Re: [PATCH v2] e1000e: Add a delay to let ME unconfigure s0ix when DPG_EXIT_DONE is already flagged

On 11/2/2021 05:27, Kai-Heng Feng wrote:

On Fri, Oct 29, 2021 at 5:14 PM Sasha Neftin [off-list ref] wrote:

quoted

On 10/27/2021 01:50, Kai-Heng Feng wrote:

quoted

On Tue, Oct 26, 2021 at 4:48 PM Sasha Neftin [off-list ref] wrote:

quoted

On 10/26/2021 09:51, Kai-Heng Feng wrote:

quoted

On some ADL platforms, DPG_EXIT_DONE is always flagged so e1000e resume
polling logic doesn't wait until ME really unconfigures s0ix.

So check DPG_EXIT_DONE before issuing EXIT_DPG, and if it's already
flagged, wait for 1 second to let ME unconfigure s0ix.

Fixes: 3e55d231716e ("e1000e: Add handshake with the CSME to support S0ix")
Bugzilla: https://bugzilla.kernel.org/show_bug.cgi?id=214821
Signed-off-by: Kai-Heng Feng <redacted>
---
v2:
    Add missing "Fixes:" tag

    drivers/net/ethernet/intel/e1000e/netdev.c | 7 +++++++
    1 file changed, 7 insertions(+)

diff --git a/drivers/net/ethernet/intel/e1000e/netdev.c b/drivers/net/ethernet/intel/e1000e/netdev.c
index 44e2dc8328a22..cd81ba00a6bc9 100644
--- a/drivers/net/ethernet/intel/e1000e/netdev.c
+++ b/drivers/net/ethernet/intel/e1000e/netdev.c

@@ -6493,14 +6493,21 @@ static void e1000e_s0ix_exit_flow(struct e1000_adapter *adapter)
        u32 mac_data;
        u16 phy_data;
        u32 i = 0;
+     bool dpg_exit_done;

        if (er32(FWSM) & E1000_ICH_FWSM_FW_VALID) {
+             dpg_exit_done = er32(EXFWSM) & E1000_EXFWSM_DPG_EXIT_DONE;
                /* Request ME unconfigure the device from S0ix */
                mac_data = er32(H2ME);
                mac_data &= ~E1000_H2ME_START_DPG;
                mac_data |= E1000_H2ME_EXIT_DPG;
                ew32(H2ME, mac_data);

+             if (dpg_exit_done) {
+                     e_warn("DPG_EXIT_DONE is already flagged. This is a firmware bug\n");
+                     msleep(1000);
+             }

Thanks for working on the enablement.
The delay approach is fragile. We need to work with CSME folks to
understand why _DPG_EXIT_DONE indication is wrong on some ADL platforms.
Could you provide CSME/BIOS version? dmidecode -t 0 and cat
/sys/class/mei/mei0/fw_ver

$ sudo dmidecode -t 0
# dmidecode 3.2
Getting SMBIOS data from sysfs.
SMBIOS 3.4 present.
# SMBIOS implementations newer than version 3.2.0 are not
# fully supported by this version of dmidecode.

Handle 0x0001, DMI type 0, 26 bytes
BIOS Information
          Vendor: Dell Inc.
          Version: 0.12.68
          Release Date: 10/01/2021
          ROM Size: 48 MB
          Characteristics:
                  PCI is supported
                  PNP is supported
                  BIOS is upgradeable
                  BIOS shadowing is allowed
                  Boot from CD is supported
                  Selectable boot is supported
                  EDD is supported
                  Print screen service is supported (int 5h)
                  8042 keyboard services are supported (int 9h)
                  Serial services are supported (int 14h)
                  Printer services are supported (int 17h)
                  ACPI is supported
                  USB legacy is supported
                  BIOS boot specification is supported
                  Function key-initiated network boot is supported
                  Targeted content distribution is supported
                  UEFI is supported
          BIOS Revision: 0.12

$ cat /sys/class/mei/mei0/fw_ver
0:16.0.15.1518
0:16.0.15.1518
0:16.0.15.1518

Thank you Kai-Heng. The _DPG_EXIT_DONE bit indication comes from the
EXFWSM register controlled by the CSME. We have only read access.  I
realized that this indication was set to 1 even before our request to
unconfigure the s0ix settings from the CSME. I am wondering. Does after
a ~ 1s delay (or less, or more) _DPG_EXIT_DONE indication eventually
change by CSME to 0? (is it consistently)

Never. It's consistently being 1.

no. On my TGL platform is cleared by CSME:
[Sun Oct 31 08:54:40 2021] s0ix exit: EXFWSM register: 0x00000000
[Sun Oct 31 08:54:40 2021] s0ix exit (right after sent H2ME): EXFWSM 
register: 0x00000000
[Sun Oct 31 08:54:40 2021] s0ix exit(after polling): EXFWSM register: 
0x00000001
[Sun Oct 31 08:54:40 2021] e1000e 0000:00:1f.6 enp0s31f6: DPG_EXIT_DONE 
cleared after 130 msec

Right now we are seeing the same issue on TGL, so I wonder if it's
better to just revert the CSME series?

no. We need to investigate it and understand what is CSME state we hit. 
Meanwhile few options:
1. use privilege flags to disable s0ix flow for problematic system 
(power will increased)
ethtool --set-priv-flags enp0s31f6 s0ix-enabled off
ethtool --show-priv-flags enp0s31f6
Private flags for enp0s31f6:
s0ix-enabled: off
2. delay as you suggested - less preferable I though
3. I would like to suggest (need to check it) in case the DPG_EXIT_DONE 
is 1 (and polling will be exit immediately) - let's perform enforce 
settings to the CSME, before write request to CSME unconfigure the 
device from s0ix :

if (er32(EXFWSM) & E1000_EXFWSM_DPG_EXIT_DONE)
	mac_data |= E1000_H2ME_ENFORCE_SETTINGS;

I will update Bugzilla: 
https://bugzilla.kernel.org/show_bug.cgi?id=214821 with this information.

I also will need some another information regards SMB state in this case.

Kai-Heng

quoted

quoted

quoted

quoted

                /* Poll up to 2.5 seconds for ME to unconfigure DPG.
                 * If this takes more than 1 second, show a warning indicating a
                 * firmware bug