On Tue, Feb 20, 2024 at 01:22:11PM +0100, Niklas Schnelle wrote:
> On Mon, 2024-02-19 at 11:08 +0100, Heiko Carstens wrote:
> > On Fri, Feb 16, 2024 at 08:48:14PM -0400, Jason Gunthorpe wrote:
> > > The signature for __iowrite64_copy() requires the number of 64 bit
> > > quantities, not bytes. Multiple by 8 to get to a byte length before
> > > invoking zpci_memcpy_toio()
> > >
> > > Fixes: 87bc359b9822 ("s390/pci: speed up __iowrite64_copy by using pci store block insn")
> > > Acked-by: Niklas Schnelle <schnelle@xxxxxxxxxxxxx>
> > > Signed-off-by: Jason Gunthorpe <jgg@xxxxxxxxxx>
> > > ---
> > > arch/s390/pci/pci.c | 2 +-
> > > 1 file changed, 1 insertion(+), 1 deletion(-)
> > >
> > > diff --git a/arch/s390/pci/pci.c b/arch/s390/pci/pci.c
> > > index 676ac74026a82b..52a44e353796c0 100644
> > > --- a/arch/s390/pci/pci.c
> > > +++ b/arch/s390/pci/pci.c
> > > @@ -252,7 +252,7 @@ resource_size_t pcibios_align_resource(void *data, const struct resource *res,
> > > /* combine single writes by using store-block insn */
> > > void __iowrite64_copy(void __iomem *to, const void *from, size_t count)
> > > {
> > > - zpci_memcpy_toio(to, from, count);
> > > + zpci_memcpy_toio(to, from, count * 8);
> > > }
> >
> > Odd, this bug is 11 years old, and there was never any bug report.
> > Is this never called, or how is this possible?
> >
> > Niklas, would you happen to have an idea?
>
> I did ask myself the same question when I first saw this change but
> still don't have a satisfactory answer.
>
> As far as I can tell there is only one call that could be relevant for
> us in drivers/net/ethernet/mellanox/mlx4/en_tx.c:mlx4_bf_copy(). That
> call looks like it would be done during transmit.
I'm pretty sure a short write to even mlx4 will trigger the NIC to do
the slow path anyhow, so it would have been hidden. (assuming anyone
ever tested mlx4 on a kernel with this, it sure is old at this point)
Jason
