On Sunday, 17 of August 2008, Rafael J. Wysocki wrote:

On Sunday, 17 of August 2008, Yinghai Lu wrote:

quoted

On Sun, Aug 17, 2008 at 12:34 PM, Rafael J. Wysocki [off-list ref] wrote:

quoted

On Sunday, 17 of August 2008, Rafael J. Wysocki wrote:

quoted

On Sunday, 17 of August 2008, Yinghai Lu wrote:

quoted

On Sun, Aug 17, 2008 at 9:55 AM, Rafael J. Wysocki [off-list ref] wrote:

quoted

On Sunday, 17 of August 2008, Rafael J. Wysocki wrote:

quoted

On Sunday, 17 of August 2008, Yinghai Lu wrote:

quoted

after

| commit f735a2a1a4f2a0f5cd823ce323e82675990469e2
| Author: Tobias Diedrich [off-list ref]
| Date:   Sun May 18 15:02:37 2008 +0200
|
|    [netdrvr] forcedeth: setup wake-on-lan before shutting down
|
|    When hibernating in 'shutdown' mode, after saving the image the suspend hook
|    is not called again.
|    However, if the device is in promiscous mode, wake-on-lan will not work.
|    This adds a shutdown hook to setup wake-on-lan before the final shutdown.
|
|    Signed-off-by: Tobias Diedrich [off-list ref]
|    Signed-off-by: Jeff Garzik [off-list ref]

my servers with nvidia mcp55 nic doesn't work with msi in second kernel by kexec

after remove pci_set_power_state(, PCI_D3hot) that nic/msi will work again.

check with e1000 is using pci_choose_state in _shutdown.

This is wrong.

quoted

quoted

So change that pci_choose_state(pdev, ...), and it works.

Well, this doesn't look like a good solution to me, because you're putting
PMSG_SUSPEND in there, which is not correct for shutdown.  The right thing to
do would be to avoid changing the device power state if nv_shutdown() is
used for kexec or to rework the initialization of the adapter to handle the
case when it's initially in D3.

Does it help if you just remove the pci_set_power_state(pdev, PCI_D3hot)
altogether?

Ah, sorry.  I see it does.

Actually, I think you can use pci_prepare_to_sleep() instead of
pci_enable_wake() / pci_set_power_state() combo.  It wasn't designed for this
purpose, but should work nevertheless.

Can you please check if the appended patch works instead of your one?

Rafael

---
Fix the problem that boxes with NVidia MCP55 don't work with MSI
in a kexeced kernel.

Signed-off-by: Rafael J. Wysocki <redacted>
---
 drivers/net/forcedeth.c |    4 +---
 1 file changed, 1 insertion(+), 3 deletions(-)

Index: linux-2.6/drivers/net/forcedeth.c
===================================================================

--- linux-2.6.orig/drivers/net/forcedeth.c
+++ linux-2.6/drivers/net/forcedeth.c

@@ -5975,10 +5975,8 @@ static void nv_shutdown(struct pci_dev *
       if (netif_running(dev))
               nv_close(dev);

-       pci_enable_wake(pdev, PCI_D3hot, np->wolenabled);
-       pci_enable_wake(pdev, PCI_D3cold, np->wolenabled);
       pci_disable_device(pdev);
-       pci_set_power_state(pdev, PCI_D3hot);
+       pci_prepare_to_sleep(pdev);
 }
 #else
 #define nv_suspend NULL

your patch doesn't work...  it seems that silicon has problem with D3Hot

Okay, so perhaps it's better to do something like this:

---
 drivers/net/forcedeth.c |    8 +++++---
 1 file changed, 5 insertions(+), 3 deletions(-)

Index: linux-2.6/drivers/net/forcedeth.c
===================================================================

--- linux-2.6.orig/drivers/net/forcedeth.c
+++ linux-2.6/drivers/net/forcedeth.c

@@ -5975,10 +5975,12 @@ static void nv_shutdown(struct pci_dev *
      if (netif_running(dev))
              nv_close(dev);

-     pci_enable_wake(pdev, PCI_D3hot, np->wolenabled);
-     pci_enable_wake(pdev, PCI_D3cold, np->wolenabled);
      pci_disable_device(pdev);
-     pci_set_power_state(pdev, PCI_D3hot);
+     if (system_state == SYS_POWER_OFF) {

Sorry, it should be SYSTEM_POWER_OFF here.  Corrected patch is appended.

quoted

+             if (pci_enable_wake(pdev, PCI_D3cold, np->wolenabled))
+                     pci_enable_wake(pdev, PCI_D3hot, np->wolenabled);
+             pci_set_power_state(pdev, PCI_D3hot);
+     }
 }
 #else
 #define nv_suspend NULL
--

---
 drivers/net/forcedeth.c |    8 +++++---
 1 file changed, 5 insertions(+), 3 deletions(-)

Index: linux-2.6/drivers/net/forcedeth.c
===================================================================

--- linux-2.6.orig/drivers/net/forcedeth.c
+++ linux-2.6/drivers/net/forcedeth.c

@@ -5975,10 +5975,12 @@ static void nv_shutdown(struct pci_dev *
       if (netif_running(dev))
               nv_close(dev);

-       pci_enable_wake(pdev, PCI_D3hot, np->wolenabled);
-       pci_enable_wake(pdev, PCI_D3cold, np->wolenabled);
       pci_disable_device(pdev);
-       pci_set_power_state(pdev, PCI_D3hot);
+       if (system_state == SYSTEM_POWER_OFF) {
+               if (pci_enable_wake(pdev, PCI_D3cold, np->wolenabled))
+                       pci_enable_wake(pdev, PCI_D3hot, np->wolenabled);
+               pci_set_power_state(pdev, PCI_D3hot);
+       }
 }
 #else
 #define nv_suspend NULL

why not like e1000 to
pci_set_power_state(pdev, pci_choose_state(pdev, PMSG_SUSPEND));

pci_choose_state would check if platform support those state...

... but ->shutdown() only involves the platform in the
'system_state == SYSTEM_POWER_OFF' case!

In fact, using pci_choose_state() in ->shutdown() is a very convoluted way of
checking the 'system_state' value, and not a 100% reliable one. :-)

Namely, the fact that pci_choose_state() returns you D0 instead of D3hot for
'system_state != SYSTEM_POWER_OFF' is a pure coincidence (accidentally, there
is an ACPI handle for the device, but there need not be one) and you should
not rely on that in general.  Generally speaking, pci_choose_state() is not to
be used outside of the drivers' ->suspend() routines.

[Yes, this means e1000 will have to be fixed, as I said before.]

Apart from this, pci_choose_state() is broken and will shortly be deprecated.
Moreover, in future all direct references to PMSG_SUSPEND from the drivers
will be removed.

Does the last patch work for you BTW?

Rafael