[PATCH] pci/probe: Ignore the device id while handling CRS

[PATCH] pci/probe: Ignore the device id while handling CRS

[Rajat Jain]

Some root ports may return the actual device id returned by the
endpoint, instead of faking up a synthetic id (0xFFFF) as specified
by the PCIe spec (sec 2.3.2). We have seen such broken devices in the
past that match this behaviour:
(Ref: https://lkml.org/lkml/2007/11/1/198)

To take care of the same, CRS was disabled then, however, we want to
enable CRS now (https://lkml.org/lkml/2014/9/2/509), but still do not
want to break such devices. Thus check only against the
(vendor id == 1) and ignore the device id.

Signed-off-by: Rajat Jain <rajatxjain@gmail.com>
Signed-off-by: Rajat Jain <rajatjain@juniper.net>
Signed-off-by: Guenter Roeck <groeck@juniper.net>
---
 drivers/pci/probe.c |    9 +++++++--
 1 file changed, 7 insertions(+), 2 deletions(-)

diff --git a/drivers/pci/probe.c b/drivers/pci/probe.c
index 3c4c35c..0076a9b 100644
--- a/drivers/pci/probe.c
+++ b/drivers/pci/probe.c
@@ -1305,8 +1305,13 @@ bool pci_bus_read_dev_vendor_id(struct pci_bus *bus, int devfn, u32 *l,
 	    *l == 0x0000ffff || *l == 0xffff0000)
 		return false;
 
-	/* Configuration request Retry Status */
-	while (*l == 0xffff0001) {
+	/*
+	 * Configuration request Retry Status. Note that some root ports
+	 * are broken and may return actual device ID instead of a synthetic ID
+	 * (0xFFFF) to be faked as per the PCIe spec. Hence ignore the device ID
+	 * and only check for (vendor id == 1)
+	 */
+	while (*l & 0xffff == 0x0001) {
 		if (!crs_timeout)
 			return false;
 
-- 
1.7.9.5

Re: [PATCH] pci/probe: Ignore the device id while handling CRS

[Bjorn Helgaas]

On Mon, Sep 08, 2014 at 02:19:49PM -0700, Rajat Jain wrote:
> Some root ports may return the actual device id returned by the
> endpoint, instead of faking up a synthetic id (0xFFFF) as specified
> by the PCIe spec (sec 2.3.2). We have seen such broken devices in the
> past that match this behaviour:
> (Ref: https://lkml.org/lkml/2007/11/1/198)
> 
> To take care of the same, CRS was disabled then, however, we want to
> enable CRS now (https://lkml.org/lkml/2014/9/2/509), but still do not
> want to break such devices. Thus check only against the
> (vendor id == 1) and ignore the device id.
> 
> Signed-off-by: Rajat Jain <rajatxjain@gmail.com>
> Signed-off-by: Rajat Jain <rajatjain@juniper.net>
> Signed-off-by: Guenter Roeck <groeck@juniper.net>

Applied to pci/enumeration for v3.18, thanks!

> ---
>  drivers/pci/probe.c |    9 +++++++--
>  1 file changed, 7 insertions(+), 2 deletions(-)
> 
> diff --git a/drivers/pci/probe.c b/drivers/pci/probe.c
> index 3c4c35c..0076a9b 100644
> --- a/drivers/pci/probe.c
> +++ b/drivers/pci/probe.c
> @@ -1305,8 +1305,13 @@ bool pci_bus_read_dev_vendor_id(struct pci_bus *bus, int devfn, u32 *l,
>  	    *l == 0x0000ffff || *l == 0xffff0000)
>  		return false;
>  
> -	/* Configuration request Retry Status */
> -	while (*l == 0xffff0001) {
> +	/*
> +	 * Configuration request Retry Status. Note that some root ports
> +	 * are broken and may return actual device ID instead of a synthetic ID
> +	 * (0xFFFF) to be faked as per the PCIe spec. Hence ignore the device ID
> +	 * and only check for (vendor id == 1)
> +	 */
> +	while (*l & 0xffff == 0x0001) {
>  		if (!crs_timeout)
>  			return false;
>  
> -- 
> 1.7.9.5
>