On Tue, May 06, 2025 at 06:01:15PM +0800, Keguang Zhang wrote:
> On Tue, May 6, 2025 at 5:32 PM Dan Carpenter <dan.carpenter@xxxxxxxxxx> wrote:
> >
> > On Tue, May 06, 2025 at 05:16:03PM +0800, Keguang Zhang wrote:
> > > On Fri, May 2, 2025 at 4:39 PM Dan Carpenter <dan.carpenter@xxxxxxxxxx> wrote:
> > > >
> > > > The "desc" variable is NULL and PTR_ERR(NULL) is zero/success. Return
> > > > a negative error code instead.
> > > >
> > > > Fixes: d2d10ede04b1 ("mtd: rawnand: Add Loongson-1 NAND Controller Driver")
> > > > Signed-off-by: Dan Carpenter <dan.carpenter@xxxxxxxxxx>
> > > > ---
> > > > It's hard to know what the patch prefix should be here. Ideally when we
> > > > add a new driver we would use the patch prefix for the driver.
> > > >
> > > > Tired: subsystem: Add driver XXX
> > > > Wired: subsystem: XXX: Add driver for XXX
> > > >
> > > > drivers/mtd/nand/raw/loongson1-nand-controller.c | 2 +-
> > > > 1 file changed, 1 insertion(+), 1 deletion(-)
> > > >
> > > > diff --git a/drivers/mtd/nand/raw/loongson1-nand-controller.c b/drivers/mtd/nand/raw/loongson1-nand-controller.c
> > > > index 6a369b1c7d86..8754bb4f8b56 100644
> > > > --- a/drivers/mtd/nand/raw/loongson1-nand-controller.c
> > > > +++ b/drivers/mtd/nand/raw/loongson1-nand-controller.c
> > > > @@ -371,7 +371,7 @@ static int ls1x_nand_dma_transfer(struct ls1x_nand_host *host, struct ls1x_nand_
> > > > desc = dmaengine_prep_slave_single(chan, dma_addr, op->len, xfer_dir, DMA_PREP_INTERRUPT);
> > > > if (!desc) {
> > > > dev_err(dev, "failed to prepare DMA descriptor\n");
> > > > - ret = PTR_ERR(desc);
> > > > + ret = -ENOMEM;
> > >
> > > Thank you for fixing this issue.
> > > However, I believe -EIO is more appropriate than -ENOMEM, since
> > > dmaengine_prep_slave_single() can return errors other than -ENOMEM.
> > >
> >
> > It's not an I/O error so -EIO isn't correct.
> >
> > There are a bunch of reasons it could fail but most likely
> > dma_pool_alloc() failed. I think -ENOMEM is correct.
> >
>
> Have you reviewed the implementation of ls1x_dma_prep_slave_sg()?
The ls1x_dma_prep_slave_sg() has basically two error cases, buggy drivers
and allocation errors. Someone could argue that if people pass invalid
data then the correct return is -EINVAL but in these cases it's a buggy
driver and we fix bugs, we don't work around them. So the correct
return is -ENOMEM.
> Errors in this function can be caused not only by -ENOMEM, but also by -EINVAL.
> Moreover, in most cases, the error handling logic for
> dmaengine_prep_slave_single() returns -EIO when the function returns
> NULL.
There are some that return -EIO but hardly the majority. Other places
return -ENOMEM or -EINVAL. It's not worth going back and fixing all of
these but really -ENOMEM is the correct return.
regards,
dan carpenter
