[V2] PCI/ASPM: reconfigure ASPM following hotplug for POLICY_DEFAULT
diff mbox

Message ID 1485891928-14573-1-git-send-email-okaya@codeaurora.org
State Superseded
Headers show

Commit Message

Sinan Kaya Jan. 31, 2017, 7:45 p.m. UTC
When the operating system is booted with the default ASPM policy
(POLICY_DEFAULT), the current code is determining the ASPM policy
set up by the BIOS by querying the enable/disable states from ASPM
registers.

In the case of hotplug removal, the ASPM registers get cleared by
calling the pcie_aspm_exit_link_state() function.

An insertion following hotplug removal reads incorrect policy as
ASPM disabled even though ASPM was enabled during boot.

Adding a flag to the struct pci_dev and saving the power up policy
in the bridge to be reused during hotplug insertion. Bridge's enable
counter is used as a switch to determine when to use saved value.

Signed-off-by: Sinan Kaya <okaya@codeaurora.org>
---
 drivers/pci/pcie/aspm.c | 21 +++++++++++++++++----
 include/linux/pci.h     |  1 +
 2 files changed, 18 insertions(+), 4 deletions(-)

Comments

Patel, Mayurkumar Feb. 28, 2017, 11:03 a.m. UTC | #1
>
>When the operating system is booted with the default ASPM policy
>(POLICY_DEFAULT), the current code is determining the ASPM policy
>set up by the BIOS by querying the enable/disable states from ASPM
>registers.
>
>In the case of hotplug removal, the ASPM registers get cleared by
>calling the pcie_aspm_exit_link_state() function.
>

I see the same problem. So cherry-picked your change to resolve the problem
on top of Linus mainline.

>An insertion following hotplug removal reads incorrect policy as
>ASPM disabled even though ASPM was enabled during boot.
>
>Adding a flag to the struct pci_dev and saving the power up policy
>in the bridge to be reused during hotplug insertion. Bridge's enable
>counter is used as a switch to determine when to use saved value.
>
>Signed-off-by: Sinan Kaya <okaya@codeaurora.org>
>---
> drivers/pci/pcie/aspm.c | 21 +++++++++++++++++----
> include/linux/pci.h     |  1 +
> 2 files changed, 18 insertions(+), 4 deletions(-)
>
>diff --git a/drivers/pci/pcie/aspm.c b/drivers/pci/pcie/aspm.c
>index 17ac1dc..5cfcc6d 100644
>--- a/drivers/pci/pcie/aspm.c
>+++ b/drivers/pci/pcie/aspm.c
>@@ -338,8 +338,9 @@ static void pcie_aspm_check_latency(struct pci_dev *endpoint)
> 	}
> }
>
>-static void pcie_aspm_cap_init(struct pcie_link_state *link, int blacklist)
>+static void pcie_aspm_cap_init(struct pci_dev *pdev, int blacklist)
> {
>+	struct pcie_link_state *link = pdev->link_state;
> 	struct pci_dev *child, *parent = link->pdev;
> 	struct pci_bus *linkbus = parent->subordinate;
> 	struct aspm_register_info upreg, dwreg;
>@@ -397,8 +398,20 @@ static void pcie_aspm_cap_init(struct pcie_link_state *link, int blacklist)
> 	link->latency_up.l1 = calc_l1_latency(upreg.latency_encoding_l1);
> 	link->latency_dw.l1 = calc_l1_latency(dwreg.latency_encoding_l1);
>
>-	/* Save default state */
>-	link->aspm_default = link->aspm_enabled;

But, I am finding a problem with this change, if Policy is set to default,
BIOS enables ASPM L1, but pcie_config_aspm_link() disables ASPM L1
due to link->aspm_enabled is different than link->aspm_default while
called from pcie_aspm_init_link_state() during hotplug insertion.


>+	/*
>+	 * Save default state from FW when enabling ASPM for the first time
>+	 * during boot by looking at the calculated link->aspm_enabled bits
>+	 * above and enable_cnt will be zero.
>+	 *
>+	 * If this path is getting called for the second/third time
>+	 * (enable_cnt will be non-zero). Assume that the current state of the
>+	 * ASPM registers may not necessarily match what FW asked us to do as
>+	 * in the case of hotplug insertion/removal.
>+	 */
>+	if (!atomic_read(&pdev->enable_cnt))
>+		pdev->aspm_default = link->aspm_default = link->aspm_enabled;
>+	else
>+		link->aspm_default = pdev->aspm_default;
>
> 	/* Setup initial capable state. Will be updated later */
> 	link->aspm_capable = link->aspm_support;
>@@ -599,7 +612,7 @@ void pcie_aspm_init_link_state(struct pci_dev *pdev)
> 	 * upstream links also because capable state of them can be
> 	 * update through pcie_aspm_cap_init().
> 	 */
>-	pcie_aspm_cap_init(link, blacklist);
>+	pcie_aspm_cap_init(pdev, blacklist);
>
> 	/* Setup initial Clock PM state */
> 	pcie_clkpm_cap_init(link, blacklist);
>diff --git a/include/linux/pci.h b/include/linux/pci.h
>index e2d1a12..521f88c 100644
>--- a/include/linux/pci.h
>+++ b/include/linux/pci.h
>@@ -321,6 +321,7 @@ struct pci_dev {
>
> #ifdef CONFIG_PCIEASPM
> 	struct pcie_link_state	*link_state;	/* ASPM link state */
>+	unsigned int	aspm_default;		/* ASPM policy set by BIOS */
> #endif
>
> 	pci_channel_state_t error_state;	/* current connectivity state */
>--
>1.9.1
>
>--
>To unsubscribe from this list: send the line "unsubscribe linux-pci" in
>the body of a message to majordomo@vger.kernel.org
>More majordomo info at  http://vger.kernel.org/majordomo-info.html
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
Sinan Kaya Feb. 28, 2017, 3:04 p.m. UTC | #2
On 2/28/2017 6:03 AM, Patel, Mayurkumar wrote:
>> -	/* Save default state */
>> -	link->aspm_default = link->aspm_enabled;
> But, I am finding a problem with this change, if Policy is set to default,
> BIOS enables ASPM L1, but pcie_config_aspm_link() disables ASPM L1
> due to link->aspm_enabled is different than link->aspm_default while
> called from pcie_aspm_init_link_state() during hotplug insertion.
> 

Thanks for testing. I think it means the enable count logic I put below
is not working.

I was trying to figure out when to use saved values vs. the values in
registers by looking at the enable_cnt.

enable_cnt is 0 during boot on my system.

The enable_cnt should have been non-zero following an insertion following
remove and default and enabled values should have been identical. 

Can you print out the values of pdev->aspm_default, link->aspm_default, 
and link->aspm_enabled in this function along with atomic_read(&pdev->enable_cnt)?

> 
>> +	/*
>> +	 * Save default state from FW when enabling ASPM for the first time
>> +	 * during boot by looking at the calculated link->aspm_enabled bits
>> +	 * above and enable_cnt will be zero.
>> +	 *
>> +	 * If this path is getting called for the second/third time
>> +	 * (enable_cnt will be non-zero). Assume that the current state of the
>> +	 * ASPM registers may not necessarily match what FW asked us to do as
>> +	 * in the case of hotplug insertion/removal.
>> +	 */
>> +	if (!atomic_read(&pdev->enable_cnt))
>> +		pdev->aspm_default = link->aspm_default = link->aspm_enabled;
>> +	else
>> +		link->aspm_default = pdev->aspm_default;
Patel, Mayurkumar Feb. 28, 2017, 6:57 p.m. UTC | #3
>
>On 2/28/2017 6:03 AM, Patel, Mayurkumar wrote:
>>> -	/* Save default state */
>>> -	link->aspm_default = link->aspm_enabled;
>> But, I am finding a problem with this change, if Policy is set to default,
>> BIOS enables ASPM L1, but pcie_config_aspm_link() disables ASPM L1
>> due to link->aspm_enabled is different than link->aspm_default while
>> called from pcie_aspm_init_link_state() during hotplug insertion.
>>
>
>Thanks for testing. I think it means the enable count logic I put below
>is not working.

Yes I believe so.

>
>I was trying to figure out when to use saved values vs. the values in
>registers by looking at the enable_cnt.

>enable_cnt is 0 during boot on my system.

enable_cnt for the root port on my system is set to 1 for "root port" already without saving
any ASPM related settings. 


>
>The enable_cnt should have been non-zero following an insertion following
>remove and default and enabled values should have been identical.
>

pcie_aspm_init_link_state() is called on the "root port" from hotplug routine when I plug in PCIE Endpoint.
And in my case root port "enable_cnt" is already set to 1 because of root port driver probed already. So
link->aspm_default by  default always sets to pdev->aspm_default which is 0 as it was never set to
BIOS values before. I will continue dig in further to see how the "root port" probe gets triggered
and why it does not call pcie_aspm_init_link_state() to store aspm policy before "enable_cnt" becomes 1 for it.

>Can you print out the values of pdev->aspm_default, link->aspm_default,
>and link->aspm_enabled in this function along with atomic_read(&pdev->enable_cnt)?
>
>>

Yes. Also printed link control register(lnkctl) set by BIOS for upstream port and downstream port
Which gets disabled afterwards due to the link->aspm_default is 0. The prints are in pcie_aspm_cap_init()
Function.

[  199.287732] pci 0000:03:00.0: reg 0x10: [mem 0x00000000-0x00000fff 64bit]
[  199.287748] pci 0000:03:00.0: reg 0x18: [mem 0x00000000-0x000000ff 64bit]
[  199.287864] pci 0000:03:00.0: PME# supported from D0 D3hot D3cold
[  199.288000] pci 0000:03:00.0: System wakeup disabled by ACPI
[  199.299868] upreg lnkcap 00000003 lnkctl=00000042
[  199.299996] dwreg lnkcap  00000002 lnkctl=00000142
[  199.300076] pdev_aspm_default: 0x00000000 link_aspm_default: 0x00000000 aspm_enabled: 0x00000004 enable_cnt: 0x00000001 
[  199.300402] pci 0000:03:00.0: BAR 0: assigned [mem 0xaa000000-0xaa000fff 64bit]
[  199.300413] pci 0000:03:00.0: BAR 2: assigned [mem 0xaa001000-0xaa0010ff 64bit]


>>> +	/*
>>> +	 * Save default state from FW when enabling ASPM for the first time
>>> +	 * during boot by looking at the calculated link->aspm_enabled bits
>>> +	 * above and enable_cnt will be zero.
>>> +	 *
>>> +	 * If this path is getting called for the second/third time
>>> +	 * (enable_cnt will be non-zero). Assume that the current state of the
>>> +	 * ASPM registers may not necessarily match what FW asked us to do as
>>> +	 * in the case of hotplug insertion/removal.
>>> +	 */
>>> +	if (!atomic_read(&pdev->enable_cnt))
>>> +		pdev->aspm_default = link->aspm_default = link->aspm_enabled;
>>> +	else
>>> +		link->aspm_default = pdev->aspm_default;
>
>
>--
>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
Sinan Kaya March 1, 2017, 5:58 p.m. UTC | #4
Hi Bjorn,

On 2/28/2017 1:57 PM, Patel, Mayurkumar wrote:
>> I was trying to figure out when to use saved values vs. the values in
>> registers by looking at the enable_cnt.
>> enable_cnt is 0 during boot on my system.
> enable_cnt for the root port on my system is set to 1 for "root port" already without saving
> any ASPM related settings. 
> 
> 

What would be the best way to figure out when to save power-on values from
the registers?
Patel, Mayurkumar March 2, 2017, 4:05 p.m. UTC | #5
>
>Hi Bjorn,
>
>On 2/28/2017 1:57 PM, Patel, Mayurkumar wrote:
>>> I was trying to figure out when to use saved values vs. the values in
>>> registers by looking at the enable_cnt.
>>> enable_cnt is 0 during boot on my system.
>> enable_cnt for the root port on my system is set to 1 for "root port" already without saving
>> any ASPM related settings.
>>
>>
>
>What would be the best way to figure out when to save power-on values from
>the registers?
>

So in my case, pci_enable_device() for root port is called from pcie_port_device_register() at the boot time which
Is called from pcie_portdrv_probe() function which does not call pcie_aspm_init_link_state() but
Increment's enable_cnt field. When the Endpoint get plugged in, pcie_aspm_init_link_state() is called
from hotplug routine on "Root port" but due to enable_cnt is already incremented we do not set
right policy and aspm gets disabled by aspm driver.

Shall we introduce new atomic variable along with aspm_default in struct pci_dev and increment it in when called
first time pcie_aspm_init_link_state()  to save power-on values ( I have tested it locally, if you think could be good approach
I can upload the diffs) because enable_cnt in pci_enable_device() can be triggered
from multiple places at boot time so it might not be safe to use it? 
or adding pcie_aspm_init_link_state() in pcie_port_device_register() for ports before calling pci_enable_device()
although I am not sure whether it's the right approach and that would work for all the devices!

@Kaya/Bjorn: Do you have any other suggestions or Could you also please help by comment what would make sense?


>--
>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
Sinan Kaya March 2, 2017, 6:10 p.m. UTC | #6
Hi Mayurkumar

On 3/2/2017 11:05 AM, Patel, Mayurkumar wrote:
>>
>> Hi Bjorn,
>>
>> On 2/28/2017 1:57 PM, Patel, Mayurkumar wrote:
>>>> I was trying to figure out when to use saved values vs. the values in
>>>> registers by looking at the enable_cnt.
>>>> enable_cnt is 0 during boot on my system.
>>> enable_cnt for the root port on my system is set to 1 for "root port" already without saving
>>> any ASPM related settings.
>>>
>>>
>>
>> What would be the best way to figure out when to save power-on values from
>> the registers?
>>
> 
> I can upload the diffs) because enable_cnt in pci_enable_device() can be triggered
> from multiple places at boot time so it might not be safe to use it? 

Go ahead and share your diff. It doesn't hurt to look at other alternatives.

> 
> @Kaya/Bjorn: Do you have any other suggestions or Could you also please help by comment what would make sense?
> 
> 
>> --
>> 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
> 
>

Patch
diff mbox

diff --git a/drivers/pci/pcie/aspm.c b/drivers/pci/pcie/aspm.c
index 17ac1dc..5cfcc6d 100644
--- a/drivers/pci/pcie/aspm.c
+++ b/drivers/pci/pcie/aspm.c
@@ -338,8 +338,9 @@  static void pcie_aspm_check_latency(struct pci_dev *endpoint)
 	}
 }
 
-static void pcie_aspm_cap_init(struct pcie_link_state *link, int blacklist)
+static void pcie_aspm_cap_init(struct pci_dev *pdev, int blacklist)
 {
+	struct pcie_link_state *link = pdev->link_state;
 	struct pci_dev *child, *parent = link->pdev;
 	struct pci_bus *linkbus = parent->subordinate;
 	struct aspm_register_info upreg, dwreg;
@@ -397,8 +398,20 @@  static void pcie_aspm_cap_init(struct pcie_link_state *link, int blacklist)
 	link->latency_up.l1 = calc_l1_latency(upreg.latency_encoding_l1);
 	link->latency_dw.l1 = calc_l1_latency(dwreg.latency_encoding_l1);
 
-	/* Save default state */
-	link->aspm_default = link->aspm_enabled;
+	/*
+	 * Save default state from FW when enabling ASPM for the first time
+	 * during boot by looking at the calculated link->aspm_enabled bits
+	 * above and enable_cnt will be zero.
+	 *
+	 * If this path is getting called for the second/third time
+	 * (enable_cnt will be non-zero). Assume that the current state of the
+	 * ASPM registers may not necessarily match what FW asked us to do as
+	 * in the case of hotplug insertion/removal.
+	 */
+	if (!atomic_read(&pdev->enable_cnt))
+		pdev->aspm_default = link->aspm_default = link->aspm_enabled;
+	else
+		link->aspm_default = pdev->aspm_default;
 
 	/* Setup initial capable state. Will be updated later */
 	link->aspm_capable = link->aspm_support;
@@ -599,7 +612,7 @@  void pcie_aspm_init_link_state(struct pci_dev *pdev)
 	 * upstream links also because capable state of them can be
 	 * update through pcie_aspm_cap_init().
 	 */
-	pcie_aspm_cap_init(link, blacklist);
+	pcie_aspm_cap_init(pdev, blacklist);
 
 	/* Setup initial Clock PM state */
 	pcie_clkpm_cap_init(link, blacklist);
diff --git a/include/linux/pci.h b/include/linux/pci.h
index e2d1a12..521f88c 100644
--- a/include/linux/pci.h
+++ b/include/linux/pci.h
@@ -321,6 +321,7 @@  struct pci_dev {
 
 #ifdef CONFIG_PCIEASPM
 	struct pcie_link_state	*link_state;	/* ASPM link state */
+	unsigned int	aspm_default;		/* ASPM policy set by BIOS */
 #endif
 
 	pci_channel_state_t error_state;	/* current connectivity state */