mac: improve posix_fadvise shim

This builds upon the best effort macOS cache invalidation work by
DeveloperEcosystemEngineering@apple.com's by shimming even more of
posix_advise() on macOS:
- POSIX_FADV_NORMAL is ignored
- POSIX_FADV_SEQUENTIAL/POSIX_FADV_RANDOM are mapped to enable/disable
  readahead via fcntl(..., F_RDAHEAD, ...)

To paraphrase from the commit that was done for similar reasons
targeting Windows (see 8300eba ("windowsaio: add best effort cache
invalidation")):

This change may make default bandwidth speeds on macOS look lower
compared to older versions of fio but this matches the behaviour of fio
on other platforms with invalidation (such as Linux) because we are
trying to avoid measuring cache reuse (unless invalidate=0 is set).

The impact of invalidation is demonstrated by the bandwidths achieved by
the following jobs running on a SSD of an otherwise idle Intel Mac with
16GBytes of RAM:

./fio --stonewall --size=128M --ioengine=posixaio --filename=fio.tmp --iodepth=64 \
  --bs=4k --invalidate=0 --direct=0 \
  --name=create --rw=write \
  --name=cached --rw=randread --loops=2 \
  --name=invalidated --rw=randread --loops=2 --invalidate=1

[...]
cached: (groupid=1, jobs=1): err= 0: pid=32517: Sat May 17 18:37:57 2025
  read: IOPS=228k, BW=892MiB/s (935MB/s)(256MiB/287msec)
[...]
invalidated: (groupid=2, jobs=1): err= 0: pid=32518: Sat May 17 18:37:57 2025
  read: IOPS=43.1k, BW=168MiB/s (176MB/s)(256MiB/1522msec)

The impact of being able to tune readahead can be seen below:

./fio --filename=fio.tmp --stonewall --size=128M --filename=fio.tmp --bs=4k \
  --name=precache --rw=read --bs=128k \
  --name=cached --rw=read \
  --name=uncached-readahead --rw=read --invalidate=1 \
  --name=uncached-no-readahead --rw=read --invalidate=1 --fadvise=random

[...]
cached: (groupid=1, jobs=1): err= 0: pid=32600: Sat May 17 18:49:22 2025
  read: IOPS=349k, BW=1362MiB/s (1428MB/s)(128MiB/94msec)
--
uncached-readahead: (groupid=2, jobs=1): err= 0: pid=32601: Sat May 17 18:49:22 2025
  read: IOPS=315k, BW=1231MiB/s (1291MB/s)(128MiB/104msec)
--
uncached-no-readahead: (groupid=3, jobs=1): err= 0: pid=32602: Sat May 17 18:49:22 2025
  read: IOPS=40.8k, BW=159MiB/s (167MB/s)(128MiB/803msec)

Signed-off-by: Sitsofe Wheeler <sitsofe@yahoo.com>

Author: sitsofe

helpers.c

@@ -26,19 +26,9 @@ int sync_file_range(int fd, uint64_t offset, uint64_t nbytes,
 }
 #endif
 
-#if defined (__APPLE__)
-int posix_fadvise(int fd, off_t offset, off_t len, int advice)
-{
-	if (advice == POSIX_FADV_DONTNEED)
-		return discard_pages(fd, offset, len);
-        errno = EINVAL; 
-	return -1;
-}
-#else
 #ifndef CONFIG_POSIX_FADVISE
 int posix_fadvise(int fd, off_t offset, off_t len, int advice)
 {
 	return 0;
 }
 #endif
-#endif /* defined (__APPLE__) */

os/os-mac.h

@@ -6,6 +6,7 @@
 #include <errno.h>
 #include <fcntl.h>
 #include <sys/disk.h>
+#include <sys/mman.h>
 #include <sys/sysctl.h>
 #include <sys/time.h>
 #include <unistd.h>
@@ -14,8 +15,6 @@
 #include <machine/endian.h>
 #include <libkern/OSByteOrder.h>
 
-#include <sys/stat.h>
-#include <sys/mman.h>
 #include <stdio.h>
 
 
@@ -135,13 +134,6 @@ static inline int discard_pages(int fd, off_t offset, off_t size)
 	caddr_t *addr;
 	uint64_t chunk_size = MMAP_CHUNK_SIZE;
 
-	if (fsync(fd) < 0) {
-                int __err = errno; 
-		log_err("%s: Cannot fsync file: %s \n", __func__, strerror(errno));
-                errno = __err;
-		return -1;
-	}	    
-    
 	/*
 	 * mmap the file in 1GB chunks and msync(MS_INVALIDATE).
 	 */
@@ -154,22 +146,22 @@ static inline int discard_pages(int fd, off_t offset, off_t size)
 			fprintf(stderr, "Failed to mmap (%s), offset = %llu, size = %llu\n",
 				strerror(errno), offset, mmap_size);
                         errno = __map_errno;
-			return -1;
+			return errno;
 		}
 
 		if (msync(addr, mmap_size, MS_INVALIDATE)) {
                         int __msync_errno = errno;
 			fprintf(stderr, "msync failed to free cache pages.\n");
                         errno = __msync_errno;
-			return -1;
+			return errno;
 		}
 
 		/* Destroy the above mappings used to invalidate cache - cleaning up */
 		if (munmap(addr, mmap_size) < 0) {
                         int __munmap_errno = errno;
 			fprintf(stderr, "munmap failed, error = %d.\n", errno);
                         errno = __munmap_errno;
-			return -1;
+			return errno;
 		}
 
 		size -= mmap_size;
@@ -178,3 +170,43 @@ static inline int discard_pages(int fd, off_t offset, off_t size)
 
 	return 0;
 }
+
+static inline int set_readhead(int fd, bool enabled) {
+	int ret;
+
+	ret = fcntl(fd, F_RDAHEAD, enabled ? 1 : 0);
+	if (ret == -1) {
+		ret = errno;
+	}
+
+	return ret;
+}
+
+#define CONFIG_POSIX_FADVISE
+#define POSIX_FADV_NORMAL       (0)
+#define POSIX_FADV_RANDOM       (1)
+#define POSIX_FADV_SEQUENTIAL   (2)
+#define POSIX_FADV_DONTNEED     (4)
+static inline int posix_fadvise(int fd, off_t offset, off_t len, int advice)
+{
+	int ret;
+
+	switch(advice) {
+	case POSIX_FADV_NORMAL:
+		ret = 0;
+		break;
+	case POSIX_FADV_RANDOM:
+		ret = set_readhead(fd, false);
+		break;
+	case POSIX_FADV_SEQUENTIAL:
+		ret = set_readhead(fd, true);
+		break;
+	case POSIX_FADV_DONTNEED:
+		ret = discard_pages(fd, offset, len);
+		break;
+        default:
+		ret = EINVAL;
+        }
+
+	return ret;
+}

t/run-fio-tests.py

@@ -946,7 +946,7 @@ def check_result(self):
         'test_id':          35,
         'test_class':       FioJobFileTest,
         'job':              't0035.fio',
-        'success':          SUCCESS_STDERR,
+        'success':          SUCCESS_DEFAULT,
         'pre_job':          None,
         'pre_success':      None,
         'requirements':     [],