mirror of
https://github.com/apache/nuttx.git
synced 2026-08-01 20:28:58 +00:00
b2b78d2f8a
This adds a kernel-level performance profiler for the VFS. By enabling CONFIG_FS_PROFILER, the core VFS system calls (file_read, file_write, file_open, and file_close) are instrumented to track high-resolution execution times using clock_systime_timespec() seamlessly. The collected statistics are exposed dynamically via a new procfs node at /proc/fs/profile, allowing CI regression testing without needing external debugging tools. Signed-off-by: Sumit6307 <sumitkesar6307@gmail.com>
38 lines
1.4 KiB
ReStructuredText
38 lines
1.4 KiB
ReStructuredText
==============================
|
|
VFS Performance Profiler
|
|
==============================
|
|
|
|
The Virtual File System (VFS) Performance Profiler provides a simple, in-kernel
|
|
mechanism to track execution times and invocation counts for core VFS operations
|
|
(read, write, open, close) seamlessly. This is highly suitable for
|
|
CI/CD automated regression testing and performance bottleneck identification.
|
|
|
|
Configuration
|
|
=============
|
|
|
|
To enable the profiler, select ``CONFIG_FS_PROFILER`` in your Kconfig.
|
|
To expose the metrics dynamically via procfs, ensure ``CONFIG_FS_PROCFS`` is enabled, and
|
|
the profiler node is included via ``CONFIG_FS_PROCFS_PROFILER``.
|
|
|
|
Usage
|
|
=====
|
|
|
|
When enabled, the profiler automatically intercepts calls to the underlying
|
|
inode operations and records the execution elapsed times using ``perf_gettime()``.
|
|
Since no blocking mutexes are used during updates (fast ``atomic.h`` operations
|
|
are utilized instead), the overhead is extremely minimal and safely scales on SMP.
|
|
|
|
To view the current statistics collectively from the NuttShell (NSH), simply
|
|
read the node:
|
|
|
|
.. code-block:: bash
|
|
|
|
nsh> cat /proc/fs/profile
|
|
VFS Performance Profile:
|
|
Reads: 12 (Total time: 4500120 ns)
|
|
Writes: 3 (Total time: 95050 ns)
|
|
Opens: 15 (Total time: 1005000 ns)
|
|
Closes: 15 (Total time: 45000 ns)
|
|
|
|
The reported times are in the raw ticks/units provided by ``perf_gettime()`` on
|
|
your specific architecture.
|