloc
+	gfp_flags = $gfp_flags
+	gfp_flag_name = __gfp_flag_str($gfp_flags)
+	ptr = $ptr
+}
+
+probe __vm.kmem_cache_alloc_node.kp = kernel.function("kmem_cache_alloc_node").return? {
+	name = "kmem_cache_alloc_node"
+	call_site = 0
+	caller_function = "unknown"
+	bytes_req = $size
+	bytes_alloc = "unknown"
+	gfp_flags = $gfp_flags
+	gfp_flag_name = __gfp_flag_str($gfp_flags)
+	ptr = $return
+}
+
+/**
+ * probe vm.kmem_cache_alloc_node - Fires when \
+ *              kmem_cache_alloc_node is requested.
+ * @call_site: Address of the function calling this kmemory function.
+ * @caller_function: Name of the caller function.
+ * @bytes_req: Requested Bytes
+ * @bytes_alloc: Allocated Bytes
+ * @gfp_flags: type of kmemory to allocate
+ * @gfp_flag_name: Type of kmemory to allocate(in string format)
+ * @ptr: Pointer to the kmemory allocated
+ */
+probe vm.kmem_cache_alloc_node = __vm.kmem_cache_alloc_node.tp !,
+				__vm.kmem_cache_alloc_node.kp
+{}
+
+
+probe __vm.kfree.tp = kernel.trace("kfree") {
+	name = "kfree"
+	call_site = $call_site
+	caller_function = symname(call_site)
+	ptr = $ptr
+}
+
+probe __vm.kfree.kp = kernel.function("kfree").return {
+	name = "kfree"
+	call_site = 0
+	caller_function = "unknown"
+	ptr = $return
+}
+
+/**
+ * probe vm.kfree - Fires when kfree is requested.
+ * @call_site: Address of the function calling this kmemory function.
+ * @caller_function: Name of the caller function.
+ * @ptr: Pointer to the kmemory allocated which is returned by kmalloc
+ */
+probe vm.kfree = __vm.kfree.tp !,
+		__vm.kfree.kp
+{}
+
+probe __vm.kmem_cache_free.tp = kernel.trace("kmem_cache_free") {
+	name = "kmem_cache_free"
+	call_site = $call_site
+	caller_function = symname(call_site)
+	ptr = $ptr
+}
+probe __vm.kmem_cache_free.kp = kernel.function("kmem_cache_free").return {
+	name = "kmem_cache_free"
+	call_site = 0
+	caller_function = "unknown"
+	ptr = $return
+}
+/**
+ * probe vm.kmem_cache_free - Fires when \
+ *              kmem_cache_free is requested.
+ * @call_site: Address of the function calling this kmemory function.
+ * @caller_function: Name of the caller function.
+ * @ptr: Pointer to the kmemory allocated which is returned by kmem_cache
+ */
+probe vm.kmem_cache_free = __vm.kmem_cache_free.tp !,
+				__vm.kmem_cache_free.kp
+{}
diff -rupaN a/testsuite/buildok/vm.tracepoints.stp b/testsuite/buildok/vm.tracepoints.stp
--- a/testsuite/buildok/vm.tracepoints.stp	1969-12-31 19:00:00.000000000 -0500
+++ b/testsuite/buildok/vm.tracepoints.stp	2009-12-04 06:20:12.000000000 -0500
@@ -0,0 +1,31 @@
+#! stap -up4
+
+probe vm.kfree {
+        println(name)
+        printf("%-15s %-15p %-15s %-15p \n", execname(), call_site, caller_function, ptr)
+}
+
+probe vm.kmalloc {
+        println(name)
+        printf("%-15s %-15p %-15s %-15p %-15d %-15d %-15d %-15s \n", execname(), call_site, caller_function, ptr, bytes_req, bytes_alloc, gfp_flags, gfp_flag_name)
+}
+
+probe vm.kmem_cache_alloc {
+        println(name)
+        printf("%-15s %-15p %-15s %-15p %-15d %-15d %-15d %-15s \n", execname(), call_site, caller_function, ptr, bytes_req, bytes_alloc, gfp_flags, gfp_flag_name)
+}
+
+probe vm.kmalloc_node {
+        println(name)
+        printf("%-15s %-15p %-15s %-15p %-15d %-15d %-15d %-15s \n", execname(), call_site, caller_function, ptr, bytes_req, bytes_alloc, gfp_flags, gfp_flag_name)
+}
+
+probe vm.kmem_cache_alloc_node {
+        println(name)
+        printf("%-15s %-15p %-15s %-15p %-15d %-15d %-15d %-15s \n", execname(), call_site, caller_function, ptr, bytes_req, bytes_alloc, gfp_flags, gfp_flag_name)
+}
+
+probe vm.kmem_cache_free {
+        println(name)
+        printf("%-15s %-15p %-15s %-15p \n", execname(), call_site, caller_function, ptr)
+}
diff -rupaN a/testsuite/systemtap.examples/memory/vm.tracepoints.meta b/testsuite/systemtap.examples/memory/vm.tracepoints.meta
--- a/testsuite/systemtap.examples/memory/vm.tracepoints.meta	1969-12-31 19:00:00.000000000 -0500
+++ b/testsuite/systemtap.examples/memory/vm.tracepoints.meta	2009-12-04 06:22:05.000000000 -0500
@@ -0,0 +1,13 @@
+title: Collect slab allocation statistics
+name: vm.tracepoints.stp
+version: 1.0
+author: Rajasekhar
+keywords: memory slab allocator
+subsystem: memory
+status: production
+exit: user-controlled
+output: sorted-list
+scope: system-wide
+description: The script will probe all memory slab/slub allocations and collects information about the size of the object (bytes requested) and user-space process in execution. When run over a period of time, it helps to correlate kernel-space memory consumption owing to user-space processes.
+test_check: stap -p4 vm.tracepoints.stp
+test_installcheck: stap vm.tracepoints.stp -c "sleep 10"
diff -rupaN a/testsuite/systemtap.examples/memory/vm.tracepoints.stp b/testsuite/systemtap.examples/memory/vm.tracepoints.stp
--- a/testsuite/systemtap.examples/memory/vm.tracepoints.stp	1969-12-31 19:00:00.000000000 -0500
+++ b/testsuite/systemtap.examples/memory/vm.tracepoints.stp	2009-12-04 06:21:38.000000000 -0500
@@ -0,0 +1,18 @@
+global slabs
+
+probe vm.kmem_cache_alloc {
+	slabs [execname(), bytes_req]<<<1
+}
+
+probe timer.ms(10000)
+{
+	dummy = "";
+	foreach ([name, bytes] in slabs) {
+		if (dummy != name)
+			printf("\nProcess:%s\n", name);
+        	printf("Slab_size:%d\tCount:%d\n", bytes, @count(slabs[name, bytes]));
+		dummy = name;
+	}
+	delete slabs
+	printf("\n-------------------------------------------------------\n\n")
+}

Thanks
-- 
Rajasekhar Duddu (rajduddu@linux.vnet.ibm.com),
Linux on System z - CSVT, IBM LTC, Bangalore.