Re: [PATCH 7/9] PCI: endpoint: pci-epf-test: Advertise reserved BARs

[Niklas Cassel <cassel@kernel.org>]

On Tue, Feb 17, 2026 at 06:02:55PM -0500, Frank Li wrote:
> On Tue, Feb 17, 2026 at 10:27:13PM +0100, Niklas Cassel wrote:
> > Advertise reserved BARs as reserved in the Capabilities register,
> > such that the host side driver will be able to skip reserved BARs.
> >
> > Signed-off-by: Niklas Cassel <cassel@kernel.org>
> > ---
> >  drivers/pci/endpoint/functions/pci-epf-test.c | 24 +++++++++++++++++++
> >  1 file changed, 24 insertions(+)
> >
> > diff --git a/drivers/pci/endpoint/functions/pci-epf-test.c b/drivers/pci/endpoint/functions/pci-epf-test.c
> > index 0cb7af0919dc..4c7e42fe5d8e 100644
> > --- a/drivers/pci/endpoint/functions/pci-epf-test.c
> > +++ b/drivers/pci/endpoint/functions/pci-epf-test.c
> > @@ -64,6 +64,12 @@
> >  #define CAP_MSIX			BIT(2)
> >  #define CAP_INTX			BIT(3)
> >  #define CAP_SUBRANGE_MAPPING		BIT(4)
> > +#define CAP_BAR0_RESERVED		BIT(5)
> > +#define CAP_BAR1_RESERVED		BIT(6)
> > +#define CAP_BAR2_RESERVED		BIT(7)
> > +#define CAP_BAR3_RESERVED		BIT(8)
> > +#define CAP_BAR4_RESERVED		BIT(9)
> > +#define CAP_BAR5_RESERVED		BIT(10)
> >
> >  #define PCI_EPF_TEST_BAR_SUBRANGE_NSUB	2
> >
> > @@ -1106,6 +1112,24 @@ static void pci_epf_test_set_capabilities(struct pci_epf *epf)
> >  	    epf_test->epc_features->subrange_mapping)
> >  		caps |= CAP_SUBRANGE_MAPPING;
> >
> > +	if (epf_test->epc_features->bar[BAR_0].type == BAR_RESERVED)
> > +		caps |= CAP_BAR0_RESERVED;
> 
> Is it BAR_DISABLE? suppose BAR_RESERVED redefine as hardware MMIO mapped
> bar


A BAR that is defined as BAR_DISABLE will always be disabled by the EPC
driver, so the host side driver will not see the BAR in the first place.

If you try to run pci_endpoint_test against a disabled BAR, the size will
be zero, and pci_endpoint_test will return -ENODATA:
https://github.com/torvalds/linux/blob/v6.19/drivers/misc/pci_endpoint_test.c#L300-L302

and the pci_endpoint selftest will print SKIP for that test case.


BAR_RESERVED is describing a BAR that has hardware backed resources, e.g. eDMA
registers, iATU registers, MSI-X table. These BARs are never disabled by an
EPC driver by default. See the kdoc for these enum values here:
https://lore.kernel.org/linux-pci/20260217212707.2450423-15-cassel@kernel.org/


Thus, we need to tell pci_endpoint_test that these reserved BARs (which are
not disabled) has to be skipped, because if we perfrom READ/WRITE tests
against these BARs, bad things will happen. E.g. if you write to a BAR that
exposes iATU registers, you will clear/overwrite the current inbound addresss
translation configured by the endpoint.


Kind regards,
Niklas