OSDN Git Service

hw/pci-bridge: Fix release ordering by embedding PCIBridgeWindows within PCIBridge
authorJonathan Cameron <Jonathan.Cameron@huawei.com>
Fri, 21 Apr 2023 12:25:50 +0000 (13:25 +0100)
committerThomas Huth <thuth@redhat.com>
Tue, 16 May 2023 07:14:18 +0000 (09:14 +0200)
commitb2999ed8ad6d9793b8310a1c7138a0232f637d1d
treee3dfc902a9692babb7e930f86cad2122ad14ab66
parent855436dbf756014a024f3e415001ead37301ef95
hw/pci-bridge: Fix release ordering by embedding PCIBridgeWindows within PCIBridge

The lifetime of the PCIBridgeWindows instance accessed via the windows pointer
in struct PCIBridge is managed separately from the PCIBridge itself.

Triggered by ./qemu-system-x86_64 -M x-remote -display none -monitor stdio
QEMU monitor: device_add cxl-downstream

In some error handling paths (such as the above due to attaching a cxl-downstream
port anything other than a cxl-upstream port) the g_free() of the PCIBridge
windows in pci_bridge_region_cleanup() is called before the final call of
flatview_uref() in address_space_set_flatview() ultimately from
drain_call_rcu()

At one stage this resulted in a crash, currently can still be observed using
valgrind which records a use after free.

When present, only one instance is allocated. pci_bridge_update_mappings()
can operate directly on an instance rather than creating a new one and
swapping it in.  Thus there appears to be no reason to not directly
couple the lifetimes of the two structures by embedding the PCIBridgeWindows
within the PCIBridge removing the need for the problematic separate free.

Patch is same as was posted deep in the discussion.
https://lore.kernel.org/qemu-devel/20230403171232.000020bb@huawei.com/

Reported-by: Thomas Huth <thuth@redhat.com>
Signed-off-by: Jonathan Cameron <Jonathan.Cameron@huawei.com>
Message-Id: <20230421122550.28234-1-Jonathan.Cameron@huawei.com>
Reviewed-by: Thomas Huth <thuth@redhat.com>
Signed-off-by: Thomas Huth <thuth@redhat.com>
hw/pci/pci_bridge.c
include/hw/pci/pci_bridge.h