ata: ahci: Do not enable LPM if no LPM states are supported by the HBA
Subsystems (2)
Niklas Cassel
maintainerPATCH v2 1/1 ata: ahci: Do not enable LPM if no LPM states are supported by the HBA
LPM consists of HIPM (host initiated power management) and DIPM (device initiated power management).
ata_eh_set_lpm() will only enable HIPM if both the HBA and the device supports it.
However, DIPM will be enabled as long as the device supports it. The HBA will later reject the device's request to enter a power state that it does not support (Slumber/Partial/DevSleep) (DevSleep is never initiated by the device).
For a HBA that doesn't support any LPM states, simply don't set a LPM policy such that all the HIPM/DIPM probing/enabling will be skipped.
Not enabling HIPM or DIPM in the first place is safer than relying on the device following the AHCI specification and respecting the NAK. (There are comments in the code that some devices misbehave when receiving a NAK.)
Performing this check in ahci_update_initial_lpm_policy() also has the advantage that a HBA that doesn't support any LPM states will take the exact same code paths as a port that is external/hot plug capable.
Side note: the port in ata_port_dbg() has not been given a unique id yet, but this is not overly important as the debug print is disabled unless explicitly enabled using dynamic debug. A follow-up series will make sure that the unique id assignment will be done earlier. For now, the important thing is that the function returns before setting the LPM policy.
---
Changes since v1: Add debug print as suggested by Mika.
drivers/ata/ahci.c | 8 ++++++++ 1 file changed, 8 insertions(+)
drivers/ata/ahci.c +8 -0
@@ -1735,6 +1735,14 @@ static void ahci_update_initial_lpm_policy(struct ata_port *ap)if (ap->pflags &ATA_PFLAG_EXTERNAL )return ; +/* If no LPM states are supported by the HBA, do not bother with LPM */ +if ((ap->host->flags &ATA_HOST_NO_PART ) && + (ap->host->flags &ATA_HOST_NO_SSC ) && + (ap->host->flags &ATA_HOST_NO_DEVSLP )) { +ata_port_dbg (ap,"no LPM states supported, not enabling LPM\n" ); +return ; + } +/* user modified policy via module param */ if (mobile_lpm_policy !=-1 ) { policy = mobile_lpm_policy;
Signature · 1 line
2.45.2
Quoted 29 lines · Niklas Cassel
LPM consists of HIPM (host initiated power management) and DIPM (device initiated power management).
ata_eh_set_lpm()will only enable HIPM if both the HBA and the device supports it.However, DIPM will be enabled as long as the device supports it. The HBA will later reject the device's request to enter a power state that it does not support (Slumber/Partial/DevSleep) (DevSleep is never initiated by the device).
For a HBA that doesn't support any LPM states, simply don't set a LPM policy such that all the HIPM/DIPM probing/enabling will be skipped.
Not enabling HIPM or DIPM in the first place is safer than relying on the device following the AHCI specification and respecting the NAK. (There are comments in the code that some devices misbehave when receiving a NAK.)
Performing this check in
ahci_update_initial_lpm_policy()also has the advantage that a HBA that doesn't support any LPM states will take the exact same code paths as a port that is external/hot plug capable.Side note: the port in
ata_port_dbg()has not been given a unique id yet, but this is not overly important as the debug print is disabled unless explicitly enabled using dynamic debug. A follow-up series will make sure that the unique id assignment will be done earlier. For now, the important thing is that the function returns before setting the LPM policy.
Quoted 6 lines
Fixes 7627a0edef54 ("ata: ahci: Drop low power policy board type")Signed-off-by Niklas Cassel <[email protected]>---
Changes since v1: Add debug print as suggested by Mika.
Quoted 22 lines
drivers/ata/ahci.c | 8 ++++++++ 1 file changed, 8 insertions(+)drivers/ata/ahci.c +8
-0@@ -1735,6 +1735,14 @@ static void ahci_update_initial_lpm_policy(struct ata_port *ap)if (ap->pflags &ATA_PFLAG_EXTERNAL )return ; +/* If no LPM states are supported by the HBA, do not bother with LPM */ +if ((ap->host->flags &ATA_HOST_NO_PART ) && + (ap->host->flags &ATA_HOST_NO_SSC ) && + (ap->host->flags &ATA_HOST_NO_DEVSLP )) { +ata_port_dbg (ap,"no LPM states supported, not enabling LPM\n" ); +return ; + } +/* user modified policy via module param */ if (mobile_lpm_policy !=-1 ) { policy = mobile_lpm_policy;
Quoted 29 lines · Niklas Cassel
LPM consists of HIPM (host initiated power management) and DIPM (device initiated power management).
ata_eh_set_lpm()will only enable HIPM if both the HBA and the device supports it.However, DIPM will be enabled as long as the device supports it. The HBA will later reject the device's request to enter a power state that it does not support (Slumber/Partial/DevSleep) (DevSleep is never initiated by the device).
For a HBA that doesn't support any LPM states, simply don't set a LPM policy such that all the HIPM/DIPM probing/enabling will be skipped.
Not enabling HIPM or DIPM in the first place is safer than relying on the device following the AHCI specification and respecting the NAK. (There are comments in the code that some devices misbehave when receiving a NAK.)
Performing this check in
ahci_update_initial_lpm_policy()also has the advantage that a HBA that doesn't support any LPM states will take the exact same code paths as a port that is external/hot plug capable.Side note: the port in
ata_port_dbg()has not been given a unique id yet, but this is not overly important as the debug print is disabled unless explicitly enabled using dynamic debug. A follow-up series will make sure that the unique id assignment will be done earlier. For now, the important thing is that the function returns before setting the LPM policy.
Quoted 3 lines
Fixes 7627a0edef54 ("ata: ahci: Drop low power policy board type")Signed-off-by Niklas Cassel <[email protected]>
Quoted 43 lines · Niklas Cassel
LPM consists of HIPM (host initiated power management) and DIPM (device initiated power management).
ata_eh_set_lpm()will only enable HIPM if both the HBA and the device supports it.However, DIPM will be enabled as long as the device supports it. The HBA will later reject the device's request to enter a power state that it does not support (Slumber/Partial/DevSleep) (DevSleep is never initiated by the device).
For a HBA that doesn't support any LPM states, simply don't set a LPM policy such that all the HIPM/DIPM probing/enabling will be skipped.
Not enabling HIPM or DIPM in the first place is safer than relying on the device following the AHCI specification and respecting the NAK. (There are comments in the code that some devices misbehave when receiving a NAK.)
Performing this check in
ahci_update_initial_lpm_policy()also has the advantage that a HBA that doesn't support any LPM states will take the exact same code paths as a port that is external/hot plug capable.Side note: the port in
ata_port_dbg()has not been given a unique id yet, but this is not overly important as the debug print is disabled unless explicitly enabled using dynamic debug. A follow-up series will make sure that the unique id assignment will be done earlier. For now, the important thing is that the function returns before setting the LPM policy.Fixes 7627a0edef54 ("ata: ahci: Drop low power policy board type")Signed-off-by Niklas Cassel <[email protected]>---
Changes since v1: Add debug print as suggested by Mika.drivers/ata/ahci.c | 8 ++++++++ 1 file changed, 8 insertions(+)drivers/ata/ahci.c +0
-0
drivers/ata/ahci.c static void ahci_update_initial_lpm_policy(struct ata_port *ap) · line 1741
Quoted 7 lines
drivers/ata/ahci.c +4
-0@@ -1735,6 +1735,14 @@ static void ahci_update_initial_lpm_policy(struct ata_port *ap)if (ap->pflags &ATA_PFLAG_EXTERNAL )return ; +/* If no LPM states are supported by the HBA, do not bother with LPM */ +if ((ap->host->flags &ATA_HOST_NO_PART ) && + (ap->host->flags &ATA_HOST_NO_SSC ) && + (ap->host->flags &ATA_HOST_NO_DEVSLP )) {
Nit: Maybe:
#define ATA_HOST_NO_LPM \
(ATA_HOST_NO_PART | ATA_HOST_NO_SSC | ATA_HOST_NO_DEVSLP)
and then the if becomes:
if ((ap->host->flags & ATA_HOST_NO_LPM) == ATA_HOST_NO_LPM) {But no strong feelings about it. So:
Quoted 7 lines
drivers/ata/ahci.c +4
-0+ata_port_dbg (ap,"no LPM states supported, not enabling LPM\n" ); +return ; + } +/* user modified policy via module param */ if (mobile_lpm_policy !=-1 ) { policy = mobile_lpm_policy;
Signature · 2 lines
Damien Le Moal Western Digital Research
Niklas Cassel
maintainerQuoted 57 lines · Damien Le Moal
Quoted 43 lines · Niklas Cassel
LPM consists of HIPM (host initiated power management) and DIPM (device initiated power management).
ata_eh_set_lpm()will only enable HIPM if both the HBA and the device supports it.However, DIPM will be enabled as long as the device supports it. The HBA will later reject the device's request to enter a power state that it does not support (Slumber/Partial/DevSleep) (DevSleep is never initiated by the device).
For a HBA that doesn't support any LPM states, simply don't set a LPM policy such that all the HIPM/DIPM probing/enabling will be skipped.
Not enabling HIPM or DIPM in the first place is safer than relying on the device following the AHCI specification and respecting the NAK. (There are comments in the code that some devices misbehave when receiving a NAK.)
Performing this check in
ahci_update_initial_lpm_policy()also has the advantage that a HBA that doesn't support any LPM states will take the exact same code paths as a port that is external/hot plug capable.Side note: the port in
ata_port_dbg()has not been given a unique id yet, but this is not overly important as the debug print is disabled unless explicitly enabled using dynamic debug. A follow-up series will make sure that the unique id assignment will be done earlier. For now, the important thing is that the function returns before setting the LPM policy.Fixes 7627a0edef54 ("ata: ahci: Drop low power policy board type")Signed-off-by Niklas Cassel <[email protected]>---
Changes since v1: Add debug print as suggested by Mika.drivers/ata/ahci.c | 8 ++++++++ 1 file changed, 8 insertions(+)drivers/ata/ahci.c +0
-0drivers/ata/ahci.c static void ahci_update_initial_lpm_policy(struct ata_port *ap) · line 1741
Quoted 7 lines
drivers/ata/ahci.c +4
-0@@ -1735,6 +1735,14 @@ static void ahci_update_initial_lpm_policy(struct ata_port *ap)if (ap->pflags &ATA_PFLAG_EXTERNAL )return ; +/* If no LPM states are supported by the HBA, do not bother with LPM */ +if ((ap->host->flags &ATA_HOST_NO_PART ) && + (ap->host->flags &ATA_HOST_NO_SSC ) && + (ap->host->flags &ATA_HOST_NO_DEVSLP )) {Nit: Maybe:
#define
ATA_HOST_NO_LPM\
(ATA_HOST_NO_PART|ATA_HOST_NO_SSC|ATA_HOST_NO_DEVSLP)
Quoted 7 lines
and then the if becomes:
if ((ap->host->flags & ATA_HOST_NO_LPM) == ATA_HOST_NO_LPM) {But no strong feelings about it. So:
Reviewed-by Damien Le Moal <[email protected]>
Thank you for the R-b and your suggestion.
Personally, I do not think that your suggestion is significantly easier to read than what is already there (especially with the comment to give context).
My brain always has to read a:
if ((foo & bar) == bar)
twice anyway.
I guess a:
if (!ata_host_has_lpm(ap->host))
would be clearer, but considering that we wouldn't be able to use this helper function anywhere else in the libata subsystem, I'm not sure if it is worth it, so I will just apply this patch as is.
Kind regards,
Niklas
Quoted 6 lines · Niklas Cassel
LPM consists of HIPM (host initiated power management) and DIPM (device initiated power management).
ata_eh_set_lpm()will only enable HIPM if both the HBA and the device supports it.
Quoted 6 lines
However, DIPM will be enabled as long as the device supports it. The HBA will later reject the device's request to enter a power state that it does not support (Slumber/Partial/DevSleep) (DevSleep is never initiated by the device).
[...]
Applied to libata/linux.git (for-6.10-fixes), thanks!
[1/1] ata: ahci: Do not enable LPM if no LPM states are supported by the HBA
https://git.kernel.org/libata/linux/c/fa997b05
Kind regards,
Signature · 1 line
Niklas Cassel <[email protected]>