On 8/8/24 23:31, Octavian Purdila wrote:
+static Property flexspi_properties[] = {
+ DEFINE_PROP_UINT32("mmap_base", FlexSpiState, mmap_base, 0),
+ DEFINE_PROP_UINT32("mmap_size", FlexSpiState, mmap_size, 0),
Preferably simply 'size'.
+ DEFINE_PROP_END_OF_LIST(),
+};
+
+static void flexspi_init(Object *obj)
+{
+ FlexSpiState *s = FLEXSPI(obj);
+
+ memory_region_init_io(&s->mmio, obj, &flexspi_ops, s, TYPE_FLEXSPI,
+ sizeof(FLEXSPI_Type));
+ sysbus_init_mmio(SYS_BUS_DEVICE(obj), &s->mmio);
+}
+
+static void flexspi_realize(DeviceState *dev, Error **errp)
+{
+ FlexSpiState *s = FLEXSPI(dev);
+
+ if (s->mmap_size) {
+ memory_region_init_ram(&s->mem, OBJECT(s), DEVICE(s)->id, s->mmap_size,
+ NULL);
+ memory_region_add_subregion(get_system_memory(), s->mmap_base,
&s->mem);
Where is this region used?
These regions are enabled in rt500.c when instantiating the flexspi
peripherals. As implemented now they are backed by RAM, the full
implementation should translate accesses to spi commands to FLASH or
PSRAM devices.
We need the memory regions because even the simplest NXP SDK examples
are using the memory mapped flexspi0 region.
Devices shouldn't access get_system_memory() directly (it is
documented as kind of deprecated =) ). Since you implement a
sysbus device, you need to export the region with sysbus_init_mmio()
then the upper layer (SoC) instantiating it gets the regions calling
sysbus_mmio_get_region() and maps it. Then you don't need the
'mmap_base' property.
