Hi Vinod, Thanks for looking at this closely. On Fri, Oct 5, 2018 at 11:57 PM Vinod <vkoul@xxxxxxxxxx> wrote: > > On 13-09-18, 09:51, Masahiro Yamada wrote: > > > +#define UNIPHIER_MDMAC_CH_IRQ_STAT 0x010 // current hw status (RO) > > +#define UNIPHIER_MDMAC_CH_IRQ_REQ 0x014 // latched STAT (WOC) > > +#define UNIPHIER_MDMAC_CH_IRQ_EN 0x018 // IRQ enable mask > > +#define UNIPHIER_MDMAC_CH_IRQ_DET 0x01c // REQ & EN (RO) > > +#define UNIPHIER_MDMAC_CH_IRQ__ABORT BIT(13) > > +#define UNIPHIER_MDMAC_CH_IRQ__DONE BIT(1) > > why notation if UNIPHIER_MDMAC_CH_IRQ__FOO ? Macros without double-underscore are just register offsets. Macros with double-underscore are bit flags in the corresponding register. I often use this notation, and I also see somebody else did so. For example, see drivers/mtd/nand/raw/denali.h Please let me know if you do not like it. I can adjust myself to your preference. > > +#define UNIPHIER_MDMAC_CH_SRC_MODE 0x020 // mode of source > > +#define UNIPHIER_MDMAC_CH_DEST_MODE 0x024 // mode of destination > > +#define UNIPHIER_MDMAC_CH_MODE__ADDR_INC (0 << 4) > > +#define UNIPHIER_MDMAC_CH_MODE__ADDR_DEC (1 << 4) > > +#define UNIPHIER_MDMAC_CH_MODE__ADDR_FIXED (2 << 4) > > +#define UNIPHIER_MDMAC_CH_SRC_ADDR 0x028 // source address > > +#define UNIPHIER_MDMAC_CH_DEST_ADDR 0x02c // destination address > > +#define UNIPHIER_MDMAC_CH_SIZE 0x030 // transfer bytes > > Please use /* comment */ style for these I just thought people are getting tolerant of C++ comments. Linus is so: https://lkml.org/lkml/2017/11/25/133 However, C++ is not officially allowed in the Linux coding style. Will fix (without odd closing */ alignment) > > +/* mc->vc.lock must be held by caller */ > > +static struct uniphier_mdmac_desc *__uniphier_mdmac_next_desc( > > + struct uniphier_mdmac_chan *mc) > > this can be made to look better by: > > static struct uniphier_mdmac_desc * > __uniphier_mdmac_next_desc(struct uniphier_mdmac_chan *mc) OK. This is not mentioned in the coding style doc at least, but common enough. Will fix. > Btw why leading __ for function name here and other places? Just a reminder of "mc->vc.lock must be held by caller". It is common to use double-underscore prefixing for functions that should be used with care. However, I am happy to adjust myself to the maintainer. Please let me know if you do not like it, then I will remove them out. > > +{ > > + struct virt_dma_desc *vd; > > + > > + vd = vchan_next_desc(&mc->vc); > > + if (!vd) { > > + mc->md = NULL; > > + return NULL; > > + } > > + > > + list_del(&vd->node); > > + > > + mc->md = to_uniphier_mdmac_desc(vd); > > + > > + return mc->md; > > +} > > + > > +/* mc->vc.lock must be held by caller */ > > +static void __uniphier_mdmac_handle(struct uniphier_mdmac_chan *mc, > > + struct uniphier_mdmac_desc *md) > > please align this to previous line opening brace (hint checkpatch with > --strict option will give you the warning) This is already aligned. Perhaps, due to your mailer. > > +static irqreturn_t uniphier_mdmac_interrupt(int irq, void *dev_id) > > +{ > > + struct uniphier_mdmac_chan *mc = dev_id; > > + struct uniphier_mdmac_desc *md; > > + irqreturn_t ret = IRQ_HANDLED; > > + u32 irq_stat; > > + > > + spin_lock(&mc->vc.lock); > > + > > + irq_stat = readl(mc->reg_ch_base + UNIPHIER_MDMAC_CH_IRQ_DET); > > + > > + /* > > + * Some channels share a single interrupt line. If the IRQ status is 0, > > + * this is probably triggered by a different channel. > > + */ > > + if (!irq_stat) { > > + ret = IRQ_NONE; > > + goto out; > > + } > > + > > + /* write 1 to clear */ > > + writel(irq_stat, mc->reg_ch_base + UNIPHIER_MDMAC_CH_IRQ_REQ); > > + > > + /* > > + * UNIPHIER_MDMAC_CH_IRQ__DONE interrupt is asserted even when the DMA > > + * is aborted. To distinguish the normal completion and the abort, > ^^^^ > double space.. OK, will fix. > > +static int uniphier_mdmac_config(struct dma_chan *chan, > > + struct dma_slave_config *config) > > +{ > > + /* Nothing in struct dma_slave_config is configurable. */ > > + return 0; > > +} > > I dont think config callback is mandatory, so we can drop this Will remove. > > +static enum dma_status uniphier_mdmac_tx_status(struct dma_chan *chan, > > + dma_cookie_t cookie, > > + struct dma_tx_state *txstate) > > +{ > > + struct virt_dma_chan *vc; > > + struct virt_dma_desc *vd; > > + struct uniphier_mdmac_chan *mc; > > + struct uniphier_mdmac_desc *md = NULL; > > + enum dma_status stat; > > + unsigned long flags; > > + > > + stat = dma_cookie_status(chan, cookie, txstate); > > + /* Return immediately if we do not need to compute the residue. */ > > + if (stat == DMA_COMPLETE || !txstate) > > + return stat; > > + > > + vc = to_virt_chan(chan); > > + > > + spin_lock_irqsave(&vc->lock, flags); > > + > > + mc = to_uniphier_mdmac_chan(vc); > > + > > + if (mc->md && mc->md->vd.tx.cookie == cookie) > > + md = mc->md; > > + > > + if (!md) { > > + vd = vchan_find_desc(vc, cookie); > > + if (vd) > > + md = to_uniphier_mdmac_desc(vd); > > + } > > in both of these cases you are calling __uniphier_mdmac_get_residue() > which reads the register and updates. But I think you should read > register only in the first case when descriptor is submitted and not in > latter case when descriptor is queued Good catch! Will fix. > > +static int uniphier_mdmac_probe(struct platform_device *pdev) > > +{ > > + struct device *dev = &pdev->dev; > > + struct uniphier_mdmac_device *mdev; > > + struct dma_device *ddev; > > + struct resource *res; > > + int nr_chans, ret, i; > > + > > + nr_chans = platform_irq_count(pdev); > > + if (nr_chans < 0) > > + return nr_chans; > > + > > + ret = dma_set_mask(dev, DMA_BIT_MASK(32)); > > + if (ret) > > + return ret; > > + > > + mdev = devm_kzalloc(dev, struct_size(mdev, channels, nr_chans), > > + GFP_KERNEL); > > + if (!mdev) > > + return -ENOMEM; > > + > > + res = platform_get_resource(pdev, IORESOURCE_MEM, 0); > > + mdev->reg_base = devm_ioremap_resource(dev, res); > > + if (IS_ERR(mdev->reg_base)) > > + return PTR_ERR(mdev->reg_base); > > + > > + mdev->clk = devm_clk_get(dev, NULL); > > + if (IS_ERR(mdev->clk)) { > > + dev_err(dev, "failed to get clock\n"); > > + return PTR_ERR(mdev->clk); > > + } > > + > > + ret = clk_prepare_enable(mdev->clk); > > + if (ret) > > + return ret; > > + > > + ddev = &mdev->ddev; > > + ddev->dev = dev; > > + dma_cap_set(DMA_PRIVATE, ddev->cap_mask); > > + ddev->src_addr_widths = BIT(DMA_SLAVE_BUSWIDTH_UNDEFINED); > > + ddev->dst_addr_widths = BIT(DMA_SLAVE_BUSWIDTH_UNDEFINED); > > undefined? Precisely, I do not know the *_addr_widths. As far as I read dmaengine/provider.rst this represents the data bytes that are read/written at a time. Really I do not know (care about) the transfer width. As I commented in v2, the connection of the device side is hard-wired. The transfer width cannot be observed from SW view. What should I do? > > +static int uniphier_mdmac_remove(struct platform_device *pdev) > > +{ > > + struct uniphier_mdmac_device *mdev = platform_get_drvdata(pdev); > > + > > + of_dma_controller_free(pdev->dev.of_node); > > + dma_async_device_unregister(&mdev->ddev); > > + clk_disable_unprepare(mdev->clk); > > at this point your irq is registered and can be fired, the tasklets are > not killed :( Please let me clarify the concerns here. Before the .remove hook is called, all the consumers should have already put the dma channels. So, no new descriptor is coming in. However, Some already-issued descriptors might be remaining, and being processed. [1] This DMA engine might be still running when clk_disable_unprepare() is being called. The register access with its clock disabled would cause the system crash. [2] vchan_cookie_complete() might being called at this point and schedule the tasklet. It might call uniphier_mdmac_desc_free() after the reference disapperrs. Is this correct? Do you have recommendation for module removal guideline? -- Best Regards Masahiro Yamada
