Re: iwlwifi devices disappear after suspend on kernel 5.17

["Grumbach, Emmanuel" <emmanuel.grumbach@intel.com>]

On Tue, 2022-04-19 at 13:33 +0200, Toke Høiland-Jørgensen wrote:
> "Grumbach, Emmanuel" <emmanuel.grumbach@intel.com> writes:
> 
> > On Tue, 2022-04-19 at 09:41 +0300, Grumbach, Emmanuel wrote:
> > > Hi Toke,
> > > 
> > > On Tue, 2022-04-19 at 06:24 +0000, Coelho, Luciano wrote:
> > > > On Mon, 2022-04-18 at 13:36 +0200, Toke Høiland-Jørgensen wrote:
> > > > > Hi Luca
> > > > 
> > > > Hi Toke,
> > > > 
> > > > > I've started seeing issues with my iwlwifi interface going away after
> > > > > suspend. I get errors like these in dmesg:
> > > > > 
> > > > > [104393.142264] wlan0: deauthenticating from 4c:60:de:ea:b8:58 by local choice (Reason:
> > > > > 3=DEAUTH_LEAVING)
> > > > > [104393.347775] iwlmei 0000:00:16.0-13280904-7792-4fcb-a1aa-5e70cbb1e865: Couldn't get ACK
> > > > > from
> > > > > CSME on HOST_GOES_DOWN message
> > > > > [104393.347876] iwlmei 0000:00:16.0-13280904-7792-4fcb-a1aa-5e70cbb1e865: failed to send
> > > > > the
> > > > > SAP_ME_MSG_CHECK_SHARED_AREA message -19
> > > > > 
> > > > > And when the host comes back up, there is no connectivity. Restarting
> > > > > iwd fixes the problem.
> > > > > 
> > > > > This is on a 5.17.3 kernel (Arch Linux distribution kernel), and lspci
> > > > > says the WiFi device is an "Intel Corporation Wi-Fi 6 AX201".
> > > > > 
> > > > > Any ideas? :)
> > > > 
> > > > This seems to be related to iwlmei, so I added Emmanuel to the thread.
> > > > 
> > > 
> > > Can we have the full dmesg output?
> > > What NIC / platform do you have?
> > > Do you have AMT configured in the BIOS?
> > > Did you enable wireless operation in AMT?
> > 
> > Ah - this is AX201, ok, but I still need the platform and thefull dmesg :-)
> 
> It's a Lenovo ThinkPad X1 Carbon Gen 9, model 20XXS3HC26; AMT is set to
> "Disabled" in the BIOS. I rebooted and did a suspend/wake cycle after
> connecting to the WiFi network, which produced the dmesg below.

Thanks.
This is a Skylake platform and we didn't have a handshake with CSME which is not enabled. We are
missing a check before we shut down iwlmei.
Can you check this?

diff --git a/drivers/net/wireless/intel/iwlwifi/mei/main.c
b/drivers/net/wireless/intel/iwlwifi/mei/main.c
index 8d61044a6ed3..7b45382c813c 100644
--- a/drivers/net/wireless/intel/iwlwifi/mei/main.c
+++ b/drivers/net/wireless/intel/iwlwifi/mei/main.c
@@ -1923,29 +1923,33 @@ static void iwl_mei_remove(struct mei_cl_device *cldev)
 
        mutex_lock(&iwl_mei_mutex);
 
-       /*
-        * Tell CSME that we are going down so that it won't access the
-        * memory anymore, make sure this message goes through immediately.
-        */
-       mei->csa_throttled = false;
-       iwl_mei_send_sap_msg(mei->cldev,
-                            SAP_MSG_NOTIF_HOST_GOES_DOWN);
+       if (iwl_mei_is_connected())
+       {
+               /*
+                * Tell CSME that we are going down so that it won't access the
+                * memory anymore, make sure this message goes through immediately.
+                */
+               mei->csa_throttled = false;
+               iwl_mei_send_sap_msg(mei->cldev,
+                                    SAP_MSG_NOTIF_HOST_GOES_DOWN);
 
-       for (i = 0; i < SEND_SAP_MAX_WAIT_ITERATION; i++) {
-               if (!iwl_mei_host_to_me_data_pending(mei))
-                       break;
+               for (i = 0; i < SEND_SAP_MAX_WAIT_ITERATION; i++) {
+                       if (!iwl_mei_host_to_me_data_pending(mei))
+                               break;
 
-               msleep(5);
-       }
+                       msleep(5);
+               }
 
-       /*
-        * If we couldn't make sure that CSME saw the HOST_GOES_DOWN message,
-        * it means that it will probably keep reading memory that we are going
-        * to unmap and free, expect IOMMU error messages.
-        */
-       if (i == SEND_SAP_MAX_WAIT_ITERATION)
-               dev_err(&mei->cldev->dev,
-                       "Couldn't get ACK from CSME on HOST_GOES_DOWN message\n");
+               /*
+                * If we couldn't make sure that CSME saw the HOST_GOES_DOWN
+                * message, it means that it will probably keep reading memory
+                * that we are going to unmap and free, expect IOMMU error
+                * messages.
+                */
+               if (i == SEND_SAP_MAX_WAIT_ITERATION)
+                       dev_err(&mei->cldev->dev,
+                               "Couldn't get ACK from CSME on HOST_GOES_DOWN message\n");
+       }
 
        mutex_unlock(&iwl_mei_mutex);
 

Although I think it'll just fix the error print and nothing more than this