[PATCH v2 0/2] msi: extend msi_irqs sysfs entries to platform devices

[PATCH v2 0/2] msi: extend msi_irqs sysfs entries to platform devices

[Barry Song]

From: Barry Song <song.bao.hua@hisilicon.com>

Just like pci devices have msi_irqs which can be used by userspace irq affinity
tools or applications to bind irqs, platform devices also widely support msi
irqs.
For platform devices, for example ARM SMMU, userspaces also care about its msi
irqs as applications can know the mapping between devices and irqs and then
make smarter decision on handling irq affinity. For example, for SVA mode,
it is better to pin io page fault to the numa node applications are running
on. Otherwise, io page fault will get a remote page from the node iopf happens
rather than from the node applications are running on.

The first patch extracts the sysfs populate/destory code from PCI to
MSI core. The 2nd patch lets platform-msi export msi_irqs entry so that
userspace can know the mapping between devices and irqs for platform
devices.

-v2:
  extract common code for msi_irqs sysfs populate/destory from PCI to MSI core,
  platform_device can directly reuse common code;

-v1:
  https://lore.kernel.org/lkml/20210811105020.12980-1-song.bao.hua@hisilicon.com/

Barry Song (2):
  genirq/msi: extract common sysfs populate entries to msi core from pci
  platform-msi: Add ABI to show msi_irqs of platform devices

 Documentation/ABI/testing/sysfs-bus-platform |  14 +++
 drivers/base/platform-msi.c                  |  10 ++
 drivers/pci/msi.c                            | 124 ++-----------------------
 include/linux/msi.h                          |   4 +
 kernel/irq/msi.c                             | 134 +++++++++++++++++++++++++++
 5 files changed, 171 insertions(+), 115 deletions(-)

-- 
1.8.3.1

[PATCH v2 1/2] genirq/msi: extract common sysfs populate entries to msi core from pci

[Barry Song]

From: Barry Song <song.bao.hua@hisilicon.com>

This patch makes PCI's msi sysfs code common in msi so that other busses
such as platform can reuse it.

Signed-off-by: Barry Song <song.bao.hua@hisilicon.com>
---
 drivers/pci/msi.c   | 124 ++++--------------------------------------------
 include/linux/msi.h |   4 ++
 kernel/irq/msi.c    | 134 ++++++++++++++++++++++++++++++++++++++++++++++++++++
 3 files changed, 147 insertions(+), 115 deletions(-)

diff --git a/drivers/pci/msi.c b/drivers/pci/msi.c
index 9232255..4b9c0bb 100644
--- a/drivers/pci/msi.c
+++ b/drivers/pci/msi.c
@@ -338,9 +338,7 @@ static void free_msi_irqs(struct pci_dev *dev)
 {
 	struct list_head *msi_list = dev_to_msi_list(&dev->dev);
 	struct msi_desc *entry, *tmp;
-	struct attribute **msi_attrs;
-	struct device_attribute *dev_attr;
-	int i, count = 0;
+	int i;
 
 	for_each_pci_msi_entry(entry, dev)
 		if (entry->irq)
@@ -360,18 +358,7 @@ static void free_msi_irqs(struct pci_dev *dev)
 	}
 
 	if (dev->msi_irq_groups) {
-		sysfs_remove_groups(&dev->dev.kobj, dev->msi_irq_groups);
-		msi_attrs = dev->msi_irq_groups[0]->attrs;
-		while (msi_attrs[count]) {
-			dev_attr = container_of(msi_attrs[count],
-						struct device_attribute, attr);
-			kfree(dev_attr->attr.name);
-			kfree(dev_attr);
-			++count;
-		}
-		kfree(msi_attrs);
-		kfree(dev->msi_irq_groups[0]);
-		kfree(dev->msi_irq_groups);
+		msi_destroy_sysfs(&dev->dev, dev->msi_irq_groups);
 		dev->msi_irq_groups = NULL;
 	}
 }
@@ -452,102 +439,6 @@ void pci_restore_msi_state(struct pci_dev *dev)
 }
 EXPORT_SYMBOL_GPL(pci_restore_msi_state);
 
-static ssize_t msi_mode_show(struct device *dev, struct device_attribute *attr,
-			     char *buf)
-{
-	struct msi_desc *entry;
-	unsigned long irq;
-	int retval;
-
-	retval = kstrtoul(attr->attr.name, 10, &irq);
-	if (retval)
-		return retval;
-
-	entry = irq_get_msi_desc(irq);
-	if (!entry)
-		return -ENODEV;
-
-	return sysfs_emit(buf, "%s\n",
-			  entry->msi_attrib.is_msix ? "msix" : "msi");
-}
-
-static int populate_msi_sysfs(struct pci_dev *pdev)
-{
-	struct attribute **msi_attrs;
-	struct attribute *msi_attr;
-	struct device_attribute *msi_dev_attr;
-	struct attribute_group *msi_irq_group;
-	const struct attribute_group **msi_irq_groups;
-	struct msi_desc *entry;
-	int ret = -ENOMEM;
-	int num_msi = 0;
-	int count = 0;
-	int i;
-
-	/* Determine how many msi entries we have */
-	for_each_pci_msi_entry(entry, pdev)
-		num_msi += entry->nvec_used;
-	if (!num_msi)
-		return 0;
-
-	/* Dynamically create the MSI attributes for the PCI device */
-	msi_attrs = kcalloc(num_msi + 1, sizeof(void *), GFP_KERNEL);
-	if (!msi_attrs)
-		return -ENOMEM;
-	for_each_pci_msi_entry(entry, pdev) {
-		for (i = 0; i < entry->nvec_used; i++) {
-			msi_dev_attr = kzalloc(sizeof(*msi_dev_attr), GFP_KERNEL);
-			if (!msi_dev_attr)
-				goto error_attrs;
-			msi_attrs[count] = &msi_dev_attr->attr;
-
-			sysfs_attr_init(&msi_dev_attr->attr);
-			msi_dev_attr->attr.name = kasprintf(GFP_KERNEL, "%d",
-							    entry->irq + i);
-			if (!msi_dev_attr->attr.name)
-				goto error_attrs;
-			msi_dev_attr->attr.mode = S_IRUGO;
-			msi_dev_attr->show = msi_mode_show;
-			++count;
-		}
-	}
-
-	msi_irq_group = kzalloc(sizeof(*msi_irq_group), GFP_KERNEL);
-	if (!msi_irq_group)
-		goto error_attrs;
-	msi_irq_group->name = "msi_irqs";
-	msi_irq_group->attrs = msi_attrs;
-
-	msi_irq_groups = kcalloc(2, sizeof(void *), GFP_KERNEL);
-	if (!msi_irq_groups)
-		goto error_irq_group;
-	msi_irq_groups[0] = msi_irq_group;
-
-	ret = sysfs_create_groups(&pdev->dev.kobj, msi_irq_groups);
-	if (ret)
-		goto error_irq_groups;
-	pdev->msi_irq_groups = msi_irq_groups;
-
-	return 0;
-
-error_irq_groups:
-	kfree(msi_irq_groups);
-error_irq_group:
-	kfree(msi_irq_group);
-error_attrs:
-	count = 0;
-	msi_attr = msi_attrs[count];
-	while (msi_attr) {
-		msi_dev_attr = container_of(msi_attr, struct device_attribute, attr);
-		kfree(msi_attr->name);
-		kfree(msi_dev_attr);
-		++count;
-		msi_attr = msi_attrs[count];
-	}
-	kfree(msi_attrs);
-	return ret;
-}
-
 static struct msi_desc *
 msi_setup_entry(struct pci_dev *dev, int nvec, struct irq_affinity *affd)
 {
@@ -648,8 +539,9 @@ static int msi_capability_init(struct pci_dev *dev, int nvec,
 		return ret;
 	}
 
-	ret = populate_msi_sysfs(dev);
-	if (ret) {
+	dev->msi_irq_groups = msi_populate_sysfs(&dev->dev);
+	if (IS_ERR(dev->msi_irq_groups)) {
+		ret = PTR_ERR(dev->msi_irq_groups);
 		msi_mask_irq(entry, mask, ~mask);
 		free_msi_irqs(dev);
 		return ret;
@@ -804,9 +696,11 @@ static int msix_capability_init(struct pci_dev *dev, struct msix_entry *entries,
 
 	msix_program_entries(dev, entries);
 
-	ret = populate_msi_sysfs(dev);
-	if (ret)
+	dev->msi_irq_groups = msi_populate_sysfs(&dev->dev);
+	if (IS_ERR(dev->msi_irq_groups)) {
+		ret = PTR_ERR(dev->msi_irq_groups);
 		goto out_free;
+	}
 
 	/* Set MSI-X enabled bits and unmask the function */
 	pci_intx_for_msi(dev, 0);
diff --git a/include/linux/msi.h b/include/linux/msi.h
index 6aff469..375f1f9 100644
--- a/include/linux/msi.h
+++ b/include/linux/msi.h
@@ -237,6 +237,10 @@ struct msi_desc *alloc_msi_entry(struct device *dev, int nvec,
 void pci_msi_mask_irq(struct irq_data *data);
 void pci_msi_unmask_irq(struct irq_data *data);
 
+const struct attribute_group **msi_populate_sysfs(struct device *dev);
+void msi_destroy_sysfs(struct device *dev,
+		       const struct attribute_group **msi_irq_groups);
+
 /*
  * The arch hooks to setup up msi irqs. Default functions are implemented
  * as weak symbols so that they /can/ be overriden by architecture specific
diff --git a/kernel/irq/msi.c b/kernel/irq/msi.c
index c41965e..f153bca 100644
--- a/kernel/irq/msi.c
+++ b/kernel/irq/msi.c
@@ -14,6 +14,7 @@
 #include <linux/irqdomain.h>
 #include <linux/msi.h>
 #include <linux/slab.h>
+#include <linux/pci.h>
 
 #include "internals.h"
 
@@ -69,6 +70,139 @@ void get_cached_msi_msg(unsigned int irq, struct msi_msg *msg)
 }
 EXPORT_SYMBOL_GPL(get_cached_msi_msg);
 
+static ssize_t msi_mode_show(struct device *dev, struct device_attribute *attr,
+			     char *buf)
+{
+	struct msi_desc *entry;
+	unsigned long irq;
+	int retval;
+	bool is_msix = false;
+
+	retval = kstrtoul(attr->attr.name, 10, &irq);
+	if (retval)
+		return retval;
+
+	entry = irq_get_msi_desc(irq);
+	if (!entry)
+		return -ENODEV;
+
+	if (dev_is_pci(dev))
+		is_msix = entry->msi_attrib.is_msix;
+
+	return sysfs_emit(buf, "%s\n", is_msix ? "msix" : "msi");
+}
+
+/**
+ * msi_populate_sysfs - Populate msi_irqs sysfs entries for devices
+ * @dev:	The device(PCI, platform etc) who will get sysfs entries
+ *
+ * Return attribute_group ** so that specific bus MSI can save it to
+ * somewhere during initilizing msi irqs. If devices has no MSI irq,
+ * return NULL; if it fails to populate sysfs, return ERR_PTR
+ */
+const struct attribute_group **msi_populate_sysfs(struct device *dev)
+{
+	struct attribute **msi_attrs;
+	struct attribute *msi_attr;
+	struct device_attribute *msi_dev_attr;
+	struct attribute_group *msi_irq_group;
+	const struct attribute_group **msi_irq_groups;
+	struct msi_desc *entry;
+	int ret = -ENOMEM;
+	int num_msi = 0;
+	int count = 0;
+	int i;
+
+	/* Determine how many msi entries we have */
+	for_each_msi_entry(entry, dev)
+		num_msi += entry->nvec_used;
+	if (!num_msi)
+		return NULL;
+
+	/* Dynamically create the MSI attributes for the device */
+	msi_attrs = kcalloc(num_msi + 1, sizeof(void *), GFP_KERNEL);
+	if (!msi_attrs)
+		return ERR_PTR(-ENOMEM);
+	for_each_msi_entry(entry, dev) {
+		for (i = 0; i < entry->nvec_used; i++) {
+			msi_dev_attr = kzalloc(sizeof(*msi_dev_attr), GFP_KERNEL);
+			if (!msi_dev_attr)
+				goto error_attrs;
+			msi_attrs[count] = &msi_dev_attr->attr;
+
+			sysfs_attr_init(&msi_dev_attr->attr);
+			msi_dev_attr->attr.name = kasprintf(GFP_KERNEL, "%d",
+							    entry->irq + i);
+			if (!msi_dev_attr->attr.name)
+				goto error_attrs;
+			msi_dev_attr->attr.mode = 0444;
+			msi_dev_attr->show = msi_mode_show;
+			++count;
+		}
+	}
+
+	msi_irq_group = kzalloc(sizeof(*msi_irq_group), GFP_KERNEL);
+	if (!msi_irq_group)
+		goto error_attrs;
+	msi_irq_group->name = "msi_irqs";
+	msi_irq_group->attrs = msi_attrs;
+
+	msi_irq_groups = kcalloc(2, sizeof(void *), GFP_KERNEL);
+	if (!msi_irq_groups)
+		goto error_irq_group;
+	msi_irq_groups[0] = msi_irq_group;
+
+	ret = sysfs_create_groups(&dev->kobj, msi_irq_groups);
+	if (ret)
+		goto error_irq_groups;
+
+	return msi_irq_groups;
+
+error_irq_groups:
+	kfree(msi_irq_groups);
+error_irq_group:
+	kfree(msi_irq_group);
+error_attrs:
+	count = 0;
+	msi_attr = msi_attrs[count];
+	while (msi_attr) {
+		msi_dev_attr = container_of(msi_attr, struct device_attribute, attr);
+		kfree(msi_attr->name);
+		kfree(msi_dev_attr);
+		++count;
+		msi_attr = msi_attrs[count];
+	}
+	kfree(msi_attrs);
+	return ERR_PTR(ret);
+}
+
+/**
+ * msi_destroy_sysfs - Destroy msi_irqs sysfs entries for devices
+ * @dev:		The device(PCI, platform etc) who will remove sysfs entries
+ * @msi_irq_groups:	attribute_group for device msi_irqs entries
+ */
+void msi_destroy_sysfs(struct device *dev, const struct attribute_group **msi_irq_groups)
+{
+	struct attribute **msi_attrs;
+	struct device_attribute *dev_attr;
+	int count = 0;
+
+	if (msi_irq_groups) {
+		sysfs_remove_groups(&dev->kobj, msi_irq_groups);
+		msi_attrs = msi_irq_groups[0]->attrs;
+		while (msi_attrs[count]) {
+			dev_attr = container_of(msi_attrs[count],
+					struct device_attribute, attr);
+			kfree(dev_attr->attr.name);
+			kfree(dev_attr);
+			++count;
+		}
+		kfree(msi_attrs);
+		kfree(msi_irq_groups[0]);
+		kfree(msi_irq_groups);
+	}
+}
+
 #ifdef CONFIG_GENERIC_MSI_IRQ_DOMAIN
 static inline void irq_chip_write_msi_msg(struct irq_data *data,
 					  struct msi_msg *msg)
-- 
1.8.3.1

[PATCH v2 2/2] platform-msi: Add ABI to show msi_irqs of platform devices

[Barry Song]

From: Barry Song <song.bao.hua@hisilicon.com>

Just like pci devices have msi_irqs which can be used by userspace irq affinity
tools or applications to bind irqs, platform devices also widely support msi
irqs. For platform devices, for example ARM SMMU, userspaces also care about
its msi irqs as applications can know the mapping between devices and irqs
and then make smarter decision on handling irq affinity. For example, for
SVA mode, it is better to pin io page fault to the numa node applications
are running on. Otherwise, io page fault will get a remote page from the
node iopf happens.
With this patch, a system with multiple arm SMMUs in multiple different numa
node can get the mapping between devices and irqs now:
root@ubuntu:/sys/devices/platform# ls -l arm-smmu-v3.*/msi_irqs/*
-r--r--r-- 1 root root 4096 Aug 11 10:29 arm-smmu-v3.0.auto/msi_irqs/25
-r--r--r-- 1 root root 4096 Aug 11 10:29 arm-smmu-v3.0.auto/msi_irqs/26
-r--r--r-- 1 root root 4096 Aug 11 10:28 arm-smmu-v3.1.auto/msi_irqs/27
-r--r--r-- 1 root root 4096 Aug 11 10:28 arm-smmu-v3.1.auto/msi_irqs/28
-r--r--r-- 1 root root 4096 Aug 11 10:29 arm-smmu-v3.2.auto/msi_irqs/29
-r--r--r-- 1 root root 4096 Aug 11 10:29 arm-smmu-v3.2.auto/msi_irqs/30
-r--r--r-- 1 root root 4096 Aug 11 10:29 arm-smmu-v3.3.auto/msi_irqs/31
-r--r--r-- 1 root root 4096 Aug 11 10:29 arm-smmu-v3.3.auto/msi_irqs/32
-r--r--r-- 1 root root 4096 Aug 11 10:29 arm-smmu-v3.4.auto/msi_irqs/33
-r--r--r-- 1 root root 4096 Aug 11 10:29 arm-smmu-v3.4.auto/msi_irqs/34
-r--r--r-- 1 root root 4096 Aug 11 10:29 arm-smmu-v3.5.auto/msi_irqs/35
-r--r--r-- 1 root root 4096 Aug 11 10:29 arm-smmu-v3.5.auto/msi_irqs/36
-r--r--r-- 1 root root 4096 Aug 11 10:29 arm-smmu-v3.6.auto/msi_irqs/37
-r--r--r-- 1 root root 4096 Aug 11 10:29 arm-smmu-v3.6.auto/msi_irqs/38
-r--r--r-- 1 root root 4096 Aug 11 10:29 arm-smmu-v3.7.auto/msi_irqs/39
-r--r--r-- 1 root root 4096 Aug 11 10:29 arm-smmu-v3.7.auto/msi_irqs/40

Applications can use the mapping and the numa node information to pin
irqs by leveraging the numa information which has also been exported:
root@ubuntu:/sys/devices/platform# cat arm-smmu-v3.0.auto/numa_node
0
root@ubuntu:/sys/devices/platform# cat arm-smmu-v3.4.auto/numa_node
2

Cc: Zhou Wang <wangzhou1@hisilicon.com>
Signed-off-by: Barry Song <song.bao.hua@hisilicon.com>
---
 Documentation/ABI/testing/sysfs-bus-platform | 14 ++++++++++++++
 drivers/base/platform-msi.c                  | 10 ++++++++++
 2 files changed, 24 insertions(+)

diff --git a/Documentation/ABI/testing/sysfs-bus-platform b/Documentation/ABI/testing/sysfs-bus-platform
index 194ca70..ff30728 100644
--- a/Documentation/ABI/testing/sysfs-bus-platform
+++ b/Documentation/ABI/testing/sysfs-bus-platform
@@ -28,3 +28,17 @@ Description:
 		value comes from an ACPI _PXM method or a similar firmware
 		source. Initial users for this file would be devices like
 		arm smmu which are populated by arm64 acpi_iort.
+
+What:		/sys/bus/platform/devices/.../msi_irqs/
+Date:		August 2021
+Contact:	Barry Song <song.bao.hua@hisilicon.com>
+Description:
+		The /sys/devices/.../msi_irqs directory contains a variable set
+		of files, with each file being named after a corresponding msi
+		irq vector allocated to that device.
+
+What:		/sys/bus/platform/devices/.../msi_irqs/<N>
+Date:		August 2021
+Contact:	Barry Song <song.bao.hua@hisilicon.com>
+Description:
+		This attribute will show "msi" if <N> is a valid msi irq
diff --git a/drivers/base/platform-msi.c b/drivers/base/platform-msi.c
index 0b72b13..a3bf910 100644
--- a/drivers/base/platform-msi.c
+++ b/drivers/base/platform-msi.c
@@ -23,6 +23,7 @@
 struct platform_msi_priv_data {
 	struct device		*dev;
 	void 			*host_data;
+	const struct attribute_group    **msi_irq_groups;
 	msi_alloc_info_t	arg;
 	irq_write_msi_msg_t	write_msg;
 	int			devid;
@@ -272,8 +273,16 @@ int platform_msi_domain_alloc_irqs(struct device *dev, unsigned int nvec,
 	if (err)
 		goto out_free_desc;
 
+	priv_data->msi_irq_groups = msi_populate_sysfs(dev);
+	if (IS_ERR(priv_data->msi_irq_groups)) {
+		err = PTR_ERR(priv_data->msi_irq_groups);
+		goto out_free_irqs;
+	}
+
 	return 0;
 
+out_free_irqs:
+	msi_domain_free_irqs(dev->msi_domain, dev);
 out_free_desc:
 	platform_msi_free_descs(dev, 0, nvec);
 out_free_priv_data:
@@ -293,6 +302,7 @@ void platform_msi_domain_free_irqs(struct device *dev)
 		struct msi_desc *desc;
 
 		desc = first_msi_entry(dev);
+		msi_destroy_sysfs(dev, desc->platform.msi_priv_data->msi_irq_groups);
 		platform_msi_free_priv_data(desc->platform.msi_priv_data);
 	}
 
-- 
1.8.3.1

Re: [PATCH v2 0/2] msi: extend msi_irqs sysfs entries to platform devices

[Greg KH]

On Thu, Aug 12, 2021 at 10:53:39PM +1200, Barry Song wrote:
> From: Barry Song <song.bao.hua@hisilicon.com>
> 
> Just like pci devices have msi_irqs which can be used by userspace irq affinity
> tools or applications to bind irqs, platform devices also widely support msi
> irqs.
> For platform devices, for example ARM SMMU, userspaces also care about its msi
> irqs as applications can know the mapping between devices and irqs and then
> make smarter decision on handling irq affinity. For example, for SVA mode,
> it is better to pin io page fault to the numa node applications are running
> on. Otherwise, io page fault will get a remote page from the node iopf happens
> rather than from the node applications are running on.
> 
> The first patch extracts the sysfs populate/destory code from PCI to
> MSI core. The 2nd patch lets platform-msi export msi_irqs entry so that
> userspace can know the mapping between devices and irqs for platform
> devices.

Acked-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>

Re: [PATCH v2 0/2] msi: extend msi_irqs sysfs entries to platform devices

[Bjorn Helgaas]

On Thu, Aug 12, 2021 at 10:53:39PM +1200, Barry Song wrote:
> From: Barry Song <song.bao.hua@hisilicon.com>
> 
> Just like pci devices have msi_irqs which can be used by userspace irq affinity
> tools or applications to bind irqs, platform devices also widely support msi
> irqs.
> For platform devices, for example ARM SMMU, userspaces also care about its msi
> irqs as applications can know the mapping between devices and irqs and then
> make smarter decision on handling irq affinity. For example, for SVA mode,
> it is better to pin io page fault to the numa node applications are running
> on. Otherwise, io page fault will get a remote page from the node iopf happens
> rather than from the node applications are running on.
> 
> The first patch extracts the sysfs populate/destory code from PCI to
> MSI core. The 2nd patch lets platform-msi export msi_irqs entry so that
> userspace can know the mapping between devices and irqs for platform
> devices.
> 
> -v2:
>   extract common code for msi_irqs sysfs populate/destory from PCI to MSI core,
>   platform_device can directly reuse common code;
> 
> -v1:
>   https://lore.kernel.org/lkml/20210811105020.12980-1-song.bao.hua@hisilicon.com/
> 
> Barry Song (2):
>   genirq/msi: extract common sysfs populate entries to msi core from pci
>   platform-msi: Add ABI to show msi_irqs of platform devices
> 
>  Documentation/ABI/testing/sysfs-bus-platform |  14 +++
>  drivers/base/platform-msi.c                  |  10 ++
>  drivers/pci/msi.c                            | 124 ++-----------------------
>  include/linux/msi.h                          |   4 +
>  kernel/irq/msi.c                             | 134 +++++++++++++++++++++++++++
>  5 files changed, 171 insertions(+), 115 deletions(-)

I assume the IRQ guys will take care of this.

For the drivers/pci/ part:

Acked-by: Bjorn Helgaas <bhelgaas@google.com>

But I would update the commit logs to consistently capitalize
initialisms.  Currently it's a mix of "PCI", "pci", "MSI", "msi",
"numa", "irq", "io", etc.

Also, if you rewrap the 2/2 commit log to fit in 75 columns, you won't
have a line that becomes 83 columns when "git log" indents it.  Maybe
also indent quoted things like the "ls" output by 2 spaces and add a
blank line before so the text doesn't run into them.