Endless gc spining on nearly full JFFS2 fs

Endless gc spining on nearly full JFFS2 fs

[Frederic Giasson]

Hi,
 
    I have noticed that the gc sometimes seems to spin-around for a very
long time, if not endlessly, when files are copied to a nearly full JFFS2
drive.  

	For example, I have done some tests copying a huge, uncompressable
file whose size is a little less than the total space left on a jffs2 drive.
When I copy the file on an empty, freshly-mounted file system, the "cp"
command returns after a normal time.  But if I erase the file and try to
copy it back, the gc seems to spin-around, erasing the same sectors many
times.  I have put traces in my chip driver which shown that the same
sectors are erased many times during the process.

	I have done the same test with a set of smaller files (5 binary
executable files ranging from 300 KB to 1.7MB).  I let the test running
overnight and when I returned the next morning, it had completed after more
than 5 hours!  With this set of files, my JFFS2 drive is 91% used.

	I am using kernel 2.4.19 pre-7 and JFFS2 code included in this
version.  I have a 4MB AMD flash device, from which I withdraw 2 sectors
(128KB) for my boot loader and stuff, using a map driver.  This leaves
4063232 bytes for the JFFS2 drive.


Is it normal according to what JFFS2 has been designed for?  Does someone
else have ran similar tests with large files and obtained same results?   Is
there a way to avoid that?
 
Thanks,
 
Frédéric Giasson

Re: Endless gc spining on nearly full JFFS2 fs

[David Woodhouse]

fgiasson@mediatrix.com said:
>     I have noticed that the gc sometimes seems to spin-around for a
> very long time, if not endlessly, when files are copied to a nearly
> full JFFS2 drive.  

Hmmm. This shouldn't happen. Can you provide a log with the debugging 
enabled? Somehow, some space is being marked dirty when we're 
garbage-collecting, so it thinks it can still make some progress.

--
dwmw2