Ok, let's try the following patch as a test, it will kind of let me know if I'm barking up the wrong tree or not. This basically turns off the TSB path entirely in the TLB miss handler for THP mappings. Let me know if this changes the behavior on your Ultra-I/II machines. Thanks! diff --git a/arch/sparc/kernel/tsb.S b/arch/sparc/kernel/tsb.S index 14158d4..cc3d688 100644 --- a/arch/sparc/kernel/tsb.S +++ b/arch/sparc/kernel/tsb.S @@ -45,7 +45,7 @@ tsb_miss_itlb: */ tsb_miss_page_table_walk: TRAP_LOAD_TRAP_BLOCK(%g7, %g5) - +#if 0 /* Before committing to a full page table walk, * check the huge page TSB. */ @@ -94,7 +94,7 @@ tsb_miss_page_table_walk: 80: stx %g5, [%g7 + TRAP_PER_CPU_TSB_HUGE_TEMP] #endif - +#endif ldx [%g7 + TRAP_PER_CPU_PGD_PADDR], %g7 /* At this point we have: @@ -135,6 +135,11 @@ tsb_miss_page_table_walk_sun4v_fastpath: bne,pt %xcc, 60f nop +#if 1 + ba,pt %xcc, tsb_tlb_reload + nop +#endif +#if 0 /* It is a huge page, use huge page TSB entry address we * calculated above. If the huge page TSB has not been * allocated, setup a trap stack and call hugetlb_setup() @@ -172,6 +177,7 @@ tsb_miss_page_table_walk_sun4v_fastpath: add %sp, PTREGS_OFF, %o0 ba,pt %xcc, rtrap nop +#endif 60: #endif -- 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
