Re: [PATCH 13/36] next-cube: move ESCC to be QOM child of next-pc device

[Mark Cave-Ayland]

On 28/10/2024 16:39, Thomas Huth wrote:
Am Wed, 23 Oct 2024 09:58:29 +0100
schrieb Mark Cave-Ayland <mark.cave-ayl...@ilande.co.uk>:
Since the ESCC is part of the next-pc device, move the ESCC to be a QOM child
of the next-pc device.

Signed-off-by: Mark Cave-Ayland <mark.cave-ayl...@ilande.co.uk>
---
  hw/m68k/next-cube.c | 54 ++++++++++++++++++++++-----------------------
  1 file changed, 26 insertions(+), 28 deletions(-)

diff --git a/hw/m68k/next-cube.c b/hw/m68k/next-cube.c
index 7f714640da..915dd80f6f 100644
--- a/hw/m68k/next-cube.c
+++ b/hw/m68k/next-cube.c
@@ -124,6 +124,8 @@ struct NeXTPC {
      qemu_irq scsi_reset;
      qemu_irq scsi_dma;
  
+    ESCCState escc;
+
      NextRtc rtc;
  };
  
@@ -978,31 +980,6 @@ static const MemoryRegionOps next_floppy_ops = {
      .endianness = DEVICE_BIG_ENDIAN,
  };
  
-static void next_escc_init(DeviceState *pcdev)
-{
-    NeXTPC *next_pc = NEXT_PC(pcdev);
-    DeviceState *dev;
-    SysBusDevice *s;
-
-    dev = qdev_new(TYPE_ESCC);
-    qdev_prop_set_uint32(dev, "disabled", 0);
-    qdev_prop_set_uint32(dev, "frequency", 9600 * 384);
-    qdev_prop_set_uint32(dev, "it_shift", 0);
-    qdev_prop_set_bit(dev, "bit_swap", true);
-    qdev_prop_set_chr(dev, "chrB", serial_hd(1));
-    qdev_prop_set_chr(dev, "chrA", serial_hd(0));
-    qdev_prop_set_uint32(dev, "chnBtype", escc_serial);
-    qdev_prop_set_uint32(dev, "chnAtype", escc_serial);
-
-    s = SYS_BUS_DEVICE(dev);
-    sysbus_realize_and_unref(s, &error_fatal);
-    sysbus_connect_irq(s, 0, qdev_get_gpio_in(pcdev, NEXT_SCC_I));
-    sysbus_connect_irq(s, 1, qdev_get_gpio_in(pcdev, NEXT_SCC_DMA_I));
-
-    memory_region_add_subregion(&next_pc->scrmem, 0x18000,
-                                sysbus_mmio_get_region(s, 0));
-}
-
  static void next_pc_reset(DeviceState *dev)
  {
      NeXTPC *s = NEXT_PC(dev);
@@ -1043,6 +1020,28 @@ static void next_pc_realize(DeviceState *dev, Error 
**errp)
      /* Floppy */
      memory_region_add_subregion(&s->scrmem, 0x14108,
                                  &s->floppy_mem);
+
+    /* ESCC */
+    d = DEVICE(object_resolve_path_component(OBJECT(dev), "escc"));
+    qdev_prop_set_uint32(d, "disabled", 0);
+    qdev_prop_set_uint32(d, "frequency", 9600 * 384);
+    qdev_prop_set_uint32(d, "it_shift", 0);
+    qdev_prop_set_bit(d, "bit_swap", true);
+    qdev_prop_set_chr(d, "chrB", serial_hd(1));
+    qdev_prop_set_chr(d, "chrA", serial_hd(0));
+    qdev_prop_set_uint32(d, "chnBtype", escc_serial);
+    qdev_prop_set_uint32(d, "chnAtype", escc_serial);
+
+    sbd = SYS_BUS_DEVICE(d);
+    if (!sysbus_realize(sbd, errp)) {
+        return;
+    }
+    sysbus_connect_irq(sbd, 0, qdev_get_gpio_in(dev, NEXT_SCC_I));
+    sysbus_connect_irq(sbd, 1, qdev_get_gpio_in(dev, NEXT_SCC_DMA_I));
+
+    memory_region_add_subregion(&s->scrmem, 0x18000,
+                                sysbus_mmio_get_region(sbd, 0));

You could also keep the next_escc_init() function, and call next_escc_init()
here?

Normally a non-QOM _init() function suffix is used to both init() and realize() a 
device, whereas here since the ESCC device is a child of the next-pc device these 
operations must be separate. I think I can see why this convention is used elsewhere 
in the codebase, as otherwise you end up calling a function with a _init() prefix 
from _realize() which can get confusing with respect to the QOM model...


ATB,

Mark.

  }
  
  static void next_pc_init(Object *obj)
@@ -1064,6 +1063,8 @@ static void next_pc_init(Object *obj)
  
      memory_region_init_io(&s->floppy_mem, OBJECT(s), &next_floppy_ops, s,
                            "next.floppy", 4);
+
+    object_initialize_child(obj, "escc", &s->escc, TYPE_ESCC);
  }
  
  /*
@@ -1201,9 +1202,6 @@ static void next_cube_init(MachineState *machine)
          }
      }
  
-    /* Serial */
-    next_escc_init(pcdev);
-
      /* TODO: */
      /* Network */