NVMeFix - Use to force a low power state when entering sleep?

[herb2k]

I was reviewing debug info and noticed these entries when entering sleep, is it possible to force the change to a low power state on entering sleep, similar to Apple SSDs?

While in sleep - before the device enters standby, I notice the SSD stays powered up - ie it's warm and drawing current the whole time, which leads me to believe it's not being idled properly?

2020-06-14 23:53:52.981921-0700  localhost kernel[0]: (Lilu) NVMeFix        pm: @ (DBG) powerStateDidChangeTo 0x0
2020-06-14 23:53:52.981929-0700  localhost kernel[0]: (Lilu) NVMeFix        pm: @ (DBG) Ignoring transition to non-usable state 0x0

Also grabbed the APST details from linux for comparison:

get-feature:0xc (Autonomous Power State Transition), Current value:0x000001
Autonomous Power State Transition Enable (APSTE): Enabled
Auto PST Entries .................
Entry[ 0]  
.................
Idle Time Prior to Transition (ITPT): 500 ms
Idle Transition Power State   (ITPS): 3
.................
Entry[ 1]  
.................
Idle Time Prior to Transition (ITPT): 500 ms
Idle Transition Power State   (ITPS): 3
.................
Entry[ 2]  
.................
Idle Time Prior to Transition (ITPT): 500 ms
Idle Transition Power State   (ITPS): 3
.................
Entry[ 3]  
.................
Idle Time Prior to Transition (ITPT): 700 ms
Idle Transition Power State   (ITPS): 4
.................
Entry[ 4]  
.................
Idle Time Prior to Transition (ITPT): 0 ms
Idle Transition Power State   (ITPS): 0
.................
Entry[ 5]  
.................
Idle Time Prior to Transition (ITPT): 0 ms
Idle Transition Power State   (ITPS): 0

[07151129]

The log message just states that the controller is not waking up from sleep, so APST re-enable is not needed. There is nothing wrong with this.

For system sleep, ACPI S1-S4 low-power states should be used that disable the controller. There is also PCI ASPM support, but you may need to inject an extra property to force it. There was a recent broadcom-related issue that discussed this in detail, see #794. Note that some controllers become unreliable with ASPM, so there may be a potential risk of data loss.

In addition to PCI-level power management, the NVMe specification also has support for the following power management modes:

• Active host-driven state switching, i.e. when the host explicitly tells the controller to switch to a certain state.

• Autonomous power state transition (APST), i.e. when the controller decides to switch to a certain state on its own.

The latter is the preferred method of NVMeFix because it is a lot easier to integrate into IONVMeFamily, and it should perform better, as the controller can accurately keep track of the I/O performed. Exactly the same logic as in the Linux NVMe driver is used for APST config, resulting in exactly the same table as you posted.

IONVMeFamily exposes three power states (indexed 0 to 2): off, low-power and active. The log says that state 0 has been entered, so the controller should be off.

To summarise, this issue has nothing to do with APST; rather, it is a problem at PCI level that NVMeFix was not attempting to solve to begin with, because Apple drivers should be adequate for PCI-level power management.

Incidentally, you may want to try hibernation to see if that helps.

[herb2k]

Thank you! This makes a lot more sense now, appreciate taking the time to explain to me!