Linux kernel -stable discussions
 help / color / mirror / Atom feed
* [PATCH] platform/x86: hp-wmi: Enable cTGP in performance mode on OMEN 15-en1xxx
@ 2026-10-30  0:57 David William Richmond Davies-Jones
  2026-11-22 17:59 ` Krishna Chomal
  0 siblings, 1 reply; 2+ messages in thread
From: David William Richmond Davies-Jones @ 2026-10-30  0:57 UTC (permalink / raw)
  To: platform-driver-x86, Ilpo Järvinen
  Cc: Hans de Goede, Krishna Chomal, linux-kernel,
	David William Richmond Davies-Jones, stable

On the OMEN 15-en1xxx (board 88D2) the GPU power limit stays at its
80 W default in every platform profile, including performance.

The firmware's SET_PERFORMANCE_MODE handler (GM1A in the DSDT) stores
the mode byte in EC register 0x95 (HPCM) and enables NVIDIA cTGP through
the NPCF device only when that byte is 0x01:

	If ((HPCM == One)) { NPCF.CTGP = One ... }
	Else { NPCF.CTGP = Zero ... }

0x01 is the thermal policy v0 performance value. This board reports
policy v1 (byte 3 of GET_SYSTEM_DESIGN_DATA is 1), so the driver sends
0x31 for performance and cTGP is never enabled. NPCF only reports the
cTGP offset to nvidia-powerd (ACBT = 0xA0, i.e. +20 W) when CTGP is set.

For boards in a new omen_ctgp_boards[] list, set cTGP explicitly with
SET_GPU_THERMAL_MODES after setting the profile: enabled in performance,
disabled otherwise, with Dynamic Boost (PPAB) left enabled. The Victus S
code already uses this command. platform_profile_omen_set_ec() is also
called at probe time and when AC power returns, so cTGP follows the
profile in those cases too.

A report for the Victus 16 board 8934 shows the same HPCM check in its
GM1A method, so other v1 boards may be affected. Only 88D2 has been
tested, hence the list.

Tested on an OMEN Laptop 15-en1xxx (board 88D2, BIOS F.20, RTX 3070
Laptop GPU) on AC: nvidia-smi reports a 100 W power limit in
performance and 80 W in balanced and cool. Without this change it
reports 80 W in every profile.

The firmware analysis and this patch were prepared with the help of an
AI coding assistant (Claude Opus 5.5 in Claude Code). The result was
reviewed and tested on the hardware by the submitter.

Link: https://lore.kernel.org/platform-driver-x86/CA+LW3S2g4E971by7WhfFZGuyLT47141_Zn+eJ=wyqxB_+4G95A@mail.gmail.com/
Fixes: 286e937efbc7 ("platform/x86: hp-wmi: support omen thermal profile policy v1")
Cc: stable@vger.kernel.org
Signed-off-by: David William Richmond Davies-Jones <david@exultantmonkey.co.uk>
---
 drivers/platform/x86/hp/hp-wmi.c | 34 ++++++++++++++++++++++++++++++++
 1 file changed, 34 insertions(+)

diff --git a/drivers/platform/x86/hp/hp-wmi.c b/drivers/platform/x86/hp/hp-wmi.c
index 52077833ac89..e09a6d261144 100644
--- a/drivers/platform/x86/hp/hp-wmi.c
+++ b/drivers/platform/x86/hp/hp-wmi.c
@@ -235,6 +235,14 @@ static const char * const omen_timed_thermal_profile_boards[] = {
 	"8BAD",
 };
 
+/* DMI board names of Omen laptops whose firmware only enables NVIDIA cTGP
+ * from SET_PERFORMANCE_MODE for the v0 performance value. With v1 values cTGP
+ * stays disabled, so it has to be set with SET_GPU_THERMAL_MODES instead.
+ */
+static const char * const omen_ctgp_boards[] = {
+	"88D2",
+};
+
 /* DMI Board names of Victus 16-d and 16-e laptops */
 static const char * const victus_thermal_profile_boards[] = {
 	"88EE",
@@ -2035,6 +2043,17 @@ static bool has_omen_thermal_profile_ec_timer(void)
 			    board_name) >= 0;
 }
 
+static bool has_omen_ctgp_control(void)
+{
+	const char *board_name = dmi_get_system_info(DMI_BOARD_NAME);
+
+	if (!board_name)
+		return false;
+
+	return match_string(omen_ctgp_boards, ARRAY_SIZE(omen_ctgp_boards),
+			    board_name) >= 0;
+}
+
 inline int omen_thermal_profile_ec_flags_set(enum hp_thermal_profile_omen_flags flags)
 {
 	return ec_write(HP_OMEN_EC_THERMAL_PROFILE_FLAGS_OFFSET, flags);
@@ -2045,6 +2064,10 @@ inline int omen_thermal_profile_ec_timer_set(u8 value)
 	return ec_write(HP_OMEN_EC_THERMAL_PROFILE_TIMER_OFFSET, value);
 }
 
+static int victus_s_gpu_thermal_profile_set(bool ctgp_enable,
+					    bool ppab_enable,
+					    u8 dstate);
+
 static int platform_profile_omen_set_ec(enum platform_profile_option profile)
 {
 	int err, tp, tp_version;
@@ -2096,6 +2119,17 @@ static int platform_profile_omen_set_ec(enum platform_profile_option profile)
 			return err;
 	}
 
+	/*
+	 * Enable cTGP only in performance mode and keep Dynamic Boost enabled.
+	 * dstate 1 means 100%, as on Victus S.
+	 */
+	if (has_omen_ctgp_control()) {
+		err = victus_s_gpu_thermal_profile_set(profile == PLATFORM_PROFILE_PERFORMANCE,
+						       true, 1);
+		if (err)
+			pr_warn("Failed to set GPU cTGP state: %d\n", err);
+	}
+
 	return 0;
 }
 

base-commit: c339f4010dea8b0d754f84ace70cd1b089878ef8
-- 
2.56.0


^ permalink raw reply related	[flat|nested] 2+ messages in thread

* Re: [PATCH] platform/x86: hp-wmi: Enable cTGP in performance mode on OMEN 15-en1xxx
  2026-10-30  0:57 [PATCH] platform/x86: hp-wmi: Enable cTGP in performance mode on OMEN 15-en1xxx David William Richmond Davies-Jones
@ 2026-11-22 17:59 ` Krishna Chomal
  0 siblings, 0 replies; 2+ messages in thread
From: Krishna Chomal @ 2026-11-22 17:59 UTC (permalink / raw)
  To: David William Richmond Davies-Jones
  Cc: platform-driver-x86, Ilpo Järvinen, Hans de Goede,
	linux-kernel, stable

Hi David,

Thanks for sending the patch.

On Fri, Oct 30, 2026 at 12:57:24AM +0000, David William Richmond Davies-Jones wrote:
> On the OMEN 15-en1xxx (board 88D2) the GPU power limit stays at its
> 80 W default in every platform profile, including performance.
>
> The firmware's SET_PERFORMANCE_MODE handler (GM1A in the DSDT) stores
> the mode byte in EC register 0x95 (HPCM) and enables NVIDIA cTGP through
> the NPCF device only when that byte is 0x01:
>
> 	If ((HPCM == One)) { NPCF.CTGP = One ... }
> 	Else { NPCF.CTGP = Zero ... }
>
> 0x01 is the thermal policy v0 performance value. This board reports
> policy v1 (byte 3 of GET_SYSTEM_DESIGN_DATA is 1), so the driver sends
> 0x31 for performance and cTGP is never enabled. NPCF only reports the
> cTGP offset to nvidia-powerd (ACBT = 0xA0, i.e. +20 W) when CTGP is set.

This is interesting because the patch does not make any changes to the
HPWMI_SET_PERFORMANCE_MODE path. So, if I'm following correctly, the
call tree would roughly look like this:
platform_profile_omen_set()
   platform_profile_omen_set_ec()
     omen_get_thermal_policy_version()		(you say this returns 1)
       switch-case (performance) => tp = HP_OMEN_V1_THERMAL_PROFILE_PERFORMANCE
       omen_thermal_profile_set(0x31)

So perhaps policy v1 is not entirely wrong? The HPCM == 1 check in GM1A
only tells us how this particular method handles the performance mode
value. Have you checked the other consumers of HPCM, particularly GM22
and the NVIDIA-related SSDT methods? They may interpret the v1 value
(0x31) differently. Given that the firmware reports thermal policy v1,
I'd be hesitant to conclude that it is incorrect based solely on this
check.

> For boards in a new omen_ctgp_boards[] list, set cTGP explicitly with
> SET_GPU_THERMAL_MODES after setting the profile: enabled in performance,
> disabled otherwise, with Dynamic Boost (PPAB) left enabled. The Victus S
> code already uses this command. platform_profile_omen_set_ec() is also
> called at probe time and when AC power returns, so cTGP follows the
> profile in those cases too.

The diff follows Victus S code consistently. However, have you considered
adding an entry to hp_wmi_feature_boards[] and using the corresponding
omen_v1_legacy_thermal_params mapping instead? I'd prefer reusing
the existing feature-board handling rather than introducing another
board-specific list for functionality that already has an implementation.

If I'm not mistaken, board 88D2 should also support fan curves, right? By
keeping the board ID in omen_thermal_profile_boards[], wouldn't you also
be missing the manual PWM mode support?

> A report for the Victus 16 board 8934 shows the same HPCM check in its
> GM1A method, so other v1 boards may be affected. Only 88D2 has been
> tested, hence the list.
>
> Tested on an OMEN Laptop 15-en1xxx (board 88D2, BIOS F.20, RTX 3070
> Laptop GPU) on AC: nvidia-smi reports a 100 W power limit in
> performance and 80 W in balanced and cool. Without this change it
> reports 80 W in every profile.
>
> The firmware analysis and this patch were prepared with the help of an
> AI coding assistant (Claude Opus 5.5 in Claude Code). The result was
> reviewed and tested on the hardware by the submitter.
>
> Link: https://lore.kernel.org/platform-driver-x86/CA+LW3S2g4E971by7WhfFZGuyLT47141_Zn+eJ=wyqxB_+4G95A@mail.gmail.com/
> Fixes: 286e937efbc7 ("platform/x86: hp-wmi: support omen thermal profile policy v1")
> Cc: stable@vger.kernel.org
> Signed-off-by: David William Richmond Davies-Jones <david@exultantmonkey.co.uk>
> ---
> drivers/platform/x86/hp/hp-wmi.c | 34 ++++++++++++++++++++++++++++++++
> 1 file changed, 34 insertions(+)
>
> diff --git a/drivers/platform/x86/hp/hp-wmi.c b/drivers/platform/x86/hp/hp-wmi.c
> index 52077833ac89..e09a6d261144 100644
> --- a/drivers/platform/x86/hp/hp-wmi.c
> +++ b/drivers/platform/x86/hp/hp-wmi.c
> @@ -235,6 +235,14 @@ static const char * const omen_timed_thermal_profile_boards[] = {
> 	"8BAD",
> };
>
> +/* DMI board names of Omen laptops whose firmware only enables NVIDIA cTGP
> + * from SET_PERFORMANCE_MODE for the v0 performance value. With v1 values cTGP
> + * stays disabled, so it has to be set with SET_GPU_THERMAL_MODES instead.
> + */
> +static const char * const omen_ctgp_boards[] = {
> +	"88D2",
> +};
> +
> /* DMI Board names of Victus 16-d and 16-e laptops */
> static const char * const victus_thermal_profile_boards[] = {
> 	"88EE",
> @@ -2035,6 +2043,17 @@ static bool has_omen_thermal_profile_ec_timer(void)
> 			    board_name) >= 0;
> }
>
> +static bool has_omen_ctgp_control(void)
> +{
> +	const char *board_name = dmi_get_system_info(DMI_BOARD_NAME);
> +
> +	if (!board_name)
> +		return false;
> +
> +	return match_string(omen_ctgp_boards, ARRAY_SIZE(omen_ctgp_boards),
> +			    board_name) >= 0;
> +}
> +
> inline int omen_thermal_profile_ec_flags_set(enum hp_thermal_profile_omen_flags flags)
> {
> 	return ec_write(HP_OMEN_EC_THERMAL_PROFILE_FLAGS_OFFSET, flags);
> @@ -2045,6 +2064,10 @@ inline int omen_thermal_profile_ec_timer_set(u8 value)
> 	return ec_write(HP_OMEN_EC_THERMAL_PROFILE_TIMER_OFFSET, value);
> }
>
> +static int victus_s_gpu_thermal_profile_set(bool ctgp_enable,
> +					    bool ppab_enable,
> +					    u8 dstate);
> +
> static int platform_profile_omen_set_ec(enum platform_profile_option profile)
> {
> 	int err, tp, tp_version;
> @@ -2096,6 +2119,17 @@ static int platform_profile_omen_set_ec(enum platform_profile_option profile)
> 			return err;
> 	}
>
> +	/*
> +	 * Enable cTGP only in performance mode and keep Dynamic Boost enabled.
> +	 * dstate 1 means 100%, as on Victus S.
> +	 */
> +	if (has_omen_ctgp_control()) {
> +		err = victus_s_gpu_thermal_profile_set(profile == PLATFORM_PROFILE_PERFORMANCE,
> +						       true, 1);

Why enable Dynamic Boost unconditionally when the firmware would almost
certainly disable it in case of low-power/equivalent modes? Wouldn't it
be better to make PPAB follow the selected platform profile instead?

> +		if (err)
> +			pr_warn("Failed to set GPU cTGP state: %d\n", err);
> +	}
> +
> 	return 0;
> }
>
>
> base-commit: c339f4010dea8b0d754f84ace70cd1b089878ef8
> -- 
> 2.56.0
>

^ permalink raw reply	[flat|nested] 2+ messages in thread

end of thread, other threads:[~2026-11-22 17:59 UTC | newest]

Thread overview: 2+ messages (download: mbox.gz follow: Atom feed
-- links below jump to the message on this page --
2026-10-30  0:57 [PATCH] platform/x86: hp-wmi: Enable cTGP in performance mode on OMEN 15-en1xxx David William Richmond Davies-Jones
2026-11-22 17:59 ` Krishna Chomal

This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox