Re: [PATCH V2] PCI/ASPM: Save/restore L1SS Capability for suspend/resume

From: Vidya Sagar
Date: Tue Aug 30 2022 - 10:03:04 EST

Next message: Alan Stern: "Re: [PATCH] usb: Improves USB2.0 write performance"
Previous message: Jens Axboe: "Re: [RFC PATCH] blk-mq: change bio_set_ioprio to inline"
In reply to: Lukasz Majczak: "Re: [PATCH V2] PCI/ASPM: Save/restore L1SS Capability for suspend/resume"
Messages sorted by: [ date ] [ thread ] [ subject ] [ author ]

On 8/30/2022 4:45 PM, Lukasz Majczak wrote:

External email: Use caution opening links or attachments

pt., 26 sie 2022 o 15:00 Vidya Sagar <vidyas@xxxxxxxxxx> napisał(a):

On 8/23/2022 8:25 PM, Kai-Heng Feng wrote:

External email: Use caution opening links or attachments

Hi Vidya,

On Tue, Aug 9, 2022 at 12:17 AM Vidya Sagar <vidyas@xxxxxxxxxx> wrote:

Thanks Lukasz for the update.
I think confirms that there is no issue with the patch as such.
Bjorn, could you please define the next step for this patch?

I think the L1SS cap went away _after_ L1SS registers are restored,
since your patch already check the cap before doing any write:
+ aspm_l1ss = pci_find_ext_capability(dev, PCI_EXT_CAP_ID_L1SS);
+ if (!aspm_l1ss)
+ return;

That means it's more likely to be caused by the following change:
+ pci_write_config_dword(dev, aspm_l1ss + PCI_L1SS_CTL2, *cap++);
+ pci_write_config_dword(dev, aspm_l1ss + PCI_L1SS_CTL1, *cap++);

So is it possible to clear PCI_L1SS_CTL1 before setting PCI_L1SS_CTL2,
like what aspm_calc_l1ss_info() does?

I posted a new patch
https://patchwork.kernel.org/project/linux-pci/patch/20220826125526.28859-1-vidyas@xxxxxxxxxx/
keeping L1.2 disabled while restoring the rest of the fields in
Control-1 register and restoring the L1.2 enable bits later. Could you
please try this new patch on your setup and update your observations?

Thanks & Regards,
Vidya Sagar

Kai-Heng

Thanks,
Vidya Sagar

On 8/8/2022 7:37 PM, Lukasz Majczak wrote:

External email: Use caution opening links or attachments

śr., 3 sie 2022 o 14:55 Vidya Sagar <vidyas@xxxxxxxxxx> napisał(a):

Thanks Lukasz for the logs.
I still that the L1SS capability in the root port (00:14.0) disappeared
after resume.
I still don't understand how this patch can make the capability register
itself disappear. Honestly, I still see this as a HW issue.
Bjorn, could you please throw some light on this?

Thanks,
Vidya Sagar

On 8/3/2022 5:34 PM, Lukasz Majczak wrote:

External email: Use caution opening links or attachments

pt., 29 lip 2022 o 16:36 Vidya Sagar <vidyas@xxxxxxxxxx> napisał(a):

Hi Lukasz,
Thanks for sharing your observations.

Could you please also share the output of 'sudo lspci -vvvv' before and
after suspend-resume cycle with the latest linux-next?
Do we still see the L1SS capabilities getting disappeared post resume?

Thanks,
Vidya Sagar

On 7/29/2022 3:09 PM, Lukasz Majczak wrote:

External email: Use caution opening links or attachments

wt., 26 lip 2022 o 09:20 Lukasz Majczak <lma@xxxxxxxxxxxx> napisał(a):

wt., 26 lip 2022 o 00:51 Rajat Jain <rajatja@xxxxxxxxxx> napisał(a):

Hello,

On Sat, Jul 23, 2022 at 10:03 AM Vidya Sagar <vidyas@xxxxxxxxxx> wrote:

Agree with Bjorn's observations.
The fact that the L1SS capability registers themselves disappeared in
the root port post resume indicates that there seems to be something
wrong with the BIOS itself.
Could you please check from that perspective?

ChromeOS Intel platforms use S0ix (suspend-to-idle) for suspend. This
is a shallower sleep state that preserves more state than, for e.g. S3
(suspend-to-RAM). When we use S0ix, then BIOS does not come in picture
at all. i.e. after the kernel runs its suspend routines, it just puts
the CPU into S0ix state. So I do not think there is a BIOS angle to
this.

Thanks,
Vidya Sagar

On 7/22/2022 11:12 PM, Bjorn Helgaas wrote:

External email: Use caution opening links or attachments

On Fri, Jul 22, 2022 at 11:41:14AM +0200, Lukasz Majczak wrote:

pt., 22 lip 2022 o 09:31 Kai-Heng Feng <kai.heng.feng@xxxxxxxxxxxxx> napisał(a):

On Fri, Jul 15, 2022 at 6:38 PM Ben Chuang <benchuanggli@xxxxxxxxx> wrote:

On Tue, Jul 5, 2022 at 2:00 PM Vidya Sagar <vidyas@xxxxxxxxxx> wrote:

Previously ASPM L1 Substates control registers (CTL1 and CTL2) weren't
saved and restored during suspend/resume leading to L1 Substates
configuration being lost post-resume.

Save the L1 Substates control registers so that the configuration is
retained post-resume.

Signed-off-by: Vidya Sagar <vidyas@xxxxxxxxxx>
Tested-by: Abhishek Sahu <abhsahu@xxxxxxxxxx>

Hi Vidya,

I tested this patch on kernel v5.19-rc6.
The test device is GL9755 card reader controller on Intel i5-10210U RVP.
This patch can restore L1SS after suspend/resume.

The test results are as follows:

After Boot:
#lspci -d 17a0:9755 -vvv | grep -A5 "L1 PM Substates"
Capabilities: [110 v1] L1 PM Substates
L1SubCap: PCI-PM_L1.2+ PCI-PM_L1.1+ ASPM_L1.2+
ASPM_L1.1+ L1_PM_Substates+
PortCommonModeRestoreTime=255us
PortTPowerOnTime=3100us
L1SubCtl1: PCI-PM_L1.2+ PCI-PM_L1.1+ ASPM_L1.2+ ASPM_L1.1+
T_CommonMode=0us LTR1.2_Threshold=3145728ns
L1SubCtl2: T_PwrOn=3100us

After suspend/resume without this patch.
#lspci -d 17a0:9755 -vvv | grep -A5 "L1 PM Substates"
Capabilities: [110 v1] L1 PM Substates
L1SubCap: PCI-PM_L1.2+ PCI-PM_L1.1+ ASPM_L1.2+
ASPM_L1.1+ L1_PM_Substates+
PortCommonModeRestoreTime=255us
PortTPowerOnTime=3100us
L1SubCtl1: PCI-PM_L1.2- PCI-PM_L1.1- ASPM_L1.2- ASPM_L1.1-
T_CommonMode=0us LTR1.2_Threshold=0ns
L1SubCtl2: T_PwrOn=10us

After suspend/resume with this patch.
#lspci -d 17a0:9755 -vvv | grep -A5 "L1 PM Substates"
Capabilities: [110 v1] L1 PM Substates
L1SubCap: PCI-PM_L1.2+ PCI-PM_L1.1+ ASPM_L1.2+
ASPM_L1.1+ L1_PM_Substates+
PortCommonModeRestoreTime=255us
PortTPowerOnTime=3100us
L1SubCtl1: PCI-PM_L1.2+ PCI-PM_L1.1+ ASPM_L1.2+ ASPM_L1.1+
T_CommonMode=0us LTR1.2_Threshold=3145728ns
L1SubCtl2: T_PwrOn=3100us

Tested-by: Ben Chuang <benchuanggli@xxxxxxxxx>

Forgot to add mine:
Tested-by: Kai-Heng Feng <kai.heng.feng@xxxxxxxxxxxxx>

Best regards,
Ben Chuang

---
Hi,
Kenneth R. Crudup <kenny@xxxxxxxxx>, Could you please verify this patch
on your laptop (Dell XPS 13) one last time?
IMHO, the regression observed on your laptop with an old version of the patch
could be due to a buggy old version BIOS in the laptop.

Thanks,
Vidya Sagar

drivers/pci/pci.c | 7 +++++++
drivers/pci/pci.h | 4 ++++
drivers/pci/pcie/aspm.c | 44 +++++++++++++++++++++++++++++++++++++++++
3 files changed, 55 insertions(+)

diff --git a/drivers/pci/pci.c b/drivers/pci/pci.c
index cfaf40a540a8..aca05880aaa3 100644
--- a/drivers/pci/pci.c
+++ b/drivers/pci/pci.c
@@ -1667,6 +1667,7 @@ int pci_save_state(struct pci_dev *dev)
return i;

pci_save_ltr_state(dev);
+ pci_save_aspm_l1ss_state(dev);
pci_save_dpc_state(dev);
pci_save_aer_state(dev);
pci_save_ptm_state(dev);
@@ -1773,6 +1774,7 @@ void pci_restore_state(struct pci_dev *dev)
* LTR itself (in the PCIe capability).
*/
pci_restore_ltr_state(dev);
+ pci_restore_aspm_l1ss_state(dev);

pci_restore_pcie_state(dev);
pci_restore_pasid_state(dev);
@@ -3489,6 +3491,11 @@ void pci_allocate_cap_save_buffers(struct pci_dev *dev)
if (error)
pci_err(dev, "unable to allocate suspend buffer for LTR\n");

+ error = pci_add_ext_cap_save_buffer(dev, PCI_EXT_CAP_ID_L1SS,
+ 2 * sizeof(u32));
+ if (error)
+ pci_err(dev, "unable to allocate suspend buffer for ASPM-L1SS\n");
+
pci_allocate_vc_save_buffers(dev);
}

diff --git a/drivers/pci/pci.h b/drivers/pci/pci.h
index e10cdec6c56e..92d8c92662a4 100644
--- a/drivers/pci/pci.h
+++ b/drivers/pci/pci.h
@@ -562,11 +562,15 @@ void pcie_aspm_init_link_state(struct pci_dev *pdev);
void pcie_aspm_exit_link_state(struct pci_dev *pdev);
void pcie_aspm_pm_state_change(struct pci_dev *pdev);
void pcie_aspm_powersave_config_link(struct pci_dev *pdev);
+void pci_save_aspm_l1ss_state(struct pci_dev *dev);
+void pci_restore_aspm_l1ss_state(struct pci_dev *dev);
#else
static inline void pcie_aspm_init_link_state(struct pci_dev *pdev) { }
static inline void pcie_aspm_exit_link_state(struct pci_dev *pdev) { }
static inline void pcie_aspm_pm_state_change(struct pci_dev *pdev) { }
static inline void pcie_aspm_powersave_config_link(struct pci_dev *pdev) { }
+static inline void pci_save_aspm_l1ss_state(struct pci_dev *dev) { }
+static inline void pci_restore_aspm_l1ss_state(struct pci_dev *dev) { }
#endif

#ifdef CONFIG_PCIE_ECRC
diff --git a/drivers/pci/pcie/aspm.c b/drivers/pci/pcie/aspm.c
index a96b7424c9bc..2c29fdd20059 100644
--- a/drivers/pci/pcie/aspm.c
+++ b/drivers/pci/pcie/aspm.c
@@ -726,6 +726,50 @@ static void pcie_config_aspm_l1ss(struct pcie_link_state *link, u32 state)
PCI_L1SS_CTL1_L1SS_MASK, val);
}

+void pci_save_aspm_l1ss_state(struct pci_dev *dev)
+{
+ int aspm_l1ss;
+ struct pci_cap_saved_state *save_state;
+ u32 *cap;
+
+ if (!pci_is_pcie(dev))
+ return;
+
+ aspm_l1ss = pci_find_ext_capability(dev, PCI_EXT_CAP_ID_L1SS);
+ if (!aspm_l1ss)
+ return;
+
+ save_state = pci_find_saved_ext_cap(dev, PCI_EXT_CAP_ID_L1SS);
+ if (!save_state)
+ return;
+
+ cap = (u32 *)&save_state->cap.data[0];
+ pci_read_config_dword(dev, aspm_l1ss + PCI_L1SS_CTL2, cap++);
+ pci_read_config_dword(dev, aspm_l1ss + PCI_L1SS_CTL1, cap++);
+}
+
+void pci_restore_aspm_l1ss_state(struct pci_dev *dev)
+{
+ int aspm_l1ss;
+ struct pci_cap_saved_state *save_state;
+ u32 *cap;
+
+ if (!pci_is_pcie(dev))
+ return;
+
+ aspm_l1ss = pci_find_ext_capability(dev, PCI_EXT_CAP_ID_L1SS);
+ if (!aspm_l1ss)
+ return;
+
+ save_state = pci_find_saved_ext_cap(dev, PCI_EXT_CAP_ID_L1SS);
+ if (!save_state)
+ return;
+
+ cap = (u32 *)&save_state->cap.data[0];
+ pci_write_config_dword(dev, aspm_l1ss + PCI_L1SS_CTL2, *cap++);
+ pci_write_config_dword(dev, aspm_l1ss + PCI_L1SS_CTL1, *cap++);
+}
+
static void pcie_config_aspm_dev(struct pci_dev *pdev, u32 val)
{
pcie_capability_clear_and_set_word(pdev, PCI_EXP_LNKCTL,
--
2.17.1

Hi,

With this patch (and also mentioned
https://lore.kernel.org/all/20220509073639.2048236-1-kai.heng.feng@xxxxxxxxxxxxx/)
applied on 5.10 (chromeos-5.10) I am observing problems after
suspend/resume with my WiFi card - it looks like whole communication
via PCI fails. Attaching logs (dmesg, lspci -vvv before suspend/resume
and after) https://gist.github.com/semihalf-majczak-lukasz/fb36dfa2eff22911109dfb91ab0fc0e3

I played a little bit with this code and it looks like the
pci_write_config_dword() to the PCI_L1SS_CTL1 breaks it (don't know
why, not a PCI expert).

Thanks a lot for testing this! I'm not quite sure what to make of the
results since v5.10 is fairly old (Dec 2020) and I don't know what
other changes are in chromeos-5.10.

Lukasz: I assume you are running this on Atlas and are seeing this bug
when uprev'ving it to 5.10 kernel. Can you please try it on a newer
Intel platform that have the latest upstream kernel running already
and see if this can be reproduced there too?
Note that the wifi PCI device is different on newer Intel platforms,
but platform design is similar enough that I suspect we should see
similar bug on those too. The other option is to try the latest
ustream kernel on Atlas. Perhaps if we just care about wifi (and
ignore bringing up the graphics stack and GUI), it may come up
sufficiently enough to try this patch?

Thanks,

Rajat

Random observations, no analysis below. This from your dmesg
certainly looks like PCI reads failing and returning ~0:

Timeout waiting for hardware access (CSR_GP_CNTRL 0xffffffff)
iwlwifi 0000:01:00.0: 00000000: ffffffff ffffffff ffffffff ffffffff ffffffff ffffffff ffffffff ffffffff
iwlwifi 0000:01:00.0: Device gone - attempting removal
Hardware became unavailable upon resume. This could be a software issue prior to suspend or a hardware issue.

And then we re-enumerate 01:00.0 and it looks like it may have been
reset (BAR is 0):

pci 0000:01:00.0: [8086:095a] type 00 class 0x028000
pci 0000:01:00.0: reg 0x10: [mem 0x00000000-0x00001fff 64bit]

lspci diffs from before/after suspend:

00:14.0 PCI bridge: Intel Corporation Celeron N3350/Pentium N4200/Atom E3900 Series PCI Express Port B #1 (rev fb) (prog-if 00 [Normal decode])
Bus: primary=00, secondary=01, subordinate=01, sec-latency=64
- DevSta: CorrErr- NonFatalErr+ FatalErr- UnsupReq+ AuxPwr+ TransPend-
+ DevSta: CorrErr+ NonFatalErr- FatalErr- UnsupReq- AuxPwr+ TransPend-
- LnkCtl: ASPM L1 Enabled; RCB 64 bytes, Disabled- CommClk+
+ LnkCtl: ASPM Disabled; RCB 64 bytes, Disabled- CommClk+
- LnkSta2: Current De-emphasis Level: -6dB, EqualizationComplete- EqualizationPhase1-
+ LnkSta2: Current De-emphasis Level: -3.5dB, EqualizationComplete- EqualizationPhase1-
- Capabilities: [150 v0] Null
- Capabilities: [200 v1] L1 PM Substates
- L1SubCap: PCI-PM_L1.2+ PCI-PM_L1.1+ ASPM_L1.2+ ASPM_L1.1+ L1_PM_Substates+
- PortCommonModeRestoreTime=40us PortTPowerOnTime=10us
- L1SubCtl1: PCI-PM_L1.2+ PCI-PM_L1.1+ ASPM_L1.2+ ASPM_L1.1+
- T_CommonMode=40us LTR1.2_Threshold=98304ns
- L1SubCtl2: T_PwrOn=60us

The DevSta differences might be BIOS bugs, probably not relevant.
Interesting that ASPM is disabled, maybe didn't get enabled after
re-enumerating 01:00.0? Strange that the L1 PM Substates capability
disappeared.

01:00.0 Network controller: Intel Corporation Wireless 7265 (rev 59)
LnkCtl: ASPM L1 Enabled; RCB 64 bytes, Disabled- CommClk+
- ExtSynch- ClockPM+ AutWidDis- BWInt- AutBWInt-
+ ExtSynch- ClockPM- AutWidDis- BWInt- AutBWInt-
Capabilities: [154 v1] L1 PM Substates
L1SubCap: PCI-PM_L1.2+ PCI-PM_L1.1+ ASPM_L1.2+ ASPM_L1.1+ L1_PM_Substates+
PortCommonModeRestoreTime=30us PortTPowerOnTime=60us
- L1SubCtl1: PCI-PM_L1.2+ PCI-PM_L1.1+ ASPM_L1.2+ ASPM_L1.1+
- T_CommonMode=0us LTR1.2_Threshold=98304ns
+ L1SubCtl1: PCI-PM_L1.2- PCI-PM_L1.1- ASPM_L1.2- ASPM_L1.1-
+ T_CommonMode=0us LTR1.2_Threshold=0ns

Dmesg claimed we reconfigured common clock config. Maybe ASPM didn't
get reinitialized after re-enumeration? Looks like we didn't restore
L1SubCtl1.

Bjorn

Hi,

Thank you all for the response and input! As Rajat mentioned I'm using
chromebook - but not Atlas (Amberlake) - in this case it is Babymega
(Apollolake) - I will try to load most recent kernel and give it a
try once again.

Best regards,
Lukasz

Hi,

I have applied this patch on top of v5.19-rc7 (chromeos) and I'm
still getting same results:
https://gist.github.com/semihalf-majczak-lukasz/4b716704c21a3758d6711b2030ea34b9

Best regards,
Lukasz

Hi Vidya,

Sorry for the long delay, I have retested your patch on top of
linux-next/master (next-20220802) - the results for my device remain
the same.
Here are the logs (lspci -vvv before suspend, lspci -vvv after resume and dmesg)
https://gist.github.com/semihalf-majczak-lukasz/c7bfd811359f23278034056a8002b3ef
Let me know if you need any more logs and/or tests.

Best regards,
Lukasz

Hi Vidya,

After your last email, I've re-tested my setup and (without your
patch) the capability register also disappears - so it looks there is
- in fact - some problem in my setup and your patch just brings it to
the top as after resume tries to write to a register that is no longer
present. I'm very sorry for the confusion here and I've not notice
that at the very beginning.

Best regards,
Lukasz

Hi Vidya,

For me (on Apollolake devices) the results remain the same, but as
I've mentioned earlier - it looks very much related exactly to the
Apollolake and is not directly related to your patch (e.g. I'm losing
L1SS capabilities even without your patch).
As a counter example, I don't observe any issues with your patach
(v3) on Amberlake devices - lspci -vvv before suspend and after resume
are exactly the same.

Thanks for the update Lukasz.
Anyway, i sent V3 fore review. Could you please review it and also test it on your platform?

Thanks,
Vidya Sagar

Best regards,
Lukasz

Next message: Alan Stern: "Re: [PATCH] usb: Improves USB2.0 write performance"
Previous message: Jens Axboe: "Re: [RFC PATCH] blk-mq: change bio_set_ioprio to inline"
In reply to: Lukasz Majczak: "Re: [PATCH V2] PCI/ASPM: Save/restore L1SS Capability for suspend/resume"
Messages sorted by: [ date ] [ thread ] [ subject ] [ author ]