From: Jarod Wilson <jwilson@redhat.com>
Subject: [RHEL5.1 PATCH] ia64/xen: (updated) blktap does not build on ia64 (bz#216293)
Date: Thu, 28 Jun 2007 17:47:51 -0400
Bugzilla: 216293
Message-Id: <46842C87.7030800@redhat.com>
Changelog: [xen] ia64: enable blktap driver


The attached is an updated ia64 blktap patch that includes one
additional line of code over the earlier v2 version of the same patch,
which kludges around the 'tapdisk memory growing without bound' problem
in an ia64-specific fashion. Actually, the entire patch is
ia64-specific, as all changes, while in common code, are conditional on
XENFEAT_auto_translated_physmap, which is only found on ia64.

-- 
Jarod Wilson
jwilson@redhat.com


diff -Naurp linux-2.6.18.ia64.orig/drivers/xen/blktap/blktapmain.c linux-2.6.18.ia64/drivers/xen/blktap/blktapmain.c
--- linux-2.6.18.ia64.orig/drivers/xen/blktap/blktapmain.c	2007-06-28 17:32:15.000000000 -0400
+++ linux-2.6.18.ia64/drivers/xen/blktap/blktapmain.c	2007-06-28 17:38:41.000000000 -0400
@@ -840,6 +840,16 @@ static void fast_flush_area(pending_req_
 		WPRINTK("fast_flush: Couldn't get info!\n");
 		return;
 	}
+
+	if (info->vma != NULL &&
+	    xen_feature(XENFEAT_auto_translated_physmap)) {
+		down_write(&info->vma->vm_mm->mmap_sem);
+		zap_page_range(info->vma,
+			       MMAP_VADDR(info->user_vstart, u_idx, 0),
+			       req->nr_pages << PAGE_SHIFT, NULL);
+		up_write(&info->vma->vm_mm->mmap_sem);
+	}
+
 	mmap_idx = req->mem_idx;
 
 	for (i = 0; i < req->nr_pages; i++) {
@@ -856,6 +866,7 @@ static void fast_flush_area(pending_req_
 		}
 
 		if (khandle->user != INVALID_GRANT_HANDLE) {
+			BUG_ON(xen_feature(XENFEAT_auto_translated_physmap));
 			if (create_lookup_pte_addr(
 		    		info->vma->vm_mm,
 		    		MMAP_VADDR(info->user_vstart, u_idx, i),
@@ -878,7 +889,7 @@ static void fast_flush_area(pending_req_
 		GNTTABOP_unmap_grant_ref, unmap, invcount);
 	BUG_ON(ret);
 	
-	if (info->vma != NULL)
+	if (info->vma != NULL && !xen_feature(XENFEAT_auto_translated_physmap))
 		zap_page_range(info->vma, 
 			       MMAP_VADDR(info->user_vstart, u_idx, 0), 
 			       req->nr_pages << PAGE_SHIFT, NULL);
@@ -1210,70 +1221,105 @@ static void dispatch_rw_block_io(blkif_t
 				  req->seg[i].gref, blkif->domid);
 		op++;
 
-		/* Now map it to user. */
-		ret = create_lookup_pte_addr(info->vma->vm_mm, 
-					     uvaddr, &ptep);
-		if (ret) {
-			WPRINTK("Couldn't get a pte addr!\n");
-			fast_flush_area(pending_req, pending_idx, usr_idx, 
-					blkif->dev_num);
-			goto fail_flush;
-		}
+		if (!xen_feature(XENFEAT_auto_translated_physmap)) {
+			/* Now map it to user. */
+			ret = create_lookup_pte_addr(info->vma->vm_mm, 
+						     uvaddr, &ptep);
+			if (ret) {
+				WPRINTK("Couldn't get a pte addr!\n");
+				fast_flush_area(pending_req, pending_idx, usr_idx, 
+						blkif->dev_num);
+				goto fail_flush;
+			}
 
-		flags = GNTMAP_host_map | GNTMAP_application_map
-			| GNTMAP_contains_pte;
-		if (operation == WRITE)
-			flags |= GNTMAP_readonly;
-		gnttab_set_map_op(&map[op], ptep, flags,
-				  req->seg[i].gref, blkif->domid);
-		op++;
+			flags = GNTMAP_host_map | GNTMAP_application_map
+				| GNTMAP_contains_pte;
+			if (operation == WRITE)
+				flags |= GNTMAP_readonly;
+			gnttab_set_map_op(&map[op], ptep, flags,
+					  req->seg[i].gref, blkif->domid);
+			op++;
+		}
 	}
 
 	ret = HYPERVISOR_grant_table_op(GNTTABOP_map_grant_ref, map, op);
 	BUG_ON(ret);
 
-	for (i = 0; i < (nseg*2); i+=2) {
-		unsigned long uvaddr;
-		unsigned long kvaddr;
-		unsigned long offset;
-		struct page *pg;
+	if (!xen_feature(XENFEAT_auto_translated_physmap)) {
+		for (i = 0; i < (nseg*2); i+=2) {
+			unsigned long uvaddr;
+			unsigned long kvaddr;
+			unsigned long offset;
+			struct page *pg;
 
-		uvaddr = MMAP_VADDR(info->user_vstart, usr_idx, i/2);
-		kvaddr = idx_to_kaddr(mmap_idx, pending_idx, i/2);
+			uvaddr = MMAP_VADDR(info->user_vstart, usr_idx, i/2);
+			kvaddr = idx_to_kaddr(mmap_idx, pending_idx, i/2);
 
-		if (unlikely(map[i].status != 0)) {
-			WPRINTK("invalid kernel buffer -- "
-				"could not remap it\n");
-			ret |= 1;
-			map[i].handle = INVALID_GRANT_HANDLE;
-		}
-
-		if (unlikely(map[i+1].status != 0)) {
-			WPRINTK("invalid user buffer -- "
-				"could not remap it\n");
-			ret |= 1;
-			map[i+1].handle = INVALID_GRANT_HANDLE;
-		}
-
-		pending_handle(mmap_idx, pending_idx, i/2).kernel 
-			= map[i].handle;
-		pending_handle(mmap_idx, pending_idx, i/2).user   
-			= map[i+1].handle;
+			if (unlikely(map[i].status != 0)) {
+				WPRINTK("invalid kernel buffer -- "
+					"could not remap it\n");
+				ret |= 1;
+				map[i].handle = INVALID_GRANT_HANDLE;
+			}
 
-		if (ret)
-			continue;
+			if (unlikely(map[i+1].status != 0)) {
+				WPRINTK("invalid user buffer -- "
+					"could not remap it\n");
+				ret |= 1;
+				map[i+1].handle = INVALID_GRANT_HANDLE;
+			}
 
-		set_phys_to_machine(__pa(kvaddr) >> PAGE_SHIFT,
-			FOREIGN_FRAME(map[i].dev_bus_addr >> PAGE_SHIFT));
-		offset = (uvaddr - info->vma->vm_start) >> PAGE_SHIFT;
-		pg = pfn_to_page(__pa(kvaddr) >> PAGE_SHIFT);
-		((struct page **)info->vma->vm_private_data)[offset] =
-			pg;
+			pending_handle(mmap_idx, pending_idx, i/2).kernel 
+				= map[i].handle;
+			pending_handle(mmap_idx, pending_idx, i/2).user   
+				= map[i+1].handle;
+
+			if (ret)
+				continue;
+
+			set_phys_to_machine(__pa(kvaddr) >> PAGE_SHIFT,
+				FOREIGN_FRAME(map[i].dev_bus_addr >> PAGE_SHIFT));
+			offset = (uvaddr - info->vma->vm_start) >> PAGE_SHIFT;
+			pg = pfn_to_page(__pa(kvaddr) >> PAGE_SHIFT);
+			((struct page **)info->vma->vm_private_data)[offset] =
+				pg;
+		}
+	} else {
+		for (i = 0; i < nseg; i++) {
+			unsigned long uvaddr;
+			unsigned long kvaddr;
+			unsigned long offset;
+			struct page *pg;
+
+			uvaddr = MMAP_VADDR(info->user_vstart, usr_idx, i);
+			kvaddr = idx_to_kaddr(mmap_idx, pending_idx, i);
+
+			if (unlikely(map[i].status != 0)) {
+				WPRINTK("invalid kernel buffer -- "
+					"could not remap it\n");
+				ret |= 1;
+				map[i].handle = INVALID_GRANT_HANDLE;
+			}
+
+			pending_handle(mmap_idx, pending_idx, i).kernel 
+				= map[i].handle;
+
+			if (ret)
+				continue;
+
+			offset = (uvaddr - info->vma->vm_start) >> PAGE_SHIFT;
+			pg = pfn_to_page(__pa(kvaddr) >> PAGE_SHIFT);
+			((struct page **)info->vma->vm_private_data)[offset] =
+				pg;
+		}
 	}
 
 	if (ret)
 		goto fail_flush;
 
+	if (xen_feature(XENFEAT_auto_translated_physmap))
+		down_write(&info->vma->vm_mm->mmap_sem);
+
 	/* Mark mapped pages as reserved: */
 	for (i = 0; i < req->nr_segments; i++) {
 		unsigned long kvaddr;
@@ -1282,7 +1328,27 @@ static void dispatch_rw_block_io(blkif_t
 		kvaddr = idx_to_kaddr(mmap_idx, pending_idx, i);
 		pg = pfn_to_page(__pa(kvaddr) >> PAGE_SHIFT);
 		SetPageReserved(pg);
+		if (xen_feature(XENFEAT_auto_translated_physmap)) {
+			ret = vm_insert_page(info->vma,
+					     MMAP_VADDR(info->user_vstart,
+							usr_idx, i), pg);
+			if (ret) {
+				up_write(&info->vma->vm_mm->mmap_sem);
+				goto fail_flush;
+			}
+
+			/*
+			 * kludge
+			 * vm_insert_page() increments file_rss but
+			 * we don't want to increment so that compensate here.
+			 * The right fix is to our own vm_insert_page() which
+			 * doesn't increment file_rss.
+			 */
+			dec_mm_counter(info->vma->vm_mm, file_rss);
+		}
 	}
+	if (xen_feature(XENFEAT_auto_translated_physmap))
+		up_write(&info->vma->vm_mm->mmap_sem);
 	
 	/*record [mmap_idx,pending_idx] to [usr_idx] mapping*/
 	info->idx_map[usr_idx] = MAKE_ID(mmap_idx, pending_idx);
diff -Naurp linux-2.6.18.ia64.orig/include/asm-ia64/hypervisor.h linux-2.6.18.ia64/include/asm-ia64/hypervisor.h
--- linux-2.6.18.ia64.orig/include/asm-ia64/hypervisor.h	2007-06-28 17:32:16.000000000 -0400
+++ linux-2.6.18.ia64/include/asm-ia64/hypervisor.h	2007-06-28 17:38:30.000000000 -0400
@@ -192,6 +192,22 @@ MULTI_grant_table_op(multicall_entry_t *
 	mcl->args[2] = count;
 }
 
+/*
+ * for blktap.c
+ * int create_lookup_pte_addr(struct mm_struct *mm, 
+ *                            unsigned long address,
+ *                            uint64_t *ptep);
+ */
+#define create_lookup_pte_addr(mm, address, ptep)			\
+	({								\
+		printk(KERN_EMERG					\
+		       "%s:%d "						\
+		       "create_lookup_pte_addr() isn't supported.\n",	\
+		       __func__, __LINE__);				\
+		BUG();							\
+		(-ENOSYS);						\
+	})
+
 // for debug
 asmlinkage int xprintk(const char *fmt, ...);
 #define xprintd(fmt, ...)	xprintk("%s:%d " fmt, __func__, __LINE__, \