| 1 |
On Sat, Aug 31, 2013 at 10:55 PM, Walter Dnes <waltdnes@××××××××.org> wrote: |
| 2 |
>> I usally use ext4 as filesystem. |
| 3 |
>> |
| 4 |
>> # lsmod|grep ext |
| 5 |
>> ext3 100768 0 |
| 6 |
>> jbd 39586 1 ext3 |
| 7 |
>> ext2 49572 0 |
| 8 |
>> ext4 263621 1 |
| 9 |
>> crc16 1255 2 ext4,bluetooth |
| 10 |
>> mbcache 4450 3 ext2,ext3,ext4 |
| 11 |
>> jbd2 48679 1 ext4 |
| 12 |
>> |
| 13 |
>> Isn't great what an initramfs can do? |
| 14 |
> |
| 15 |
> In this case, initramfs is your root filesystem, from which you load |
| 16 |
> another fs and then transfer (pivot root?) to it. You have to build |
| 17 |
> initramfs support into the kernel, to boot an initramfs. So my argument |
| 18 |
> still stands, regardless of whether your *INITIAL* filesystem is ext4fs, |
| 19 |
> or ZFS, or initramfs, that *INITIAL* filesystem has to be built into the |
| 20 |
> kernel. |
| 21 |
|
| 22 |
Interesting perspective. Of course, support for an initramfs is not |
| 23 |
actually a file system (it's not even in the File systems section of |
| 24 |
the kernel configuration, is in General setup); it's not possible to |
| 25 |
have initramfs as a module (that would make no sense at all); and it's |
| 26 |
code that is several orders of magnitude more simpler than the one |
| 27 |
used by ext4 (or any other journal file system). |
| 28 |
|
| 29 |
But you are right that for booting with an initramfs, you need |
| 30 |
initramfs support. |
| 31 |
|
| 32 |
> Also, I really wonder what the point is in having to use |
| 33 |
> initramfs on a system where /usr is part of /. |
| 34 |
|
| 35 |
Well, since some months ago I've been running as a module almost |
| 36 |
everything that can be compiled as a module. This allows me to run a |
| 37 |
*truly* minimal kernel, and only the necessary modules autoload |
| 38 |
automatically (one big exception: binfmt_script, I compiled that into |
| 39 |
the kernel because it was not loading automatically). I can also |
| 40 |
unload some modules when not in use anymore (and this is great to |
| 41 |
debug sometimes). |
| 42 |
|
| 43 |
This also lets me to add a lot of stuff in the kernel, as long as I |
| 44 |
add them as modules, without me worrying about bloating my kernel. |
| 45 |
Only when they are needed they are loaded. I have USB speakers, but I |
| 46 |
almost never use them; no problem, they (like almost everything else) |
| 47 |
live as modules, and only are loaded (automagically, thanks to udev) |
| 48 |
when needed. And again, I can unload them when not in use. |
| 49 |
|
| 50 |
And also, it turns out that by using dracut+systemd you could boot |
| 51 |
faster than without initramfs (although I can't find the link |
| 52 |
anymore). |
| 53 |
|
| 54 |
Finally, using only modules and dracut liberates me from thinking what |
| 55 |
should it be compiled in and what not; I just put *everything* as a |
| 56 |
module, and the kernel, udev and dracut take care of loading what's |
| 57 |
necessary. Thus, my kernel (the one running in memory) is as minimal |
| 58 |
as it can be, all the time. |
| 59 |
|
| 60 |
Oh, and one more thing; by having everything as a module, if suddenly |
| 61 |
I need support for new hardware, usually I can do a quick "make |
| 62 |
menuconfig; make modules_install", and the new module can be |
| 63 |
modprobe'd into the kernel without needing a reboot. That's |
| 64 |
convenient. |
| 65 |
|
| 66 |
Regards. |
| 67 |
-- |
| 68 |
Canek Peláez Valdés |
| 69 |
Posgrado en Ciencia e Ingeniería de la Computación |
| 70 |
Universidad Nacional Autónoma de México |
Archives