Speculating on page faults

Posted Jan 13, 2010 9:18 UTC (Wed) by intgr (subscriber, #39733)
Parent article: Speculating on page faults

So basically he took the rwsem implementation from x86 and ported it
to x86_64?

Why did the x86 and x86_64 implementations differ so significantly in
the first place?

to post comments

Speculating on page faults

Posted Jan 13, 2010 12:01 UTC (Wed) by mlankhorst (subscriber, #52260) [Link] (1 responses)

It just wasn't implemented for 64-bits and as fallback it used the generic
arch-neutral mechanism that used a spinlock, which is of course a lot
slower.

Speculating on page faults

Posted Feb 9, 2010 5:59 UTC (Tue) by yuhong (guest, #57183) [Link]

Yea, the more I read on LKML and the Linux x86-64 mailing lists, the more I think that developing the x86-64 port separately from the main x86 port was a big mistake in the first place. Look at the way the ACPI SRAT support was developed on the 32-bit and 64-bit side for an example, ACPI SRAT is supposed to be a *standard*, but before it was finally fixed in the year 2008, 32-bit SRAT/NUMA only worked for Summit systems, and not Opterons. Another example is AMD64 IOMMU support, which could have been supported on 32-bit Linux with PAE, but never was. It is now being gradually being fixed, but it was still stupid to do it in the first place.