Add support for us to power on the USB on the bcm283x

[andrew]

Summary

U-Boot may not have run so we need to ask the firmware to power on the
hardware.

[• hselasky]

sys/arm/broadcom/bcm2835/bcm283x_dwc_fdt.c

118 ↗

(On Diff #3910)

You should forward detach and shutdown methods too.

[rpaulo]

Should probably be approved by one more person.

sys/arm/broadcom/bcm2835/bcm283x_dwc_fdt.c

83 ↗

(On Diff #3910)

Shouldn't the __aligned() come after "msg" ?

118 ↗

(On Diff #3910)

Shutdown is just bus_generic_shutdown, right? It doesn't look like it needs to be forwarded.

[imp]

Despite my catty comments, this looks good to me, assuming a decent answer for the why 16 and not 32, 64 or 8.

sys/arm/broadcom/bcm2835/bcm283x_dwc_fdt.c

83 ↗

(On Diff #3910)

and why 16?

sys/dev/usb/controller/dwc_otg_fdt.c

81 ↗

(On Diff #3910)

Now that's a mouthful :)

But also correct.

[mmel]

This patch breaks coherency for multicore ARMs.

sys/arm/broadcom/bcm2835/bcm283x_dwc_fdt.c

104 ↗

(On Diff #3910)

cpu_dcache_wbinv_all() is not SMP coherent and cannot be used outside init/shutdown code. Please, use range based cache operations here.

[• hselasky]

sys/arm/broadcom/bcm2835/bcm283x_dwc_fdt.c

118 ↗

(On Diff #3910)

Just ensure it looks something like this:

DEVMETHOD(device_detach, dwc_otg_detach),
DEVMETHOD(device_suspend, bus_generic_suspend),
DEVMETHOD(device_resume, bus_generic_resume),
DEVMETHOD(device_shutdown, bus_generic_shutdown),

--HPS

[andrew]

sys/arm/broadcom/bcm2835/bcm283x_dwc_fdt.c

83 ↗

(On Diff #3910)

I got it from somewhere, but don't remember the details.

118 ↗

(On Diff #3910)

Anything missing from here will call into the dwc_otg_fdt driver. We don't need shutdown as it's not implemented here. This is why I'm subclassing the existing bus, so I don't need to handle all of these.

[• ian]

sys/arm/broadcom/bcm2835/bcm283x_dwc_fdt.c

104 ↗

(On Diff #3910)

Even better, use busdma to allocate and sync the buffer. (Hmmm, I wonder what happened to jhb's proposed change to make it easier to allocate busdma buffers for stuff like this? Probably got sidetracked when I asked for changes and didn't offer a patch.)

[andrew]

Update to use busdma

[mmel]

Please see inline comment.
I'm fine with this patch, if missing bus_dmamap_sync() will be included.

sys/arm/broadcom/bcm2835/bcm283x_dwc_fdt.c

154 ↗

(On Diff #4049)

bus_dmamap_sync(msg_tag, msg_map, BUS_DMASYNC_POSTWRITE | BUS_DMASYNC_POSTREAD);

[• ian]

sys/arm/broadcom/bcm2835/bcm283x_dwc_fdt.c

110 ↗

(On Diff #4049)

doesn't the alignment need to be 16? it looks like the mailbox interface uses the low 4 bits for the channel number.

154 ↗

(On Diff #4049)

maybe along with a comment that says it's only because we need to match PRE ops with corresponding POST ops (since we don't actually read any data coming back via the mailbox).

[andrew]

Closed by commit rS279778 (authored by @andrew).