[PATCH v9 0/4] Loongarch irq-redirect support

[PATCH v9 0/4] Loongarch irq-redirect support

[Tianyang Zhang]

This series of patches introduces support for interrupt-redirect
controllers, and this hardware feature will be supported on 3C6000
for the first time

change log:
        v0->v1:
        1.Rename the model names in the document.
        2.Adjust the code format.
        3.Remove architecture - specific prefixes.
        4.Refactor the initialization logic, and IR driver no longer set 
	  AVEC_ENABLE.
        5.Enhance compatibility under certain configurations.

        v1->v2:
        1.Fixed an erroneous enabling issue.

        v2->v3
        1.Replace smp_call with address mapping to access registers
        2.Fix some code style issues

        v3->v4
        1.Provide reasonable comments on the modifications made to
	  IRQ_SET_MASK_OK_DONE
        2.Replace meaningless empty functions with parent_mask/unmask/ack
        3.Added and indeed released resources
        4.Added judgment for data structure initialization completion to 
          avoid duplicate creation during cpuhotplug
        5.Fixed the code style and some unnecessary troubles

        v4->v5
	1.when it is detected in avecintc_set_affinity that the current affinity
	remains valid, the return value is modified to IRQ_SET_MASK_OK_DONE.
	  After the introduction of redirect-domain, for each interrupt source, 
	avecintc-domain only provides the CPU/interrupt vector, while redirect-domain 
	provides other operations to synchronize interrupt affinity information 
	among multiple cores. 	  The original intention is to notify the cascaded
	redirect_set_affinity that multi-core synchronization is not required. 
	  However, this introduces some compatibility issues, such as the new return
	value causing msi_domain_set_affinity to no longer perform irq_chip_write_msi_msg.
	  1) When redirect exist in the system, the msi msg_address and msg_data no 
	longer changes after the allocation phase, so it does not actually require updating
	the MSI message info.
	  2) When only avecintc exists in the system, the irq_domain_activate_irq
	interface will be responsible for the initial configuration of the MSI message,
	which is unconditional. After that, if unnecessary, no modification to the MSI
	message is alse correctly.

	2.Restructured the macro definitions to make them appear more logical.

	3.Adjusted the layout of members struct redirect_queue\struct redirect_table and 
	struct redirect_item, making redirect_item the primary interface for accessing
	other members.

	4.The method of accessing registers has been standardized to MMIO.

	5.Initialize variables at declaration whenever possible.

	6.Replaced the the "struct page" in redirect_table and redirect_queue with "struct folio".

	7.Adjusted the initialization process so that all irq_desc configurations are completed
	during driver initialization, no longer relying on specific CPUs being online.

	8.Refactored portions of the code to make them more concise and logical.

	v5->v6
	Fix the warning messages reported by the test bot.

	v6->v7:
	1 Split patch:
 	 1) Docs/LoongArch: Add Advanced Extended-Redirect IRQ model description
	 2) LoongArch: Architectural preparation for Redirect irqchip
	 3) irqchip/irq-loongson.h:irq-loongson.h preparation for Redirect irqchip
	 4) irqchip/loongarch-avec.c:return IRQ_SET_MASK_OK_DONE when keep affinity
	 5) irqchip/irq-loongarch-ir:Add Redirect irqchip support

	2 Use sizeof() to replace fixed-size macro definitions.

	3 Unify the data types of the parameters for redirect_write/read_reg*.

	4 rename irde_invalid_entry_node to irde_invalid_entry and add comments 
	  explaining the 'raddr'.

	5 Fix the critical condition check bug in redirect_table_alloc.

	6 Use clear_bit to replace bitmap_release_region

	7 Delete some goto and handle the failure when it occurs.

	8 Removed the check for the `CONFIG_ACPI` macro, as CONFIG_ACPI 
          is selected by the arch/loongarch/Kconfig.

	9 Fixed the incorrect error flow in redirect_acpi_init.

	v7->v8:
	1 Apologies for the chaotic email delivery due to some network issues earlier.

	2 redirect_table_alloc now allocates nr_irqs consecutive redirect table entries to 
	  support multiple MSI devices.

	v8->v9:
	1 Rebased and reorganized the patches on the latest irq/core branch.

Tianyang Zhang (4):
  Docs/LoongArch: Add Advanced Extended-Redirect IRQ model description
  irqchip/irq-loongson.h:irq-loongson.h preparation for Redirect irqchip
  irqchip/loongarch-avec.c:return IRQ_SET_MASK_OK_DONE when keep
    affinity
  irqchip/irq-loongarch-ir:Add Redirect irqchip support

 .../arch/loongarch/irq-chip-model.rst         |  38 ++
 .../zh_CN/arch/loongarch/irq-chip-model.rst   |  37 ++
 drivers/irqchip/Makefile                      |   2 +-
 drivers/irqchip/irq-loongarch-avec.c          |  20 +-
 drivers/irqchip/irq-loongarch-ir.c            | 534 ++++++++++++++++++
 drivers/irqchip/irq-loongson.h                |  19 +
 6 files changed, 636 insertions(+), 14 deletions(-)
 create mode 100644 drivers/irqchip/irq-loongarch-ir.c

-- 
2.41.0

[PATCH v9 1/4] Docs/LoongArch: Add Advanced Extended-Redirect IRQ model description

[Tianyang Zhang]

Introduce the redirect interrupt controllers.When the redirect interrupt
controller is enabled, the routing target of MSI interrupts is no longer
a specific CPU and vector number, but a specific redirect entry. The
actual CPU and vector number used are described by the redirect entry.

Signed-off-by: Tianyang Zhang <zhangtianyang@loongson.cn>
---
 .../arch/loongarch/irq-chip-model.rst         | 38 +++++++++++++++++++
 .../zh_CN/arch/loongarch/irq-chip-model.rst   | 37 ++++++++++++++++++
 2 files changed, 75 insertions(+)

diff --git a/Documentation/arch/loongarch/irq-chip-model.rst b/Documentation/arch/loongarch/irq-chip-model.rst
index 8f5c3345109e..88474edcea48 100644
--- a/Documentation/arch/loongarch/irq-chip-model.rst
+++ b/Documentation/arch/loongarch/irq-chip-model.rst
@@ -181,6 +181,44 @@ go to PCH-PIC/PCH-LPC and gathered by EIOINTC, and then go to CPUINTC directly::
              | Devices |
              +---------+
 
+Advanced Extended IRQ model (with redirection)
+==============================================
+
+In this model, IPI (Inter-Processor Interrupt) and CPU Local Timer interrupt go
+to CPUINTC directly, CPU UARTS interrupts go to LIOINTC, PCH-MSI interrupts go
+to REDIRECT for remapping it to AVEC, and then go to CPUINTC directly, while all
+other devices interrupts go to PCH-PIC/PCH-LPC and gathered by EIOINTC, and then
+go to CPUINTC directly::
+
+ +-----+     +-----------------------+     +-------+
+ | IPI | --> |        CPUINTC        | <-- | Timer |
+ +-----+     +-----------------------+     +-------+
+              ^          ^          ^
+              |          |          |
+       +---------+ +----------+ +---------+     +-------+
+       | EIOINTC | | AVECINTC | | LIOINTC | <-- | UARTs |
+       +---------+ +----------+ +---------+     +-------+
+            ^            ^
+            |            |
+            |      +----------+
+            |      | REDIRECT |
+            |      +----------+
+            |            ^
+            |            |
+       +---------+  +---------+
+       | PCH-PIC |  | PCH-MSI |
+       +---------+  +---------+
+         ^     ^           ^
+         |     |           |
+ +---------+ +---------+ +---------+
+ | Devices | | PCH-LPC | | Devices |
+ +---------+ +---------+ +---------+
+                  ^
+                  |
+             +---------+
+             | Devices |
+             +---------+
+
 ACPI-related definitions
 ========================
 
diff --git a/Documentation/translations/zh_CN/arch/loongarch/irq-chip-model.rst b/Documentation/translations/zh_CN/arch/loongarch/irq-chip-model.rst
index d4ff80de47b6..7e4e3e55c7ad 100644
--- a/Documentation/translations/zh_CN/arch/loongarch/irq-chip-model.rst
+++ b/Documentation/translations/zh_CN/arch/loongarch/irq-chip-model.rst
@@ -174,6 +174,43 @@ CPU串口（UARTs）中断发送到LIOINTC，PCH-MSI中断发送到AVECINTC，
              | Devices |
              +---------+
 
+高级扩展IRQ模型 (带重定向)
+==========================
+
+在这种模型里面，IPI（Inter-Processor Interrupt）和CPU本地时钟中断直接发送到CPUINTC，
+CPU串口（UARTs）中断发送到LIOINTC，PCH-MSI中断首先发送到REDIRECT模块,完成重定向后发
+送到AVECINTC，而后通过AVECINTC直接送达CPUINTC，而其他所有设备的中断则分别发送到所连
+接的PCH-PIC/PCH-LPC，然后由EIOINTC统一收集，再直接到达CPUINTC::
+
+ +-----+     +-----------------------+     +-------+
+ | IPI | --> |        CPUINTC        | <-- | Timer |
+ +-----+     +-----------------------+     +-------+
+              ^          ^          ^
+              |          |          |
+       +---------+ +----------+ +---------+     +-------+
+       | EIOINTC | | AVECINTC | | LIOINTC | <-- | UARTs |
+       +---------+ +----------+ +---------+     +-------+
+            ^            ^
+            |            |
+            |      +----------+
+            |      | REDIRECT |
+            |      +----------+
+            |            ^
+            |            |
+       +---------+  +---------+
+       | PCH-PIC |  | PCH-MSI |
+       +---------+  +---------+
+         ^     ^           ^
+         |     |           |
+ +---------+ +---------+ +---------+
+ | Devices | | PCH-LPC | | Devices |
+ +---------+ +---------+ +---------+
+                  ^
+                  |
+             +---------+
+             | Devices |
+             +---------+
+
 ACPI相关的定义
 ==============
 
-- 
2.41.0

[PATCH v9 2/4] irqchip/irq-loongson.h:irq-loongson.h preparation for Redirect irqchip

[Tianyang Zhang]

Adjusted definitions and macro from loongson-avec.c to irq-loongson.h,
including:
1 marco AVEC_MSG_OFFSET
2 struct avecintc_data
3 Make avecintc_sync public

Signed-off-by: Tianyang Zhang <zhangtianyang@loongson.cn>
---
 drivers/irqchip/irq-loongarch-avec.c | 12 +-----------
 drivers/irqchip/irq-loongson.h       | 19 +++++++++++++++++++
 2 files changed, 20 insertions(+), 11 deletions(-)

diff --git a/drivers/irqchip/irq-loongarch-avec.c b/drivers/irqchip/irq-loongarch-avec.c
index ba556c008cf3..70aa34439370 100644
--- a/drivers/irqchip/irq-loongarch-avec.c
+++ b/drivers/irqchip/irq-loongarch-avec.c
@@ -24,7 +24,6 @@
 #define VECTORS_PER_REG		64
 #define IRR_VECTOR_MASK		0xffUL
 #define IRR_INVALID_MASK	0x80000000UL
-#define AVEC_MSG_OFFSET		0x100000
 
 #ifdef CONFIG_SMP
 struct pending_list {
@@ -47,15 +46,6 @@ struct avecintc_chip {
 
 static struct avecintc_chip loongarch_avec;
 
-struct avecintc_data {
-	struct list_head	entry;
-	unsigned int		cpu;
-	unsigned int		vec;
-	unsigned int		prev_cpu;
-	unsigned int		prev_vec;
-	unsigned int		moving;
-};
-
 static inline void avecintc_enable(void)
 {
 	u64 value;
@@ -85,7 +75,7 @@ static inline void pending_list_init(int cpu)
 	INIT_LIST_HEAD(&plist->head);
 }
 
-static void avecintc_sync(struct avecintc_data *adata)
+void avecintc_sync(struct avecintc_data *adata)
 {
 	struct pending_list *plist;
 
diff --git a/drivers/irqchip/irq-loongson.h b/drivers/irqchip/irq-loongson.h
index 11fa138d1f44..1f11d799bf0a 100644
--- a/drivers/irqchip/irq-loongson.h
+++ b/drivers/irqchip/irq-loongson.h
@@ -6,6 +6,17 @@
 #ifndef _DRIVERS_IRQCHIP_IRQ_LOONGSON_H
 #define _DRIVERS_IRQCHIP_IRQ_LOONGSON_H
 
+#define AVEC_MSG_OFFSET		0x100000
+
+struct avecintc_data {
+	struct list_head        entry;
+	unsigned int            cpu;
+	unsigned int            vec;
+	unsigned int            prev_cpu;
+	unsigned int            prev_vec;
+	unsigned int            moving;
+};
+
 int find_pch_pic(u32 gsi);
 
 int liointc_acpi_init(struct irq_domain *parent,
@@ -24,4 +35,12 @@ int pch_msi_acpi_init(struct irq_domain *parent,
 					struct acpi_madt_msi_pic *acpi_pchmsi);
 int pch_msi_acpi_init_avec(struct irq_domain *parent);
 
+int redirect_acpi_init(struct irq_domain *parent);
+
+#ifdef CONFIG_SMP
+void avecintc_sync(struct avecintc_data *adata);
+#else
+static inline void avecintc_sync(struct avecintc_data *adata) {}
+#endif
+
 #endif /* _DRIVERS_IRQCHIP_IRQ_LOONGSON_H */
-- 
2.41.0

Re: [PATCH v9 2/4] irqchip/irq-loongson.h:irq-loongson.h preparation for Redirect irqchip

[Thomas Gleixner]

On Fri, Jan 30 2026 at 10:59, Tianyang Zhang wrote:

The subject line is broken and incomprehensible. Something like:

    irqchip/irq-loonarch-avec: Prepare for interrupt redirection support

is concise and gives context.

> Adjusted definitions and macro from loongson-avec.c to irq-loongson.h,
> including:
> 1 marco AVEC_MSG_OFFSET
> 2 struct avecintc_data
> 3 Make avecintc_sync public

That's not a change log because it only tells _what_ the patch is doing,
but not _why_. Enumerating the details of what is moved is a pointless
exercise because that can bee seen from the diff itself.

So you want something like:

  Interrupt redirection support requires a new interrupt chip, which
  needs to share data structures, constants and functions with the AVEC
  code.

  Move them to the header file and make the required functions public.

See?

[PATCH v9 3/4] irqchip/loongarch-avec.c:return IRQ_SET_MASK_OK_DONE when keep affinity

[Tianyang Zhang]

When it is detected in avecintc_set_affinity that the current affinity
remains valid, the return value is modified to IRQ_SET_MASK_OK_DONE.

After the introduction of redirect-domain, for each interrupt source,
avecintc-domain only provides the CPU/interrupt vector, while redirect-domain
provides other operations to synchronize interrupt affinity information
among multiple cores. The original intention is to notify the cascaded
redirect_set_affinity that multi-core synchronization is not required.

However, this introduces some compatibility issues, such as the new return
value causing msi_domain_set_affinity to no longer perform irq_chip_write_msi_msg.
  1) When redirect exist in the system, the msi msg_address and msg_data no
longer changes after the allocation phase, so it does not actually require updating
the MSI message info.
  2) When only avecintc exists in the system, the irq_domain_activate_irq
interface will be responsible for the initial configuration of the MSI message,
which is unconditional. After that, if unnecessary, no modification to the MSI
message is alse correctly.

Signed-off-by: Tianyang Zhang <zhangtianyang@loongson.cn>
---
 drivers/irqchip/irq-loongarch-avec.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/drivers/irqchip/irq-loongarch-avec.c b/drivers/irqchip/irq-loongarch-avec.c
index 70aa34439370..7bdf3f678db4 100644
--- a/drivers/irqchip/irq-loongarch-avec.c
+++ b/drivers/irqchip/irq-loongarch-avec.c
@@ -99,7 +99,7 @@ static int avecintc_set_affinity(struct irq_data *data, const struct cpumask *de
 			return -EBUSY;
 
 		if (cpu_online(adata->cpu) && cpumask_test_cpu(adata->cpu, dest))
-			return 0;
+			return IRQ_SET_MASK_OK_DONE;
 
 		cpumask_and(&intersect_mask, dest, cpu_online_mask);
 
-- 
2.41.0

Re: [PATCH v9 3/4] irqchip/loongarch-avec.c:return IRQ_SET_MASK_OK_DONE when keep affinity

[Thomas Gleixner]

On Fri, Jan 30 2026 at 10:59, Tianyang Zhang wrote:

  irqchip/irq-loonarch-avec: Return IRQ_SET_MASK_OK_DONE when affinity is unchanged

Note the space between 'prefix: ' and the shortlog sentence

> When it is detected in avecintc_set_affinity that the current affinity

Functions are denoted with fname() to make it clear. See

https://www.kernel.org/doc/html/latest/process/maintainer-tip.html#patch-submission-notes

> remains valid, the return value is modified to IRQ_SET_MASK_OK_DONE.

That's the very wrong order of explaining the change. You again explain
the _what_ first without giving context and reason.

See the above link. It provides guidance about structuring a change log.

> After the introduction of redirect-domain, for each interrupt source,
> avecintc-domain only provides the CPU/interrupt vector, while redirect-domain
> provides other operations to synchronize interrupt affinity information
> among multiple cores. The original intention is to notify the cascaded
> redirect_set_affinity that multi-core synchronization is not required.

"After the introduction.." is confusing at best.

> However, this introduces some compatibility issues, such as the new return
> value causing msi_domain_set_affinity to no longer perform irq_chip_write_msi_msg.
>   1) When redirect exist in the system, the msi msg_address and msg_data no
> longer changes after the allocation phase, so it does not actually require updating
> the MSI message info.
>   2) When only avecintc exists in the system, the irq_domain_activate_irq
> interface will be responsible for the initial configuration of the MSI message,
> which is unconditional. After that, if unnecessary, no modification to the MSI
> message is alse correctly.

Spell checkers exist for a reason.

Thanks

        tglx

Re: [PATCH v9 3/4] irqchip/loongarch-avec.c:return IRQ_SET_MASK_OK_DONE when keep affinity

[Tianyang Zhang]

Hi, Thomas

在 2026/1/30 下午4:20, Thomas Gleixner 写道:
> On Fri, Jan 30 2026 at 10:59, Tianyang Zhang wrote:
>
>    irqchip/irq-loonarch-avec: Return IRQ_SET_MASK_OK_DONE when affinity is unchanged
>
> Note the space between 'prefix: ' and the shortlog sentence
Ok, I got it
>
>> When it is detected in avecintc_set_affinity that the current affinity
> Functions are denoted with fname() to make it clear. See
>
> https://www.kernel.org/doc/html/latest/process/maintainer-tip.html#patch-submission-notes
Ok, I got it
>
>> remains valid, the return value is modified to IRQ_SET_MASK_OK_DONE.
> That's the very wrong order of explaining the change. You again explain
> the _what_ first without giving context and reason.
>
> See the above link. It provides guidance about structuring a change log.
Ok, I will rewrite this changelog
>
>> After the introduction of redirect-domain, for each interrupt source,
>> avecintc-domain only provides the CPU/interrupt vector, while redirect-domain
>> provides other operations to synchronize interrupt affinity information
>> among multiple cores. The original intention is to notify the cascaded
>> redirect_set_affinity that multi-core synchronization is not required.
> "After the introduction.." is confusing at best.
Ok, I got it
>
>> However, this introduces some compatibility issues, such as the new return
>> value causing msi_domain_set_affinity to no longer perform irq_chip_write_msi_msg.
>>    1) When redirect exist in the system, the msi msg_address and msg_data no
>> longer changes after the allocation phase, so it does not actually require updating
>> the MSI message info.
>>    2) When only avecintc exists in the system, the irq_domain_activate_irq
>> interface will be responsible for the initial configuration of the MSI message,
>> which is unconditional. After that, if unnecessary, no modification to the MSI
>> message is alse correctly.
> Spell checkers exist for a reason.

Ok, I got it

Thanks

>
> Thanks
>
>          tglx

[PATCH v9 4/4] irqchip/irq-loongarch-ir:Add Redirect irqchip support

[Tianyang Zhang]

The main function of the Redirected interrupt controller is to manage
the redirected-interrupt table, which consists of many redirected entries.
When MSI interrupts are requested, the driver creates a corresponding
redirected entry that describes the target CPU/vector number and the
operating mode of the interrupt. The redirected interrupt module has an
independent cache, and during the interrupt routing process, it will
prioritize the redirected entries that hit the cache. The driver
invalidates certain entry caches via a command queue.

Co-developed-by: Liupu Wang <wangliupu@loongson.cn>
Signed-off-by: Liupu Wang <wangliupu@loongson.cn>
Signed-off-by: Tianyang Zhang <zhangtianyang@loongson.cn>
---
 drivers/irqchip/Makefile             |   2 +-
 drivers/irqchip/irq-loongarch-avec.c |   6 +-
 drivers/irqchip/irq-loongarch-ir.c   | 534 +++++++++++++++++++++++++++
 3 files changed, 540 insertions(+), 2 deletions(-)
 create mode 100644 drivers/irqchip/irq-loongarch-ir.c

diff --git a/drivers/irqchip/Makefile b/drivers/irqchip/Makefile
index 6a229443efe0..2fc15a8e6ada 100644
--- a/drivers/irqchip/Makefile
+++ b/drivers/irqchip/Makefile
@@ -118,7 +118,7 @@ obj-$(CONFIG_LS1X_IRQ)			+= irq-ls1x.o
 obj-$(CONFIG_TI_SCI_INTR_IRQCHIP)	+= irq-ti-sci-intr.o
 obj-$(CONFIG_TI_SCI_INTA_IRQCHIP)	+= irq-ti-sci-inta.o
 obj-$(CONFIG_TI_PRUSS_INTC)		+= irq-pruss-intc.o
-obj-$(CONFIG_IRQ_LOONGARCH_CPU)		+= irq-loongarch-cpu.o irq-loongarch-avec.o
+obj-$(CONFIG_IRQ_LOONGARCH_CPU)		+= irq-loongarch-cpu.o irq-loongarch-avec.o irq-loongarch-ir.o
 obj-$(CONFIG_LOONGSON_LIOINTC)		+= irq-loongson-liointc.o
 obj-$(CONFIG_LOONGSON_EIOINTC)		+= irq-loongson-eiointc.o
 obj-$(CONFIG_LOONGSON_HTPIC)		+= irq-loongson-htpic.o
diff --git a/drivers/irqchip/irq-loongarch-avec.c b/drivers/irqchip/irq-loongarch-avec.c
index 7bdf3f678db4..972203182079 100644
--- a/drivers/irqchip/irq-loongarch-avec.c
+++ b/drivers/irqchip/irq-loongarch-avec.c
@@ -111,7 +111,8 @@ static int avecintc_set_affinity(struct irq_data *data, const struct cpumask *de
 		adata->cpu = cpu;
 		adata->vec = vector;
 		per_cpu_ptr(irq_map, adata->cpu)[adata->vec] = irq_data_to_desc(data);
-		avecintc_sync(adata);
+		if (!cpu_has_redirectint)
+			avecintc_sync(adata);
 	}
 
 	irq_data_update_effective_affinity(data, cpumask_of(cpu));
@@ -403,6 +404,9 @@ static int __init pch_msi_parse_madt(union acpi_subtable_headers *header,
 
 static inline int __init acpi_cascade_irqdomain_init(void)
 {
+	if (cpu_has_redirectint)
+		return redirect_acpi_init(loongarch_avec.domain);
+
 	return acpi_table_parse_madt(ACPI_MADT_TYPE_MSI_PIC, pch_msi_parse_madt, 1);
 }
 
diff --git a/drivers/irqchip/irq-loongarch-ir.c b/drivers/irqchip/irq-loongarch-ir.c
new file mode 100644
index 000000000000..0ee7c37487f7
--- /dev/null
+++ b/drivers/irqchip/irq-loongarch-ir.c
@@ -0,0 +1,534 @@
+// SPDX-License-Identifier: GPL-2.0
+/*
+ * Copyright (C) 2020 Loongson Technologies, Inc.
+ */
+
+#include <linux/cpuhotplug.h>
+#include <linux/init.h>
+#include <linux/interrupt.h>
+#include <linux/kernel.h>
+#include <linux/irq.h>
+#include <linux/irqchip.h>
+#include <linux/irqchip/irq-msi-lib.h>
+#include <linux/irqdomain.h>
+#include <linux/spinlock.h>
+#include <linux/msi.h>
+
+#include <asm/irq.h>
+#include <asm/loongarch.h>
+#include <asm/loongson.h>
+#include <asm/numa.h>
+#include <asm/setup.h>
+#include <larchintrin.h>
+
+#include "irq-loongson.h"
+
+#define LOONGARCH_IOCSR_REDIRECT_CFG           0x15e0
+#define LOONGARCH_IOCSR_REDIRECT_TBR           0x15e8  /* IRT BASE REG*/
+#define LOONGARCH_IOCSR_REDIRECT_CQB           0x15f0  /* IRT CACHE QUEUE BASE */
+#define LOONGARCH_IOCSR_REDIRECT_CQH           0x15f8  /* IRT CACHE QUEUE HEAD, 32bit */
+#define LOONGARCH_IOCSR_REDIRECT_CQT           0x15fc  /* IRT CACHE QUEUE TAIL, 32bit */
+
+#define GPID_ADDR_MASK			GENMASK(47, 6)
+#define GPID_ADDR_SHIFT 		6
+
+#define CQB_SIZE_MASK			0xf
+#define CQB_ADDR_MASK			GENMASK(47, 12)
+
+#define CFG_DISABLE_IDLE		2
+#define INVALID_INDEX			0
+
+#define MAX_IR_ENGINES			16
+
+struct redirect_entry {
+	struct  {
+		u64	valid	: 1,
+			res1	: 5,
+			gpid	: 42,
+			res2	: 8,
+			vector	: 8;
+	}	lo;
+	u64	hi;
+};
+#define IRD_ENTRY_SIZE			sizeof(struct redirect_entry)
+#define IRD_ENTRIES			65536
+#define IRD_TABLE_PAGE_ORDER		get_order(IRD_ENTRIES * IRD_ENTRY_SIZE)
+
+struct redirect_gpid {
+	u64	pir[4];      // Pending interrupt requested
+	u8	en	: 1, // doorbell
+		res0	: 7;
+	u8	irqnum;
+	u16	res1;
+	u32	dst;
+	u32	rsvd[6];
+};
+
+struct irde_inv_cmd {
+	union {
+		u64	cmd_info;
+		struct {
+			u64	res1		: 4,
+				type		: 1,
+				need_notice	: 1,
+				pad		: 2,
+				index		: 16,
+				pad2		: 40;
+		}	index;
+	};
+	u64		notice_addr;
+};
+#define IRD_CMD_SIZE			sizeof(struct irde_inv_cmd)
+#define INVALID_QUEUE_SIZE		4096
+#define INV_QUEUE_PAGE_ORDER		get_order(INVALID_QUEUE_SIZE * IRD_CMD_SIZE)
+
+struct redirect_table {
+	struct redirect_entry	*table;
+	unsigned long		*bitmap;
+	raw_spinlock_t		lock;
+};
+
+struct redirect_queue {
+	struct irde_inv_cmd	*cmd_base;
+	int			head;
+	int			tail;
+	raw_spinlock_t		lock;
+};
+
+struct irde_desc {
+	struct	redirect_table	ird_table;
+	struct	redirect_queue	inv_queue;
+	int			node;
+};
+
+struct redirect_item {
+	int			index;
+	struct irde_desc	*irde;
+	struct redirect_gpid	*gpid;
+};
+
+
+static struct irq_domain *redirect_domain;
+static struct irde_desc irde_descs[MAX_IR_ENGINES];
+static phys_addr_t msi_base_addr;
+static phys_addr_t redirect_reg_base = LOONGSON_REG_BASE;
+
+#define REDIRECT_REG(reg, node) \
+	((void __iomem *)(IO_BASE | redirect_reg_base | (u64)(node) << NODE_ADDRSPACE_SHIFT | (reg)))
+
+static inline void redirect_write_reg64(u32 node, u64 val, u32 reg)
+{
+	void __iomem *reg_addr = REDIRECT_REG(reg, node);
+
+	return writeq(val, reg_addr);
+}
+
+static inline void redirect_write_reg32(u32 node, u32 val, u32 reg)
+{
+	void __iomem *reg_addr = REDIRECT_REG(reg, node);
+
+	return writel(val, reg_addr);
+}
+
+static inline u32 redirect_read_reg32(u32 node, u32 reg)
+{
+	void __iomem *reg_addr = REDIRECT_REG(reg, node);
+
+	return readl(reg_addr);
+}
+
+static inline struct redirect_entry *item_get_entry(struct redirect_item *item)
+{
+	return item->irde->ird_table.table + item->index;
+}
+
+static inline bool invalid_queue_is_full(int node, u32 *tail)
+{
+	u32 head = redirect_read_reg32(node, LOONGARCH_IOCSR_REDIRECT_CQH);
+
+	*tail = redirect_read_reg32(node, LOONGARCH_IOCSR_REDIRECT_CQT);
+
+	return head == ((*tail + 1) % INVALID_QUEUE_SIZE);
+}
+
+static void invalid_enqueue(struct redirect_item *item, struct irde_inv_cmd *cmd)
+{
+	struct redirect_queue *inv_queue = &item->irde->inv_queue;
+	u32 tail;
+
+	guard(raw_spinlock_irqsave)(&inv_queue->lock);
+
+	while (invalid_queue_is_full(item->irde->node, &tail))
+		cpu_relax();
+
+	memcpy(&inv_queue->cmd_base[tail], cmd, sizeof(*cmd));
+
+	redirect_write_reg32(item->irde->node, (tail + 1) % INVALID_QUEUE_SIZE,
+			     LOONGARCH_IOCSR_REDIRECT_CQT);
+}
+
+static void irde_invalid_entry(struct redirect_item *item)
+{
+	struct irde_inv_cmd cmd;
+	u64 raddr = 0;
+
+	cmd.cmd_info = 0;
+	cmd.index.type = INVALID_INDEX;
+	cmd.index.need_notice = 1;
+	cmd.index.index = item->index;
+	cmd.notice_addr = (u64)(__pa(&raddr));
+
+	invalid_enqueue(item, &cmd);
+
+	/*
+	 * CPU needs to wait here for cmd to complete, and it determines this
+	 * by checking whether invalid queue has already written a valid value
+	 * to cmd.notice_addr.
+	 */
+	while (!raddr)
+		cpu_relax();
+
+}
+
+static inline struct avecintc_data *irq_data_get_avec_data(struct irq_data *data)
+{
+	return data->parent_data->chip_data;
+}
+
+static int redirect_table_alloc(int node, u32 nr_irqs)
+{
+	struct redirect_table *ird_table = &irde_descs[node].ird_table;
+	unsigned int index, order;
+
+	if (nr_irqs > 1) {
+		nr_irqs = __roundup_pow_of_two(nr_irqs);
+		order = ilog2(nr_irqs);
+	}
+
+	guard(raw_spinlock_irqsave)(&ird_table->lock);
+
+	index = bitmap_find_free_region(ird_table->bitmap,
+					IRD_ENTRIES, order);
+	if (index < 0) {
+		pr_err("No redirect entry to use\n");
+		return -ENOMEM;
+	}
+
+	return index;
+}
+
+static void redirect_table_free(struct redirect_item *item)
+{
+	struct redirect_table *ird_table = &item->irde->ird_table;
+	struct redirect_entry *entry = item_get_entry(item);
+
+	memset(entry, 0, sizeof(*entry));
+
+	scoped_guard(raw_spinlock_irq, &ird_table->lock)
+		clear_bit(item->index, ird_table->bitmap);
+
+	kfree(item->gpid);
+
+	irde_invalid_entry(item);
+}
+
+static inline void redirect_domain_prepare_entry(struct redirect_item *item,
+						 struct avecintc_data *adata)
+{
+	struct redirect_entry *entry = item_get_entry(item);
+
+	item->gpid->en = 1;
+	item->gpid->irqnum = adata->vec;
+	item->gpid->dst = adata->cpu;
+
+	entry->lo.valid = 1;
+	entry->lo.gpid = ((u64)item->gpid & GPID_ADDR_MASK) >> GPID_ADDR_SHIFT;
+	entry->lo.vector = 0xff;
+}
+
+static int redirect_set_affinity(struct irq_data *data, const struct cpumask *dest, bool force)
+{
+	struct redirect_item *item = data->chip_data;
+	int ret;
+
+	ret = irq_chip_set_affinity_parent(data, dest, force);
+	if (ret == IRQ_SET_MASK_OK_DONE) {
+		return ret;
+	} else if (ret) {
+		pr_err("IRDE:set_affinity error %d\n", ret);
+		return ret;
+	}
+	struct avecintc_data *adata = irq_data_get_avec_data(data);
+
+	redirect_domain_prepare_entry(item, adata);
+	irde_invalid_entry(item);
+	avecintc_sync(adata);
+
+	return IRQ_SET_MASK_OK;
+}
+
+static void redirect_compose_msi_msg(struct irq_data *d, struct msi_msg *msg)
+{
+	struct redirect_item *item = irq_data_get_irq_chip_data(d);
+
+	msg->address_lo = (msi_base_addr | 1 << 2) ;
+	msg->address_hi = 0x0;
+	msg->data = item->index;
+}
+
+static struct irq_chip loongarch_redirect_chip = {
+	.name			= "REDIRECT",
+	.irq_ack		= irq_chip_ack_parent,
+	.irq_mask		= irq_chip_mask_parent,
+	.irq_unmask		= irq_chip_unmask_parent,
+	.irq_set_affinity	= redirect_set_affinity,
+	.irq_compose_msi_msg	= redirect_compose_msi_msg,
+};
+
+static void redirect_free_resources(struct irq_domain *domain, unsigned int virq,
+				    unsigned int nr_irqs)
+{
+	for (int i = 0; i < nr_irqs; i++) {
+		struct irq_data *irq_data = irq_domain_get_irq_data(domain, virq + i);
+
+		if (irq_data && irq_data->chip_data) {
+			struct redirect_item *item = irq_data->chip_data;
+
+			redirect_table_free(item);
+			kfree(item);
+		}
+	}
+}
+
+static int redirect_domain_alloc(struct irq_domain *domain, unsigned int virq,
+				 unsigned int nr_irqs, void *arg)
+{
+	msi_alloc_info_t *info = arg;
+	int ret, i, node, index;
+
+	node = dev_to_node(info->desc->dev);
+
+	ret = irq_domain_alloc_irqs_parent(domain, virq, nr_irqs, arg);
+	if (ret < 0)
+		return ret;
+
+	index = redirect_table_alloc(node, nr_irqs);
+	if (index < 0) {
+		pr_err("Alloc redirect table entry failed\n");
+		return -ENOMEM;
+	}
+
+	for (i = 0; i < nr_irqs; i++) {
+		struct irq_data *irq_data = irq_domain_get_irq_data(domain, virq + i);
+		struct redirect_item *item;
+
+		item = kzalloc(sizeof(*item), GFP_KERNEL);
+		if (!item) {
+			pr_err("Alloc redirect descriptor failed\n");
+			goto out_free_resources;
+		}
+		item->irde = &irde_descs[node];
+
+		/*
+		 * Only bits 47:6 of the GPID are passed to the controller,
+		 * 64-byte alignment must be guarantee and make kzalloc can
+		 * align to the respective size.
+		 */
+		static_assert(sizeof(*item->gpid) == 64);
+		item->gpid = kzalloc_node(sizeof(*item->gpid), GFP_KERNEL, node);
+		if (!item->gpid) {
+			pr_err("Alloc redirect GPID failed\n");
+			goto out_free_resources;
+		}
+		item->index = index + i;
+
+		irq_data->chip_data = item;
+		irq_data->chip = &loongarch_redirect_chip;
+
+		redirect_domain_prepare_entry(item, irq_data_get_avec_data(irq_data));
+	}
+	return 0;
+
+out_free_resources:
+	redirect_free_resources(domain, virq, nr_irqs);
+	irq_domain_free_irqs_common(domain, virq, nr_irqs);
+
+	return -ENOMEM;
+}
+
+static void redirect_domain_free(struct irq_domain *domain, unsigned int virq, unsigned int nr_irqs)
+{
+	redirect_free_resources(domain, virq, nr_irqs);
+	return irq_domain_free_irqs_common(domain, virq, nr_irqs);
+}
+
+static const struct irq_domain_ops redirect_domain_ops = {
+	.alloc		= redirect_domain_alloc,
+	.free		= redirect_domain_free,
+	.select		= msi_lib_irq_domain_select,
+};
+
+static int redirect_queue_init(struct irde_desc *irde)
+{
+	struct redirect_queue *inv_queue = &irde->inv_queue;
+	struct folio *folio;
+
+	folio = __folio_alloc_node(GFP_KERNEL | __GFP_ZERO, INV_QUEUE_PAGE_ORDER, irde->node);
+	if (!folio) {
+		pr_err("Node [%d] invalid queue alloc pages failed!\n", irde->node);
+		return -ENOMEM;
+	}
+
+	inv_queue->cmd_base = folio_address(folio);
+	inv_queue->head = 0;
+	inv_queue->tail = 0;
+	raw_spin_lock_init(&inv_queue->lock);
+
+	return 0;
+}
+
+static int redirect_table_init(struct irde_desc *irde)
+{
+	struct redirect_table *ird_table = &irde->ird_table;
+	unsigned long *bitmap;
+	struct folio *folio;
+	int node = irde->node;
+
+	folio = __folio_alloc_node(GFP_KERNEL | __GFP_ZERO, IRD_TABLE_PAGE_ORDER, node);
+	if (!folio) {
+		pr_err("Node [%d] redirect table alloc pages failed!\n", node);
+		return -ENOMEM;
+	}
+	ird_table->table = folio_address(folio);
+
+	bitmap = bitmap_zalloc(IRD_ENTRIES, GFP_KERNEL);
+	if (!bitmap) {
+		pr_err("Node [%d] redirect table bitmap alloc pages failed!\n", node);
+		folio_put(folio);
+		ird_table->table = NULL;
+		return -ENOMEM;
+	}
+
+	ird_table->bitmap = bitmap;
+
+	raw_spin_lock_init(&ird_table->lock);
+
+	return 0;
+}
+
+static void redirect_irde_cfg(struct irde_desc *irde)
+{
+	redirect_write_reg64(irde->node, CFG_DISABLE_IDLE, LOONGARCH_IOCSR_REDIRECT_CFG);
+	redirect_write_reg64(irde->node, __pa(irde->ird_table.table), LOONGARCH_IOCSR_REDIRECT_TBR);
+	redirect_write_reg32(irde->node, 0, LOONGARCH_IOCSR_REDIRECT_CQH);
+	redirect_write_reg32(irde->node, 0, LOONGARCH_IOCSR_REDIRECT_CQT);
+	redirect_write_reg64(irde->node, ((u64)irde->inv_queue.cmd_base & CQB_ADDR_MASK) |
+			     CQB_SIZE_MASK, LOONGARCH_IOCSR_REDIRECT_CQB);
+}
+
+static void __redirect_irde_fini(struct irde_desc *irde)
+{
+	struct redirect_table *ird_table = &irde_descs->ird_table;
+	struct redirect_queue *inv_queue = &irde_descs->inv_queue;
+
+	if (ird_table->table) {
+		folio_put(virt_to_folio(ird_table->table));
+		ird_table->table = NULL;
+	}
+
+	if (ird_table->bitmap) {
+		bitmap_free(ird_table->bitmap);
+		ird_table->bitmap = NULL;
+	}
+
+	if (inv_queue->cmd_base) {
+		folio_put(virt_to_folio(inv_queue->cmd_base));
+		inv_queue->cmd_base = NULL;
+	}
+}
+
+static inline void redirect_irde_fini(int node)
+{
+	__redirect_irde_fini(&irde_descs[node]);
+}
+
+static int redirect_irde_init(int node)
+{
+	struct irde_desc *irde = &irde_descs[node];
+	int ret;
+
+	irde->node = node;
+
+	ret = redirect_table_init(irde);
+	if (ret)
+		return ret;
+
+	ret = redirect_queue_init(irde);
+	if (ret) {
+		__redirect_irde_fini(irde);
+		return ret;
+	}
+
+	redirect_irde_cfg(irde);
+
+	return 0;
+}
+
+static int __init pch_msi_parse_madt(union acpi_subtable_headers *header, const unsigned long end)
+{
+	struct acpi_madt_msi_pic *pchmsi_entry = (struct acpi_madt_msi_pic *)header;
+
+	msi_base_addr = pchmsi_entry->msg_address - AVEC_MSG_OFFSET;
+
+	return pch_msi_acpi_init_avec(redirect_domain);
+}
+
+static int __init acpi_cascade_irqdomain_init(void)
+{
+	return acpi_table_parse_madt(ACPI_MADT_TYPE_MSI_PIC, pch_msi_parse_madt, 1);
+}
+
+int __init redirect_acpi_init(struct irq_domain *parent)
+{
+	struct fwnode_handle *fwnode;
+	int ret = -EINVAL, node;
+
+	fwnode = irq_domain_alloc_named_fwnode("redirect");
+	if (!fwnode) {
+		pr_err("Unable to alloc redirect domain handle\n");
+		goto fail;
+	}
+
+	redirect_domain = irq_domain_create_hierarchy(parent, 0, IRD_ENTRIES, fwnode,
+						      &redirect_domain_ops, irde_descs);
+	if (!redirect_domain) {
+		pr_err("Unable to alloc redirect domain\n");
+		goto out_free_fwnode;
+	}
+
+
+	for_each_node_mask(node, node_possible_map) {
+		ret = redirect_irde_init(node);
+		if (ret)
+			goto out_clear_irde;
+	}
+
+	ret = acpi_cascade_irqdomain_init();
+	if (ret < 0) {
+		pr_err("Failed to cascade IRQ domain, ret=%d\n", ret);
+		goto out_clear_irde;
+	}
+
+	pr_info("loongarch irq redirect modules init succeeded\n");
+	return 0;
+
+out_clear_irde:
+	for_each_node_mask(node, node_possible_map)
+		redirect_irde_fini(node);
+
+	irq_domain_remove(redirect_domain);
+out_free_fwnode:
+	irq_domain_free_fwnode(fwnode);
+fail:
+	return ret;
+}
-- 
2.41.0

Re: [PATCH v9 4/4] irqchip/irq-loongarch-ir:Add Redirect irqchip support

[kernel test robot]

Hi Tianyang,

kernel test robot noticed the following build warnings:

[auto build test WARNING on tip/irq/core]
[also build test WARNING on linus/master v6.19-rc7 next-20260129]
[If your patch is applied to the wrong git tree, kindly drop us a note.
And when submitting patch, we suggest to use '--base' as documented in
https://git-scm.com/docs/git-format-patch#_base_tree_information]

url:    https://github.com/intel-lab-lkp/linux/commits/Tianyang-Zhang/Docs-LoongArch-Add-Advanced-Extended-Redirect-IRQ-model-description/20260130-110249
base:   tip/irq/core
patch link:    https://lore.kernel.org/r/20260130025941.2140582-5-zhangtianyang%40loongson.cn
patch subject: [PATCH v9 4/4] irqchip/irq-loongarch-ir:Add Redirect irqchip support
config: loongarch-allnoconfig (https://download.01.org/0day-ci/archive/20260130/202601301601.EonMu52D-lkp@intel.com/config)
compiler: clang version 22.0.0git (https://github.com/llvm/llvm-project 9b8addffa70cee5b2acc5454712d9cf78ce45710)
reproduce (this is a W=1 build): (https://download.01.org/0day-ci/archive/20260130/202601301601.EonMu52D-lkp@intel.com/reproduce)

If you fix the issue in a separate patch/commit (i.e. not just a new version of
the same patch/commit), kindly add following tags
| Reported-by: kernel test robot <lkp@intel.com>
| Closes: https://lore.kernel.org/oe-kbuild-all/202601301601.EonMu52D-lkp@intel.com/

All warnings (new ones prefixed by >>):

>> drivers/irqchip/irq-loongarch-ir.c:203:6: warning: variable 'order' is used uninitialized whenever 'if' condition is false [-Wsometimes-uninitialized]
     203 |         if (nr_irqs > 1) {
         |             ^~~~~~~~~~~
   drivers/irqchip/irq-loongarch-ir.c:211:19: note: uninitialized use occurs here
     211 |                                         IRD_ENTRIES, order);
         |                                                      ^~~~~
   drivers/irqchip/irq-loongarch-ir.c:203:2: note: remove the 'if' if its condition is always true
     203 |         if (nr_irqs > 1) {
         |         ^~~~~~~~~~~~~~~~
   drivers/irqchip/irq-loongarch-ir.c:201:27: note: initialize the variable 'order' to silence this warning
     201 |         unsigned int index, order;
         |                                  ^
         |                                   = 0
   1 warning generated.


vim +203 drivers/irqchip/irq-loongarch-ir.c

   197	
   198	static int redirect_table_alloc(int node, u32 nr_irqs)
   199	{
   200		struct redirect_table *ird_table = &irde_descs[node].ird_table;
   201		unsigned int index, order;
   202	
 > 203		if (nr_irqs > 1) {
   204			nr_irqs = __roundup_pow_of_two(nr_irqs);
   205			order = ilog2(nr_irqs);
   206		}
   207	
   208		guard(raw_spinlock_irqsave)(&ird_table->lock);
   209	
   210		index = bitmap_find_free_region(ird_table->bitmap,
   211						IRD_ENTRIES, order);
   212		if (index < 0) {
   213			pr_err("No redirect entry to use\n");
   214			return -ENOMEM;
   215		}
   216	
   217		return index;
   218	}
   219	

-- 
0-DAY CI Kernel Test Service
https://github.com/intel/lkp-tests/wiki

Re: [PATCH v9 4/4] irqchip/irq-loongarch-ir:Add Redirect irqchip support

[Thomas Gleixner]

On Fri, Jan 30 2026 at 10:59, Tianyang Zhang wrote:
> +// SPDX-License-Identifier: GPL-2.0

GPL-2.0-only please

> +/*
> + * Copyright (C) 2020 Loongson Technologies, Inc.

This was written 6 years ago already?

> + */
> +
> +#include <linux/cpuhotplug.h>
> +#include <linux/init.h>
> +#include <linux/interrupt.h>
> +#include <linux/kernel.h>
> +#include <linux/irq.h>
> +#include <linux/irqchip.h>
> +#include <linux/irqchip/irq-msi-lib.h>
> +#include <linux/irqdomain.h>
> +#include <linux/spinlock.h>
> +#include <linux/msi.h>

Includes should be alphabetically ordered.

> +#define REDIRECT_REG(reg, node) \
> +	((void __iomem *)(IO_BASE | redirect_reg_base | (u64)(node) << NODE_ADDRSPACE_SHIFT | (reg)))
> +
> +static inline void redirect_write_reg64(u32 node, u64 val, u32 reg)
> +{
> +	void __iomem *reg_addr = REDIRECT_REG(reg, node);
> +
> +	return writeq(val, reg_addr);

Bogus return and you can simplify this to

      writeq(val, REDIRECT_REG(reg, node));

No?

> +}
> +
> +static inline void redirect_write_reg32(u32 node, u32 val, u32 reg)
> +{
> +	void __iomem *reg_addr = REDIRECT_REG(reg, node);
> +
> +	return writel(val, reg_addr);

Ditto

> +}
> +
> +static inline u32 redirect_read_reg32(u32 node, u32 reg)
> +{
> +	void __iomem *reg_addr = REDIRECT_REG(reg, node);
> +
> +	return readl(reg_addr);

Condense to single line as well

> +static void irde_invalid_entry(struct redirect_item *item)

This should be named irde_invalidate_entry() as that's what the function
is about. irq_invalid_entry() reads more like a function which check for
an invalid entry.

> +{
> +	struct irde_inv_cmd cmd;
> +	u64 raddr = 0;
> +
> +	cmd.cmd_info = 0;
> +	cmd.index.type = INVALID_INDEX;
> +	cmd.index.need_notice = 1;
> +	cmd.index.index = item->index;
> +	cmd.notice_addr = (u64)(__pa(&raddr));
> +
> +	invalid_enqueue(item, &cmd);
> +
> +	/*
> +	 * CPU needs to wait here for cmd to complete, and it determines this

The CPU 

> +	 * by checking whether invalid queue has already written a valid value

whether the invalidation queue

> +	 * to cmd.notice_addr.
> +	 */

> +static int redirect_table_alloc(int node, u32 nr_irqs)
> +{
> +	struct redirect_table *ird_table = &irde_descs[node].ird_table;
> +	unsigned int index, order;
> +
> +	if (nr_irqs > 1) {
> +		nr_irqs = __roundup_pow_of_two(nr_irqs);
> +		order = ilog2(nr_irqs);
> +	}
> +
> +	guard(raw_spinlock_irqsave)(&ird_table->lock);
> +
> +	index = bitmap_find_free_region(ird_table->bitmap,
> +					IRD_ENTRIES, order);

Get rid of this pointless line break. You have 100 characters and the
above even fits into 80 

> +	if (index < 0) {
> +		pr_err("No redirect entry to use\n");
> +		return -ENOMEM;
> +	}

> +static int redirect_domain_alloc(struct irq_domain *domain, unsigned int virq,
> +				 unsigned int nr_irqs, void *arg)
> +{
> +	msi_alloc_info_t *info = arg;
> +	int ret, i, node, index;
> +
> +	node = dev_to_node(info->desc->dev);
> +
> +	ret = irq_domain_alloc_irqs_parent(domain, virq, nr_irqs, arg);
> +	if (ret < 0)
> +		return ret;
> +
> +	index = redirect_table_alloc(node, nr_irqs);
> +	if (index < 0) {
> +		pr_err("Alloc redirect table entry failed\n");
> +		return -ENOMEM;
> +	}
> +
> +	for (i = 0; i < nr_irqs; i++) {
> +		struct irq_data *irq_data = irq_domain_get_irq_data(domain, virq + i);
> +		struct redirect_item *item;
> +
> +		item = kzalloc(sizeof(*item), GFP_KERNEL);
> +		if (!item) {
> +			pr_err("Alloc redirect descriptor failed\n");
> +			goto out_free_resources;
> +		}
> +		item->irde = &irde_descs[node];
> +
> +		/*
> +		 * Only bits 47:6 of the GPID are passed to the controller,
> +		 * 64-byte alignment must be guarantee and make kzalloc can
> +		 * align to the respective size.

-ENOPARSE

> +		 */

> +static const struct irq_domain_ops redirect_domain_ops = {
> +	.alloc		= redirect_domain_alloc,
> +	.free		= redirect_domain_free,
> +	.select		= msi_lib_irq_domain_select,
> +};

> +static void __redirect_irde_fini(struct irde_desc *irde)

This schoolbook _fini() naming is just lame and nondescriptive. Please
use descriptive function names which make it clear what this is about,
e.g. redirect_free_irde() or something like that.

Also this should be __init, no?

> +{
> +	struct redirect_table *ird_table = &irde_descs->ird_table;
> +	struct redirect_queue *inv_queue = &irde_descs->inv_queue;

> +static inline void redirect_irde_fini(int node)
> +{
> +	__redirect_irde_fini(&irde_descs[node]);

This indirection is really pointless. Just move the '&irde_descs[node]' to
the only caller.

> +int __init redirect_acpi_init(struct irq_domain *parent)
> +{
> +	struct fwnode_handle *fwnode;
> +	int ret = -EINVAL, node;
> +
> +	fwnode = irq_domain_alloc_named_fwnode("redirect");
> +	if (!fwnode) {
> +		pr_err("Unable to alloc redirect domain handle\n");
> +		goto fail;
> +	}
> +
> +	redirect_domain = irq_domain_create_hierarchy(parent, 0, IRD_ENTRIES, fwnode,
> +						      &redirect_domain_ops, irde_descs);
> +	if (!redirect_domain) {
> +		pr_err("Unable to alloc redirect domain\n");
> +		goto out_free_fwnode;
> +	}
> +
> +

stray newline

> +	for_each_node_mask(node, node_possible_map) {
> +		ret = redirect_irde_init(node);
> +		if (ret)
> +			goto out_clear_irde;
> +	}
> +
> +	ret = acpi_cascade_irqdomain_init();
> +	if (ret < 0) {
> +		pr_err("Failed to cascade IRQ domain, ret=%d\n", ret);
> +		goto out_clear_irde;
> +	}
> +
> +	pr_info("loongarch irq redirect modules init succeeded\n");

You really want to have:

#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt

at the top of the file, so that all printk()s in this file are properly
prefixed.

Thanks,

        tglx

Re: [PATCH v9 4/4] irqchip/irq-loongarch-ir:Add Redirect irqchip support

[Tianyang Zhang]

Hi, Thomas

在 2026/1/30 下午4:54, Thomas Gleixner 写道:
> On Fri, Jan 30 2026 at 10:59, Tianyang Zhang wrote:
>> +// SPDX-License-Identifier: GPL-2.0
> GPL-2.0-only please
Ok, I got it
>
>> +/*
>> + * Copyright (C) 2020 Loongson Technologies, Inc.
> This was written 6 years ago already?
This was copied from other files, and I will rework it, thanks
>
>> + */
>> +
>> +#include <linux/cpuhotplug.h>
>> +#include <linux/init.h>
>> +#include <linux/interrupt.h>
>> +#include <linux/kernel.h>
>> +#include <linux/irq.h>
>> +#include <linux/irqchip.h>
>> +#include <linux/irqchip/irq-msi-lib.h>
>> +#include <linux/irqdomain.h>
>> +#include <linux/spinlock.h>
>> +#include <linux/msi.h>
> Includes should be alphabetically ordered.
Ok, I got it
>
>> +#define REDIRECT_REG(reg, node) \
>> +	((void __iomem *)(IO_BASE | redirect_reg_base | (u64)(node) << NODE_ADDRSPACE_SHIFT | (reg)))
>> +
>> +static inline void redirect_write_reg64(u32 node, u64 val, u32 reg)
>> +{
>> +	void __iomem *reg_addr = REDIRECT_REG(reg, node);
>> +
>> +	return writeq(val, reg_addr);
> Bogus return and you can simplify this to
>
>        writeq(val, REDIRECT_REG(reg, node));
>
> No?
Ok, this is more reasonable.
>
>> +}
>> +
>> +static inline u32 redirect_read_reg32(u32 node, u32 reg)
>> +{
>> +	void __iomem *reg_addr = REDIRECT_REG(reg, node);
>> +
>> +	return readl(reg_addr);
> Condense to single line as well
Ok, I got it
>
>> +static void irde_invalid_entry(struct redirect_item *item)
> This should be named irde_invalidate_entry() as that's what the function
> is about. irq_invalid_entry() reads more like a function which check for
> an invalid entry.
Ok, I got it
>
>> +{
>> +	struct irde_inv_cmd cmd;
>> +	u64 raddr = 0;
>> +
>> +	cmd.cmd_info = 0;
>> +	cmd.index.type = INVALID_INDEX;
>> +	cmd.index.need_notice = 1;
>> +	cmd.index.index = item->index;
>> +	cmd.notice_addr = (u64)(__pa(&raddr));
>> +
>> +	invalid_enqueue(item, &cmd);
>> +
>> +	/*
>> +	 * CPU needs to wait here for cmd to complete, and it determines this
> The CPU
Ok, I got it, thanks
>
>> +	 * by checking whether invalid queue has already written a valid value
> whether the invalidation queue
Ok, I got it, thanks
>
>> +static int redirect_table_alloc(int node, u32 nr_irqs)
>> +{
>> +	struct redirect_table *ird_table = &irde_descs[node].ird_table;
>> +	unsigned int index, order;
>> +
>> +	if (nr_irqs > 1) {
>> +		nr_irqs = __roundup_pow_of_two(nr_irqs);
>> +		order = ilog2(nr_irqs);
>> +	}
>> +
>> +	guard(raw_spinlock_irqsave)(&ird_table->lock);
>> +
>> +	index = bitmap_find_free_region(ird_table->bitmap,
>> +					IRD_ENTRIES, order);
> Get rid of this pointless line break. You have 100 characters and the
> above even fits into 80
Ok, I got it
>
>> +	if (index < 0) {
>> +		pr_err("No redirect entry to use\n");
>> +		return -ENOMEM;
>> +	}
>> +static int redirect_domain_alloc(struct irq_domain *domain, unsigned int virq,
>> +				 unsigned int nr_irqs, void *arg)
>> +{
>> +	msi_alloc_info_t *info = arg;
>> +	int ret, i, node, index;
>> +
>> +	node = dev_to_node(info->desc->dev);
>> +
>> +	ret = irq_domain_alloc_irqs_parent(domain, virq, nr_irqs, arg);
>> +	if (ret < 0)
>> +		return ret;
>> +
>> +	index = redirect_table_alloc(node, nr_irqs);
>> +	if (index < 0) {
>> +		pr_err("Alloc redirect table entry failed\n");
>> +		return -ENOMEM;
>> +	}
>> +
>> +	for (i = 0; i < nr_irqs; i++) {
>> +		struct irq_data *irq_data = irq_domain_get_irq_data(domain, virq + i);
>> +		struct redirect_item *item;
>> +
>> +		item = kzalloc(sizeof(*item), GFP_KERNEL);
>> +		if (!item) {
>> +			pr_err("Alloc redirect descriptor failed\n");
>> +			goto out_free_resources;
>> +		}
>> +		item->irde = &irde_descs[node];
>> +
>> +		/*
>> +		 * Only bits 47:6 of the GPID are passed to the controller,
>> +		 * 64-byte alignment must be guarantee and make kzalloc can
>> +		 * align to the respective size.
> -ENOPARSE
Ok ,I got it
>> +		 */
>> +static const struct irq_domain_ops redirect_domain_ops = {
>> +	.alloc		= redirect_domain_alloc,
>> +	.free		= redirect_domain_free,
>> +	.select		= msi_lib_irq_domain_select,
>> +};
>> +static void __redirect_irde_fini(struct irde_desc *irde)
> This schoolbook _fini() naming is just lame and nondescriptive. Please
> use descriptive function names which make it clear what this is about,
> e.g. redirect_free_irde() or something like that.
>
> Also this should be __init, no?
Yes , This code style is indeed not very appropriate, I will adjust the 
function name
>
>> +{
>> +	struct redirect_table *ird_table = &irde_descs->ird_table;
>> +	struct redirect_queue *inv_queue = &irde_descs->inv_queue;
>> +static inline void redirect_irde_fini(int node)
>> +{
>> +	__redirect_irde_fini(&irde_descs[node]);
> This indirection is really pointless. Just move the '&irde_descs[node]' to
> the only caller.
Ok, I got it
>
>> +int __init redirect_acpi_init(struct irq_domain *parent)
>> +{
>> +	struct fwnode_handle *fwnode;
>> +	int ret = -EINVAL, node;
>> +
>> +	fwnode = irq_domain_alloc_named_fwnode("redirect");
>> +	if (!fwnode) {
>> +		pr_err("Unable to alloc redirect domain handle\n");
>> +		goto fail;
>> +	}
>> +
>> +	redirect_domain = irq_domain_create_hierarchy(parent, 0, IRD_ENTRIES, fwnode,
>> +						      &redirect_domain_ops, irde_descs);
>> +	if (!redirect_domain) {
>> +		pr_err("Unable to alloc redirect domain\n");
>> +		goto out_free_fwnode;
>> +	}
>> +
>> +
> stray newline
Ok, I got it
>
>> +	for_each_node_mask(node, node_possible_map) {
>> +		ret = redirect_irde_init(node);
>> +		if (ret)
>> +			goto out_clear_irde;
>> +	}
>> +
>> +	ret = acpi_cascade_irqdomain_init();
>> +	if (ret < 0) {
>> +		pr_err("Failed to cascade IRQ domain, ret=%d\n", ret);
>> +		goto out_clear_irde;
>> +	}
>> +
>> +	pr_info("loongarch irq redirect modules init succeeded\n");
> You really want to have:
>
> #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
>
> at the top of the file, so that all printk()s in this file are properly
> prefixed.
Ok, I got it, thanks
>
> Thanks,
>
>          tglx

Tiyang