MFC r284928: speed up ar(1) on UFS file systems

Fault in the buffer prior to writing to workaround poor performance due
to interaction with kernel fs deadlock avoidance code. See the comment
prior to vn_io_fault_doio() in sys/kern/vfs_vnops.c for details of the
issue.

On my stable/10 desktop with a 16MB obj.o and "ar r out.a obj.o" I see
the following run times (seconds):

x ar.r284891
+ ar.patched
+----------------------------------------------------------------------+
|+                                                                     |
|+                                                                    x|
|+                                                                   xx|
|A                                                                   |A|
+----------------------------------------------------------------------+
    N         Min          Max        Median           Avg        Stddev
x   3       1.307        1.321         1.315     1.3143333  0.0070237692
+   3       0.020        0.023         0.022   0.021666667  0.0015275252
Difference at 95.0% confidence
        -1.29267 +/- 0.0115203
        -98.3515% +/- 0.876513%
        (Student's t, pooled s = 0.00508265)

Thanks to kib for diagnosing and explaining the issue and suggesting
the workaround.

Approved by:	re (gjb)
Sponsored by:	The FreeBSD Foundation

1 files changed, 26 insertions, 0 deletions

diff --git a/usr.bin/ar/write.c b/usr.bin/ar/write.c
index 9ce1806..10b912e 100644
--- a/usr.bin/ar/write.c
+++ b/usr.bin/ar/write.c
@@ -41,6 +41,7 @@ __FBSDID("$FreeBSD$");
 #include <stdlib.h>
 #include <string.h>
 #include <sysexits.h>
+#include <unistd.h>
 
 #include "ar.h"
 
@@ -61,6 +62,7 @@ static void	create_symtab_entry(struct bsdar *bsdar, void *maddr,
 static void	free_obj(struct bsdar *bsdar, struct ar_obj *obj);
 static void	insert_obj(struct bsdar *bsdar, struct ar_obj *obj,
 		    struct ar_obj *pos);
+static void	prefault_buffer(const char *buf, size_t s);
 static void	read_objs(struct bsdar *bsdar, const char *archive,
 		    int checkargv);
 static void	write_archive(struct bsdar *bsdar, char mode);
@@ -551,11 +553,35 @@ write_cleanup(struct bsdar *bsdar)
 }
 
 /*
+ * Fault in the buffer prior to writing as a workaround for poor performance
+ * due to interaction with kernel fs deadlock avoidance code. See the comment
+ * above vn_io_fault_doio() in sys/kern/vfs_vnops.c for details of the issue.
+ */
+static void
+prefault_buffer(const char *buf, size_t s)
+{
+	volatile const char *p;
+	size_t page_size;
+
+	if (s == 0)
+		return;
+	page_size = sysconf(_SC_PAGESIZE);
+	for (p = buf; p < buf + s; p += page_size)
+		*p;
+	/*
+	 * Ensure we touch the last page as well, in case the buffer is not
+	 * page-aligned.
+	 */
+	*(volatile const char *)(buf + s - 1);
+}
+
+/*
  * Wrapper for archive_write_data().
  */
 static void
 write_data(struct bsdar *bsdar, struct archive *a, const void *buf, size_t s)
 {
+	prefault_buffer(buf, s);
 	if (archive_write_data(a, buf, s) != (ssize_t)s)
 		bsdar_errc(bsdar, EX_SOFTWARE, 0, "%s",
 		    archive_error_string(a));