Re: [fuse-devel] Some patches for embedded systems...

[Miklos S. <mi...@sz...>]

> I'm compiling it with 2.4.20 and 2.6.9-uc0 without support for sysfs,
> and the kernel/dev.c doesn't check if CONFIG_SYSFS is defined. Is it ok
> to check that in linux/autoconf.h ?  Does it work for all platforms?

It doesn't need to check I think, because if CONFIG_SYSFS isn't
defined all sysfs functions become noop (see linux/sysfs.h).

> alloc_inode, destroy_inode and inode_init_once is new from kernel
> 2.4.21. I'm compiling it with 2.4.20 on the WRT54G router.

Hmm.  I'll see how hard it is to use old inode alloc code for old 2.4
kernels.  Maybe the best strategy, is just to use FUSE 1.4 for such
old kernels, and when kernel is upgraded, then use the new FUSE.

> check_version() should look for "/proc/fs/fuse/version" if
> "/sys/fs/fuse/version" doesn't exist.

It does, doesn't it?

> Timed mutexes doesn't work 100% (which is defaulted from the call
> pthread_mutex_init(&m, NULL); ) with uClibc's libpthread support
> on Coldfire processors at least.

So why is that the default if it doesn't work?  This sounds like a
uClibc bug: they should either make it work properly or make the
default different.

> fork() can't be used if processor is without MMU and uClibc is
> compiled without UCLIBC_HAS_MMU define.
> 
> mtab and capabilities isn't available for embedded systems with
> uClibc. (read-only root-filesystem)

OK.  Do you have a patch?

Thanks,
Miklos