[Qemu-devel] [RFC PATCH v3 15/19] Add _OST dimm support

[Vasilis Liaskovitis <vasilis.liaskovitis@profitbricks.com>]

This allows qemu to receive notifications from the guest OS on success or
failure of a memory hotplug request. The guest OS needs to implement the _OST
functionality for this to work (linux-next: http://lkml.org/lkml/2012/6/25/321)

This patch also updates dimm bitmap state and hot-remove pending flag
on hot-remove fail.  This allows failed hot operations to be retried at
anytime. This only works for guests that use _OST notification.
Also adds new _OST registers in  docs/specs/acpi_hotplug.txt

Signed-off-by: Vasilis Liaskovitis <vasilis.liaskovitis@profitbricks.com>
---
 docs/specs/acpi_hotplug.txt |   25 +++++++++++++++++++++++++
 hw/acpi_piix4.c             |   35 ++++++++++++++++++++++++++++++++++-
 hw/dimm.c                   |   28 +++++++++++++++++++++++++++-
 hw/dimm.h                   |   10 +++++++++-
 4 files changed, 95 insertions(+), 3 deletions(-)

diff --git a/docs/specs/acpi_hotplug.txt b/docs/specs/acpi_hotplug.txt
index cf86242..536da16 100644
--- a/docs/specs/acpi_hotplug.txt
+++ b/docs/specs/acpi_hotplug.txt
@@ -20,3 +20,28 @@ ejected.
 
 Written by ACPI memory device _EJ0 method to notify qemu of successfull
 hot-removal.  Write-only.
+
+Memory Dimm ejection failure notification (IO port 0xafa1, 1-byte access):
+---------------------------------------------------------------
+Dimm hot-remove _OST notification. Byte value indicates Dimm slot for which
+ejection failed.
+
+Written by ACPI memory device _OST method to notify qemu of failed
+hot-removal.  Write-only.
+
+Memory Dimm insertion success notification (IO port 0xafa2, 1-byte access):
+---------------------------------------------------------------
+Dimm hot-remove _OST notification. Byte value indicates Dimm slot for which
+insertion succeeded.
+
+Written by ACPI memory device _OST method to notify qemu of failed
+hot-add.  Write-only.
+
+Memory Dimm insertion failure notification (IO port 0xafa3, 1-byte access):
+---------------------------------------------------------------
+Dimm hot-remove _OST notification. Byte value indicates Dimm slot for which
+insertion failed.
+
+Written by ACPI memory device _OST method to notify qemu of failed
+hot-add.  Write-only.
+
diff --git a/hw/acpi_piix4.c b/hw/acpi_piix4.c
index 8776669..f7220d4 100644
--- a/hw/acpi_piix4.c
+++ b/hw/acpi_piix4.c
@@ -49,6 +49,9 @@
 #define PCI_RMV_BASE 0xae0c
 #define MEM_BASE 0xaf80
 #define MEM_EJ_BASE 0xafa0
+#define MEM_OST_REMOVE_FAIL 0xafa1
+#define MEM_OST_ADD_SUCCESS 0xafa2
+#define MEM_OST_ADD_FAIL 0xafa3
 
 #define PIIX4_MEM_HOTPLUG_STATUS 8
 #define PIIX4_PCI_HOTPLUG_STATUS 2
@@ -87,6 +90,7 @@ typedef struct PIIX4PMState {
     uint8_t s4_val;
 } PIIX4PMState;
 
+static int piix4_dimm_revert(DeviceState *qdev, DimmDevice *dev, int add);
 static void piix4_acpi_system_hot_add_init(PCIBus *bus, PIIX4PMState *s);
 
 #define ACPI_ENABLE 0xf1
@@ -531,6 +535,15 @@ static void gpe_writeb(void *opaque, uint32_t addr, uint32_t val)
         case MEM_EJ_BASE:
             dimm_notify(val, DIMM_REMOVE_SUCCESS);
             break;
+        case MEM_OST_REMOVE_FAIL:
+            dimm_notify(val, DIMM_REMOVE_FAIL);
+            break;
+        case MEM_OST_ADD_SUCCESS:
+            dimm_notify(val, DIMM_ADD_SUCCESS);
+            break;
+        case MEM_OST_ADD_FAIL:
+            dimm_notify(val, DIMM_ADD_FAIL);
+            break;
         default:
             acpi_gpe_ioport_writeb(&s->ar, addr, val);
     }
@@ -604,13 +617,16 @@ static void piix4_acpi_system_hot_add_init(PCIBus *bus, PIIX4PMState *s)
 
     register_ioport_read(MEM_BASE, DIMM_BITMAP_BYTES, 1,  gpe_readb, s);
     register_ioport_write(MEM_EJ_BASE, 1, 1,  gpe_writeb, s);
+    register_ioport_write(MEM_OST_REMOVE_FAIL, 1, 1,  gpe_writeb, s);
+    register_ioport_write(MEM_OST_ADD_SUCCESS, 1, 1,  gpe_writeb, s);
+    register_ioport_write(MEM_OST_ADD_FAIL, 1, 1,  gpe_writeb, s);
 
     for(i = 0; i < DIMM_BITMAP_BYTES; i++) {
         s->gperegs.mems_sts[i] = 0;
     }
 
     pci_bus_hotplug(bus, piix4_device_hotplug, &s->dev.qdev);
-    dimm_bus_hotplug(piix4_dimm_hotplug, &s->dev.qdev);
+    dimm_bus_hotplug(piix4_dimm_hotplug, piix4_dimm_revert, &s->dev.qdev);
 }
 
 static void enable_device(PIIX4PMState *s, int slot)
@@ -656,6 +672,23 @@ static int piix4_dimm_hotplug(DeviceState *qdev, DimmDevice *dev, int
     return 0;
 }
 
+static int piix4_dimm_revert(DeviceState *qdev, DimmDevice *dev, int add)
+{
+    PCIDevice *pci_dev = DO_UPCAST(PCIDevice, qdev, qdev);
+    PIIX4PMState *s = DO_UPCAST(PIIX4PMState, dev, pci_dev);
+    struct gpe_regs *g = &s->gperegs;
+    DimmDevice *slot = DIMM(dev);
+    int idx = slot->idx;
+
+    if (add) {
+        g->mems_sts[idx/8] &= ~(1 << (idx%8));
+    }
+    else {
+        g->mems_sts[idx/8] |= (1 << (idx%8));
+    }
+    return 0;
+}
+
 static int piix4_device_hotplug(DeviceState *qdev, PCIDevice *dev,
 				PCIHotplugState state)
 {
diff --git a/hw/dimm.c b/hw/dimm.c
index 21626f6..1521462 100644
--- a/hw/dimm.c
+++ b/hw/dimm.c
@@ -126,12 +126,14 @@ void dimm_config_create(char *id, uint64_t size, uint64_t node, uint32_t
     QTAILQ_INSERT_TAIL(&dimmconfig_list, dimm_cfg, nextdimmcfg);
 }
 
-void dimm_bus_hotplug(dimm_hotplug_fn hotplug, DeviceState *qdev)
+void dimm_bus_hotplug(dimm_hotplug_fn hotplug, dimm_hotplug_fn revert,
+        DeviceState *qdev)
 {
     DimmBus *bus = main_memory_bus;
     bus->qbus.allow_hotplug = 1;
     bus->dimm_hotplug_qdev = qdev;
     bus->dimm_hotplug = hotplug;
+    bus->dimm_revert = revert;
 }
 
 static void dimm_plug_device(DimmDevice *slot)
@@ -141,6 +143,7 @@ static void dimm_plug_device(DimmDevice *slot)
     dimm_populate(slot);
     if (bus->dimm_hotplug)
         bus->dimm_hotplug(bus->dimm_hotplug_qdev, slot, 1);
+    slot->pending = DIMM_ADD_PENDING;
 }
 
 static int dimm_unplug_device(DeviceState *qdev)
@@ -149,6 +152,7 @@ static int dimm_unplug_device(DeviceState *qdev)
 
     if (bus->dimm_hotplug)
         bus->dimm_hotplug(bus->dimm_hotplug_qdev, DIMM(qdev), 0);
+    DIMM(qdev)->pending = DIMM_REMOVE_PENDING;
     return 1;
 }
 
@@ -266,12 +270,33 @@ void dimm_notify(uint32_t idx, uint32_t event)
     result = g_malloc0(sizeof(*result));
     slotcfg = dimmcfg_find_from_name(DEVICE(s)->id);
     result->dimmname = slotcfg->name;
+    result->ret = event;
 
     switch(event) {
         case DIMM_REMOVE_SUCCESS:
             dimm_depopulate(s);
             QTAILQ_REMOVE(&bus->dimmlist, s, nextdimm);
             qdev_simple_unplug_cb((DeviceState*)s);
+            s->pending = DIMM_NO_PENDING;
+            QTAILQ_INSERT_TAIL(&bus->dimm_hp_result_queue, result, next);
+            break;
+        case DIMM_REMOVE_FAIL:
+            s->pending = DIMM_NO_PENDING;
+            if (bus->dimm_revert)
+                bus->dimm_revert(bus->dimm_hotplug_qdev, s, 0);
+            QTAILQ_INSERT_TAIL(&bus->dimm_hp_result_queue, result, next);
+            break;
+        case DIMM_ADD_SUCCESS:
+            s->pending = DIMM_NO_PENDING;
+            QTAILQ_INSERT_TAIL(&bus->dimm_hp_result_queue, result, next);
+            break;
+        case DIMM_ADD_FAIL:
+            dimm_depopulate(s);
+            s->pending = DIMM_NO_PENDING;
+            if (bus->dimm_revert)
+                bus->dimm_revert(bus->dimm_hotplug_qdev, s, 1);
+            QTAILQ_REMOVE(&bus->dimmlist, s, nextdimm);
+            qdev_simple_unplug_cb((DeviceState*)s);
             QTAILQ_INSERT_TAIL(&bus->dimm_hp_result_queue, result, next);
             break;
         default:
@@ -352,6 +377,7 @@ static int dimm_init(DeviceState *s)
     slot->start = slotcfg->start;
     slot->size = slotcfg->size;
     slot->node = slotcfg->node;
+    slot->pending = DIMM_NO_PENDING;
 
     QTAILQ_INSERT_TAIL(&bus->dimmlist, slot, nextdimm);
     dimm_plug_device(slot);
diff --git a/hw/dimm.h b/hw/dimm.h
index 21225be..4f696d8 100644
--- a/hw/dimm.h
+++ b/hw/dimm.h
@@ -18,6 +18,12 @@ typedef enum {
     DIMM_ADD_FAIL = 3
 } dimm_hp_result_code;
 
+typedef enum {
+    DIMM_NO_PENDING = 0,
+    DIMM_ADD_PENDING = 1,
+    DIMM_REMOVE_PENDING = 2,
+} dimm_hp_pending_code;
+
 #define TYPE_DIMM "dimm"
 #define DIMM(obj) \
     OBJECT_CHECK(DimmDevice, (obj), TYPE_DIMM)
@@ -42,6 +48,7 @@ typedef struct DimmDevice {
     ram_addr_t size;
     uint32_t node; /* numa node proximity */
     MemoryRegion *mr; /* MemoryRegion for this slot. !NULL only if populated */
+    dimm_hp_pending_code pending; /* indicates if a hot operation is pending for this dimm */
     QTAILQ_ENTRY (DimmDevice) nextdimm;
 } DimmDevice;
 
@@ -66,6 +73,7 @@ typedef struct DimmBus {
     BusState qbus;
     DeviceState *dimm_hotplug_qdev;
     dimm_hotplug_fn dimm_hotplug;
+    dimm_hotplug_fn dimm_revert;
     dimm_calcoffset_fn dimm_calcoffset;
     DimmConfiglist dimmconfig_list;
     QTAILQ_HEAD(Dimmlist, DimmDevice) dimmlist;
@@ -80,7 +88,7 @@ struct dimm_hp_result {
 
 void dimm_calc_offsets(dimm_calcoffset_fn calcfn);
 void dimm_notify(uint32_t idx, uint32_t event);
-void dimm_bus_hotplug(dimm_hotplug_fn hotplug, DeviceState *qdev);
+void dimm_bus_hotplug(dimm_hotplug_fn hotplug, dimm_hotplug_fn revert, DeviceState *qdev);
 void setup_fwcfg_hp_dimms(uint64_t *fw_cfg_slots);
 int dimm_add(char *id);
 void main_memory_bus_create(Object *parent);
-- 
1.7.9