On Sun, Aug 09, 2015 at 03:19:18PM -0700, David Miller wrote: > From: Sam Ravnborg <sam@xxxxxxxxxxxx> > Date: Sun, 9 Aug 2015 21:45:24 +0200 > > > When browsing sparc64 assembler I can see most > > handwritten assembler routines are aligned with ".align 32". > > The required alignment is 4 bytes. > > Some routines are aligned to 32 bytes in order to have them begin at > exactly an I-cache line boundary to increase I-cache utilization. > This is done especially for "hot" routines like memcpy, memset, etc. Makes perfect sense - thanks. I may try to dive into this a little to make this a bit more explicit. Something like: ENTRY(foo) - 4 byte aligned. Use 0x0 as filler. ENTRY_CACHE_ALIGNED - 32 bytes aligned, use 0x0 as filler. Today ENTRY() uses 0x90 as fille but as Kjetill writes in another mail this is a weird instruction. Before I can change the filler to 0x0 then I need to audit all uses of ENTRY(), because if anyone uses fallthrough then this would not work. Let me come up with a patch later then we can judgde if this is worthwhile then. Sam -- To unsubscribe from this list: send the line "unsubscribe sparclinux" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
