mlx5: Print driver version only when creating an interface

[• hselasky]

Summary

Otherwise mlx5 unconditionally prints messages to the dmesg during boot,
even when there is no hardware for it to drive.

[• hselasky]

Need to check this a bit. I will come back on monday with a final answer.

BTW: There is a macro in the LinuxKPI, PRINTK ONCE or something like that, which does exactly like you do. Maybe use that?

[• hselasky]

I suspect we may just put the printf() under if (bootverbose) .

[markj]

Predicate the printf() on bootverbose.

[• hselasky]

@markj: Do you mind if I commander this revision?

[markj]

In D29100#652197, @hselasky wrote:

@markj: Do you mind if I commander this revision?

Sure, go ahead.

[• hselasky]

Update patch.

[markj]

sys/dev/mlx5/mlx5_core/mlx5_main.c

1328 ↗

(On Diff #85772)

This reintroduces the problem. The line is printed even if no mlx5 hw is present. We already call device_set_desc() here, that ensures that mlx5_version will be printed immediately before attach.

Also there is a newline missing I think.

[• hselasky]

sys/dev/mlx5/mlx5_core/mlx5_main.c

1328 ↗

(On Diff #85772)

No, this is not the device_probe(). This is the device_attach(), and the device_set_desc() is only visible to the sysctl tree. We actually set it too late for it to show up in dmesg :-(

Should work. Can you test it? Or are you sure?

[markj]

sys/dev/mlx5/mlx5_core/mlx5_main.c

1328 ↗

(On Diff #85772)

Sorry, I tested it, you are right. I missed that Linux "probe" is FreeBSD "attach".

We should remove the device_set_desc() call then.

[• hselasky]

Right now we want both, due to some internal QA/regression tools. Else you are right! Thanks for reviewing.

[• hselasky]

This will be upstreamed soon.