| Message ID | 1441913164-3785-1-git-send-email-robert.jarzmik@free.fr |
|---|---|
| State | Accepted, archived |
| Delegated to: | David Miller |
| Headers | show |
From: Robert Jarzmik <robert.jarzmik@free.fr> Date: Thu, 10 Sep 2015 21:26:04 +0200 > Convert the dma transfers to be dmaengine based, now pxa has a dmaengine > slave driver. This makes this driver a bit more PXA agnostic. > > The driver was tested on pxa27x (mainstone) and pxa310 (zylonite), > ie. only pxa platforms. > > Signed-off-by: Robert Jarzmik <robert.jarzmik@free.fr> > Cc: Russell King <linux@arm.linux.org.uk> > Cc: Arnd Bergmann <arnd@arndb.de> > --- > This has potential to break other platform such as Neponset, Idp, > halibut and qsd8x50, so I added Russell and Arnd as they were discussing > smc91x support last February. Is someone testing whether such platforms break or not? I'm waiting for that before I consider applying this patch. -- To unsubscribe from this list: send the line "unsubscribe netdev" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
David Miller <davem@davemloft.net> writes: > From: Robert Jarzmik <robert.jarzmik@free.fr> > Date: Thu, 10 Sep 2015 21:26:04 +0200 > >> Convert the dma transfers to be dmaengine based, now pxa has a dmaengine >> slave driver. This makes this driver a bit more PXA agnostic. >> >> The driver was tested on pxa27x (mainstone) and pxa310 (zylonite), >> ie. only pxa platforms. >> >> Signed-off-by: Robert Jarzmik <robert.jarzmik@free.fr> >> Cc: Russell King <linux@arm.linux.org.uk> >> Cc: Arnd Bergmann <arnd@arndb.de> >> --- >> This has potential to break other platform such as Neponset, Idp, >> halibut and qsd8x50, so I added Russell and Arnd as they were discussing >> smc91x support last February. > > Is someone testing whether such platforms break or not? I'm waiting for > that before I consider applying this patch. My understanding is that Russell is the only one left testing them, or at least he was the only one complaining about a breakage lately on neponset. I can wait several weeks for Russell to have a bit of time to try : I know it will compile correctly at least for neponset, and I know almost all the code is under #ifdef CONFIG_ARCH_PXA. And still I would feel far more comfortable if it was tested, just as you. Cheers.
From: Robert Jarzmik <robert.jarzmik@free.fr> Date: Wed, 16 Sep 2015 11:41:54 +0200 > David Miller <davem@davemloft.net> writes: > >> From: Robert Jarzmik <robert.jarzmik@free.fr> >> Date: Thu, 10 Sep 2015 21:26:04 +0200 >> >>> Convert the dma transfers to be dmaengine based, now pxa has a dmaengine >>> slave driver. This makes this driver a bit more PXA agnostic. >>> >>> The driver was tested on pxa27x (mainstone) and pxa310 (zylonite), >>> ie. only pxa platforms. >>> >>> Signed-off-by: Robert Jarzmik <robert.jarzmik@free.fr> >>> Cc: Russell King <linux@arm.linux.org.uk> >>> Cc: Arnd Bergmann <arnd@arndb.de> >>> --- >>> This has potential to break other platform such as Neponset, Idp, >>> halibut and qsd8x50, so I added Russell and Arnd as they were discussing >>> smc91x support last February. >> > >> Is someone testing whether such platforms break or not? I'm waiting for >> that before I consider applying this patch. > > My understanding is that Russell is the only one left testing them, or at least > he was the only one complaining about a breakage lately on neponset. > > I can wait several weeks for Russell to have a bit of time to try : I know it > will compile correctly at least for neponset, and I know almost all the code is > under #ifdef CONFIG_ARCH_PXA. And still I would feel far more comfortable if it > was tested, just as you. Oh well, I've waited long enough.... patch applied, thanks. -- To unsubscribe from this list: send the line "unsubscribe netdev" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
On Thu, Sep 17, 2015 at 01:37:22PM -0700, David Miller wrote: > From: Robert Jarzmik <robert.jarzmik@free.fr> > Date: Wed, 16 Sep 2015 11:41:54 +0200 > > > David Miller <davem@davemloft.net> writes: > > > >> From: Robert Jarzmik <robert.jarzmik@free.fr> > >> Date: Thu, 10 Sep 2015 21:26:04 +0200 > >> > >>> Convert the dma transfers to be dmaengine based, now pxa has a dmaengine > >>> slave driver. This makes this driver a bit more PXA agnostic. > >>> > >>> The driver was tested on pxa27x (mainstone) and pxa310 (zylonite), > >>> ie. only pxa platforms. > >>> > >>> Signed-off-by: Robert Jarzmik <robert.jarzmik@free.fr> > >>> Cc: Russell King <linux@arm.linux.org.uk> > >>> Cc: Arnd Bergmann <arnd@arndb.de> > >>> --- > >>> This has potential to break other platform such as Neponset, Idp, > >>> halibut and qsd8x50, so I added Russell and Arnd as they were discussing > >>> smc91x support last February. > >> > > > >> Is someone testing whether such platforms break or not? I'm waiting for > >> that before I consider applying this patch. > > > > My understanding is that Russell is the only one left testing them, or at least > > he was the only one complaining about a breakage lately on neponset. > > > > I can wait several weeks for Russell to have a bit of time to try : I know it > > will compile correctly at least for neponset, and I know almost all the code is > > under #ifdef CONFIG_ARCH_PXA. And still I would feel far more comfortable if it > > was tested, just as you. > > Oh well, I've waited long enough.... patch applied, thanks. Well, I'm unlikely to get around to testing on the neponset any time in the next month, and I don't think there's much that would go wrong as a result of these changes: neponset doesn't use any of this DMA code.
diff --git a/drivers/net/ethernet/smsc/smc91x.c b/drivers/net/ethernet/smsc/smc91x.c index 630f0b7800e4..0e2fc1a844ab 100644 --- a/drivers/net/ethernet/smsc/smc91x.c +++ b/drivers/net/ethernet/smsc/smc91x.c @@ -2018,10 +2018,18 @@ static int smc_probe(struct net_device *dev, void __iomem *ioaddr, lp->cfg.flags |= SMC91X_USE_DMA; # endif if (lp->cfg.flags & SMC91X_USE_DMA) { - int dma = pxa_request_dma(dev->name, DMA_PRIO_LOW, - smc_pxa_dma_irq, NULL); - if (dma >= 0) - dev->dma = dma; + dma_cap_mask_t mask; + struct pxad_param param; + + dma_cap_zero(mask); + dma_cap_set(DMA_SLAVE, mask); + param.prio = PXAD_PRIO_LOWEST; + param.drcmr = -1UL; + + lp->dma_chan = + dma_request_slave_channel_compat(mask, pxad_filter_fn, + ¶m, &dev->dev, + "data"); } #endif @@ -2032,8 +2040,8 @@ static int smc_probe(struct net_device *dev, void __iomem *ioaddr, version_string, revision_register & 0x0f, lp->base, dev->irq); - if (dev->dma != (unsigned char)-1) - pr_cont(" DMA %d", dev->dma); + if (lp->dma_chan) + pr_cont(" DMA %p", lp->dma_chan); pr_cont("%s%s\n", lp->cfg.flags & SMC91X_NOWAIT ? " [nowait]" : "", @@ -2058,8 +2066,8 @@ static int smc_probe(struct net_device *dev, void __iomem *ioaddr, err_out: #ifdef CONFIG_ARCH_PXA - if (retval && dev->dma != (unsigned char)-1) - pxa_free_dma(dev->dma); + if (retval && lp->dma_chan) + dma_release_channel(lp->dma_chan); #endif return retval; } @@ -2370,6 +2378,7 @@ static int smc_drv_probe(struct platform_device *pdev) struct smc_local *lp = netdev_priv(ndev); lp->device = &pdev->dev; lp->physaddr = res->start; + } #endif @@ -2406,8 +2415,8 @@ static int smc_drv_remove(struct platform_device *pdev) free_irq(ndev->irq, ndev); #ifdef CONFIG_ARCH_PXA - if (ndev->dma != (unsigned char)-1) - pxa_free_dma(ndev->dma); + if (lp->dma_chan) + dma_release_channel(lp->dma_chan); #endif iounmap(lp->base); diff --git a/drivers/net/ethernet/smsc/smc91x.h b/drivers/net/ethernet/smsc/smc91x.h index 3a18501d1068..a3c129e1e40a 100644 --- a/drivers/net/ethernet/smsc/smc91x.h +++ b/drivers/net/ethernet/smsc/smc91x.h @@ -33,6 +33,7 @@ #ifndef _SMC91X_H_ #define _SMC91X_H_ +#include <linux/dmaengine.h> #include <linux/smc91x.h> /* @@ -244,6 +245,7 @@ struct smc_local { u_long physaddr; struct device *device; #endif + struct dma_chan *dma_chan; void __iomem *base; void __iomem *datacs; @@ -265,21 +267,47 @@ struct smc_local { * as RX which can overrun memory and lose packets. */ #include <linux/dma-mapping.h> -#include <mach/dma.h> +#include <linux/dma/pxa-dma.h> #ifdef SMC_insl #undef SMC_insl #define SMC_insl(a, r, p, l) \ smc_pxa_dma_insl(a, lp, r, dev->dma, p, l) static inline void +smc_pxa_dma_inpump(struct smc_local *lp, u_char *buf, int len) +{ + dma_addr_t dmabuf; + struct dma_async_tx_descriptor *tx; + dma_cookie_t cookie; + enum dma_status status; + struct dma_tx_state state; + + dmabuf = dma_map_single(lp->device, buf, len, DMA_FROM_DEVICE); + tx = dmaengine_prep_slave_single(lp->dma_chan, dmabuf, len, + DMA_DEV_TO_MEM, 0); + if (tx) { + cookie = dmaengine_submit(tx); + dma_async_issue_pending(lp->dma_chan); + do { + status = dmaengine_tx_status(lp->dma_chan, cookie, + &state); + cpu_relax(); + } while (status != DMA_COMPLETE && status != DMA_ERROR && + state.residue); + dmaengine_terminate_all(lp->dma_chan); + } + dma_unmap_single(lp->device, dmabuf, len, DMA_FROM_DEVICE); +} + +static inline void smc_pxa_dma_insl(void __iomem *ioaddr, struct smc_local *lp, int reg, int dma, u_char *buf, int len) { - u_long physaddr = lp->physaddr; - dma_addr_t dmabuf; + struct dma_slave_config config; + int ret; /* fallback if no DMA available */ - if (dma == (unsigned char)-1) { + if (!lp->dma_chan) { readsl(ioaddr + reg, buf, len); return; } @@ -291,18 +319,22 @@ smc_pxa_dma_insl(void __iomem *ioaddr, struct smc_local *lp, int reg, int dma, len--; } + memset(&config, 0, sizeof(config)); + config.src_addr_width = DMA_SLAVE_BUSWIDTH_4_BYTES; + config.dst_addr_width = DMA_SLAVE_BUSWIDTH_4_BYTES; + config.src_addr = lp->physaddr + reg; + config.dst_addr = lp->physaddr + reg; + config.src_maxburst = 32; + config.dst_maxburst = 32; + ret = dmaengine_slave_config(lp->dma_chan, &config); + if (ret) { + dev_err(lp->device, "dma channel configuration failed: %d\n", + ret); + return; + } + len *= 4; - dmabuf = dma_map_single(lp->device, buf, len, DMA_FROM_DEVICE); - DCSR(dma) = DCSR_NODESC; - DTADR(dma) = dmabuf; - DSADR(dma) = physaddr + reg; - DCMD(dma) = (DCMD_INCTRGADDR | DCMD_BURST32 | - DCMD_WIDTH4 | (DCMD_LENGTH & len)); - DCSR(dma) = DCSR_NODESC | DCSR_RUN; - while (!(DCSR(dma) & DCSR_STOPSTATE)) - cpu_relax(); - DCSR(dma) = 0; - dma_unmap_single(lp->device, dmabuf, len, DMA_FROM_DEVICE); + smc_pxa_dma_inpump(lp, buf, len); } #endif @@ -314,11 +346,11 @@ static inline void smc_pxa_dma_insw(void __iomem *ioaddr, struct smc_local *lp, int reg, int dma, u_char *buf, int len) { - u_long physaddr = lp->physaddr; - dma_addr_t dmabuf; + struct dma_slave_config config; + int ret; /* fallback if no DMA available */ - if (dma == (unsigned char)-1) { + if (!lp->dma_chan) { readsw(ioaddr + reg, buf, len); return; } @@ -330,26 +362,25 @@ smc_pxa_dma_insw(void __iomem *ioaddr, struct smc_local *lp, int reg, int dma, len--; } + memset(&config, 0, sizeof(config)); + config.src_addr_width = DMA_SLAVE_BUSWIDTH_2_BYTES; + config.dst_addr_width = DMA_SLAVE_BUSWIDTH_2_BYTES; + config.src_addr = lp->physaddr + reg; + config.dst_addr = lp->physaddr + reg; + config.src_maxburst = 32; + config.dst_maxburst = 32; + ret = dmaengine_slave_config(lp->dma_chan, &config); + if (ret) { + dev_err(lp->device, "dma channel configuration failed: %d\n", + ret); + return; + } + len *= 2; - dmabuf = dma_map_single(lp->device, buf, len, DMA_FROM_DEVICE); - DCSR(dma) = DCSR_NODESC; - DTADR(dma) = dmabuf; - DSADR(dma) = physaddr + reg; - DCMD(dma) = (DCMD_INCTRGADDR | DCMD_BURST32 | - DCMD_WIDTH2 | (DCMD_LENGTH & len)); - DCSR(dma) = DCSR_NODESC | DCSR_RUN; - while (!(DCSR(dma) & DCSR_STOPSTATE)) - cpu_relax(); - DCSR(dma) = 0; - dma_unmap_single(lp->device, dmabuf, len, DMA_FROM_DEVICE); + smc_pxa_dma_inpump(lp, buf, len); } #endif -static void -smc_pxa_dma_irq(int dma, void *dummy) -{ - DCSR(dma) = 0; -} #endif /* CONFIG_ARCH_PXA */
Convert the dma transfers to be dmaengine based, now pxa has a dmaengine slave driver. This makes this driver a bit more PXA agnostic. The driver was tested on pxa27x (mainstone) and pxa310 (zylonite), ie. only pxa platforms. Signed-off-by: Robert Jarzmik <robert.jarzmik@free.fr> Cc: Russell King <linux@arm.linux.org.uk> Cc: Arnd Bergmann <arnd@arndb.de> --- This has potential to break other platform such as Neponset, Idp, halibut and qsd8x50, so I added Russell and Arnd as they were discussing smc91x support last February. --- drivers/net/ethernet/smsc/smc91x.c | 29 +++++++---- drivers/net/ethernet/smsc/smc91x.h | 99 +++++++++++++++++++++++++------------- 2 files changed, 84 insertions(+), 44 deletions(-)