[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
[Qemu-devel] [PATCH 31/39] pcie: do not recreate mmcfg I/O region, use a
From: |
Paolo Bonzini |
Subject: |
[Qemu-devel] [PATCH 31/39] pcie: do not recreate mmcfg I/O region, use an alias instead |
Date: |
Tue, 4 Jun 2013 20:52:26 +0200 |
The MMIO regions should not be destroyed and recreated on the fly,
because they might have I/O in progress. Instead, create an area that
is as large as possible, and then map a window of it into system memory.
This is possible because containers and aliases are never the destination
of I/O and can be safely deleted at any time. It is also similar to
how PAM maps windows of system RAM.
Signed-off-by: Paolo Bonzini <address@hidden>
---
hw/pci/pcie_host.c | 22 +++++++++++++++++-----
include/hw/pci/pcie_host.h | 1 +
2 files changed, 18 insertions(+), 5 deletions(-)
diff --git a/hw/pci/pcie_host.c b/hw/pci/pcie_host.c
index b2d942b..43aeeba 100644
--- a/hw/pci/pcie_host.c
+++ b/hw/pci/pcie_host.c
@@ -110,19 +110,26 @@ static const MemoryRegionOps pcie_mmcfg_ops = {
int pcie_host_init(PCIExpressHost *e)
{
e->base_addr = PCIE_BASE_ADDR_UNMAPPED;
-
+ memory_region_init_io(&e->mmio, &pcie_mmcfg_ops, e, "pcie-mmcfg",
+ PCIE_MMCFG_SIZE_MAX);
return 0;
}
void pcie_host_mmcfg_unmap(PCIExpressHost *e)
{
if (e->base_addr != PCIE_BASE_ADDR_UNMAPPED) {
- memory_region_del_subregion(get_system_memory(), &e->mmio);
- memory_region_destroy(&e->mmio);
+ memory_region_del_subregion(get_system_memory(), &e->mapped_mmio);
+ memory_region_destroy(&e->mapped_mmio);
e->base_addr = PCIE_BASE_ADDR_UNMAPPED;
}
}
+static void pcie_host_instance_finalize(Object *obj)
+{
+ PCIExpressHost *e = PCIE_HOST_BRIDGE(obj);
+ memory_region_destroy(&e->mmio);
+}
+
void pcie_host_mmcfg_map(PCIExpressHost *e, hwaddr addr,
uint32_t size)
{
@@ -130,9 +137,13 @@ void pcie_host_mmcfg_map(PCIExpressHost *e, hwaddr addr,
assert(size >= PCIE_MMCFG_SIZE_MIN);
assert(size <= PCIE_MMCFG_SIZE_MAX);
e->size = size;
- memory_region_init_io(&e->mmio, &pcie_mmcfg_ops, e, "pcie-mmcfg", e->size);
e->base_addr = addr;
- memory_region_add_subregion(get_system_memory(), e->base_addr, &e->mmio);
+
+ memory_region_init_alias(&e->mapped_mmio,
+ "pci_bridge_vga_io_lo", &e->mmio,
+ 0, e->size);
+
+ memory_region_add_subregion(get_system_memory(), e->base_addr,
&e->mapped_mmio);
}
void pcie_host_mmcfg_update(PCIExpressHost *e,
@@ -151,6 +162,7 @@ static const TypeInfo pcie_host_type_info = {
.parent = TYPE_PCI_HOST_BRIDGE,
.abstract = true,
.instance_size = sizeof(PCIExpressHost),
+ .instance_finalize = pcie_host_instance_finalize,
};
static void pcie_host_register_types(void)
diff --git a/include/hw/pci/pcie_host.h b/include/hw/pci/pcie_host.h
index 1228e36..1656bde 100644
--- a/include/hw/pci/pcie_host.h
+++ b/include/hw/pci/pcie_host.h
@@ -41,6 +41,7 @@ struct PCIExpressHost {
/* MMCONFIG mmio area */
MemoryRegion mmio;
+ MemoryRegion mapped_mmio;
};
int pcie_host_init(PCIExpressHost *e);
--
1.8.1.4
- [Qemu-devel] [PATCH 21/39] pcnet: use instance_finalize instead of exit, (continued)
- [Qemu-devel] [PATCH 21/39] pcnet: use instance_finalize instead of exit, Paolo Bonzini, 2013/06/04
- [Qemu-devel] [PATCH 22/39] rtl8139: use instance_finalize instead of exit, Paolo Bonzini, 2013/06/04
- [Qemu-devel] [PATCH 23/39] vmxnet3: split exit and instance_finalize, Paolo Bonzini, 2013/06/04
- [Qemu-devel] [PATCH 24/39] shpc: split shpc_free out of shpc_cleanup, Paolo Bonzini, 2013/06/04
- [Qemu-devel] [PATCH 25/39] pci_bridge: split pci_bridge_free from pci_bridge_exitfn, Paolo Bonzini, 2013/06/04
- [Qemu-devel] [PATCH 26/39] pcie_aer: pcie_aer_exit really frees stuff, Paolo Bonzini, 2013/06/04
- [Qemu-devel] [PATCH 27/39] pci_bridge: split exit and instance_finalize, Paolo Bonzini, 2013/06/04
- [Qemu-devel] [PATCH 28/39] ioh4320: split exit and instance_finalize, Paolo Bonzini, 2013/06/04
- [Qemu-devel] [PATCH 29/39] xio3130-downstream: split exit and instance_finalize, Paolo Bonzini, 2013/06/04
- [Qemu-devel] [PATCH 30/39] xio3130-upstream: split exit and instance_finalize, Paolo Bonzini, 2013/06/04
- [Qemu-devel] [PATCH 31/39] pcie: do not recreate mmcfg I/O region, use an alias instead,
Paolo Bonzini <=
- [Qemu-devel] [PATCH 32/39] esp: use instance_finalize instead of exit, Paolo Bonzini, 2013/06/04
- [Qemu-devel] [PATCH 33/39] lsi: use instance_finalize instead of exit, Paolo Bonzini, 2013/06/04
- [Qemu-devel] [PATCH 34/39] pvscsi: split exit and instance_finalize, Paolo Bonzini, 2013/06/04
- [Qemu-devel] [PATCH 35/39] usb-uhci: use instance_finalize instead of exit, Paolo Bonzini, 2013/06/04
- [Qemu-devel] [PATCH 36/39] virtio-pci: split exit and instance_finalize, Paolo Bonzini, 2013/06/04
- [Qemu-devel] [PATCH 37/39] wdt_i6300esb: use instance_finalize instead of exit, Paolo Bonzini, 2013/06/04
- [Qemu-devel] [PATCH 38/39] xen_pt: use instance_finalize instead of exit, Paolo Bonzini, 2013/06/04
- [Qemu-devel] [PATCH 39/39] tpm: move add/del_subregion to realize/unrealize, Paolo Bonzini, 2013/06/04
- Re: [Qemu-devel] [PATCH 00/39] Delay destruction of memory regions to instance_finalize, Andreas Färber, 2013/06/05