Hello Bjorn,
On Thu, May 16, 2024 at 03:49:07PM -0500, Bjorn Helgaas wrote:
> [+cc Shawn for Rockchip question]
>
> On Tue, Mar 12, 2024 at 11:51:47AM +0100, Niklas Cassel wrote:
> > Ever since commit f25b5fae29d4 ("PCI: endpoint: Setting a BAR size > 4 GB
> > is invalid if 64-bit flag is not set") it has been impossible to get the
> > .set_bar() callback with a BAR size > 4 GB, if the BAR was also not
> > requested to be configured as a 64-bit BAR.
> >
> > Thus, forcing setting the 64-bit flag for BARs larger than 4 GB in the
> > lower level driver is dead code and can be removed.
> >
> > It is however possible that an EPF driver configures a BAR as 64-bit,
> > even if the requested size is < 4 GB.
> >
> > Respect the requested BAR configuration, just like how it is already
> > repected with regards to the prefetchable bit.
> >
> > Signed-off-by: Niklas Cassel <cassel@xxxxxxxxxx>
> > ---
> > drivers/pci/controller/cadence/pcie-cadence-ep.c | 5 +----
> > 1 file changed, 1 insertion(+), 4 deletions(-)
> >
> > diff --git a/drivers/pci/controller/cadence/pcie-cadence-ep.c b/drivers/pci/controller/cadence/pcie-cadence-ep.c
> > index 2d0a8d78bffb..de10e5edd1b0 100644
> > --- a/drivers/pci/controller/cadence/pcie-cadence-ep.c
> > +++ b/drivers/pci/controller/cadence/pcie-cadence-ep.c
> > @@ -99,14 +99,11 @@ static int cdns_pcie_ep_set_bar(struct pci_epc *epc, u8 fn, u8 vfn,
> > ctrl = CDNS_PCIE_LM_BAR_CFG_CTRL_IO_32BITS;
> > } else {
> > bool is_prefetch = !!(flags & PCI_BASE_ADDRESS_MEM_PREFETCH);
> > - bool is_64bits = sz > SZ_2G;
> > + bool is_64bits = !!(flags & PCI_BASE_ADDRESS_MEM_TYPE_64);
> >
> > if (is_64bits && (bar & 1))
> > return -EINVAL;
>
> Not relevant to *this* patch, but it looks like this code assumes that
> a 64-bit BAR must consist of an even-numbered DWORD followed by an
> odd-numbered one, e.g., it could be BAR 0 and BAR 1, but not BAR 1 and
> BAR 2.
>
> I don't think the PCIe spec requires that. Does the Cadence hardware
> require this?
The PCIe spec does not seems to require this:
"A Type 0 Configuration Space Header has six DWORD locations allocated for
Base Address registers starting at offset 10h in Configuration Space. [...]
A Function may use any of the locations to implement Base Address registers.
An implemented 64-bit Base Address register consumes two consecutive DWORD
locations."
> What about Rockchip, which has similar code in
> rockchip_pcie_ep_set_bar()?
>
> FWIW, dw_pcie_ep_set_bar() doesn't enforce this restriction.
The Synopsys PCIe controller does have this limitation:
>From the DWC databook:
3.5.7.1.1 Overview
Base Address Registers (Offset: 0x100x24)
The controller provides three pairs of 32-bit BARs for each implemented
function. Each pair (BARs 0 and 1, BARs 2 and 3, BARs 4 and 5) can be
configured as follows:
-One 64-bit BAR: For example, BARs 0 and 1 are combined to form a single
64-bit BAR.
-Two 32-bit BARs: For example, BARs 0 and 1 are two independent 32-bit BARs.
-One 32-bit BAR: For example, BAR0 is a 32-bit BAR and BAR1 is either
disabled or removed from the controller altogether to reduce gate count.
So dw_pcie_ep_set_bar() should probably be fixed to enforce this requirement.
(PCI patches for DWC appear to have a tendency to be redirected to /dev/null,
so personally I'm not planning on sending out a patch to enforce this.)
Kind regards,
Niklas
