[REGRESSION] libata: SATA LPM forcibly disabled on Intel Raptor Lake since Linux 6.14.

[REGRESSION] libata: SATA LPM forcibly disabled on Intel Raptor Lake since Linux 6.14.

[wolf]

Hello,

I am reporting a power-management regression in libata affecting the 
Intel Raptor Lake SATA AHCI Controller, introduced after Linux 6.14.

Hardware:
- CPU: Intel Core i5-12500
- SATA controller: Intel Corporation Raptor Lake SATA AHCI Controller 
[8086:7a62]
- Motherboard: Gigabyte B760 GAMING X DDR4

Distributions tested:
- Proxmox / Debian 13 (Trixie)
- Chimera Linux Live USB (kernel 6.18)

Kernel versions tested:
- 6.14.x (Proxmox): OK
- 6.17.x (Proxmox): REGRESSION
- 6.18.x (Chimera): REGRESSION

Problem description:
On kernel versions newer than 6.14, SATA Link Power Management (LPM) is 
forcibly disabled on part of the controller.
Out of 8 SCSI hosts, 4 hosts are stuck with link_power_management_policy 
set to "max_performance" and cannot be changed:

max_performance
max_performance
max_performance
max_performance
med_power_with_dipm
med_power_with_dipm
med_power_with_dipm
med_power_with_dipm

The motherboard only has 4 physical SATA ports, so the presence of 8 
hosts and 4 apparent dummy ports might be unexpected?

Manually setting LPM on the affected hosts fails with:
echo med_power_with_dipm > 
/sys/class/scsi_host/host*/link_power_management_policy
-bash: echo: write error: Operation not supported

This worked correctly on 6.14.x and earlier.

Because LPM cannot be enabled on all ports, the system is unable to 
enter deeper package C-states.

Observed effects:
- CPU package C-states are stuck at C3 (normally reaches C8 or deeper)
- Idle power consumption increased by approximately 4 watts
- powertop reports 4 out of 8 SATA LPM tunables as permanently "Bad"

SATA LPM worked reliably on this hardware on previous kernels

dmesg logs:
Working (6.14):
https://www.dropbox.com/scl/fi/5ryfjrp9fgx1dkxmszm5l/6.14-proxmox.txt?rlkey=75y3hue5whdmxbfvxq1piyofg&raw=1

Not working:
6.17 (Proxmox):
https://www.dropbox.com/scl/fi/ja2w0qtfummkmh406rxc7/6.17-proxmox.txt?rlkey=qxqpi7w2rn0ecbhdf2t4ju9kh&raw=1

6.18 (Chimera):
https://www.dropbox.com/scl/fi/s7412dfcv3pjk63mqfrep/6.18-chimera.txt?rlkey=9dd5wylsxb45l1b9zxgxadaw7&raw=1

While reviewing the logs, I noticed:
The presence of 4 apparent dummy SATA ports
Exactly 4 out of 8 SATA LPM entries disabled in both sysfs and powertop
This suggests the issue may be related to SATA port enumeration or 
libata power-management handling in newer kernels.


Expected behavior:
The ability to enable SATA Link Power Management on all SATA hosts 
should be restored, allowing the system to enter deeper package C-states 
and return to previous idle power levels.


Best regards,
Wolf

Re: [REGRESSION] libata: SATA LPM forcibly disabled on Intel Raptor Lake since Linux 6.14.

[Niklas Cassel]

On Tue, Jan 06, 2026 at 08:57:07PM +0100, wolf@yoxt.cc wrote:
> Hello,
> 
> I am reporting a power-management regression in libata affecting the Intel
> Raptor Lake SATA AHCI Controller, introduced after Linux 6.14.

(snip)

> and 4 apparent dummy ports might be unexpected?
> 
> Manually setting LPM on the affected hosts fails with:
> echo med_power_with_dipm >
> /sys/class/scsi_host/host*/link_power_management_policy
> -bash: echo: write error: Operation not supported

This is most likely because your ports are marked as External / Hotplug
capable by BIOS.

LPM and Hotplug are mutually exlusive in SATA.
So in order allow your ports to detect a hotplug event, LPM cannot be
disabled.


Things you can do:
-Go into your BIOS settings and disable the hot-plugging capability for the
 ports.
-If you don't have such a setting, you can use the mask_port_ext AHCI module
 parameter. Simply add:
 ahci.mask_port_ext=<your portmask here> to your kernel command line


For more info, see this thread:
https://lore.kernel.org/linux-ide/5a74106a-d9db-4e63-b3e6-3857f30a1190@interia.pl/T/


Kind regards,
Niklas

Re: [REGRESSION] libata: SATA LPM forcibly disabled on Intel Raptor Lake since Linux 6.14.

[Damien Le Moal]

On 1/7/26 11:14, Niklas Cassel wrote:
> On Tue, Jan 06, 2026 at 08:57:07PM +0100, wolf@yoxt.cc wrote:
>> Hello,
>>
>> I am reporting a power-management regression in libata affecting the Intel
>> Raptor Lake SATA AHCI Controller, introduced after Linux 6.14.
> 
> (snip)
> 
>> and 4 apparent dummy ports might be unexpected?
>>
>> Manually setting LPM on the affected hosts fails with:
>> echo med_power_with_dipm >
>> /sys/class/scsi_host/host*/link_power_management_policy
>> -bash: echo: write error: Operation not supported

This will not work with the dummy ports. Can you do that successfully with the
valid ports ?

> 
> This is most likely because your ports are marked as External / Hotplug
> capable by BIOS.

I do not think so. There is no "ext" advertized in the log for the valid ports.
But still worth trying to see if the BIOS has a hotplug option and disable it if
it does.

> 
> LPM and Hotplug are mutually exlusive in SATA.
> So in order allow your ports to detect a hotplug event, LPM cannot be
> disabled.
> 
> 
> Things you can do:
> -Go into your BIOS settings and disable the hot-plugging capability for the
>  ports.
> -If you don't have such a setting, you can use the mask_port_ext AHCI module
>  parameter. Simply add:
>  ahci.mask_port_ext=<your portmask here> to your kernel command line
> 
> 
> For more info, see this thread:
> https://lore.kernel.org/linux-ide/5a74106a-d9db-4e63-b3e6-3857f30a1190@interia.pl/T/
> 
> 
> Kind regards,
> Niklas


-- 
Damien Le Moal
Western Digital Research

Re: [REGRESSION] libata: SATA LPM forcibly disabled on Intel Raptor Lake since Linux 6.14.

[Damien Le Moal]

On 1/7/26 11:28, wolf@yoxt.cc wrote:
> On 2026-01-07 11:19, Damien Le Moal wrote:
>> On 1/7/26 11:14, Niklas Cassel wrote:
>>> On Tue, Jan 06, 2026 at 08:57:07PM +0100, wolf@yoxt.cc wrote:
>>>> Hello,
>>>>
>>>> I am reporting a power-management regression in libata affecting the 
>>>> Intel
>>>> Raptor Lake SATA AHCI Controller, introduced after Linux 6.14.
>>>
>>> (snip)
>>>
>>>> and 4 apparent dummy ports might be unexpected?
>>>>
>>>> Manually setting LPM on the affected hosts fails with:
>>>> echo med_power_with_dipm >
>>>> /sys/class/scsi_host/host*/link_power_management_policy
>>>> -bash: echo: write error: Operation not supported
>>
>> This will not work with the dummy ports. Can you do that successfully 
>> with the
>> valid ports ?
>>
>>>
>>> This is most likely because your ports are marked as External / 
>>> Hotplug
>>> capable by BIOS.
>>
>> I do not think so. There is no "ext" advertized in the log for the 
>> valid ports.
>> But still worth trying to see if the BIOS has a hotplug option and 
>> disable it if
>> it does.
>>
>>>
>>> LPM and Hotplug are mutually exlusive in SATA.
>>> So in order allow your ports to detect a hotplug event, LPM cannot be
>>> disabled.
>>>
>>>
>>> Things you can do:
>>> -Go into your BIOS settings and disable the hot-plugging capability 
>>> for the
>>>  ports.
>>> -If you don't have such a setting, you can use the mask_port_ext AHCI 
>>> module
>>>  parameter. Simply add:
>>>  ahci.mask_port_ext=<your portmask here> to your kernel command line
>>>
>>>
>>> For more info, see this thread:
>>> https://lore.kernel.org/linux-ide/5a74106a-d9db-4e63-b3e6-3857f30a1190@interia.pl/T/
>>>
>>>
>>> Kind regards,
>>> Niklas
> 
> Hello Damien,
> 
> I can apply med_power_with_dipm without error successfully to 4 out of 
> the 8 available.
> 
> /sys/class/scsi_host/host0
> /sys/class/scsi_host/host1
> /sys/class/scsi_host/host2
> /sys/class/scsi_host/host3
> - these do not work, those give an error.

Yep, that is expected since these are dummy ports. So all good.
These come from the adapter indicating that the ports are not actually physical
ones. I wonder if something is off here. How many SATA plugs does your main
board has ? Only 4 ? More ?



-- 
Damien Le Moal
Western Digital Research

Re: [REGRESSION] libata: SATA LPM forcibly disabled on Intel Raptor Lake since Linux 6.14.

[Niklas Cassel]

On Wed, Jan 07, 2026 at 11:19:42AM +0100, Damien Le Moal wrote:
> > 
> > This is most likely because your ports are marked as External / Hotplug
> > capable by BIOS.
> 
> I do not think so. There is no "ext" advertized in the log for the valid ports.
> But still worth trying to see if the BIOS has a hotplug option and disable it if
> it does.

Good catch.

One thing that caught my eye:
[    1.433187] ata6.00: ATAPI: PIONEER BD-RW  BDR-S12JX, 1.01, max UDMA/100
[    1.433272] ata5.00: ATAPI: PIONEER BD-RW  BDR-S12JX, 1.01, max UDMA/100

We've had so many problems with PIONEER...


So, since we concluded that you don't have any ports marked as External,
to get deeper sleep states, all SATA ports will need to be in a power
saving state.

Ports that don't have any device connected will get powered off, so they
are no problem.


For LPM to work, you need to use either HIPM (host-initiated) or DIPM
(device initiated). You can have both enabled, but the device needs to
support one of them to enter a power saving state.


I think the problem is:

[    3.010059] ata8.00: ATA-11: ST16000NM000J-2TW103, SN06, max UDMA/133
[    3.010079] ata7.00: ATA-11: ST16000NM000J-2TW103, SN06, max UDMA/133
[    3.012748] ata6.00: ATAPI: PIONEER BD-RW  BDR-S12JX, 1.01, max UDMA/100
[    3.012779] ata5.00: ATAPI: PIONEER BD-RW  BDR-S12JX, 1.01, max UDMA/100
[    3.061873] ata8.00: Features: DIPM NCQ-sndrcv
[    3.065319] ata7.00: Features: DIPM NCQ-sndrcv

The ST drives advertise support for DIPM, so these two ports should enter
a lower power state.


For the PIONEER optical drives however, we don't seem to get any "Features: "
print at all. My guess is that the support neither HIPM nor DIPM, and thus
stops the whole system from entering a lower power state.


What I think happens:

For a device that does not support HIPM, we will clear flag ATA_LPM_HIPM from
hints:
https://github.com/torvalds/linux/blob/v6.19-rc4/drivers/ata/libata-eh.c#L2168-L2169

If flag ATA_LPM_HIPM is not set, we will disable HIPM (ALPE) for the port:
https://github.com/torvalds/linux/blob/v6.19-rc4/drivers/ata/libahci.c#L835-L847

Thus the HBA will never tell the device to enter a low power state.
(And if DIPM is not supported, the device itself will never enter a low
power state.)


Questions:
-Why was this not a problem on older kernels?

If you are willing try some patches from us, we could give you a patch with more
debug prints, to show for each device, when we {enable,disable} {HIPM,DIPM}.


-Why don't we get the Features: print for ATAPI devices?

We probably need something like this:

diff --git a/drivers/ata/libata-core.c b/drivers/ata/libata-core.c
index fc250b494ce3..a8af200ac24f 100644
--- a/drivers/ata/libata-core.c
+++ b/drivers/ata/libata-core.c
@@ -3161,11 +3161,13 @@ int ata_dev_configure(struct ata_device *dev)
                /* print device info to dmesg */
                if (print_info)
                        ata_dev_info(dev,
-                                    "ATAPI: %s, %s, max %s%s%s%s\n",
+                                    "ATAPI: %s, %s, max %s%s%s%s%s%s\n",
                                     modelbuf, fwrevbuf,
                                     ata_mode_string(xfer_mask),
                                     cdb_intr_string, atapi_an_string,
-                                    dma_dir_string);
+                                    dma_dir_string,
+                                    ata_id_has_hipm(dev->id) ? " HIPM" : "",
+                                    ata_id_has_dipm(dev->id) ? " DIPM" : "");
        }
 
        /* determine max_sectors */



Kind regards,
Niklas

Re: [REGRESSION] libata: SATA LPM forcibly disabled on Intel Raptor Lake since Linux 6.14.

[Damien Le Moal]

On 1/7/26 12:20, Niklas Cassel wrote:
> I think the problem is:
> 
> [    3.010059] ata8.00: ATA-11: ST16000NM000J-2TW103, SN06, max UDMA/133
> [    3.010079] ata7.00: ATA-11: ST16000NM000J-2TW103, SN06, max UDMA/133
> [    3.012748] ata6.00: ATAPI: PIONEER BD-RW  BDR-S12JX, 1.01, max UDMA/100
> [    3.012779] ata5.00: ATAPI: PIONEER BD-RW  BDR-S12JX, 1.01, max UDMA/100
> [    3.061873] ata8.00: Features: DIPM NCQ-sndrcv
> [    3.065319] ata7.00: Features: DIPM NCQ-sndrcv
> 
> The ST drives advertise support for DIPM, so these two ports should enter
> a lower power state.
> 
> 
> For the PIONEER optical drives however, we don't seem to get any "Features: "
> print at all. My guess is that the support neither HIPM nor DIPM, and thus
> stops the whole system from entering a lower power state.
> 
> 
> What I think happens:
> 
> For a device that does not support HIPM, we will clear flag ATA_LPM_HIPM from
> hints:
> https://github.com/torvalds/linux/blob/v6.19-rc4/drivers/ata/libata-eh.c#L2168-L2169
> 
> If flag ATA_LPM_HIPM is not set, we will disable HIPM (ALPE) for the port:
> https://github.com/torvalds/linux/blob/v6.19-rc4/drivers/ata/libahci.c#L835-L847
> 
> Thus the HBA will never tell the device to enter a low power state.
> (And if DIPM is not supported, the device itself will never enter a low
> power state.)
> 
> 
> Questions:
> -Why was this not a problem on older kernels?
> 
> If you are willing try some patches from us, we could give you a patch with more
> debug prints, to show for each device, when we {enable,disable} {HIPM,DIPM}.
> 
> 
> -Why don't we get the Features: print for ATAPI devices?
> 
> We probably need something like this:

Before doing this, one simple test to do would be to unplug the 2 ATAPI devices
(CD/DVD drives) and reboot to see if LPM works then.

> 
> diff --git a/drivers/ata/libata-core.c b/drivers/ata/libata-core.c
> index fc250b494ce3..a8af200ac24f 100644
> --- a/drivers/ata/libata-core.c
> +++ b/drivers/ata/libata-core.c
> @@ -3161,11 +3161,13 @@ int ata_dev_configure(struct ata_device *dev)
>                 /* print device info to dmesg */
>                 if (print_info)
>                         ata_dev_info(dev,
> -                                    "ATAPI: %s, %s, max %s%s%s%s\n",
> +                                    "ATAPI: %s, %s, max %s%s%s%s%s%s\n",
>                                      modelbuf, fwrevbuf,
>                                      ata_mode_string(xfer_mask),
>                                      cdb_intr_string, atapi_an_string,
> -                                    dma_dir_string);
> +                                    dma_dir_string,
> +                                    ata_id_has_hipm(dev->id) ? " HIPM" : "",
> +                                    ata_id_has_dipm(dev->id) ? " DIPM" : "");
>         }
>  
>         /* determine max_sectors */
> 
> 
> 
> Kind regards,
> Niklas


-- 
Damien Le Moal
Western Digital Research