CFT: Re: linux libusb again, I made an updated port...

[Wojciech A. Koszek]

On Tue, Feb 11, 2014 at 12:11:09AM +0100, Juergen Lock wrote:
> On Mon, Feb 10, 2014 at 11:06:27AM +0000, Bjoern A. Zeeb wrote:
> > 
> > On 10 Feb 2014, at 04:18 , Wojciech A. Koszek <wkoszek at FreeBSD.org> wrote:
> > 
> > > On nie, lut 09, 2014 at 02:59:06 +0100, Juergen Lock wrote:
> > >> On Sun, Feb 09, 2014 at 02:56:24AM +0000, Wojciech A. Koszek wrote:
> > >>> On sob, lut 08, 2014 at 09:45:46 +0100, Juergen Lock wrote:
> > >>>> On Fri, Feb 07, 2014 at 08:49:28PM +0000, Wojciech A. Koszek wrote:
> > >>>>> On pi??, lut 07, 2014 at 09:12:08 +0100, Juergen Lock wrote:
> > >>>>>> Hi!
> > >>>>>> 
> > >>>>>> This came up on irc so I tried to build a linux libusb port (before
> > >>>>>> I learned about ports/146895), mine uses linux_base-gentoo-stage3
> > >>>>>> like linux_kdump with a src/lib/libusb head snapshot so it's more
> > >>>>>> up to date than wkoszek's build (ports/146895), and it's really
> > >>>>>> easy to update it again.  Also maybe it can be used as linux
> > >>>>>> libusb-1.0.so too; I didn't actually test it tho.
> > >>>>>> 
> > >>>>>> Should this be committed?  Is wkoszek's version better since it
> > >>>>>> also builds on < 10.x?  Comments welcome...
> > >>>>>> 
> > >>>>>> wkoszek's version:
> > >>>>>> 
> > >>>>>> 	http://www.freebsd.org/cgi/query-pr.cgi?pr=146895
> > >>>>>> 
> > >>>>>> Mine:
> > >>>>>> 
> > >>>>>> 	http://people.freebsd.org/~nox/tmp/linux_libusb.shar
> > >>>>>> 
> > >>>>>> Distfile:
> > >>>>>> 
> > >>>>>> 	http://people.freebsd.org/~nox/tmp/distfiles/linux_libusb-11.0r261448.tar.bz2
> > >>>>>> 
> > >>>>>> 10/amd64 package:
> > >>>>>> 
> > >>>>>> 	http://people.freebsd.org/~nox/tmp/packages/10amd64/linux_libusb-11.0r261448.txz
> > >>>>>> 
> > >>>>>> (built via:
> > >>>>>> 
> > >>>>>> 	poudriere bulk -v -j 10amd64 -p custom devel/linux_libusb
> > >>>>>> 
> > >>>>>> - btw for some reason the dependency emulators/linux_base-gentoo-stage3
> > >>>>>> doesn't build for 10i386 in poudriere bulk, I get a pkg segfault.  bapt
> > >>>>>> Cc'd...)
> > >>>>>> 
> > >>>>> 
> > >>>>> Juergen,
> > >>>> Hi!
> > >>>>> 
> > >>>>> What would be the reason for this update?
> > >>>>> 
> > >>>>> My stuff may be out of date, but it was all tested and working. I verified
> > >>>>> it with Linux'ish lsusb(1) and USB-based FPGA JTAG programmer, for which
> > >>>>> this stuff was written.
> > >>>>> 
> > >>>> I was just thinking an updated version may be useful, but if it's
> > >>>> already working for everyone maybe less so...
> > >>>> 
> > >>>> Or would it work as a linux libusb-1.0.so too?  I know the libusb 1.0
> > >>>> stuff added some functions since 9.x at least... maybe hps would know
> > >>>> (Cc'd.)
> > >>>> 
> > >>> 
> > >>> Juergen,
> > >>> 
> > >>> I think this package is useful and is looking for maintainer, so if you have
> > >>> time and energy, I'm OK with upgrading it, but I suggest testing it first.
> > >>> Bjoern might be interested too.
> > >>> 
> > >> You mean bz@ ?  Cc'd.  I tried testing lsusb from debian sid but it printed
> > >> nothing, neither with my nor with your older version, but maybe it's just
> > >> `too new' for our current linuxolator.
> > > 
> > > I assume you have at least 1 USB device while trying this. I don't remember
> > > exactly, but while trying within Linuxolator, you may need devfs/procfs to
> > > be mounted under Linuxolator's root directory.
> > 
> > My understanding and from looking at trace is that if we cannot find it in /compat/linux we ale search in /; so no need for an extra mount unless maybe you run chrooted.
> > 
> > 
> > > So you'll have to figure this out.
> > > 
> > > Does it return with 0 exit code?
> > > 
> > > If not, lsusb should be simple enough to let you place printf() all over the
> > > place and understand out when it's failing. 
> > 
> > For me the problem was a clock_gettime() call in the libusb which my glibc did not provide.   That made all things fail (silently) until I used linux ?rtld" tracing to see the unresolved symbol from libusb/the commercial 3rd party software dynamically loading libusb.
> > 
> And my linux_kdump ends like this:
> 
>  35607 lsusb    CALL  munmap(0x28247000,0x8000)
>  35607 lsusb    RET   munmap 0
>  35607 lsusb    CALL  linux_open(0x28090c31,0,0x28081e00)
>  35607 lsusb    NAMI  "/compat/linux/dev/usbctl"
>  35607 lsusb    NAMI  "/dev/usbctl"
>  35607 lsusb    RET   linux_open 3
>  35607 lsusb    CALL  linux_ioctl(0x3,0xffe5 ,0xffffc5d8)
>  35607 lsusb    RET   linux_ioctl 0
>  35607 lsusb    CALL  close(0x3)
>  35607 lsusb    RET   close 0
>  35607 lsusb    CALL  linux_exit_group(0x1)
>  35607 lsusb    UNKNOWN(11)
> 
> and the exit code is 1, so I guess it's failing on an unhandled
> syscall or something like that.

Juergen

Can you send the full trace with all eventual child processes? If you could
copy&paste:

	open libusb
	for each bus
		for each device
			open device

to a separate program, we could make it libusb/Linuxolator regression test.
This would also help us match ptrace<->line number output.

Which FreeBSD are you using?

This may indicate some relationship with

	http://www.freebsd.org/cgi/query-pr.cgi?pr=141439

-- 
Wojciech A. Koszek
wkoszek at FreeBSD.czest.pl
http://FreeBSD.czest.pl/~wkoszek/