[PATCH RFC 0/3] PCI/ASPM: add sysfs attribute for controlling ASPM

[PATCH RFC 0/3] PCI/ASPM: add sysfs attribute for controlling ASPM

[Heiner Kallweit]

Background of this extension is a problem with the r8169 network driver.
Several combinations of board chipsets and network chip versions have
problems if ASPM is enabled, therefore we have to disable ASPM per
default. However especially on notebooks ASPM can provide significant
power-saving, therefore we want to give users the option to enable
ASPM. With the new sysfs attribute users can control which ASPM
link-states are disabled.

This is a first RFC version, therefore documentation of attribute is
still missing. The attribute handling was inspired by the protocol
attribute handling in drivers/media/rc/rc-main.c.
Attribute syntax in a few words:
none: allow all supported ASPM states
all: disable all ASPM states
+<state>: add state to list of disabled ASPM states
-<state>: re-enable ASPM state if supported

Heiner Kallweit (3):
  PCI/ASPM: add L1 sub-state support to pci_disable_link_state
  PCI/ASPM: allow to re-enable Clock PM
  PCI/ASPM: add sysfs attribute for controlling ASPM

 drivers/pci/pci.h        |   8 +-
 drivers/pci/pcie/aspm.c  | 204 ++++++++++++++++++++++++++++++++++++---
 include/linux/pci-aspm.h |   8 +-
 3 files changed, 199 insertions(+), 21 deletions(-)

-- 
2.21.0

[PATCH RFC 1/3] PCI/ASPM: add L1 sub-state support to pci_disable_link_state

[Heiner Kallweit]

Add support for disabling states L1.1 and L1.2 to pci_disable_link_state.

Signed-off-by: Heiner Kallweit <hkallweit1@gmail.com>
---
 drivers/pci/pcie/aspm.c  | 13 ++++++++++---
 include/linux/pci-aspm.h |  8 +++++---
 2 files changed, 15 insertions(+), 6 deletions(-)

diff --git a/drivers/pci/pcie/aspm.c b/drivers/pci/pcie/aspm.c
index fd4cb7508..511f3e018 100644
--- a/drivers/pci/pcie/aspm.c
+++ b/drivers/pci/pcie/aspm.c
@@ -35,9 +35,9 @@
 #define ASPM_STATE_L1_1_PCIPM	(0x20)	/* PCI PM L1.1 state */
 #define ASPM_STATE_L1_2_PCIPM	(0x40)	/* PCI PM L1.2 state */
 #define ASPM_STATE_L1_SS_PCIPM	(ASPM_STATE_L1_1_PCIPM | ASPM_STATE_L1_2_PCIPM)
+#define ASPM_STATE_L1_1_MASK	(ASPM_STATE_L1_1 | ASPM_STATE_L1_1_PCIPM)
 #define ASPM_STATE_L1_2_MASK	(ASPM_STATE_L1_2 | ASPM_STATE_L1_2_PCIPM)
-#define ASPM_STATE_L1SS		(ASPM_STATE_L1_1 | ASPM_STATE_L1_1_PCIPM |\
-				 ASPM_STATE_L1_2_MASK)
+#define ASPM_STATE_L1SS		(ASPM_STATE_L1_1_MASK | ASPM_STATE_L1_2_MASK)
 #define ASPM_STATE_L0S		(ASPM_STATE_L0S_UP | ASPM_STATE_L0S_DW)
 #define ASPM_STATE_ALL		(ASPM_STATE_L0S | ASPM_STATE_L1 |	\
 				 ASPM_STATE_L1SS)
@@ -1094,8 +1094,15 @@ static void __pci_disable_link_state(struct pci_dev *pdev, int state, bool sem)
 	link = parent->link_state;
 	if (state & PCIE_LINK_STATE_L0S)
 		link->aspm_disable |= ASPM_STATE_L0S;
-	if (state & PCIE_LINK_STATE_L1)
+	if (state & PCIE_LINK_STATE_L1) {
 		link->aspm_disable |= ASPM_STATE_L1;
+		/* sub-states require L1 */
+		link->aspm_disable |= ASPM_STATE_L1SS;
+	}
+	if (state & PCIE_LINK_STATE_L1_1)
+		link->aspm_disable |= ASPM_STATE_L1_1_MASK;
+	if (state & PCIE_LINK_STATE_L1_2)
+		link->aspm_disable |= ASPM_STATE_L1_2_MASK;
 	pcie_config_aspm_link(link, policy_to_aspm_state(link));
 
 	if (state & PCIE_LINK_STATE_CLKPM) {
diff --git a/include/linux/pci-aspm.h b/include/linux/pci-aspm.h
index df28af5ce..e66c3e3d8 100644
--- a/include/linux/pci-aspm.h
+++ b/include/linux/pci-aspm.h
@@ -19,9 +19,11 @@
 
 #include <linux/pci.h>
 
-#define PCIE_LINK_STATE_L0S	1
-#define PCIE_LINK_STATE_L1	2
-#define PCIE_LINK_STATE_CLKPM	4
+#define PCIE_LINK_STATE_L0S	BIT(0)
+#define PCIE_LINK_STATE_L1	BIT(1)
+#define PCIE_LINK_STATE_CLKPM	BIT(2)
+#define PCIE_LINK_STATE_L1_1	BIT(3)
+#define PCIE_LINK_STATE_L1_2	BIT(4)
 
 #ifdef CONFIG_PCIEASPM
 void pci_disable_link_state(struct pci_dev *pdev, int state);
-- 
2.21.0

[PATCH RFC 2/3] PCI/ASPM: allow to re-enable Clock PM

[Heiner Kallweit]

So far Clock PM can't be re-enabled once it has been disabled with a
call to pci_disable_link_state(). Reason is that clkpm_capable is
reset. Change this by adding a clkpm_disable field similar to
aspm_disable.

Signed-off-by: Heiner Kallweit <hkallweit1@gmail.com>
---
 drivers/pci/pcie/aspm.c | 18 +++++++++++-------
 1 file changed, 11 insertions(+), 7 deletions(-)

diff --git a/drivers/pci/pcie/aspm.c b/drivers/pci/pcie/aspm.c
index 511f3e018..7847be38e 100644
--- a/drivers/pci/pcie/aspm.c
+++ b/drivers/pci/pcie/aspm.c
@@ -65,6 +65,7 @@ struct pcie_link_state {
 	u32 clkpm_capable:1;		/* Clock PM capable? */
 	u32 clkpm_enabled:1;		/* Current Clock PM state */
 	u32 clkpm_default:1;		/* Default Clock PM state by BIOS */
+	u32 clkpm_disable:1;		/* Clock PM disabled */
 
 	/* Exit latencies */
 	struct aspm_latency latency_up;	/* Upstream direction exit latency */
@@ -162,8 +163,11 @@ static void pcie_set_clkpm_nocheck(struct pcie_link_state *link, int enable)
 
 static void pcie_set_clkpm(struct pcie_link_state *link, int enable)
 {
-	/* Don't enable Clock PM if the link is not Clock PM capable */
-	if (!link->clkpm_capable)
+	/*
+	 * Don't enable Clock PM if the link is not Clock PM capable
+	 * or Clock PM is disabled
+	 */
+	if (!link->clkpm_capable || link->clkpm_disable)
 		enable = 0;
 	/* Need nothing if the specified equals to current state */
 	if (link->clkpm_enabled == enable)
@@ -193,7 +197,8 @@ static void pcie_clkpm_cap_init(struct pcie_link_state *link, int blacklist)
 	}
 	link->clkpm_enabled = enabled;
 	link->clkpm_default = enabled;
-	link->clkpm_capable = (blacklist) ? 0 : capable;
+	link->clkpm_capable = capable;
+	link->clkpm_disable = blacklist ? 1 : 0;
 }
 
 static bool pcie_retrain_link(struct pcie_link_state *link)
@@ -1105,10 +1110,9 @@ static void __pci_disable_link_state(struct pci_dev *pdev, int state, bool sem)
 		link->aspm_disable |= ASPM_STATE_L1_2_MASK;
 	pcie_config_aspm_link(link, policy_to_aspm_state(link));
 
-	if (state & PCIE_LINK_STATE_CLKPM) {
-		link->clkpm_capable = 0;
-		pcie_set_clkpm(link, 0);
-	}
+	if (state & PCIE_LINK_STATE_CLKPM)
+		link->clkpm_disable = 1;
+	pcie_set_clkpm(link, policy_to_clkpm_state(link));
 	mutex_unlock(&aspm_lock);
 	if (sem)
 		up_read(&pci_bus_sem);
-- 
2.21.0

[PATCH RFC 3/3] PCI/ASPM: add sysfs attribute for controlling ASPM

[Heiner Kallweit]

Background of this extension is a problem with the r8169 network driver.
Several combinations of board chipsets and network chip versions have
problems if ASPM is enabled, therefore we have to disable ASPM per
default. However especially on notebooks ASPM can provide significant
power-saving, therefore we want to give users the option to enable
ASPM. With the new sysfs attribute users can control which ASPM
link-states are disabled.

This is a first RFC version, therefore documentation of attribute is
still missing. The attribute handling was inspired by the protocol
attribute handling in drivers/media/rc/rc-main.c.
Attribute syntax in a few words:
none: allow all supported ASPM states
all: disable all ASPM states
+<state>: add state to list of disabled ASPM states
-<state>: re-enable ASPM state if supported

Signed-off-by: Heiner Kallweit <hkallweit1@gmail.com>
---
 drivers/pci/pci.h       |   8 +-
 drivers/pci/pcie/aspm.c | 173 +++++++++++++++++++++++++++++++++++++++-
 2 files changed, 173 insertions(+), 8 deletions(-)

diff --git a/drivers/pci/pci.h b/drivers/pci/pci.h
index d994839a3..6d4e5fc24 100644
--- a/drivers/pci/pci.h
+++ b/drivers/pci/pci.h
@@ -499,17 +499,13 @@ void pcie_aspm_init_link_state(struct pci_dev *pdev);
 void pcie_aspm_exit_link_state(struct pci_dev *pdev);
 void pcie_aspm_pm_state_change(struct pci_dev *pdev);
 void pcie_aspm_powersave_config_link(struct pci_dev *pdev);
+void pcie_aspm_create_sysfs_dev_files(struct pci_dev *pdev);
+void pcie_aspm_remove_sysfs_dev_files(struct pci_dev *pdev);
 #else
 static inline void pcie_aspm_init_link_state(struct pci_dev *pdev) { }
 static inline void pcie_aspm_exit_link_state(struct pci_dev *pdev) { }
 static inline void pcie_aspm_pm_state_change(struct pci_dev *pdev) { }
 static inline void pcie_aspm_powersave_config_link(struct pci_dev *pdev) { }
-#endif
-
-#ifdef CONFIG_PCIEASPM_DEBUG
-void pcie_aspm_create_sysfs_dev_files(struct pci_dev *pdev);
-void pcie_aspm_remove_sysfs_dev_files(struct pci_dev *pdev);
-#else
 static inline void pcie_aspm_create_sysfs_dev_files(struct pci_dev *pdev) { }
 static inline void pcie_aspm_remove_sysfs_dev_files(struct pci_dev *pdev) { }
 #endif
diff --git a/drivers/pci/pcie/aspm.c b/drivers/pci/pcie/aspm.c
index 7847be38e..bfbfd9134 100644
--- a/drivers/pci/pcie/aspm.c
+++ b/drivers/pci/pcie/aspm.c
@@ -42,6 +42,8 @@
 #define ASPM_STATE_ALL		(ASPM_STATE_L0S | ASPM_STATE_L1 |	\
 				 ASPM_STATE_L1SS)
 
+static const char power_group[] = "power";
+
 struct aspm_latency {
 	u32 l0s;			/* L0s latency (nsec) */
 	u32 l1;				/* L1 latency (nsec) */
@@ -1251,8 +1253,164 @@ static ssize_t clk_ctl_store(struct device *dev,
 
 static DEVICE_ATTR_RW(link_state);
 static DEVICE_ATTR_RW(clk_ctl);
+#endif
+
+struct aspm_sysfs_state {
+	const char *name;
+	int disable_mask;
+};
+
+static const struct aspm_sysfs_state aspm_sysfs_states[] = {
+	{ "L0S",	ASPM_STATE_L0S		},
+	{ "L1",		ASPM_STATE_L1		},
+	{ "L1.1",	ASPM_STATE_L1_1_MASK	},
+	{ "L1.2",	ASPM_STATE_L1_2_MASK	},
+};
+
+static struct pci_dev *aspm_get_parent(struct pci_dev *pdev)
+{
+	struct pci_dev *parent = pdev->bus->self;
+
+	if (pdev->has_secondary_link)
+		parent = pdev;
+
+	return parent && parent->link_state ? parent : NULL;
+}
+
+static ssize_t aspm_disable_link_state_show(struct device *dev,
+					    struct device_attribute *attr,
+					    char *buf)
+{
+	struct pci_dev *parent, *pdev = to_pci_dev(dev);
+	struct pcie_link_state *link;
+	int len = 0, i;
+
+	parent = aspm_get_parent(pdev);
+	if (!parent)
+		return -EOPNOTSUPP;
+
+	link = parent->link_state;
+
+	mutex_lock(&aspm_lock);
+
+	for (i = 0; i < ARRAY_SIZE(aspm_sysfs_states); i++) {
+		const struct aspm_sysfs_state *st = aspm_sysfs_states + i;
+
+		if (link->aspm_disable & st->disable_mask)
+			len += scnprintf(buf + len, PAGE_SIZE - len, "[%s] ",
+					 st->name);
+		else
+			len += scnprintf(buf + len, PAGE_SIZE - len, "%s ",
+					 st->name);
+	}
+
+	if (link->clkpm_disable)
+		len += scnprintf(buf + len, PAGE_SIZE - len, "[CLKPM] ");
+	else
+		len += scnprintf(buf + len, PAGE_SIZE - len, "CLKPM ");
+
+	mutex_unlock(&aspm_lock);
+
+	len += scnprintf(buf + len, PAGE_SIZE - len, "\n");
+
+	return len;
+}
+
+static ssize_t aspm_disable_link_state_store(struct device *dev,
+					     struct device_attribute *attr,
+					     const char *buf, size_t len)
+{
+	struct pci_dev *parent, *pdev = to_pci_dev(dev);
+	struct pcie_link_state *link;
+	char *buftmp = (char *)buf, *tok;
+	unsigned int disable_aspm, disable_clkpm;
+	bool first = true, add;
+	int err = 0, i;
+
+	if (aspm_disabled)
+		return -EPERM;
+
+	parent = aspm_get_parent(pdev);
+	if (!parent)
+		return -EOPNOTSUPP;
+
+	link = parent->link_state;
+
+	down_read(&pci_bus_sem);
+	mutex_lock(&aspm_lock);
+
+	disable_aspm = link->aspm_disable;
+	disable_clkpm = link->clkpm_disable;
+
+	while ((tok = strsep(&buftmp, " \n")) != NULL) {
+		bool found = false;
+
+		if (!*tok)
+			continue;
+
+		if (first) {
+			if (!strcasecmp(tok, "none")) {
+				disable_aspm = 0;
+				disable_clkpm = 0;
+				break;
+			}
+			if (!strcasecmp(tok, "all")) {
+				disable_aspm = ASPM_STATE_ALL;
+				disable_clkpm = 1;
+				break;
+			}
+			first = false;
+		}
+
+		if (*tok != '+' && *tok != '-') {
+			err = -EINVAL;
+			goto out;
+		}
+
+		add = *tok++ == '+';
+
+		for (i = 0; i < ARRAY_SIZE(aspm_sysfs_states); i++) {
+			const struct aspm_sysfs_state *st =
+						aspm_sysfs_states + i;
+
+			if (!strcasecmp(tok, st->name)) {
+				if (add)
+					disable_aspm |= st->disable_mask;
+				else
+					disable_aspm &= ~st->disable_mask;
+				found = true;
+				break;
+			}
+		}
+
+		if (!found && !strcasecmp(tok, "clkpm")) {
+			disable_clkpm = add ? 1 : 0;
+			found = true;
+		}
+
+		if (!found) {
+			err = -EINVAL;
+			goto out;
+		}
+	}
+
+	if (disable_aspm & ASPM_STATE_L1)
+		disable_aspm |= ASPM_STATE_L1SS;
+
+	link->aspm_disable = disable_aspm;
+	link->clkpm_disable = disable_clkpm;
+
+	pcie_config_aspm_link(link, policy_to_aspm_state(link));
+	pcie_set_clkpm(link, policy_to_clkpm_state(link));
+out:
+	mutex_unlock(&aspm_lock);
+	up_read(&pci_bus_sem);
+
+	return err ?: len;
+}
+
+static DEVICE_ATTR_RW(aspm_disable_link_state);
 
-static char power_group[] = "power";
 void pcie_aspm_create_sysfs_dev_files(struct pci_dev *pdev)
 {
 	struct pcie_link_state *link_state = pdev->link_state;
@@ -1260,12 +1418,18 @@ void pcie_aspm_create_sysfs_dev_files(struct pci_dev *pdev)
 	if (!link_state)
 		return;
 
+	if (link_state->aspm_support)
+		sysfs_add_file_to_group(&pdev->dev.kobj,
+			&dev_attr_aspm_disable_link_state.attr, power_group);
+
+#ifdef CONFIG_PCIEASPM_DEBUG
 	if (link_state->aspm_support)
 		sysfs_add_file_to_group(&pdev->dev.kobj,
 			&dev_attr_link_state.attr, power_group);
 	if (link_state->clkpm_capable)
 		sysfs_add_file_to_group(&pdev->dev.kobj,
 			&dev_attr_clk_ctl.attr, power_group);
+#endif
 }
 
 void pcie_aspm_remove_sysfs_dev_files(struct pci_dev *pdev)
@@ -1275,14 +1439,19 @@ void pcie_aspm_remove_sysfs_dev_files(struct pci_dev *pdev)
 	if (!link_state)
 		return;
 
+	if (link_state->aspm_support)
+		sysfs_remove_file_from_group(&pdev->dev.kobj,
+			&dev_attr_aspm_disable_link_state.attr, power_group);
+
+#ifdef CONFIG_PCIEASPM_DEBUG
 	if (link_state->aspm_support)
 		sysfs_remove_file_from_group(&pdev->dev.kobj,
 			&dev_attr_link_state.attr, power_group);
 	if (link_state->clkpm_capable)
 		sysfs_remove_file_from_group(&pdev->dev.kobj,
 			&dev_attr_clk_ctl.attr, power_group);
-}
 #endif
+}
 
 static int __init pcie_aspm_disable(char *str)
 {
-- 
2.21.0

Re: [PATCH RFC 3/3] PCI/ASPM: add sysfs attribute for controlling ASPM

[Bjorn Helgaas]

On Sat, Apr 13, 2019 at 11:12:41AM +0200, Heiner Kallweit wrote:
> Background of this extension is a problem with the r8169 network driver.
> Several combinations of board chipsets and network chip versions have
> problems if ASPM is enabled, therefore we have to disable ASPM per
> default. However especially on notebooks ASPM can provide significant
> power-saving, therefore we want to give users the option to enable
> ASPM. With the new sysfs attribute users can control which ASPM
> link-states are disabled.

> +void pcie_aspm_create_sysfs_dev_files(struct pci_dev *pdev);
> +void pcie_aspm_remove_sysfs_dev_files(struct pci_dev *pdev);
>  #else
>  static inline void pcie_aspm_init_link_state(struct pci_dev *pdev) { }
>  static inline void pcie_aspm_exit_link_state(struct pci_dev *pdev) { }
>  static inline void pcie_aspm_pm_state_change(struct pci_dev *pdev) { }
>  static inline void pcie_aspm_powersave_config_link(struct pci_dev *pdev) { }
> -#endif
> -
> -#ifdef CONFIG_PCIEASPM_DEBUG
> -void pcie_aspm_create_sysfs_dev_files(struct pci_dev *pdev);
> -void pcie_aspm_remove_sysfs_dev_files(struct pci_dev *pdev);
> -#else

I like the idea of exposing these sysfs control files all the time,
instead of only when CONFIG_PCIEASPM_DEBUG=y, but I think when we do
that, we should put the files at the downstream end of the link (e.g.,
an endpoint) instead of at the upstream end (e.g., a root port or
switch downstream port).  We had some conversation about this here:

https://lore.kernel.org/lkml/20180727202619.GD173328@bhelgaas-glaptop.roam.corp.google.com

Doing it at the downstream end would require more changes, of course,
and probably raises some locking issues, but I think we have a small
window of opportunity here where we can tweak the sysfs structure
before we're committed to supporting something forever.

Bjorn

Re: [PATCH RFC 3/3] PCI/ASPM: add sysfs attribute for controlling ASPM

[Heiner Kallweit]

On 30.04.2019 19:53, Bjorn Helgaas wrote:
> On Sat, Apr 13, 2019 at 11:12:41AM +0200, Heiner Kallweit wrote:
>> Background of this extension is a problem with the r8169 network driver.
>> Several combinations of board chipsets and network chip versions have
>> problems if ASPM is enabled, therefore we have to disable ASPM per
>> default. However especially on notebooks ASPM can provide significant
>> power-saving, therefore we want to give users the option to enable
>> ASPM. With the new sysfs attribute users can control which ASPM
>> link-states are disabled.
> 
>> +void pcie_aspm_create_sysfs_dev_files(struct pci_dev *pdev);
>> +void pcie_aspm_remove_sysfs_dev_files(struct pci_dev *pdev);
>>  #else
>>  static inline void pcie_aspm_init_link_state(struct pci_dev *pdev) { }
>>  static inline void pcie_aspm_exit_link_state(struct pci_dev *pdev) { }
>>  static inline void pcie_aspm_pm_state_change(struct pci_dev *pdev) { }
>>  static inline void pcie_aspm_powersave_config_link(struct pci_dev *pdev) { }
>> -#endif
>> -
>> -#ifdef CONFIG_PCIEASPM_DEBUG
>> -void pcie_aspm_create_sysfs_dev_files(struct pci_dev *pdev);
>> -void pcie_aspm_remove_sysfs_dev_files(struct pci_dev *pdev);
>> -#else
> 
> I like the idea of exposing these sysfs control files all the time,
> instead of only when CONFIG_PCIEASPM_DEBUG=y, but I think when we do
> that, we should put the files at the downstream end of the link (e.g.,
> an endpoint) instead of at the upstream end (e.g., a root port or
> switch downstream port).  We had some conversation about this here:
> 
> https://lore.kernel.org/lkml/20180727202619.GD173328@bhelgaas-glaptop.roam.corp.google.com
> 
> Doing it at the downstream end would require more changes, of course,
> and probably raises some locking issues, but I think we have a small
> window of opportunity here where we can tweak the sysfs structure
> before we're committed to supporting something forever.
> 
Thanks for the feedback. I'll check the mentioned conversation and try
to get the control moved to the endpoint.

> Bjorn
> 
Heiner

Re: [PATCH RFC 3/3] PCI/ASPM: add sysfs attribute for controlling ASPM

[Heiner Kallweit]

On 30.04.2019 19:53, Bjorn Helgaas wrote:
> On Sat, Apr 13, 2019 at 11:12:41AM +0200, Heiner Kallweit wrote:
>> Background of this extension is a problem with the r8169 network driver.
>> Several combinations of board chipsets and network chip versions have
>> problems if ASPM is enabled, therefore we have to disable ASPM per
>> default. However especially on notebooks ASPM can provide significant
>> power-saving, therefore we want to give users the option to enable
>> ASPM. With the new sysfs attribute users can control which ASPM
>> link-states are disabled.
> 
>> +void pcie_aspm_create_sysfs_dev_files(struct pci_dev *pdev);
>> +void pcie_aspm_remove_sysfs_dev_files(struct pci_dev *pdev);
>>  #else
>>  static inline void pcie_aspm_init_link_state(struct pci_dev *pdev) { }
>>  static inline void pcie_aspm_exit_link_state(struct pci_dev *pdev) { }
>>  static inline void pcie_aspm_pm_state_change(struct pci_dev *pdev) { }
>>  static inline void pcie_aspm_powersave_config_link(struct pci_dev *pdev) { }
>> -#endif
>> -
>> -#ifdef CONFIG_PCIEASPM_DEBUG
>> -void pcie_aspm_create_sysfs_dev_files(struct pci_dev *pdev);
>> -void pcie_aspm_remove_sysfs_dev_files(struct pci_dev *pdev);
>> -#else
> 
> I like the idea of exposing these sysfs control files all the time,
> instead of only when CONFIG_PCIEASPM_DEBUG=y, but I think when we do
> that, we should put the files at the downstream end of the link (e.g.,
> an endpoint) instead of at the upstream end (e.g., a root port or
> switch downstream port).  We had some conversation about this here:
> 
> https://lore.kernel.org/lkml/20180727202619.GD173328@bhelgaas-glaptop.roam.corp.google.com
> 
> Doing it at the downstream end would require more changes, of course,
> and probably raises some locking issues, but I think we have a small

This isn't obvious to me as I'm not that familiar with the PCIe subsystem.
- Why do we need more changes on the downstream end?
- Which locking is affected?

> window of opportunity here where we can tweak the sysfs structure
> before we're committed to supporting something forever.
> 
> Bjorn
> 
Heiner

Re: [PATCH RFC 3/3] PCI/ASPM: add sysfs attribute for controlling ASPM

[Heiner Kallweit]

On 30.04.2019 19:53, Bjorn Helgaas wrote:
> On Sat, Apr 13, 2019 at 11:12:41AM +0200, Heiner Kallweit wrote:
>> Background of this extension is a problem with the r8169 network driver.
>> Several combinations of board chipsets and network chip versions have
>> problems if ASPM is enabled, therefore we have to disable ASPM per
>> default. However especially on notebooks ASPM can provide significant
>> power-saving, therefore we want to give users the option to enable
>> ASPM. With the new sysfs attribute users can control which ASPM
>> link-states are disabled.
> 
>> +void pcie_aspm_create_sysfs_dev_files(struct pci_dev *pdev);
>> +void pcie_aspm_remove_sysfs_dev_files(struct pci_dev *pdev);
>>  #else
>>  static inline void pcie_aspm_init_link_state(struct pci_dev *pdev) { }
>>  static inline void pcie_aspm_exit_link_state(struct pci_dev *pdev) { }
>>  static inline void pcie_aspm_pm_state_change(struct pci_dev *pdev) { }
>>  static inline void pcie_aspm_powersave_config_link(struct pci_dev *pdev) { }
>> -#endif
>> -
>> -#ifdef CONFIG_PCIEASPM_DEBUG
>> -void pcie_aspm_create_sysfs_dev_files(struct pci_dev *pdev);
>> -void pcie_aspm_remove_sysfs_dev_files(struct pci_dev *pdev);
>> -#else
> 
> I like the idea of exposing these sysfs control files all the time,
> instead of only when CONFIG_PCIEASPM_DEBUG=y, but I think when we do
> that, we should put the files at the downstream end of the link (e.g.,
> an endpoint) instead of at the upstream end (e.g., a root port or
> switch downstream port).  We had some conversation about this here:
> 
> https://lore.kernel.org/lkml/20180727202619.GD173328@bhelgaas-glaptop.roam.corp.google.com
> 
> Doing it at the downstream end would require more changes, of course,
> and probably raises some locking issues, but I think we have a small
> window of opportunity here where we can tweak the sysfs structure
> before we're committed to supporting something forever.
> 
> Bjorn
> 

Here comes an updated version of patch 3. Unfortunately my test systems
don't allow OS ASPM control. At least the sysfs files are created for
the right devices (endpoints), and the attribute can be properly
controlled.


---
 drivers/pci/pci.h       |   8 +-
 drivers/pci/pcie/aspm.c | 171 +++++++++++++++++++++++++++++++++++++++-
 2 files changed, 171 insertions(+), 8 deletions(-)

diff --git a/drivers/pci/pci.h b/drivers/pci/pci.h
index 9cb99380c..06642b7de 100644
--- a/drivers/pci/pci.h
+++ b/drivers/pci/pci.h
@@ -499,17 +499,13 @@ void pcie_aspm_init_link_state(struct pci_dev *pdev);
 void pcie_aspm_exit_link_state(struct pci_dev *pdev);
 void pcie_aspm_pm_state_change(struct pci_dev *pdev);
 void pcie_aspm_powersave_config_link(struct pci_dev *pdev);
+void pcie_aspm_create_sysfs_dev_files(struct pci_dev *pdev);
+void pcie_aspm_remove_sysfs_dev_files(struct pci_dev *pdev);
 #else
 static inline void pcie_aspm_init_link_state(struct pci_dev *pdev) { }
 static inline void pcie_aspm_exit_link_state(struct pci_dev *pdev) { }
 static inline void pcie_aspm_pm_state_change(struct pci_dev *pdev) { }
 static inline void pcie_aspm_powersave_config_link(struct pci_dev *pdev) { }
-#endif
-
-#ifdef CONFIG_PCIEASPM_DEBUG
-void pcie_aspm_create_sysfs_dev_files(struct pci_dev *pdev);
-void pcie_aspm_remove_sysfs_dev_files(struct pci_dev *pdev);
-#else
 static inline void pcie_aspm_create_sysfs_dev_files(struct pci_dev *pdev) { }
 static inline void pcie_aspm_remove_sysfs_dev_files(struct pci_dev *pdev) { }
 #endif
diff --git a/drivers/pci/pcie/aspm.c b/drivers/pci/pcie/aspm.c
index 7847be38e..f73539e5b 100644
--- a/drivers/pci/pcie/aspm.c
+++ b/drivers/pci/pcie/aspm.c
@@ -42,6 +42,8 @@
 #define ASPM_STATE_ALL		(ASPM_STATE_L0S | ASPM_STATE_L1 |	\
 				 ASPM_STATE_L1SS)
 
+static const char power_group[] = "power";
+
 struct aspm_latency {
 	u32 l0s;			/* L0s latency (nsec) */
 	u32 l1;				/* L1 latency (nsec) */
@@ -1251,38 +1253,203 @@ static ssize_t clk_ctl_store(struct device *dev,
 
 static DEVICE_ATTR_RW(link_state);
 static DEVICE_ATTR_RW(clk_ctl);
+#endif
+
+struct aspm_sysfs_state {
+	const char *name;
+	int disable_mask;
+};
+
+static const struct aspm_sysfs_state aspm_sysfs_states[] = {
+	{ "L0S",	ASPM_STATE_L0S		},
+	{ "L1",		ASPM_STATE_L1		},
+	{ "L1.1",	ASPM_STATE_L1_1_MASK	},
+	{ "L1.2",	ASPM_STATE_L1_2_MASK	},
+};
+
+static struct pcie_link_state *aspm_get_parent_link(struct pci_dev *pdev)
+{
+	struct pci_dev *parent = pdev->bus->self;
+
+	if (pdev->has_secondary_link)
+		parent = pdev;
+
+	return parent ? parent->link_state : NULL;
+}
+
+static ssize_t aspm_disable_link_state_show(struct device *dev,
+					    struct device_attribute *attr,
+					    char *buf)
+{
+	struct pci_dev *pdev = to_pci_dev(dev);
+	struct pcie_link_state *link;
+	int len = 0, i;
+
+	link = aspm_get_parent_link(pdev);
+	if (!link)
+		return -EOPNOTSUPP;
+
+	mutex_lock(&aspm_lock);
+
+	for (i = 0; i < ARRAY_SIZE(aspm_sysfs_states); i++) {
+		const struct aspm_sysfs_state *st = aspm_sysfs_states + i;
+
+		if (link->aspm_disable & st->disable_mask)
+			len += scnprintf(buf + len, PAGE_SIZE - len, "[%s] ",
+					 st->name);
+		else
+			len += scnprintf(buf + len, PAGE_SIZE - len, "%s ",
+					 st->name);
+	}
+
+	if (link->clkpm_disable)
+		len += scnprintf(buf + len, PAGE_SIZE - len, "[CLKPM] ");
+	else
+		len += scnprintf(buf + len, PAGE_SIZE - len, "CLKPM ");
+
+	mutex_unlock(&aspm_lock);
+
+	len += scnprintf(buf + len, PAGE_SIZE - len, "\n");
+
+	return len;
+}
+
+static ssize_t aspm_disable_link_state_store(struct device *dev,
+					     struct device_attribute *attr,
+					     const char *buf, size_t len)
+{
+	struct pci_dev *pdev = to_pci_dev(dev);
+	struct pcie_link_state *link;
+	char *buftmp = (char *)buf, *tok;
+	unsigned int disable_aspm, disable_clkpm;
+	bool first = true, add;
+	int err = 0, i;
+
+	if (aspm_disabled)
+		return -EPERM;
+
+	link = aspm_get_parent_link(pdev);
+	if (!link)
+		return -EOPNOTSUPP;
+
+	down_read(&pci_bus_sem);
+	mutex_lock(&aspm_lock);
+
+	disable_aspm = link->aspm_disable;
+	disable_clkpm = link->clkpm_disable;
+
+	while ((tok = strsep(&buftmp, " \n")) != NULL) {
+		bool found = false;
+
+		if (!*tok)
+			continue;
+
+		if (first) {
+			if (!strcasecmp(tok, "none")) {
+				disable_aspm = 0;
+				disable_clkpm = 0;
+				break;
+			}
+			if (!strcasecmp(tok, "all")) {
+				disable_aspm = ASPM_STATE_ALL;
+				disable_clkpm = 1;
+				break;
+			}
+			first = false;
+		}
+
+		if (*tok != '+' && *tok != '-') {
+			err = -EINVAL;
+			goto out;
+		}
+
+		add = *tok++ == '+';
+
+		for (i = 0; i < ARRAY_SIZE(aspm_sysfs_states); i++) {
+			const struct aspm_sysfs_state *st =
+						aspm_sysfs_states + i;
+
+			if (!strcasecmp(tok, st->name)) {
+				if (add)
+					disable_aspm |= st->disable_mask;
+				else
+					disable_aspm &= ~st->disable_mask;
+				found = true;
+				break;
+			}
+		}
+
+		if (!found && !strcasecmp(tok, "clkpm")) {
+			disable_clkpm = add ? 1 : 0;
+			found = true;
+		}
+
+		if (!found) {
+			err = -EINVAL;
+			goto out;
+		}
+	}
+
+	if (disable_aspm & ASPM_STATE_L1)
+		disable_aspm |= ASPM_STATE_L1SS;
+
+	link->aspm_disable = disable_aspm;
+	link->clkpm_disable = disable_clkpm;
+
+	pcie_config_aspm_link(link, policy_to_aspm_state(link));
+	pcie_set_clkpm(link, policy_to_clkpm_state(link));
+out:
+	mutex_unlock(&aspm_lock);
+	up_read(&pci_bus_sem);
+
+	return err ?: len;
+}
+
+static DEVICE_ATTR_RW(aspm_disable_link_state);
 
-static char power_group[] = "power";
 void pcie_aspm_create_sysfs_dev_files(struct pci_dev *pdev)
 {
 	struct pcie_link_state *link_state = pdev->link_state;
 
+	if (pci_pcie_type(pdev) == PCI_EXP_TYPE_ENDPOINT &&
+	    aspm_get_parent_link(pdev))
+		sysfs_add_file_to_group(&pdev->dev.kobj,
+			&dev_attr_aspm_disable_link_state.attr, power_group);
+
 	if (!link_state)
 		return;
 
+#ifdef CONFIG_PCIEASPM_DEBUG
 	if (link_state->aspm_support)
 		sysfs_add_file_to_group(&pdev->dev.kobj,
 			&dev_attr_link_state.attr, power_group);
 	if (link_state->clkpm_capable)
 		sysfs_add_file_to_group(&pdev->dev.kobj,
 			&dev_attr_clk_ctl.attr, power_group);
+#endif
 }
 
 void pcie_aspm_remove_sysfs_dev_files(struct pci_dev *pdev)
 {
 	struct pcie_link_state *link_state = pdev->link_state;
 
+	if (pci_pcie_type(pdev) == PCI_EXP_TYPE_ENDPOINT &&
+	    aspm_get_parent_link(pdev))
+		sysfs_remove_file_from_group(&pdev->dev.kobj,
+			&dev_attr_aspm_disable_link_state.attr, power_group);
+
 	if (!link_state)
 		return;
 
+#ifdef CONFIG_PCIEASPM_DEBUG
 	if (link_state->aspm_support)
 		sysfs_remove_file_from_group(&pdev->dev.kobj,
 			&dev_attr_link_state.attr, power_group);
 	if (link_state->clkpm_capable)
 		sysfs_remove_file_from_group(&pdev->dev.kobj,
 			&dev_attr_clk_ctl.attr, power_group);
-}
 #endif
+}
 
 static int __init pcie_aspm_disable(char *str)
 {
-- 
2.21.0

Re: [PATCH RFC 3/3] PCI/ASPM: add sysfs attribute for controlling ASPM

[Bjorn Helgaas]

On Sat, May 04, 2019 at 05:31:04PM +0200, Heiner Kallweit wrote:
> On 30.04.2019 19:53, Bjorn Helgaas wrote:
> > On Sat, Apr 13, 2019 at 11:12:41AM +0200, Heiner Kallweit wrote:

> > I like the idea of exposing these sysfs control files all the time,
> > instead of only when CONFIG_PCIEASPM_DEBUG=y, but I think when we do
> > that, we should put the files at the downstream end of the link (e.g.,
> > an endpoint) instead of at the upstream end (e.g., a root port or
> > switch downstream port).  We had some conversation about this here:
> > 
> > https://lore.kernel.org/lkml/20180727202619.GD173328@bhelgaas-glaptop.roam.corp.google.com
> > 
> > Doing it at the downstream end would require more changes, of course,
> > and probably raises some locking issues, but I think we have a small
> 
> This isn't obvious to me as I'm not that familiar with the PCIe subsystem.
> - Why do we need more changes on the downstream end?

I just meant that it might be a little more complicated because the
existing PCIEASPM_DEBUG code works at the upstream end so you can't
use it directly.

> - Which locking is affected?

Normally if we operate on device X, we only touch registers in device
X.  Here we need to also touch things in the bridge upstream from X,
so it's outside the usual model.  We've had locking issues in the
device enable path where we sometimes have to enable the upstream
bridge before enabling the device itself.  I don't know if there are
similar issues here or not, so I mentioned it as a possibility.

> > window of opportunity here where we can tweak the sysfs structure
> > before we're committed to supporting something forever.
> > 
> > Bjorn
> > 
> Heiner