From: Bolarinwa Olayemi Saheed <[email protected]>
On failure pcie_capability_read_dword() sets it's last parameter,
val to 0. In this case dn and up will be 0, so aspm_hw_l1_supported()
will return false.
However, with Patch 14/14, it is possible that val is set to ~0 on
failure. This would introduce a bug because (x & x) == (~0 & x). So with
dn and up being 0x02, a true value is return when the read has actually
failed.
This bug can be avoided if the return value of pcie_capability_read_dword
is checked to confirm success. The behaviour of the function remains
intact.
Check the return value of pcie_capability_read_dword() to ensure success.
Suggested-by: Bjorn Helgaas <[email protected]>
Signed-off-by: Bolarinwa Olayemi Saheed <[email protected]>
---
drivers/infiniband/hw/hfi1/aspm.c | 7 ++++---
1 file changed, 4 insertions(+), 3 deletions(-)
diff --git a/drivers/infiniband/hw/hfi1/aspm.c b/drivers/infiniband/hw/hfi1/aspm.c
index a3c53be4072c..80d0b3edd983 100644
--- a/drivers/infiniband/hw/hfi1/aspm.c
+++ b/drivers/infiniband/hw/hfi1/aspm.c
@@ -24,6 +24,7 @@ static bool aspm_hw_l1_supported(struct hfi1_devdata *dd)
{
struct pci_dev *parent = dd->pcidev->bus->self;
u32 up, dn;
+ int ret_up, ret_dn;
/*
* If the driver does not have access to the upstream component,
@@ -32,14 +33,14 @@ static bool aspm_hw_l1_supported(struct hfi1_devdata *dd)
if (!parent)
return false;
- pcie_capability_read_dword(dd->pcidev, PCI_EXP_LNKCAP, &dn);
+ ret_dn = pcie_capability_read_dword(dd->pcidev, PCI_EXP_LNKCAP, &dn);
dn = ASPM_L1_SUPPORTED(dn);
- pcie_capability_read_dword(parent, PCI_EXP_LNKCAP, &up);
+ ret_up = pcie_capability_read_dword(parent, PCI_EXP_LNKCAP, &up);
up = ASPM_L1_SUPPORTED(up);
/* ASPM works on A-step but is reported as not supported */
- return (!!dn || is_ax(dd)) && !!up;
+ return !!ret_dn && !!ret_up && (!!dn || is_ax(dd)) && !!up;
}
/* Set L1 entrance latency for slower entry to L1 */
--
2.18.2
On Mon, Jul 13, 2020 at 07:55:25PM +0200, Saheed O. Bolarinwa wrote:
> From: Bolarinwa Olayemi Saheed <[email protected]>
>
> On failure pcie_capability_read_dword() sets it's last parameter,
> val to 0. In this case dn and up will be 0, so aspm_hw_l1_supported()
> will return false.
> However, with Patch 14/14, it is possible that val is set to ~0 on
> failure. This would introduce a bug because (x & x) == (~0 & x). So with
> dn and up being 0x02, a true value is return when the read has actually
> failed.
>
> This bug can be avoided if the return value of pcie_capability_read_dword
> is checked to confirm success. The behaviour of the function remains
> intact.
>
> Check the return value of pcie_capability_read_dword() to ensure success.
>
> Suggested-by: Bjorn Helgaas <[email protected]>
> Signed-off-by: Bolarinwa Olayemi Saheed <[email protected]>
> drivers/infiniband/hw/hfi1/aspm.c | 7 ++++---
> 1 file changed, 4 insertions(+), 3 deletions(-)
>
> diff --git a/drivers/infiniband/hw/hfi1/aspm.c b/drivers/infiniband/hw/hfi1/aspm.c
> index a3c53be4072c..80d0b3edd983 100644
> +++ b/drivers/infiniband/hw/hfi1/aspm.c
> @@ -24,6 +24,7 @@ static bool aspm_hw_l1_supported(struct hfi1_devdata *dd)
> {
> struct pci_dev *parent = dd->pcidev->bus->self;
> u32 up, dn;
> + int ret_up, ret_dn;
>
> /*
> * If the driver does not have access to the upstream component,
> @@ -32,14 +33,14 @@ static bool aspm_hw_l1_supported(struct hfi1_devdata *dd)
> if (!parent)
> return false;
>
> - pcie_capability_read_dword(dd->pcidev, PCI_EXP_LNKCAP, &dn);
> + ret_dn = pcie_capability_read_dword(dd->pcidev, PCI_EXP_LNKCAP, &dn);
> dn = ASPM_L1_SUPPORTED(dn);
>
> - pcie_capability_read_dword(parent, PCI_EXP_LNKCAP, &up);
> + ret_up = pcie_capability_read_dword(parent, PCI_EXP_LNKCAP, &up);
> up = ASPM_L1_SUPPORTED(up);
>
> /* ASPM works on A-step but is reported as not supported */
> - return (!!dn || is_ax(dd)) && !!up;
> + return !!ret_dn && !!ret_up && (!!dn || is_ax(dd)) && !!up;
> }
what is all the !! for? boolean contexts already coerce to boolean
Jason