On POWER the lshw version of beaker-system-scan is reporting the identical amount of memory as the smolt version did -- which appears to be just the free memory reported by the kernel, not the amount of memory physically installed. For example 15790 MB for a P8 LPAR that (presumably) has 16GB allocated.
The bug is that it's 15790MB but it should be 16384MB. (Actually I'm just guessing about that, but I doubt that the LPAR was defined as having 15790MB of memory, it was more likely 16GB = 16384MB.)
Anyway this is really just a low-priority RFE since the current data produced by lshw is no worse than smolt. The task here is to figure out how much actual "physical" memory is allocated to the LPAR (16384MB? something else?) and then figure out what in the kernel is reserving it and see if we can make lshw account for that, so that the amount of memory reported more closely matches how much is actually allocated. This is essentially the same bug or closely related to bug 1223961.
There are some IBM-specific properties under memory-controller in the device tree which we can use to grab DIMM info.
We can get most of the way there by looking at the hotplug info in /sys/devices/system/memory: http://gerrit.beaker-project.org/4290 In particular, that gives us better info on S/390 guests where we have no other means of checking memory size. We can also grab DIMM info from the /memory-controller node in device tree although that seems to only exist on the JS20. http://gerrit.beaker-project.org/4291
Beaker 21.0 has been released.