The last quarter was rather… peaceful. But nevertheless there were a few noteworthy things.
We always asserted that our library is multi-platform and hardware agnostic… as long as your platform is a recent distribution of Linux (or Windows) on x86 hardware :)
Two things happened that intend to change the current status quo:
I’ve talked about this previously, so I won’t repeat myself here, but the allocation classes were finally merged last quarter.
For anyone to whom this is a topic of interest, I suggest looking at the slab allocator example that shows the exact use case this feature is intended for.
Here’s a short, but hopefully interesting, bit of news: the happy-case
pmemobj_open() has increased by two orders of magnitude due to
strategic usage of lazy initialization.
Just as in traditional disks, NVDIMMs can have bad physical blocks on the medium. The specification on how those blocks are exposed has been recently published with ACPI 6.2.
From linux kernel file system perspective, the NVDIMM badblocks are exposed like in traditional storage, and can be cleared through normal fs interfaces.
The story changes a bit when the memory is mapped in the address space of the
application - unsurprisingly, the semantics are similar to those of bad volatile
memory - a
SIGBUS with appropriate flags is generated if a poisoned (bad) page
is touched by the program.
Regular software doesn’t really have to worry about poisoned pages because memory
is volatile and a
SIGBUS normally terminates the application, and when started
again, everything will be back in order.
Our initial approach focuses on letting the application terminate when a
is encountered, and then provide a set of offline tools that allow seamless
recovery of data from replicas.
The work is still in progress and the relevant PR is here.
Vacation season might appear to have taken a toll this time around, but that’s mostly just that - an appearance. The vast majority of changes this quarter were fixes and tiny improvements.