On Thu, Apr 4, 2019 at 1:56 PM Keith Busch <kbusch@xxxxxxxxxx> wrote: > > On Thu, Apr 04, 2019 at 12:08:44PM -0700, Dan Williams wrote: > > As of ACPI 6.3 the HMAT no longer advertises the physical memory address > > range for its entries. Instead, the expectation is the corresponding > > entry in the SRAT is looked up by the target proximity domain. > > > > Given there may be multiple distinct address ranges that share the same > > performance profile (sparse address space), find_mem_target() is updated > > to also consider the start address of the memory range. Target property > > updates are also adjusted to loop over all possible 'struct target' > > instances that may share the same proximity domain identification. > > Since this may allocate multiple targets with the same PXM, > hmat_register_targets() will attempt to register the same node multiple > times. > > Would it make sense if the existing struct memory_target adds a resource > list that we can append to as we parse SRAT? That way we have one target > per memory node, and also track the ranges. That sounds reasonable to me.
