[patch V2 00/21] genirq, PCI/MSI: Support for per device MSI and PCI/IMS - Part 2 API rework

[patch V2 00/21] genirq, PCI/MSI: Support for per device MSI and PCI/IMS - Part 2 API rework

[Thomas Gleixner]

This is V2 of the second part of effort to provide support for per device
MSI interrupt domains.

Version 1 of this second part can be found here:

  https://lore.kernel.org/all/20221111132706.104870257@linutronix.de

The first part is available here:

  https://lore.kernel.org/all/20221111120501.026511281@linutronix.de

and has been merged into:

 git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip.git irq/core

This part on top of the tip irq/core branch is also available here:

   git://git.kernel.org/pub/scm/linux/kernel/git/tglx/devel.git devmsi-v2-part2

This part is the main preparation step for per device MSI domains:

  1) Introduce irqdomain pointer storage space in device::msi::data
     and related helpers.

  2) Convert interfaces to handle multiple per device MSI domains
     based on a domain ID

  3) Provide new interfaces for allocation/free which are domain ID
     based and provide range allocation/free which is a prerequisite
     for post MSI-X enable alloc/free.

  4) Switch all existing call sites of msi allocation/free interfaces
     over to the new interfaces

  5) Remove the old interfaces

Changes vs. V1:

  - Split the kernel doc changes from the rename (Jason)

  - Rename msi_ctrl_valid() to msi_ctrl_range_valid() (Kevin)

  - Fix the off by one vs. MSI_MAX_INDEX (Kevin)

  - Update changelogs and comments (Bjorn, Jason, Kevin)

  - Fix the kernel robot fallout from randconfig builds

  - Picked up Reviewed/Acked-by tags where appropriate

Thanks,

	tglx
---
 arch/x86/kernel/apic/msi.c       |    5 
 drivers/base/platform-msi.c      |    4 
 drivers/bus/fsl-mc/fsl-mc-msi.c  |   25 -
 drivers/irqchip/irq-gic.c        |    4 
 drivers/pci/msi/irqdomain.c      |    4 
 drivers/pci/msi/msi.c            |    4 
 drivers/soc/ti/ti_sci_inta_msi.c |   12 
 include/linux/irqdomain.h        |  112 ++++----
 include/linux/msi.h              |  125 +++++++--
 include/linux/msi_api.h          |   35 ++
 kernel/irq/chip.c                |    8 
 kernel/irq/msi.c                 |  541 +++++++++++++++++++++++++++++----------
 12 files changed, 631 insertions(+), 248 deletions(-)

[patch V2 01/21] genirq/msi: Move IRQ_DOMAIN_MSI_NOMASK_QUIRK to MSI flags

[Thomas Gleixner]

It's truly a MSI only flag and for the upcoming per device MSI domains this
must be in the MSI flags so it can be set during domain setup without
exposing this quirk outside of x86.

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Reviewed-by: Jason Gunthorpe <jgg@nvidia.com>
---
 arch/x86/kernel/apic/msi.c |    5 ++---
 include/linux/irqdomain.h  |    9 +--------
 include/linux/msi.h        |    6 ++++++
 kernel/irq/msi.c           |    2 +-
 4 files changed, 10 insertions(+), 12 deletions(-)

--- a/arch/x86/kernel/apic/msi.c
+++ b/arch/x86/kernel/apic/msi.c
@@ -176,7 +176,8 @@ static struct msi_domain_ops pci_msi_dom
 
 static struct msi_domain_info pci_msi_domain_info = {
 	.flags		= MSI_FLAG_USE_DEF_DOM_OPS | MSI_FLAG_USE_DEF_CHIP_OPS |
-			  MSI_FLAG_PCI_MSIX,
+			  MSI_FLAG_PCI_MSIX | MSI_FLAG_NOMASK_QUIRK,
+
 	.ops		= &pci_msi_domain_ops,
 	.chip		= &pci_msi_controller,
 	.handler	= handle_edge_irq,
@@ -200,8 +201,6 @@ struct irq_domain * __init native_create
 	if (!d) {
 		irq_domain_free_fwnode(fn);
 		pr_warn("Failed to initialize PCI-MSI irqdomain.\n");
-	} else {
-		d->flags |= IRQ_DOMAIN_MSI_NOMASK_QUIRK;
 	}
 	return d;
 }
--- a/include/linux/irqdomain.h
+++ b/include/linux/irqdomain.h
@@ -186,15 +186,8 @@ enum {
 	/* Irq domain implements MSI remapping */
 	IRQ_DOMAIN_FLAG_MSI_REMAP	= (1 << 5),
 
-	/*
-	 * Quirk to handle MSI implementations which do not provide
-	 * masking. Currently known to affect x86, but partially
-	 * handled in core code.
-	 */
-	IRQ_DOMAIN_MSI_NOMASK_QUIRK	= (1 << 6),
-
 	/* Irq domain doesn't translate anything */
-	IRQ_DOMAIN_FLAG_NO_MAP		= (1 << 7),
+	IRQ_DOMAIN_FLAG_NO_MAP		= (1 << 6),
 
 	/*
 	 * Flags starting from IRQ_DOMAIN_FLAG_NONCORE are reserved
--- a/include/linux/msi.h
+++ b/include/linux/msi.h
@@ -399,6 +399,12 @@ enum {
 	MSI_FLAG_ALLOC_SIMPLE_MSI_DESCS	= (1 << 9),
 	/* Free MSI descriptors */
 	MSI_FLAG_FREE_MSI_DESCS		= (1 << 10),
+	/*
+	 * Quirk to handle MSI implementations which do not provide
+	 * masking. Currently known to affect x86, but has to be partially
+	 * handled in the core MSI code.
+	 */
+	MSI_FLAG_NOMASK_QUIRK		= (1 << 11),
 };
 
 int msi_domain_set_affinity(struct irq_data *data, const struct cpumask *mask,
--- a/kernel/irq/msi.c
+++ b/kernel/irq/msi.c
@@ -875,7 +875,7 @@ static int __msi_domain_alloc_irqs(struc
 		 * MSI affinity setting requires a special quirk (X86) when
 		 * reservation mode is active.
 		 */
-		if (domain->flags & IRQ_DOMAIN_MSI_NOMASK_QUIRK)
+		if (info->flags & MSI_FLAG_NOMASK_QUIRK)
 			vflags |= VIRQ_NOMASK_QUIRK;
 	}
 

Re: [patch V2 01/21] genirq/msi: Move IRQ_DOMAIN_MSI_NOMASK_QUIRK to MSI flags

[Marc Zyngier]

On Mon, 21 Nov 2022 14:36:19 +0000,
Thomas Gleixner <tglx@linutronix.de> wrote:
> 
> It's truly a MSI only flag and for the upcoming per device MSI domains this
> must be in the MSI flags so it can be set during domain setup without
> exposing this quirk outside of x86.
> 
> Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
> Reviewed-by: Jason Gunthorpe <jgg@nvidia.com>
> ---
>  arch/x86/kernel/apic/msi.c |    5 ++---
>  include/linux/irqdomain.h  |    9 +--------
>  include/linux/msi.h        |    6 ++++++
>  kernel/irq/msi.c           |    2 +-
>  4 files changed, 10 insertions(+), 12 deletions(-)

Acked-by: Marc Zyngier <maz@kernel.org>

	M.

-- 
Without deviation from the norm, progress is not possible.

[patch V2 02/21] genirq/irqdomain: Make struct irqdomain readable

[Thomas Gleixner]

Tabular alignment of both kernel-doc and the actual struct declaration make
visual parsing way more conveniant.

No functional change.

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
---
V2: Split out from the irqdomain::dev rename patch (Jason)
---
 include/linux/irqdomain.h |   74 +++++++++++++++++++++++-----------------------
 1 file changed, 37 insertions(+), 37 deletions(-)

--- a/include/linux/irqdomain.h
+++ b/include/linux/irqdomain.h
@@ -117,53 +117,53 @@ struct irq_domain_chip_generic;
 
 /**
  * struct irq_domain - Hardware interrupt number translation object
- * @link: Element in global irq_domain list.
- * @name: Name of interrupt domain
- * @ops: pointer to irq_domain methods
- * @host_data: private data pointer for use by owner.  Not touched by irq_domain
- *             core code.
- * @flags: host per irq_domain flags
- * @mapcount: The number of mapped interrupts
+ * @link:	Element in global irq_domain list.
+ * @name:	Name of interrupt domain
+ * @ops:	Pointer to irq_domain methods
+ * @host_data:	Private data pointer for use by owner.  Not touched by irq_domain
+ *		core code.
+ * @flags:	Per irq_domain flags
+ * @mapcount:	The number of mapped interrupts
  *
- * Optional elements
- * @fwnode: Pointer to firmware node associated with the irq_domain. Pretty easy
- *          to swap it for the of_node via the irq_domain_get_of_node accessor
- * @gc: Pointer to a list of generic chips. There is a helper function for
- *      setting up one or more generic chips for interrupt controllers
- *      drivers using the generic chip library which uses this pointer.
- * @dev: Pointer to a device that the domain represent, and that will be
- *       used for power management purposes.
- * @parent: Pointer to parent irq_domain to support hierarchy irq_domains
+ * Optional elements:
+ * @fwnode:	Pointer to firmware node associated with the irq_domain. Pretty easy
+ *		to swap it for the of_node via the irq_domain_get_of_node accessor
+ * @gc:		Pointer to a list of generic chips. There is a helper function for
+ *		setting up one or more generic chips for interrupt controllers
+ *		drivers using the generic chip library which uses this pointer.
+ * @dev:	Pointer to a device that can be utilized for power management
+ *		purposes related to the irq domain.
+ * @parent:	Pointer to parent irq_domain to support hierarchy irq_domains
  *
- * Revmap data, used internally by irq_domain
- * @revmap_size: Size of the linear map table @revmap[]
- * @revmap_tree: Radix map tree for hwirqs that don't fit in the linear map
- * @revmap_mutex: Lock for the revmap
- * @revmap: Linear table of irq_data pointers
+ * Revmap data, used internally by the irq domain code:
+ * @revmap_size:	Size of the linear map table @revmap[]
+ * @revmap_tree:	Radix map tree for hwirqs that don't fit in the linear map
+ * @revmap_mutex:	Lock for the revmap
+ * @revmap:		Linear table of irq_data pointers
  */
 struct irq_domain {
-	struct list_head link;
-	const char *name;
-	const struct irq_domain_ops *ops;
-	void *host_data;
-	unsigned int flags;
-	unsigned int mapcount;
+	struct list_head		link;
+	const char			*name;
+	const struct irq_domain_ops	*ops;
+	void				*host_data;
+	unsigned int			flags;
+	unsigned int			mapcount;
 
 	/* Optional data */
-	struct fwnode_handle *fwnode;
-	enum irq_domain_bus_token bus_token;
-	struct irq_domain_chip_generic *gc;
-	struct device *dev;
+	struct fwnode_handle		*fwnode;
+	enum irq_domain_bus_token	bus_token;
+	struct irq_domain_chip_generic	*gc;
+	struct device			*dev;
 #ifdef	CONFIG_IRQ_DOMAIN_HIERARCHY
-	struct irq_domain *parent;
+	struct irq_domain		*parent;
 #endif
 
 	/* reverse map data. The linear map gets appended to the irq_domain */
-	irq_hw_number_t hwirq_max;
-	unsigned int revmap_size;
-	struct radix_tree_root revmap_tree;
-	struct mutex revmap_mutex;
-	struct irq_data __rcu *revmap[];
+	irq_hw_number_t			hwirq_max;
+	unsigned int			revmap_size;
+	struct radix_tree_root		revmap_tree;
+	struct mutex			revmap_mutex;
+	struct irq_data __rcu		*revmap[];
 };
 
 /* Irq domain flags */

Re: [patch V2 02/21] genirq/irqdomain: Make struct irqdomain readable

[Marc Zyngier]

On Mon, 21 Nov 2022 14:36:20 +0000,
Thomas Gleixner <tglx@linutronix.de> wrote:
> 
> Tabular alignment of both kernel-doc and the actual struct declaration make
> visual parsing way more conveniant.
> 
> No functional change.
> 
> Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
> ---
> V2: Split out from the irqdomain::dev rename patch (Jason)
> ---
>  include/linux/irqdomain.h |   74 +++++++++++++++++++++++-----------------------
>  1 file changed, 37 insertions(+), 37 deletions(-)

Acked-by: Marc Zyngier <maz@kernel.org>

	M.

-- 
Without deviation from the norm, progress is not possible.

[patch V2 03/21] genirq/irqdomain: Rename irq_domain::dev to irq_domain::pm_dev

[Thomas Gleixner]

irq_domain::dev is a misnomer as it's usually the rule that a device
pointer points to something which is directly related to the instance.

irq_domain::dev can point to some other device for power management to
ensure that this underlying device is not powered down when an interrupt is
allocated.

The upcoming per device MSI domains really require a pointer to the device
which instantiated the irq domain and not to some random other device which
is required for power management down the chain.

Rename irq_domain::dev to irq_domain::pm_dev and fixup the few sites which
use that pointer.

Conversion was done with the help of coccinelle.

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Reviewed-by: Jason Gunthorpe <jgg@nvidia.com>
---
 drivers/irqchip/irq-gic.c |    4 ++--
 include/linux/irqdomain.h |    6 +++---
 kernel/irq/chip.c         |    8 ++++----
 3 files changed, 9 insertions(+), 9 deletions(-)

--- a/drivers/irqchip/irq-gic.c
+++ b/drivers/irqchip/irq-gic.c
@@ -401,8 +401,8 @@ static void gic_irq_print_chip(struct ir
 {
 	struct gic_chip_data *gic = irq_data_get_irq_chip_data(d);
 
-	if (gic->domain->dev)
-		seq_printf(p, gic->domain->dev->of_node->name);
+	if (gic->domain->pm_dev)
+		seq_printf(p, gic->domain->pm_dev->of_node->name);
 	else
 		seq_printf(p, "GIC-%d", (int)(gic - &gic_data[0]));
 }
--- a/include/linux/irqdomain.h
+++ b/include/linux/irqdomain.h
@@ -131,7 +131,7 @@ struct irq_domain_chip_generic;
  * @gc:		Pointer to a list of generic chips. There is a helper function for
  *		setting up one or more generic chips for interrupt controllers
  *		drivers using the generic chip library which uses this pointer.
- * @dev:	Pointer to a device that can be utilized for power management
+ * @pm_dev:	Pointer to a device that can be utilized for power management
  *		purposes related to the irq domain.
  * @parent:	Pointer to parent irq_domain to support hierarchy irq_domains
  *
@@ -153,7 +153,7 @@ struct irq_domain {
 	struct fwnode_handle		*fwnode;
 	enum irq_domain_bus_token	bus_token;
 	struct irq_domain_chip_generic	*gc;
-	struct device			*dev;
+	struct device			*pm_dev;
 #ifdef	CONFIG_IRQ_DOMAIN_HIERARCHY
 	struct irq_domain		*parent;
 #endif
@@ -206,7 +206,7 @@ static inline void irq_domain_set_pm_dev
 					    struct device *dev)
 {
 	if (d)
-		d->dev = dev;
+		d->pm_dev = dev;
 }
 
 #ifdef CONFIG_IRQ_DOMAIN
--- a/kernel/irq/chip.c
+++ b/kernel/irq/chip.c
@@ -1561,10 +1561,10 @@ int irq_chip_compose_msi_msg(struct irq_
 	return 0;
 }
 
-static struct device *irq_get_parent_device(struct irq_data *data)
+static struct device *irq_get_pm_device(struct irq_data *data)
 {
 	if (data->domain)
-		return data->domain->dev;
+		return data->domain->pm_dev;
 
 	return NULL;
 }
@@ -1578,7 +1578,7 @@ static struct device *irq_get_parent_dev
  */
 int irq_chip_pm_get(struct irq_data *data)
 {
-	struct device *dev = irq_get_parent_device(data);
+	struct device *dev = irq_get_pm_device(data);
 	int retval = 0;
 
 	if (IS_ENABLED(CONFIG_PM) && dev)
@@ -1597,7 +1597,7 @@ int irq_chip_pm_get(struct irq_data *dat
  */
 int irq_chip_pm_put(struct irq_data *data)
 {
-	struct device *dev = irq_get_parent_device(data);
+	struct device *dev = irq_get_pm_device(data);
 	int retval = 0;
 
 	if (IS_ENABLED(CONFIG_PM) && dev)

Re: [patch V2 03/21] genirq/irqdomain: Rename irq_domain::dev to irq_domain::pm_dev

[Marc Zyngier]

On Mon, 21 Nov 2022 14:36:22 +0000,
Thomas Gleixner <tglx@linutronix.de> wrote:
> 
> irq_domain::dev is a misnomer as it's usually the rule that a device
> pointer points to something which is directly related to the instance.
> 
> irq_domain::dev can point to some other device for power management to
> ensure that this underlying device is not powered down when an interrupt is
> allocated.
> 
> The upcoming per device MSI domains really require a pointer to the device
> which instantiated the irq domain and not to some random other device which
> is required for power management down the chain.
> 
> Rename irq_domain::dev to irq_domain::pm_dev and fixup the few sites which
> use that pointer.
> 
> Conversion was done with the help of coccinelle.
> 
> Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
> Reviewed-by: Jason Gunthorpe <jgg@nvidia.com>
> ---
>  drivers/irqchip/irq-gic.c |    4 ++--
>  include/linux/irqdomain.h |    6 +++---
>  kernel/irq/chip.c         |    8 ++++----
>  3 files changed, 9 insertions(+), 9 deletions(-)

Reviewed-by: Marc Zyngier <maz@kernel.org>

	M.

-- 
Without deviation from the norm, progress is not possible.

[patch V2 04/21] genirq/msi: Create msi_api.h

[Thomas Gleixner]

Create a API header for MSI specific functions which are relevant to device
drivers.

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Reviewed-by: Jason Gunthorpe <jgg@nvidia.com>
---
 include/linux/msi.h     |    6 ++++--
 include/linux/msi_api.h |   15 +++++++++++++++
 2 files changed, 19 insertions(+), 2 deletions(-)

--- a/include/linux/msi.h
+++ b/include/linux/msi.h
@@ -13,11 +13,14 @@
  *
  * Regular device drivers have no business with any of these functions and
  * especially storing MSI descriptor pointers in random code is considered
- * abuse. The only function which is relevant for drivers is msi_get_virq().
+ * abuse.
+ *
+ * Device driver relevant functions are available in <linux/msi_api.h>
  */
 
 #include <linux/irqdomain_defs.h>
 #include <linux/cpumask.h>
+#include <linux/msi_api.h>
 #include <linux/xarray.h>
 #include <linux/mutex.h>
 #include <linux/list.h>
@@ -188,7 +191,6 @@ struct msi_device_data {
 
 int msi_setup_device_data(struct device *dev);
 
-unsigned int msi_get_virq(struct device *dev, unsigned int index);
 void msi_lock_descs(struct device *dev);
 void msi_unlock_descs(struct device *dev);
 
--- /dev/null
+++ b/include/linux/msi_api.h
@@ -0,0 +1,15 @@
+/* SPDX-License-Identifier: GPL-2.0 */
+#ifndef LINUX_MSI_API_H
+#define LINUX_MSI_API_H
+
+/*
+ * APIs which are relevant for device driver code for allocating and
+ * freeing MSI interrupts and querying the associations between
+ * hardware/software MSI indices and the Linux interrupt number.
+ */
+
+struct device;
+
+unsigned int msi_get_virq(struct device *dev, unsigned int index);
+
+#endif

Re: [patch V2 04/21] genirq/msi: Create msi_api.h

[Marc Zyngier]

On Mon, 21 Nov 2022 14:36:23 +0000,
Thomas Gleixner <tglx@linutronix.de> wrote:
> 
> Create a API header for MSI specific functions which are relevant to device
> drivers.
> 
> Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
> Reviewed-by: Jason Gunthorpe <jgg@nvidia.com>
> ---
>  include/linux/msi.h     |    6 ++++--
>  include/linux/msi_api.h |   15 +++++++++++++++
>  2 files changed, 19 insertions(+), 2 deletions(-)

Acked-by: Marc Zyngier <maz@kernel.org>

	M.

-- 
Without deviation from the norm, progress is not possible.

[patch V2 05/21] genirq/irqdomain: Provide IRQ_DOMAIN_FLAG_MSI_PARENT

[Thomas Gleixner]

The new PCI/IMS (Interrupt Message Store) functionality is allowing
hardware vendors to provide implementation specific storage for the MSI
messages. This can be device memory and also host/guest memory, e.g. in
queue memory which is shared with the hardware.

This requires device specific MSI interrupt domains, which cannot be
achieved by expanding the existing PCI/MSI interrupt domain concept which is
a global interrupt domain shared by all PCI devices on a particular (IOMMU)
segment:

                                         |--- device 1
     [Vector]---[Remapping]---[PCI/MSI]--|...
                                         |--- device N

This works because the PCI/MSI[-X] space is uniform, but falls apart with
PCI/IMS which is implementation defined and must be available along with
PCI/MSI[-X] on the same device.

To support PCI/MSI[-X] plus PCI/IMS on the same device it is required to
rework the PCI/MSI interrupt domain hierarchy concept in the following way:

                              |--- [PCI/MSI] device 1
     [Vector]---[Remapping]---|...
                              |--- [PCI/MSI] device N

That allows in the next step to create multiple interrupt domains per device:


                              |--- [PCI/MSI] device 1
                              |--- [PCI/IMS] device 1
     [Vector]---[Remapping]---|...
                              |--- [PCI/MSI] device N
                              |--- [PCI/IMS] device N

So the domain which previously created the global PCI/MSI domain must now
act as parent domain for the per device domains.

The hierarchy depth is the same as before, but the PCI/MSI domains are then
device specific and not longer global.

Provide IRQ_DOMAIN_FLAG_MSI_PARENT, which allows to identify these parent
domains, along with helpers to query it.

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Reviewed-by: Jason Gunthorpe <jgg@nvidia.com>
---
 include/linux/irqdomain.h |   14 ++++++++++++++
 1 file changed, 14 insertions(+)
--- a/include/linux/irqdomain.h
+++ b/include/linux/irqdomain.h
@@ -189,6 +189,9 @@ enum {
 	/* Irq domain doesn't translate anything */
 	IRQ_DOMAIN_FLAG_NO_MAP		= (1 << 6),
 
+	/* Irq domain is a MSI parent domain */
+	IRQ_DOMAIN_FLAG_MSI_PARENT	= (1 << 8),
+
 	/*
 	 * Flags starting from IRQ_DOMAIN_FLAG_NONCORE are reserved
 	 * for implementation specific purposes and ignored by the
@@ -551,6 +554,11 @@ static inline bool irq_domain_is_msi_rem
 
 extern bool irq_domain_hierarchical_is_msi_remap(struct irq_domain *domain);
 
+static inline bool irq_domain_is_msi_parent(struct irq_domain *domain)
+{
+	return domain->flags & IRQ_DOMAIN_FLAG_MSI_PARENT;
+}
+
 #else	/* CONFIG_IRQ_DOMAIN_HIERARCHY */
 static inline int irq_domain_alloc_irqs(struct irq_domain *domain,
 			unsigned int nr_irqs, int node, void *arg)
@@ -596,6 +604,12 @@ irq_domain_hierarchical_is_msi_remap(str
 {
 	return false;
 }
+
+static inline bool irq_domain_is_msi_parent(struct irq_domain *domain)
+{
+	return false;
+}
+
 #endif	/* CONFIG_IRQ_DOMAIN_HIERARCHY */
 
 #else /* CONFIG_IRQ_DOMAIN */

[patch V2 06/21] genirq/irqdomain: Provide IRQ_DOMAIN_FLAG_MSI_DEVICE

[Thomas Gleixner]

Similar to marking parent MSI domains it's required to identify per device
domains. Add flag and helpers.

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
---
 include/linux/irqdomain.h |   13 +++++++++++++
 1 file changed, 13 insertions(+)

--- a/include/linux/irqdomain.h
+++ b/include/linux/irqdomain.h
@@ -192,6 +192,9 @@ enum {
 	/* Irq domain is a MSI parent domain */
 	IRQ_DOMAIN_FLAG_MSI_PARENT	= (1 << 8),
 
+	/* Irq domain is a MSI device domain */
+	IRQ_DOMAIN_FLAG_MSI_DEVICE	= (1 << 9),
+
 	/*
 	 * Flags starting from IRQ_DOMAIN_FLAG_NONCORE are reserved
 	 * for implementation specific purposes and ignored by the
@@ -559,6 +562,11 @@ static inline bool irq_domain_is_msi_par
 	return domain->flags & IRQ_DOMAIN_FLAG_MSI_PARENT;
 }
 
+static inline bool irq_domain_is_msi_device(struct irq_domain *domain)
+{
+	return domain->flags & IRQ_DOMAIN_FLAG_MSI_DEVICE;
+}
+
 #else	/* CONFIG_IRQ_DOMAIN_HIERARCHY */
 static inline int irq_domain_alloc_irqs(struct irq_domain *domain,
 			unsigned int nr_irqs, int node, void *arg)
@@ -609,6 +617,11 @@ static inline bool irq_domain_is_msi_par
 {
 	return false;
 }
+
+static inline bool irq_domain_is_msi_device(struct irq_domain *domain)
+{
+	return false;
+}
 
 #endif	/* CONFIG_IRQ_DOMAIN_HIERARCHY */
 

[patch V2 07/21] genirq/msi: Check for invalid MSI parent domain usage

[Thomas Gleixner]

In the upcoming per device MSI domain concept the MSI parent domains are
not allowed to be used as regular MSI domains where the MSI allocation/free
operations are applicable.

Add appropriate checks.

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
---
V2: Made the error return understandable. (Kevin)
---
 kernel/irq/msi.c |   17 ++++++++++++++---
 1 file changed, 14 insertions(+), 3 deletions(-)

--- a/kernel/irq/msi.c
+++ b/kernel/irq/msi.c
@@ -937,13 +937,21 @@ int msi_domain_alloc_irqs_descs_locked(s
 
 	lockdep_assert_held(&dev->msi.data->mutex);
 
+	if (WARN_ON_ONCE(irq_domain_is_msi_parent(domain))) {
+		ret = -EINVAL;
+		goto free;
+	}
+
+	/* Frees allocated descriptors in case of failure. */
 	ret = msi_domain_add_simple_msi_descs(info, dev, nvec);
 	if (ret)
-		return ret;
+		goto free;
 
 	ret = ops->domain_alloc_irqs(domain, dev, nvec);
-	if (ret)
-		msi_domain_free_irqs_descs_locked(domain, dev);
+	if (!ret)
+		return 0;
+free:
+	msi_domain_free_irqs_descs_locked(domain, dev);
 	return ret;
 }
 
@@ -1013,6 +1021,9 @@ void msi_domain_free_irqs_descs_locked(s
 
 	lockdep_assert_held(&dev->msi.data->mutex);
 
+	if (WARN_ON_ONCE(irq_domain_is_msi_parent(domain)))
+		return;
+
 	ops->domain_free_irqs(domain, dev);
 	if (ops->msi_post_free)
 		ops->msi_post_free(domain, dev);

[patch V2 08/21] genirq/msi: Add pointers for per device irq domains

[Thomas Gleixner]

With the upcoming per device MSI interrupt domain support it is necessary
to store the domain pointers per device.

Instead of delegating that storage to device drivers or subsystems create a
storage array in struct msi_device_data which will also take care of
tearing down the irq domains when msi_device_data is cleaned up via devres.

The interfaces into the MSI core will be changed from irqdomain pointer
based interfaces to domain id based interfaces to support multiple MSI
domains on a single device (e.g. PCI/MSI[-X] and PCI/IMS.

Once the per device domain support is complete the irq domain pointer in
struct device::msi.domain will not longer contain a pointer to the "global"
MSI domain. It will contain a pointer to the MSI parent domain instead.

It would be a horrible maze of conditionals to evaluate all over the place
which domain pointer should be used, i.e. the "global" one in
device::msi::domain or one from the internal pointer array.

To avoid this evaluate in msi_setup_device_data() whether the irq domain
which is associated to a device is a "global" or a parent MSI domain. If it
is global then copy the pointer into the first entry in the irqdomain
pointer array.

This allows to convert interfaces and implementation to domain ids while
keeping everything existing working.

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
---
 include/linux/msi.h     |    3 +++
 include/linux/msi_api.h |    8 ++++++++
 kernel/irq/msi.c        |   14 ++++++++++++++
 3 files changed, 25 insertions(+)

--- a/include/linux/msi.h
+++ b/include/linux/msi.h
@@ -77,6 +77,7 @@ struct msi_desc;
 struct pci_dev;
 struct platform_msi_priv_data;
 struct device_attribute;
+struct irq_domain;
 
 void __get_cached_msi_msg(struct msi_desc *entry, struct msi_msg *msg);
 #ifdef CONFIG_GENERIC_MSI_IRQ
@@ -180,6 +181,7 @@ enum msi_desc_filter {
  * @mutex:		Mutex protecting the MSI descriptor store
  * @__store:		Xarray for storing MSI descriptor pointers
  * @__iter_idx:		Index to search the next entry for iterators
+ * @__irqdomains:	Per device interrupt domains
  */
 struct msi_device_data {
 	unsigned long			properties;
@@ -187,6 +189,7 @@ struct msi_device_data {
 	struct mutex			mutex;
 	struct xarray			__store;
 	unsigned long			__iter_idx;
+	struct irq_domain		*__irqdomains[MSI_MAX_DEVICE_IRQDOMAINS];
 };
 
 int msi_setup_device_data(struct device *dev);
--- a/include/linux/msi_api.h
+++ b/include/linux/msi_api.h
@@ -10,6 +10,14 @@
 
 struct device;
 
+/*
+ * Per device interrupt domain related constants.
+ */
+enum msi_domain_ids {
+	MSI_DEFAULT_DOMAIN,
+	MSI_MAX_DEVICE_IRQDOMAINS,
+};
+
 unsigned int msi_get_virq(struct device *dev, unsigned int index);
 
 #endif
--- a/kernel/irq/msi.c
+++ b/kernel/irq/msi.c
@@ -21,6 +21,18 @@
 
 static inline int msi_sysfs_create_group(struct device *dev);
 
+static inline void msi_setup_default_irqdomain(struct device *dev, struct msi_device_data *md)
+{
+	if (!dev->msi.domain)
+		return;
+	/*
+	 * If @dev::msi::domain is a global MSI domain, copy the pointer
+	 * into the domain array to avoid conditionals all over the place.
+	 */
+	if (!irq_domain_is_msi_parent(dev->msi.domain))
+		md->__irqdomains[MSI_DEFAULT_DOMAIN] = dev->msi.domain;
+}
+
 /**
  * msi_alloc_desc - Allocate an initialized msi_desc
  * @dev:	Pointer to the device for which this is allocated
@@ -213,6 +225,8 @@ int msi_setup_device_data(struct device
 		return ret;
 	}
 
+	msi_setup_default_irqdomain(dev, md);
+
 	xa_init(&md->__store);
 	mutex_init(&md->mutex);
 	dev->msi.data = md;

Re: [patch V2 08/21] genirq/msi: Add pointers for per device irq domains

[Marc Zyngier]

On Mon, 21 Nov 2022 14:36:28 +0000,
Thomas Gleixner <tglx@linutronix.de> wrote:
> 
> With the upcoming per device MSI interrupt domain support it is necessary
> to store the domain pointers per device.
> 
> Instead of delegating that storage to device drivers or subsystems create a
> storage array in struct msi_device_data which will also take care of
> tearing down the irq domains when msi_device_data is cleaned up via devres.
> 
> The interfaces into the MSI core will be changed from irqdomain pointer
> based interfaces to domain id based interfaces to support multiple MSI
> domains on a single device (e.g. PCI/MSI[-X] and PCI/IMS.
> 
> Once the per device domain support is complete the irq domain pointer in
> struct device::msi.domain will not longer contain a pointer to the "global"
> MSI domain. It will contain a pointer to the MSI parent domain instead.
> 
> It would be a horrible maze of conditionals to evaluate all over the place
> which domain pointer should be used, i.e. the "global" one in
> device::msi::domain or one from the internal pointer array.
> 
> To avoid this evaluate in msi_setup_device_data() whether the irq domain
> which is associated to a device is a "global" or a parent MSI domain. If it
> is global then copy the pointer into the first entry in the irqdomain
> pointer array.
> 
> This allows to convert interfaces and implementation to domain ids while
> keeping everything existing working.
> 
> Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
> ---
>  include/linux/msi.h     |    3 +++
>  include/linux/msi_api.h |    8 ++++++++
>  kernel/irq/msi.c        |   14 ++++++++++++++
>  3 files changed, 25 insertions(+)
> 
> --- a/include/linux/msi.h
> +++ b/include/linux/msi.h
> @@ -77,6 +77,7 @@ struct msi_desc;
>  struct pci_dev;
>  struct platform_msi_priv_data;
>  struct device_attribute;
> +struct irq_domain;
>  
>  void __get_cached_msi_msg(struct msi_desc *entry, struct msi_msg *msg);
>  #ifdef CONFIG_GENERIC_MSI_IRQ
> @@ -180,6 +181,7 @@ enum msi_desc_filter {
>   * @mutex:		Mutex protecting the MSI descriptor store
>   * @__store:		Xarray for storing MSI descriptor pointers
>   * @__iter_idx:		Index to search the next entry for iterators
> + * @__irqdomains:	Per device interrupt domains
>   */
>  struct msi_device_data {
>  	unsigned long			properties;
> @@ -187,6 +189,7 @@ struct msi_device_data {
>  	struct mutex			mutex;
>  	struct xarray			__store;
>  	unsigned long			__iter_idx;
> +	struct irq_domain		*__irqdomains[MSI_MAX_DEVICE_IRQDOMAINS];
>  };
>  
>  int msi_setup_device_data(struct device *dev);
> --- a/include/linux/msi_api.h
> +++ b/include/linux/msi_api.h
> @@ -10,6 +10,14 @@
>  
>  struct device;
>  
> +/*
> + * Per device interrupt domain related constants.
> + */
> +enum msi_domain_ids {
> +	MSI_DEFAULT_DOMAIN,
> +	MSI_MAX_DEVICE_IRQDOMAINS,
> +};
> +
>  unsigned int msi_get_virq(struct device *dev, unsigned int index);
>  
>  #endif
> --- a/kernel/irq/msi.c
> +++ b/kernel/irq/msi.c
> @@ -21,6 +21,18 @@
>  
>  static inline int msi_sysfs_create_group(struct device *dev);
>  
> +static inline void msi_setup_default_irqdomain(struct device *dev, struct msi_device_data *md)

Do we really need this to be inline? I'm sure the compiler can figure
it out.

> +{
> +	if (!dev->msi.domain)
> +		return;
> +	/*
> +	 * If @dev::msi::domain is a global MSI domain, copy the pointer
> +	 * into the domain array to avoid conditionals all over the place.
> +	 */
> +	if (!irq_domain_is_msi_parent(dev->msi.domain))
> +		md->__irqdomains[MSI_DEFAULT_DOMAIN] = dev->msi.domain;
> +}
> +
>  /**
>   * msi_alloc_desc - Allocate an initialized msi_desc
>   * @dev:	Pointer to the device for which this is allocated
> @@ -213,6 +225,8 @@ int msi_setup_device_data(struct device
>  		return ret;
>  	}
>  
> +	msi_setup_default_irqdomain(dev, md);
> +

nit: if you move the setup below the msi.data assignment, you could
only pass dev as a parameter. Or pass both and move the assignment in
the function?

>  	xa_init(&md->__store);
>  	mutex_init(&md->mutex);
>  	dev->msi.data = md;
> 
> 

Irrespective of the above,

Reviewed-by: Marc Zyngier <maz@kernel.org>

	M.

-- 
Without deviation from the norm, progress is not possible.

Re: [patch V2 08/21] genirq/msi: Add pointers for per device irq domains

[Thomas Gleixner]

On Thu, Nov 24 2022 at 14:56, Marc Zyngier wrote:
> On Mon, 21 Nov 2022 14:36:28 +0000,
> Thomas Gleixner <tglx@linutronix.de> wrote:
>>  static inline int msi_sysfs_create_group(struct device *dev);
>>  
>> +static inline void msi_setup_default_irqdomain(struct device *dev, struct msi_device_data *md)
>
> Do we really need this to be inline? I'm sure the compiler can figure
> it out.

No. I'll fix that up.

>> +{
>> +	if (!dev->msi.domain)
>> +		return;
>> +	/*
>> +	 * If @dev::msi::domain is a global MSI domain, copy the pointer
>> +	 * into the domain array to avoid conditionals all over the place.
>> +	 */
>> +	if (!irq_domain_is_msi_parent(dev->msi.domain))
>> +		md->__irqdomains[MSI_DEFAULT_DOMAIN] = dev->msi.domain;
>> +}
>> +
>>  /**
>>   * msi_alloc_desc - Allocate an initialized msi_desc
>>   * @dev:	Pointer to the device for which this is allocated
>> @@ -213,6 +225,8 @@ int msi_setup_device_data(struct device
>>  		return ret;
>>  	}
>>  
>> +	msi_setup_default_irqdomain(dev, md);
>> +
>
> nit: if you move the setup below the msi.data assignment, you could
> only pass dev as a parameter. Or pass both and move the assignment in
> the function?

In which order we do that it all looks wrong :)

>>  	xa_init(&md->__store);
>>  	mutex_init(&md->mutex);
>>  	dev->msi.data = md;
>> 
>> 
>
> Irrespective of the above,
>
> Reviewed-by: Marc Zyngier <maz@kernel.org>
>
> 	M.

[patch V2 09/21] genirq/msi: Make MSI descriptor iterators device domain aware

[Thomas Gleixner]

To support multiple MSI interrupt domains per device it is necessary to
segment the xarray MSI descriptor storage. Each domain gets up to
MSI_MAX_INDEX entries.

Change the iterators so they operate with domain ids and take the domain
offsets into account.

The publicly available iterators which are mostly used in legacy
implementations and the PCI/MSI core default to MSI_DEFAULT_DOMAIN (0)
which is the id for the existing "global" domains.

No functional change.

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
---
V2: Fix the off by one so the index space is including MSI_MAX_INDEX (Kevin)
---
 include/linux/msi.h |   45 +++++++++++++++++++++++++++++++++++++++++----
 kernel/irq/msi.c    |   43 +++++++++++++++++++++++++++++++++++--------
 2 files changed, 76 insertions(+), 12 deletions(-)

--- a/include/linux/msi.h
+++ b/include/linux/msi.h
@@ -181,6 +181,7 @@ enum msi_desc_filter {
  * @mutex:		Mutex protecting the MSI descriptor store
  * @__store:		Xarray for storing MSI descriptor pointers
  * @__iter_idx:		Index to search the next entry for iterators
+ * @__iter_max:		Index to limit the search
  * @__irqdomains:	Per device interrupt domains
  */
 struct msi_device_data {
@@ -189,6 +190,7 @@ struct msi_device_data {
 	struct mutex			mutex;
 	struct xarray			__store;
 	unsigned long			__iter_idx;
+	unsigned long			__iter_max;
 	struct irq_domain		*__irqdomains[MSI_MAX_DEVICE_IRQDOMAINS];
 };
 
@@ -197,14 +199,34 @@ int msi_setup_device_data(struct device
 void msi_lock_descs(struct device *dev);
 void msi_unlock_descs(struct device *dev);
 
-struct msi_desc *msi_first_desc(struct device *dev, enum msi_desc_filter filter);
+struct msi_desc *msi_domain_first_desc(struct device *dev, unsigned int domid,
+				       enum msi_desc_filter filter);
+
+/**
+ * msi_first_desc - Get the first MSI descriptor of the default irqdomain
+ * @dev:	Device to operate on
+ * @filter:	Descriptor state filter
+ *
+ * Must be called with the MSI descriptor mutex held, i.e. msi_lock_descs()
+ * must be invoked before the call.
+ *
+ * Return: Pointer to the first MSI descriptor matching the search
+ *	   criteria, NULL if none found.
+ */
+static inline struct msi_desc *msi_first_desc(struct device *dev,
+					      enum msi_desc_filter filter)
+{
+	return msi_domain_first_desc(dev, MSI_DEFAULT_DOMAIN, filter);
+}
+
 struct msi_desc *msi_next_desc(struct device *dev, enum msi_desc_filter filter);
 
 /**
- * msi_for_each_desc - Iterate the MSI descriptors
+ * msi_domain_for_each_desc - Iterate the MSI descriptors in a specific domain
  *
  * @desc:	struct msi_desc pointer used as iterator
  * @dev:	struct device pointer - device to iterate
+ * @domid:	The id of the interrupt domain which should be walked.
  * @filter:	Filter for descriptor selection
  *
  * Notes:
@@ -212,10 +234,25 @@ struct msi_desc *msi_next_desc(struct de
  *    pair.
  *  - It is safe to remove a retrieved MSI descriptor in the loop.
  */
-#define msi_for_each_desc(desc, dev, filter)			\
-	for ((desc) = msi_first_desc((dev), (filter)); (desc);	\
+#define msi_domain_for_each_desc(desc, dev, domid, filter)			\
+	for ((desc) = msi_domain_first_desc((dev), (domid), (filter)); (desc);	\
 	     (desc) = msi_next_desc((dev), (filter)))
 
+/**
+ * msi_for_each_desc - Iterate the MSI descriptors in the default irqdomain
+ *
+ * @desc:	struct msi_desc pointer used as iterator
+ * @dev:	struct device pointer - device to iterate
+ * @filter:	Filter for descriptor selection
+ *
+ * Notes:
+ *  - The loop must be protected with a msi_lock_descs()/msi_unlock_descs()
+ *    pair.
+ *  - It is safe to remove a retrieved MSI descriptor in the loop.
+ */
+#define msi_for_each_desc(desc, dev, filter)					\
+	msi_domain_for_each_desc((desc), (dev), MSI_DEFAULT_DOMAIN, (filter))
+
 #define msi_desc_to_dev(desc)		((desc)->dev)
 
 #ifdef CONFIG_IRQ_MSI_IOMMU
--- a/kernel/irq/msi.c
+++ b/kernel/irq/msi.c
@@ -21,6 +21,10 @@
 
 static inline int msi_sysfs_create_group(struct device *dev);
 
+/* Invalid XA index which is outside of any searchable range */
+#define MSI_XA_MAX_INDEX	(ULONG_MAX - 1)
+#define MSI_XA_DOMAIN_SIZE	(MSI_MAX_INDEX + 1)
+
 static inline void msi_setup_default_irqdomain(struct device *dev, struct msi_device_data *md)
 {
 	if (!dev->msi.domain)
@@ -33,6 +37,20 @@ static inline void msi_setup_default_irq
 		md->__irqdomains[MSI_DEFAULT_DOMAIN] = dev->msi.domain;
 }
 
+static int msi_get_domain_base_index(struct device *dev, unsigned int domid)
+{
+	lockdep_assert_held(&dev->msi.data->mutex);
+
+	if (WARN_ON_ONCE(domid >= MSI_MAX_DEVICE_IRQDOMAINS))
+		return -ENODEV;
+
+	if (WARN_ON_ONCE(!dev->msi.data->__irqdomains[domid]))
+		return -ENODEV;
+
+	return domid * MSI_XA_DOMAIN_SIZE;
+}
+
+
 /**
  * msi_alloc_desc - Allocate an initialized msi_desc
  * @dev:	Pointer to the device for which this is allocated
@@ -229,6 +247,7 @@ int msi_setup_device_data(struct device
 
 	xa_init(&md->__store);
 	mutex_init(&md->mutex);
+	md->__iter_idx = MSI_XA_MAX_INDEX;
 	dev->msi.data = md;
 	devres_add(dev, md);
 	return 0;
@@ -251,7 +270,7 @@ EXPORT_SYMBOL_GPL(msi_lock_descs);
 void msi_unlock_descs(struct device *dev)
 {
 	/* Invalidate the index wich was cached by the iterator */
-	dev->msi.data->__iter_idx = MSI_MAX_INDEX;
+	dev->msi.data->__iter_idx = MSI_XA_MAX_INDEX;
 	mutex_unlock(&dev->msi.data->mutex);
 }
 EXPORT_SYMBOL_GPL(msi_unlock_descs);
@@ -260,17 +279,18 @@ static struct msi_desc *msi_find_desc(st
 {
 	struct msi_desc *desc;
 
-	xa_for_each_start(&md->__store, md->__iter_idx, desc, md->__iter_idx) {
+	xa_for_each_range(&md->__store, md->__iter_idx, desc, md->__iter_idx, md->__iter_max) {
 		if (msi_desc_match(desc, filter))
 			return desc;
 	}
-	md->__iter_idx = MSI_MAX_INDEX;
+	md->__iter_idx = MSI_XA_MAX_INDEX;
 	return NULL;
 }
 
 /**
- * msi_first_desc - Get the first MSI descriptor of a device
+ * msi_domain_first_desc - Get the first MSI descriptor of an irqdomain associated to a device
  * @dev:	Device to operate on
+ * @domid:	The id of the interrupt domain which should be walked.
  * @filter:	Descriptor state filter
  *
  * Must be called with the MSI descriptor mutex held, i.e. msi_lock_descs()
@@ -279,19 +299,26 @@ static struct msi_desc *msi_find_desc(st
  * Return: Pointer to the first MSI descriptor matching the search
  *	   criteria, NULL if none found.
  */
-struct msi_desc *msi_first_desc(struct device *dev, enum msi_desc_filter filter)
+struct msi_desc *msi_domain_first_desc(struct device *dev, unsigned int domid,
+				       enum msi_desc_filter filter)
 {
 	struct msi_device_data *md = dev->msi.data;
+	int baseidx;
 
 	if (WARN_ON_ONCE(!md))
 		return NULL;
 
 	lockdep_assert_held(&md->mutex);
 
-	md->__iter_idx = 0;
+	baseidx = msi_get_domain_base_index(dev, domid);
+	if (baseidx < 0)
+		return NULL;
+
+	md->__iter_idx = baseidx;
+	md->__iter_max = baseidx + MSI_MAX_INDEX;
 	return msi_find_desc(md, filter);
 }
-EXPORT_SYMBOL_GPL(msi_first_desc);
+EXPORT_SYMBOL_GPL(msi_domain_first_desc);
 
 /**
  * msi_next_desc - Get the next MSI descriptor of a device
@@ -315,7 +342,7 @@ struct msi_desc *msi_next_desc(struct de
 
 	lockdep_assert_held(&md->mutex);
 
-	if (md->__iter_idx >= (unsigned long)MSI_MAX_INDEX)
+	if (md->__iter_idx >= md->__iter_max)
 		return NULL;
 
 	md->__iter_idx++;

Re: [patch V2 09/21] genirq/msi: Make MSI descriptor iterators device domain aware

[Marc Zyngier]

On Mon, 21 Nov 2022 14:36:29 +0000,
Thomas Gleixner <tglx@linutronix.de> wrote:
> 
> To support multiple MSI interrupt domains per device it is necessary to
> segment the xarray MSI descriptor storage. Each domain gets up to
> MSI_MAX_INDEX entries.
> 
> Change the iterators so they operate with domain ids and take the domain
> offsets into account.
> 
> The publicly available iterators which are mostly used in legacy
> implementations and the PCI/MSI core default to MSI_DEFAULT_DOMAIN (0)
> which is the id for the existing "global" domains.
> 
> No functional change.
> 
> Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
> ---
> V2: Fix the off by one so the index space is including MSI_MAX_INDEX (Kevin)
> ---
>  include/linux/msi.h |   45 +++++++++++++++++++++++++++++++++++++++++----
>  kernel/irq/msi.c    |   43 +++++++++++++++++++++++++++++++++++--------
>  2 files changed, 76 insertions(+), 12 deletions(-)
> 
> --- a/include/linux/msi.h
> +++ b/include/linux/msi.h
> @@ -181,6 +181,7 @@ enum msi_desc_filter {
>   * @mutex:		Mutex protecting the MSI descriptor store
>   * @__store:		Xarray for storing MSI descriptor pointers
>   * @__iter_idx:		Index to search the next entry for iterators
> + * @__iter_max:		Index to limit the search
>   * @__irqdomains:	Per device interrupt domains
>   */
>  struct msi_device_data {
> @@ -189,6 +190,7 @@ struct msi_device_data {
>  	struct mutex			mutex;
>  	struct xarray			__store;
>  	unsigned long			__iter_idx;
> +	unsigned long			__iter_max;
>  	struct irq_domain		*__irqdomains[MSI_MAX_DEVICE_IRQDOMAINS];
>  };
>  
> @@ -197,14 +199,34 @@ int msi_setup_device_data(struct device
>  void msi_lock_descs(struct device *dev);
>  void msi_unlock_descs(struct device *dev);
>  
> -struct msi_desc *msi_first_desc(struct device *dev, enum msi_desc_filter filter);
> +struct msi_desc *msi_domain_first_desc(struct device *dev, unsigned int domid,
> +				       enum msi_desc_filter filter);
> +
> +/**
> + * msi_first_desc - Get the first MSI descriptor of the default irqdomain
> + * @dev:	Device to operate on
> + * @filter:	Descriptor state filter
> + *
> + * Must be called with the MSI descriptor mutex held, i.e. msi_lock_descs()
> + * must be invoked before the call.
> + *
> + * Return: Pointer to the first MSI descriptor matching the search
> + *	   criteria, NULL if none found.
> + */
> +static inline struct msi_desc *msi_first_desc(struct device *dev,
> +					      enum msi_desc_filter filter)
> +{
> +	return msi_domain_first_desc(dev, MSI_DEFAULT_DOMAIN, filter);
> +}
> +
>  struct msi_desc *msi_next_desc(struct device *dev, enum msi_desc_filter filter);
>  
>  /**
> - * msi_for_each_desc - Iterate the MSI descriptors
> + * msi_domain_for_each_desc - Iterate the MSI descriptors in a specific domain
>   *
>   * @desc:	struct msi_desc pointer used as iterator
>   * @dev:	struct device pointer - device to iterate
> + * @domid:	The id of the interrupt domain which should be walked.
>   * @filter:	Filter for descriptor selection
>   *
>   * Notes:
> @@ -212,10 +234,25 @@ struct msi_desc *msi_next_desc(struct de
>   *    pair.
>   *  - It is safe to remove a retrieved MSI descriptor in the loop.
>   */
> -#define msi_for_each_desc(desc, dev, filter)			\
> -	for ((desc) = msi_first_desc((dev), (filter)); (desc);	\
> +#define msi_domain_for_each_desc(desc, dev, domid, filter)			\
> +	for ((desc) = msi_domain_first_desc((dev), (domid), (filter)); (desc);	\
>  	     (desc) = msi_next_desc((dev), (filter)))
>  
> +/**
> + * msi_for_each_desc - Iterate the MSI descriptors in the default irqdomain
> + *
> + * @desc:	struct msi_desc pointer used as iterator
> + * @dev:	struct device pointer - device to iterate
> + * @filter:	Filter for descriptor selection
> + *
> + * Notes:
> + *  - The loop must be protected with a msi_lock_descs()/msi_unlock_descs()
> + *    pair.
> + *  - It is safe to remove a retrieved MSI descriptor in the loop.
> + */
> +#define msi_for_each_desc(desc, dev, filter)					\
> +	msi_domain_for_each_desc((desc), (dev), MSI_DEFAULT_DOMAIN, (filter))
> +
>  #define msi_desc_to_dev(desc)		((desc)->dev)
>  
>  #ifdef CONFIG_IRQ_MSI_IOMMU
> --- a/kernel/irq/msi.c
> +++ b/kernel/irq/msi.c
> @@ -21,6 +21,10 @@
>  
>  static inline int msi_sysfs_create_group(struct device *dev);
>  
> +/* Invalid XA index which is outside of any searchable range */
> +#define MSI_XA_MAX_INDEX	(ULONG_MAX - 1)
> +#define MSI_XA_DOMAIN_SIZE	(MSI_MAX_INDEX + 1)
> +
>  static inline void msi_setup_default_irqdomain(struct device *dev, struct msi_device_data *md)
>  {
>  	if (!dev->msi.domain)
> @@ -33,6 +37,20 @@ static inline void msi_setup_default_irq
>  		md->__irqdomains[MSI_DEFAULT_DOMAIN] = dev->msi.domain;
>  }
>  
> +static int msi_get_domain_base_index(struct device *dev, unsigned int domid)
> +{
> +	lockdep_assert_held(&dev->msi.data->mutex);
> +
> +	if (WARN_ON_ONCE(domid >= MSI_MAX_DEVICE_IRQDOMAINS))
> +		return -ENODEV;
> +
> +	if (WARN_ON_ONCE(!dev->msi.data->__irqdomains[domid]))
> +		return -ENODEV;
> +
> +	return domid * MSI_XA_DOMAIN_SIZE;
> +}

So what I understand of this is that we split the index space into
segments, one per msi_domain_ids, MSI_XA_DOMAIN_SIZE apart.

Why didn't you decide to go all the way and have one xarray per
irqdomain? It's not that big a structure, and it would make the whole
thing a bit more straightforward.

Or do you anticipate cases where you'd walk the __store xarray across
irqdomains?

Thanks,

	M.

-- 
Without deviation from the norm, progress is not possible.

Re: [patch V2 09/21] genirq/msi: Make MSI descriptor iterators device domain aware

[Thomas Gleixner]

On Thu, Nov 24 2022 at 15:46, Marc Zyngier wrote:
> On Mon, 21 Nov 2022 14:36:29 +0000,
> Thomas Gleixner <tglx@linutronix.de> wrote:
>> +static int msi_get_domain_base_index(struct device *dev, unsigned int domid)
>> +{
>> +	lockdep_assert_held(&dev->msi.data->mutex);
>> +
>> +	if (WARN_ON_ONCE(domid >= MSI_MAX_DEVICE_IRQDOMAINS))
>> +		return -ENODEV;
>> +
>> +	if (WARN_ON_ONCE(!dev->msi.data->__irqdomains[domid]))
>> +		return -ENODEV;
>> +
>> +	return domid * MSI_XA_DOMAIN_SIZE;
>> +}
>
> So what I understand of this is that we split the index space into
> segments, one per msi_domain_ids, MSI_XA_DOMAIN_SIZE apart.
>
> Why didn't you decide to go all the way and have one xarray per
> irqdomain? It's not that big a structure, and it would make the whole
> thing a bit more straightforward.
>
> Or do you anticipate cases where you'd walk the __store xarray across
> irqdomains?

Not really. I just found it conveniant to deal with one, but yes we
could do the same thing with two xarrays.

But at the very end it does not make a huge difference. Fine with me
either way.

Thanks,

        tglx

[patch V2 10/21] genirq/msi: Make msi_get_virq() device domain aware

[Thomas Gleixner]

From: Ahmed S. Darwish <darwi@linutronix.de>

In preparation of the upcoming per device multi MSI domain support, change
the interface to support lookups based on domain id and zero based index
within the domain.

Signed-off-by: Ahmed S. Darwish <darwi@linutronix.de>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
---
V2: Fix the locking leak and the operator precedence issue (kernel robot)
---
 include/linux/msi_api.h |   14 +++++++++++++-
 kernel/irq/msi.c        |   27 +++++++++++++++++++++------
 2 files changed, 34 insertions(+), 7 deletions(-)

--- a/include/linux/msi_api.h
+++ b/include/linux/msi_api.h
@@ -18,6 +18,18 @@ enum msi_domain_ids {
 	MSI_MAX_DEVICE_IRQDOMAINS,
 };
 
-unsigned int msi_get_virq(struct device *dev, unsigned int index);
+unsigned int msi_domain_get_virq(struct device *dev, unsigned int domid, unsigned int index);
+
+/**
+ * msi_get_virq - Lookup the Linux interrupt number for a MSI index on the default interrupt domain
+ * @dev:	Device for which the lookup happens
+ * @index:	The MSI index to lookup
+ *
+ * Return: The Linux interrupt number on success (> 0), 0 if not found
+ */
+static inline unsigned int msi_get_virq(struct device *dev, unsigned int index)
+{
+	return msi_domain_get_virq(dev, MSI_DEFAULT_DOMAIN, index);
+}
 
 #endif
--- a/kernel/irq/msi.c
+++ b/kernel/irq/msi.c
@@ -351,25 +351,38 @@ struct msi_desc *msi_next_desc(struct de
 EXPORT_SYMBOL_GPL(msi_next_desc);
 
 /**
- * msi_get_virq - Return Linux interrupt number of a MSI interrupt
+ * msi_domain_get_virq - Lookup the Linux interrupt number for a MSI index on a interrupt domain
  * @dev:	Device to operate on
+ * @domid:	Domain ID of the interrupt domain associated to the device
  * @index:	MSI interrupt index to look for (0-based)
  *
  * Return: The Linux interrupt number on success (> 0), 0 if not found
  */
-unsigned int msi_get_virq(struct device *dev, unsigned int index)
+unsigned int msi_domain_get_virq(struct device *dev, unsigned int domid, unsigned int index)
 {
 	struct msi_desc *desc;
 	unsigned int ret = 0;
-	bool pcimsi;
+	bool pcimsi = false;
+	int base;
 
 	if (!dev->msi.data)
 		return 0;
 
-	pcimsi = dev_is_pci(dev) ? to_pci_dev(dev)->msi_enabled : false;
+	if (WARN_ON_ONCE(index > MSI_MAX_INDEX))
+		return 0;
+
+	/* This check is only valid for the PCI default MSI domain */
+	if (dev_is_pci(dev) && domid == MSI_DEFAULT_DOMAIN)
+		pcimsi = to_pci_dev(dev)->msi_enabled;
 
 	msi_lock_descs(dev);
-	desc = xa_load(&dev->msi.data->__store, pcimsi ? 0 : index);
+
+	base = msi_get_domain_base_index(dev, domid);
+	if (base < 0)
+		goto unlock;
+
+	base += pcimsi ? 0 : index;
+	desc = xa_load(&dev->msi.data->__store, base);
 	if (desc && desc->irq) {
 		/*
 		 * PCI-MSI has only one descriptor for multiple interrupts.
@@ -383,10 +396,12 @@ unsigned int msi_get_virq(struct device
 			ret = desc->irq;
 		}
 	}
+
+unlock:
 	msi_unlock_descs(dev);
 	return ret;
 }
-EXPORT_SYMBOL_GPL(msi_get_virq);
+EXPORT_SYMBOL_GPL(msi_domain_get_virq);
 
 #ifdef CONFIG_SYSFS
 static struct attribute *msi_dev_attrs[] = {

[patch V2 11/21] genirq/msi: Rename msi_add_msi_desc() to msi_insert_msi_desc()

[Thomas Gleixner]

This reflects the functionality better. No functional change.

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Reviewed-by: Jason Gunthorpe <jgg@nvidia.com>
---
 drivers/pci/msi/msi.c            |    4 ++--
 drivers/soc/ti/ti_sci_inta_msi.c |    4 ++--
 include/linux/msi.h              |    2 +-
 kernel/irq/msi.c                 |    6 ++++--
 4 files changed, 9 insertions(+), 7 deletions(-)

--- a/drivers/pci/msi/msi.c
+++ b/drivers/pci/msi/msi.c
@@ -308,7 +308,7 @@ static int msi_setup_msi_desc(struct pci
 	if (desc.pci.msi_attrib.can_mask)
 		pci_read_config_dword(dev, desc.pci.mask_pos, &desc.pci.msi_mask);
 
-	return msi_add_msi_desc(&dev->dev, &desc);
+	return msi_insert_msi_desc(&dev->dev, &desc);
 }
 
 static int msi_verify_entries(struct pci_dev *dev)
@@ -591,7 +591,7 @@ static int msix_setup_msi_descs(struct p
 			desc.pci.msix_ctrl = readl(addr + PCI_MSIX_ENTRY_VECTOR_CTRL);
 		}
 
-		ret = msi_add_msi_desc(&dev->dev, &desc);
+		ret = msi_insert_msi_desc(&dev->dev, &desc);
 		if (ret)
 			break;
 	}
--- a/drivers/soc/ti/ti_sci_inta_msi.c
+++ b/drivers/soc/ti/ti_sci_inta_msi.c
@@ -73,13 +73,13 @@ static int ti_sci_inta_msi_alloc_descs(s
 	for (set = 0; set < res->sets; set++) {
 		for (i = 0; i < res->desc[set].num; i++, count++) {
 			msi_desc.msi_index = res->desc[set].start + i;
-			if (msi_add_msi_desc(dev, &msi_desc))
+			if (msi_insert_msi_desc(dev, &msi_desc))
 				goto fail;
 		}
 
 		for (i = 0; i < res->desc[set].num_sec; i++, count++) {
 			msi_desc.msi_index = res->desc[set].start_sec + i;
-			if (msi_add_msi_desc(dev, &msi_desc))
+			if (msi_insert_msi_desc(dev, &msi_desc))
 				goto fail;
 		}
 	}
--- a/include/linux/msi.h
+++ b/include/linux/msi.h
@@ -278,7 +278,7 @@ static inline void msi_desc_set_iommu_co
 }
 #endif
 
-int msi_add_msi_desc(struct device *dev, struct msi_desc *init_desc);
+int msi_insert_msi_desc(struct device *dev, struct msi_desc *init_desc);
 void msi_free_msi_descs_range(struct device *dev, unsigned int first_index, unsigned int last_index);
 
 /**
--- a/kernel/irq/msi.c
+++ b/kernel/irq/msi.c
@@ -100,13 +100,15 @@ static int msi_insert_desc(struct msi_de
 }
 
 /**
- * msi_add_msi_desc - Allocate and initialize a MSI descriptor
+ * msi_insert_msi_desc - Allocate and initialize a MSI descriptor and
+ *			 insert it at @init_desc->msi_index
+ *
  * @dev:	Pointer to the device for which the descriptor is allocated
  * @init_desc:	Pointer to an MSI descriptor to initialize the new descriptor
  *
  * Return: 0 on success or an appropriate failure code.
  */
-int msi_add_msi_desc(struct device *dev, struct msi_desc *init_desc)
+int msi_insert_msi_desc(struct device *dev, struct msi_desc *init_desc)
 {
 	struct msi_desc *desc;
 

[patch V2 12/21] genirq/msi: Make descriptor allocation device domain aware

[Thomas Gleixner]

Change the descriptor allocation and insertion functions to take a domain
id to prepare for the upcoming multi MSI domain per device support.

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
---
 include/linux/msi.h |   16 +++++++++++++++-
 kernel/irq/msi.c    |   34 +++++++++++++++++++++++++---------
 2 files changed, 40 insertions(+), 10 deletions(-)

--- a/include/linux/msi.h
+++ b/include/linux/msi.h
@@ -278,7 +278,21 @@ static inline void msi_desc_set_iommu_co
 }
 #endif
 
-int msi_insert_msi_desc(struct device *dev, struct msi_desc *init_desc);
+int msi_domain_insert_msi_desc(struct device *dev, unsigned int domid,
+			       struct msi_desc *init_desc);
+/**
+ * msi_insert_msi_desc - Allocate and initialize a MSI descriptor in the
+ *			 default irqdomain and insert it at @init_desc->msi_index
+ * @dev:	Pointer to the device for which the descriptor is allocated
+ * @init_desc:	Pointer to an MSI descriptor to initialize the new descriptor
+ *
+ * Return: 0 on success or an appropriate failure code.
+ */
+static inline int msi_insert_msi_desc(struct device *dev, struct msi_desc *init_desc)
+{
+	return msi_domain_insert_msi_desc(dev, MSI_DEFAULT_DOMAIN, init_desc);
+}
+
 void msi_free_msi_descs_range(struct device *dev, unsigned int first_index, unsigned int last_index);
 
 /**
--- a/kernel/irq/msi.c
+++ b/kernel/irq/msi.c
@@ -63,7 +63,7 @@ static int msi_get_domain_base_index(str
  * Return: pointer to allocated &msi_desc on success or %NULL on failure
  */
 static struct msi_desc *msi_alloc_desc(struct device *dev, int nvec,
-					const struct irq_affinity_desc *affinity)
+				       const struct irq_affinity_desc *affinity)
 {
 	struct msi_desc *desc = kzalloc(sizeof(*desc), GFP_KERNEL);
 
@@ -88,27 +88,42 @@ static void msi_free_desc(struct msi_des
 	kfree(desc);
 }
 
-static int msi_insert_desc(struct msi_device_data *md, struct msi_desc *desc, unsigned int index)
+static int msi_insert_desc(struct device *dev, struct msi_desc *desc,
+			   unsigned int domid, unsigned int index)
 {
-	int ret;
+	struct msi_device_data *md = dev->msi.data;
+	int baseidx, ret;
+
+	baseidx = msi_get_domain_base_index(dev, domid);
+	if (baseidx < 0) {
+		ret = baseidx;
+		goto fail;
+	}
 
 	desc->msi_index = index;
+	index += baseidx;
 	ret = xa_insert(&md->__store, index, desc, GFP_KERNEL);
 	if (ret)
-		msi_free_desc(desc);
+		goto fail;
+	return 0;
+
+fail:
+	msi_free_desc(desc);
 	return ret;
 }
 
 /**
- * msi_insert_msi_desc - Allocate and initialize a MSI descriptor and
- *			 insert it at @init_desc->msi_index
+ * msi_domain_insert_msi_desc - Allocate and initialize a MSI descriptor and
+ *				insert it at @init_desc->msi_index
  *
  * @dev:	Pointer to the device for which the descriptor is allocated
+ * @domid:	The id of the interrupt domain to which the desriptor is added
  * @init_desc:	Pointer to an MSI descriptor to initialize the new descriptor
  *
  * Return: 0 on success or an appropriate failure code.
  */
-int msi_insert_msi_desc(struct device *dev, struct msi_desc *init_desc)
+int msi_domain_insert_msi_desc(struct device *dev, unsigned int domid,
+			       struct msi_desc *init_desc)
 {
 	struct msi_desc *desc;
 
@@ -120,7 +135,8 @@ int msi_insert_msi_desc(struct device *d
 
 	/* Copy type specific data to the new descriptor. */
 	desc->pci = init_desc->pci;
-	return msi_insert_desc(dev->msi.data, desc, init_desc->msi_index);
+
+	return msi_insert_desc(dev, desc, domid, init_desc->msi_index);
 }
 
 /**
@@ -143,7 +159,7 @@ static int msi_add_simple_msi_descs(stru
 		desc = msi_alloc_desc(dev, 1, NULL);
 		if (!desc)
 			goto fail_mem;
-		ret = msi_insert_desc(dev->msi.data, desc, idx);
+		ret = msi_insert_desc(dev, desc, MSI_DEFAULT_DOMAIN, idx);
 		if (ret)
 			goto fail;
 	}

[patch V2 13/21] genirq/msi: Make descriptor freeing domain aware

[Thomas Gleixner]

Change the descriptor free functions to take a domain id to prepare for the
upcoming multi MSI domain per device support.

To avoid changing and extending the interfaces over and over use an core
internal control struct and hand the pointer through the various functions.

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
---
 include/linux/msi.h |   19 ++++++++++++++--
 kernel/irq/msi.c    |   59 ++++++++++++++++++++++++++++++++++++++++++++--------
 2 files changed, 67 insertions(+), 11 deletions(-)

--- a/include/linux/msi.h
+++ b/include/linux/msi.h
@@ -293,10 +293,25 @@ static inline int msi_insert_msi_desc(st
 	return msi_domain_insert_msi_desc(dev, MSI_DEFAULT_DOMAIN, init_desc);
 }
 
-void msi_free_msi_descs_range(struct device *dev, unsigned int first_index, unsigned int last_index);
+void msi_domain_free_msi_descs_range(struct device *dev, unsigned int domid,
+				     unsigned int first, unsigned int last);
 
 /**
- * msi_free_msi_descs - Free MSI descriptors of a device
+ * msi_free_msi_descs_range - Free a range of MSI descriptors of a device
+ *			      in the default irqdomain
+ *
+ * @dev:	Device for which to free the descriptors
+ * @first:	Index to start freeing from (inclusive)
+ * @last:	Last index to be freed (inclusive)
+ */
+static inline void msi_free_msi_descs_range(struct device *dev, unsigned int first,
+					    unsigned int last)
+{
+	msi_domain_free_msi_descs_range(dev, MSI_DEFAULT_DOMAIN, first, last);
+}
+
+/**
+ * msi_free_msi_descs - Free all MSI descriptors of a device in the default irqdomain
  * @dev:	Device to free the descriptors
  */
 static inline void msi_free_msi_descs(struct device *dev)
--- a/kernel/irq/msi.c
+++ b/kernel/irq/msi.c
@@ -19,6 +19,18 @@
 
 #include "internals.h"
 
+/**
+ * struct msi_ctrl - MSI internal management control structure
+ * @domid:	ID of the domain on which management operations should be done
+ * @first:	First (hardware) slot index to operate on
+ * @last:	Last (hardware) slot index to operate on
+ */
+struct msi_ctrl {
+	unsigned int			domid;
+	unsigned int			first;
+	unsigned int			last;
+};
+
 static inline int msi_sysfs_create_group(struct device *dev);
 
 /* Invalid XA index which is outside of any searchable range */
@@ -186,22 +198,32 @@ static bool msi_desc_match(struct msi_de
 	return false;
 }
 
-/**
- * msi_free_msi_descs_range - Free MSI descriptors of a device
- * @dev:		Device to free the descriptors
- * @first_index:	Index to start freeing from
- * @last_index:		Last index to be freed
- */
-void msi_free_msi_descs_range(struct device *dev, unsigned int first_index,
-			      unsigned int last_index)
+static bool msi_ctrl_range_valid(struct device *dev, struct msi_ctrl *ctrl)
+{
+	if (WARN_ON_ONCE(ctrl->first > ctrl->last ||
+			 ctrl->first > MSI_MAX_INDEX ||
+			 ctrl->last > MSI_MAX_INDEX))
+		return false;
+	return true;
+}
+
+static void msi_domain_free_descs(struct device *dev, struct msi_ctrl *ctrl)
 {
 	struct xarray *xa = &dev->msi.data->__store;
 	struct msi_desc *desc;
 	unsigned long idx;
+	int base;
 
 	lockdep_assert_held(&dev->msi.data->mutex);
 
-	xa_for_each_range(xa, idx, desc, first_index, last_index) {
+	if (!msi_ctrl_range_valid(dev, ctrl))
+		return;
+
+	base = msi_get_domain_base_index(dev, ctrl->domid);
+	if (base < 0)
+		return;
+
+	xa_for_each_range(xa, idx, desc, ctrl->first + base, ctrl->last + base) {
 		xa_erase(xa, idx);
 
 		/* Leak the descriptor when it is still referenced */
@@ -211,6 +233,25 @@ void msi_free_msi_descs_range(struct dev
 	}
 }
 
+/**
+ * msi_domain_free_msi_descs_range - Free a range of MSI descriptors of a device in an irqdomain
+ * @dev:	Device for which to free the descriptors
+ * @domid:	Id of the domain to operate on
+ * @first:	Index to start freeing from (inclusive)
+ * @last:	Last index to be freed (inclusive)
+ */
+void msi_domain_free_msi_descs_range(struct device *dev, unsigned int domid,
+				     unsigned int first, unsigned int last)
+{
+	struct msi_ctrl ctrl = {
+		.domid	= domid,
+		.first	= first,
+		.last	= last,
+	};
+
+	msi_domain_free_descs(dev, &ctrl);
+}
+
 void __get_cached_msi_msg(struct msi_desc *entry, struct msi_msg *msg)
 {
 	*msg = entry->msg;

[patch V2 14/21] genirq/msi: Make msi_add_simple_msi_descs() device domain aware

[Thomas Gleixner]

Allocating simple interrupt descriptors in the core code has to be multi
device irqdomain aware for the upcoming PCI/IMS support.

Change the interfaces to take a domain id into account. Use the internal
control struct for transport of arguments.

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
---
 kernel/irq/msi.c |   99 ++++++++++++++++++++++++++++++++-----------------------
 1 file changed, 59 insertions(+), 40 deletions(-)

--- a/kernel/irq/msi.c
+++ b/kernel/irq/msi.c
@@ -151,39 +151,6 @@ int msi_domain_insert_msi_desc(struct de
 	return msi_insert_desc(dev, desc, domid, init_desc->msi_index);
 }
 
-/**
- * msi_add_simple_msi_descs - Allocate and initialize MSI descriptors
- * @dev:	Pointer to the device for which the descriptors are allocated
- * @index:	Index for the first MSI descriptor
- * @ndesc:	Number of descriptors to allocate
- *
- * Return: 0 on success or an appropriate failure code.
- */
-static int msi_add_simple_msi_descs(struct device *dev, unsigned int index, unsigned int ndesc)
-{
-	unsigned int idx, last = index + ndesc - 1;
-	struct msi_desc *desc;
-	int ret;
-
-	lockdep_assert_held(&dev->msi.data->mutex);
-
-	for (idx = index; idx <= last; idx++) {
-		desc = msi_alloc_desc(dev, 1, NULL);
-		if (!desc)
-			goto fail_mem;
-		ret = msi_insert_desc(dev, desc, MSI_DEFAULT_DOMAIN, idx);
-		if (ret)
-			goto fail;
-	}
-	return 0;
-
-fail_mem:
-	ret = -ENOMEM;
-fail:
-	msi_free_msi_descs_range(dev, index, last);
-	return ret;
-}
-
 static bool msi_desc_match(struct msi_desc *desc, enum msi_desc_filter filter)
 {
 	switch (filter) {
@@ -252,6 +219,45 @@ void msi_domain_free_msi_descs_range(str
 	msi_domain_free_descs(dev, &ctrl);
 }
 
+/**
+ * msi_domain_add_simple_msi_descs - Allocate and initialize MSI descriptors
+ * @dev:	Pointer to the device for which the descriptors are allocated
+ * @ctrl:	Allocation control struct
+ *
+ * Return: 0 on success or an appropriate failure code.
+ */
+static int msi_domain_add_simple_msi_descs(struct device *dev, struct msi_ctrl *ctrl)
+{
+	struct msi_desc *desc;
+	unsigned int idx;
+	int ret, baseidx;
+
+	lockdep_assert_held(&dev->msi.data->mutex);
+
+	if (!msi_ctrl_range_valid(dev, ctrl))
+		return -EINVAL;
+
+	baseidx = msi_get_domain_base_index(dev, ctrl->domid);
+	if (baseidx < 0)
+		return baseidx;
+
+	for (idx = ctrl->first; idx <= ctrl->last; idx++) {
+		desc = msi_alloc_desc(dev, 1, NULL);
+		if (!desc)
+			goto fail_mem;
+		ret = msi_insert_desc(dev, desc, ctrl->domid, idx);
+		if (ret)
+			goto fail;
+	}
+	return 0;
+
+fail_mem:
+	ret = -ENOMEM;
+fail:
+	msi_domain_free_descs(dev, ctrl);
+	return ret;
+}
+
 void __get_cached_msi_msg(struct msi_desc *entry, struct msi_msg *msg)
 {
 	*msg = entry->msg;
@@ -821,11 +827,19 @@ int msi_domain_populate_irqs(struct irq_
 {
 	struct msi_domain_info *info = domain->host_data;
 	struct msi_domain_ops *ops = info->ops;
+	struct msi_ctrl ctrl = {
+		.domid	= MSI_DEFAULT_DOMAIN,
+		.first  = virq_base,
+		.last	= virq_base + nvec - 1,
+	};
 	struct msi_desc *desc;
 	int ret, virq;
 
+	if (!msi_ctrl_range_valid(dev, &ctrl))
+		return -EINVAL;
+
 	msi_lock_descs(dev);
-	ret = msi_add_simple_msi_descs(dev, virq_base, nvec);
+	ret = msi_domain_add_simple_msi_descs(dev, &ctrl);
 	if (ret)
 		goto unlock;
 
@@ -846,7 +860,7 @@ int msi_domain_populate_irqs(struct irq_
 fail:
 	for (--virq; virq >= virq_base; virq--)
 		irq_domain_free_irqs_common(domain, virq, 1);
-	msi_free_msi_descs_range(dev, virq_base, virq_base + nvec - 1);
+	msi_domain_free_descs(dev, &ctrl);
 unlock:
 	msi_unlock_descs(dev);
 	return ret;
@@ -1020,14 +1034,19 @@ static int __msi_domain_alloc_irqs(struc
 	return 0;
 }
 
-static int msi_domain_add_simple_msi_descs(struct msi_domain_info *info,
-					   struct device *dev,
-					   unsigned int num_descs)
+static int msi_domain_alloc_simple_msi_descs(struct device *dev,
+					     struct msi_domain_info *info,
+					     unsigned int num_descs)
 {
+	struct msi_ctrl ctrl = {
+		.domid	= MSI_DEFAULT_DOMAIN,
+		.last	= num_descs - 1,
+	};
+
 	if (!(info->flags & MSI_FLAG_ALLOC_SIMPLE_MSI_DESCS))
 		return 0;
 
-	return msi_add_simple_msi_descs(dev, 0, num_descs);
+	return msi_domain_add_simple_msi_descs(dev, &ctrl);
 }
 
 /**
@@ -1058,7 +1077,7 @@ int msi_domain_alloc_irqs_descs_locked(s
 	}
 
 	/* Frees allocated descriptors in case of failure. */
-	ret = msi_domain_add_simple_msi_descs(info, dev, nvec);
+	ret = msi_domain_alloc_simple_msi_descs(dev, info, nvec);
 	if (ret)
 		goto free;
 

[patch V2 15/21] genirq/msi: Provide new domain id based interfaces for freeing interrupts

[Thomas Gleixner]

Provide two sorts of interfaces to handle the different use cases:

  - msi_domain_free_irqs_range():

	Handles a caller defined precise range

  - msi_domain_free_irqs_all():

	Frees all interrupts associated to a domain

The latter is useful for device teardown and to handle the legacy MSI support
which does not have any range information available.

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
---
 include/linux/msi.h |    9 +++
 kernel/irq/msi.c    |  146 +++++++++++++++++++++++++++++++++++++++++++---------
 2 files changed, 132 insertions(+), 23 deletions(-)

--- a/include/linux/msi.h
+++ b/include/linux/msi.h
@@ -490,6 +490,15 @@ int msi_domain_alloc_irqs(struct irq_dom
 			  int nvec);
 void msi_domain_free_irqs_descs_locked(struct irq_domain *domain, struct device *dev);
 void msi_domain_free_irqs(struct irq_domain *domain, struct device *dev);
+
+void msi_domain_free_irqs_range_locked(struct device *dev, unsigned int domid,
+				       unsigned int first, unsigned int last);
+void msi_domain_free_irqs_range(struct device *dev, unsigned int domid,
+				unsigned int first, unsigned int last);
+
+void msi_domain_free_irqs_all_locked(struct device *dev, unsigned int domid);
+void msi_domain_free_irqs_all(struct device *dev, unsigned int domid);
+
 struct msi_domain_info *msi_get_domain_info(struct irq_domain *domain);
 
 struct irq_domain *platform_msi_create_irq_domain(struct fwnode_handle *fwnode,
--- a/kernel/irq/msi.c
+++ b/kernel/irq/msi.c
@@ -583,7 +583,25 @@ static inline void msi_sysfs_remove_desc
 #endif /* !CONFIG_SYSFS */
 
 static int __msi_domain_alloc_irqs(struct irq_domain *domain, struct device *dev, int nvec);
-static void __msi_domain_free_irqs(struct irq_domain *domain, struct device *dev);
+
+static struct irq_domain *msi_get_device_domain(struct device *dev, unsigned int domid)
+{
+	struct irq_domain *domain;
+
+	lockdep_assert_held(&dev->msi.data->mutex);
+
+	if (WARN_ON_ONCE(domid >= MSI_MAX_DEVICE_IRQDOMAINS))
+		return NULL;
+
+	domain = dev->msi.data->__irqdomains[domid];
+	if (!domain)
+		return NULL;
+
+	if (WARN_ON_ONCE(irq_domain_is_msi_parent(domain)))
+		return NULL;
+
+	return domain;
+}
 
 static inline void irq_chip_write_msi_msg(struct irq_data *data,
 					  struct msi_msg *msg)
@@ -744,7 +762,6 @@ static struct msi_domain_ops msi_domain_
 	.msi_prepare		= msi_domain_ops_prepare,
 	.set_desc		= msi_domain_ops_set_desc,
 	.domain_alloc_irqs	= __msi_domain_alloc_irqs,
-	.domain_free_irqs	= __msi_domain_free_irqs,
 };
 
 static void msi_domain_update_dom_ops(struct msi_domain_info *info)
@@ -758,8 +775,6 @@ static void msi_domain_update_dom_ops(st
 
 	if (ops->domain_alloc_irqs == NULL)
 		ops->domain_alloc_irqs = msi_domain_ops_default.domain_alloc_irqs;
-	if (ops->domain_free_irqs == NULL)
-		ops->domain_free_irqs = msi_domain_ops_default.domain_free_irqs;
 
 	if (!(info->flags & MSI_FLAG_USE_DEF_DOM_OPS))
 		return;
@@ -1108,15 +1123,23 @@ int msi_domain_alloc_irqs(struct irq_dom
 	return ret;
 }
 
-static void __msi_domain_free_irqs(struct irq_domain *domain, struct device *dev)
+static void __msi_domain_free_irqs(struct device *dev, struct irq_domain *domain,
+				   struct msi_ctrl *ctrl)
 {
 	struct msi_domain_info *info = domain->host_data;
+	struct xarray *xa = &dev->msi.data->__store;
 	struct irq_data *irqd;
 	struct msi_desc *desc;
-	int i;
+	unsigned long idx;
+	int i, base;
+
+	base = ctrl->domid * MSI_XA_DOMAIN_SIZE;
+
+	xa_for_each_range(xa, idx, desc, ctrl->first + base, ctrl->last + base) {
+		/* Only handle MSI entries which have an interrupt associated */
+		if (!msi_desc_match(desc, MSI_DESC_ASSOCIATED))
+			continue;
 
-	/* Only handle MSI entries which have an interrupt associated */
-	msi_for_each_desc(desc, dev, MSI_DESC_ASSOCIATED) {
 		/* Make sure all interrupts are deactivated */
 		for (i = 0; i < desc->nvec_used; i++) {
 			irqd = irq_domain_get_irq_data(domain, desc->irq + i);
@@ -1131,11 +1154,99 @@ static void __msi_domain_free_irqs(struc
 	}
 }
 
-static void msi_domain_free_msi_descs(struct msi_domain_info *info,
-				      struct device *dev)
+static void msi_domain_free_locked(struct device *dev, struct msi_ctrl *ctrl)
 {
+	struct msi_domain_info *info;
+	struct msi_domain_ops *ops;
+	struct irq_domain *domain;
+
+	if (!msi_ctrl_range_valid(dev, ctrl))
+		return;
+
+	domain = msi_get_device_domain(dev, ctrl->domid);
+	if (!domain)
+		return;
+
+	info = domain->host_data;
+	ops = info->ops;
+
+	if (ops->domain_free_irqs)
+		ops->domain_free_irqs(domain, dev);
+	else
+		__msi_domain_free_irqs(dev, domain, ctrl);
+
+	if (ops->msi_post_free)
+		ops->msi_post_free(domain, dev);
+
 	if (info->flags & MSI_FLAG_FREE_MSI_DESCS)
-		msi_free_msi_descs(dev);
+		msi_domain_free_descs(dev, ctrl);
+}
+
+/**
+ * msi_domain_free_irqs_range_locked - Free a range of interrupts from a MSI interrupt domain
+ *				       associated to @dev with msi_lock held
+ * @dev:	Pointer to device struct of the device for which the interrupts
+ *		are freed
+ * @domid:	Id of the interrupt domain to operate on
+ * @first:	First index to free (inclusive)
+ * @last:	Last index to free (inclusive)
+ */
+void msi_domain_free_irqs_range_locked(struct device *dev, unsigned int domid,
+				       unsigned int first, unsigned int last)
+{
+	struct msi_ctrl ctrl = {
+		.domid	= domid,
+		.first	= first,
+		.last	= last,
+	};
+	msi_domain_free_locked(dev, &ctrl);
+}
+
+/**
+ * msi_domain_free_irqs_range - Free a range of interrupts from a MSI interrupt domain
+ *				associated to @dev
+ * @dev:	Pointer to device struct of the device for which the interrupts
+ *		are freed
+ * @domid:	Id of the interrupt domain to operate on
+ * @first:	First index to free (inclusive)
+ * @last:	Last index to free (inclusive)
+ */
+void msi_domain_free_irqs_range(struct device *dev, unsigned int domid,
+				unsigned int first, unsigned int last)
+{
+	msi_lock_descs(dev);
+	msi_domain_free_irqs_range_locked(dev, domid, first, last);
+	msi_unlock_descs(dev);
+}
+
+/**
+ * msi_domain_free_irqs_all_locked - Free all interrupts from a MSI interrupt domain
+ *				     associated to a device
+ * @dev:	Pointer to device struct of the device for which the interrupts
+ *		are freed
+ * @domid:	The id of the domain to operate on
+ *
+ * Must be invoked from within a msi_lock_descs() / msi_unlock_descs()
+ * pair. Use this for MSI irqdomains which implement their own vector
+ * allocation.
+ */
+void msi_domain_free_irqs_all_locked(struct device *dev, unsigned int domid)
+{
+	msi_domain_free_irqs_range_locked(dev, domid, 0, MSI_MAX_INDEX);
+}
+
+/**
+ * msi_domain_free_irqs_all - Free all interrupts from a MSI interrupt domain
+ *			      associated to a device
+ * @dev:	Pointer to device struct of the device for which the interrupts
+ *		are freed
+ * @domid:	The id of the domain to operate on
+ */
+void msi_domain_free_irqs_all(struct device *dev, unsigned int domid)
+{
+	msi_lock_descs(dev);
+	msi_domain_free_irqs_all_locked(dev, domid);
+	msi_unlock_descs(dev);
 }
 
 /**
@@ -1150,18 +1261,7 @@ static void msi_domain_free_msi_descs(st
  */
 void msi_domain_free_irqs_descs_locked(struct irq_domain *domain, struct device *dev)
 {
-	struct msi_domain_info *info = domain->host_data;
-	struct msi_domain_ops *ops = info->ops;
-
-	lockdep_assert_held(&dev->msi.data->mutex);
-
-	if (WARN_ON_ONCE(irq_domain_is_msi_parent(domain)))
-		return;
-
-	ops->domain_free_irqs(domain, dev);
-	if (ops->msi_post_free)
-		ops->msi_post_free(domain, dev);
-	msi_domain_free_msi_descs(info, dev);
+	msi_domain_free_irqs_range_locked(dev, MSI_DEFAULT_DOMAIN, 0, MSI_MAX_INDEX);
 }
 
 /**

[patch V2 16/21] genirq/msi: Provide new domain id allocation functions

[Thomas Gleixner]

Provide two sorts of interfaces to handle the different use cases:

  - msi_domain_alloc_irqs_range():

	Handles a caller defined precise range

  - msi_domain_alloc_irqs_all():

	Allocates all interrupts associated to a domain by scanning the
    	allocated MSI descriptors

The latter is useful for the existing PCI/MSI support which does not have
range information available.

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
---
 include/linux/msi.h |   18 ++---
 kernel/irq/msi.c    |  179 ++++++++++++++++++++++++++++++++++++++++------------
 2 files changed, 149 insertions(+), 48 deletions(-)

--- a/include/linux/msi.h
+++ b/include/linux/msi.h
@@ -375,8 +375,8 @@ struct msi_domain_info;
  * @get_hwirq, @msi_init and @msi_free are callbacks used by the underlying
  * irqdomain.
  *
- * @msi_check, @msi_prepare and @set_desc are callbacks used by
- * msi_domain_alloc/free_irqs().
+ * @msi_check, @msi_prepare and @set_desc are callbacks used by the
+ * msi_domain_alloc/free_irqs*() variants.
  *
  * @domain_alloc_irqs, @domain_free_irqs can be used to override the
  * default allocation/free functions (__msi_domain_alloc/free_irqs). This
@@ -384,11 +384,6 @@ struct msi_domain_info;
  * be wrapped into the regular irq domains concepts by mere mortals.  This
  * allows to universally use msi_domain_alloc/free_irqs without having to
  * special case XEN all over the place.
- *
- * Contrary to other operations @domain_alloc_irqs and @domain_free_irqs
- * are set to the default implementation if NULL and even when
- * MSI_FLAG_USE_DEF_DOM_OPS is not set to avoid breaking existing users and
- * because these callbacks are obviously mandatory.
  */
 struct msi_domain_ops {
 	irq_hw_number_t	(*get_hwirq)(struct msi_domain_info *info,
@@ -488,14 +483,21 @@ int msi_domain_alloc_irqs_descs_locked(s
 				       int nvec);
 int msi_domain_alloc_irqs(struct irq_domain *domain, struct device *dev,
 			  int nvec);
+
 void msi_domain_free_irqs_descs_locked(struct irq_domain *domain, struct device *dev);
 void msi_domain_free_irqs(struct irq_domain *domain, struct device *dev);
 
+int msi_domain_alloc_irqs_range_locked(struct device *dev, unsigned int domid,
+				       unsigned int first, unsigned int last);
+int msi_domain_alloc_irqs_range(struct device *dev, unsigned int domid,
+				unsigned int first, unsigned int last);
+int msi_domain_alloc_irqs_all_locked(struct device *dev, unsigned int domid, int nirqs);
+
+
 void msi_domain_free_irqs_range_locked(struct device *dev, unsigned int domid,
 				       unsigned int first, unsigned int last);
 void msi_domain_free_irqs_range(struct device *dev, unsigned int domid,
 				unsigned int first, unsigned int last);
-
 void msi_domain_free_irqs_all_locked(struct device *dev, unsigned int domid);
 void msi_domain_free_irqs_all(struct device *dev, unsigned int domid);
 
--- a/kernel/irq/msi.c
+++ b/kernel/irq/msi.c
@@ -24,13 +24,17 @@
  * @domid:	ID of the domain on which management operations should be done
  * @first:	First (hardware) slot index to operate on
  * @last:	Last (hardware) slot index to operate on
+ * @nirqs:	The number of Linux interrupts to allocate. Can be larger
+ *		than the range due to PCI/multi-MSI.
  */
 struct msi_ctrl {
 	unsigned int			domid;
 	unsigned int			first;
 	unsigned int			last;
+	unsigned int			nirqs;
 };
 
+static void msi_domain_free_locked(struct device *dev, struct msi_ctrl *ctrl);
 static inline int msi_sysfs_create_group(struct device *dev);
 
 /* Invalid XA index which is outside of any searchable range */
@@ -582,8 +586,6 @@ static inline int msi_sysfs_populate_des
 static inline void msi_sysfs_remove_desc(struct device *dev, struct msi_desc *desc) { }
 #endif /* !CONFIG_SYSFS */
 
-static int __msi_domain_alloc_irqs(struct irq_domain *domain, struct device *dev, int nvec);
-
 static struct irq_domain *msi_get_device_domain(struct device *dev, unsigned int domid)
 {
 	struct irq_domain *domain;
@@ -761,7 +763,6 @@ static struct msi_domain_ops msi_domain_
 	.msi_init		= msi_domain_ops_init,
 	.msi_prepare		= msi_domain_ops_prepare,
 	.set_desc		= msi_domain_ops_set_desc,
-	.domain_alloc_irqs	= __msi_domain_alloc_irqs,
 };
 
 static void msi_domain_update_dom_ops(struct msi_domain_info *info)
@@ -773,9 +774,6 @@ static void msi_domain_update_dom_ops(st
 		return;
 	}
 
-	if (ops->domain_alloc_irqs == NULL)
-		ops->domain_alloc_irqs = msi_domain_ops_default.domain_alloc_irqs;
-
 	if (!(info->flags & MSI_FLAG_USE_DEF_DOM_OPS))
 		return;
 
@@ -986,18 +984,19 @@ static int msi_init_virq(struct irq_doma
 	return 0;
 }
 
-static int __msi_domain_alloc_irqs(struct irq_domain *domain, struct device *dev,
-				   int nvec)
+static int __msi_domain_alloc_irqs(struct device *dev, struct irq_domain *domain,
+				   struct msi_ctrl *ctrl)
 {
 	struct msi_domain_info *info = domain->host_data;
+	struct xarray *xa = &dev->msi.data->__store;
 	struct msi_domain_ops *ops = info->ops;
+	unsigned int vflags = 0, allocated = 0;
 	msi_alloc_info_t arg = { };
-	unsigned int vflags = 0;
+	int i, ret, virq, base;
 	struct msi_desc *desc;
-	int allocated = 0;
-	int i, ret, virq;
+	unsigned long idx;
 
-	ret = msi_domain_prepare_irqs(domain, dev, nvec, &arg);
+	ret = msi_domain_prepare_irqs(domain, dev, ctrl->nirqs, &arg);
 	if (ret)
 		return ret;
 
@@ -1023,7 +1022,16 @@ static int __msi_domain_alloc_irqs(struc
 			vflags |= VIRQ_NOMASK_QUIRK;
 	}
 
-	msi_for_each_desc(desc, dev, MSI_DESC_NOTASSOCIATED) {
+	base = ctrl->domid * MSI_XA_DOMAIN_SIZE;
+
+	xa_for_each_range(xa, idx, desc, ctrl->first + base, ctrl->last + base) {
+		if (!msi_desc_match(desc, MSI_DESC_NOTASSOCIATED))
+			continue;
+
+		/* This should return -ECONFUSED... */
+		if (WARN_ON_ONCE(allocated >= ctrl->nirqs))
+			return -EINVAL;
+
 		ops->set_desc(&arg, desc);
 
 		virq = __irq_domain_alloc_irqs(domain, -1, desc->nvec_used,
@@ -1051,17 +1059,122 @@ static int __msi_domain_alloc_irqs(struc
 
 static int msi_domain_alloc_simple_msi_descs(struct device *dev,
 					     struct msi_domain_info *info,
-					     unsigned int num_descs)
+					     struct msi_ctrl *ctrl)
+{
+	if (!(info->flags & MSI_FLAG_ALLOC_SIMPLE_MSI_DESCS))
+		return 0;
+
+	return msi_domain_add_simple_msi_descs(dev, ctrl);
+}
+
+static int __msi_domain_alloc_locked(struct device *dev, struct msi_ctrl *ctrl)
+{
+	struct msi_domain_info *info;
+	struct msi_domain_ops *ops;
+	struct irq_domain *domain;
+	int ret;
+
+	if (!msi_ctrl_range_valid(dev, ctrl))
+		return -EINVAL;
+
+	domain = msi_get_device_domain(dev, ctrl->domid);
+	if (!domain)
+		return -ENODEV;
+
+	info = domain->host_data;
+
+	ret = msi_domain_alloc_simple_msi_descs(dev, info, ctrl);
+	if (ret)
+		return ret;
+
+	ops = info->ops;
+	if (ops->domain_alloc_irqs)
+		return ops->domain_alloc_irqs(domain, dev, ctrl->nirqs);
+
+	return __msi_domain_alloc_irqs(dev, domain, ctrl);
+}
+
+static int msi_domain_alloc_locked(struct device *dev, struct msi_ctrl *ctrl)
+{
+	int ret = __msi_domain_alloc_locked(dev, ctrl);
+
+	if (ret)
+		msi_domain_free_locked(dev, ctrl);
+	return ret;
+}
+
+/**
+ * msi_domain_alloc_irqs_range_locked - Allocate interrupts from a MSI interrupt domain
+ * @dev:	Pointer to device struct of the device for which the interrupts
+ *		are allocated
+ * @domid:	Id of the interrupt domain to operate on
+ * @first:	First index to allocate (inclusive)
+ * @last:	Last index to allocate (inclusive)
+ *
+ * Must be invoked from within a msi_lock_descs() / msi_unlock_descs()
+ * pair. Use this for MSI irqdomains which implement their own descriptor
+ * allocation/free.
+ *
+ * Return: %0 on success or an error code.
+ */
+int msi_domain_alloc_irqs_range_locked(struct device *dev, unsigned int domid,
+				       unsigned int first, unsigned int last)
 {
 	struct msi_ctrl ctrl = {
-		.domid	= MSI_DEFAULT_DOMAIN,
-		.last	= num_descs - 1,
+		.domid	= domid,
+		.first	= first,
+		.last	= last,
+		.nirqs	= last + 1 - first,
 	};
 
-	if (!(info->flags & MSI_FLAG_ALLOC_SIMPLE_MSI_DESCS))
-		return 0;
+	return msi_domain_alloc_locked(dev, &ctrl);
+}
+
+/**
+ * msi_domain_alloc_irqs_range - Allocate interrupts from a MSI interrupt domain
+ * @dev:	Pointer to device struct of the device for which the interrupts
+ *		are allocated
+ * @domid:	Id of the interrupt domain to operate on
+ * @first:	First index to allocate (inclusive)
+ * @last:	Last index to allocate (inclusive)
+ *
+ * Return: %0 on success or an error code.
+ */
+int msi_domain_alloc_irqs_range(struct device *dev, unsigned int domid,
+				unsigned int first, unsigned int last)
+{
+	int ret;
 
-	return msi_domain_add_simple_msi_descs(dev, &ctrl);
+	msi_lock_descs(dev);
+	ret = msi_domain_alloc_irqs_range_locked(dev, domid, first, last);
+	msi_unlock_descs(dev);
+	return ret;
+}
+
+/**
+ * msi_domain_alloc_irqs_all_locked - Allocate all interrupts from a MSI interrupt domain
+ *
+ * @dev:	Pointer to device struct of the device for which the interrupts
+ *		are allocated
+ * @domid:	Id of the interrupt domain to operate on
+ * @nirqs:	The number of interrupts to allocate
+ *
+ * This function scans all MSI descriptors of the MSI domain and allocates interrupts
+ * for all unassigned ones. That function is to be used for MSI domain usage where
+ * the descriptor allocation is handled at the call site, e.g. PCI/MSI[X].
+ *
+ * Return: %0 on success or an error code.
+ */
+int msi_domain_alloc_irqs_all_locked(struct device *dev, unsigned int domid, int nirqs)
+{
+	struct msi_ctrl ctrl = {
+		.domid	= domid,
+		.first	= 0,
+		.last	= MSI_MAX_INDEX,
+		.nirqs	= nirqs,
+	};
+
+	return msi_domain_alloc_locked(dev, &ctrl);
 }
 
 /**
@@ -1080,28 +1193,14 @@ static int msi_domain_alloc_simple_msi_d
 int msi_domain_alloc_irqs_descs_locked(struct irq_domain *domain, struct device *dev,
 				       int nvec)
 {
-	struct msi_domain_info *info = domain->host_data;
-	struct msi_domain_ops *ops = info->ops;
-	int ret;
-
-	lockdep_assert_held(&dev->msi.data->mutex);
-
-	if (WARN_ON_ONCE(irq_domain_is_msi_parent(domain))) {
-		ret = -EINVAL;
-		goto free;
-	}
-
-	/* Frees allocated descriptors in case of failure. */
-	ret = msi_domain_alloc_simple_msi_descs(dev, info, nvec);
-	if (ret)
-		goto free;
+	struct msi_ctrl ctrl = {
+		.domid	= MSI_DEFAULT_DOMAIN,
+		.first	= 0,
+		.last	= MSI_MAX_INDEX,
+		.nirqs	= nvec,
+	};
 
-	ret = ops->domain_alloc_irqs(domain, dev, nvec);
-	if (!ret)
-		return 0;
-free:
-	msi_domain_free_irqs_descs_locked(domain, dev);
-	return ret;
+	return msi_domain_alloc_locked(dev, &ctrl);
 }
 
 /**

[patch V2 17/21] PCI/MSI: Use msi_domain_alloc/free_irqs_all_locked()

[Thomas Gleixner]

Switch to the new domain id aware interfaces to phase out the previous
ones.

This uses the _all() variants of the MSI allocation/free interfaces
because due to potentially sparse indices of the preallocated MSI
descriptors describing the range precisely is not possible.

No functional change.

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Acked-by: Bjorn Helgaas <bhelgaas@google.com>
---
 drivers/pci/msi/irqdomain.c |    4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

--- a/drivers/pci/msi/irqdomain.c
+++ b/drivers/pci/msi/irqdomain.c
@@ -14,7 +14,7 @@ int pci_msi_setup_msi_irqs(struct pci_de
 
 	domain = dev_get_msi_domain(&dev->dev);
 	if (domain && irq_domain_is_hierarchy(domain))
-		return msi_domain_alloc_irqs_descs_locked(domain, &dev->dev, nvec);
+		return msi_domain_alloc_irqs_all_locked(&dev->dev, MSI_DEFAULT_DOMAIN, nvec);
 
 	return pci_msi_legacy_setup_msi_irqs(dev, nvec, type);
 }
@@ -25,7 +25,7 @@ void pci_msi_teardown_msi_irqs(struct pc
 
 	domain = dev_get_msi_domain(&dev->dev);
 	if (domain && irq_domain_is_hierarchy(domain)) {
-		msi_domain_free_irqs_descs_locked(domain, &dev->dev);
+		msi_domain_free_irqs_all_locked(&dev->dev, MSI_DEFAULT_DOMAIN);
 	} else {
 		pci_msi_legacy_teardown_msi_irqs(dev);
 		msi_free_msi_descs(&dev->dev);

[patch V2 18/21] platform-msi: Switch to the domain id aware MSI interfaces

[Thomas Gleixner]

From: Ahmed S. Darwish <darwi@linutronix.de>

Switch to the new domain id aware interfaces to phase out the previous
ones. Use the range interface for allocation as the MSI descriptors
are allocate by the MSI core and therefore a precise range has to be
given. On free just tear down all MSI interrupts for simplicity.

No functional change.

Signed-off-by: Ahmed S. Darwish <darwi@linutronix.de>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
---
 drivers/base/platform-msi.c |    4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

--- a/drivers/base/platform-msi.c
+++ b/drivers/base/platform-msi.c
@@ -213,7 +213,7 @@ int platform_msi_domain_alloc_irqs(struc
 	if (err)
 		return err;
 
-	err = msi_domain_alloc_irqs(dev->msi.domain, dev, nvec);
+	err = msi_domain_alloc_irqs_range(dev, MSI_DEFAULT_DOMAIN, 0, nvec - 1);
 	if (err)
 		platform_msi_free_priv_data(dev);
 
@@ -227,7 +227,7 @@ EXPORT_SYMBOL_GPL(platform_msi_domain_al
  */
 void platform_msi_domain_free_irqs(struct device *dev)
 {
-	msi_domain_free_irqs(dev->msi.domain, dev);
+	msi_domain_free_irqs_all(dev, MSI_DEFAULT_DOMAIN);
 	platform_msi_free_priv_data(dev);
 }
 EXPORT_SYMBOL_GPL(platform_msi_domain_free_irqs);

[patch V2 19/21] bus: fsl-mc-msi: Switch to domain id aware interfaces

[Thomas Gleixner]

Switch to the new domain id aware interfaces to phase out the previous
ones.

Get rid of the MSI descriptor and domain checks as the core code detects
these issues anyway.

No functional change.

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
---
 drivers/bus/fsl-mc/fsl-mc-msi.c |   25 +++----------------------
 1 file changed, 3 insertions(+), 22 deletions(-)

--- a/drivers/bus/fsl-mc/fsl-mc-msi.c
+++ b/drivers/bus/fsl-mc/fsl-mc-msi.c
@@ -213,21 +213,8 @@ struct irq_domain *fsl_mc_find_msi_domai
 
 int fsl_mc_msi_domain_alloc_irqs(struct device *dev,  unsigned int irq_count)
 {
-	struct irq_domain *msi_domain;
-	int error;
+	int error = msi_setup_device_data(dev);
 
-	msi_domain = dev_get_msi_domain(dev);
-	if (!msi_domain)
-		return -EINVAL;
-
-	error = msi_setup_device_data(dev);
-	if (error)
-		return error;
-
-	msi_lock_descs(dev);
-	if (msi_first_desc(dev, MSI_DESC_ALL))
-		error = -EINVAL;
-	msi_unlock_descs(dev);
 	if (error)
 		return error;
 
@@ -235,7 +222,7 @@ int fsl_mc_msi_domain_alloc_irqs(struct
 	 * NOTE: Calling this function will trigger the invocation of the
 	 * its_fsl_mc_msi_prepare() callback
 	 */
-	error = msi_domain_alloc_irqs(msi_domain, dev, irq_count);
+	error = msi_domain_alloc_irqs_range(dev, MSI_DEFAULT_DOMAIN, 0, irq_count - 1);
 
 	if (error)
 		dev_err(dev, "Failed to allocate IRQs\n");
@@ -244,11 +231,5 @@ int fsl_mc_msi_domain_alloc_irqs(struct
 
 void fsl_mc_msi_domain_free_irqs(struct device *dev)
 {
-	struct irq_domain *msi_domain;
-
-	msi_domain = dev_get_msi_domain(dev);
-	if (!msi_domain)
-		return;
-
-	msi_domain_free_irqs(msi_domain, dev);
+	msi_domain_free_irqs_all(dev, MSI_DEFAULT_DOMAIN);
 }

[patch V2 20/21] oc: ti: ti_sci_inta_msi: Switch to domain id aware MSI functions

[Thomas Gleixner]

From: Ahmed S. Darwish <darwi@linutronix.de>

Switch to the new domain id aware interfaces to phase out the previous
ones. Remove the domain check as it happens in the core code now.

No functional change.

Signed-off-by: Ahmed S. Darwish <darwi@linutronix.de>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
---
 drivers/soc/ti/ti_sci_inta_msi.c |    8 ++------
 1 file changed, 2 insertions(+), 6 deletions(-)

--- a/drivers/soc/ti/ti_sci_inta_msi.c
+++ b/drivers/soc/ti/ti_sci_inta_msi.c
@@ -93,13 +93,8 @@ int ti_sci_inta_msi_domain_alloc_irqs(st
 				      struct ti_sci_resource *res)
 {
 	struct platform_device *pdev = to_platform_device(dev);
-	struct irq_domain *msi_domain;
 	int ret, nvec;
 
-	msi_domain = dev_get_msi_domain(dev);
-	if (!msi_domain)
-		return -EINVAL;
-
 	if (pdev->id < 0)
 		return -ENODEV;
 
@@ -114,7 +109,8 @@ int ti_sci_inta_msi_domain_alloc_irqs(st
 		goto unlock;
 	}
 
-	ret = msi_domain_alloc_irqs_descs_locked(msi_domain, dev, nvec);
+	/* Use alloc ALL as it's unclear whether there are gaps in the indices */
+	ret = msi_domain_alloc_irqs_all_locked(dev, MSI_DEFAULT_DOMAIN, nvec);
 	if (ret)
 		dev_err(dev, "Failed to allocate IRQs %d\n", ret);
 unlock:

[patch V2 21/21] genirq/msi: Remove unused alloc/free interfaces

[Thomas Gleixner]

Now that all users are converted remove the old interfaces.

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
---
 include/linux/msi.h |    7 ----
 kernel/irq/msi.c    |   73 ----------------------------------------------------
 2 files changed, 80 deletions(-)

--- a/include/linux/msi.h
+++ b/include/linux/msi.h
@@ -479,13 +479,6 @@ int msi_domain_set_affinity(struct irq_d
 struct irq_domain *msi_create_irq_domain(struct fwnode_handle *fwnode,
 					 struct msi_domain_info *info,
 					 struct irq_domain *parent);
-int msi_domain_alloc_irqs_descs_locked(struct irq_domain *domain, struct device *dev,
-				       int nvec);
-int msi_domain_alloc_irqs(struct irq_domain *domain, struct device *dev,
-			  int nvec);
-
-void msi_domain_free_irqs_descs_locked(struct irq_domain *domain, struct device *dev);
-void msi_domain_free_irqs(struct irq_domain *domain, struct device *dev);
 
 int msi_domain_alloc_irqs_range_locked(struct device *dev, unsigned int domid,
 				       unsigned int first, unsigned int last);
--- a/kernel/irq/msi.c
+++ b/kernel/irq/msi.c
@@ -1177,51 +1177,6 @@ int msi_domain_alloc_irqs_all_locked(str
 	return msi_domain_alloc_locked(dev, &ctrl);
 }
 
-/**
- * msi_domain_alloc_irqs_descs_locked - Allocate interrupts from a MSI interrupt domain
- * @domain:	The domain to allocate from
- * @dev:	Pointer to device struct of the device for which the interrupts
- *		are allocated
- * @nvec:	The number of interrupts to allocate
- *
- * Must be invoked from within a msi_lock_descs() / msi_unlock_descs()
- * pair. Use this for MSI irqdomains which implement their own vector
- * allocation/free.
- *
- * Return: %0 on success or an error code.
- */
-int msi_domain_alloc_irqs_descs_locked(struct irq_domain *domain, struct device *dev,
-				       int nvec)
-{
-	struct msi_ctrl ctrl = {
-		.domid	= MSI_DEFAULT_DOMAIN,
-		.first	= 0,
-		.last	= MSI_MAX_INDEX,
-		.nirqs	= nvec,
-	};
-
-	return msi_domain_alloc_locked(dev, &ctrl);
-}
-
-/**
- * msi_domain_alloc_irqs - Allocate interrupts from a MSI interrupt domain
- * @domain:	The domain to allocate from
- * @dev:	Pointer to device struct of the device for which the interrupts
- *		are allocated
- * @nvec:	The number of interrupts to allocate
- *
- * Return: %0 on success or an error code.
- */
-int msi_domain_alloc_irqs(struct irq_domain *domain, struct device *dev, int nvec)
-{
-	int ret;
-
-	msi_lock_descs(dev);
-	ret = msi_domain_alloc_irqs_descs_locked(domain, dev, nvec);
-	msi_unlock_descs(dev);
-	return ret;
-}
-
 static void __msi_domain_free_irqs(struct device *dev, struct irq_domain *domain,
 				   struct msi_ctrl *ctrl)
 {
@@ -1348,34 +1303,6 @@ void msi_domain_free_irqs_all(struct dev
 	msi_unlock_descs(dev);
 }
 
-/**
- * msi_domain_free_irqs_descs_locked - Free interrupts from a MSI interrupt @domain associated to @dev
- * @domain:	The domain to managing the interrupts
- * @dev:	Pointer to device struct of the device for which the interrupts
- *		are free
- *
- * Must be invoked from within a msi_lock_descs() / msi_unlock_descs()
- * pair. Use this for MSI irqdomains which implement their own vector
- * allocation.
- */
-void msi_domain_free_irqs_descs_locked(struct irq_domain *domain, struct device *dev)
-{
-	msi_domain_free_irqs_range_locked(dev, MSI_DEFAULT_DOMAIN, 0, MSI_MAX_INDEX);
-}
-
-/**
- * msi_domain_free_irqs - Free interrupts from a MSI interrupt @domain associated to @dev
- * @domain:	The domain to managing the interrupts
- * @dev:	Pointer to device struct of the device for which the interrupts
- *		are free
- */
-void msi_domain_free_irqs(struct irq_domain *domain, struct device *dev)
-{
-	msi_lock_descs(dev);
-	msi_domain_free_irqs_descs_locked(domain, dev);
-	msi_unlock_descs(dev);
-}
-
 /**
  * msi_get_domain_info - Get the MSI interrupt domain info for @domain
  * @domain:	The interrupt domain to retrieve data from

RE: [patch V2 00/21] genirq, PCI/MSI: Support for per device MSI and PCI/IMS - Part 2 API rework

[Tian, Kevin]

> From: Thomas Gleixner <tglx@linutronix.de>
> Sent: Monday, November 21, 2022 10:36 PM
> 
> This is V2 of the second part of effort to provide support for per device
> MSI interrupt domains.
> 

for the entire series:

Reviewed-by: Kevin Tian <kevin.tian@intel.com>