[linuxsh-dev] Question on SH/kernel discrepancies

[NIIBE Y. <gn...@m1...>]

Jeremy Siegel wrote:
 > We've been considering SH support in future distributions, and I've
 > noticed a couple differences between the SourceForge CVS and kernel.org
 > in non-SH-specific areas.  In particular:
 > 
 >   - In mm/memory.c:do_swap_page(), a pair of page/cache flush calls were
 > 
 >     moved out of an "if (!page)" block down toward the function end, so
 >     they are executed in more cases and at a different time.  This
 >     change was apparently put into the SH CVS source at version 1.7
 >     nearly 6 months ago, but isn't in kernel.org v2.4.1.  I'm not that
 >     familiar with the memory-management internals for linux (or SH); is
 >     this something that SH requires, or is it a "leftover" artifact of
 >     earlier cache/MMU testing?  If it's needed, is there some reason why
 > 
 >     it has been rejected by the main linux tree?

I believe it's needed for SH-4.  SH-4 has virtually indexed and
physically tagged cache system with write back.  For such architecture
the changes are required.

I've reported two times last summer, but got no interest.  Nowadays,
we have bugzilla for linux-mm, I will report it again.

 >   - In driver/ide/ide.c:ide_wait_cmd(), the SH code includes a delay
 >     immediately before function return.  Again, is this required for the
 > 
 >     SH, or is it leftover test/debug code?

IIRC, it is required for Hitachi SolutionEngine only.  It's kinda of
artifact, which should be removed (when some work around is found).
--