From: "任海 " <rh_king@163.com>
To: "Bjorn Helgaas" <helgaas@kernel.org>
Cc: "Bjorn Helgaas" <bhelgaas@google.com>,
linux-pci@vger.kernel.org,
"linux-kernel@vger.kernel.org" <linux-kernel@vger.kernel.org>,
"Ryan Brue" <ryanbrue.dev@gmail.com>,
"derekjohn.clark@gmail.com" <derekjohn.clark@gmail.com>,
"mpearson-lenovo@squebb.ca" <mpearson-lenovo@squebb.ca>,
"Timur Tabi" <ttabi@nvidia.com>,
"John Hubbard" <jhubbard@nvidia.com>,
dri-devel@lists.freedesktop.org, nouveau@lists.freedesktop.org,
"Lyude Paul" <lyude@redhat.com>
Subject: Re:Re: [PATCH v3] PCI: Disable spurious s2idle wake from the NVMe root port on ThinkPad P1 Gen 9
Date: Tue, 17 Nov 2026 14:01:33 +0800 (CST) [thread overview]
Message-ID: <6f0ee424.4816.1a1e818acdf.Coremail.rh_king@163.com> (raw)
In-Reply-To: <20261116202943.GA204660@bhelgaas>
[-- Attachment #1: Type: text/plain, Size: 6540 bytes --]
Hi Bjorn,
Thanks for the pointer and for Cc'ing the relevant people.
I looked at the Legion patch. While both issues involve firmware firing a spurious GPE on a PCIe port, I believe they are distinct problems and should not be merged:
Different trigger context. The Legion quirk targets a runtime D3Cold wake loop: the GPU enters D3Cold, firmware fires a GPE on its parent port, the GPU wakes, and the cycle repeats. It is explicitly driver-independent (reproduces on nouveau/nova, possibly without any driver) . My P1 Gen 9 issue is a system s2idle abort: the NVMe controller enters D3 during suspend, a spurious PME sets GPE 0x6F during the noirq phase, and s2idle never reaches the low-power state .
Different target and scope. Legion matches NVIDIA GPU device IDs and acts on the GPU’s parent ACPI device . My quirk matches the 21UF DMI and disables wakeup on a specific root port (00:1c.4) associated with the primary NVMe SSD .
Different fix mechanism. Legion removes the parent ACPI device’s PM notifier so the spurious GPE no longer causes a wakeup . My quirk calls device_set_wakeup_enable(..., false) on the root port after checking device_may_wakeup() .
So while the underlying firmware misbehavior is similar in spirit, the affected hardware, power-management context, and workaround are different. I don’t think they can be combined.
And more important, I tested that patch not wroking for me.
Thanks,
Kean
At 2026-11-17 04:29:43, "Bjorn Helgaas" <helgaas@kernel.org> wrote:
>[+cc others from https://lore.kernel.org/linux-pci/20261110-broken-gpe-legion-v4-1-c672bb1c4d91@gmail.com/;
>is this the same issue? That is for GPUs, not NVMe, but sounds similar]
>
>On Mon, Nov 16, 2026 at 02:56:12PM +0800, Kean Ren wrote:
>> On Lenovo ThinkPad P1 Gen 9 (21UF, Panther Lake), firmware enables the
>> ACPI wake capability (_PRW) of PCIe root port 00:1c.4, which is the root
>> port of the primary NVMe SSD. When the NVMe controller is put into D3
>> during system suspend, a spurious PME is raised on that root port, which
>> sets ACPI GPE 0x6F (SCI). The SCI fires during the noirq suspend phase
>> and aborts s2idle before it reaches the low-power state: every suspend
>> is followed by a wake within ~2.5 seconds with practically zero s2idle
>> residency.
>>
>> The downstream NVMe device is not a wake source (its own ACPI wake is
>> disabled), so the spurious event comes from the root port's wake/PME
>> path. Disable the root port's wake capability as well.
>>
>> No firmware with a fix is available: the issue reproduces with BIOS
>> N4SET31W (1.10) and N4SET32W (1.11), the latest version at the time of
>> writing. Match the 21UF model prefix; all 21UF SKUs share the same
>> platform and firmware, and the remaining MTM characters encode the
>> factory configuration and region. If a fixed BIOS is released later,
>> this quirk can be removed or restricted to the affected BIOS versions.
>>
>> Verified on 21UFZE0PCN:
>> - gpe6F increments by one per suspend/resume on both BIOS versions;
>> - masking gpe6F, or disabling RP05 wake via /proc/acpi/wakeup, makes
>> s2idle stay asleep;
>> - with this quirk, /proc/acpi/wakeup reports RP05 as disabled and
>> gpe6F no longer increments across repeated suspends.
>>
>> Signed-off-by: Kean Ren <rh_king@163.com>
>> ---
>> Changes in v3:
>> - Check device_may_wakeup() instead of device_can_wakeup(): the latter is
>> set for all PME-capable devices, whether or not wakeup is enabled, so the
>> quirk could log a disable that did not actually happen. Drop the
>> corresponding warning; there is nothing to do if wakeup is not enabled.
>>
>> Changes in v2:
>> - Match the 21UF model prefix instead of the full 21UFZE0PCN MTM so the
>> quirk also covers the other ThinkPad P1 Gen 9 factory configurations
>> that share the same platform and firmware.
>>
>> drivers/pci/quirks.c | 56 ++++++++++++++++++++++++++++++++++++++++++++
>> 1 file changed, 56 insertions(+)
>>
>> diff --git a/drivers/pci/quirks.c b/drivers/pci/quirks.c
>> index 7f1f47ea2ee4..6ede2ea49b3a 100644
>> --- a/drivers/pci/quirks.c
>> +++ b/drivers/pci/quirks.c
>> @@ -6595,3 +6595,59 @@ static void pci_mask_replay_timer_timeout(struct pci_dev *pdev)
>> DECLARE_PCI_FIXUP_FINAL(PCI_VENDOR_ID_GLI, 0x9750, pci_mask_replay_timer_timeout);
>> DECLARE_PCI_FIXUP_FINAL(PCI_VENDOR_ID_GLI, 0x9755, pci_mask_replay_timer_timeout);
>> #endif
>> +
>> +/*
>> + * Lenovo ThinkPad P1 Gen 9 (21UF): the firmware enables the ACPI wake
>> + * capability of PCIe root port 00:1c.4, which is the root port of the
>> + * primary NVMe SSD. When the NVMe controller is put into D3 during
>> + * system suspend, a spurious PME is raised on that root port, which in
>> + * turn fires ACPI GPE 0x6F (SCI) and aborts s2idle before the system
>> + * reaches the low-power state. The NVMe device itself is not a wake
>> + * source (its own ACPI wake is disabled), so disable the root port's
>> + * wake capability as well.
>> + *
>> + * No firmware with a fix is available: the issue reproduces with BIOS
>> + * N4SET31W (1.10) and N4SET32W (1.11), the latest version at the time
>> + * of writing. Match the 21UF model prefix (the remaining MTM
>> + * characters encode the factory configuration and region); if a fixed
>> + * BIOS is released later, this quirk can be removed or restricted to
>> + * the affected BIOS versions.
>> + */
>> +static const struct dmi_system_id p1g9_rp05_wakeup_dmi[] __initconst = {
>> + {
>> + .ident = "Lenovo ThinkPad P1 Gen 9 (21UF)",
>> + .matches = {
>> + DMI_MATCH(DMI_SYS_VENDOR, "LENOVO"),
>> + /* 21UF = ThinkPad P1 Gen 9 model; suffix is the SKU. */
>> + DMI_MATCH(DMI_PRODUCT_NAME, "21UF"),
>> + },
>> + },
>> + { }
>> +};
>> +
>> +static int __init pci_fixup_p1g9_rp05_wakeup(void)
>> +{
>> + struct pci_dev *rp;
>> +
>> + if (!dmi_check_system(p1g9_rp05_wakeup_dmi))
>> + return 0;
>> +
>> + /* Root port of the primary NVMe SSD (05:00.0). */
>> + rp = pci_get_domain_bus_and_slot(0, 0, PCI_DEVFN(0x1c, 4));
>> + if (!rp)
>> + return 0;
>> +
>> + /*
>> + * Disable wakeup only if it is enabled: device_can_wakeup() is set
>> + * for all PME-capable devices, even when wakeup is disabled.
>> + */
>> + if (device_may_wakeup(&rp->dev)) {
>> + device_set_wakeup_enable(&rp->dev, false);
>> + pci_info(rp, "disabled spurious s2idle wake (BIOS %s)\n",
>> + dmi_get_system_info(DMI_BIOS_VERSION));
>> + }
>> + pci_dev_put(rp);
>> +
>> + return 0;
>> +}
>> +late_initcall(pci_fixup_p1g9_rp05_wakeup);
>> --
>> 2.43.0
>>
[-- Attachment #2: Type: text/html, Size: 12868 bytes --]
WARNING: multiple messages have this Message-ID (diff)
From: "任海 " <rh_king@163.com>
To: "Bjorn Helgaas" <helgaas@kernel.org>
Cc: Bjorn Helgaas <bhelgaas@google.com>,
linux-pci@vger.kernel.org,
"linux-kernel@vger.kernel.org" <linux-kernel@vger.kernel.org>,
Ryan Brue <ryanbrue.dev@gmail.com>,
"derekjohn.clark@gmail.com" <derekjohn.clark@gmail.com>,
"mpearson-lenovo@squebb.ca" <mpearson-lenovo@squebb.ca>,
dri-devel@lists.freedesktop.org, nouveau@lists.freedesktop.org
Subject: Re:Re: [PATCH v3] PCI: Disable spurious s2idle wake from the NVMe root port on ThinkPad P1 Gen 9
Date: Tue, 17 Nov 2026 14:01:33 +0800 (CST) [thread overview]
Message-ID: <6f0ee424.4816.1a1e818acdf.Coremail.rh_king@163.com> (raw)
In-Reply-To: <20261116202943.GA204660@bhelgaas>
[-- Attachment #1: Type: text/plain, Size: 6540 bytes --]
Hi Bjorn,
Thanks for the pointer and for Cc'ing the relevant people.
I looked at the Legion patch. While both issues involve firmware firing a spurious GPE on a PCIe port, I believe they are distinct problems and should not be merged:
Different trigger context. The Legion quirk targets a runtime D3Cold wake loop: the GPU enters D3Cold, firmware fires a GPE on its parent port, the GPU wakes, and the cycle repeats. It is explicitly driver-independent (reproduces on nouveau/nova, possibly without any driver) . My P1 Gen 9 issue is a system s2idle abort: the NVMe controller enters D3 during suspend, a spurious PME sets GPE 0x6F during the noirq phase, and s2idle never reaches the low-power state .
Different target and scope. Legion matches NVIDIA GPU device IDs and acts on the GPU’s parent ACPI device . My quirk matches the 21UF DMI and disables wakeup on a specific root port (00:1c.4) associated with the primary NVMe SSD .
Different fix mechanism. Legion removes the parent ACPI device’s PM notifier so the spurious GPE no longer causes a wakeup . My quirk calls device_set_wakeup_enable(..., false) on the root port after checking device_may_wakeup() .
So while the underlying firmware misbehavior is similar in spirit, the affected hardware, power-management context, and workaround are different. I don’t think they can be combined.
And more important, I tested that patch not wroking for me.
Thanks,
Kean
At 2026-11-17 04:29:43, "Bjorn Helgaas" <helgaas@kernel.org> wrote:
>[+cc others from https://lore.kernel.org/linux-pci/20261110-broken-gpe-legion-v4-1-c672bb1c4d91@gmail.com/;
>is this the same issue? That is for GPUs, not NVMe, but sounds similar]
>
>On Mon, Nov 16, 2026 at 02:56:12PM +0800, Kean Ren wrote:
>> On Lenovo ThinkPad P1 Gen 9 (21UF, Panther Lake), firmware enables the
>> ACPI wake capability (_PRW) of PCIe root port 00:1c.4, which is the root
>> port of the primary NVMe SSD. When the NVMe controller is put into D3
>> during system suspend, a spurious PME is raised on that root port, which
>> sets ACPI GPE 0x6F (SCI). The SCI fires during the noirq suspend phase
>> and aborts s2idle before it reaches the low-power state: every suspend
>> is followed by a wake within ~2.5 seconds with practically zero s2idle
>> residency.
>>
>> The downstream NVMe device is not a wake source (its own ACPI wake is
>> disabled), so the spurious event comes from the root port's wake/PME
>> path. Disable the root port's wake capability as well.
>>
>> No firmware with a fix is available: the issue reproduces with BIOS
>> N4SET31W (1.10) and N4SET32W (1.11), the latest version at the time of
>> writing. Match the 21UF model prefix; all 21UF SKUs share the same
>> platform and firmware, and the remaining MTM characters encode the
>> factory configuration and region. If a fixed BIOS is released later,
>> this quirk can be removed or restricted to the affected BIOS versions.
>>
>> Verified on 21UFZE0PCN:
>> - gpe6F increments by one per suspend/resume on both BIOS versions;
>> - masking gpe6F, or disabling RP05 wake via /proc/acpi/wakeup, makes
>> s2idle stay asleep;
>> - with this quirk, /proc/acpi/wakeup reports RP05 as disabled and
>> gpe6F no longer increments across repeated suspends.
>>
>> Signed-off-by: Kean Ren <rh_king@163.com>
>> ---
>> Changes in v3:
>> - Check device_may_wakeup() instead of device_can_wakeup(): the latter is
>> set for all PME-capable devices, whether or not wakeup is enabled, so the
>> quirk could log a disable that did not actually happen. Drop the
>> corresponding warning; there is nothing to do if wakeup is not enabled.
>>
>> Changes in v2:
>> - Match the 21UF model prefix instead of the full 21UFZE0PCN MTM so the
>> quirk also covers the other ThinkPad P1 Gen 9 factory configurations
>> that share the same platform and firmware.
>>
>> drivers/pci/quirks.c | 56 ++++++++++++++++++++++++++++++++++++++++++++
>> 1 file changed, 56 insertions(+)
>>
>> diff --git a/drivers/pci/quirks.c b/drivers/pci/quirks.c
>> index 7f1f47ea2ee4..6ede2ea49b3a 100644
>> --- a/drivers/pci/quirks.c
>> +++ b/drivers/pci/quirks.c
>> @@ -6595,3 +6595,59 @@ static void pci_mask_replay_timer_timeout(struct pci_dev *pdev)
>> DECLARE_PCI_FIXUP_FINAL(PCI_VENDOR_ID_GLI, 0x9750, pci_mask_replay_timer_timeout);
>> DECLARE_PCI_FIXUP_FINAL(PCI_VENDOR_ID_GLI, 0x9755, pci_mask_replay_timer_timeout);
>> #endif
>> +
>> +/*
>> + * Lenovo ThinkPad P1 Gen 9 (21UF): the firmware enables the ACPI wake
>> + * capability of PCIe root port 00:1c.4, which is the root port of the
>> + * primary NVMe SSD. When the NVMe controller is put into D3 during
>> + * system suspend, a spurious PME is raised on that root port, which in
>> + * turn fires ACPI GPE 0x6F (SCI) and aborts s2idle before the system
>> + * reaches the low-power state. The NVMe device itself is not a wake
>> + * source (its own ACPI wake is disabled), so disable the root port's
>> + * wake capability as well.
>> + *
>> + * No firmware with a fix is available: the issue reproduces with BIOS
>> + * N4SET31W (1.10) and N4SET32W (1.11), the latest version at the time
>> + * of writing. Match the 21UF model prefix (the remaining MTM
>> + * characters encode the factory configuration and region); if a fixed
>> + * BIOS is released later, this quirk can be removed or restricted to
>> + * the affected BIOS versions.
>> + */
>> +static const struct dmi_system_id p1g9_rp05_wakeup_dmi[] __initconst = {
>> + {
>> + .ident = "Lenovo ThinkPad P1 Gen 9 (21UF)",
>> + .matches = {
>> + DMI_MATCH(DMI_SYS_VENDOR, "LENOVO"),
>> + /* 21UF = ThinkPad P1 Gen 9 model; suffix is the SKU. */
>> + DMI_MATCH(DMI_PRODUCT_NAME, "21UF"),
>> + },
>> + },
>> + { }
>> +};
>> +
>> +static int __init pci_fixup_p1g9_rp05_wakeup(void)
>> +{
>> + struct pci_dev *rp;
>> +
>> + if (!dmi_check_system(p1g9_rp05_wakeup_dmi))
>> + return 0;
>> +
>> + /* Root port of the primary NVMe SSD (05:00.0). */
>> + rp = pci_get_domain_bus_and_slot(0, 0, PCI_DEVFN(0x1c, 4));
>> + if (!rp)
>> + return 0;
>> +
>> + /*
>> + * Disable wakeup only if it is enabled: device_can_wakeup() is set
>> + * for all PME-capable devices, even when wakeup is disabled.
>> + */
>> + if (device_may_wakeup(&rp->dev)) {
>> + device_set_wakeup_enable(&rp->dev, false);
>> + pci_info(rp, "disabled spurious s2idle wake (BIOS %s)\n",
>> + dmi_get_system_info(DMI_BIOS_VERSION));
>> + }
>> + pci_dev_put(rp);
>> +
>> + return 0;
>> +}
>> +late_initcall(pci_fixup_p1g9_rp05_wakeup);
>> --
>> 2.43.0
>>
[-- Attachment #2: Type: text/html, Size: 12868 bytes --]
next prev parent reply other threads:[~2026-11-17 8:07 UTC|newest]
Thread overview: 11+ messages / expand[flat|nested] mbox.gz Atom feed top
2026-11-13 1:48 [PATCH] PCI: Disable spurious s2idle wake from the NVMe root port on ThinkPad P1 Gen 9 Kean Ren
2026-11-13 2:12 ` sashiko-bot
2026-11-16 1:42 ` [PATCH v2] " Kean Ren
2026-11-16 2:08 ` sashiko-bot
2026-11-16 6:56 ` [PATCH v3] " Kean Ren
2026-11-16 7:06 ` sashiko-bot
2026-11-16 20:29 ` Bjorn Helgaas
2026-11-16 20:29 ` Bjorn Helgaas
2026-11-17 6:01 ` 任海 [this message]
2026-11-17 6:01 ` 任海
2026-11-17 6:37 ` Kean Ren
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=6f0ee424.4816.1a1e818acdf.Coremail.rh_king@163.com \
--to=rh_king@163.com \
--cc=bhelgaas@google.com \
--cc=derekjohn.clark@gmail.com \
--cc=dri-devel@lists.freedesktop.org \
--cc=helgaas@kernel.org \
--cc=jhubbard@nvidia.com \
--cc=linux-kernel@vger.kernel.org \
--cc=linux-pci@vger.kernel.org \
--cc=lyude@redhat.com \
--cc=mpearson-lenovo@squebb.ca \
--cc=nouveau@lists.freedesktop.org \
--cc=ryanbrue.dev@gmail.com \
--cc=ttabi@nvidia.com \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is an external index of several public inboxes,
see mirroring instructions on how to clone and mirror
all data and code used by this external index.