Re: [PATCH] platform/x86/intel/pmc: Add Sunrise Point-H support

From: Xi Pardee

Date: Mon Nov 16 2026 - 17:12:10 EST


Hi Angel,

Thanks for the patch, and for the detailed testing notes.
The change looks reasonable to me. Without it, SPT-H systems fall into the Cannon Point path, and the debugfs output (e.g. ltr_show) is wrong. Your comparison against direct PWRMBASE reads is helpful.

A few comments:
1. Fixes tag: This adds support for a new device ID rather than fixing a regression introduced by 661405bd817b, so I'd drop the Fixes tag. It would also pull the patch into stable backports.
2. PFEAR map: As you note, spt_pfear_map is the SPT-LP layout and SPT-H has two more PCIe controllers (SPD/SPE). I couldn't find public documentation for the PCH-H PFEAR bits either, so I'm fine leaving the map as is. Please state in the commit message that the map is reused from SPT-LP and that SPD/SPE are not covered, so pch_ip_power_gating_status is incomplete for those controllers. If you can, a cross-check of PFEAR against the root ports present on your system would be useful, but it isn't required. A follow-up fixing the map is welcome once the bits can be confirmed.

With the Fixes tag dropped and the commit message updated, please send a v2. I'm happy to give a Reviewed-by after that.
Thanks,
Xi

On 10/5/2026 1:00 PM, Angel Lindo via B4 Relay wrote:
From: Angel Lindo <angellindo@xxxxxxxxx>

spt_core_init() decides between the Sunrise Point and Cannon Point
register maps by looking for the Sunrise Point PMC PCI device, but only
the SPT-LP ID (0x9d21) is listed. On Skylake-H/Kaby Lake-H systems with
a Sunrise Point-H PCH (PMC 8086:a121), pci_dev_present() fails and the
driver takes the Coffee Lake path, using cnp_pmc_dev. That map does not
describe this PCH, so the debugfs output is wrong.

Seen on a MacBookPro14,3 (i7-7820HQ, PCH-H, PMC 8086:a121). With the
stock driver, ltr_show lists Cannon Point IP names (SOUTHPORT_A, CNV,
EVA, ...) and every entry reads RAW 0xffffffff. I was debugging why
the package never goes deeper than PC3 on this machine, and without a
working PMC map there is no way to see which PCH IP blocks it.

Add the SPT-H PMC ID to spt_pmc_pci_id[] so these systems use the
Sunrise Point map.

Tested on MacBookPro14,3 (Apple firmware 529.140.2.0.0), kernel 7.2.5,
with the module built with this ID:
- ltr_show lists the Sunrise Point IP names, and the values match a
direct read of the PWRMBASE registers through /dev/mem bit for bit.
- pch_ip_power_gating_status follows device state: LPSS reads On while
the Bluetooth UART (00:1e.0) is bound and turns Off once it is
unbound and the UART enters D3hot.
- No warnings on load.

Limitations:
- spt_pfear_map is the SPT-LP layout and is reused as is. SPT-H has
two more PCIe controllers (SPD and SPE). Bits 13 and 21 are RSVD in
the SPT map but SPD/SPE in the Cannon Point map; I could not verify
this on PCH-H, so the map is left unchanged.
- slp_s0_residency_usec reads 0 on this machine.

Fixes: 661405bd817b ("platform/x86: intel_pmc_core: Special case for Coffeelake")
Assisted-by: claude-opus-5-5
Signed-off-by: Angel Lindo <angellindo@xxxxxxxxx>
---
On the same machine and firmware, macOS reaches PC10 (~81%) at idle,
while Linux never goes deeper than PC3. This patch only fixes the PMC
detection so pch_ip_power_gating_status and ltr_show become usable for
that investigation. With it, the PCH IPs report gating states consistent
with device state (LPSS confirmed by toggling the Bluetooth UART); PC6
and SLP_S0 residency are still 0.
I plan to send a separate report on the PC3 limit to linux-pm.
---
drivers/platform/x86/intel/pmc/core.h | 1 +
drivers/platform/x86/intel/pmc/spt.c | 1 +
2 files changed, 2 insertions(+)

diff --git a/drivers/platform/x86/intel/pmc/core.h b/drivers/platform/x86/intel/pmc/core.h
index b4c7399f8..5822dbbda 100644
--- a/drivers/platform/x86/intel/pmc/core.h
+++ b/drivers/platform/x86/intel/pmc/core.h
@@ -36,6 +36,7 @@ DEFINE_FREE(pmc_acpi_free, void *, if (_T) ACPI_FREE(_T))
/* Sunrise Point Power Management Controller PCI Device ID */
#define SPT_PMC_PCI_DEVICE_ID 0x9d21
+#define SPT_H_PMC_PCI_DEVICE_ID 0xa121
#define SPT_PMC_BASE_ADDR_OFFSET 0x48
#define SPT_PMC_SLP_S0_RES_COUNTER_OFFSET 0x13c
#define SPT_PMC_PM_CFG_OFFSET 0x18
diff --git a/drivers/platform/x86/intel/pmc/spt.c b/drivers/platform/x86/intel/pmc/spt.c
index b50534aa2..5c4b8d536 100644
--- a/drivers/platform/x86/intel/pmc/spt.c
+++ b/drivers/platform/x86/intel/pmc/spt.c
@@ -138,6 +138,7 @@ static const struct pmc_reg_map spt_reg_map = {
static const struct pci_device_id spt_pmc_pci_id[] = {
{ PCI_VDEVICE(INTEL, SPT_PMC_PCI_DEVICE_ID) },
+ { PCI_VDEVICE(INTEL, SPT_H_PMC_PCI_DEVICE_ID) },
{ }
};

---
base-commit: cf963c9359dd6c5d0bc6af16df5cf1f691bba4f3
change-id: 20261005-intel-pmc-spt-h-3ebe43d206fb

Best regards,
--
Angel Lindo <angellindo@xxxxxxxxx>