On 16/02/2017 09:53, Rick Stevens wrote:
On 02/15/2017 01:27 PM, Stephen Morris wrote:
> On 15/02/2017 04:42, Rick Stevens wrote:
>> On 02/13/2017 04:38 PM, Rick Stevens wrote:
>>> On 02/13/2017 03:39 PM, Rick Stevens wrote:
>>>> On 02/13/2017 12:22 PM, Stephen Morris wrote:
>>>>> On 13/02/2017 19:21, Stephen Morris wrote:
>>>>> I've removed the driver from DKMS in Ubuntu and re-added it,
then
>>>>> run a
>>>>> DKMS AUTOINSTALL to rebuild and re-install the driver back into the
>>>>> running kernel and under Ubuntu the modprobe command to active the
>>>>> driver is not issued either, but also under Ubuntu is seems that
when
>>>>> the rmmod command is issued to remove the driver from the running
>>>>> kernel
>>>>> which removes wifi access, the dns resolver service seems to also be
>>>>> shutdown, because after the driver is rebuilt by DKMS and the
modprobe
>>>>> issued a sudo apt-get update fails to be able to resolve the
>>>>> repository
>>>>> names.
>>>>> I've tried the same thing under Fedora to see if Fedora has the
same
>>>>> functionality, which fortunately Fedora doesn't, after the DKMS
>>>>> process
>>>>> is done and the modprobe issued a sudo dnf update successfully
>>>>> refreshes
>>>>> the repository lists and installs any available updates after
>>>>> prompting
>>>>> whether or not to do so.
>>>> I've been watching this thread a bit and I'm not a DKMS user,
but
>>>> from what I can gather, DKMS should do an automatic modprobe when
>>>> a kernel is upgraded or whatever based on certain criteria. Section 2
>>>> of the webpage at:
>>>>
>>>>
https://wiki.centos.org/HowTos/BuildingKernelModules
>>>>
>>>> describes the process (the example is a patched CIFS filesystem). In a
>>>> nutshell:
>>>>
>>>> 1. The module's source directory must be in
>>>>
>>>> /usr/src/<module>-<version>
>>>>
>>>> 2. There is a "dkms.conf" file in that directory that
contains
>>>> some specific info.
>>>>
>>>> 3. It has been added to the DKMS tree via "dkms add"
>>>>
>>>> 4. It has been built under DKMS control via "dkms build"
>>>>
>>>> 5. It has been installed under DKMS control via "dkms
install"
>>>>
>>>> Note that I've never tried a DKMS add-on module, but that seems to
be
>>>> the way it's done. I think that dkms.conf file is pretty critical,
>>>> along with the "dkms add", "dkms build" and
"dkms install". I don't
>>>> believe "dkms autoinstall" would modprobe modules that
weren't added,
>>>> built and install using dkms as they weren't fully inserted under
dkms'
>>>> care. That's just a guess.
>>> I just built a module (evdi) under DKMS for a Mimo USB-based display. I
>>> don't have the hardware handy (it's here somewhere), but I copied
the
>>> driver source to /usr/src/evdi-1.2.65 and created a dkms.conf file in
>>> there that contains:
>>>
>>> PACKAGE_NAME="evdi"
>>> PACKAGE_VERSION=1.0.0
>>> AUTOINSTALL=yes
>>>
>>> MAKE[0]="make all INCLUDEDIR=/lib/modules/$kernelver/build/include
>>> KVERSION=$kernelver DKMS_BUILD=1"
>>> DEST_MODULE_LOCATION[0]="/kernel/drivers/gpu/drm/evdi"
>>> BUILT_MODULE_NAME[0]="evdi"
>>> CLEAN="make clean DKMS_BUILD=1"
>>>
>>> Then I did (as root):
>>>
>>> dkms add -m evdi -v 1.2.65
>>>
>>> That added it to dkms' build stack as witnessed by dkms creating a
>>> /var/lib/dkms/evdi/1.2.65 directory with a subdirectory named after the
>>> current kernel version and a symlink, "source", pointing at
>>> /usr/src/evdi-1.2.65. The next command:
>>>
>>> dkms build -m evdi -v 1.2.65
>>>
>>> actually built the driver, creating an "x86_64" subdirectory below
the
>>> kernel version directory with "log" and "module" subdirs
below that.
>>> The log subdirectory contains the build log of the module and the
>>> module subdir contains the actual module (evdi.ko). I then ran:
>>>
>>> dkms install -m evdi -v 1.2.65
>>>
>>> and saw dkms install the module into /lib/modules/`uname -r`/extra/ and
>>> do a full depmod:
>>>
>>> # dkms install -m evdi -v 1.2.65
>>>
>>> evdi:
>>> Running module version sanity check.
>>> - Original module
>>> - No original module exists within this kernel
>>> - Installation
>>> - Installing to /lib/modules/4.9.7-201.fc25.x86_64/extra/
>>> Adding any weak-modules
>>>
>>> depmod.......
>>>
>>> DKMS: install completed.
>>>
>>> Doing a "modinfo evdi" returns the data I'd expect.
>>>
>>> I won't know if it auto-modprobes until I get the new kernel update
>>> and find the Mimo display and plug it in. Just thought you'd like to
>>> see what I've done.
>> UPDATE: Just got the new kernel and dkms did build the module for the
>> kernel without me telling it to.
>>
>> I located the display device, plugged it in and it "just worked." I
>> did NOT see the module in the output of "lsmod" but the device did
>> work. I don't know why it doesn't show up in lsmod, but it may be that
>> this is an X/Wayland module only, even though it's a ".ko" file.
> Thanks for the info Rick, what you did to manually use dkms to build and
> install the driver in to the kernel is the same process I went through
> which didn't issue a modprobe to make the network device usable, so I
> couldn't access the network until I manually issued a modprobe which
> then made the device immediately usable.
Can't speak to that. The MIMO device worked when I plugged it in,
but the module was NOT listed in "lsmod", so the module may be X or
Wayland-specific and loaded by that software...not the kernel in
general.
My module is listed in lsmod as using cfg80211. Your device may have
worked when you plugged it in after dkms building it for the new kernel
because of the nature of plug-and-play. The whole nature of
plug-and-play is to sense a device has been attached to the pc and load
any necessary drivers for it to make it usable without the end user
having to do anything special.
> I also tried it a different way by issuing the same manual steps but
> each time specifying the -c parameter to point it to the dkms.conf file
> but that achieved the same results. This method though did highlight
> what I consider the be a bug in DKMS, in that I consider that if I
> specify the -c parameter I should not have to specify the -m and -v
> parameters, but my experience with it is that DKMS complains about
> insufficient parameters when running the build and install steps if -m
> and -v are not supplied.
I think you still have to specify the -m and -v because you don't
specify WHERE the module source is--the conf file could be for any
module and you still need to tell it what you're building.
You shouldn't
have to, the -m and -v parameters specify the name of the
directory in /usr/src that contains the source code (/usr/src is
compulsory) and within that directory it has to be able to find a
dkms.conf file to tell it how to build the module, but with the -c
parameter you have to specify the full path to the dkms.conf, so in your
case the parameter would be -c /usr/src/evdi-1.2.65/dkms.conf, and the
first two statements in dkms.conf specify the same information that are
specified with the -m and -v anyway, hence they should not be required.
Also the module that is being built it irrelevant to dkms, it doesn't
need to know, those details are taken care of by specifications in the
dkms.conf file, which in your case and my case is via make and the
associated source makefile plus the LOCATION[0] statement which may or
may not be honoured depending on the makefile.
> Just one comment on your dkms.conf file contents you have listed above,
> I'm surprised that your module gets compiled at boot time as the
> documentation I have read on the dkms.conf file dictates that the value
> for PACKAGE_VERSION must match the version part of the directory
> containing the source code, so in your case it should be 1.2.65 not 1.0.0.
Well, yes they probably should be the same, but the .conf file
specifies the package name and version--not the module name and
version. I think this is just used by dkms to track things. The module
builds just fine.
At boot time when the dkms autoinstall is run it goes to
/var/lib/dkms
to find any directories that are defined in there, which in my case
there is only one which is rtl8814AU, then it navigates into that
directory looking for sub directories, which in my case is a long name
that was specified in the -v, within that directory is a directory for
the kernel the module has been compiled into and a symlink called
'source' which points to the physical directory containing the source
(in your case it points to /usr/src/evdi-1.2.65). Having found that it
then tries to navigate to that symlink path and load the dkms.conf file,
it is this part that if there are multiple version directories with
source symlinks and the linked to paths don't physically exist or don't
contain a dkms.conf file that the message get produced and the process
stops. Having found the dkms.conf file it then uses the contents of that
to know how to build the module. The name of the module being built is
irrelevant, as in my case even though I'm compiling an rtl8814AU driver
the name of the module that gets built and loaded for the device is
8814au. This process get repeated for every directory it can find in
/var/lib/dkms where the contents of those directories indicate the
associated module has yet to be built for the current kernel and the
associated dkms.conf file has AUTOINSTALL="yes".
> Just as a side info, there is a known bug in DKMS (which I encountered
> in my testing trying to get things to work) that if, in your case, there
> are multiple version directories in /var/lib/dkms/evdi and any of those
> directories don't physically exist, dkms fails with an error that it
> can't find dkms.conf.
Well, that, uh, sucks. :-)
> From the documentation I have read, when DKMS runs at boot time to check
> if it needs to build and install the driver into the running kernel, the
> command that is issued is DKMS AUTOINSTALL, which is why AUTOINSTALL=yes
> in dkms.conf is critical.
I believe autoinstall just makes sure the module gets stuffed into
/lib/modules/`uname -r`/extra.
There are situations, from what I have read, where
location information
specified in dkms.conf can cause the module to be placed elsewhere, for
example, I believe it is possible to get the module placed in the
appropriate low level driver directory rather than in extra.
> Just for reference I have listed the contents of my dkms.conf file. On
>
git.com where I obtained my driver code from there was also a branch for
> and rtl8812au network driver, where they supplied a dkms.conf file, and
> my dkms.conf is based on that.
>
> PACKAGE_NAME="rtl8814AU"
> PACKAGE_VERSION="4.3.21_17997.20160531"
> BUILT_MODULE_NAME[0]="8814au"
> MAKE="'make' -j3 KVER=${kernelver}"
> CLEAN="'make' clean"
> DEST_MODULE_LOCATION[0]="/updates/dkms"
> AUTOINSTALL="YES"
>
> For me DKMS loses a lot of it usefulness if the driver is not loaded
> (modprobed) at boot time if a compile is required for the running kernel
> as I have network mounts in /etc/fstab that I would prefer to not have
> to mount manually every time a kernel update occurs if the driver is not
> compiled at kernel install time. I also don't understand why these
> issues are occurring under both F25 and Ubuntu 16.10 as up to the point
> of where I stopped using Mandriva (when it was forked to Mageia) 3 - 5
> years ago, I was using dkms all the time for multiple drivers and didn't
> encounter any of these problems.
I don't know that's a dkms issue. If the module builds, gets inserted
into your "/lib/modules/`uname -r`/extra" and a "depmod" is done,
then
dkms has done all it's supposed to do. I don't know you can direct it
to create a modprobe file to force a probe at boot.
To do that, I think you need to put a file in /etc/modprobe.d, e.g.
echo "8814au" >/etc/modprobe.d/rtl8814AU.conf
That should force a modprobe on boot. dkms should run before the
modprobe does, so the module should build, get installed, then modprobed
when a new kernel is installed.
Thanks Rick, I'll try that and see what happens
next time a new kernel
is provided. I'll force a recompile at boot if the new kernel install
compiles it.
regards,
Steve
----------------------------------------------------------------------
- Rick Stevens, Systems Engineer, AllDigital ricks(a)alldigital.com -
- AIM/Skype: therps2 ICQ: 226437340 Yahoo: origrps2 -
- -
- I doubt, therefore I might be. -
----------------------------------------------------------------------
_______________________________________________
users mailing list -- users(a)lists.fedoraproject.org
To unsubscribe send an email to users-leave(a)lists.fedoraproject.org