Re: linux-next: build failure after merge of the driver-core tree

From: Jeffrey Hugo
Date: Tue Apr 11 2023 - 11:30:21 EST

On 4/11/2023 9:26 AM, Jeffrey Hugo wrote:
On 4/11/2023 9:13 AM, Greg KH wrote:
On Tue, Apr 11, 2023 at 09:08:39AM -0600, Jeffrey Hugo wrote:
On 4/11/2023 9:01 AM, Daniel Vetter wrote:
On Tue, Apr 11, 2023 at 12:40:28PM +0200, Greg KH wrote:
On Tue, Apr 11, 2023 at 11:55:20AM +0200, Daniel Vetter wrote:
On Tue, Apr 11, 2023 at 02:38:12PM +1000, Stephen Rothwell wrote:
Hi all,

After merging the driver-core tree, today's linux-next build (x86_64
allmodconfig) failed like this:

In file included from include/linux/linkage.h:7,
                   from include/linux/kernel.h:17,
                   from drivers/accel/qaic/mhi_qaic_ctrl.c:4:
drivers/accel/qaic/mhi_qaic_ctrl.c: In function 'mhi_qaic_ctrl_init':
include/linux/export.h:27:22: error: passing argument 1 of 'class_create' from incompatible pointer type [-Werror=incompatible-pointer-types]
     27 | #define THIS_MODULE (&__this_module)
        |                     ~^~~~~~~~~~~~~~~
        |                      |
        |                      struct module *
drivers/accel/qaic/mhi_qaic_ctrl.c:544:38: note: in expansion of macro 'THIS_MODULE'
    544 |         mqc_dev_class = class_create(THIS_MODULE, MHI_QAIC_CTRL_DRIVER_NAME);
        |                                      ^~~~~~~~~~~
In file included from include/linux/device.h:31,
                   from include/linux/mhi.h:9,
                   from drivers/accel/qaic/mhi_qaic_ctrl.c:5:
include/linux/device/class.h:229:54: note: expected 'const char *' but argument is of type 'struct module *'
    229 | struct class * __must_check class_create(const char *name);
        |                                          ~~~~~~~~~~~~^~~~
drivers/accel/qaic/mhi_qaic_ctrl.c:544:25: error: too many arguments to function 'class_create'
    544 |         mqc_dev_class = class_create(THIS_MODULE, MHI_QAIC_CTRL_DRIVER_NAME);
        |                         ^~~~~~~~~~~~
include/linux/device/class.h:229:29: note: declared here
    229 | struct class * __must_check class_create(const char *name);
        |                             ^~~~~~~~~~~~

Caused by commit

    1aaba11da9aa ("driver core: class: remove module * from class_create()")

interacting with commit

    566fc96198b4 ("accel/qaic: Add mhi_qaic_cntl")

from the drm tree.

I have applied the following merge fix patch for today.

From: Stephen Rothwell <sfr@xxxxxxxxxxxxxxxx>
Date: Tue, 11 Apr 2023 14:16:57 +1000
Subject: [PATCH] fixup for "driver core: class: remove module * from class_create()"

interacting with "accel/qaic: Add mhi_qaic_cntl"

Signed-off-by: Stephen Rothwell <sfr@xxxxxxxxxxxxxxxx>

Thanks for the fixup. Since Dave is out I've made a note about this in my
handover mail so it won't get lost in the drm-next merge window pull. I
don't think we need any other coordination than mention it in each pull to
Linus, topic tree seems overkill for this. Plus there's no way I can
untangle the drm tree anyway :-).

Want me to submit a patch for the drm tree that moves this to use
class_register() instead, which will make the merge/build issue go away
for you?  That's my long-term goal here anyway, so converting this new
code to this api today would be something I have to do eventually :)

We kinda closed drm-next for feature work mostly already (just pulling
stuff in from subtrees), so won't really help for this merge window.

For everything else I think this is up to Oded, I had no idea qaic needed
it's entire own dev class and I don't want to dig into this for the risk I
might freak out :-)

Adding Oded.

Cheers, Daniel

Sorry for the mess.

I made a note to update to class_register() once my drm-misc access is
sorted out.  Looks like we'll address the conflict in the merge window, and
catch the update to the new API in the following release.

Wait, I think the large question is, "why does this need a separate
class"?  Why are you not using the accel char device and class?  That is
what everything under accel/ should be using, otherwise why put it in
there?

And what exactly are you using that class for?  Just device nodes?  If
so, how many?

thanks,

greg k-h


Remember MHI_UCI that then evolved into the WWAN subsystem?  I pointed out at the time that AIC100/QAIC would need the same functionality. You/Jakub told myself/Mani/Loic that a combined implementation is not acceptable, and every area needs to implement their own version of MHI_UCI.

We took the WWAN subsystem and simplified it to meet our needs.

The functionality is QAIC specific, so wedging it into the Accel node seems to be a poor fit as it would subject Habana and iVPU to the same.

Also, I forgot to mention. QAIC is sharing userspace components with WWAN, so we really cannot diverge from what WWAN has done and define a new API through the Accel node.


We need (eventually) 128 device nodes.  We have systems with 32 QAIC devices, and each QAIC device uses 4 device nodes (32 * 4 = 128).  WWAN subsystem would be similar.  Looks like each 5G modem is 6 nodes per device, so if you had 22 5G modems on a system, you'd have 132 device nodes.  I'm not aware of any such system, but it could exist.

-Jeff