| lists.openwall.net | lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC | |
|
Open Source and information security mailing list archives
| ||
|
Message-ID: <otdgyzdymraa3f33vyb445kmssi3mqf5z2mw7w5pib4q4sb7vz@qbrzvrojqji3>
Date: Fri, 18 Jul 2025 15:50:02 +0530
From: Manivannan Sadhasivam <mani@...nel.org>
To: Baochen Qiang <baochen.qiang@....qualcomm.com>
Cc: manivannan.sadhasivam@....qualcomm.com,
Jeff Johnson <jjohnson@...nel.org>, Lorenzo Pieralisi <lpieralisi@...nel.org>,
Krzysztof Wilczyński <kwilczynski@...nel.org>, Rob Herring <robh@...nel.org>,
Bjorn Helgaas <bhelgaas@...gle.com>, Nirmal Patel <nirmal.patel@...ux.intel.com>,
Jonathan Derrick <jonathan.derrick@...ux.dev>, linux-wireless@...r.kernel.org, linux-kernel@...r.kernel.org,
ath12k@...ts.infradead.org, ath11k@...ts.infradead.org, ath10k@...ts.infradead.org,
Bjorn Helgaas <helgaas@...nel.org>, ilpo.jarvinen@...ux.intel.com, linux-arm-msm@...r.kernel.org,
linux-pci@...r.kernel.org, Krishna Chaitanya Chundru <krishna.chundru@....qualcomm.com>,
Qiang Yu <qiang.yu@....qualcomm.com>
Subject: Re: [PATCH 4/6] wifi: ath12k: Use pci_{enable/disable}_link_state()
APIs to enable/disable ASPM states
On Fri, Jul 18, 2025 at 01:27:27PM GMT, Manivannan Sadhasivam wrote:
> On Fri, Jul 18, 2025 at 10:05:02AM GMT, Baochen Qiang wrote:
> >
> >
> > On 7/17/2025 7:29 PM, Manivannan Sadhasivam wrote:
> > > On Thu, Jul 17, 2025 at 06:46:12PM GMT, Baochen Qiang wrote:
> > >>
> > >>
> > >> On 7/17/2025 6:31 PM, Manivannan Sadhasivam wrote:
> > >>> On Thu, Jul 17, 2025 at 05:24:13PM GMT, Baochen Qiang wrote:
> > >>>
> > >>> [...]
> > >>>
> > >>>>> @@ -16,6 +16,8 @@
> > >>>>> #include "mhi.h"
> > >>>>> #include "debug.h"
> > >>>>>
> > >>>>> +#include "../ath.h"
> > >>>>> +
> > >>>>> #define ATH12K_PCI_BAR_NUM 0
> > >>>>> #define ATH12K_PCI_DMA_MASK 36
> > >>>>>
> > >>>>> @@ -928,8 +930,7 @@ static void ath12k_pci_aspm_disable(struct ath12k_pci *ab_pci)
> > >>>>> u16_get_bits(ab_pci->link_ctl, PCI_EXP_LNKCTL_ASPM_L1));
> > >>>>>
> > >>>>> /* disable L0s and L1 */
> > >>>>> - pcie_capability_clear_word(ab_pci->pdev, PCI_EXP_LNKCTL,
> > >>>>> - PCI_EXP_LNKCTL_ASPMC);
> > >>>>> + pci_disable_link_state(ab_pci->pdev, PCIE_LINK_STATE_L0S | PCIE_LINK_STATE_L1);
> > >>>>
> > >>>> Not always, but sometimes seems the 'disable' does not work:
> > >>>>
> > >>>> [ 279.920507] ath12k_pci_power_up 1475: link_ctl 0x43 //before disable
> > >>>> [ 279.920539] ath12k_pci_power_up 1482: link_ctl 0x43 //after disable
> > >>>>
> > >>>>
> > >>>>>
> > >>>>> set_bit(ATH12K_PCI_ASPM_RESTORE, &ab_pci->flags);
> > >>>>> }
> > >>>>> @@ -958,10 +959,7 @@ static void ath12k_pci_aspm_restore(struct ath12k_pci *ab_pci)
> > >>>>> {
> > >>>>> if (ab_pci->ab->hw_params->supports_aspm &&
> > >>>>> test_and_clear_bit(ATH12K_PCI_ASPM_RESTORE, &ab_pci->flags))
> > >>>>> - pcie_capability_clear_and_set_word(ab_pci->pdev, PCI_EXP_LNKCTL,
> > >>>>> - PCI_EXP_LNKCTL_ASPMC,
> > >>>>> - ab_pci->link_ctl &
> > >>>>> - PCI_EXP_LNKCTL_ASPMC);
> > >>>>> + pci_enable_link_state(ab_pci->pdev, ath_pci_aspm_state(ab_pci->link_ctl));
> > >>>>
> > >>>> always, the 'enable' is not working:
> > >>>>
> > >>>> [ 280.561762] ath12k_pci_start 1180: link_ctl 0x43 //before restore
> > >>>> [ 280.561809] ath12k_pci_start 1185: link_ctl 0x42 //after restore
> > >>>>
> > >>>
> > >>> Interesting! I applied your diff and I never see this issue so far (across 10+
> > >>> reboots):
> > >>
> > >> I was not testing reboot. Here is what I am doing:
> > >>
> > >> step1: rmmod ath12k
> > >> step2: force LinkCtrl using setpci (make sure it is 0x43, which seems more likely to see
> > >> the issue)
> > >>
> > >> sudo setpci -s 02:00.0 0x80.B=0x43
> > >>
> > >> step3: insmod ath12k and check linkctrl
> > >>
> > >
> > > So I did the same and got:
> > >
> > > [ 3283.363569] ath12k_pci_power_up 1475: link_ctl 0x43
> > > [ 3283.363769] ath12k_pci_power_up 1480: link_ctl 0x40
> > > [ 3284.007661] ath12k_pci_start 1180: link_ctl 0x40
> > > [ 3284.007826] ath12k_pci_start 1185: link_ctl 0x42
> > >
> > > My host machine is Qcom based Thinkpad T14s and it doesn't support L0s. So
> > > that's why the lnkctl value once enabled becomes 0x42. This is exactly the
> > > reason why the drivers should not muck around LNKCTL register manually.
> >
> > Thanks, then the 0x43 -> 0x40 -> 0x40 -> 0x42 sequence should not be a concern. But still
> > the random 0x43 -> 0x43 -> 0x43 -> 0x42 sequence seems problematic.
> >
> > How many iterations have you done with above steps? From my side it seems random so better
> > to do some stress test.
> >
>
> So I ran the modprobe for about 50 times on the Intel NUC that has QCA6390, but
> didn't spot the disparity. This is the script I used:
>
> for i in {1..50} ;do echo "Loop $i"; sudo setpci -s 01:00.0 0x80.B=0x43;\
> sudo modprobe -r ath11k_pci; sleep 1; sudo modprobe ath11k_pci; sleep 1;done
>
> And I always got:
>
> [ 5862.388083] ath11k_pci_aspm_disable: 609 lnkctrl: 0x43
> [ 5862.388124] ath11k_pci_aspm_disable: 614 lnkctrl: 0x40
> [ 5862.876291] ath11k_pci_start: 880 lnkctrl: 0x40
> [ 5862.876346] ath11k_pci_start: 886 lnkctrl: 0x42
>
> Also no AER messages. TBH, I'm not sure how you were able to see the random
> issues with these APIs. That looks like a race, which is scary.
>
> I do not want to ignore your scenario, but would like to reproduce and get to
> the bottom of it.
>
I synced with Baochen internally and able to repro the issue. Ths issue is due
to hand modifying the LNKCTL register from userspace. The PCI core maintains
the ASPM state internally and uses it to change the state when the
pci_{enable/disable}_link_state*() APIs are called.
So if the userspace or a client driver modifies the LNKCTL register manually, it
makes the PCI cached ASPM states invalid. So while this series fixes the driver
from doing that, nothing prevents userspace from doing so using 'setpci' and
other tools. Userspace should only use sysfs attributes to change the state and
avoid modifying the PCI registers when the PCI core is controlling the device.
So this is the reason behind the errantic behavior of the API and it is not due
to the issue with the API or the PCI core.
- Mani
--
மணிவண்ணன் சதாசிவம்
Powered by blists - more mailing lists