[PATCH 0/2] acpi, gicv3-its, numa: Adding numa node mapping for its

ACPI 6.2 have added SRAT subtable to define proximity domain for ITS
devices. This patchset Updates acpi header file from acpica repo and
adds numa node mapping for its devices.

Ganapatrao Kulkarni (2):
  ACPICA: ACPI 6.2: Add support for new SRAT subtable
  acpi, gicv3-its, numa: Adding numa node mapping for gic-its devices

 arch/arm64/include/asm/acpi.h    |  2 ++
 arch/arm64/kernel/acpi_numa.c    | 59 ++++++++++++++++++++++++++++++++++++++++
 drivers/acpi/numa.c              | 32 +++++++++++++++++++++-
 drivers/irqchip/irq-gic-v3-its.c |  3 +-
 include/acpi/actbl1.h            | 12 +++++++-
 include/linux/acpi.h             |  3 ++
 6 files changed, 108 insertions(+), 3 deletions(-)

-- 
1.8.1.4

Add GIC ITS Affinity (ACPI 6.2) subtable to SRAT table.

ACPICA commit 5bc67f63918da249bfe279ee461d152bb3e6f55b
Link: https://github.com/acpica/acpica/commit/5bc67f6

Signed-off-by: Ganapatrao Kulkarni <redacted>
---
 include/acpi/actbl1.h | 12 +++++++++++-
 1 file changed, 11 insertions(+), 1 deletion(-)

diff --git a/include/acpi/actbl1.h b/include/acpi/actbl1.h
index b4ce55c..253c9db 100644
--- a/include/acpi/actbl1.h
+++ b/include/acpi/actbl1.h

@@ -1192,7 +1192,8 @@ enum acpi_srat_type {
 	ACPI_SRAT_TYPE_MEMORY_AFFINITY = 1,
 	ACPI_SRAT_TYPE_X2APIC_CPU_AFFINITY = 2,
 	ACPI_SRAT_TYPE_GICC_AFFINITY = 3,
-	ACPI_SRAT_TYPE_RESERVED = 4	/* 4 and greater are reserved */
+	ACPI_SRAT_TYPE_GIC_ITS_AFFINITY = 4,	/* ACPI 6.2 */
+	ACPI_SRAT_TYPE_RESERVED = 5	/* 5 and greater are reserved */
 };
 
 /*

@@ -1260,6 +1261,15 @@ struct acpi_srat_gicc_affinity {
 	u32 clock_domain;
 };
 
+/* 4: GIC ITS Affinity (ACPI 6.2) */
+
+struct acpi_srat_its_affinity {
+	struct acpi_subtable_header header;
+	u32 proximity_domain;
+	u16 reserved;
+	u32 its_id;
+};
+
 /* Flags for struct acpi_srat_gicc_affinity */
 
 #define ACPI_SRAT_GICC_ENABLED     (1)	/* 00: Use affinity structure */

-- 
1.8.1.4

Add code to parse SRAT ITS Affinity sub table as defined in ACPI 6.2
Later in driver probe, its devices are mapped to numa node
using its id to proximity domain mapping.

Signed-off-by: Ganapatrao Kulkarni <redacted>
---
 arch/arm64/include/asm/acpi.h    |  2 ++
 arch/arm64/kernel/acpi_numa.c    | 59 ++++++++++++++++++++++++++++++++++++++++
 drivers/acpi/numa.c              | 32 +++++++++++++++++++++-
 drivers/irqchip/irq-gic-v3-its.c |  3 +-
 include/linux/acpi.h             |  3 ++
 5 files changed, 97 insertions(+), 2 deletions(-)

diff --git a/arch/arm64/include/asm/acpi.h b/arch/arm64/include/asm/acpi.h
index 0e99978..60ea7d1 100644
--- a/arch/arm64/include/asm/acpi.h
+++ b/arch/arm64/include/asm/acpi.h

@@ -143,9 +143,11 @@ static inline void arch_apei_flush_tlb_one(unsigned long addr)
 #ifdef CONFIG_ACPI_NUMA
 int arm64_acpi_numa_init(void);
 int acpi_numa_get_nid(unsigned int cpu, u64 hwid);
+int acpi_numa_get_its_nid(u32 its_id);
 #else
 static inline int arm64_acpi_numa_init(void) { return -ENOSYS; }
 static inline int acpi_numa_get_nid(unsigned int cpu, u64 hwid) { return NUMA_NO_NODE; }
+static inline int acpi_numa_get_its_nid(u32 its_id) { return NUMA_NO_NODE; }
 #endif /* CONFIG_ACPI_NUMA */
 
 #define ACPI_TABLE_UPGRADE_MAX_PHYS MEMBLOCK_ALLOC_ACCESSIBLE

diff --git a/arch/arm64/kernel/acpi_numa.c b/arch/arm64/kernel/acpi_numa.c
index f01fab6..f7b2ea6 100644
--- a/arch/arm64/kernel/acpi_numa.c
+++ b/arch/arm64/kernel/acpi_numa.c

@@ -29,15 +29,24 @@
 #include <asm/numa.h>
 
 static int cpus_in_srat;
+static int its_in_srat;
 
 struct __node_cpu_hwid {
 	u32 node_id;    /* logical node containing this CPU */
 	u64 cpu_hwid;   /* MPIDR for this CPU */
 };
 
+struct __node_its_id {
+	u32 node_id;    /* numa node id */
+	u32 its_id ;   /* GIC ITS ID */
+};
+
 static struct __node_cpu_hwid early_node_cpu_hwid[NR_CPUS] = {
 [0 ... NR_CPUS - 1] = {NUMA_NO_NODE, PHYS_CPUID_INVALID} };
 
+static struct __node_its_id early_node_its_id[MAX_NUMNODES] = {
+[0 ... MAX_NUMNODES - 1] = {NUMA_NO_NODE, UINT_MAX} };
+
 int acpi_numa_get_nid(unsigned int cpu, u64 hwid)
 {
 	int i;

@@ -50,6 +59,18 @@ int acpi_numa_get_nid(unsigned int cpu, u64 hwid)
 	return NUMA_NO_NODE;
 }
 
+int acpi_numa_get_its_nid(u32 its_id)
+{
+	int i;
+
+	for (i = 0; i < its_in_srat; i++) {
+		if (its_id == early_node_its_id[i].its_id)
+			return early_node_its_id[i].node_id;
+	}
+
+	return NUMA_NO_NODE;
+}
+
 /* Callback for Proximity Domain -> ACPI processor UID mapping */
 void __init acpi_numa_gicc_affinity_init(struct acpi_srat_gicc_affinity *pa)
 {

@@ -100,6 +121,44 @@ void __init acpi_numa_gicc_affinity_init(struct acpi_srat_gicc_affinity *pa)
 		pxm, mpidr, node);
 }
 
+/* Callback for ITS ACPI Proximity Domain mapping */
+void __init acpi_numa_its_affinity_init(struct acpi_srat_its_affinity *pa)
+{
+	int pxm, node;
+
+	if (srat_disabled())
+		return;
+
+	if (pa->header.length < sizeof(struct acpi_srat_its_affinity)) {
+		pr_err("SRAT: Invalid SRAT header length: %d\n",
+			pa->header.length);
+		bad_srat();
+		return;
+	}
+
+	if (its_in_srat >= MAX_NUMNODES) {
+		pr_warn_once("SRAT: its count exceeding max count[%d]\n",
+			     MAX_NUMNODES);
+		return;
+	}
+
+	pxm = pa->proximity_domain;
+	node = acpi_map_pxm_to_node(pxm);
+
+	if (node == NUMA_NO_NODE || node >= MAX_NUMNODES) {
+		pr_err("SRAT: Too many proximity domains %d\n", pxm);
+		bad_srat();
+		return;
+	}
+
+	early_node_its_id[its_in_srat].node_id = node;
+	early_node_its_id[its_in_srat].its_id =  its_in_srat;
+	node_set(node, numa_nodes_parsed);
+	pr_info("SRAT: PXM %d -> ITS %d -> Node %d\n",
+		pxm, its_in_srat, node);
+	its_in_srat++;
+}
+
 int __init arm64_acpi_numa_init(void)
 {
 	int ret;

diff --git a/drivers/acpi/numa.c b/drivers/acpi/numa.c
index edb0c79..9c04dba 100644
--- a/drivers/acpi/numa.c
+++ b/drivers/acpi/numa.c

@@ -182,6 +182,16 @@ int acpi_map_pxm_to_online_node(int pxm)
 		}
 		break;
 
+	case ACPI_SRAT_TYPE_GIC_ITS_AFFINITY:
+		{
+			struct acpi_srat_its_affinity *p =
+			    (struct acpi_srat_its_affinity *)header;
+			pr_debug("SRAT ITS [0x%04x]) in proximity domain %d\n",
+				 p->its_id,
+				 p->proximity_domain);
+		}
+		break;
+
 	default:
 		pr_warn("Found unsupported SRAT entry (type = 0x%x)\n",
 			header->type);

@@ -390,6 +400,24 @@ static int __init acpi_parse_slit(struct acpi_table_header *table)
 	return 0;
 }
 
+static int __init
+acpi_parse_its_affinity(struct acpi_subtable_header *header,
+			 const unsigned long end)
+{
+	struct acpi_srat_its_affinity *its_affinity;
+
+	its_affinity = (struct acpi_srat_its_affinity *)header;
+	if (!its_affinity)
+		return -EINVAL;
+
+	acpi_table_print_srat_entry(header);
+
+	/* let architecture-dependent part to do it */
+	acpi_numa_its_affinity_init(its_affinity);
+
+	return 0;
+}
+
 static int __initdata parsed_numa_memblks;
 
 static int __init

@@ -445,7 +473,7 @@ int __init acpi_numa_init(void)
 
 	/* SRAT: Static Resource Affinity Table */
 	if (!acpi_table_parse(ACPI_SIG_SRAT, acpi_parse_srat)) {
-		struct acpi_subtable_proc srat_proc[3];
+		struct acpi_subtable_proc srat_proc[4];
 
 		memset(srat_proc, 0, sizeof(srat_proc));
 		srat_proc[0].id = ACPI_SRAT_TYPE_CPU_AFFINITY;

@@ -454,6 +482,8 @@ int __init acpi_numa_init(void)
 		srat_proc[1].handler = acpi_parse_x2apic_affinity;
 		srat_proc[2].id = ACPI_SRAT_TYPE_GICC_AFFINITY;
 		srat_proc[2].handler = acpi_parse_gicc_affinity;
+		srat_proc[3].id = ACPI_SRAT_TYPE_GIC_ITS_AFFINITY;
+		srat_proc[3].handler = acpi_parse_its_affinity;
 
 		acpi_table_parse_entries_array(ACPI_SIG_SRAT,
 					sizeof(struct acpi_table_srat),

diff --git a/drivers/irqchip/irq-gic-v3-its.c b/drivers/irqchip/irq-gic-v3-its.c
index 45ea1933..84936da 100644
--- a/drivers/irqchip/irq-gic-v3-its.c
+++ b/drivers/irqchip/irq-gic-v3-its.c

@@ -1861,7 +1861,8 @@ static int __init gic_acpi_parse_madt_its(struct acpi_subtable_header *header,
 		goto dom_err;
 	}
 
-	err = its_probe_one(&res, dom_handle, NUMA_NO_NODE);
+	err = its_probe_one(&res, dom_handle,
+			acpi_numa_get_its_nid(its_entry->translation_id));
 	if (!err)
 		return 0;
 

diff --git a/include/linux/acpi.h b/include/linux/acpi.h
index 137e4a3..e5e8ae3 100644
--- a/include/linux/acpi.h
+++ b/include/linux/acpi.h

@@ -260,9 +260,12 @@ int acpi_table_parse_madt(enum acpi_madt_type id,
 
 #ifdef CONFIG_ARM64
 void acpi_numa_gicc_affinity_init(struct acpi_srat_gicc_affinity *pa);
+void acpi_numa_its_affinity_init(struct acpi_srat_its_affinity *pa);
 #else
 static inline void
 acpi_numa_gicc_affinity_init(struct acpi_srat_gicc_affinity *pa) { }
+static inline void
+acpi_numa_its_affinity_init(struct acpi_srat_its_affinity *pa) { }
 #endif
 
 int acpi_numa_memory_affinity_init (struct acpi_srat_mem_affinity *ma);

-- 
1.8.1.4

Adding  Lorenzo and  Hanzun.

On Tue, Jun 6, 2017 at 6:26 PM, Ganapatrao Kulkarni
[off-list ref] wrote:

On Tue, Jun 06, 2017 at 06:26:27PM +0530, Ganapatrao Kulkarni wrote:

Add code to parse SRAT ITS Affinity sub table as defined in ACPI 6.2
Later in driver probe, its devices are mapped to numa node

Nit: Capitalize "its".

using its id to proximity domain mapping.

Ditto.

                  ^
Remove space.

+};
+
 static struct __node_cpu_hwid early_node_cpu_hwid[NR_CPUS] = {
 [0 ... NR_CPUS - 1] = {NUMA_NO_NODE, PHYS_CPUID_INVALID} };
 
+static struct __node_its_id early_node_its_id[MAX_NUMNODES] = {

Mind explaining to us where does MAX_NUMNODES as an array size limit
stem from here ?

I do not understand why maxcount == MAX_NUMNODES in the first place.

+		pr_warn_once("SRAT: its count exceeding max count[%d]\n",
+			     MAX_NUMNODES);
+		return;
+	}
+
+	pxm = pa->proximity_domain;
+	node = acpi_map_pxm_to_node(pxm);
+
+	if (node == NUMA_NO_NODE || node >= MAX_NUMNODES) {
+		pr_err("SRAT: Too many proximity domains %d\n", pxm);

Wrong error log message or at least inconsistent (yes, GICC code
where you copied this code from needs patching too).

+		bad_srat();
+		return;
+	}
+
+	early_node_its_id[its_in_srat].node_id = node;
+	early_node_its_id[its_in_srat].its_id =  its_in_srat;

                                               ^
One space is enough (again - that's through for GICC code too).

If that's the only usage I wonder whether we really need all arm64
arch code/data, instead of parsing the SRAT in ITS code driver straight
away at probe, retrieve its node and be done with this.

I understand you replicated what x86/GICC does with APIC code, I would
like to understand though if we see a reason why (or better, why we keep
the relevant stashed data in arch/arm64 instead of the ITS driver).

Thanks,
Lorenzo

Hi Lorenzo,

On Mon, Jun 12, 2017 at 4:23 PM, Lorenzo Pieralisi
[off-list ref] wrote:

On Tue, Jun 06, 2017 at 06:26:27PM +0530, Ganapatrao Kulkarni wrote:

quoted

Add code to parse SRAT ITS Affinity sub table as defined in ACPI 6.2
Later in driver probe, its devices are mapped to numa node

Nit: Capitalize "its".

ok.

quoted

using its id to proximity domain mapping.

Ditto.

quoted

Signed-off-by: Ganapatrao Kulkarni <redacted>
---
 arch/arm64/include/asm/acpi.h    |  2 ++
 arch/arm64/kernel/acpi_numa.c    | 59 ++++++++++++++++++++++++++++++++++++++++
 drivers/acpi/numa.c              | 32 +++++++++++++++++++++-
 drivers/irqchip/irq-gic-v3-its.c |  3 +-
 include/linux/acpi.h             |  3 ++
 5 files changed, 97 insertions(+), 2 deletions(-)

diff --git a/arch/arm64/include/asm/acpi.h b/arch/arm64/include/asm/acpi.h
index 0e99978..60ea7d1 100644
--- a/arch/arm64/include/asm/acpi.h
+++ b/arch/arm64/include/asm/acpi.h

@@ -143,9 +143,11 @@ static inline void arch_apei_flush_tlb_one(unsigned long addr)
 #ifdef CONFIG_ACPI_NUMA
 int arm64_acpi_numa_init(void);
 int acpi_numa_get_nid(unsigned int cpu, u64 hwid);
+int acpi_numa_get_its_nid(u32 its_id);
 #else
 static inline int arm64_acpi_numa_init(void) { return -ENOSYS; }
 static inline int acpi_numa_get_nid(unsigned int cpu, u64 hwid) { return NUMA_NO_NODE; }
+static inline int acpi_numa_get_its_nid(u32 its_id) { return NUMA_NO_NODE; }
 #endif /* CONFIG_ACPI_NUMA */

 #define ACPI_TABLE_UPGRADE_MAX_PHYS MEMBLOCK_ALLOC_ACCESSIBLE

diff --git a/arch/arm64/kernel/acpi_numa.c b/arch/arm64/kernel/acpi_numa.c
index f01fab6..f7b2ea6 100644
--- a/arch/arm64/kernel/acpi_numa.c
+++ b/arch/arm64/kernel/acpi_numa.c

@@ -29,15 +29,24 @@
 #include <asm/numa.h>

 static int cpus_in_srat;
+static int its_in_srat;

 struct __node_cpu_hwid {
      u32 node_id;    /* logical node containing this CPU */
      u64 cpu_hwid;   /* MPIDR for this CPU */
 };

+struct __node_its_id {
+     u32 node_id;    /* numa node id */
+     u32 its_id ;   /* GIC ITS ID */

                  ^
Remove space.

ok, thanks.

quoted

+};
+
 static struct __node_cpu_hwid early_node_cpu_hwid[NR_CPUS] = {
 [0 ... NR_CPUS - 1] = {NUMA_NO_NODE, PHYS_CPUID_INVALID} };

+static struct __node_its_id early_node_its_id[MAX_NUMNODES] = {

Mind explaining to us where does MAX_NUMNODES as an array size limit
stem from here ?

this is heuristic!
AFAIK, there is no limit on ITS devices per socket, however almost all
SoC designs has one ITS device per socket,
IMHO, assuming max ITS count to MAX_NUMNODES to init static array
should be reasonable i guess?

quoted

+[0 ... MAX_NUMNODES - 1] = {NUMA_NO_NODE, UINT_MAX} };
+
 int acpi_numa_get_nid(unsigned int cpu, u64 hwid)
 {
      int i;

@@ -50,6 +59,18 @@ int acpi_numa_get_nid(unsigned int cpu, u64 hwid)
      return NUMA_NO_NODE;
 }

+int acpi_numa_get_its_nid(u32 its_id)
+{
+     int i;
+
+     for (i = 0; i < its_in_srat; i++) {
+             if (its_id == early_node_its_id[i].its_id)
+                     return early_node_its_id[i].node_id;
+     }
+
+     return NUMA_NO_NODE;
+}
+
 /* Callback for Proximity Domain -> ACPI processor UID mapping */
 void __init acpi_numa_gicc_affinity_init(struct acpi_srat_gicc_affinity *pa)
 {

@@ -100,6 +121,44 @@ void __init acpi_numa_gicc_affinity_init(struct acpi_srat_gicc_affinity *pa)
              pxm, mpidr, node);
 }

+/* Callback for ITS ACPI Proximity Domain mapping */
+void __init acpi_numa_its_affinity_init(struct acpi_srat_its_affinity *pa)
+{
+     int pxm, node;
+
+     if (srat_disabled())
+             return;
+
+     if (pa->header.length < sizeof(struct acpi_srat_its_affinity)) {
+             pr_err("SRAT: Invalid SRAT header length: %d\n",
+                     pa->header.length);
+             bad_srat();
+             return;
+     }
+
+     if (its_in_srat >= MAX_NUMNODES) {

I do not understand why maxcount == MAX_NUMNODES in the first place.

quoted

+             pr_warn_once("SRAT: its count exceeding max count[%d]\n",
+                          MAX_NUMNODES);
+             return;
+     }
+
+     pxm = pa->proximity_domain;
+     node = acpi_map_pxm_to_node(pxm);
+
+     if (node == NUMA_NO_NODE || node >= MAX_NUMNODES) {
+             pr_err("SRAT: Too many proximity domains %d\n", pxm);

Wrong error log message or at least inconsistent (yes, GICC code
where you copied this code from needs patching too).

ok, i will update.

quoted

+             bad_srat();
+             return;
+     }
+
+     early_node_its_id[its_in_srat].node_id = node;
+     early_node_its_id[its_in_srat].its_id =  its_in_srat;

                                               ^
One space is enough (again - that's through for GICC code too).

ok, thanks

quoted

+     node_set(node, numa_nodes_parsed);
+     pr_info("SRAT: PXM %d -> ITS %d -> Node %d\n",
+             pxm, its_in_srat, node);
+     its_in_srat++;
+}
+
 int __init arm64_acpi_numa_init(void)
 {
      int ret;

diff --git a/drivers/acpi/numa.c b/drivers/acpi/numa.c
index edb0c79..9c04dba 100644
--- a/drivers/acpi/numa.c
+++ b/drivers/acpi/numa.c

@@ -182,6 +182,16 @@ int acpi_map_pxm_to_online_node(int pxm)
              }
              break;

+     case ACPI_SRAT_TYPE_GIC_ITS_AFFINITY:
+             {
+                     struct acpi_srat_its_affinity *p =
+                         (struct acpi_srat_its_affinity *)header;
+                     pr_debug("SRAT ITS [0x%04x]) in proximity domain %d\n",
+                              p->its_id,
+                              p->proximity_domain);
+             }
+             break;
+
      default:
              pr_warn("Found unsupported SRAT entry (type = 0x%x)\n",
                      header->type);

@@ -390,6 +400,24 @@ static int __init acpi_parse_slit(struct acpi_table_header *table)
      return 0;
 }

+static int __init
+acpi_parse_its_affinity(struct acpi_subtable_header *header,
+                      const unsigned long end)
+{
+     struct acpi_srat_its_affinity *its_affinity;
+
+     its_affinity = (struct acpi_srat_its_affinity *)header;
+     if (!its_affinity)
+             return -EINVAL;
+
+     acpi_table_print_srat_entry(header);
+
+     /* let architecture-dependent part to do it */
+     acpi_numa_its_affinity_init(its_affinity);
+
+     return 0;
+}
+
 static int __initdata parsed_numa_memblks;

 static int __init

@@ -445,7 +473,7 @@ int __init acpi_numa_init(void)

      /* SRAT: Static Resource Affinity Table */
      if (!acpi_table_parse(ACPI_SIG_SRAT, acpi_parse_srat)) {
-             struct acpi_subtable_proc srat_proc[3];
+             struct acpi_subtable_proc srat_proc[4];

              memset(srat_proc, 0, sizeof(srat_proc));
              srat_proc[0].id = ACPI_SRAT_TYPE_CPU_AFFINITY;

@@ -454,6 +482,8 @@ int __init acpi_numa_init(void)
              srat_proc[1].handler = acpi_parse_x2apic_affinity;
              srat_proc[2].id = ACPI_SRAT_TYPE_GICC_AFFINITY;
              srat_proc[2].handler = acpi_parse_gicc_affinity;
+             srat_proc[3].id = ACPI_SRAT_TYPE_GIC_ITS_AFFINITY;
+             srat_proc[3].handler = acpi_parse_its_affinity;

              acpi_table_parse_entries_array(ACPI_SIG_SRAT,
                                      sizeof(struct acpi_table_srat),

diff --git a/drivers/irqchip/irq-gic-v3-its.c b/drivers/irqchip/irq-gic-v3-its.c
index 45ea1933..84936da 100644
--- a/drivers/irqchip/irq-gic-v3-its.c
+++ b/drivers/irqchip/irq-gic-v3-its.c

@@ -1861,7 +1861,8 @@ static int __init gic_acpi_parse_madt_its(struct acpi_subtable_header *header,
              goto dom_err;
      }

-     err = its_probe_one(&res, dom_handle, NUMA_NO_NODE);
+     err = its_probe_one(&res, dom_handle,
+                     acpi_numa_get_its_nid(its_entry->translation_id));

If that's the only usage I wonder whether we really need all arm64
arch code/data, instead of parsing the SRAT in ITS code driver straight
away at probe, retrieve its node and be done with this.

I understand you replicated what x86/GICC does with APIC code, I would
like to understand though if we see a reason why (or better, why we keep
the relevant stashed data in arch/arm64 instead of the ITS driver).

it is been thought to do ITS sub table parse along with other SRAT tables.
and use the mapping later when ITS devices are initialised/probed.
IMO, it is more appropriate to keep all SRAT sub table parsing to same
function/place
rather than moving to driver.

Thanks,
Lorenzo

quoted

      if (!err)
              return 0;

diff --git a/include/linux/acpi.h b/include/linux/acpi.h
index 137e4a3..e5e8ae3 100644
--- a/include/linux/acpi.h
+++ b/include/linux/acpi.h

@@ -260,9 +260,12 @@ int acpi_table_parse_madt(enum acpi_madt_type id,

 #ifdef CONFIG_ARM64
 void acpi_numa_gicc_affinity_init(struct acpi_srat_gicc_affinity *pa);
+void acpi_numa_its_affinity_init(struct acpi_srat_its_affinity *pa);
 #else
 static inline void
 acpi_numa_gicc_affinity_init(struct acpi_srat_gicc_affinity *pa) { }
+static inline void
+acpi_numa_its_affinity_init(struct acpi_srat_its_affinity *pa) { }
 #endif

 int acpi_numa_memory_affinity_init (struct acpi_srat_mem_affinity *ma);
--

1.8.1.4

--
To unsubscribe from this list: send the line "unsubscribe linux-acpi" in
the body of a message to majordomo at vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

thanks
Ganapat

On Wed, Jun 14, 2017 at 03:23:17PM +0530, Ganapatrao Kulkarni wrote:

[...]

quoted

quoted

+static int __init
+acpi_parse_its_affinity(struct acpi_subtable_header *header,
+                      const unsigned long end)
+{
+     struct acpi_srat_its_affinity *its_affinity;
+
+     its_affinity = (struct acpi_srat_its_affinity *)header;
+     if (!its_affinity)
+             return -EINVAL;
+
+     acpi_table_print_srat_entry(header);

You can leave this info printing but see below.

quoted

quoted

+
+     /* let architecture-dependent part to do it */
+     acpi_numa_its_affinity_init(its_affinity);
+
+     return 0;
+}
+
 static int __initdata parsed_numa_memblks;

 static int __init

@@ -445,7 +473,7 @@ int __init acpi_numa_init(void)

      /* SRAT: Static Resource Affinity Table */
      if (!acpi_table_parse(ACPI_SIG_SRAT, acpi_parse_srat)) {
-             struct acpi_subtable_proc srat_proc[3];
+             struct acpi_subtable_proc srat_proc[4];

              memset(srat_proc, 0, sizeof(srat_proc));
              srat_proc[0].id = ACPI_SRAT_TYPE_CPU_AFFINITY;

@@ -454,6 +482,8 @@ int __init acpi_numa_init(void)
              srat_proc[1].handler = acpi_parse_x2apic_affinity;
              srat_proc[2].id = ACPI_SRAT_TYPE_GICC_AFFINITY;
              srat_proc[2].handler = acpi_parse_gicc_affinity;
+             srat_proc[3].id = ACPI_SRAT_TYPE_GIC_ITS_AFFINITY;
+             srat_proc[3].handler = acpi_parse_its_affinity;

              acpi_table_parse_entries_array(ACPI_SIG_SRAT,
                                      sizeof(struct acpi_table_srat),

diff --git a/drivers/irqchip/irq-gic-v3-its.c b/drivers/irqchip/irq-gic-v3-its.c
index 45ea1933..84936da 100644
--- a/drivers/irqchip/irq-gic-v3-its.c
+++ b/drivers/irqchip/irq-gic-v3-its.c

@@ -1861,7 +1861,8 @@ static int __init gic_acpi_parse_madt_its(struct acpi_subtable_header *header,
              goto dom_err;
      }

-     err = its_probe_one(&res, dom_handle, NUMA_NO_NODE);
+     err = its_probe_one(&res, dom_handle,
+                     acpi_numa_get_its_nid(its_entry->translation_id));

If that's the only usage I wonder whether we really need all arm64
arch code/data, instead of parsing the SRAT in ITS code driver straight
away at probe, retrieve its node and be done with this.

I understand you replicated what x86/GICC does with APIC code, I would
like to understand though if we see a reason why (or better, why we keep
the relevant stashed data in arch/arm64 instead of the ITS driver).

it is been thought to do ITS sub table parse along with other SRAT
tables.  and use the mapping later when ITS devices are
initialised/probed.  IMO, it is more appropriate to keep all SRAT sub
table parsing to same function/place rather than moving to driver.

I do not follow. If it is just used in ITS driver code to set the ITS
affinity node what's the point of stashing data and adding callbacks
when you can simply parse the SRAT and be done with it ?

Or you have something on top of these patches that require ITS node
information and the calls you added ? If so post the code please.

Regardless, it's ITS specific information, ITS is managed through an
irqchip driver on ARM64 so even if you decided to stash the SRAT ITS
information it does not belong in arch/arm64 IMO, you can implement
acpi_numa_its_affinity_init() in the ITS driver but AFAICS for the time
being it would be just useless that's the point I am making.

Thanks,
Lorenzo

On 15/06/17 11:35, Lorenzo Pieralisi wrote:

On Wed, Jun 14, 2017 at 03:23:17PM +0530, Ganapatrao Kulkarni wrote:

[...]

quoted

quoted

quoted

+static int __init
+acpi_parse_its_affinity(struct acpi_subtable_header *header,
+                      const unsigned long end)
+{
+     struct acpi_srat_its_affinity *its_affinity;
+
+     its_affinity = (struct acpi_srat_its_affinity *)header;
+     if (!its_affinity)
+             return -EINVAL;
+
+     acpi_table_print_srat_entry(header);

You can leave this info printing but see below.

quoted

quoted

quoted

+
+     /* let architecture-dependent part to do it */
+     acpi_numa_its_affinity_init(its_affinity);
+
+     return 0;
+}
+
 static int __initdata parsed_numa_memblks;

 static int __init

@@ -445,7 +473,7 @@ int __init acpi_numa_init(void)

      /* SRAT: Static Resource Affinity Table */
      if (!acpi_table_parse(ACPI_SIG_SRAT, acpi_parse_srat)) {
-             struct acpi_subtable_proc srat_proc[3];
+             struct acpi_subtable_proc srat_proc[4];

              memset(srat_proc, 0, sizeof(srat_proc));
              srat_proc[0].id = ACPI_SRAT_TYPE_CPU_AFFINITY;

@@ -454,6 +482,8 @@ int __init acpi_numa_init(void)
              srat_proc[1].handler = acpi_parse_x2apic_affinity;
              srat_proc[2].id = ACPI_SRAT_TYPE_GICC_AFFINITY;
              srat_proc[2].handler = acpi_parse_gicc_affinity;
+             srat_proc[3].id = ACPI_SRAT_TYPE_GIC_ITS_AFFINITY;
+             srat_proc[3].handler = acpi_parse_its_affinity;

              acpi_table_parse_entries_array(ACPI_SIG_SRAT,
                                      sizeof(struct acpi_table_srat),

diff --git a/drivers/irqchip/irq-gic-v3-its.c b/drivers/irqchip/irq-gic-v3-its.c
index 45ea1933..84936da 100644
--- a/drivers/irqchip/irq-gic-v3-its.c
+++ b/drivers/irqchip/irq-gic-v3-its.c

@@ -1861,7 +1861,8 @@ static int __init gic_acpi_parse_madt_its(struct acpi_subtable_header *header,
              goto dom_err;
      }

-     err = its_probe_one(&res, dom_handle, NUMA_NO_NODE);
+     err = its_probe_one(&res, dom_handle,
+                     acpi_numa_get_its_nid(its_entry->translation_id));

If that's the only usage I wonder whether we really need all arm64
arch code/data, instead of parsing the SRAT in ITS code driver straight
away at probe, retrieve its node and be done with this.

I understand you replicated what x86/GICC does with APIC code, I would
like to understand though if we see a reason why (or better, why we keep
the relevant stashed data in arch/arm64 instead of the ITS driver).

it is been thought to do ITS sub table parse along with other SRAT
tables.  and use the mapping later when ITS devices are
initialised/probed.  IMO, it is more appropriate to keep all SRAT sub
table parsing to same function/place rather than moving to driver.

I do not follow. If it is just used in ITS driver code to set the ITS
affinity node what's the point of stashing data and adding callbacks
when you can simply parse the SRAT and be done with it ?

Or you have something on top of these patches that require ITS node
information and the calls you added ? If so post the code please.

Regardless, it's ITS specific information, ITS is managed through an
irqchip driver on ARM64 so even if you decided to stash the SRAT ITS
information it does not belong in arch/arm64 IMO, you can implement
acpi_numa_its_affinity_init() in the ITS driver but AFAICS for the time
being it would be just useless that's the point I am making.

Agreed. As long as there is only the ITS as a consumer of that
information, there is no need to pollute the rest of the kernel with it.
Once we have another consumer, we can look at making that code common.
In the meantime, keeping it in the ITS code is the right thing to do.

Thanks,

	M.
-- 
Jazz is not dead. It just smells funny...

On Thu, Jun 15, 2017 at 4:13 PM, Marc Zyngier [off-list ref] wrote:

On 15/06/17 11:35, Lorenzo Pieralisi wrote:

quoted

On Wed, Jun 14, 2017 at 03:23:17PM +0530, Ganapatrao Kulkarni wrote:

[...]

quoted

quoted

quoted

+static int __init
+acpi_parse_its_affinity(struct acpi_subtable_header *header,
+                      const unsigned long end)
+{
+     struct acpi_srat_its_affinity *its_affinity;
+
+     its_affinity = (struct acpi_srat_its_affinity *)header;
+     if (!its_affinity)
+             return -EINVAL;
+
+     acpi_table_print_srat_entry(header);

You can leave this info printing but see below.

ok.

quoted

quoted

quoted

quoted

+
+     /* let architecture-dependent part to do it */
+     acpi_numa_its_affinity_init(its_affinity);
+
+     return 0;
+}
+
 static int __initdata parsed_numa_memblks;

 static int __init

@@ -445,7 +473,7 @@ int __init acpi_numa_init(void)

      /* SRAT: Static Resource Affinity Table */
      if (!acpi_table_parse(ACPI_SIG_SRAT, acpi_parse_srat)) {
-             struct acpi_subtable_proc srat_proc[3];
+             struct acpi_subtable_proc srat_proc[4];

              memset(srat_proc, 0, sizeof(srat_proc));
              srat_proc[0].id = ACPI_SRAT_TYPE_CPU_AFFINITY;

@@ -454,6 +482,8 @@ int __init acpi_numa_init(void)
              srat_proc[1].handler = acpi_parse_x2apic_affinity;
              srat_proc[2].id = ACPI_SRAT_TYPE_GICC_AFFINITY;
              srat_proc[2].handler = acpi_parse_gicc_affinity;
+             srat_proc[3].id = ACPI_SRAT_TYPE_GIC_ITS_AFFINITY;
+             srat_proc[3].handler = acpi_parse_its_affinity;

              acpi_table_parse_entries_array(ACPI_SIG_SRAT,
                                      sizeof(struct acpi_table_srat),

diff --git a/drivers/irqchip/irq-gic-v3-its.c b/drivers/irqchip/irq-gic-v3-its.c
index 45ea1933..84936da 100644
--- a/drivers/irqchip/irq-gic-v3-its.c
+++ b/drivers/irqchip/irq-gic-v3-its.c

@@ -1861,7 +1861,8 @@ static int __init gic_acpi_parse_madt_its(struct acpi_subtable_header *header,
              goto dom_err;
      }

-     err = its_probe_one(&res, dom_handle, NUMA_NO_NODE);
+     err = its_probe_one(&res, dom_handle,
+                     acpi_numa_get_its_nid(its_entry->translation_id));

If that's the only usage I wonder whether we really need all arm64
arch code/data, instead of parsing the SRAT in ITS code driver straight
away at probe, retrieve its node and be done with this.

I understand you replicated what x86/GICC does with APIC code, I would
like to understand though if we see a reason why (or better, why we keep
the relevant stashed data in arch/arm64 instead of the ITS driver).

it is been thought to do ITS sub table parse along with other SRAT
tables.  and use the mapping later when ITS devices are
initialised/probed.  IMO, it is more appropriate to keep all SRAT sub
table parsing to same function/place rather than moving to driver.

I do not follow. If it is just used in ITS driver code to set the ITS
affinity node what's the point of stashing data and adding callbacks
when you can simply parse the SRAT and be done with it ?

Or you have something on top of these patches that require ITS node
information and the calls you added ? If so post the code please.

Regardless, it's ITS specific information, ITS is managed through an
irqchip driver on ARM64 so even if you decided to stash the SRAT ITS
information it does not belong in arch/arm64 IMO, you can implement
acpi_numa_its_affinity_init() in the ITS driver but AFAICS for the time
being it would be just useless that's the point I am making.

Agreed. As long as there is only the ITS as a consumer of that
information, there is no need to pollute the rest of the kernel with it.
Once we have another consumer, we can look at making that code common.
In the meantime, keeping it in the ITS code is the right thing to do.

agreed, i will change it in next version.

Thanks,

        M.
--
Jazz is not dead. It just smells funny...

thanks
Ganapat