[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: [Qemu-devel] unique (or otherwise) RAM block names
From: |
Dr. David Alan Gilbert |
Subject: |
Re: [Qemu-devel] unique (or otherwise) RAM block names |
Date: |
Fri, 1 Jun 2018 18:42:43 +0100 |
User-agent: |
Mutt/1.9.5 (2018-04-13) |
* Peter Maydell (address@hidden) wrote:
> I was going through trying to fix up various devices that currently
> fail to register RAM blocks for migration, or register the RAM
> block globally rather than locally, and I ran into something
> unexpected.
>
> We name RAM blocks in qemu_ram_set_idstr() like this:
> * if the user passed a device pointer, then call qdev_get_dev_path(),
> and if that returns non-NULL, use "path/name"
> * otherwise, use "name"
>
> Unfortunately, it turns out that there's no guarantee that
> qdev_get_dev_path() will return anything useful. If the device
> isn't on a bus, or the bus's class doesn't implement the get_dev_path
> method, then it'll return NULL.
>
> In particular, this means that if you create what you expect to
> be a local-to-this-device RAM memory region in a SysBusDevice,
> then (because SysBus doesn't implement get_dev_path), there is
> no per-device qualification added to the region name, and so the
> code silently creates a globally-namespaced RAM region.
> Trying to create multiple instances of the device therefore fails.
For device we also have a nmumerical 'instance_id' parameter that
some odd setups use.
> How can we make this work (preferably without breaking migration
> compat for existing devices) ?
>
> (Sysbus isn't the only bus that doesn't implement get_dev_path,
> it's just the first one I found.)
My preference would be that for new machine types we should
define get_dev_path on the busses that are missing it.
Dave
> thanks
> -- PMM
--
Dr. David Alan Gilbert / address@hidden / Manchester, UK