--- On Mon, 6/21/10, Rajaram R <rajaram.officemail@xxxxxxxxx> wrote: > From: Rajaram R <rajaram.officemail@xxxxxxxxx> > Subject: Re: how OUT/IN request from host are propogated on gadget side Through the usb_request entities sent from the gadget driver ... you won't see packets, only whole transfers (one or more packet each). For example, if maxpacket is 64 bytes, and the buffer in the usb_request is 512 bytes, the controller driver will fill (OUT) or empty (IN) that buffer using four USB packets. That's true whether DMA or PIO is used to empty the FIFO. > > I am looking at gadget driver code. > > I have a question understanding how OUT/IN request, > sent by USB host, travels on Gadget Side Stack. The requests > will be received in terms of OUT/IN token by gadget > controller hardware. Gadget controller driver will receive > an interrupt for the same. Not necessarily. DMA is often in use, so any IRQ will signify the end of the DMA transfer rather than a single USB packet. EP0 is usually packet-at-a-time PIO though, as you imply; especially SETUP packets Look at the at91_udc driver for a relatively straightforward PIO driver. Look at the omap_udc driver for a driver with both PIO and DMA support that may be relatively easy to understand. Or not ... all USB peripheral drivers are different. Tricky bits relate to DMA and FIFO handling;failover from DMA modes to PIO modes (and vice versa). What happens afterwards ? In any > case, gadget controller driver should notify gadget side > driver about such request. But, I do not see any such > mechanism. You have it backwards. I/O notifications are driven by the usb_request entities from the gadget driver. The controller driver just obeys usb_request commands, in conjunction with packets from the host. If there's no usb_request then nothing hapens. (Except for SETUP packets, ISTR ... it's late at night here though, so I might not be thinking clearly on that.) > > > > Going through, file_storage.c, I do see that gadget > driver always polls for any OUT/IN request. Is it the *only* > mechanism ? No. Polling in that driver is a special case and is unusual. Use g_zero as a more typical example. > > Is there any other in-kernel mechanism through which > gadget controller driver can notify about OUT/IN request to > gadget driver ? Explained above, and in the kerneldoc you reference. > > Did you had a chance to read this document ? > > http://tali.admingilde.org/linux-docbook/gadget.pdf > > In case of net2280(gadget/net2280.c), pls see net2280_irq > and the > following routines take care of it. > > /* handle disconnect, dma, and more */ > 2709 > handle_stat1_irqs (dev, readl > (&dev->regs->irqstat1)); > 2710 > 2711 /* control > requests and PIO */ > 2712 > handle_stat0_irqs (dev, readl > (&dev->regs->irqstat0)); > The net2280 peripheral controller driver is one of the less straightforward ones, in part because of the workarounds it has for early chip errata. > HTH > > > > > Thanks, > > Viral > > -- To unsubscribe from this list: send the line "unsubscribe linux-usb" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
