RE: [PATCH V8 4/5] PCI/ASPM: save power on values during bridge init

[Patel, Mayurkumar]

Hi Bjorn

Thanks a lot for your reply and explanations. Sorry for my late reply due to 
some other emergencies.
>
>On Tue, May 02, 2017 at 12:02:53PM +, Patel, Mayurkumar wrote:
>> >On Fri, Apr 21, 2017 at 2:46 AM, Patel, Mayurkumar
>> > wrote:
>> >>>On 4/17/2017 12:38 PM, Bjorn Helgaas wrote:
>> > Like you said, what do we do by default is the question. Should we opt
>> > for safe like we are doing, or try to save some power.
>>  I think safety is paramount.  Every user should be able to boot safely
>>  without any kernel parameters.  We don't want users to have a problem
>>  booting and then have to search for a workaround like booting with
>>  "pcie_aspm=off".  Most users will never do that.
>> >>>
>> >>>OK, no problem with leaving the behavior as it is.
>> >>>
>> >>>My initial approach was #2. We knew this way that user had full control
>> >>>over the ASPM policy by changing the BIOS option. Then, Mayurkumar
>> >>>complained that ASPM is not enabled following a hotplug insertion to an
>> >>>empty slot. That's when I switched to #3 as it sounded like a good thing
>> >>>to have for us.
>> >>>
>>  Here's a long-term strawman proposal, see what you think:
>> 
>>    - Deprecate CONFIG_PCIEASPM_DEFAULT, CONFIG_PCIEASPM_POWERSAVE, etc.
>>    - Default aspm_policy is POLICY_DEFAULT always.
>>    - POLICY_DEFAULT means Linux doesn't touch anything: if BIOS enabled
>>  ASPM, we leave it that way; we leave ASPM disabled on hot-added
>>  devices.
>> >>>
>> >> I am also ok with leaving the same behavior as now.  But still
>> >> following is something open I feel besides, Which may be there in
>> >> your comments redundantly.  The current problem is,
>> >> pcie_aspm_exit_link_state() disables the ASPM configuration even
>> >> if POLICY_DEFAULT was set.
>> >
>> >We call pcie_aspm_exit_link_state() when removing an endpoint.  When
>> >we remove an endpoint, I think disabling ASPM is the right thing to
>> >do.  The spec (PCIe r3.1, sec 5.4.1.3) says "Software must not enable
>> >L0s in either direction on a given Link unless components on both
>> >sides of the Link each support L0s; otherwise, the result is
>> >undefined."
>>
>> Yes, you are right and per spec also it makes sense that ASPM needs
>> to be disabled.  But, if POLICY_DEFAULT is set then, shouldn't BIOS
>> take care of disabling ASPM?
>
>No, I don't think so.  POLICY_DEFAULT is a Linux thing and BIOS
>doesn't know anything about it.
>
>ASPM can be configured by BIOS before handoff to Linux, but after
>handoff it should be managed either entirely by BIOS or entirely by
>Linux.  If BIOS wants to retain ASPM control, it would have to tell
>the OS *not* to use ASPM, and it would have to use ACPI hotplug.  In
>this case, POLICY_DEFAULT is irrelevant because Linux shouldn't do
>anything with ASPM.
>
>But normally BIOS allows Linux to control ASPM, and we would use
>native PCIe hotplug (pciehp) instead of ACPI hotplug, and BIOS has no
>opportunity to enable or disable ASPM on hotplug events.
>

BIOS that I am having, has an SMI handler Which gets triggered upon
Hotplug (Data Link Layer State Changed) Interrupt Which configures ASPM L1/L1SS 
in BIOS
and We are still using Native Hotplug driver. Sounds like BIOS we have in our 
System,
does not inform OS that it wants control ASPM.

>> >> I am seeing already following problem(or may be influence) with
>> >> it. The Endpoint I have does not have does not have "Presence
>> >> detect change" mechanism. Hot plug is working with Link status
>> >> events.  When link is in L1 or L1SS and if EP is powered off, no
>> >> Link status change event are triggered (It might be the expected
>> >> behavior in L1 or L1SS).  When next time EP is powered on there
>> >> are link down and link up events coming one after other. BIOS
>> >> enables ASPM on Root port and Endpoint, but while processing link
>> >> status down, pcie_aspm_exit_link_state() clears the ASPM already
>> >> which were enabled by BIOS.  If we want to follow above approach
>> >> then shall we consider having something similar as following?
>> >
>> >The proposal was to leave ASPM disabled on hot-added devices.  If
>> >the endpoint was powered off and powered back on again, I think
>> >that device looks like a hot-added device, doesn't it?
>>
>> Yes, it is hot-added device. Also, I understand, for POLICY_DEFAULT,
>> OS would/should not touch ASPM(enable/disable), but BIOS could still
>> (enable/disable), right?
>
>Maybe I'm misunderstanding your question.  There are two questions
>here:
>
>  1) Does the BIOS allow Linux to manage ASPM?
>
>  2) If Linux does manage ASPM, what policy does it use?
>
>If BIOS doesn't allow Linux to manage ASPM, POLICY_DEFAULT is
>irrelevant.  If BIOS does allow Linux to manage ASPM, POLICY_DEFAULT
>means Linux should use the settings made by BIOS.  The user could
>select a different policy, and then Linux would change the ASPM
>configuration accordingly.
>

RE: [PATCH V8 4/5] PCI/ASPM: save power on values during bridge init

[Patel, Mayurkumar]

Hi Bjorn

Thanks a lot for your reply and explanations. Sorry for my late reply due to 
some other emergencies.
>
>On Tue, May 02, 2017 at 12:02:53PM +, Patel, Mayurkumar wrote:
>> >On Fri, Apr 21, 2017 at 2:46 AM, Patel, Mayurkumar
>> > wrote:
>> >>>On 4/17/2017 12:38 PM, Bjorn Helgaas wrote:
>> > Like you said, what do we do by default is the question. Should we opt
>> > for safe like we are doing, or try to save some power.
>>  I think safety is paramount.  Every user should be able to boot safely
>>  without any kernel parameters.  We don't want users to have a problem
>>  booting and then have to search for a workaround like booting with
>>  "pcie_aspm=off".  Most users will never do that.
>> >>>
>> >>>OK, no problem with leaving the behavior as it is.
>> >>>
>> >>>My initial approach was #2. We knew this way that user had full control
>> >>>over the ASPM policy by changing the BIOS option. Then, Mayurkumar
>> >>>complained that ASPM is not enabled following a hotplug insertion to an
>> >>>empty slot. That's when I switched to #3 as it sounded like a good thing
>> >>>to have for us.
>> >>>
>>  Here's a long-term strawman proposal, see what you think:
>> 
>>    - Deprecate CONFIG_PCIEASPM_DEFAULT, CONFIG_PCIEASPM_POWERSAVE, etc.
>>    - Default aspm_policy is POLICY_DEFAULT always.
>>    - POLICY_DEFAULT means Linux doesn't touch anything: if BIOS enabled
>>  ASPM, we leave it that way; we leave ASPM disabled on hot-added
>>  devices.
>> >>>
>> >> I am also ok with leaving the same behavior as now.  But still
>> >> following is something open I feel besides, Which may be there in
>> >> your comments redundantly.  The current problem is,
>> >> pcie_aspm_exit_link_state() disables the ASPM configuration even
>> >> if POLICY_DEFAULT was set.
>> >
>> >We call pcie_aspm_exit_link_state() when removing an endpoint.  When
>> >we remove an endpoint, I think disabling ASPM is the right thing to
>> >do.  The spec (PCIe r3.1, sec 5.4.1.3) says "Software must not enable
>> >L0s in either direction on a given Link unless components on both
>> >sides of the Link each support L0s; otherwise, the result is
>> >undefined."
>>
>> Yes, you are right and per spec also it makes sense that ASPM needs
>> to be disabled.  But, if POLICY_DEFAULT is set then, shouldn't BIOS
>> take care of disabling ASPM?
>
>No, I don't think so.  POLICY_DEFAULT is a Linux thing and BIOS
>doesn't know anything about it.
>
>ASPM can be configured by BIOS before handoff to Linux, but after
>handoff it should be managed either entirely by BIOS or entirely by
>Linux.  If BIOS wants to retain ASPM control, it would have to tell
>the OS *not* to use ASPM, and it would have to use ACPI hotplug.  In
>this case, POLICY_DEFAULT is irrelevant because Linux shouldn't do
>anything with ASPM.
>
>But normally BIOS allows Linux to control ASPM, and we would use
>native PCIe hotplug (pciehp) instead of ACPI hotplug, and BIOS has no
>opportunity to enable or disable ASPM on hotplug events.
>

BIOS that I am having, has an SMI handler Which gets triggered upon
Hotplug (Data Link Layer State Changed) Interrupt Which configures ASPM L1/L1SS 
in BIOS
and We are still using Native Hotplug driver. Sounds like BIOS we have in our 
System,
does not inform OS that it wants control ASPM.

>> >> I am seeing already following problem(or may be influence) with
>> >> it. The Endpoint I have does not have does not have "Presence
>> >> detect change" mechanism. Hot plug is working with Link status
>> >> events.  When link is in L1 or L1SS and if EP is powered off, no
>> >> Link status change event are triggered (It might be the expected
>> >> behavior in L1 or L1SS).  When next time EP is powered on there
>> >> are link down and link up events coming one after other. BIOS
>> >> enables ASPM on Root port and Endpoint, but while processing link
>> >> status down, pcie_aspm_exit_link_state() clears the ASPM already
>> >> which were enabled by BIOS.  If we want to follow above approach
>> >> then shall we consider having something similar as following?
>> >
>> >The proposal was to leave ASPM disabled on hot-added devices.  If
>> >the endpoint was powered off and powered back on again, I think
>> >that device looks like a hot-added device, doesn't it?
>>
>> Yes, it is hot-added device. Also, I understand, for POLICY_DEFAULT,
>> OS would/should not touch ASPM(enable/disable), but BIOS could still
>> (enable/disable), right?
>
>Maybe I'm misunderstanding your question.  There are two questions
>here:
>
>  1) Does the BIOS allow Linux to manage ASPM?
>
>  2) If Linux does manage ASPM, what policy does it use?
>
>If BIOS doesn't allow Linux to manage ASPM, POLICY_DEFAULT is
>irrelevant.  If BIOS does allow Linux to manage ASPM, POLICY_DEFAULT
>means Linux should use the settings made by BIOS.  The user could
>select a different policy, and then Linux would change the ASPM
>configuration accordingly.
>

Ok understood.

>> 

Re: [PATCH V8 4/5] PCI/ASPM: save power on values during bridge init

[Bjorn Helgaas]

On Tue, May 02, 2017 at 12:02:53PM +, Patel, Mayurkumar wrote:
> >On Fri, Apr 21, 2017 at 2:46 AM, Patel, Mayurkumar
> > wrote:
> >>>On 4/17/2017 12:38 PM, Bjorn Helgaas wrote:
> > Like you said, what do we do by default is the question. Should we opt
> > for safe like we are doing, or try to save some power.
>  I think safety is paramount.  Every user should be able to boot safely
>  without any kernel parameters.  We don't want users to have a problem
>  booting and then have to search for a workaround like booting with
>  "pcie_aspm=off".  Most users will never do that.
> >>>
> >>>OK, no problem with leaving the behavior as it is.
> >>>
> >>>My initial approach was #2. We knew this way that user had full control
> >>>over the ASPM policy by changing the BIOS option. Then, Mayurkumar
> >>>complained that ASPM is not enabled following a hotplug insertion to an
> >>>empty slot. That's when I switched to #3 as it sounded like a good thing
> >>>to have for us.
> >>>
>  Here's a long-term strawman proposal, see what you think:
> 
>    - Deprecate CONFIG_PCIEASPM_DEFAULT, CONFIG_PCIEASPM_POWERSAVE, etc.
>    - Default aspm_policy is POLICY_DEFAULT always.
>    - POLICY_DEFAULT means Linux doesn't touch anything: if BIOS enabled
>  ASPM, we leave it that way; we leave ASPM disabled on hot-added
>  devices.
> >>>
> >> I am also ok with leaving the same behavior as now.  But still
> >> following is something open I feel besides, Which may be there in
> >> your comments redundantly.  The current problem is,
> >> pcie_aspm_exit_link_state() disables the ASPM configuration even
> >> if POLICY_DEFAULT was set.
> >
> >We call pcie_aspm_exit_link_state() when removing an endpoint.  When
> >we remove an endpoint, I think disabling ASPM is the right thing to
> >do.  The spec (PCIe r3.1, sec 5.4.1.3) says "Software must not enable
> >L0s in either direction on a given Link unless components on both
> >sides of the Link each support L0s; otherwise, the result is
> >undefined."
> 
> Yes, you are right and per spec also it makes sense that ASPM needs
> to be disabled.  But, if POLICY_DEFAULT is set then, shouldn't BIOS
> take care of disabling ASPM?

No, I don't think so.  POLICY_DEFAULT is a Linux thing and BIOS
doesn't know anything about it.

ASPM can be configured by BIOS before handoff to Linux, but after
handoff it should be managed either entirely by BIOS or entirely by
Linux.  If BIOS wants to retain ASPM control, it would have to tell
the OS *not* to use ASPM, and it would have to use ACPI hotplug.  In
this case, POLICY_DEFAULT is irrelevant because Linux shouldn't do
anything with ASPM.

But normally BIOS allows Linux to control ASPM, and we would use
native PCIe hotplug (pciehp) instead of ACPI hotplug, and BIOS has no
opportunity to enable or disable ASPM on hotplug events.

> >> I am seeing already following problem(or may be influence) with
> >> it. The Endpoint I have does not have does not have "Presence
> >> detect change" mechanism. Hot plug is working with Link status
> >> events.  When link is in L1 or L1SS and if EP is powered off, no
> >> Link status change event are triggered (It might be the expected
> >> behavior in L1 or L1SS).  When next time EP is powered on there
> >> are link down and link up events coming one after other. BIOS
> >> enables ASPM on Root port and Endpoint, but while processing link
> >> status down, pcie_aspm_exit_link_state() clears the ASPM already
> >> which were enabled by BIOS.  If we want to follow above approach
> >> then shall we consider having something similar as following?
> >
> >The proposal was to leave ASPM disabled on hot-added devices.  If
> >the endpoint was powered off and powered back on again, I think
> >that device looks like a hot-added device, doesn't it?
> 
> Yes, it is hot-added device. Also, I understand, for POLICY_DEFAULT,
> OS would/should not touch ASPM(enable/disable), but BIOS could still
> (enable/disable), right?

Maybe I'm misunderstanding your question.  There are two questions
here:

  1) Does the BIOS allow Linux to manage ASPM?

  2) If Linux does manage ASPM, what policy does it use?

If BIOS doesn't allow Linux to manage ASPM, POLICY_DEFAULT is
irrelevant.  If BIOS does allow Linux to manage ASPM, POLICY_DEFAULT
means Linux should use the settings made by BIOS.  The user could
select a different policy, and then Linux would change the ASPM
configuration accordingly.

> Currently, what happens in my system is as following, (each 2nd
> power cycle/hotplug of Endpoint disables ASPM):
> 
> 
> First Power cycle (When ASPM L1 is already enabled): device gets
> powered off -> there are no Link status events, so no pcie hotplug
> interrupt and pcie_aspm_exit_link_state() triggered.  

If the Downstream Port leading to your Endpoint is hotplug capable,
doesn't the spec require that it can report link state changes (PCIe
r3.1, sec 7.8.6, 7.8.10, 7.8.11)?

> When 

Re: [PATCH V8 4/5] PCI/ASPM: save power on values during bridge init

[Bjorn Helgaas]

On Tue, May 02, 2017 at 12:02:53PM +, Patel, Mayurkumar wrote:
> >On Fri, Apr 21, 2017 at 2:46 AM, Patel, Mayurkumar
> > wrote:
> >>>On 4/17/2017 12:38 PM, Bjorn Helgaas wrote:
> > Like you said, what do we do by default is the question. Should we opt
> > for safe like we are doing, or try to save some power.
>  I think safety is paramount.  Every user should be able to boot safely
>  without any kernel parameters.  We don't want users to have a problem
>  booting and then have to search for a workaround like booting with
>  "pcie_aspm=off".  Most users will never do that.
> >>>
> >>>OK, no problem with leaving the behavior as it is.
> >>>
> >>>My initial approach was #2. We knew this way that user had full control
> >>>over the ASPM policy by changing the BIOS option. Then, Mayurkumar
> >>>complained that ASPM is not enabled following a hotplug insertion to an
> >>>empty slot. That's when I switched to #3 as it sounded like a good thing
> >>>to have for us.
> >>>
>  Here's a long-term strawman proposal, see what you think:
> 
>    - Deprecate CONFIG_PCIEASPM_DEFAULT, CONFIG_PCIEASPM_POWERSAVE, etc.
>    - Default aspm_policy is POLICY_DEFAULT always.
>    - POLICY_DEFAULT means Linux doesn't touch anything: if BIOS enabled
>  ASPM, we leave it that way; we leave ASPM disabled on hot-added
>  devices.
> >>>
> >> I am also ok with leaving the same behavior as now.  But still
> >> following is something open I feel besides, Which may be there in
> >> your comments redundantly.  The current problem is,
> >> pcie_aspm_exit_link_state() disables the ASPM configuration even
> >> if POLICY_DEFAULT was set.
> >
> >We call pcie_aspm_exit_link_state() when removing an endpoint.  When
> >we remove an endpoint, I think disabling ASPM is the right thing to
> >do.  The spec (PCIe r3.1, sec 5.4.1.3) says "Software must not enable
> >L0s in either direction on a given Link unless components on both
> >sides of the Link each support L0s; otherwise, the result is
> >undefined."
> 
> Yes, you are right and per spec also it makes sense that ASPM needs
> to be disabled.  But, if POLICY_DEFAULT is set then, shouldn't BIOS
> take care of disabling ASPM?

No, I don't think so.  POLICY_DEFAULT is a Linux thing and BIOS
doesn't know anything about it.

ASPM can be configured by BIOS before handoff to Linux, but after
handoff it should be managed either entirely by BIOS or entirely by
Linux.  If BIOS wants to retain ASPM control, it would have to tell
the OS *not* to use ASPM, and it would have to use ACPI hotplug.  In
this case, POLICY_DEFAULT is irrelevant because Linux shouldn't do
anything with ASPM.

But normally BIOS allows Linux to control ASPM, and we would use
native PCIe hotplug (pciehp) instead of ACPI hotplug, and BIOS has no
opportunity to enable or disable ASPM on hotplug events.

> >> I am seeing already following problem(or may be influence) with
> >> it. The Endpoint I have does not have does not have "Presence
> >> detect change" mechanism. Hot plug is working with Link status
> >> events.  When link is in L1 or L1SS and if EP is powered off, no
> >> Link status change event are triggered (It might be the expected
> >> behavior in L1 or L1SS).  When next time EP is powered on there
> >> are link down and link up events coming one after other. BIOS
> >> enables ASPM on Root port and Endpoint, but while processing link
> >> status down, pcie_aspm_exit_link_state() clears the ASPM already
> >> which were enabled by BIOS.  If we want to follow above approach
> >> then shall we consider having something similar as following?
> >
> >The proposal was to leave ASPM disabled on hot-added devices.  If
> >the endpoint was powered off and powered back on again, I think
> >that device looks like a hot-added device, doesn't it?
> 
> Yes, it is hot-added device. Also, I understand, for POLICY_DEFAULT,
> OS would/should not touch ASPM(enable/disable), but BIOS could still
> (enable/disable), right?

Maybe I'm misunderstanding your question.  There are two questions
here:

  1) Does the BIOS allow Linux to manage ASPM?

  2) If Linux does manage ASPM, what policy does it use?

If BIOS doesn't allow Linux to manage ASPM, POLICY_DEFAULT is
irrelevant.  If BIOS does allow Linux to manage ASPM, POLICY_DEFAULT
means Linux should use the settings made by BIOS.  The user could
select a different policy, and then Linux would change the ASPM
configuration accordingly.

> Currently, what happens in my system is as following, (each 2nd
> power cycle/hotplug of Endpoint disables ASPM):
> 
> 
> First Power cycle (When ASPM L1 is already enabled): device gets
> powered off -> there are no Link status events, so no pcie hotplug
> interrupt and pcie_aspm_exit_link_state() triggered.  

If the Downstream Port leading to your Endpoint is hotplug capable,
doesn't the spec require that it can report link state changes (PCIe
r3.1, sec 7.8.6, 7.8.10, 7.8.11)?

> When the device gets powered on 

RE: [PATCH V8 4/5] PCI/ASPM: save power on values during bridge init

[Patel, Mayurkumar]

Hi Bjorn

>
>On Fri, Apr 21, 2017 at 2:46 AM, Patel, Mayurkumar
> wrote:
>> Hi Bjorn/Kaya,
>>
>>
>>>
>>>On 4/17/2017 12:38 PM, Bjorn Helgaas wrote:
> Like you said, what do we do by default is the question. Should we opt
> for safe like we are doing, or try to save some power.
 I think safety is paramount.  Every user should be able to boot safely
 without any kernel parameters.  We don't want users to have a problem
 booting and then have to search for a workaround like booting with
 "pcie_aspm=off".  Most users will never do that.

>>>
>>>OK, no problem with leaving the behavior as it is.
>>>
>>>My initial approach was #2. We knew this way that user had full control
>>>over the ASPM policy by changing the BIOS option. Then, Mayurkumar
>>>complained that ASPM is not enabled following a hotplug insertion to an
>>>empty slot. That's when I switched to #3 as it sounded like a good thing
>>>to have for us.
>>>
 Here's a long-term strawman proposal, see what you think:

   - Deprecate CONFIG_PCIEASPM_DEFAULT, CONFIG_PCIEASPM_POWERSAVE, etc.
   - Default aspm_policy is POLICY_DEFAULT always.
   - POLICY_DEFAULT means Linux doesn't touch anything: if BIOS enabled
 ASPM, we leave it that way; we leave ASPM disabled on hot-added
 devices.
>>>
>> I am also ok with leaving the same behavior as now.
>> But still following is something open I feel besides, Which may be there in 
>> your comments redundantly.
>> The current problem is, pcie_aspm_exit_link_state() disables the ASPM 
>> configuration even
>> if POLICY_DEFAULT was set.
>
>We call pcie_aspm_exit_link_state() when removing an endpoint.  When
>we remove an endpoint, I think disabling ASPM is the right thing to
>do.  The spec (PCIe r3.1, sec 5.4.1.3) says "Software must not enable
>L0s in either direction on a given Link unless components on both
>sides of the Link each support L0s; otherwise, the result is
>undefined."
>

Yes, you are right and per spec also it makes sense that ASPM needs to be 
disabled.
But, if POLICY_DEFAULT is set then, shouldn't BIOS take care of disabling ASPM?



>> I am seeing already following problem(or may be influence) with it. The 
>> Endpoint I have does not have
>> does not have "Presence detect change" mechanism. Hot plug is working with 
>> Link status events.
>> When link is in L1 or L1SS and if EP is powered off, no Link status change 
>> event are triggered (It might be
>> the expected behavior in L1 or L1SS).  When next time EP is powered on there 
>> are link down and
>> link up events coming one after other. BIOS enables ASPM on Root port and 
>> Endpoint, but while
>> processing link status down, pcie_aspm_exit_link_state() clears the ASPM 
>> already which were enabled by BIOS.
>> If we want to follow above approach then shall we consider having something 
>> similar as following?
>
>The proposal was to leave ASPM disabled on hot-added devices.  If the
>endpoint was powered off and powered back on again, I think that
>device looks like a hot-added device, doesn't it?
>

Yes, it is hot-added device. Also, I understand, for POLICY_DEFAULT, OS 
would/should not touch ASPM(enable/disable),
but BIOS could still (enable/disable), right?

Currently, what happens in my system is as following, (each 2nd power 
cycle/hotplug of Endpoint disables ASPM):


First Power cycle (When ASPM L1 is already enabled):
device gets powered off -> there are no Link status events, so no pcie hotplug 
interrupt and pcie_aspm_exit_link_state() triggered.
When the device gets powered on again -> Link down/Link up events are coming 
back to back. 
First Link down is served. (BIOS checks for the Link status and enables ASPM 
already, as the device is
actually powered back). OS calls pcie_aspm_exit_link_state() and ASPM gets 
disabled by OS.

Second Power cycle (When ASPM L1 is disabled after above):
device gets powered off -> there are link status events, pcie hotplug interrupt 
is triggered and pcie_aspm_exit_link_state() triggered.
OS disables ASPM. BIOS checks Link status and disables ASPM too.
When the device gets powered on -> BIOS enables ASPM and as this is pcie 
hotplug insertion, OS
does not interfere and we have ASPM enabled.

The above sequence happens each 2nd power cycle of the hotplug device.

So One could still argue if POLICY_DEFAULT is set, then why OS disables ASPM if 
it is not meant to touch configuration.
This is why I proposed following kind of change, so that OS would not touch 
ASPM, if POLICY_DEFAULT is set.
Also, With the below change, everything relies on BIOS for ASPM when 
POLICY_DEFAULT is set and I see above problem
gets resolved. Also, the existing ASPM behavior does not have impact, unless 
specific BIOS does not disable ASPM on
Root Port when device gets removed.



>Bjorn
Intel Deutschland GmbH
Registered Address: Am Campeon 10-12, 85579 Neubiberg, Germany
Tel: +49 89 99 8853-0, www.intel.de
Managing Directors: Christin 

RE: [PATCH V8 4/5] PCI/ASPM: save power on values during bridge init

[Patel, Mayurkumar]

Hi Bjorn

>
>On Fri, Apr 21, 2017 at 2:46 AM, Patel, Mayurkumar
> wrote:
>> Hi Bjorn/Kaya,
>>
>>
>>>
>>>On 4/17/2017 12:38 PM, Bjorn Helgaas wrote:
> Like you said, what do we do by default is the question. Should we opt
> for safe like we are doing, or try to save some power.
 I think safety is paramount.  Every user should be able to boot safely
 without any kernel parameters.  We don't want users to have a problem
 booting and then have to search for a workaround like booting with
 "pcie_aspm=off".  Most users will never do that.

>>>
>>>OK, no problem with leaving the behavior as it is.
>>>
>>>My initial approach was #2. We knew this way that user had full control
>>>over the ASPM policy by changing the BIOS option. Then, Mayurkumar
>>>complained that ASPM is not enabled following a hotplug insertion to an
>>>empty slot. That's when I switched to #3 as it sounded like a good thing
>>>to have for us.
>>>
 Here's a long-term strawman proposal, see what you think:

   - Deprecate CONFIG_PCIEASPM_DEFAULT, CONFIG_PCIEASPM_POWERSAVE, etc.
   - Default aspm_policy is POLICY_DEFAULT always.
   - POLICY_DEFAULT means Linux doesn't touch anything: if BIOS enabled
 ASPM, we leave it that way; we leave ASPM disabled on hot-added
 devices.
>>>
>> I am also ok with leaving the same behavior as now.
>> But still following is something open I feel besides, Which may be there in 
>> your comments redundantly.
>> The current problem is, pcie_aspm_exit_link_state() disables the ASPM 
>> configuration even
>> if POLICY_DEFAULT was set.
>
>We call pcie_aspm_exit_link_state() when removing an endpoint.  When
>we remove an endpoint, I think disabling ASPM is the right thing to
>do.  The spec (PCIe r3.1, sec 5.4.1.3) says "Software must not enable
>L0s in either direction on a given Link unless components on both
>sides of the Link each support L0s; otherwise, the result is
>undefined."
>

Yes, you are right and per spec also it makes sense that ASPM needs to be 
disabled.
But, if POLICY_DEFAULT is set then, shouldn't BIOS take care of disabling ASPM?



>> I am seeing already following problem(or may be influence) with it. The 
>> Endpoint I have does not have
>> does not have "Presence detect change" mechanism. Hot plug is working with 
>> Link status events.
>> When link is in L1 or L1SS and if EP is powered off, no Link status change 
>> event are triggered (It might be
>> the expected behavior in L1 or L1SS).  When next time EP is powered on there 
>> are link down and
>> link up events coming one after other. BIOS enables ASPM on Root port and 
>> Endpoint, but while
>> processing link status down, pcie_aspm_exit_link_state() clears the ASPM 
>> already which were enabled by BIOS.
>> If we want to follow above approach then shall we consider having something 
>> similar as following?
>
>The proposal was to leave ASPM disabled on hot-added devices.  If the
>endpoint was powered off and powered back on again, I think that
>device looks like a hot-added device, doesn't it?
>

Yes, it is hot-added device. Also, I understand, for POLICY_DEFAULT, OS 
would/should not touch ASPM(enable/disable),
but BIOS could still (enable/disable), right?

Currently, what happens in my system is as following, (each 2nd power 
cycle/hotplug of Endpoint disables ASPM):


First Power cycle (When ASPM L1 is already enabled):
device gets powered off -> there are no Link status events, so no pcie hotplug 
interrupt and pcie_aspm_exit_link_state() triggered.
When the device gets powered on again -> Link down/Link up events are coming 
back to back. 
First Link down is served. (BIOS checks for the Link status and enables ASPM 
already, as the device is
actually powered back). OS calls pcie_aspm_exit_link_state() and ASPM gets 
disabled by OS.

Second Power cycle (When ASPM L1 is disabled after above):
device gets powered off -> there are link status events, pcie hotplug interrupt 
is triggered and pcie_aspm_exit_link_state() triggered.
OS disables ASPM. BIOS checks Link status and disables ASPM too.
When the device gets powered on -> BIOS enables ASPM and as this is pcie 
hotplug insertion, OS
does not interfere and we have ASPM enabled.

The above sequence happens each 2nd power cycle of the hotplug device.

So One could still argue if POLICY_DEFAULT is set, then why OS disables ASPM if 
it is not meant to touch configuration.
This is why I proposed following kind of change, so that OS would not touch 
ASPM, if POLICY_DEFAULT is set.
Also, With the below change, everything relies on BIOS for ASPM when 
POLICY_DEFAULT is set and I see above problem
gets resolved. Also, the existing ASPM behavior does not have impact, unless 
specific BIOS does not disable ASPM on
Root Port when device gets removed.



>Bjorn
Intel Deutschland GmbH
Registered Address: Am Campeon 10-12, 85579 Neubiberg, Germany
Tel: +49 89 99 8853-0, www.intel.de
Managing Directors: Christin Eisenschmid, Christian Lamprechter

Re: [PATCH V8 4/5] PCI/ASPM: save power on values during bridge init

[Bjorn Helgaas]

On Fri, Apr 21, 2017 at 2:46 AM, Patel, Mayurkumar
 wrote:
> Hi Bjorn/Kaya,
>
>
>>
>>On 4/17/2017 12:38 PM, Bjorn Helgaas wrote:
 Like you said, what do we do by default is the question. Should we opt
 for safe like we are doing, or try to save some power.
>>> I think safety is paramount.  Every user should be able to boot safely
>>> without any kernel parameters.  We don't want users to have a problem
>>> booting and then have to search for a workaround like booting with
>>> "pcie_aspm=off".  Most users will never do that.
>>>
>>
>>OK, no problem with leaving the behavior as it is.
>>
>>My initial approach was #2. We knew this way that user had full control
>>over the ASPM policy by changing the BIOS option. Then, Mayurkumar
>>complained that ASPM is not enabled following a hotplug insertion to an
>>empty slot. That's when I switched to #3 as it sounded like a good thing
>>to have for us.
>>
>>> Here's a long-term strawman proposal, see what you think:
>>>
>>>   - Deprecate CONFIG_PCIEASPM_DEFAULT, CONFIG_PCIEASPM_POWERSAVE, etc.
>>>   - Default aspm_policy is POLICY_DEFAULT always.
>>>   - POLICY_DEFAULT means Linux doesn't touch anything: if BIOS enabled
>>> ASPM, we leave it that way; we leave ASPM disabled on hot-added
>>> devices.
>>
> I am also ok with leaving the same behavior as now.
> But still following is something open I feel besides, Which may be there in 
> your comments redundantly.
> The current problem is, pcie_aspm_exit_link_state() disables the ASPM 
> configuration even
> if POLICY_DEFAULT was set.

We call pcie_aspm_exit_link_state() when removing an endpoint.  When
we remove an endpoint, I think disabling ASPM is the right thing to
do.  The spec (PCIe r3.1, sec 5.4.1.3) says "Software must not enable
L0s in either direction on a given Link unless components on both
sides of the Link each support L0s; otherwise, the result is
undefined."

> I am seeing already following problem(or may be influence) with it. The 
> Endpoint I have does not have
> does not have "Presence detect change" mechanism. Hot plug is working with 
> Link status events.
> When link is in L1 or L1SS and if EP is powered off, no Link status change 
> event are triggered (It might be
> the expected behavior in L1 or L1SS).  When next time EP is powered on there 
> are link down and
> link up events coming one after other. BIOS enables ASPM on Root port and 
> Endpoint, but while
> processing link status down, pcie_aspm_exit_link_state() clears the ASPM 
> already which were enabled by BIOS.
> If we want to follow above approach then shall we consider having something 
> similar as following?

The proposal was to leave ASPM disabled on hot-added devices.  If the
endpoint was powered off and powered back on again, I think that
device looks like a hot-added device, doesn't it?

Bjorn

Re: [PATCH V8 4/5] PCI/ASPM: save power on values during bridge init

[Bjorn Helgaas]

On Fri, Apr 21, 2017 at 2:46 AM, Patel, Mayurkumar
 wrote:
> Hi Bjorn/Kaya,
>
>
>>
>>On 4/17/2017 12:38 PM, Bjorn Helgaas wrote:
 Like you said, what do we do by default is the question. Should we opt
 for safe like we are doing, or try to save some power.
>>> I think safety is paramount.  Every user should be able to boot safely
>>> without any kernel parameters.  We don't want users to have a problem
>>> booting and then have to search for a workaround like booting with
>>> "pcie_aspm=off".  Most users will never do that.
>>>
>>
>>OK, no problem with leaving the behavior as it is.
>>
>>My initial approach was #2. We knew this way that user had full control
>>over the ASPM policy by changing the BIOS option. Then, Mayurkumar
>>complained that ASPM is not enabled following a hotplug insertion to an
>>empty slot. That's when I switched to #3 as it sounded like a good thing
>>to have for us.
>>
>>> Here's a long-term strawman proposal, see what you think:
>>>
>>>   - Deprecate CONFIG_PCIEASPM_DEFAULT, CONFIG_PCIEASPM_POWERSAVE, etc.
>>>   - Default aspm_policy is POLICY_DEFAULT always.
>>>   - POLICY_DEFAULT means Linux doesn't touch anything: if BIOS enabled
>>> ASPM, we leave it that way; we leave ASPM disabled on hot-added
>>> devices.
>>
> I am also ok with leaving the same behavior as now.
> But still following is something open I feel besides, Which may be there in 
> your comments redundantly.
> The current problem is, pcie_aspm_exit_link_state() disables the ASPM 
> configuration even
> if POLICY_DEFAULT was set.

We call pcie_aspm_exit_link_state() when removing an endpoint.  When
we remove an endpoint, I think disabling ASPM is the right thing to
do.  The spec (PCIe r3.1, sec 5.4.1.3) says "Software must not enable
L0s in either direction on a given Link unless components on both
sides of the Link each support L0s; otherwise, the result is
undefined."

> I am seeing already following problem(or may be influence) with it. The 
> Endpoint I have does not have
> does not have "Presence detect change" mechanism. Hot plug is working with 
> Link status events.
> When link is in L1 or L1SS and if EP is powered off, no Link status change 
> event are triggered (It might be
> the expected behavior in L1 or L1SS).  When next time EP is powered on there 
> are link down and
> link up events coming one after other. BIOS enables ASPM on Root port and 
> Endpoint, but while
> processing link status down, pcie_aspm_exit_link_state() clears the ASPM 
> already which were enabled by BIOS.
> If we want to follow above approach then shall we consider having something 
> similar as following?

The proposal was to leave ASPM disabled on hot-added devices.  If the
endpoint was powered off and powered back on again, I think that
device looks like a hot-added device, doesn't it?

Bjorn

RE: [PATCH V8 4/5] PCI/ASPM: save power on values during bridge init

[Patel, Mayurkumar]

>
>On 4/21/2017 3:46 AM, Patel, Mayurkumar wrote:
>> If we want to follow above approach then shall we consider having something 
>> similar as following?
>
>Do you see this problem if you boot with pcie_aspm.policy=powersave option?
>

No problems. with pcie_aspm.policy=powersave(L1SS are not enabled in this case
but L1 stays ok all the time after many Power(hotplug) cycles but I think that 
is expected with this policy)
and pcie_aspm.policy=powersupersave (L1/L1SS both stays ok all the time).

>
>--
>Sinan Kaya
>Qualcomm Datacenter Technologies, Inc. as an affiliate of Qualcomm 
>Technologies, Inc.
>Qualcomm Technologies, Inc. is a member of the Code Aurora Forum, a Linux 
>Foundation Collaborative Project.
Intel Deutschland GmbH
Registered Address: Am Campeon 10-12, 85579 Neubiberg, Germany
Tel: +49 89 99 8853-0, www.intel.de
Managing Directors: Christin Eisenschmid, Christian Lamprechter
Chairperson of the Supervisory Board: Nicole Lau
Registered Office: Munich
Commercial Register: Amtsgericht Muenchen HRB 186928

RE: [PATCH V8 4/5] PCI/ASPM: save power on values during bridge init

[Patel, Mayurkumar]

>
>On 4/21/2017 3:46 AM, Patel, Mayurkumar wrote:
>> If we want to follow above approach then shall we consider having something 
>> similar as following?
>
>Do you see this problem if you boot with pcie_aspm.policy=powersave option?
>

No problems. with pcie_aspm.policy=powersave(L1SS are not enabled in this case
but L1 stays ok all the time after many Power(hotplug) cycles but I think that 
is expected with this policy)
and pcie_aspm.policy=powersupersave (L1/L1SS both stays ok all the time).

>
>--
>Sinan Kaya
>Qualcomm Datacenter Technologies, Inc. as an affiliate of Qualcomm 
>Technologies, Inc.
>Qualcomm Technologies, Inc. is a member of the Code Aurora Forum, a Linux 
>Foundation Collaborative Project.
Intel Deutschland GmbH
Registered Address: Am Campeon 10-12, 85579 Neubiberg, Germany
Tel: +49 89 99 8853-0, www.intel.de
Managing Directors: Christin Eisenschmid, Christian Lamprechter
Chairperson of the Supervisory Board: Nicole Lau
Registered Office: Munich
Commercial Register: Amtsgericht Muenchen HRB 186928

Re: [PATCH V8 4/5] PCI/ASPM: save power on values during bridge init

[Sinan Kaya]

On 4/21/2017 3:46 AM, Patel, Mayurkumar wrote:
> If we want to follow above approach then shall we consider having something 
> similar as following?

Do you see this problem if you boot with pcie_aspm.policy=powersave option?


-- 
Sinan Kaya
Qualcomm Datacenter Technologies, Inc. as an affiliate of Qualcomm 
Technologies, Inc.
Qualcomm Technologies, Inc. is a member of the Code Aurora Forum, a Linux 
Foundation Collaborative Project.

Re: [PATCH V8 4/5] PCI/ASPM: save power on values during bridge init

[Sinan Kaya]

On 4/21/2017 3:46 AM, Patel, Mayurkumar wrote:
> If we want to follow above approach then shall we consider having something 
> similar as following?

Do you see this problem if you boot with pcie_aspm.policy=powersave option?


-- 
Sinan Kaya
Qualcomm Datacenter Technologies, Inc. as an affiliate of Qualcomm 
Technologies, Inc.
Qualcomm Technologies, Inc. is a member of the Code Aurora Forum, a Linux 
Foundation Collaborative Project.

RE: [PATCH V8 4/5] PCI/ASPM: save power on values during bridge init

[Patel, Mayurkumar]

Hi Bjorn/Kaya,


>
>On 4/17/2017 12:38 PM, Bjorn Helgaas wrote:
>>> Like you said, what do we do by default is the question. Should we opt
>>> for safe like we are doing, or try to save some power.
>> I think safety is paramount.  Every user should be able to boot safely
>> without any kernel parameters.  We don't want users to have a problem
>> booting and then have to search for a workaround like booting with
>> "pcie_aspm=off".  Most users will never do that.
>>
>
>OK, no problem with leaving the behavior as it is.
>
>My initial approach was #2. We knew this way that user had full control
>over the ASPM policy by changing the BIOS option. Then, Mayurkumar
>complained that ASPM is not enabled following a hotplug insertion to an
>empty slot. That's when I switched to #3 as it sounded like a good thing
>to have for us.
>
>> Here's a long-term strawman proposal, see what you think:
>>
>>   - Deprecate CONFIG_PCIEASPM_DEFAULT, CONFIG_PCIEASPM_POWERSAVE, etc.
>>   - Default aspm_policy is POLICY_DEFAULT always.
>>   - POLICY_DEFAULT means Linux doesn't touch anything: if BIOS enabled
>> ASPM, we leave it that way; we leave ASPM disabled on hot-added
>> devices.
>
I am also ok with leaving the same behavior as now.
But still following is something open I feel besides, Which may be there in 
your comments redundantly.
The current problem is, pcie_aspm_exit_link_state() disables the ASPM 
configuration even
if POLICY_DEFAULT was set.
I am seeing already following problem(or may be influence) with it. The 
Endpoint I have does not have
does not have "Presence detect change" mechanism. Hot plug is working with Link 
status events.
When link is in L1 or L1SS and if EP is powered off, no Link status change 
event are triggered (It might be
the expected behavior in L1 or L1SS).  When next time EP is powered on there 
are link down and
link up events coming one after other. BIOS enables ASPM on Root port and 
Endpoint, but while
processing link status down, pcie_aspm_exit_link_state() clears the ASPM 
already which were enabled by BIOS. 
If we want to follow above approach then shall we consider having something 
similar as following?

diff --git a/drivers/pci/pcie/aspm.c b/drivers/pci/pcie/aspm.c
index 1dfa10c..bf5be6d 100644
--- a/drivers/pci/pcie/aspm.c
+++ b/drivers/pci/pcie/aspm.c
@@ -940,7 +940,8 @@ void pcie_aspm_exit_link_state(struct pci_dev *pdev)
parent_link = link->parent;

/* All functions are removed, so just disable ASPM for the link */
-   pcie_config_aspm_link(link, 0);
+   if (aspm_policy != POLICY_DEFAULT)
+   pcie_config_aspm_link(link, 0);
list_del(>sibling);
list_del(>link);
/* Clock PM is for endpoint device */
 

>I can easily see people complaining the other way around. There
>could be some boot FW that doesn't know what ASPM is and this particular
>system could rely on the compile time option for enabling power options.
>Maybe, a command line option will be required for them to keep the existing
>behavior.
>
>>   - Deprecate kernel boot parameters (possibly keep pcie_aspm=off for
>> debugging use).
>>   - Use /sys/module/pcie_aspm/parameters/policy for run-time
>> system-wide  control, including for future hot-added devices.
>>   - Remove CONFIG_PCIEASPM_DEBUG and enable that code always, so we
>> have fine-grained run-time control.
>>
>
>Runtime control sounds like a better plan. We need hooks into the system
>power management policy.
>
>>> Maybe, we are missing a HPP option from the PCI spec.
>> That's an interesting idea.  _HPX does have provision for manipulating
>> Link Control bits (see ACPI r5.0, sec 6.2.8.3), but I don't think very
>> many systems implement it.  And there's currently no connection
>> between program_hpp_type2() and aspm.c, so I'm a little worried that
>> we might have issues if a system did implement an _HPX that sets any
>> of the ASPM bits.
>
>I looked at the spec some more. These are there to restore the register
>settings following hotplug insertion. I agree it won't play nice with ASPM
>as the control bits need to be enabled in coordination with the upstream
>device.
>
>I think the right approach is to let the userspace feed the required
>policy to the kernel like you suggested. Then, it needs to be per port
>via link_state to have the most flexibility.
>
>
>--
>Sinan Kaya
>Qualcomm Datacenter Technologies, Inc. as an affiliate of Qualcomm 
>Technologies, Inc.
>Qualcomm Technologies, Inc. is a member of the Code Aurora Forum, a Linux 
>Foundation Collaborative Project.
Intel Deutschland GmbH
Registered Address: Am Campeon 10-12, 85579 Neubiberg, Germany
Tel: +49 89 99 8853-0, www.intel.de
Managing Directors: Christin Eisenschmid, Christian Lamprechter
Chairperson of the Supervisory Board: Nicole Lau
Registered Office: Munich
Commercial Register: Amtsgericht Muenchen HRB 186928

RE: [PATCH V8 4/5] PCI/ASPM: save power on values during bridge init

[Patel, Mayurkumar]

Hi Bjorn/Kaya,


>
>On 4/17/2017 12:38 PM, Bjorn Helgaas wrote:
>>> Like you said, what do we do by default is the question. Should we opt
>>> for safe like we are doing, or try to save some power.
>> I think safety is paramount.  Every user should be able to boot safely
>> without any kernel parameters.  We don't want users to have a problem
>> booting and then have to search for a workaround like booting with
>> "pcie_aspm=off".  Most users will never do that.
>>
>
>OK, no problem with leaving the behavior as it is.
>
>My initial approach was #2. We knew this way that user had full control
>over the ASPM policy by changing the BIOS option. Then, Mayurkumar
>complained that ASPM is not enabled following a hotplug insertion to an
>empty slot. That's when I switched to #3 as it sounded like a good thing
>to have for us.
>
>> Here's a long-term strawman proposal, see what you think:
>>
>>   - Deprecate CONFIG_PCIEASPM_DEFAULT, CONFIG_PCIEASPM_POWERSAVE, etc.
>>   - Default aspm_policy is POLICY_DEFAULT always.
>>   - POLICY_DEFAULT means Linux doesn't touch anything: if BIOS enabled
>> ASPM, we leave it that way; we leave ASPM disabled on hot-added
>> devices.
>
I am also ok with leaving the same behavior as now.
But still following is something open I feel besides, Which may be there in 
your comments redundantly.
The current problem is, pcie_aspm_exit_link_state() disables the ASPM 
configuration even
if POLICY_DEFAULT was set.
I am seeing already following problem(or may be influence) with it. The 
Endpoint I have does not have
does not have "Presence detect change" mechanism. Hot plug is working with Link 
status events.
When link is in L1 or L1SS and if EP is powered off, no Link status change 
event are triggered (It might be
the expected behavior in L1 or L1SS).  When next time EP is powered on there 
are link down and
link up events coming one after other. BIOS enables ASPM on Root port and 
Endpoint, but while
processing link status down, pcie_aspm_exit_link_state() clears the ASPM 
already which were enabled by BIOS. 
If we want to follow above approach then shall we consider having something 
similar as following?

diff --git a/drivers/pci/pcie/aspm.c b/drivers/pci/pcie/aspm.c
index 1dfa10c..bf5be6d 100644
--- a/drivers/pci/pcie/aspm.c
+++ b/drivers/pci/pcie/aspm.c
@@ -940,7 +940,8 @@ void pcie_aspm_exit_link_state(struct pci_dev *pdev)
parent_link = link->parent;

/* All functions are removed, so just disable ASPM for the link */
-   pcie_config_aspm_link(link, 0);
+   if (aspm_policy != POLICY_DEFAULT)
+   pcie_config_aspm_link(link, 0);
list_del(>sibling);
list_del(>link);
/* Clock PM is for endpoint device */
 

>I can easily see people complaining the other way around. There
>could be some boot FW that doesn't know what ASPM is and this particular
>system could rely on the compile time option for enabling power options.
>Maybe, a command line option will be required for them to keep the existing
>behavior.
>
>>   - Deprecate kernel boot parameters (possibly keep pcie_aspm=off for
>> debugging use).
>>   - Use /sys/module/pcie_aspm/parameters/policy for run-time
>> system-wide  control, including for future hot-added devices.
>>   - Remove CONFIG_PCIEASPM_DEBUG and enable that code always, so we
>> have fine-grained run-time control.
>>
>
>Runtime control sounds like a better plan. We need hooks into the system
>power management policy.
>
>>> Maybe, we are missing a HPP option from the PCI spec.
>> That's an interesting idea.  _HPX does have provision for manipulating
>> Link Control bits (see ACPI r5.0, sec 6.2.8.3), but I don't think very
>> many systems implement it.  And there's currently no connection
>> between program_hpp_type2() and aspm.c, so I'm a little worried that
>> we might have issues if a system did implement an _HPX that sets any
>> of the ASPM bits.
>
>I looked at the spec some more. These are there to restore the register
>settings following hotplug insertion. I agree it won't play nice with ASPM
>as the control bits need to be enabled in coordination with the upstream
>device.
>
>I think the right approach is to let the userspace feed the required
>policy to the kernel like you suggested. Then, it needs to be per port
>via link_state to have the most flexibility.
>
>
>--
>Sinan Kaya
>Qualcomm Datacenter Technologies, Inc. as an affiliate of Qualcomm 
>Technologies, Inc.
>Qualcomm Technologies, Inc. is a member of the Code Aurora Forum, a Linux 
>Foundation Collaborative Project.
Intel Deutschland GmbH
Registered Address: Am Campeon 10-12, 85579 Neubiberg, Germany
Tel: +49 89 99 8853-0, www.intel.de
Managing Directors: Christin Eisenschmid, Christian Lamprechter
Chairperson of the Supervisory Board: Nicole Lau
Registered Office: Munich
Commercial Register: Amtsgericht Muenchen HRB 186928

Re: [PATCH V8 4/5] PCI/ASPM: save power on values during bridge init

[Sinan Kaya]

On 4/17/2017 12:38 PM, Bjorn Helgaas wrote:
>> Like you said, what do we do by default is the question. Should we opt
>> for safe like we are doing, or try to save some power.
> I think safety is paramount.  Every user should be able to boot safely
> without any kernel parameters.  We don't want users to have a problem
> booting and then have to search for a workaround like booting with
> "pcie_aspm=off".  Most users will never do that.
> 

OK, no problem with leaving the behavior as it is.

My initial approach was #2. We knew this way that user had full control
over the ASPM policy by changing the BIOS option. Then, Mayurkumar
complained that ASPM is not enabled following a hotplug insertion to an
empty slot. That's when I switched to #3 as it sounded like a good thing
to have for us.

> Here's a long-term strawman proposal, see what you think:
> 
>   - Deprecate CONFIG_PCIEASPM_DEFAULT, CONFIG_PCIEASPM_POWERSAVE, etc.
>   - Default aspm_policy is POLICY_DEFAULT always.
>   - POLICY_DEFAULT means Linux doesn't touch anything: if BIOS enabled
> ASPM, we leave it that way; we leave ASPM disabled on hot-added
> devices.

I can easily see people complaining the other way around. There
could be some boot FW that doesn't know what ASPM is and this particular
system could rely on the compile time option for enabling power options.
Maybe, a command line option will be required for them to keep the existing
behavior.

>   - Deprecate kernel boot parameters (possibly keep pcie_aspm=off for
> debugging use).
>   - Use /sys/module/pcie_aspm/parameters/policy for run-time
> system-wide  control, including for future hot-added devices.
>   - Remove CONFIG_PCIEASPM_DEBUG and enable that code always, so we
> have fine-grained run-time control.
> 

Runtime control sounds like a better plan. We need hooks into the system
power management policy.

>> Maybe, we are missing a HPP option from the PCI spec.
> That's an interesting idea.  _HPX does have provision for manipulating
> Link Control bits (see ACPI r5.0, sec 6.2.8.3), but I don't think very
> many systems implement it.  And there's currently no connection
> between program_hpp_type2() and aspm.c, so I'm a little worried that
> we might have issues if a system did implement an _HPX that sets any
> of the ASPM bits.

I looked at the spec some more. These are there to restore the register
settings following hotplug insertion. I agree it won't play nice with ASPM
as the control bits need to be enabled in coordination with the upstream
device.

I think the right approach is to let the userspace feed the required
policy to the kernel like you suggested. Then, it needs to be per port
via link_state to have the most flexibility. 


-- 
Sinan Kaya
Qualcomm Datacenter Technologies, Inc. as an affiliate of Qualcomm 
Technologies, Inc.
Qualcomm Technologies, Inc. is a member of the Code Aurora Forum, a Linux 
Foundation Collaborative Project.

Re: [PATCH V8 4/5] PCI/ASPM: save power on values during bridge init

[Sinan Kaya]

On 4/17/2017 12:38 PM, Bjorn Helgaas wrote:
>> Like you said, what do we do by default is the question. Should we opt
>> for safe like we are doing, or try to save some power.
> I think safety is paramount.  Every user should be able to boot safely
> without any kernel parameters.  We don't want users to have a problem
> booting and then have to search for a workaround like booting with
> "pcie_aspm=off".  Most users will never do that.
> 

OK, no problem with leaving the behavior as it is.

My initial approach was #2. We knew this way that user had full control
over the ASPM policy by changing the BIOS option. Then, Mayurkumar
complained that ASPM is not enabled following a hotplug insertion to an
empty slot. That's when I switched to #3 as it sounded like a good thing
to have for us.

> Here's a long-term strawman proposal, see what you think:
> 
>   - Deprecate CONFIG_PCIEASPM_DEFAULT, CONFIG_PCIEASPM_POWERSAVE, etc.
>   - Default aspm_policy is POLICY_DEFAULT always.
>   - POLICY_DEFAULT means Linux doesn't touch anything: if BIOS enabled
> ASPM, we leave it that way; we leave ASPM disabled on hot-added
> devices.

I can easily see people complaining the other way around. There
could be some boot FW that doesn't know what ASPM is and this particular
system could rely on the compile time option for enabling power options.
Maybe, a command line option will be required for them to keep the existing
behavior.

>   - Deprecate kernel boot parameters (possibly keep pcie_aspm=off for
> debugging use).
>   - Use /sys/module/pcie_aspm/parameters/policy for run-time
> system-wide  control, including for future hot-added devices.
>   - Remove CONFIG_PCIEASPM_DEBUG and enable that code always, so we
> have fine-grained run-time control.
> 

Runtime control sounds like a better plan. We need hooks into the system
power management policy.

>> Maybe, we are missing a HPP option from the PCI spec.
> That's an interesting idea.  _HPX does have provision for manipulating
> Link Control bits (see ACPI r5.0, sec 6.2.8.3), but I don't think very
> many systems implement it.  And there's currently no connection
> between program_hpp_type2() and aspm.c, so I'm a little worried that
> we might have issues if a system did implement an _HPX that sets any
> of the ASPM bits.

I looked at the spec some more. These are there to restore the register
settings following hotplug insertion. I agree it won't play nice with ASPM
as the control bits need to be enabled in coordination with the upstream
device.

I think the right approach is to let the userspace feed the required
policy to the kernel like you suggested. Then, it needs to be per port
via link_state to have the most flexibility. 


-- 
Sinan Kaya
Qualcomm Datacenter Technologies, Inc. as an affiliate of Qualcomm 
Technologies, Inc.
Qualcomm Technologies, Inc. is a member of the Code Aurora Forum, a Linux 
Foundation Collaborative Project.

Re: [PATCH V8 4/5] PCI/ASPM: save power on values during bridge init

[Bjorn Helgaas]

On Fri, Apr 14, 2017 at 5:17 PM, Sinan Kaya  wrote:
> On 4/14/2017 5:44 PM, Bjorn Helgaas wrote:
>> I think there's an argument to be made that if we care about ASPM
>> configuration, we should be using a non-POLICY_DEFAULT setting.  And I
>> think there's value in having POLICY_DEFAULT be the absolute lowest-
>> risk setting, which I think means option 1.
>>
>> What do you think?
>
> I see your point. The counter argument is that most of the users do not
> know what an ASPM kernel command line is unless they understand PCI
> language.

I don't think the answer is using the "pcie_aspm.policy=" boot
argument.  I certainly don't want users to have to deal with that.  I
wish we didn't even have that parameter.

I think we need runtime knobs instead (and I guess we already have
/sys/module/pcie_aspm/parameters/policy and /sys/.../link_state), and
distro userspace should use them.  I'm envisioning something in
"System Settings / Power" or similar.  Basically I think the policy
doesn't *have* to be dictated by a kernel boot-time parameter, so it
should not be.

> I have been using the powersave policy option until now. I recently realized
> that nobody except me is using this option. Therefore, we are wasting
> power by default following a hotplug insertion.
>
> This is the case where I'm trying to avoid. With the introduction of NVMe
> u.2 drives, hotplug is becoming more and more mainstream. I decided to
> take the matters into my hand with this series for this very reason.
>
> Like you said, BIOS is out of the picture with pciehp. There is nobody
> to configure ASPM following a hotplug insertion.
>
> I can also claim that If user wants performance, they should boot with
> the performance policy or pcie_aspm=off parameters.
>
> I saw this recommendation in multiple DPDK tuning documents.
>
> Like you said, what do we do by default is the question. Should we opt
> for safe like we are doing, or try to save some power.

I think safety is paramount.  Every user should be able to boot safely
without any kernel parameters.  We don't want users to have a problem
booting and then have to search for a workaround like booting with
"pcie_aspm=off".  Most users will never do that.

Here's a long-term strawman proposal, see what you think:

  - Deprecate CONFIG_PCIEASPM_DEFAULT, CONFIG_PCIEASPM_POWERSAVE, etc.
  - Default aspm_policy is POLICY_DEFAULT always.
  - POLICY_DEFAULT means Linux doesn't touch anything: if BIOS enabled
ASPM, we leave it that way; we leave ASPM disabled on hot-added
devices.
  - Deprecate kernel boot parameters (possibly keep pcie_aspm=off for
debugging use).
  - Use /sys/module/pcie_aspm/parameters/policy for run-time
system-wide  control, including for future hot-added devices.
  - Remove CONFIG_PCIEASPM_DEBUG and enable that code always, so we
have fine-grained run-time control.

> Maybe, we are missing a HPP option from the PCI spec.

That's an interesting idea.  _HPX does have provision for manipulating
Link Control bits (see ACPI r5.0, sec 6.2.8.3), but I don't think very
many systems implement it.  And there's currently no connection
between program_hpp_type2() and aspm.c, so I'm a little worried that
we might have issues if a system did implement an _HPX that sets any
of the ASPM bits.

Bjorn

Re: [PATCH V8 4/5] PCI/ASPM: save power on values during bridge init

[Bjorn Helgaas]

On Fri, Apr 14, 2017 at 5:17 PM, Sinan Kaya  wrote:
> On 4/14/2017 5:44 PM, Bjorn Helgaas wrote:
>> I think there's an argument to be made that if we care about ASPM
>> configuration, we should be using a non-POLICY_DEFAULT setting.  And I
>> think there's value in having POLICY_DEFAULT be the absolute lowest-
>> risk setting, which I think means option 1.
>>
>> What do you think?
>
> I see your point. The counter argument is that most of the users do not
> know what an ASPM kernel command line is unless they understand PCI
> language.

I don't think the answer is using the "pcie_aspm.policy=" boot
argument.  I certainly don't want users to have to deal with that.  I
wish we didn't even have that parameter.

I think we need runtime knobs instead (and I guess we already have
/sys/module/pcie_aspm/parameters/policy and /sys/.../link_state), and
distro userspace should use them.  I'm envisioning something in
"System Settings / Power" or similar.  Basically I think the policy
doesn't *have* to be dictated by a kernel boot-time parameter, so it
should not be.

> I have been using the powersave policy option until now. I recently realized
> that nobody except me is using this option. Therefore, we are wasting
> power by default following a hotplug insertion.
>
> This is the case where I'm trying to avoid. With the introduction of NVMe
> u.2 drives, hotplug is becoming more and more mainstream. I decided to
> take the matters into my hand with this series for this very reason.
>
> Like you said, BIOS is out of the picture with pciehp. There is nobody
> to configure ASPM following a hotplug insertion.
>
> I can also claim that If user wants performance, they should boot with
> the performance policy or pcie_aspm=off parameters.
>
> I saw this recommendation in multiple DPDK tuning documents.
>
> Like you said, what do we do by default is the question. Should we opt
> for safe like we are doing, or try to save some power.

I think safety is paramount.  Every user should be able to boot safely
without any kernel parameters.  We don't want users to have a problem
booting and then have to search for a workaround like booting with
"pcie_aspm=off".  Most users will never do that.

Here's a long-term strawman proposal, see what you think:

  - Deprecate CONFIG_PCIEASPM_DEFAULT, CONFIG_PCIEASPM_POWERSAVE, etc.
  - Default aspm_policy is POLICY_DEFAULT always.
  - POLICY_DEFAULT means Linux doesn't touch anything: if BIOS enabled
ASPM, we leave it that way; we leave ASPM disabled on hot-added
devices.
  - Deprecate kernel boot parameters (possibly keep pcie_aspm=off for
debugging use).
  - Use /sys/module/pcie_aspm/parameters/policy for run-time
system-wide  control, including for future hot-added devices.
  - Remove CONFIG_PCIEASPM_DEBUG and enable that code always, so we
have fine-grained run-time control.

> Maybe, we are missing a HPP option from the PCI spec.

That's an interesting idea.  _HPX does have provision for manipulating
Link Control bits (see ACPI r5.0, sec 6.2.8.3), but I don't think very
many systems implement it.  And there's currently no connection
between program_hpp_type2() and aspm.c, so I'm a little worried that
we might have issues if a system did implement an _HPX that sets any
of the ASPM bits.

Bjorn

Re: [PATCH V8 4/5] PCI/ASPM: save power on values during bridge init

[Sinan Kaya]

On 4/14/2017 5:44 PM, Bjorn Helgaas wrote:
> I think there's an argument to be made that if we care about ASPM
> configuration, we should be using a non-POLICY_DEFAULT setting.  And I
> think there's value in having POLICY_DEFAULT be the absolute lowest-
> risk setting, which I think means option 1.
> 
> What do you think?

I see your point. The counter argument is that most of the users do not
know what an ASPM kernel command line is unless they understand PCI
language. 

I have been using the powersave policy option until now. I recently realized
that nobody except me is using this option. Therefore, we are wasting
power by default following a hotplug insertion.

This is the case where I'm trying to avoid. With the introduction of NVMe
u.2 drives, hotplug is becoming more and more mainstream. I decided to
take the matters into my hand with this series for this very reason.

Like you said, BIOS is out of the picture with pciehp. There is nobody
to configure ASPM following a hotplug insertion.

I can also claim that If user wants performance, they should boot with
the performance policy or pcie_aspm=off parameters. 

I saw this recommendation in multiple DPDK tuning documents.

Like you said, what do we do by default is the question. Should we opt
for safe like we are doing, or try to save some power.

Maybe, we are missing a HPP option from the PCI spec.

-- 
Sinan Kaya
Qualcomm Datacenter Technologies, Inc. as an affiliate of Qualcomm 
Technologies, Inc.
Qualcomm Technologies, Inc. is a member of the Code Aurora Forum, a Linux 
Foundation Collaborative Project.

Re: [PATCH V8 4/5] PCI/ASPM: save power on values during bridge init

[Sinan Kaya]

On 4/14/2017 5:44 PM, Bjorn Helgaas wrote:
> I think there's an argument to be made that if we care about ASPM
> configuration, we should be using a non-POLICY_DEFAULT setting.  And I
> think there's value in having POLICY_DEFAULT be the absolute lowest-
> risk setting, which I think means option 1.
> 
> What do you think?

I see your point. The counter argument is that most of the users do not
know what an ASPM kernel command line is unless they understand PCI
language. 

I have been using the powersave policy option until now. I recently realized
that nobody except me is using this option. Therefore, we are wasting
power by default following a hotplug insertion.

This is the case where I'm trying to avoid. With the introduction of NVMe
u.2 drives, hotplug is becoming more and more mainstream. I decided to
take the matters into my hand with this series for this very reason.

Like you said, BIOS is out of the picture with pciehp. There is nobody
to configure ASPM following a hotplug insertion.

I can also claim that If user wants performance, they should boot with
the performance policy or pcie_aspm=off parameters. 

I saw this recommendation in multiple DPDK tuning documents.

Like you said, what do we do by default is the question. Should we opt
for safe like we are doing, or try to save some power.

Maybe, we are missing a HPP option from the PCI spec.

-- 
Sinan Kaya
Qualcomm Datacenter Technologies, Inc. as an affiliate of Qualcomm 
Technologies, Inc.
Qualcomm Technologies, Inc. is a member of the Code Aurora Forum, a Linux 
Foundation Collaborative Project.

Re: [PATCH V8 4/5] PCI/ASPM: save power on values during bridge init

[Bjorn Helgaas]

[+cc Myron, lkml]

On Fri, Apr 14, 2017 at 03:12:35PM -0400, Sinan Kaya wrote:
> Bjorn,
> 
> On 4/12/2017 3:19 PM, Rajat Jain wrote:
> > On Fri, Apr 7, 2017 at 9:55 PM, Sinan Kaya  wrote:
> >> Now that we added a hook to be called from device_add, save the
> >> default values from the HW registers early in the boot for further
> >> reuse during hot device add/remove operations.
> >>
> >> If the link is down during boot, assume that we want to enable L0s
> >> and L1 following hotplug insertion as well as L1SS if supported.
> > 
> > IIUC, so far POLICY_DEFAULT meant that we'd just use & follow what
> > BIOS has done, and play it safe (never try to be more opportunistic).
> > With this change however, we'd be slightly overstepping and giving
> > ourselves benefit of doubt if the BIOS could not enable ASPM states
> > because the link was not up. This may be good, but I think we should
> > call it out, and add some more elaborate comment on the POLICY_DEFAULT
> > description (what to, and what not to expect in different situations).
> > 
> > It is important because existing systems today, that used to boot
> > without cards and later hotplugged them, didn't have ASPM states
> > enabled. They will now suddenly start seeing all ASPM states enabled
> > including L1 substates for the first time (if supported).
> > 
> 
> Rajat has a good point here. Would you like me to update the ASPM document
> with this new behavior for hotplug?
> 
> Do you have another behavior preference when it comes this?

That *is* a very good point.  I think the change in behavior could be
surprising.

I wonder if we should revise our understanding of what POLICY_DEFAULT
means.  If we decided it means "the kernel never changes any ASPM
config", it would be clear that we keep the BIOS configuration for
everything present at boot, and we don't enable ASPM for any hot-added
devices.

I think the motivation for this series is to apply the BIOS's power
management policy to hot-added devices.  There's no direct way to know
the BIOS's policy, so we're trying to infer it from the boot-time link
configurations.

Should we even *try* to apply the BIOS's policy?  I don't know.  If a
platform really wanted to maintain control over ASPM and apply its policy
consistently, I think it could do that by setting ACPI_FADT_NO_ASPM and
using acpiphp instead of pciehp.   Then the OS would keep its mitts off
ASPM, and the BIOS would have a chance to configure ASPM for hot-added
devices before giving them to the OS.

Here are the possibilities I see for POLICY_DEFAULT:

1) Never touch ASPM config (what we have today)

   Boot-present devices: ASPM config retained from BIOS
   Hot-added devices: ASPM disabled (poweron state)

2) Linux maintains BIOS policy (conservative)

   Boot-present devices: ASPM config retained from BIOS
   Hot-added devices (slot occupied at boot): use boot-time ASPM config
   Hot-added devices (slot empty at boot): ASPM disabled

3) Linux maintains BIOS policy (aggressive, your current patch)

   Boot-present devices: ASPM config retained from BIOS
   Hot-added devices (slot occupied at boot): use boot-time ASPM config
   Hot-added devices (slot empty at boot): ASPM enabled

I'm becoming less convinced that options 2 or 3 make sense.  For one
thing, they're both hard to describe concisely because there are too
many special cases, and that's always a red flag for me.

Even for a given BIOS power management policy, the ASPM configuration
may depend on the particular device; for example, a balanced policy
might enable ASPM for USB devices but not for NICs.  So I'm not sure
it really makes sense to remember what BIOS did for the card that was
in the slot at boot-time and apply that to a possibly different card
hot-added later.

I think there's an argument to be made that if we care about ASPM
configuration, we should be using a non-POLICY_DEFAULT setting.  And I
think there's value in having POLICY_DEFAULT be the absolute lowest-
risk setting, which I think means option 1.

What do you think?

Bjorn

Re: [PATCH V8 4/5] PCI/ASPM: save power on values during bridge init

[Bjorn Helgaas]

[+cc Myron, lkml]

On Fri, Apr 14, 2017 at 03:12:35PM -0400, Sinan Kaya wrote:
> Bjorn,
> 
> On 4/12/2017 3:19 PM, Rajat Jain wrote:
> > On Fri, Apr 7, 2017 at 9:55 PM, Sinan Kaya  wrote:
> >> Now that we added a hook to be called from device_add, save the
> >> default values from the HW registers early in the boot for further
> >> reuse during hot device add/remove operations.
> >>
> >> If the link is down during boot, assume that we want to enable L0s
> >> and L1 following hotplug insertion as well as L1SS if supported.
> > 
> > IIUC, so far POLICY_DEFAULT meant that we'd just use & follow what
> > BIOS has done, and play it safe (never try to be more opportunistic).
> > With this change however, we'd be slightly overstepping and giving
> > ourselves benefit of doubt if the BIOS could not enable ASPM states
> > because the link was not up. This may be good, but I think we should
> > call it out, and add some more elaborate comment on the POLICY_DEFAULT
> > description (what to, and what not to expect in different situations).
> > 
> > It is important because existing systems today, that used to boot
> > without cards and later hotplugged them, didn't have ASPM states
> > enabled. They will now suddenly start seeing all ASPM states enabled
> > including L1 substates for the first time (if supported).
> > 
> 
> Rajat has a good point here. Would you like me to update the ASPM document
> with this new behavior for hotplug?
> 
> Do you have another behavior preference when it comes this?

That *is* a very good point.  I think the change in behavior could be
surprising.

I wonder if we should revise our understanding of what POLICY_DEFAULT
means.  If we decided it means "the kernel never changes any ASPM
config", it would be clear that we keep the BIOS configuration for
everything present at boot, and we don't enable ASPM for any hot-added
devices.

I think the motivation for this series is to apply the BIOS's power
management policy to hot-added devices.  There's no direct way to know
the BIOS's policy, so we're trying to infer it from the boot-time link
configurations.

Should we even *try* to apply the BIOS's policy?  I don't know.  If a
platform really wanted to maintain control over ASPM and apply its policy
consistently, I think it could do that by setting ACPI_FADT_NO_ASPM and
using acpiphp instead of pciehp.   Then the OS would keep its mitts off
ASPM, and the BIOS would have a chance to configure ASPM for hot-added
devices before giving them to the OS.

Here are the possibilities I see for POLICY_DEFAULT:

1) Never touch ASPM config (what we have today)

   Boot-present devices: ASPM config retained from BIOS
   Hot-added devices: ASPM disabled (poweron state)

2) Linux maintains BIOS policy (conservative)

   Boot-present devices: ASPM config retained from BIOS
   Hot-added devices (slot occupied at boot): use boot-time ASPM config
   Hot-added devices (slot empty at boot): ASPM disabled

3) Linux maintains BIOS policy (aggressive, your current patch)

   Boot-present devices: ASPM config retained from BIOS
   Hot-added devices (slot occupied at boot): use boot-time ASPM config
   Hot-added devices (slot empty at boot): ASPM enabled

I'm becoming less convinced that options 2 or 3 make sense.  For one
thing, they're both hard to describe concisely because there are too
many special cases, and that's always a red flag for me.

Even for a given BIOS power management policy, the ASPM configuration
may depend on the particular device; for example, a balanced policy
might enable ASPM for USB devices but not for NICs.  So I'm not sure
it really makes sense to remember what BIOS did for the card that was
in the slot at boot-time and apply that to a possibly different card
hot-added later.

I think there's an argument to be made that if we care about ASPM
configuration, we should be using a non-POLICY_DEFAULT setting.  And I
think there's value in having POLICY_DEFAULT be the absolute lowest-
risk setting, which I think means option 1.

What do you think?

Bjorn

Re: [PATCH V8 4/5] PCI/ASPM: save power on values during bridge init

[Sinan Kaya]

Bjorn,

On 4/12/2017 3:19 PM, Rajat Jain wrote:
> On Fri, Apr 7, 2017 at 9:55 PM, Sinan Kaya  wrote:
>> Now that we added a hook to be called from device_add, save the
>> default values from the HW registers early in the boot for further
>> reuse during hot device add/remove operations.
>>
>> If the link is down during boot, assume that we want to enable L0s
>> and L1 following hotplug insertion as well as L1SS if supported.
> 
> IIUC, so far POLICY_DEFAULT meant that we'd just use & follow what
> BIOS has done, and play it safe (never try to be more opportunistic).
> With this change however, we'd be slightly overstepping and giving
> ourselves benefit of doubt if the BIOS could not enable ASPM states
> because the link was not up. This may be good, but I think we should
> call it out, and add some more elaborate comment on the POLICY_DEFAULT
> description (what to, and what not to expect in different situations).
> 
> It is important because existing systems today, that used to boot
> without cards and later hotplugged them, didn't have ASPM states
> enabled. They will now suddenly start seeing all ASPM states enabled
> including L1 substates for the first time (if supported).
> 

Rajat has a good point here. Would you like me to update the ASPM document
with this new behavior for hotplug?

Do you have another behavior preference when it comes this?

Sinan

-- 
Sinan Kaya
Qualcomm Datacenter Technologies, Inc. as an affiliate of Qualcomm 
Technologies, Inc.
Qualcomm Technologies, Inc. is a member of the Code Aurora Forum, a Linux 
Foundation Collaborative Project.

Re: [PATCH V8 4/5] PCI/ASPM: save power on values during bridge init

[Sinan Kaya]

Bjorn,

On 4/12/2017 3:19 PM, Rajat Jain wrote:
> On Fri, Apr 7, 2017 at 9:55 PM, Sinan Kaya  wrote:
>> Now that we added a hook to be called from device_add, save the
>> default values from the HW registers early in the boot for further
>> reuse during hot device add/remove operations.
>>
>> If the link is down during boot, assume that we want to enable L0s
>> and L1 following hotplug insertion as well as L1SS if supported.
> 
> IIUC, so far POLICY_DEFAULT meant that we'd just use & follow what
> BIOS has done, and play it safe (never try to be more opportunistic).
> With this change however, we'd be slightly overstepping and giving
> ourselves benefit of doubt if the BIOS could not enable ASPM states
> because the link was not up. This may be good, but I think we should
> call it out, and add some more elaborate comment on the POLICY_DEFAULT
> description (what to, and what not to expect in different situations).
> 
> It is important because existing systems today, that used to boot
> without cards and later hotplugged them, didn't have ASPM states
> enabled. They will now suddenly start seeing all ASPM states enabled
> including L1 substates for the first time (if supported).
> 

Rajat has a good point here. Would you like me to update the ASPM document
with this new behavior for hotplug?

Do you have another behavior preference when it comes this?

Sinan

-- 
Sinan Kaya
Qualcomm Datacenter Technologies, Inc. as an affiliate of Qualcomm 
Technologies, Inc.
Qualcomm Technologies, Inc. is a member of the Code Aurora Forum, a Linux 
Foundation Collaborative Project.

Re: [PATCH V8 4/5] PCI/ASPM: save power on values during bridge init

[Rajat Jain]

On Fri, Apr 7, 2017 at 9:55 PM, Sinan Kaya  wrote:
> Now that we added a hook to be called from device_add, save the
> default values from the HW registers early in the boot for further
> reuse during hot device add/remove operations.
>
> If the link is down during boot, assume that we want to enable L0s
> and L1 following hotplug insertion as well as L1SS if supported.

IIUC, so far POLICY_DEFAULT meant that we'd just use & follow what
BIOS has done, and play it safe (never try to be more opportunistic).
With this change however, we'd be slightly overstepping and giving
ourselves benefit of doubt if the BIOS could not enable ASPM states
because the link was not up. This may be good, but I think we should
call it out, and add some more elaborate comment on the POLICY_DEFAULT
description (what to, and what not to expect in different situations).

It is important because existing systems today, that used to boot
without cards and later hotplugged them, didn't have ASPM states
enabled. They will now suddenly start seeing all ASPM states enabled
including L1 substates for the first time (if supported).

My system is not hotplug capable (I have the EP soldered on board, so
couldn't do much testing, except for sanity. Please feel free to use
my Reviewed-by.

>
> Fixes: https://bugzilla.kernel.org/show_bug.cgi?id=194895
> Signed-off-by: Sinan Kaya 
> ---
>  drivers/pci/pcie/aspm.c | 34 ++
>  1 file changed, 30 insertions(+), 4 deletions(-)
>
> diff --git a/drivers/pci/pcie/aspm.c b/drivers/pci/pcie/aspm.c
> index e33f84b..c7da087 100644
> --- a/drivers/pci/pcie/aspm.c
> +++ b/drivers/pci/pcie/aspm.c
> @@ -505,8 +505,10 @@ static void pcie_aspm_cap_init(struct pcie_link_state 
> *link, int blacklist)
>  */
> if (dwreg.support & upreg.support & PCIE_LINK_STATE_L0S)
> link->aspm_support |= ASPM_STATE_L0S;
> -   if (dwreg.enabled & PCIE_LINK_STATE_L0S)
> +   if (dwreg.enabled & PCIE_LINK_STATE_L0S) {
> link->aspm_enabled |= ASPM_STATE_L0S_UP;
> +   link->aspm_default |= ASPM_STATE_L0S_UP;
> +   }
> if (upreg.enabled & PCIE_LINK_STATE_L0S)
> link->aspm_enabled |= ASPM_STATE_L0S_DW;
> link->latency_up.l0s = calc_l0s_latency(upreg.latency_encoding_l0s);
> @@ -542,9 +544,6 @@ static void pcie_aspm_cap_init(struct pcie_link_state 
> *link, int blacklist)
> if (link->aspm_support & ASPM_STATE_L1SS)
> aspm_calc_l1ss_info(link, , );
>
> -   /* Save default state */
> -   link->aspm_default = link->aspm_enabled;
> -
> /* Setup initial capable state. Will be updated later */
> link->aspm_capable = link->aspm_support;
> /*
> @@ -835,11 +834,38 @@ static int pci_aspm_init_downstream(struct pci_dev 
> *pdev)
>  static int pci_aspm_init_upstream(struct pci_dev *pdev)
>  {
> struct pcie_link_state *link;
> +   struct aspm_register_info upreg;
> +   u16 lnk_status;
> +   bool ret;
>
> link = alloc_pcie_link_state(pdev);
> if (!link)
> return -ENOMEM;
>
> +   pcie_capability_read_word(pdev, PCI_EXP_LNKSTA, _status);
> +   ret = !!(lnk_status & PCI_EXP_LNKSTA_DLLLA);
> +
> +   if (ret) {
> +   pcie_get_aspm_reg(pdev, );
> +   if (upreg.enabled & PCIE_LINK_STATE_L0S)
> +   link->aspm_default |= ASPM_STATE_L0S_DW;
> +   if (upreg.enabled & PCIE_LINK_STATE_L1)
> +   link->aspm_default |= ASPM_STATE_L1;
> +   if (upreg.l1ss_ctl1 & PCI_L1SS_CTL1_ASPM_L1_1)
> +   link->aspm_default |= ASPM_STATE_L1_1;
> +   if (upreg.l1ss_ctl1 & PCI_L1SS_CTL1_ASPM_L1_2)
> +   link->aspm_default |= ASPM_STATE_L1_2;
> +   if (upreg.l1ss_ctl1 & PCI_L1SS_CTL1_PCIPM_L1_1)
> +   link->aspm_default |= ASPM_STATE_L1_1_PCIPM;
> +   if (upreg.l1ss_ctl1 & PCI_L1SS_CTL1_PCIPM_L1_2)
> +   link->aspm_default |= ASPM_STATE_L1_2_PCIPM;
> +   } else {
> +   if (!pci_find_ext_capability(pdev, PCI_EXT_CAP_ID_L1SS))
> +   link->aspm_default = ASPM_STATE_L0S | ASPM_STATE_L1;
> +   else
> +   link->aspm_default = ASPM_STATE_ALL;
> +   }

Optional: May be consider moving this code (more aptly) to
pcie_aspm_cap_init() by adding a check for link-up before we start
reading downstream registers there? I guess you'll need to move the
call to pcie_aspm_cap_init() a little further up in
pcie_aspm_init_link_state().

> +
> return 0;
>  }
>
> --
> 1.9.1
>

Re: [PATCH V8 4/5] PCI/ASPM: save power on values during bridge init

[Rajat Jain]

On Fri, Apr 7, 2017 at 9:55 PM, Sinan Kaya  wrote:
> Now that we added a hook to be called from device_add, save the
> default values from the HW registers early in the boot for further
> reuse during hot device add/remove operations.
>
> If the link is down during boot, assume that we want to enable L0s
> and L1 following hotplug insertion as well as L1SS if supported.

IIUC, so far POLICY_DEFAULT meant that we'd just use & follow what
BIOS has done, and play it safe (never try to be more opportunistic).
With this change however, we'd be slightly overstepping and giving
ourselves benefit of doubt if the BIOS could not enable ASPM states
because the link was not up. This may be good, but I think we should
call it out, and add some more elaborate comment on the POLICY_DEFAULT
description (what to, and what not to expect in different situations).

It is important because existing systems today, that used to boot
without cards and later hotplugged them, didn't have ASPM states
enabled. They will now suddenly start seeing all ASPM states enabled
including L1 substates for the first time (if supported).

My system is not hotplug capable (I have the EP soldered on board, so
couldn't do much testing, except for sanity. Please feel free to use
my Reviewed-by.

>
> Fixes: https://bugzilla.kernel.org/show_bug.cgi?id=194895
> Signed-off-by: Sinan Kaya 
> ---
>  drivers/pci/pcie/aspm.c | 34 ++
>  1 file changed, 30 insertions(+), 4 deletions(-)
>
> diff --git a/drivers/pci/pcie/aspm.c b/drivers/pci/pcie/aspm.c
> index e33f84b..c7da087 100644
> --- a/drivers/pci/pcie/aspm.c
> +++ b/drivers/pci/pcie/aspm.c
> @@ -505,8 +505,10 @@ static void pcie_aspm_cap_init(struct pcie_link_state 
> *link, int blacklist)
>  */
> if (dwreg.support & upreg.support & PCIE_LINK_STATE_L0S)
> link->aspm_support |= ASPM_STATE_L0S;
> -   if (dwreg.enabled & PCIE_LINK_STATE_L0S)
> +   if (dwreg.enabled & PCIE_LINK_STATE_L0S) {
> link->aspm_enabled |= ASPM_STATE_L0S_UP;
> +   link->aspm_default |= ASPM_STATE_L0S_UP;
> +   }
> if (upreg.enabled & PCIE_LINK_STATE_L0S)
> link->aspm_enabled |= ASPM_STATE_L0S_DW;
> link->latency_up.l0s = calc_l0s_latency(upreg.latency_encoding_l0s);
> @@ -542,9 +544,6 @@ static void pcie_aspm_cap_init(struct pcie_link_state 
> *link, int blacklist)
> if (link->aspm_support & ASPM_STATE_L1SS)
> aspm_calc_l1ss_info(link, , );
>
> -   /* Save default state */
> -   link->aspm_default = link->aspm_enabled;
> -
> /* Setup initial capable state. Will be updated later */
> link->aspm_capable = link->aspm_support;
> /*
> @@ -835,11 +834,38 @@ static int pci_aspm_init_downstream(struct pci_dev 
> *pdev)
>  static int pci_aspm_init_upstream(struct pci_dev *pdev)
>  {
> struct pcie_link_state *link;
> +   struct aspm_register_info upreg;
> +   u16 lnk_status;
> +   bool ret;
>
> link = alloc_pcie_link_state(pdev);
> if (!link)
> return -ENOMEM;
>
> +   pcie_capability_read_word(pdev, PCI_EXP_LNKSTA, _status);
> +   ret = !!(lnk_status & PCI_EXP_LNKSTA_DLLLA);
> +
> +   if (ret) {
> +   pcie_get_aspm_reg(pdev, );
> +   if (upreg.enabled & PCIE_LINK_STATE_L0S)
> +   link->aspm_default |= ASPM_STATE_L0S_DW;
> +   if (upreg.enabled & PCIE_LINK_STATE_L1)
> +   link->aspm_default |= ASPM_STATE_L1;
> +   if (upreg.l1ss_ctl1 & PCI_L1SS_CTL1_ASPM_L1_1)
> +   link->aspm_default |= ASPM_STATE_L1_1;
> +   if (upreg.l1ss_ctl1 & PCI_L1SS_CTL1_ASPM_L1_2)
> +   link->aspm_default |= ASPM_STATE_L1_2;
> +   if (upreg.l1ss_ctl1 & PCI_L1SS_CTL1_PCIPM_L1_1)
> +   link->aspm_default |= ASPM_STATE_L1_1_PCIPM;
> +   if (upreg.l1ss_ctl1 & PCI_L1SS_CTL1_PCIPM_L1_2)
> +   link->aspm_default |= ASPM_STATE_L1_2_PCIPM;
> +   } else {
> +   if (!pci_find_ext_capability(pdev, PCI_EXT_CAP_ID_L1SS))
> +   link->aspm_default = ASPM_STATE_L0S | ASPM_STATE_L1;
> +   else
> +   link->aspm_default = ASPM_STATE_ALL;
> +   }

Optional: May be consider moving this code (more aptly) to
pcie_aspm_cap_init() by adding a check for link-up before we start
reading downstream registers there? I guess you'll need to move the
call to pcie_aspm_cap_init() a little further up in
pcie_aspm_init_link_state().

> +
> return 0;
>  }
>
> --
> 1.9.1
>