Kernel Small Stacks - Revision history

Tim Bird: /* Stack tracer in ftrace */

2013-03-21T18:28:24Z

‎Stack tracer in ftrace

2013-03-21T18:24:58Z

‎Stack tracer in ftrace

2011-11-29T21:30:10Z

‎Stack tracer in ftrace (or perf?)

2011-11-29T21:26:31Z

‎CONFIG_DEBUG_STACK_USAGE

2011-11-29T21:25:50Z

‎Stack limit patches

2011-11-29T21:17:31Z

‎Stack tracer in ftrace

2011-11-29T21:08:02Z

‎ARM 4K Stacks

2011-11-29T21:06:45Z

‎ARM 4K Stacks

2011-11-29T21:05:06Z

‎Stack tracer in ftrace (or perf?)

2011-11-29T21:02:01Z

‎CONFIG_DEBUG_STACK_USAGE

@@ Line 256: / Line 256: @@
 === Stack tracer in ftrace ===
-For detailed instructions, see: http://elinux.org/Ftrace#find_deepest_kernel_stack
+For detailed instructions, see: [[Ftrace#Find_deepest_kernel_stack]]
 Rough notes:

@@ Line 255: / Line 255: @@
 === Stack tracer in ftrace ===
-Rough notes: Turn on CONFIG_STACK_TRACER in kernel config
+For detailed instructions, see: http://elinux.org/Ftrace#find_deepest_kernel_stack
-pass 'stacktrace' on kernel command line, or at runtime do:
-echo 1 >/proc/sys/kernel/stack_tracer_enabled
-mount -t debugfs none /sys/kernel/debug
+Rough notes:
+* Turn on CONFIG_STACK_TRACER in kernel config
-look for values in /sys/kernel/debug/tracing/stack_trace
+* pass 'stacktrace' on kernel command line, or at runtime do:
+* echo 1 >/proc/sys/kernel/stack_tracer_enabled
-cat /sys/kernel/debug/tracing/stack_trace
+* mount -t debugfs none /sys/kernel/debug
 See http://lwn.net/Articles/295955/

@@ Line 254: / Line 254: @@
   ash             R running   1264    56      1 0x00000000
-=== Stack tracer in ftrace (or perf?) ===
+=== Stack tracer in ftrace ===
 [FIXTHIS - add this section]
@@ Line 266: / Line 266: @@
 cat /sys/kernel/debug/tracing/stack_trace
 === Stack limit patches ===

@@ Line 196: / Line 196: @@
 In more detail, it does the following:
- * at process creation time, fills the stack with zeros (kernel/fork.c)
+* at process creation time, fills the stack with zeros (kernel/fork.c)
- * on sysrq 't', show free space, from call to stack_not_used() (kernel/sched.c)
+* on sysrq 't', show free space, from call to stack_not_used() (kernel/sched.c)
-   * it shows as 0 otherwise ??
+** it shows as 0 otherwise ??
- * define check_stack_usage(), which emits printks on each low-water hit
+* define check_stack_usage(), which emits printks on each low-water hit
-   * low-water appears to be global over all stacks
+** low-water appears to be global over all stacks
-   * check_stack_usage() is only called on process exit, so you might
+** check_stack_usage() is only called on process exit, so you might not know about a problem process until very late
-   not know about a problem process until very late
+* stack_not_used() is defined in include/linux/sched.h.  It counts the number of zero bytes following the end of thread_info going up.
- * stack_not_used() is defined in include/linux/sched.h.  It counts the number of
- zero bytes following the end of thread_info going up.
 As the systems runs, any time the stack low-water mark is exceeded, then

← Older revision		Revision as of 21:25, 29 November 2011
Line 276:		Line 276:

	These patches do the following:		These patches do the following:
		+	Add the config options:
		+	* CONFIG_SNSC_DEBUG_STACK_LIMIT - perform stack layout changes
		+	* CONFIG_SNSC_SUPPORT_4KB_MAPPING - re-map kernel memory for 4K TLB mappings
		+	* CONFIG_SNSC_DEBUG_STACK_LIMIT_MANUAL - allow a user-specified starting stack size
		+
		+
		+	the patches do several things:
		+	* change the stack layout to place thread_info at the top of the stack rather than at the bottom
		+	* change the stack size to 16KB (order 2)
		+	* allow configuring the default starting position of the stack, to simulate an arbitrary stack size (default is right below thread_info)
		+	* remap the kernel memory so that 4KB mappings are used
		+	** this allows unmapping the bottom page of the stack, so that a fault occurs when the page is accessed (on a stack overflow for a manually configured small stack)
		+	** this is only turned on if you specify 'use_4kb_mapping' on the
		+	* unmap the bottom page of the stack (the guard page)
		+	** this is only turned on if you specify 'unmap_stack'
		+
		+
		+	[ FIXTHIS - need to add more to this section]

	== ARM 4K Stacks ==		== ARM 4K Stacks ==

@@ Line 289: / Line 289: @@
       * question: for operation or just reporting??
-Dave Chinner ([http://www.redhat.com/archives/rhl-list/2007-March/msg00854.html here]) wrote:
+Dave Chinner ([https://lkml.org/lkml/2011/10/19/56 here]) wrote:
 <pre>
 There's a good reason 4k stacks went away: it's

@@ Line 289: / Line 289: @@
       * question: for operation or just reporting??
-Dave Chinner ([ here]) wrote:
+Dave Chinner ([http://www.redhat.com/archives/rhl-list/2007-March/msg00854.html here]) wrote:
 <pre>
 There's a good reason 4k stacks went away: it's
@@ Line 307: / Line 307: @@
 This is a pretty good example of people denying a use case with a red
 herring.
 == Possible mixed stack size feature ==