On Thu 17 Oct, Robert Black wrote:
> On Oct 16,  8:52pm, Mark Smith wrote:
> > Subject: Re: SA110 - Current status
> > In message <9610151623.ZM7714@physig4.ph.kcl.ac.uk>
> >           "Mark Brinicombe" <amb@physig4.ph.kcl.ac.uk> wrote:
> >
> > > I dont need to clean on an interrupt but some cleaning is needed when ever
> any
> > > pages are remapped due to having a virtual cache.
> >
> > Why not do the following for each page you remap:
> >
> > 1) Disable interupts
> > 2) For each 32 byte potential data cache line within the page do an entry
> > specific clean and flush
> > 3) Remap the page, flush the data TLB entry
> > 4) Enable interrupts
> > 5) Check for any pending interrupts by calling the interrupt handler routine
> > 6) Repeat for each page
> 
> If you were to use this scheme during a context switch then for fairly typical
> large applications with 4MB of read-write mapped DRAM each you need to remap
> roughly 1024 pages of 512 cache lines. Assuming that on average you can purge
> 10 cache lines every microsecond (not an unreasonable estimate) this means that
> it takes you more than 50ms to purge a page. The standard BSD time-slice is
> 100ms so you could end up spending 50% of the CPU time in cache-flushing. For
> applications with huge data areas (eg raytracers or X) the situation would be
> much worse.
> 

This may sound a rather strange idea, and certainly isn't a short
term fix, but how easy would it be to replace the current context
switching system with a mechanism whereby all application binaries
are relocatable, and are loaded at their own address. At this 
point, a context switch would simply envolve marking all the pages
containing the current application as inaccessable, and those
containing the next as accessable, then jumping into the next
application. This would present some new difficulties---memory
allocation will probably have to be redesigned so that an 
application can own several heap blocks in totally differant
areas of memory (otherwise it will be necessary to decide
the total maximum amount of memory an application will _ever_
be allowed to use at load time.

Just a (silly) suggestion,

Thomas
-- 
But Lancelot mused a little space
He said, `she has a lovely face;
God in his mercy lend her grace,
  The Lady of Shalott.'