firmware: Allow binary files to be loaded by /boot/loader

[imp]

Summary

Files loaded "-t firmware" (or module_type="firmware"). They are
registered with the firmware system using the full path to allow
the drivers to find them. There's only one firmware per file, and it is the
entire file.

Sponsored by: Netflix

[manu]

That's awesome, would it be hard to also have this for kldload ?

[jlduran]

Minor typo.

sys/kern/subr_firmware.c

565

[imp]

In D43522#992310, @manu wrote:

That's awesome, would it be hard to also have this for kldload ?

Unsure. I'll have to poke a bit. I had the same thought but wanted to get this in to see how hard it was and how well it worked.

Ideally, though, the driver would just do firmware_get like it does today and we'd load it in from some path automatically. Maybe with some mapping file, but likely not. Likely with a sane path like we have for .ko.

[jhb]

I also still prefer double spaces after periods in monospace text (code comments).

sys/kern/subr_firmware.c

593

605–609

I like your approach from your kldxref patch here better as it avoids the magic number of 4 all over. That is:

char *t;
....
t = strrchr(name, '.');
if (t != NULL && strcmp(t, ".bin") == 0)
    *t = '\0';

[imp]

sys/kern/subr_firmware.c

605–609

Yea, I'll cut all that out.

611

We panic early in startup, here I think.... Kinda makes sense though...
Also, we need some kind of flag that says we can't free this firmware (right? since it's pre-loaded)

[imp]

Change around a little how we're loading. Save the full filename and allow
subpath searches to succeed.

[imp]

double space full stop

[imp]

I think I've addressed everything (with a forthcoming typo fix)

sys/kern/subr_firmware.c

593

I think the new code doesn't need this blank like the old code did.

605–609

Just deleted this code entirely, so OBE.

[imp]

tpyo fix

[imp]

sys/kern/subr_firmware.c

572

I think I'm going to change this to 'native' since 'native format' is defined by the creating application (in this case the hardware the firmware is for).

[imp]

Use 'native' instead of 'raw' to describe.

[imp]

Shift to binary, update per review comments

[jhb]

If you do need the relative name lookups with some path components, then this is ok as-is to me.

sys/kern/subr_firmware.c

165

Do you want to fix this when registering the file instead? That is, only register the last component of the filename down in firmware_binary_files() instead? Oh, or are there requests that actually use partial paths? If so, I guess you have to do it this way.

[imp]

sys/kern/subr_firmware.c

165

There are requests with partial paths, though they often do other lookups w/o the path. I think the safest way is to match like this, but if it causes grief later, we can shift. I'm not sure that doing it at load time would suffice...

For example, rtw88 looks up rtw88/rtw8821c_fw.bin and then rtw8821c_fw.bin and then rtw88_rtw8821c_fw.bin and a few others.

[imp]

sys/kern/subr_firmware.c

165

Also if i load /boot/precious/onering-fw.bin then i want that to work regardless of load path...