The TL;DR is basically that this patchset fixes a set of bugs in the
HPET subsystem, and should improve their handling.
I spoke with Andrew a bit about the patch about a week ago. He mentioned
that the current HPET subsystem is completely broken on Nehalem-era
hardware. The new interrupt handler is not without issues, as it can
result in a stack overflow. Hardware older than Nehalem-era hardware is
especially susceptible to this, as they only have 3 MSI channels on the
HPET, meaning you only need 4~5 interrupts at the same time to overflow
it. Nehalem-era hardware has 8 MSI channels on the HPET and I assume
newer hardware will have even more. As pre-Nehalem CPUs are from
2007-2008, I don't think anybody will be using them for Xen anyway.
So I don't think this will be a big deal for us, as the upsides outweigh
the downsides.
This is a fairly complicated set of patches and I can't summarize this properly, so I'm simply going to link to Andrew's proposal to improve the HPET subsystem: https://lore.kernel.org/xen-devel/526A6267.1080004@citrix.com/
The TL;DR is basically that this patchset fixes a set of bugs in the HPET subsystem, and should improve their handling.
I spoke with Andrew a bit about the patch about a week ago. He mentioned that the current HPET subsystem is completely broken on Nehalem-era hardware. The new interrupt handler is not without issues, as it can result in a stack overflow. Hardware older than Nehalem-era hardware is especially susceptible to this, as they only have 3 MSI channels on the HPET, meaning you only need 4~5 interrupts at the same time to overflow it. Nehalem-era hardware has 8 MSI channels on the HPET and I assume newer hardware will have even more. As pre-Nehalem CPUs are from 2007-2008, I don't think anybody will be using them for Xen anyway.
So I don't think this will be a big deal for us, as the upsides outweigh the downsides.