On 04/08/2014 07:44 AM, Ralf Baechle wrote:
On Tue, Apr 08, 2014 at 11:59:08AM +0100, Markos Chandras wrote:
This patchset removes the MIPS SMTC support.
While not really a fix I've applied this to my 3.15 fix branch. At least
it will avoid us having to fix it up for 3.15 :-)
That said, SMTC was a remarkable hack and ingenious proof of the MT
architecture.
Sigh. I was rather proud of it. If it's being excised (and I don't
begrudge people the right to deprecate it), I guess it means I can
unsubscribe from the mailing list, as supporting MT hacks was the main
reason I hung around at all after leaving MIPS.
Still a sore spot is CONFIG_MIPS_MT_FPAFF with all its uglyness it
scatters over the tree, in particular the wrapper around the syscall
entry point. I wonder if nowadays with the work that's been done on
supporting inhomogenous SMP systems (ARM biglittle) there's now a
better way to handle this sort of issue.
Even at the time, I'd identified a cleaner way to deal with this sort of
thing, but the maintainer for the code in question wouldn't give me the
time of day, so I solved the problem as well as the politics of Linux
allowed. As I understand it, the MIPS FP Affinity hack still
outperformed some rather more complex solutions that contemporary
inhomogeneous (is that different from heterogeneous?) MT machines (e.g.
Niagara) were using.
/K.
