Re: [RFC: PATCH v2] lockdep: Make MAX_STACK_TRACE_ENTRIESconfigurable.

[John Kacur]

On Fri, 16 Apr 2010, John Kacur wrote:

> 
> This patch was created against v2.6.33.2-rt13 but is also intended for 
> mainline.
> 
> Certain debug configurations tend to go over the MAX_STACK_TRACE_ENTRIES
> limit without there really being any problem.
> 
> This patch keeps the old default, but allows the user to configure a 
> higher value if desired.
> 
> I tend to bump this value up for real-time debug configurations for 
> example. This is preferrable to indiscriminately turning off the locking 
> correctness validator.
> 
> There have been some attempts to increase the default value in the past, 
> that were met with resistance by some, because of the legitimate concern 
> that this was growing too large and that we need to understand why. By 
> making it configurable, I hope to satisfy both sets of people - those who 
> believe they need to set it larger under special circumstances, and those 
> who want a reasonably small default.
> 
> From 57479e7620d312353f5f5b173742e011896a9c74 Mon Sep 17 00:00:00 2001
> From: John Kacur <jkacur@xxxxxxxxxx>
> Date: Fri, 16 Apr 2010 13:24:02 +0200
> Subject: [RFC: PATCH] lockdep: Make MAX_STACK_TRACE_ENTRIES configurable.
> 
> Certain configurations that have LOCKDEP turned on, run into the limit
> where the MAX_STACK_TRACE_ENTRIES are too small. Rather than simply
> turning of the locking correctness validator let the user configure this
> value to something reasonable for their system.
> 
> Signed-off-by: John Kacur <jkacur@xxxxxxxxxx>
> ---
>  kernel/lockdep.c           |    8 ++++----
>  kernel/lockdep_internals.h |    6 ------
>  lib/Kconfig.debug          |    9 +++++++++
>  3 files changed, 13 insertions(+), 10 deletions(-)
> 
> diff --git a/kernel/lockdep.c b/kernel/lockdep.c
> index 1199bda..6030521 100644
> --- a/kernel/lockdep.c
> +++ b/kernel/lockdep.c
> @@ -368,12 +368,12 @@ static int verbose(struct lock_class *class)
>   * addresses. Protected by the graph_lock.
>   */
>  unsigned long nr_stack_trace_entries;
> -static unsigned long stack_trace[MAX_STACK_TRACE_ENTRIES];
> +static unsigned long stack_trace[CONFIG_MAX_STACK_TRACE_ENTRIES];
>  
>  static int save_trace(struct stack_trace *trace)
>  {
>  	trace->nr_entries = 0;
> -	trace->max_entries = MAX_STACK_TRACE_ENTRIES - nr_stack_trace_entries;
> +	trace->max_entries = CONFIG_MAX_STACK_TRACE_ENTRIES - nr_stack_trace_entries;
>  	trace->entries = stack_trace + nr_stack_trace_entries;
>  
>  	trace->skip = 3;
> @@ -395,11 +395,11 @@ static int save_trace(struct stack_trace *trace)
>  
>  	nr_stack_trace_entries += trace->nr_entries;
>  
> -	if (nr_stack_trace_entries >= MAX_STACK_TRACE_ENTRIES-1) {
> +	if (nr_stack_trace_entries >= CONFIG_MAX_STACK_TRACE_ENTRIES-1) {
>  		if (!debug_locks_off_graph_unlock())
>  			return 0;
>  
> -		printk("BUG: MAX_STACK_TRACE_ENTRIES too low!\n");
> +		printk("BUG: CONFIG_MAX_STACK_TRACE_ENTRIES = %d too low!\n", CONFIG_MAX_STACK_TRACE_ENTRIES);
>  		printk("turning off the locking correctness validator.\n");
>  		dump_stack();
>  
> diff --git a/kernel/lockdep_internals.h b/kernel/lockdep_internals.h
> index a2ee95a..6887711 100644
> --- a/kernel/lockdep_internals.h
> +++ b/kernel/lockdep_internals.h
> @@ -61,12 +61,6 @@ enum {
>  
>  #define MAX_LOCKDEP_CHAIN_HLOCKS (MAX_LOCKDEP_CHAINS*5)
>  
> -/*
> - * Stack-trace: tightly packed array of stack backtrace
> - * addresses. Protected by the hash_lock.
> - */
> -#define MAX_STACK_TRACE_ENTRIES	262144UL
> -
>  extern struct list_head all_lock_classes;
>  extern struct lock_chain lock_chains[];
>  
> diff --git a/lib/Kconfig.debug b/lib/Kconfig.debug
> index cbf6e02..ad35402 100644
> --- a/lib/Kconfig.debug
> +++ b/lib/Kconfig.debug
> @@ -509,6 +509,15 @@ config LOCKDEP
>  	select KALLSYMS
>  	select KALLSYMS_ALL
>  
> +config MAX_STACK_TRACE_ENTRIES
> +	int "MAX_STACK_TRACE_ENTRIES for LOCKDEP"
> +	depends on LOCKDEP
> +	default 262144
> +	help
> +	   This option allows you to change the default MAX_STACK_TRACE_ENTRIES
> +	   used for LOCKDEP. Warning, increasing this number will increase the
> +	   size of the stack_trace array, and thus the kernel size too.
> +
>  config LOCK_STAT
>  	bool "Lock usage statistics"
>  	depends on DEBUG_KERNEL && TRACE_IRQFLAGS_SUPPORT && STACKTRACE_SUPPORT && LOCKDEP_SUPPORT
> -- 
> 1.6.6.1
> 
> 

I missed the MAX_STACK_TRACE_ENTRIES in kernel/lockdep_proc.c
This corrects that.