Hi,
On Tue, 2023-11-21 at 14:34 -0800, Stanislav Kinsburskii wrote:
>
> External email : Please do not click links or open attachments until
> you have verified the sender or the content.
> On Tue, Nov 21, 2023 at 04:23:25PM -0600, Bjorn Helgaas wrote:
> > On Tue, Nov 21, 2023 at 02:05:56PM -0800, Stanislav Kinsburskii
> wrote:
> > > On Tue, Nov 21, 2023 at 03:37:55PM -0600, Bjorn Helgaas wrote:
> > > > On Mon, Nov 20, 2023 at 03:40:33PM -0800, Stanislav Kinsburskii
> wrote:
> > > > > Explicitly cast __iomem pointer to const void* with __force
> to fix the
> > > > > following warning:
> > > > >
> > > > > warning: incorrect type in argument 1 (different address
> spaces)
> > > > > expected void const volatile *address
> > > > > got void [noderef] __iomem *
> > > >
> > > > I have two questions about this:
> > > >
> > > > 1) There's no other use of __force in drivers/pci, so I don't
> know
> > > > what's special about pcie-mediatek.c. There should be a way
> to fix
> > > > the types so it's not needed.
> > >
> > > __force suppreses the following sparse warning:
> > >
> > > warning: cast removes address space '__iomem' of expression
> >
> > I'm suggesting that the cast is a band-aid that covers up a type
> > mismatch, and there shouldn't be a mismatch in the first place.
> >
> > > > 2) virt_to_phys() is not quite right to begin with because
> what we
> > > > want is a *bus* address, not the CPU physical address we get
> from
> > > > virt_to_phys(). Obviously the current platforms that use
> this must
> > > > not apply any offset between bus and CPU physical addresses,
> but
> > > > it's not something we should rely on.
> > > >
> > > > There are only three drivers (pci-aardvark.c, pcie-xilinx.c,
> and
> > > > this one) that use virt_to_phys(), and they're all slightly
> wrong
> > > > here.
> > > >
> > > > The *_compose_msi_msg() methods could use a little more
> consistency
> > > > across the board.
> > >
> > > Could you elaborate on what do you suggest?
> > > Should virt_to_phys() be simply removed?
> >
> > The DMA API (Documentation/core-api/dma-api.rst) is the usual way
> to
> > get bus addresses, since an MSI is basically a DMA on the PCI bus.
> >
> > https://lore.kernel.org/linux-pci/20230914203146.GA77870@bhelgaas/
> >
> > Nobody is very motivated to fix these, I guess ;) I sort of hate
> to
> > just throw in a cast to shut up the warning because it doesn't
> really
> > solve the problem.
> >
>
> This is fair.
> However I have neither expertize to assess, no hardware to test such
> intrusive
> changes.
> I guess it's a no-op then, is it?
I think I can try and test 'dmam_alloc_coherent()' or other proper APIs
on the MediaTek platform, and if it works as expected, I can send the
patch to fix this.
Thanks.
>
> > > > > Signed-off-by: Stanislav Kinsburskii <
> [email protected]>
> > > > > ---
> > > > > drivers/pci/controller/pcie-mediatek.c | 4 ++--
> > > > > 1 file changed, 2 insertions(+), 2 deletions(-)
> > > > >
> > > > > diff --git a/drivers/pci/controller/pcie-mediatek.c
> b/drivers/pci/controller/pcie-mediatek.c
> > > > > index 66a8f73296fc..27f0f79810a1 100644
> > > > > --- a/drivers/pci/controller/pcie-mediatek.c
> > > > > +++ b/drivers/pci/controller/pcie-mediatek.c
> > > > > @@ -397,7 +397,7 @@ static void mtk_compose_msi_msg(struct
> irq_data *data, struct msi_msg *msg)
> > > > > phys_addr_t addr;
> > > > >
> > > > > /* MT2712/MT7622 only support 32-bit MSI addresses */
> > > > > -addr = virt_to_phys(port->base + PCIE_MSI_VECTOR);
> > > > > +addr = virt_to_phys((__force const void *)port->base +
> PCIE_MSI_VECTOR);
> > > > > msg->address_hi = 0;
> > > > > msg->address_lo = lower_32_bits(addr);
> > > > >
> > > > > @@ -520,7 +520,7 @@ static void mtk_pcie_enable_msi(struct
> mtk_pcie_port *port)
> > > > > u32 val;
> > > > > phys_addr_t msg_addr;
> > > > >
> > > > > -msg_addr = virt_to_phys(port->base + PCIE_MSI_VECTOR);
> > > > > +msg_addr = virt_to_phys((__force const void *)port->base +
> PCIE_MSI_VECTOR);
> > > > > val = lower_32_bits(msg_addr);
> > > > > writel(val, port->base + PCIE_IMSI_ADDR);