[RFC] rename nvidia-driver port binaries [ and other comments]

Mon May 15 23:14:12 UTC 2017

On Mon, 15 May 2017 10:51:51 -0700 (PDT)
"Jeffrey Bouquet" <jbtakk at iherebuywisely.com> wrote:

> 
> 
> On Mon, 15 May 2017 23:53:10 +0900, Tomoaki AOKI <junchoon at dec.sakura.ne.jp> wrote:
> 
> > Hi.
> > If including its version to nvidia.ko and nvidia-modeset.ko,
> > [hard / symbolic] link to it with current name should be needed
> > not to bother admins every time upgrading.
> > 
> > BTW, I personally using below in rc.conf for safety.
> > This only helps situations that...
> > 
> >  *Insufficient loader memory, but OK after kernel is started.
> > 
> >  *Accidentally backed to older version that doesn't have
> >   nvidia-modeset.ko and only nvidia-modeset.ko is written in
> >   /boot/loader.conf.
> > 
> > ## Temporary settings for nvidia-driver when failed loading via loader.conf
> > kldstat -q -n nvidia.ko
> > if [ 0 -ne $? ] ; then
> >   echo "Loading nvidia-driver modules via rc.conf."
> > #  kldload nvidia
> >   if [ -e /boot/modules/nvidia-modeset.ko ] ; then
> > #    kldload nvidia-modeset
> >     kld_list="${kld_list} nvidia-modeset.ko"
> >   else
> > #    kldload nvidia
> >     kld_list="${kld_list} nvidia.ko"
> >   fi
> > fi
> > 
> > 
> > On Mon, 15 May 2017 06:41:33 -0700 (PDT)
> > "Jeffrey Bouquet" <jbtakk at iherebuywisely.com> wrote:
> > 
> > >  Just had a unique to me, unbootable backup [beside the point,
> > > just a sidebar comment... ]  quandry dealing with the nvidia-driver update
> > > that  mesa-libs needed.  [ or was appurtenant to it, unsure... ]
> > > 
> > >   12.0 - CURRENT
> > > 
> > >   [ my previous 'saves' -- files to consult...  were in .jpg, so no avail for me to peruse as I was in the terminal ]
> > > 
> > > Lessons I learned, maybe
> > > 
> > > [RFC]  rename nvidia-driver.ko to include its version, for instance nvidia-modeset-37539.ko
> > >           which would make one's diagnosis a fix easier. 
> > > 
> > > [suggestions]
> > > 
> > > ... Document that the kernel will load a /boot/modules/_nvidia.ko if you'd thus made it 'unavailable'
> > > ... document better that one can load nvidia[modeset] later than /boot/loader.conf [cli, rc.conf...]
> > > 
> > > [ mini 'what fixed it for you '  and/or stalled the same ... 
> > > ] 
> > > ... I had a make.conf in place, 'trapping' a
> > >     make -DMAKE_JOBS_UNSAFE=yes CC=".../clang39"  [and the other two...] install in failure
> > > ... I had no clue if Xorg was at fault, and luckily did not have to rebuild.
> > > ... I had no access to the forums, as the desktop could not be loaded [yet]
> > > ... I had to do strings nvidia.ko |grep 26 [39] | less to detect the not usable module numbers,
> > >     THUS THE REQ FOR RENAME of the nvidia.ko binary [vs port] to include its 5 digit number
> > >     as above.
> > > ... I was taken aback by the less infomative
> > >     this client has the .39 but the module has the .26 ... 
> > >     specifically stating which file/port/kernel module the 'client' refers to
> > >    and which specific modules 'this module' was referring to.  Unknown if that is
> > >    fixable here in a .c, .h , or is closed-source upstream in nvidia [corp ] where we can
> > >     ask them to be more specific or code in some more verbose error messages.
> > > ...  Relieved I did not have to rebuild Xorg nor the kernel, just move files out of the
> > >   way and do a simple rebuild of the nvidia-driver.
> > > [... Not relieved that the nvidia-driver needed install during the mesa-lib reshuffle,  maybe
> > >    did not need to be, just a hazy recollection on my part.  So ignore this little bit ]
> > > 
> > > ... All in all a learning experience.  Howsoever, I would like this nvidia stuff to be put eventuall
> > >  in /usr/src/UPDATING so the half or third of persons who run into this yearly will have a more
> > >  authoritative flowchart of sorts to determine the next course of action.
> > >    something like
> > >   if ... this thta
> > >    if... this that
> > >    if ... this that
> > >   OTOH... error message... a... you may need to...
> > >        ...
> > >                error message ... r... you may need to...
> > > 
> > >  ... And it would have maybe saved a bit of time here, and I would almost if eventually
> > >      possible be sure to copy /usr/src/UPDATING to /usr/ports/x11/nvidia-driver for
> > >      more easy access if the desktop was broken....
> > > 
> > > 
> > > Apologies for the hurried post. Indeed, I have tasked myself to write it up here
> > > locally [ still in scribbled notations...] so I can forestall/fix more quickly this
> > > error the next time.
> > > also...
> > > Running late  in other personal matters, and out of time.
> > > 
> > > Respectfully..
> > > 
> > > J. Bouquet 
> > > _______________________________________________
> > > freebsd-current at freebsd.org mailing list
> > > https://lists.freebsd.org/mailman/listinfo/freebsd-current
> > > To unsubscribe, send any mail to "freebsd-current-unsubscribe at freebsd.org"
> > > 
> > 
> > 
> > -- 
> > Tomoaki AOKI    <junchoon at dec.sakura.ne.jp>
> 
> 
> I found just then that one can load it manually
> first, though the nvidia.ko  [ one has newly built ] THEN
> the nvidia-modeset.ko  

nvidia-modeset.ko requires nvidia.ko and loads it automatically
if not yet loaded. So trying nvidia-modeset.ko first if exists,
and if not, trying nvidia.ko would be sufficient.

> if it has not been loaded via /boot/loader.conf which did not load it
> in the session in which I am typing this, for example.
> 
> [ wondering if /usr/ports/UPDATING or the pkg-message should be 
> updated to state that...  upon a second verification ... ] 

In /usr/ports/UPDATING, stated below:

20160829:
  AFFECTS: users of x11/nvidia-driver
  AUTHOR: cem at FreeBSD.org

  The NVidia driver has been updated to version 367.35.  Starting with
  version 358.09, new kernel module was added, nvidia-modeset.ko.  This
  new driver component works in conjunction with the nvidia.ko kernel
  module to program the display engine of the GPU.

  Users that experience hangs when starting X11 server, or observe

        (II) NVIDIA(0): Validated MetaModes:
        (II) NVIDIA(0):     "NULL"

  messages in their /var/log/Xorg.0.log file should replace ``nvidia''
  with ``nvidia-modeset'' in /boot/loader.conf or /etc/rc.conf files,
  depending on how they prefer to load NVidia driver kernel module.

> 
> 
> YMMV 
> _______________________________________________
> freebsd-current at freebsd.org mailing list
> https://lists.freebsd.org/mailman/listinfo/freebsd-current
> To unsubscribe, send any mail to "freebsd-current-unsubscribe at freebsd.org"
> 

-- 
Tomoaki AOKI    <junchoon at dec.sakura.ne.jp>