[Qemu-devel] [RFC 35/42] acpi: cpuhp: provide cpu._PXM method if running in numa mode

[Igor Mammedov]

Having proximity in SRAT table is not enough in case of
linux guest as it discards SRAT table information after
it's done with booting. So hotplugged CPUs will
go to 1st node as kernel doesn't know to what node
they belong to anymore.
So do the same as with hotplugged memory and provide
_PXM method in CPU device context so that OSPM could
assign CPU to a correct node.

Signed-off-by: Igor Mammedov <address@hidden>
---
 hw/acpi/cpu.c | 48 ++++++++++++++++++++++++++++++++++++++++++++++++
 trace-events  |  1 +
 2 files changed, 49 insertions(+)

diff --git a/hw/acpi/cpu.c b/hw/acpi/cpu.c
index a238ced..a2a03b2 100644
--- a/hw/acpi/cpu.c
+++ b/hw/acpi/cpu.c
@@ -11,6 +11,7 @@
 #define ACPI_CPU_CMD_DATA_OFFSET_RW 8
 
 enum {
+    CPHP_PXM_CMD = 0,
     CPHP_CMD_MAX
 };
 
@@ -35,6 +36,12 @@ static uint64_t cpu_hotplug_rd(void *opaque, hwaddr addr, 
unsigned size)
         break;
     case ACPI_CPU_CMD_DATA_OFFSET_RW:
         switch (cpu_st->command) {
+        case CPHP_PXM_CMD: {
+           Object *o = OBJECT(CPU(cdev->cpu));
+           val = o ? object_property_get_int(o, "node", NULL) : ~0;
+           trace_cpuhp_acpi_read_pxm(cpu_st->selector, val);
+           break;
+        }
         default:
            break;
         }
@@ -243,6 +250,7 @@ const VMStateDescription vmstate_cpu_hotplug = {
 #define CPU_SCAN_METHOD   "CSCN"
 #define CPU_EJECT_METHOD  "CEJ0"
 #define CPU_NOTIFY_METHOD "CTFY"
+#define CPU_PXM_METHOD    "CPXM"
 
 #define CPU_ENABLED       "CPEN"
 #define CPU_SELECTOR      "CSEL"
@@ -321,6 +329,8 @@ void build_cpus_aml(Aml *table, MachineState *machine, bool 
acpi1_compat,
     cpus_dev = aml_device("\\_SB.CPUS");
     {
         int i;
+        bool has_pxm = true;
+        Error *local_err = NULL;
         Aml *one = aml_int(1);
         Aml *cpu_selector = aml_name("%s.%s", cphp_res_path, CPU_SELECTOR);
         Aml *ins_evt = aml_name("%s.%s", cphp_res_path, CPU_INSERT_EVENT);
@@ -328,11 +338,25 @@ void build_cpus_aml(Aml *table, MachineState *machine, 
bool acpi1_compat,
         Aml *ej_evt = aml_name("%s.%s", cphp_res_path, CPU_EJECT_EVENT);
         Aml *is_enabled = aml_name("%s.%s", cphp_res_path, CPU_ENABLED);
         Aml *ctrl_lock = aml_name("%s.%s", cphp_res_path, CPU_LOCK);
+        Aml *cpu_cmd = aml_name("%s.%s", cphp_res_path, CPU_COMMAND);
+        Aml *cpu_data = aml_name("%s.%s", cphp_res_path, CPU_DATA);
 
         aml_append(cpus_dev, aml_name_decl("_HID", aml_string("ACPI0010")));
         aml_append(cpus_dev, aml_name_decl("_CID", aml_eisaid("PNP0A05")));
 
         if (machine->cpu_hotplug) {
+            local_err = NULL;
+            /*
+             * sample 1st VCPU (BSP) if it has support for numa and use result
+             * to enable _PXM method support
+             */
+            object_property_get_int(OBJECT(arch_ids->cpus[0].cpu), "node",
+                                    &local_err);
+            if (local_err) {
+                error_free(local_err);
+                has_pxm = false;
+            }
+
             method = aml_method(CPU_NOTIFY_METHOD, 2, AML_NOTSERIALIZED);
             for (i = 0; i < arch_ids->len; i++) {
                 Aml *cpu = aml_name(CPU_NAME_FMT, i);
@@ -413,6 +437,23 @@ void build_cpus_aml(Aml *table, MachineState *machine, 
bool acpi1_compat,
                 aml_append(method, aml_release(ctrl_lock));
             }
             aml_append(cpus_dev, method);
+
+            if (has_pxm) {
+                method = aml_method(CPU_PXM_METHOD, 1, AML_SERIALIZED);
+                {
+                    Aml *uid = aml_arg(0);
+                    Aml *pxm = aml_local(0);
+                    Aml *pxm_cmd = aml_int(CPHP_PXM_CMD);
+
+                    aml_append(method, aml_acquire(ctrl_lock, 0xFFFF));
+                    aml_append(method, aml_store(uid, cpu_selector));
+                    aml_append(method, aml_store(pxm_cmd, cpu_cmd));
+                    aml_append(method, aml_store(cpu_data, pxm));
+                    aml_append(method, aml_release(ctrl_lock));
+                    aml_append(method, aml_return(pxm));
+                }
+                aml_append(cpus_dev, method);
+            }
         }
 
         /* build Processor object for each processor */
@@ -461,6 +502,13 @@ void build_cpus_aml(Aml *table, MachineState *machine, 
bool acpi1_compat,
                 method = aml_method("_EJ0", 1, AML_NOTSERIALIZED);
                 aml_append(method, aml_call1(CPU_EJECT_METHOD, uid));
                 aml_append(dev, method);
+
+                if (has_pxm) {
+                    method = aml_method("_PXM", 0, AML_SERIALIZED);
+                    aml_append(method,
+                        aml_return(aml_call1(CPU_PXM_METHOD, uid)));
+                    aml_append(dev, method);
+                }
             }
             aml_append(cpus_dev, dev);
         }
diff --git a/trace-events b/trace-events
index 108ae07..159c94d 100644
--- a/trace-events
+++ b/trace-events
@@ -1913,6 +1913,7 @@ aspeed_vic_write(uint64_t offset, unsigned size, uint32_t 
data) "To 0x%" PRIx64
 # hw/acpi/cpu.c
 cpuhp_acpi_invalid_idx_selected(uint32_t idx) "0x%"PRIx32
 cpuhp_acpi_read_flags(uint32_t idx, uint8_t flags) "idx[0x%"PRIx32"] flags: 
0x%"PRIx8
+cpuhp_acpi_read_pxm(uint32_t idx, uint32_t pxm) "idx[0x%"PRIx32"] flags: 
0x%"PRIx32
 cpuhp_acpi_clear_inserting_evt(uint32_t idx) "idx[0x%"PRIx32"]"
 cpuhp_acpi_clear_remove_evt(uint32_t idx) "idx[0x%"PRIx32"]"
 cpuhp_acpi_write_idx(uint32_t idx) "set active cpu idx: 0x%"PRIx32
-- 
1.8.3.1