Re: Preempt Threshold Measurements

[William Lee Irwin III]

On Mon, Jul 12, 2004 at 09:01:07PM -0700, Andrew Morton wrote:
>> This is a false positive.  Nothing is setting need_resched(), so
>> unmap_vmas() doesn't bother dropping the lock.

On Tue, Jul 13, 2004 at 03:08:15AM -0700, William Lee Irwin III wrote:
> I guess I sent too many updates and the whole thing got dropped. The false
> positives were fixed in this way:

> -			if (!atomic && need_resched()) {
> +			zap_bytes = ZAP_BLOCK_SIZE;
> +			if (!atomic)
> +				continue;
> +			touch_preempt_timing();
> +			if (need_resched()) {

That's not quite right. Amazing it didn't catch might_sleep() warnings.

Index: mm7-2.6.7/mm/memory.c
===================================================================
--- mm7-2.6.7.orig/mm/memory.c	2004-07-13 03:06:12.784491200 -0700
+++ mm7-2.6.7/mm/memory.c	2004-07-13 03:39:45.843459720 -0700
@@ -568,16 +568,16 @@
 			if ((long)zap_bytes > 0)
 				continue;
 			zap_bytes = ZAP_BLOCK_SIZE;
-			if (!atomic)
+			if (atomic)
 				continue;
-			touch_preempt_timing();
 			if (need_resched()) {
 				int fullmm = tlb_is_full_mm(*tlbp);
 				tlb_finish_mmu(*tlbp, tlb_start, start);
 				cond_resched_lock(&mm->page_table_lock);
 				*tlbp = tlb_gather_mmu(mm, fullmm);
 				tlb_start_valid = 0;
-			}
+			} else
+				touch_preempt_timing();
 		}
 	}
 	return ret;
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/