Wi-Fi hardware passthrough

[dreamliner787-9]

This is far from being ready but opening a PR so I don't forget about it.

Edit: after some thoughts, perhaps it would be better if we fold #180 into this as well.

[aaurandt]

I am currently trying this example and had a few comments:

• In the overlay.dts the DMA controller is re-routed to only use channel 0, but the .system file still passes through all DMA channel interrupts from 0 to 10.
• I also believe the ethernet mapping and interrupts can be removed if you edit Buildroot to not search for eth0 with the BR2_SYSTEM_DHCP property in Buildroot (i.e., delete text entirely or change to "wlan0" if you overlay the rootfs with a proper etc/wpa_supplicant.conf).

[dreamliner787-9]

In the overlay.dts the DMA controller is re-routed to only use channel 0, but the .system file still passes through all DMA channel interrupts from 0 to 10.

Ah yes that was from my debugging attempt. I'll remove them once I figure out why removing irqpoll broke the demo.

I also believe the ethernet mapping and interrupts can be removed if you edit Buildroot to not search for eth0 with the BR2_SYSTEM_DHCP property in Buildroot (i.e., delete text entirely or change to "wlan0" if you overlay the rootfs with a proper etc/wpa_supplicant.conf).

Ah thanks for the tip!

[terryzbai]

Link #62.

[aaurandt]

I was getting boot errors with emmc2 that were congesting my user-space console; I was able to just remove this and still access WiFi through passthrough. In the overlay.dts I included the following:

emmc2bus {
		mmc@7e340000 {
			status = "disabled";
		};
	};

I then removed the mapping in the .system file for this bus.

[dreamliner787-9]

I was getting boot errors with emmc2 that were congesting my user-space console; I was able to just remove this and still access WiFi through passthrough. In the overlay.dts I included the following:

emmc2bus {
		mmc@7e340000 {
			status = "disabled";
		};
	};

I then removed the mapping in the .system file for this bus.

Cool! Thanks for the feedback, I'll make sure to incorporate it in my example. I'm glad that it is working for you, please let me know if you run into other problems with libvmm.

[aaurandt]

One solution to being able to remove irqpoll from this demo is to remove the dma-controller. Here are my updates to the overlay.dts:

dma-controller@7e007000 {
            status = "disabled";
        };

mmcnr@7e300000 {
          /delete-property/ dmas;
          /delete-property/ dma-names;
        };

Then you can also remove the memory mapping and interrupts for the dma-controller from the .system file and vmm.c file.

From what I am observing, the interrupts are getting stuck when trying to load the driver using DMA. Disabling DMA forces PIO, which succeeds. In my user-land kernel boot, I now see the message mmc-bcm2835 fe300000.mmcnr: mmc-bcm2835: Unable to initialise DMA channel. Falling back to PIO.

[dreamliner787-9]

One solution to being able to remove irqpoll from this demo is to remove the dma-controller. Here are my updates to the overlay.dts:

dma-controller@7e007000 {
            status = "disabled";
        };

mmcnr@7e300000 {
          /delete-property/ dmas;
          /delete-property/ dma-names;
        };

Then you can also remove the memory mapping and interrupts for the dma-controller from the .system file and vmm.c file.

From what I am observing, the interrupts are getting stuck when trying to load the driver using DMA. Disabling DMA forces PIO, which succeeds. In my user-land kernel boot, I now see the message mmc-bcm2835 fe300000.mmcnr: mmc-bcm2835: Unable to initialise DMA channel. Falling back to PIO.

So that would require the CPU to manually shuffle data to and from the WiFi card? That isn't great for performance, I think there is a bug in our VGIC code somewhere. I'll investigate this and see if we can keep the DMA controller.