linux-kernel - Re: [PATCH v2] PCI: Fix up L1SS capability for Intel Apollolake PCIe bridge

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

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Message-ID: <20221215151604.GA107336@bhelgaas>
Date:   Thu, 15 Dec 2022 09:16:04 -0600
From:   Bjorn Helgaas <helgaas@...nel.org>
To:     Ron Lee <ron.lee.intel@...il.com>
Cc:     bhelgaas@...gle.com, linux-kernel@...r.kernel.org,
        linux-pci@...r.kernel.org, lmajczak@...gle.com, rajatja@...gle.com,
        Ron Lee <ron.lee@...el.com>
Subject: Re: [PATCH v2] PCI: Fix up L1SS capability for Intel Apollolake PCIe
 bridge

On Thu, Dec 15, 2022 at 05:13:57PM +0800, Ron Lee wrote:
> On Google Coral and Reef family chromebooks, the PCIe bridge lost its
> L1 PM Substates capability after resumed from D3cold, and identify that
> the pointer to the this capability and capapability header are missing
> from the capability list.

s/chromebooks/Chromebooks/
s/to the this/this/
s/capapability/capability/

This should say what problem we're solving.  I assume some devices
used L1 PM Substates before suspend, but after resume they do not, so
the user-visible effect is that battery life is worse after resume.

> 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+ ...
>                   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

I'm not sure what this snippet is telling me.  Based on the patch, I
guess before suspend, lspci would show:

  Capabilities: [150 v0] Null
  Capabilities: [200 v1] L1 PM Substates
  Capabilities: [220] <some other valid capability?>

but after resume, you see only:

  Capabilities: [150 v0] Null

Right?

> This patch fix up the header and the pointer to the L1SS capability
> after resuming from D3Cold.

The main problem here is that this patch covers up an issue without
saying what the root cause is.  Presumably this is a firmware issue.
Has that been identified?  Has it been fixed for future firmware
releases?

s/D3Cold/D3cold/ to match above.

Is there a bug report for this issue?  Include the URL here.

Is there a bug report for the firmware?

> Signed-off-by: Ron Lee <ron.lee@...el.com>
> Reported-by: kernel test robot <lkp@...el.com>
> ---

Nits:

  - Use "Apollo Lake" to match Intel usage.

  - Below the "---" line, mention what changed between v1 and v2 (I
    see that you added the "#ifdef CONFIG_PCIEASPM", but you should
    save readers the effort of figuring that out).

  - For work-in-progress, the "Reported-by: kernel test robot" is
    pointless and I will remove it.  This quirk is not fixing a bug
    reported by the robot.

>  drivers/pci/quirks.c | 17 +++++++++++++++++
>  1 file changed, 17 insertions(+)
> 
> diff --git a/drivers/pci/quirks.c b/drivers/pci/quirks.c
> index 285acc4aaccc..fc959be17a9d 100644
> --- a/drivers/pci/quirks.c
> +++ b/drivers/pci/quirks.c
> @@ -5992,3 +5992,20 @@ DECLARE_PCI_FIXUP_HEADER(PCI_VENDOR_ID_INTEL, 0x9a2d, dpc_log_size);
>  DECLARE_PCI_FIXUP_HEADER(PCI_VENDOR_ID_INTEL, 0x9a2f, dpc_log_size);
>  DECLARE_PCI_FIXUP_HEADER(PCI_VENDOR_ID_INTEL, 0x9a31, dpc_log_size);
>  #endif
> +
> +#ifdef CONFIG_PCIEASPM
> +static void chromeos_fixup_apl_bridge_l1ss_capability(struct pci_dev *pdev)
> +{
> +	if (!dmi_match(DMI_SYS_VENDOR, "Google") ||
> +		(!dmi_match(DMI_PRODUCT_FAMILY, "Google_Coral") &&
> +		 !dmi_match(DMI_PRODUCT_FAMILY, "Google_Reef")))
> +		return;
> +
> +	pci_info(pdev, "Fix up L1SS Capability\n");
> +	/* Fix up the L1SS Capability Header*/
> +	pci_write_config_dword(pdev, pdev->l1ss, (0x220 << 20) | (1 << 16) | (PCI_EXT_CAP_ID_L1SS));

This looks like it adds a link to another capability at offset 0x220.
What is that, and how do we know this is safe?

These registers are read-only per spec (PCIe r6.0, sec 7.8.3.1), but I
guess you have device-specific knowledge that they are writable?

> +	/* Fix up the pointer to L1SS Capability*/
> +	pci_write_config_dword(pdev, 0x150, pdev->l1ss << 20);
> +}
> +DECLARE_PCI_FIXUP_RESUME(PCI_VENDOR_ID_INTEL, 0x5ad6, chromeos_fixup_apl_bridge_l1ss_capability);
> +#endif
> 
> base-commit: e2ca6ba6ba0152361aa4fcbf6067db71b2c7a770
> -- 
> 2.17.1
>