Skip to contentSkip to conversation

usb: gadget: bcm63xx UDC driver

PATCH Superseded Kevin Cernekee 3 participants 5 messages Aug 22, 2012 – Aug 26, 2012

Subsystems (2)

v3 of 4 Superseded Latest is v4 →

lore
Share & tools
lore
#0 lore raw

PATCH v3 1/1 usb: gadget: bcm63xx UDC driver

Driver for the "USB20D" / "USBD" block on BCM6328, BCM6368, BCM6816, BCM6362, BCM3383, and others.

The hardware block was designed to support networking applications (direct connection of a home router to a PC), and the endpoint configuration is fixed.

---
 drivers/usb/gadget/Kconfig        |   12 +
 drivers/usb/gadget/Makefile       |    1 +
 drivers/usb/gadget/bcm63xx_udc.c  | 2456 +++++++++++++++++++++++++++++++++++++
 drivers/usb/gadget/gadget_chips.h |    3 +
 4 files changed, 2472 insertions(+)
 create mode 100644 drivers/usb/gadget/bcm63xx_udc.c

V3 changes:

Fix dynamic linking verbiage in Kconfig help text Abort gracefully if platform data is missing
Change ep0 logic to use a workqueue again
Rename update_* functions to bcm63xx_update_*
Fix handling of soft connect events sent from the wrong state Add delays to prevent EMI warnings on host

drivers/usb/gadget/Kconfig +12 -0
index 51ab5fd..663aae2 100644
@@ -160,6 +160,18 @@ config USB_ATMEL_USBA
 	  USBA is the integrated high-speed USB Device controller on
 	  the AT32AP700x, some AT91SAM9 and AT91CAP9 processors from Atmel.
 
+config USB_BCM63XX_UDC
+	tristate "Broadcom BCM63xx Peripheral Controller"
+	select USB_GADGET_DUALSPEED
+	depends on BCM63XX
+	help
+	   Many Broadcom BCM63xx chipsets (such as the BCM6328) have a
+	   high speed USB Device Port with support for four fixed endpoints
+	   (plus endpoint zero).
+
+	   Say "y" to link the driver statically, or "m" to build a
+	   dynamically linked module called "bcm63xx_udc".
+
 config USB_FSL_USB2
 	tristate "Freescale Highspeed USB DR Peripheral Controller"
 	depends on FSL_SOC || ARCH_MXC
drivers/usb/gadget/Makefile +1 -0
index 3fd8cd0..d84f923 100644
@@ -16,6 +16,7 @@ obj-$(CONFIG_USB_OMAP)		+= omap_udc.o
 obj-$(CONFIG_USB_S3C2410)	+= s3c2410_udc.o
 obj-$(CONFIG_USB_AT91)		+= at91_udc.o
 obj-$(CONFIG_USB_ATMEL_USBA)	+= atmel_usba_udc.o
+obj-$(CONFIG_USB_BCM63XX_UDC)	+= bcm63xx_udc.o
 obj-$(CONFIG_USB_FSL_USB2)	+= fsl_usb2_udc.o
 fsl_usb2_udc-y			:= fsl_udc_core.o
 fsl_usb2_udc-$(CONFIG_ARCH_MXC)	+= fsl_mxc_udc.o
drivers/usb/gadget/bcm63xx_udc.c new file +2456 -0

2,456 lines not on this page · raw message

drivers/usb/gadget/gadget_chips.h +3 -0
index b8b3a34..0ccca58 100644
@@ -28,6 +28,7 @@
 #define gadget_is_amd5536udc(g)		(!strcmp("amd5536udc", (g)->name))
 #define gadget_is_at91(g)		(!strcmp("at91_udc", (g)->name))
 #define gadget_is_atmel_usba(g)		(!strcmp("atmel_usba_udc", (g)->name))
+#define gadget_is_bcm63xx(g)		(!strcmp("bcm63xx_udc", (g)->name))
 #define gadget_is_ci13xxx_msm(g)	(!strcmp("ci13xxx_msm", (g)->name))
 #define gadget_is_ci13xxx_pci(g)	(!strcmp("ci13xxx_pci", (g)->name))
 #define gadget_is_dummy(g)		(!strcmp("dummy_udc", (g)->name))
@@ -121,6 +122,8 @@ static inline int usb_gadget_controller_number(struct usb_gadget *gadget)
 		return 0x32;
 	else if (gadget_is_lpc32xx(gadget))
 		return 0x33;
+	else if (gadget_is_bcm63xx(gadget))
+		return 0x34;
 
 	return -ENOENT;
 }
Signature · 1 line
1.7.11.1
#1 lore raw

On Tue, Aug 21, 2012 at 06:22:35PM -0700, Kevin Cernekee wrote:

Just one thing that bit while I was sleeping:
The HW acks SetConfig on its own. Once you notice this, you set ->ep0_req_set_cfg and set state in bcm63xx_ep0_do_idle() to EP0_IN_FAKE_STATUS_PHASE. This is I guess the workaround for mass_storage's hold with DELAYED_STATUS and continues with a zero packet.
Now two questions:
- If a gadget descides not NAK / stall the SetConfig requests. What happens here?
- What happens if the host is faster than the UDC. SetConfig returns in usb-storage with "DELAYED_STATUS". HW Acks this. Could the Host send another request before the gadget queues the ep0 request?

Sebastian

#2 lore raw
Quoted 5 lines · Sebastian Andrzej Siewior

Just one thing that bit while I was sleeping:
The HW acks SetConfig on its own. Once you notice this, you set ->ep0_req_set_cfg and set state in bcm63xx_ep0_do_idle() to EP0_IN_FAKE_STATUS_PHASE. This is I guess the workaround for mass_storage's hold with DELAYED_STATUS and continues with a zero packet.

EP0_IN_FAKE_STATUS_PHASE is there for the general case of: setup() callback returned >= 0 after a spoofed setup packet, and we're waiting for the gadget driver to send the 0-byte status reply so we can silently discard it and move on.

When bcm63xx_udc is in EP0_IN_FAKE_STATUS_PHASE, it won't issue any more setup() callbacks until the 0-byte reply arrives from the gadget driver. If the host sends a setup request, the callback will be held off until after the (unused) status reply. This keeps the gadget driver from getting confused by out-of-sequence events.

Quoted 3 lines

Now two questions:
- If a gadget descides not NAK / stall the SetConfig requests. What happens here?

If the return value from the setup() callback was negative, bcm63xx_udc should just return to EP0_IDLE as the gadget driver will never send a 0-byte reply.

I have added a new check for this condition, verified that it works as intended, and posted V4.

I am hoping that these invalid SET_CONFIGURATION / SET_INTERFACE requests are uncommon. In what sorts of situations will a host request a configuration that isn't advertised in the device's descriptors? I had trouble just convincing usb_set_interface() / usb_driver_set_configuration() to send such a request because they honor bInterfaceNumber / bConfigurationValue from the descriptors.

Quoted 3 lines

- What happens if the host is faster than the UDC. SetConfig returns in usb-storage with "DELAYED_STATUS". HW Acks this. Could the Host send another request before the gadget queues the ep0 request?

Could you please clarify if this is the sequence of events you are describing:

1) Host sends a valid SET_CONFIGURATION request to a mass storage gadget

2) Hardware instantly auto-acks the request, completing the status phase and allowing the host to proceed with another ep0 request

3) bcm63xx_udc sends a spoofed SET_CONFIGURATION setup packet to the gadget driver

4) setup() callback returns USB_GADGET_DELAYED_STATUS (0x7fff) but doesn't queue up a reply

5) Host sends another setup packet before usb_composite_setup_continue() is called to send the 0-byte status reply

If so, the next steps should look like:

6) bcm63xx_udc takes a data IRQ, and sets ep0_req_completed

7) bcm63xx_udc stays in EP0_IN_FAKE_STATUS_PHASE until the 0-byte reply is received from usb_composite_setup_continue()

8) usb_composite_setup_continue() eventually sends the 0-byte reply

9) bcm63xx_udc returns to EP0_IDLE and notices that ep0_req_completed is now set

10) bcm63xx_ep0_do_setup() looks at the new request, and performs the setup() callback for the new setup request

11) Data/status phases are handled as usual

#3 lore raw
Quoted 6 lines · Kevin Cernekee

I am hoping that these invalid SET_CONFIGURATION / SET_INTERFACE requests are uncommon. In what sorts of situations will a host request a configuration that isn't advertised in the device's descriptors? I had trouble just convincing usb_set_interface() / usb_driver_set_configuration() to send such a request because they honor bInterfaceNumber / bConfigurationValue from the descriptors.

A request doesn't have to be invalid to fail. Valid requests can fail for all sorts of reasons. -ENOMEM is the prototypical example.

Alan Stern

#4 lore raw
In reply to Kevin Cernekee
Quoted 5 lines · Kevin Cernekee

When bcm63xx_udc is in EP0_IN_FAKE_STATUS_PHASE, it won't issue any more setup() callbacks until the 0-byte reply arrives from the gadget driver. If the host sends a setup request, the callback will be held off until after the (unused) status reply. This keeps the gadget driver from getting confused by out-of-sequence events.

Okay. Then it works as requested :)

Quoted 14 lines
Quoted 3 lines

- What happens if the host is faster than the UDC. SetConfig returns in usb-storage with "DELAYED_STATUS". HW Acks this. Could the Host send another request before the gadget queues the ep0 request?

Could you please clarify if this is the sequence of events you are describing:

1) Host sends a valid SET_CONFIGURATION request to a mass storage gadget

2) Hardware instantly auto-acks the request, completing the status phase and allowing the host to proceed with another ep0 request

3) bcm63xx_udc sends a spoofed SET_CONFIGURATION setup packet to the gadget driver

Quoted 6 lines

4) setup() callback returns USB_GADGET_DELAYED_STATUS (0x7fff) but doesn't queue up a reply

5) Host sends another setup packet before usb_composite_setup_continue() is called to send the 0-byte status reply

exactly

Quoted 7 lines

If so, the next steps should look like:

6) bcm63xx_udc takes a data IRQ, and sets ep0_req_completed

7) bcm63xx_udc stays in EP0_IN_FAKE_STATUS_PHASE until the 0-byte reply is received from usb_composite_setup_continue()

Quoted 6 lines

8) usb_composite_setup_continue() eventually sends the 0-byte reply

9) bcm63xx_udc returns to EP0_IDLE and notices that ep0_req_completed is now set

10) bcm63xx_ep0_do_setup() looks at the new request, and performs the setup() callback for the new setup request

Okay. This sounds good.

Quoted 1 line

11) Data/status phases are handled as usual

Please tell you HW vendor that this auto ack feature is complete non sense unless you already have :)

Sebastian