From mboxrd@z Thu Jan 1 00:00:00 1970 From: Sergei Shtylyov Subject: Re: [PATCH 4/5] musb: use system DMA for unaligned buffers on RTL >= 1.8 Date: Wed, 12 May 2010 17:18:02 +0400 Message-ID: <4BEAAA8A.1000003@ru.mvista.com> References: <1273664979-493-1-git-send-email-ajay.gupta@ti.com> <1273664979-493-2-git-send-email-ajay.gupta@ti.com> <1273664979-493-3-git-send-email-ajay.gupta@ti.com> <1273664979-493-4-git-send-email-ajay.gupta@ti.com> Mime-Version: 1.0 Content-Type: text/plain; charset=ISO-8859-1; format=flowed Content-Transfer-Encoding: 7bit Return-path: In-Reply-To: <1273664979-493-4-git-send-email-ajay.gupta-l0cyMroinI0@public.gmane.org> Sender: linux-usb-owner-u79uwXL29TY76Z2rM5mHXA@public.gmane.org To: Ajay Kumar Gupta Cc: linux-usb-u79uwXL29TY76Z2rM5mHXA@public.gmane.org, linux-omap-u79uwXL29TY76Z2rM5mHXA@public.gmane.org List-Id: linux-omap@vger.kernel.org Hello. Ajay Kumar Gupta wrote: > MUSB RTL version 1.8 onward (OMAP3630, AM/DM37x, OMAP4) DMA requires > the buffers to be aligned on a four byte boundary. This affects USB > CDC/RNDIS class application where buffers are always unaligned. > > Use system DMA for unaligned buffers as a workaround of this issue. > > Current patch supports device side CDC/RNDIS. Host side would require > change in Tx programming path for mode-0 operation when transfer length > is more than packet size. > > Also fixed an issue in device Tx completion path where 'is_dma' is getting > set unconditionally. This would fail the io if Tx transfer is done in > mode-0. Fixed it by updating it based on 'request->actual' length. > > Signed-off-by: Ajay Kumar Gupta > [...] > diff --git a/drivers/usb/musb/musbhsdma.c b/drivers/usb/musb/musbhsdma.c > index 70342eb..6118d5f 100644 > --- a/drivers/usb/musb/musbhsdma.c > +++ b/drivers/usb/musb/musbhsdma.c > @@ -54,12 +54,18 @@ static void musb_sysdma_completion(int lch, u16 ch_status, void *data) > (struct musb_dma_channel *) data; > struct musb_dma_controller *controller = musb_channel->controller; > struct musb *musb = controller->private_data; > + void __iomem *mbase = controller->base; > + > channel = &musb_channel->channel; > > DBG(2, "lch = 0x%d, ch_status = 0x%x\n", lch, ch_status); > spin_lock_irqsave(&musb->lock, flags); > > - addr = (u32) omap_get_dma_dst_pos(musb_channel->sysdma_channel); > + if (musb_channel->transmit) > + addr = (u32) omap_get_dma_src_pos(musb_channel->sysdma_channel); > + else > + addr = (u32) omap_get_dma_dst_pos(musb_channel->sysdma_channel); > + > if (musb_channel->len == 0) > channel->actual_len = 0; > else > @@ -72,6 +78,26 @@ static void musb_sysdma_completion(int lch, u16 ch_status, void *data) > "=> reconfig 0 " : " => complete"); > > channel->status = MUSB_DMA_STATUS_FREE; > + > + /* completed */ > + if ((musb_channel->transmit) && (channel->desired_mode == 0) > + && (channel->actual_len == musb_channel->max_packet_sz)) { > + > + u8 epnum = musb_channel->epnum; > + int offset = MUSB_EP_OFFSET(epnum, > + MUSB_TXCSR); > + u16 txcsr; > + > + /* > + * The programming guide says that we > + * must clear DMAENAB before DMAMODE. > + */ > + musb_ep_select(mbase, epnum); > + txcsr = musb_readw(mbase, offset); > + txcsr |= MUSB_TXCSR_TXPKTRDY; > + musb_writew(mbase, offset, txcsr); > + } > + > musb_dma_completion(musb, musb_channel->epnum, musb_channel->transmit); > > spin_unlock_irqrestore(&musb->lock, flags); > @@ -138,8 +164,15 @@ static struct dma_channel *dma_channel_allocate(struct dma_controller *c, > * issue when TX and RX DMA channels are simultaneously > * enabled. To work around this issue, use system DMA > * for all RX channels. > + * Also on MUSB RTL version 1.8 onward (OMAP3630, OMAP4 > + * and AM/DM37x) DMA requires buffers to be aligned on > + * a four byte boundary. This affects USB CDC/RNDIS > + * class application where buffers are always unaligned. > + * Using system DMA for unaligned buffers as a > + * workaround for this issue. > */ > - if (((musb->hwvers == MUSB_HWVERS_1400) && !transmit) > + if ((((musb->hwvers == MUSB_HWVERS_1400) && !transmit) > + || (musb->hwvers >= MUSB_HWVERS_1800)) > && use_sdma_workaround()) { > int ret; > ret = omap_request_dma(OMAP24XX_DMA_NO_DEVICE, > @@ -194,11 +227,18 @@ static void configure_channel(struct dma_channel *channel, > struct musb *musb = controller->private_data; > void __iomem *mbase = controller->base; > u8 bchannel = musb_channel->idx; > + u8 buffer_is_aligned = (dma_addr & 0x3) ? 0 : 1; > + u8 use_sdma = (musb_channel->sysdma_channel == -1) ? 0 : 1; > u16 csr = 0; > > DBG(4, "%p, pkt_sz %d, addr 0x%x, len %d, mode %d\n", > channel, packet_sz, dma_addr, len, mode); > - if (musb_channel->sysdma_channel != -1) { > + > + if (buffer_is_aligned && (packet_sz >= 512) && > + (musb->hwvers >= MUSB_HWVERS_1800)) > + use_sdma = 0; > + > + if (use_sdma && !musb_channel->transmit) { > /* System DMA */ > /* RX: set src = FIFO */ > omap_set_dma_transfer_params(musb_channel->sysdma_channel, > @@ -224,6 +264,32 @@ static void configure_channel(struct dma_channel *channel, > > omap_start_dma(musb_channel->sysdma_channel); > > + } else if (use_sdma && musb_channel->transmit) { > + /* System DMA */ > + /* TX: set dst = FIFO */ > + omap_set_dma_transfer_params(musb_channel->sysdma_channel, > + OMAP_DMA_DATA_TYPE_S8, > + len ? len : 1, 1, /* One frame */ > + OMAP_DMA_SYNC_ELEMENT, > + OMAP24XX_DMA_NO_DEVICE, > + 0); /* Src Sync */ > + > + omap_set_dma_src_params(musb_channel->sysdma_channel, 0, > + OMAP_DMA_AMODE_POST_INC, dma_addr, > + 0, 0); > + > + omap_set_dma_dest_params(musb_channel->sysdma_channel, 0, > + OMAP_DMA_AMODE_CONSTANT, > + MUSB_FIFO_ADDRESS(musb->ctrl_phys_base, > + musb_channel->epnum), > + 0, 0); > + > + omap_set_dma_dest_data_pack(musb_channel->sysdma_channel, 0); > + omap_set_dma_dest_burst_mode(musb_channel->sysdma_channel, > + OMAP_DMA_DATA_BURST_DIS); > + > + omap_start_dma(musb_channel->sysdma_channel); > + > } else { /* Mentor DMA */ > > if (mode) { > @@ -299,6 +365,9 @@ static int dma_channel_abort(struct dma_channel *channel) > > if (channel->status == MUSB_DMA_STATUS_BUSY) { > if (musb_channel->transmit) { > + if (musb_channel->sysdma_channel != -1) > + omap_stop_dma(musb_channel->sysdma_channel); > + > Have you thought about e.g. Blackfin? How this is going to compile there, without omap_*() functions even existing? Or did you think that OMAP is the only user of this driver? WBR, Sergei -- To unsubscribe from this list: send the line "unsubscribe linux-usb" in the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org More majordomo info at http://vger.kernel.org/majordomo-info.html