lkl: Add FreeBSD platform support

[cemeyer]

Please don't pull this as-is! In particular, the first few commits are big hacks that worked around issues only present on FreeBSD.

For example, without "Hack NR_CPUS" I got some warning about an invalid range; "timeconst.bc" hacks around limitations in FreeBSD's bc(1) program by hardcoding 250 hz in the bc script; and "Build 64-bit" hardcodes CONFIG_64BIT (without which it seemed to try and build 32-bit on a 64-bit host).

Despite those kludges, there are some decent portability-improving patches in here that won't hurt Linux or NT support. And some fixes for compiler or other issues (e.g. statfs64 was missing a 3rd argument which resulted in -EINVAL on all calls).

With this patchset I have ext4 mounting r/w on FreeBSD w/ lklfuse. Neat!

[tavip]

Awesome ! I will test and cherry pick most of the patches and rework some - "Build 64bit" can use the OUTPUT_FORMAT to detect that host is 64bit and enable 64bit build. Thanks !

[cemeyer]

Thanks!

[tavip]

Hi Conrad,

I have created a new branch (lkl_fbsd) which has your fixes. I've added your Signed-off-by: to your commits to adhere with the Linux process, hope that is ok.

I've also squashed a few commits and reworked some. Could you take a look and give it a try? Hope I didn't broke anything :)

Thanks,
Tavi

[cemeyer]

I've added your Signed-off-by: to your commits to adhere with the Linux process, hope that is ok.

Fine with me. (Edit: if it's not too difficult, can the Signed-off-by address be changed to cem@FreeBSD.org?)

I've also squashed a few commits and reworked some. Could you take a look and give it a try? Hope I didn't broke anything :)

I'll take a look, thanks.

Edit:

Re: "lkl: convert makefile echo \t to inline tab", it worked on FreeBSD :). If the inline tab works everywhere that's even better.

Re: "error: ignoring return value of ‘write’, declared with attribute warn_unused_result [-Werror=unused-result]" (both of them) — I believe you can just cast the result to void.

(void)write(STDOUT_FILENO, str, len);

Instead of the unused stack variable.

Re: "lkl: add support for 64bit FreeBSD", excellent! It's way better than my hack, thanks!

The patches look good to me. I will try them out shortly (I suspect the bc script will still fail on FreeBSD, unfortunately). If you haven't already, I have some follow-up patches for lklfuse you could look at in https://github.com/cemeyer/linux/tree/cem/lklfuse . I don't have XFS or BTRFS actually mounting yet; currently the kernel seems to get wedged at

[    0.004000] clocksource: jiffies: mask: 0xffffffff max_cycles: 0xffffffff, max_idle_ns: 7645041785100000 ns
[    0.004000] xor: measuring software checksum speed

[cemeyer]

A test build on lkl_fbsd branch:

...
  CHK     include/generated/timeconst.h
dc: not a string
  UPD     include/generated/timeconst.h
...
In file included from include/linux/jiffies.h:10:0,
                 from include/linux/ktime.h:25,
                 from include/linux/rcupdate.h:47,
                 from include/linux/srcu.h:33,
                 from include/linux/notifier.h:15,
                 from include/linux/memory_hotplug.h:6,
                 from include/linux/mmzone.h:812,
                 from include/linux/gfp.h:5,
                 from include/linux/kmod.h:22,
                 from include/linux/module.h:13,
                 from init/main.c:15:
include/generated/timeconst.h:11:2: error: #error "include/generated/timeconst.h has the wrong HZ value!"
 #error "include/generated/timeconst.h has the wrong HZ value!"
  ^
include/generated/timeconst.h:14:2: error: #error Totally bogus HZ value!
 #error Totally bogus HZ value!
  ^
include/generated/timeconst.h:15:1: error: expected identifier or '(' before numeric constant
 0
 ^
...

Yup, the bc script is still a problem. I think it may be easiest to port GNU bc to FreeBSD — the FreeBSD native bc is too limited for the needs of this script.

[cemeyer]

With just the BC hack from before and a second patch (-largp for fs2tar got dropped from cemeyer@8c2495a6), lkl_fbsd builds again: https://github.com/cemeyer/linux/tree/lkl_fbsd_plus Edit: And I can mount a USB ext4 under FUSE mostly fine. O_CREAT is broken without either handling it in open(2)or adding the FUSE create() method (I added it in my cem/lklfuse branch: cemeyer@ef78131).

[tavip]

Fine with me. (Edit: if it's not too difficult, can the Signed-off-by address be changed to >cem@FreeBSD.org?)

Yep, I'll do that. I guess I should also change the Author address, right?

Re: "error: ignoring return value of ‘write’, declared with attribute warn_unused_result [-Werror=unused->result]" (both of them) — I believe you can just cast the result to void.

With the new compilers (at least on Ubuntu 14.04) the (void) cast trick does not work anymore apparently due to better optimizations :)

-largp for fs2tar got dropped from cemeyer/linux@8c2495a),

Oops, I'll fix that.

I suspect the bc script will still fail on FreeBSD, unfortunately

Not sure how to deal with this. Can we keep it separate for now until maybe we find a better solution?

I'll rebase the patches to address the above and force push to the same branch and way you say ok I will merge it to the main branch.

[cemeyer]

Yep, I'll do that. I guess I should also change the Author address, right?

Sure.

With the new compilers (at least on Ubuntu 14.04) the (void) cast trick does not work anymore apparently due to better optimizations :)

I think it is just GCC being very pedantic about warn_unused_result.

I suspect the bc script will still fail on FreeBSD, unfortunately

Not sure how to deal with this. Can we keep it separate for now until maybe we find a better solution?

Yes. I think I'll try and import GNU bc into FreeBSD ports. After that, we may want a patch that changes bc invocations to $(BC) and sets BC to bc on Linux and gbc on FreeBSD. I think that's a straightforward way of getting where we want.

I'll rebase the patches to address the above and force push to the same branch and way you say ok I will merge it to the main branch.

Great, thanks!