[PATCH] e1000: repost work around 82571 completion timout on pseries HW

[PATCH] e1000: repost work around 82571 completion timout on pseries HW

[Andy Gospodarek]

There seems to have been some discussion about a patch like this in the
past but I still haven't noticed any platforms fixes or noticed that
this got included, so I'd like to propose this modified version.

Thoughts?

Signed-off-by: Andy Gospodarek <andy@greyhouse.net>
---

 e1000_hw.c |    7 +++++++
 e1000_hw.h |    1 +
 2 files changed, 8 insertions(+)

diff --git a/drivers/net/e1000/e1000_hw.c b/drivers/net/e1000/e1000_hw.c
index 9be4469..7c75b8b 100644
--- a/drivers/net/e1000/e1000_hw.c
+++ b/drivers/net/e1000/e1000_hw.c
@@ -1030,6 +1030,13 @@ e1000_init_hw(struct e1000_hw *hw)
         break;
     }
 
+#if defined(CONFIG_PPC_PSERIES)
+    if (hw->mac_type == e1000_82571) {
+        uint32_t gcr = E1000_READ_REG(hw, GCR);
+        gcr |= E1000_GCR_DISABLE_TIMEOUT_MECHANISM;
+        E1000_WRITE_REG(hw, GCR, gcr);
+    }
+#endif 
 
     if (hw->mac_type == e1000_82573) {
         uint32_t gcr = E1000_READ_REG(hw, GCR);
diff --git a/drivers/net/e1000/e1000_hw.h b/drivers/net/e1000/e1000_hw.h
index bd000b8..71f4e2d 100644
--- a/drivers/net/e1000/e1000_hw.h
+++ b/drivers/net/e1000/e1000_hw.h
@@ -2211,6 +2211,7 @@ struct e1000_host_command_info {
 #define PCI_EX_82566_SNOOP_ALL PCI_EX_NO_SNOOP_ALL
 
 #define E1000_GCR_L1_ACT_WITHOUT_L0S_RX 0x08000000
+#define E1000_GCR_DISABLE_TIMEOUT_MECHANISM 0x80000000
 /* Function Active and Power State to MNG */
 #define E1000_FACTPS_FUNC0_POWER_STATE_MASK         0x00000003
 #define E1000_FACTPS_LAN0_VALID                     0x00000004

Re: [PATCH] e1000: repost work around 82571 completion timout on pseries HW

[Kok, Auke]

Andy Gospodarek wrote:
> There seems to have been some discussion about a patch like this in the
> past but I still haven't noticed any platforms fixes or noticed that
> this got included, so I'd like to propose this modified version.
> 
> Thoughts?

I've written a completely new version for IBM based on the response that 
implements this as a pci quirk (to tag the chipset, not the device).

However IBM has spotted an issue due to firmware hiding the id for linux of the 
bridge involved (on pseries), and they're working on an alternative to fix that.

So, the patch will look different, and I'll post it as soon as we have a version 
for both pseries and xseries that works.

Auke


> 
> Signed-off-by: Andy Gospodarek <andy@greyhouse.net>
> ---
> 
>  e1000_hw.c |    7 +++++++
>  e1000_hw.h |    1 +
>  2 files changed, 8 insertions(+)
> 
> diff --git a/drivers/net/e1000/e1000_hw.c b/drivers/net/e1000/e1000_hw.c
> index 9be4469..7c75b8b 100644
> --- a/drivers/net/e1000/e1000_hw.c
> +++ b/drivers/net/e1000/e1000_hw.c
> @@ -1030,6 +1030,13 @@ e1000_init_hw(struct e1000_hw *hw)
>          break;
>      }
>  
> +#if defined(CONFIG_PPC_PSERIES)
> +    if (hw->mac_type == e1000_82571) {
> +        uint32_t gcr = E1000_READ_REG(hw, GCR);
> +        gcr |= E1000_GCR_DISABLE_TIMEOUT_MECHANISM;
> +        E1000_WRITE_REG(hw, GCR, gcr);
> +    }
> +#endif 
>  
>      if (hw->mac_type == e1000_82573) {
>          uint32_t gcr = E1000_READ_REG(hw, GCR);
> diff --git a/drivers/net/e1000/e1000_hw.h b/drivers/net/e1000/e1000_hw.h
> index bd000b8..71f4e2d 100644
> --- a/drivers/net/e1000/e1000_hw.h
> +++ b/drivers/net/e1000/e1000_hw.h
> @@ -2211,6 +2211,7 @@ struct e1000_host_command_info {
>  #define PCI_EX_82566_SNOOP_ALL PCI_EX_NO_SNOOP_ALL
>  
>  #define E1000_GCR_L1_ACT_WITHOUT_L0S_RX 0x08000000
> +#define E1000_GCR_DISABLE_TIMEOUT_MECHANISM 0x80000000
>  /* Function Active and Power State to MNG */
>  #define E1000_FACTPS_FUNC0_POWER_STATE_MASK         0x00000003
>  #define E1000_FACTPS_LAN0_VALID                     0x00000004

Re: [PATCH] e1000: repost work around 82571 completion timout on pseries HW

[Andy Gospodarek]

On Tue, Jul 24, 2007 at 08:31:08AM -0700, Kok, Auke wrote:
> Andy Gospodarek wrote:
> >There seems to have been some discussion about a patch like this in the
> >past but I still haven't noticed any platforms fixes or noticed that
> >this got included, so I'd like to propose this modified version.
> >
> >Thoughts?
> 
> I've written a completely new version for IBM based on the response that 
> implements this as a pci quirk (to tag the chipset, not the device).
> 
> However IBM has spotted an issue due to firmware hiding the id for linux of 
> the bridge involved (on pseries), and they're working on an alternative to 
> fix that.
> 
> So, the patch will look different, and I'll post it as soon as we have a 
> version for both pseries and xseries that works.
> 
> Auke
> 

Sounds great.  I just wanted to make sure this didn't slip through the
cracks.