[linuxsh-dev] Odd problem

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

I've looked through the code of linux/mm, and improve the changes I've
done yesterday.  Here's the patch.

Important point is that when a page is allocated, it may have a stale
data/instruction on cache line of user space.  We need to flush the
cache.  I did "Flushing at 'end of I/O'" yesterday, but it's more than
needed.

I've found that when swap entry is read asynchronously, the cache
remains.

2000-08-17  NIIBE Yutaka  <gn...@m1...>

	* mm/memory.c (do_anonymous_page): We need to flush I-cache and
	D-cache here, as it's newly allocated page.
	(do_no_page): We need to flush D-cache.
	(do_swap_page): Flush D-cache & I-cache here.  There're cases
	where read_swap_cache is called asynchronously and pages are
	cached.

	* Revert the change for fs/buffer.c (end_buffer_io_async).  It's
	more than needed.  We only need to flush when kernel WRITES to the
	page (from I/O), not READ (to I/O).

	* Revert the changes for mm/memory.c (do_wp_page: case 1): We have
	valid PTE here (it's read-only but works).

Index: fs/buffer.c
===================================================================
RCS file: /cvsroot/linuxsh/kernel/fs/buffer.c,v
retrieving revision 1.7
diff -u -r1.7 buffer.c
--- fs/buffer.c	2000/08/16 08:34:05	1.7
+++ fs/buffer.c	2000/08/17 04:08:10
@@ -770,7 +770,6 @@
 	/* OK, the async IO on this page is complete. */
 	spin_unlock_irqrestore(&page_uptodate_lock, flags);
 
-	flush_dcache_page(page);
 	/*
 	 * if none of the buffers had errors then we can set the
 	 * page uptodate:
Index: mm/memory.c
===================================================================
RCS file: /cvsroot/linuxsh/kernel/mm/memory.c,v
retrieving revision 1.6
diff -u -r1.6 memory.c
--- mm/memory.c	2000/08/16 08:34:09	1.6
+++ mm/memory.c	2000/08/17 04:08:21
@@ -790,8 +790,8 @@
 		pte_t *page_table)
 {
 	copy_cow_page(old_page,new_page,address);
-	flush_icache_page(vma, new_page);
 	flush_dcache_page(new_page);
+	flush_icache_page(vma, new_page);
 	establish_pte(vma, address, page_table, pte_mkwrite(pte_mkdirty(mk_pte(new_page, vma->vm_page_prot))));
 }
 
@@ -849,7 +849,7 @@
 		UnlockPage(old_page);
 		/* FallThrough */
 	case 1:
-		flush_dcache_page(old_page);
+		flush_cache_page(vma, address);
 		establish_pte(vma, address, page_table, pte_mkyoung(pte_mkdirty(pte_mkwrite(pte))));
 		spin_unlock(&mm->page_table_lock);
 		return 1;	/* Minor fault */
@@ -1059,9 +1059,6 @@
 		unlock_kernel();
 		if (!page)
 			return -1;
-
-		flush_dcache_page(page);
-		flush_icache_page(vma, page);
 	}
 
 	mm->rss++;
@@ -1084,6 +1081,8 @@
 	} else
 		UnlockPage(page);
 
+	flush_dcache_page(page);
+	flush_icache_page(vma, page);
 	set_pte(page_table, pte);
 	/* No need to invalidate - it was non-present before */
 	update_mmu_cache(vma, address, pte);
@@ -1107,7 +1106,8 @@
 		clear_user_highpage(page, addr);
 		entry = pte_mkwrite(pte_mkdirty(mk_pte(page, vma->vm_page_prot)));
 		mm->rss++;
-		flush_page_to_ram(page);
+		flush_dcache_page(page);
+		flush_icache_page(vma, page);
 	}
 	set_pte(page_table, entry);
 	/* No need to invalidate - it was non-present before */
@@ -1156,7 +1156,7 @@
 	 * so we can make it writable and dirty to avoid having to
 	 * handle that later.
 	 */
-	flush_page_to_ram(new_page);
+	flush_dcache_page(new_page);
 	flush_icache_page(vma, new_page);
 	entry = mk_pte(new_page, vma->vm_page_prot);
 	if (write_access) {