On Mon, Jul 31, 2017 at 03:59:35PM +0300, Sergei Shtylyov wrote:
> On 07/31/2017 12:57 PM, Sergei Shtylyov wrote:
>
> > > > of_irq_get() may return 0 as well as negative error number on failure,
> > > > while the driver only checks for the negative values. The driver would then
> > > > call request_irq(0, ...) in tegra_adma_alloc_chan_resources() and never get
> > > > valid channel interrupt.
> > > >
> > > > Check for 'tdc->irq <= 0' instead and return -ENXIO from the driver's probe
> > > > iff of_irq_get() returned 0.
> > > >
> > > > Fixes: f46b195799b5 ("dmaengine: tegra-adma: Add support for Tegra210 ADMA")
> > > > Signed-off-by: Sergei Shtylyov <sergei.shtylyov@xxxxxxxxxxxxxxxxxx>
> [...]
>
> > > > Index: slave-dma/drivers/dma/tegra210-adma.c
> > > > ===================================================================
> > > > --- slave-dma.orig/drivers/dma/tegra210-adma.c
> > > > +++ slave-dma/drivers/dma/tegra210-adma.c
> > > > @@ -717,8 +717,8 @@ static int tegra_adma_probe(struct platf
> > > > tdc->chan_addr = tdma->base_addr + ADMA_CH_REG_OFFSET(i);
> > > > tdc->irq = of_irq_get(pdev->dev.of_node, i);
> > > > - if (tdc->irq < 0) {
> > > > - ret = tdc->irq;
> > > > + if (tdc->irq <= 0) {
> > > > + ret = tdc->irq ?: -ENXIO;
> > > > goto irq_dispose;
> > > > }
> > > >
> > >
> > > I think in this particular case it would be better to just call
> > > platform_get_irq(), which already implements the equivalent.
>
> Not that platform_get_irq() was no better (if not worse indeed, as it
> could return 0 both on failure ad success), until I fixed it back in 2016:
>
> https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/commit/?id=e330b9a6bb35dc7097a4f02cb1ae7b6f96df92af
>
> >
> > OK, I'll try looking into that...
>
> I've looked at the commit history and no, I don't agree to use
> platform_get_irq(): my platform_get_irq() fix landed in mainline later than
> this driver, so fixing this bug in stable kernels would ensue an extra
> dependency...
Well, surely the platform_get_irq() function should also be fixed in
stable kernels, so that all callers can be fixed at the same time.
That said, this does fix a bug, and it fixes it in the least invasive
way, and the damage I'm complaining about is preexisting, so:
Acked-by: Thierry Reding <treding@xxxxxxxxxx>
Attachment:
signature.asc
Description: PGP signature
