On Mon, Apr 05, 2021 at 07:22:00PM -0700, Guenter Roeck wrote:
> On Mon, Apr 05, 2021 at 04:51:55PM -0700, Guenter Roeck wrote:
> > On Mon, Apr 05, 2021 at 10:56:29AM -0700, Guenter Roeck wrote:
> > > On Mon, Apr 05, 2021 at 10:53:35AM +0200, Greg Kroah-Hartman wrote:
> > > > This is the start of the stable review cycle for the 4.9.265 release.
> > > > There are 35 patches in this series, all will be posted as a response
> > > > to this one. If anyone has any issues with these being applied, please
> > > > let me know.
> > > >
> > > > Responses should be made by Wed, 07 Apr 2021 08:50:09 +0000.
> > > > Anything received after that time might be too late.
> > > >
> > >
> > > Build results:
> > > total: 163 pass: 163 fail: 0
> > > Qemu test results:
> > > total: 383 pass: 382 fail: 1
> > > Failed tests:
> > > parisc:generic-32bit_defconfig:smp:net,pcnet:scsi[53C895A]:rootfs
> > >
> > > In the failing test, the network interfcace instantiates but fails to get
> > > an IP address. This is not a new problem but a new test. For some reason
> > > it only happens with this specific network interface, this specific SCSI
> > > controller, and with v4.9.y. No reason for concern; I'll try to track down
> > > what is going on.
> > >
> >
> > Interesting. The problem affects all kernels up to and including
> > v4.19.y. Unlike I thought initially, the problem is not associated
> > with the SCSI controller (that was coincidental) but with pcnet
> > Ethernet interfaces. It has been fixed in the upstream kernel with
> > commit 518a2f1925c3 ("dma-mapping: zero memory returned from
> > dma_alloc_*"). This patch does not apply cleanly to any of the
> > affected kernels. I backported part of it to v4.19.y and v4.9.y
> > and confirmed that it fixes the problem in those branches.
> >
> > Question is what we should do: try to backport 518a2f1925c3 to v4.19.y
> > and earlier, or stop testing against this specific problem.
> >
>
> Another update: The following code change fixes the problem as well.
> Commit 518a2f1925c3 fixes it only as side effect since it clears
> all DMA buffers.
>
> diff --git a/drivers/net/ethernet/amd/pcnet32.c b/drivers/net/ethernet/amd/pcnet32.c
> index c22bf52d3320..7a25ec8390e4 100644
> --- a/drivers/net/ethernet/amd/pcnet32.c
> +++ b/drivers/net/ethernet/amd/pcnet32.c
> @@ -1967,7 +1967,7 @@ static int pcnet32_alloc_ring(struct net_device *dev, const char *name)
> return -ENOMEM;
> }
>
> - lp->rx_ring = pci_alloc_consistent(lp->pci_dev,
> + lp->rx_ring = pci_zalloc_consistent(lp->pci_dev,
> sizeof(struct pcnet32_rx_head) *
> lp->rx_ring_size,
> &lp->rx_ring_dma_addr);
>
> I'll submit a patch implementing that; we'll see how it goes.
Sigh. That doesn't work; upstream uses dma_alloc_coherent().
We could apply the patch making the switch, but dma_alloc_coherent()
doesn't clear memory in older kernels (we are back to commit 518a2f1925c3
which is introducing that). I'll just drop pcnet tests for kernels older
than v5.4.
Guenter
