On 12/12/2010 12:03 PM, Artem Leonenko wrote: > On 12/11/2010 09:32 PM, Pavan Kondeti wrote: >> Hi Artem, >> >> On 12/11/2010 11:42 PM, Artem Leonenko wrote: >>> On 12/11/2010 06:14 AM, pkondeti@xxxxxxxxxxxxxx wrote: >>> Do you suggest to wait until complete() callback finishes? The earlier >>> we will prime EP the faster data will be transmitted. >>> >> I was saying that (b) is sufficient to fix the problem. I agree that >> priming the endpoint before calling complete gives faster data rates. > >> I was telling (b) and (c) are same. > > I'm sorry, I misread that > No problem :-) >>>>> And as a part of the fix to make g_ether (and especially booting with >>>>> NFS root) I had >>>>> to increase NET_IP_ALIGN to mips32 natural value (4). >>>> >>>> I am not sure about the fix. But you might want to do this in a separate >>>> patch. >>>> >>> >>> Relaxing requirements on alignment will subtly increase memory usage but >>> in the same will allow ci13xxx driver + g_ether to support more >>> platforms. The ci13xxx driver supports rather big family of IP cores >>> so the more we can support the better. >>> >> >>> Can you try to test g_ether with your ci13xxx core with my patches? The >>> ci13xxx IP cores have many different flavors/features depending on >>> SoC/core instantiation. I wonder whether your core works with >>> default alignment or not. >>> >> Sure. I will test it on monday. But am sure that it will work without >> alignment change. Because g_ether is working fine on MSM with another >> controller driver (not ci13xxx_udc). > > Thanks. However it isn't clear to me how can you make an assumption > about ci13xxx USB IP core based on features of other USB core. Yes > there could be a chance that both cores have same feature. > Sorry. I should have explained clearly. Initially I have posted another controller driver for MSM SoC which has chipidea USB core. I have received feedback to re-use ci13xxx_udc driver. I have seen g_ether working fine without changing NET_IP_ALIGN with my initial driver. Please see http://www.spinics.net/lists/linux-usb/msg38460.html > FYI, comment from drivers/usb/gadget/amd5536udc.c file (also search > for NET_IP_ALIGN in gadget/Kconfig): > >> * UDC DMA requires 32-bit aligned buffers so DMA with gadget ether > does not >> * work without updating NET_IP_ALIGN. Or PIO mode (module param > "use_dma=0") >> * can be used with gadget ether. > > You see working g_ether with some UDCs doesn't guarantee that it will > work with others. I have a proposal of relaxing alignment requirements > and thus increasing the number of supported ci13xxx cores/their > instantiations. Don't know if there is something better way to handle > that. Any ideas? There was some discussion happened on the same issue. But nothing is finalized. I am also not sure about changing NET_IP_ALIGN macro as it may have some other side effects (I was told changing NET_IP_ALIGN makes parsing difficult in network layers and is inefficient). http://www.spinics.net/lists/linux-usb/msg25411.html > >> Thanks, >> Pavan >> >> > > -- > Cheers, > Artem -- Sent by a consultant of the Qualcomm Innovation Center, Inc. The Qualcomm Innovation Center, Inc. is a member of the Code Aurora Forum. -- 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
