dynamic debug v2

hi,

I've updated the userspace interface and updated the implementation. I'm going to summarize what has changed below, but for more background, please see:
http://marc.info/?l=linux-kernel&m=121338359012058&w=2

INTERFACE - userspace

The file <debugfs>/dynamic_printk/modules is now formatted as:

<module_name> <enabled=0/1> <level=[0-n]> <flag_name=0/1>....
<associated module names>
.
.
.

<module_name> : Name of the module in which the debug call resides <enabled=0/1> : whether the the messages are enabled or not <level=[0-n]> : For modules that support levels <flag_name=0/1> : names of the flags that can be set <associated module names> : names of the modules that are part of
group <module_name>

Excerpt from my system:

snd_hda_intel enabled=0

fixup enabled=0

cpufreq_shared enabled=0 CPUFREQ_DEBUG_CORE=0 CPUFREQ_DEBUG_DRIVER=0 CPUFREQ_DEBUG_GOVERNOR=0
acpi_cpufreq
freq_table
cpufreq_userspace
cpufreq_performance
cpufreq

driver enabled=0

dummy enabled=0

snd_seq enabled=0

Enable a module:

$echo "set enabled=1 <module_name>" > dynamic_printk/modules

Disable a module:

$echo "set enabled=0 <module_name>" > dynamic_printk/modules

To set the level or flag value for type 'level' or 'flag':

$echo "set level=<#> <module_name>" > dynamic_printk/modules

Enable all modules:

$echo "set enabled=1 all" > dynamic_printk/modules

Disable all modules:

$echo "set enabled=0 all" > dynamic_printk/modules

IMPLEMENTATION

The set of modules that can be dynamically controlled via /sys/kernel/debug/dynamic_printk/modules are created on the fly during bootup. Meta-data is implicity associated with each 'pr_debug()', dev_dbg()' and 'dynamic_debug_enabled()' call site. This meta-data includes the module name and the debug type, (one of boolean, flag, or level). The meta-data is contained in a special "verbose" section of the kernel binary. So this section is essentially scanned during bootup or module load/unload, and an internal kernel hash data structure is created. The "verbose" section is then freed. Thus, although the text size increases with this compile option (~2% in my testing), the runtime size remains the same (module the infrastructure code).

Two internal hash table are created as a hash on the module name. Each hash table has 64 buckets. Each bucket has a 'long long' mask associated with it, where bit 'n' set, if any module which hashes into the bucket row 'n' is enabled. Only, if both bits are set in the corresponding mask do we make a function call. This function call checks to see if the level or flags are appropriately set and that the module is enabled before outputing the printk (in the case of pr_debug() or dev_dbg()), or returning true or false, in the case of 'dynamic_debug_enabled()'. The hashes at each call site, have been computed during kernel compile, thus the overhead at each call site is simply checking for bits being set in the masks. I believe this provides very low overhead when only a handfull of modules are enabled.

thanks,

-Jason