On Wed, Sep 16, 2020 at 12:17:55PM +0300, Viorel Suman (OSS) wrote: This looks mostly good, a few smallish things below but nothing major. > +static int fsl_xcvr_load_firmware(struct fsl_xcvr *xcvr) > +{ > + struct device *dev = &xcvr->pdev->dev; > + const struct firmware *fw; > + int ret = 0, rem, off, out, page = 0, size = FSL_XCVR_REG_OFFSET; > + u32 mask, val; > + > + ret = request_firmware(&fw, xcvr->fw_name, dev); > + if (ret) { > + dev_err(dev, "failed to request firmware.\n"); > + return ret; > + } > + > + rem = fw->size; It would be good to see some explicit validation of the image size, at least printing an error message if the image is bigger than can be loaded. The code should be safe in that it won't overflow the device region it's writing to but it feels like it'd be better to tell people if we spot a problem rather than just silently truncating the file. > + /* RAM is 20KiB => max 10 pages 2KiB each */ > + for (page = 0; page < 10; page++) { > + ret = regmap_update_bits(xcvr->regmap, FSL_XCVR_EXT_CTRL, > + FSL_XCVR_EXT_CTRL_PAGE_MASK, > + FSL_XCVR_EXT_CTRL_PAGE(page)); regmap does have paging support, though given that this is currently the only place where paging is used this probably doesn't matter too much. > +static irqreturn_t irq0_isr(int irq, void *devid) > +{ > + struct fsl_xcvr *xcvr = (struct fsl_xcvr *)devid; > + struct device *dev = &xcvr->pdev->dev; > + struct regmap *regmap = xcvr->regmap; > + void __iomem *reg_ctrl, *reg_buff; > + u32 isr, val, i; > + > + regmap_read(regmap, FSL_XCVR_EXT_ISR, &isr); > + regmap_write(regmap, FSL_XCVR_EXT_ISR_CLR, isr); This will unconditionally clear any interrupts, even those we don't understand - it might be better to only clear bits that are supported so the IRQ core can complain if there's something unexpected showing up. > + if (isr & FSL_XCVR_IRQ_FIFO_UOFL_ERR) > + dev_dbg(dev, "RX/TX FIFO full/empty\n"); Should this be dev_err()? > +static irqreturn_t irq1_isr(int irq, void *devid) > +{ > + struct fsl_xcvr *xcvr = (struct fsl_xcvr *)devid; > + struct device *dev = &xcvr->pdev->dev; > + > + dev_dbg(dev, "irq[1]: %d\n", irq); > + > + return IRQ_HANDLED; > +} Is there any value in even requesting this and irq2 given the lack of meaningful handling?
Attachment:
signature.asc
Description: PGP signature