[PATCH] PCI: tegra264: Fix Link Capabilities register offset

[PATCH] PCI: tegra264: Fix Link Capabilities register offset

[Linmao Li]

The PCI Express Link Capabilities and Link Status registers are at offsets
0x0c and 0x12 from the capability header, respectively.  Since the driver
uses 0x5a for Link Status, Link Capabilities starts at 0x54, not 0x56.

Reading a 16-bit value at 0x56 returns the upper half of the Link
Capabilities register.  When a hotplug-capable port has no link during
probe, tegra264_pcie_icc_set() consequently derives the maximum speed and
width from unrelated bits and requests the wrong interconnect bandwidth.

Use the correct Link Capabilities offset.

Fixes: 01c3c27a0ef6 ("PCI: tegra264: Add Tegra264 support")
Signed-off-by: Linmao Li <lilinmao@kylinos.cn>
---
 drivers/pci/controller/pcie-tegra264.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/drivers/pci/controller/pcie-tegra264.c b/drivers/pci/controller/pcie-tegra264.c
index 653136db401e0..3a672d18c4d2d 100644
--- a/drivers/pci/controller/pcie-tegra264.c
+++ b/drivers/pci/controller/pcie-tegra264.c
@@ -49,7 +49,7 @@
 #define XAL_RC_BAR_CNTL_STANDARD_64B_BAR_EN	BIT(2)
 
 /* XTL registers */
-#define XTL_RC_PCIE_CFG_LINK_CAPS		0x56
+#define XTL_RC_PCIE_CFG_LINK_CAPS		0x54
 #define XTL_RC_PCIE_CFG_LINK_STATUS		0x5a
 
 #define XTL_RC_MGMT_PERST_CONTROL		0x218
-- 
2.25.1

Re: [PATCH] PCI: tegra264: Fix Link Capabilities register offset

[Thierry Reding]

[-- Attachment #1: Type: text/plain, Size: 957 bytes --]

On Thu, Aug 27, 2026 at 05:39:19PM +0800, Linmao Li wrote:
> The PCI Express Link Capabilities and Link Status registers are at offsets
> 0x0c and 0x12 from the capability header, respectively.  Since the driver
> uses 0x5a for Link Status, Link Capabilities starts at 0x54, not 0x56.
> 
> Reading a 16-bit value at 0x56 returns the upper half of the Link
> Capabilities register.  When a hotplug-capable port has no link during
> probe, tegra264_pcie_icc_set() consequently derives the maximum speed and
> width from unrelated bits and requests the wrong interconnect bandwidth.
> 
> Use the correct Link Capabilities offset.
> 
> Fixes: 01c3c27a0ef6 ("PCI: tegra264: Add Tegra264 support")
> Signed-off-by: Linmao Li <lilinmao@kylinos.cn>
> ---
>  drivers/pci/controller/pcie-tegra264.c | 2 +-
>  1 file changed, 1 insertion(+), 1 deletion(-)

Yes, that's what the manual says, too. /o\

Acked-by: Thierry Reding <treding@nvidia.com>

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 833 bytes --]

Re: [PATCH] PCI: tegra264: Fix Link Capabilities register offset

[Bjorn Helgaas]

On Thu, Aug 27, 2026 at 05:39:19PM +0800, Linmao Li wrote:
> The PCI Express Link Capabilities and Link Status registers are at offsets
> 0x0c and 0x12 from the capability header, respectively.  Since the driver
> uses 0x5a for Link Status, Link Capabilities starts at 0x54, not 0x56.
> 
> Reading a 16-bit value at 0x56 returns the upper half of the Link
> Capabilities register.  When a hotplug-capable port has no link during
> probe, tegra264_pcie_icc_set() consequently derives the maximum speed and
> width from unrelated bits and requests the wrong interconnect bandwidth.
> 
> Use the correct Link Capabilities offset.
> 
> Fixes: 01c3c27a0ef6 ("PCI: tegra264: Add Tegra264 support")
> Signed-off-by: Linmao Li <lilinmao@kylinos.cn>
> ---
>  drivers/pci/controller/pcie-tegra264.c | 2 +-
>  1 file changed, 1 insertion(+), 1 deletion(-)
> 
> diff --git a/drivers/pci/controller/pcie-tegra264.c b/drivers/pci/controller/pcie-tegra264.c
> index 653136db401e0..3a672d18c4d2d 100644
> --- a/drivers/pci/controller/pcie-tegra264.c
> +++ b/drivers/pci/controller/pcie-tegra264.c
> @@ -49,7 +49,7 @@
>  #define XAL_RC_BAR_CNTL_STANDARD_64B_BAR_EN	BIT(2)
>  
>  /* XTL registers */
> -#define XTL_RC_PCIE_CFG_LINK_CAPS		0x56
> +#define XTL_RC_PCIE_CFG_LINK_CAPS		0x54
>  #define XTL_RC_PCIE_CFG_LINK_STATUS		0x5a

If this is a standard PCIe Capability, and since we use the standard
#defines like PCI_EXP_LNKSTA_CLS and PCI_EXP_LNKCAP_SLS for values
read from it, it appears to be, I think these XTL_RC* #defines should
somehow be based on the existing PCI_EXP_LNKSTA and PCI_EXP_LNKCAP.

E.g., we could read from pcie->ecam + XTL_RC_PCIE_CAP + PCI_EXP_LNKSTA.
That way grep would be more useful.

>  #define XTL_RC_MGMT_PERST_CONTROL		0x218
> -- 
> 2.25.1
> 

Re: [PATCH] PCI: tegra264: Fix Link Capabilities register offset

[Thierry Reding]

[-- Attachment #1: Type: text/plain, Size: 2020 bytes --]

On Thu, Aug 27, 2026 at 12:03:58PM -0500, Bjorn Helgaas wrote:
> On Thu, Aug 27, 2026 at 05:39:19PM +0800, Linmao Li wrote:
> > The PCI Express Link Capabilities and Link Status registers are at offsets
> > 0x0c and 0x12 from the capability header, respectively.  Since the driver
> > uses 0x5a for Link Status, Link Capabilities starts at 0x54, not 0x56.
> > 
> > Reading a 16-bit value at 0x56 returns the upper half of the Link
> > Capabilities register.  When a hotplug-capable port has no link during
> > probe, tegra264_pcie_icc_set() consequently derives the maximum speed and
> > width from unrelated bits and requests the wrong interconnect bandwidth.
> > 
> > Use the correct Link Capabilities offset.
> > 
> > Fixes: 01c3c27a0ef6 ("PCI: tegra264: Add Tegra264 support")
> > Signed-off-by: Linmao Li <lilinmao@kylinos.cn>
> > ---
> >  drivers/pci/controller/pcie-tegra264.c | 2 +-
> >  1 file changed, 1 insertion(+), 1 deletion(-)
> > 
> > diff --git a/drivers/pci/controller/pcie-tegra264.c b/drivers/pci/controller/pcie-tegra264.c
> > index 653136db401e0..3a672d18c4d2d 100644
> > --- a/drivers/pci/controller/pcie-tegra264.c
> > +++ b/drivers/pci/controller/pcie-tegra264.c
> > @@ -49,7 +49,7 @@
> >  #define XAL_RC_BAR_CNTL_STANDARD_64B_BAR_EN	BIT(2)
> >  
> >  /* XTL registers */
> > -#define XTL_RC_PCIE_CFG_LINK_CAPS		0x56
> > +#define XTL_RC_PCIE_CFG_LINK_CAPS		0x54
> >  #define XTL_RC_PCIE_CFG_LINK_STATUS		0x5a
> 
> If this is a standard PCIe Capability, and since we use the standard
> #defines like PCI_EXP_LNKSTA_CLS and PCI_EXP_LNKCAP_SLS for values
> read from it, it appears to be, I think these XTL_RC* #defines should
> somehow be based on the existing PCI_EXP_LNKSTA and PCI_EXP_LNKCAP.
> 
> E.g., we could read from pcie->ecam + XTL_RC_PCIE_CAP + PCI_EXP_LNKSTA.
> That way grep would be more useful.

That's a good point. And, yes, this does look like a regular PCIe
capability, so I'll try to rework this using the standard defines.

Thierry

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 833 bytes --]

Re: [PATCH] PCI: tegra264: Fix Link Capabilities register offset

[Manivannan Sadhasivam]

On Thu, 27 Aug 2026 17:39:19 +0800, Linmao Li wrote:
> The PCI Express Link Capabilities and Link Status registers are at offsets
> 0x0c and 0x12 from the capability header, respectively.  Since the driver
> uses 0x5a for Link Status, Link Capabilities starts at 0x54, not 0x56.
> 
> Reading a 16-bit value at 0x56 returns the upper half of the Link
> Capabilities register.  When a hotplug-capable port has no link during
> probe, tegra264_pcie_icc_set() consequently derives the maximum speed and
> width from unrelated bits and requests the wrong interconnect bandwidth.
> 
> [...]

Applied, thanks!

[1/1] PCI: tegra264: Fix Link Capabilities register offset
      commit: ed17593383fe3926950cae250842783630f32697

Best regards,
-- 
மணிவண்ணன் சதாசிவம்