| commit | dde522bbc5feb2862afb243bb49c590fe65dce66 | |
| author | Fei Li <sherrylf@linux.vnet.ibm.com> | |
| Thu, 24 Nov 2016 10:10:39 +0000 (24 11:10 +0100) | ||
| committer | Cornelia Huck <cornelia.huck@de.ibm.com> | |
| Fri, 21 Apr 2017 07:32:09 +0000 (21 09:32 +0200) | ||
| tree | 1823392ac35c26189f22d92b262af2a1bd783f0f | treesnapshot (tar.gz zip) |
| parent | bc66d6cbca780c5b8c237b9d355caa8274e5b312 | commitdiff |
s390x: register I/O adapters per ISC during init
The I/O adapters should exist as soon as the bus/infrastructure
exists, and not only when the guest is actually trying to do something
with them. While the lazy allocation was not wrong, allocating at init
time is cleaner, both for the architecture and the code. Let's adjust
this by having each device type (currently for PCI and virtio-ccw)
register the adapters for each ISC (as now we don't know which ISC the
guest will use) as soon as it initializes.
Use a two-dimensional array io_adapters[type][isc] to store adapters
in ChannelSubSys, so that we can conveniently get the adapter id by
the helper function css_get_adapter_id(type, isc).
Signed-off-by: Fei Li <sherrylf@linux.vnet.ibm.com>
Signed-off-by: Cornelia Huck <cornelia.huck@de.ibm.com>
The I/O adapters should exist as soon as the bus/infrastructure
exists, and not only when the guest is actually trying to do something
with them. While the lazy allocation was not wrong, allocating at init
time is cleaner, both for the architecture and the code. Let's adjust
this by having each device type (currently for PCI and virtio-ccw)
register the adapters for each ISC (as now we don't know which ISC the
guest will use) as soon as it initializes.
Use a two-dimensional array io_adapters[type][isc] to store adapters
in ChannelSubSys, so that we can conveniently get the adapter id by
the helper function css_get_adapter_id(type, isc).
Signed-off-by: Fei Li <sherrylf@linux.vnet.ibm.com>
Signed-off-by: Cornelia Huck <cornelia.huck@de.ibm.com>