X-Git-Url: https://git.kernel.dk/?a=blobdiff_plain;f=engines%2Fsg.c;h=c46b9abaec871e5b97e599f276620bb9e001eaaa;hb=051382218cbe5101a5caa83eab55ed04608f8475;hp=800f47370ed31f6dd7a8f2756556060022488de5;hpb=b4dbb3cece5de5ebc723253b05f69eff1a129a6d;p=fio.git

diff --git a/engines/sg.c b/engines/sg.c
index 800f4737..c46b9aba 100644
--- a/engines/sg.c
+++ b/engines/sg.c
@@ -126,6 +126,9 @@ static struct fio_option options[] = {
 #define MAX_10B_LBA  0xFFFFFFFFULL
 #define SCSI_TIMEOUT_MS 30000   // 30 second timeout; currently no method to override
 #define MAX_SB 64               // sense block maximum return size
+/*
+#define FIO_SGIO_DEBUG
+*/
 
 struct sgio_cmd {
 	unsigned char cdb[16];      // enhanced from 10 to support 16 byte commands
@@ -134,7 +137,7 @@ struct sgio_cmd {
 };
 
 struct sgio_trim {
-	char *unmap_param;
+	uint8_t *unmap_param;
 	unsigned int unmap_range_count;
 	struct io_u **trim_io_us;
 };
@@ -149,9 +152,42 @@ struct sgio_data {
 	int type_checked;
 	struct sgio_trim **trim_queues;
 	int current_queue;
+#ifdef FIO_SGIO_DEBUG
 	unsigned int *trim_queue_map;
+#endif
 };
 
+static inline uint32_t sgio_get_be32(uint8_t *buf)
+{
+	return be32_to_cpu(*((uint32_t *) buf));
+}
+
+static inline uint64_t sgio_get_be64(uint8_t *buf)
+{
+	return be64_to_cpu(*((uint64_t *) buf));
+}
+
+static inline void sgio_set_be16(uint16_t val, uint8_t *buf)
+{
+	uint16_t t = cpu_to_be16(val);
+
+	memcpy(buf, &t, sizeof(uint16_t));
+}
+
+static inline void sgio_set_be32(uint32_t val, uint8_t *buf)
+{
+	uint32_t t = cpu_to_be32(val);
+
+	memcpy(buf, &t, sizeof(uint32_t));
+}
+
+static inline void sgio_set_be64(uint64_t val, uint8_t *buf)
+{
+	uint64_t t = cpu_to_be64(val);
+
+	memcpy(buf, &t, sizeof(uint64_t));
+}
+
 static inline bool sgio_unbuffered(struct thread_data *td)
 {
 	return (td->o.odirect || td->o.sync_io);
@@ -329,13 +365,19 @@ re_read:
 
 			if (io_u->ddir == DDIR_TRIM) {
 				struct sgio_trim *st = sd->trim_queues[io_u->index];
+#ifdef FIO_SGIO_DEBUG
 				assert(st->trim_io_us[0] == io_u);
+				assert(sd->trim_queue_map[io_u->index] == io_u->index);
 				dprint(FD_IO, "sgio_getevents: reaping %d io_us from trim queue %d\n", st->unmap_range_count, io_u->index);
 				dprint(FD_IO, "sgio_getevents: reaped io_u %d and stored in events[%d]\n", io_u->index, i+trims);
+#endif
 				for (j = 1; j < st->unmap_range_count; j++) {
 					++trims;
 					sd->events[i + trims] = st->trim_io_us[j];
+#ifdef FIO_SGIO_DEBUG
 					dprint(FD_IO, "sgio_getevents: reaped io_u %d and stored in events[%d]\n", st->trim_io_us[j]->index, i+trims);
+					assert(sd->trim_queue_map[st->trim_io_us[j]->index] == io_u->index);
+#endif
 					if (hdr->info & SG_INFO_CHECK) {
 						/* record if an io error occurred, ignore resid */
 						memcpy(&st->trim_io_us[j]->hdr, hdr, sizeof(struct sg_io_hdr));
@@ -382,7 +424,8 @@ static enum fio_q_status fio_sgio_ioctl_doio(struct thread_data *td,
 	return FIO_Q_COMPLETED;
 }
 
-static enum fio_q_status fio_sgio_rw_doio(struct fio_file *f,
+static enum fio_q_status fio_sgio_rw_doio(struct thread_data *td,
+					  struct fio_file *f,
 					  struct io_u *io_u, int do_sync)
 {
 	struct sg_io_hdr *hdr = &io_u->hdr;
@@ -393,13 +436,32 @@ static enum fio_q_status fio_sgio_rw_doio(struct fio_file *f,
 		return ret;
 
 	if (do_sync) {
-		ret = read(f->fd, hdr, sizeof(*hdr));
-		if (ret < 0)
-			return ret;
+		/*
+		 * We can't just read back the first command that completes
+		 * and assume it's the one we need, it could be any command
+		 * that is inflight.
+		 */
+		do {
+			struct io_u *__io_u;
 
-		/* record if an io error occurred */
-		if (hdr->info & SG_INFO_CHECK)
-			io_u->error = EIO;
+			ret = read(f->fd, hdr, sizeof(*hdr));
+			if (ret < 0)
+				return ret;
+
+			__io_u = hdr->usr_ptr;
+
+			/* record if an io error occurred */
+			if (hdr->info & SG_INFO_CHECK)
+				__io_u->error = EIO;
+
+			if (__io_u == io_u)
+				break;
+
+			if (io_u_sync_complete(td, __io_u)) {
+				ret = -1;
+				break;
+			}
+		} while (1);
 
 		return FIO_Q_COMPLETED;
 	}
@@ -415,10 +477,11 @@ static enum fio_q_status fio_sgio_doio(struct thread_data *td,
 
 	if (f->filetype == FIO_TYPE_BLOCK) {
 		ret = fio_sgio_ioctl_doio(td, f, io_u);
-		td_verror(td, io_u->error, __func__);
+		if (io_u->error)
+			td_verror(td, io_u->error, __func__);
 	} else {
-		ret = fio_sgio_rw_doio(f, io_u, do_sync);
-		if (do_sync)
+		ret = fio_sgio_rw_doio(td, f, io_u, do_sync);
+		if (io_u->error && do_sync)
 			td_verror(td, io_u->error, __func__);
 	}
 
@@ -429,25 +492,11 @@ static void fio_sgio_rw_lba(struct sg_io_hdr *hdr, unsigned long long lba,
 			    unsigned long long nr_blocks)
 {
 	if (lba < MAX_10B_LBA) {
-		hdr->cmdp[2] = (unsigned char) ((lba >> 24) & 0xff);
-		hdr->cmdp[3] = (unsigned char) ((lba >> 16) & 0xff);
-		hdr->cmdp[4] = (unsigned char) ((lba >>  8) & 0xff);
-		hdr->cmdp[5] = (unsigned char) (lba & 0xff);
-		hdr->cmdp[7] = (unsigned char) ((nr_blocks >> 8) & 0xff);
-		hdr->cmdp[8] = (unsigned char) (nr_blocks & 0xff);
+		sgio_set_be32((uint32_t) lba, &hdr->cmdp[2]);
+		sgio_set_be16((uint16_t) nr_blocks, &hdr->cmdp[7]);
 	} else {
-		hdr->cmdp[2] = (unsigned char) ((lba >> 56) & 0xff);
-		hdr->cmdp[3] = (unsigned char) ((lba >> 48) & 0xff);
-		hdr->cmdp[4] = (unsigned char) ((lba >> 40) & 0xff);
-		hdr->cmdp[5] = (unsigned char) ((lba >> 32) & 0xff);
-		hdr->cmdp[6] = (unsigned char) ((lba >> 24) & 0xff);
-		hdr->cmdp[7] = (unsigned char) ((lba >> 16) & 0xff);
-		hdr->cmdp[8] = (unsigned char) ((lba >>  8) & 0xff);
-		hdr->cmdp[9] = (unsigned char) (lba & 0xff);
-		hdr->cmdp[10] = (unsigned char) ((nr_blocks >> 32) & 0xff);
-		hdr->cmdp[11] = (unsigned char) ((nr_blocks >> 16) & 0xff);
-		hdr->cmdp[12] = (unsigned char) ((nr_blocks >> 8) & 0xff);
-		hdr->cmdp[13] = (unsigned char) (nr_blocks & 0xff);
+		sgio_set_be64(lba, &hdr->cmdp[2]);
+		sgio_set_be32((uint32_t) nr_blocks, &hdr->cmdp[10]);
 	}
 
 	return;
@@ -526,29 +575,23 @@ static int fio_sgio_prep(struct thread_data *td, struct io_u *io_u)
 			sd->current_queue = io_u->index;
 			st = sd->trim_queues[sd->current_queue];
 			hdr->dxferp = st->unmap_param;
+#ifdef FIO_SGIO_DEBUG
 			assert(sd->trim_queues[io_u->index]->unmap_range_count == 0);
 			dprint(FD_IO, "sg: creating new queue based on io_u %d\n", io_u->index);
+#endif
 		}
 		else
 			st = sd->trim_queues[sd->current_queue];
 
 		dprint(FD_IO, "sg: adding io_u %d to trim queue %d\n", io_u->index, sd->current_queue);
 		st->trim_io_us[st->unmap_range_count] = io_u;
+#ifdef FIO_SGIO_DEBUG
 		sd->trim_queue_map[io_u->index] = sd->current_queue;
+#endif
 
 		offset = 8 + 16 * st->unmap_range_count;
-		st->unmap_param[offset] = (unsigned char) ((lba >> 56) & 0xff);
-		st->unmap_param[offset+1] = (unsigned char) ((lba >> 48) & 0xff);
-		st->unmap_param[offset+2] = (unsigned char) ((lba >> 40) & 0xff);
-		st->unmap_param[offset+3] = (unsigned char) ((lba >> 32) & 0xff);
-		st->unmap_param[offset+4] = (unsigned char) ((lba >> 24) & 0xff);
-		st->unmap_param[offset+5] = (unsigned char) ((lba >> 16) & 0xff);
-		st->unmap_param[offset+6] = (unsigned char) ((lba >>  8) & 0xff);
-		st->unmap_param[offset+7] = (unsigned char) (lba & 0xff);
-		st->unmap_param[offset+8] = (unsigned char) ((nr_blocks >> 32) & 0xff);
-		st->unmap_param[offset+9] = (unsigned char) ((nr_blocks >> 16) & 0xff);
-		st->unmap_param[offset+10] = (unsigned char) ((nr_blocks >> 8) & 0xff);
-		st->unmap_param[offset+11] = (unsigned char) (nr_blocks & 0xff);
+		sgio_set_be64(lba, &st->unmap_param[offset]);
+		sgio_set_be32((uint32_t) nr_blocks, &st->unmap_param[offset + 8]);
 
 		st->unmap_range_count++;
 
@@ -567,14 +610,12 @@ static int fio_sgio_prep(struct thread_data *td, struct io_u *io_u)
 
 static void fio_sgio_unmap_setup(struct sg_io_hdr *hdr, struct sgio_trim *st)
 {
-	hdr->dxfer_len = st->unmap_range_count * 16 + 8;
-	hdr->cmdp[7] = (unsigned char) (((st->unmap_range_count * 16 + 8) >> 8) & 0xff);
-	hdr->cmdp[8] = (unsigned char) ((st->unmap_range_count * 16 + 8) & 0xff);
+	uint16_t cnt = st->unmap_range_count * 16;
 
-	st->unmap_param[0] = (unsigned char) (((16 * st->unmap_range_count + 6) >> 8) & 0xff);
-	st->unmap_param[1] = (unsigned char)  ((16 * st->unmap_range_count + 6) & 0xff);
-	st->unmap_param[2] = (unsigned char) (((16 * st->unmap_range_count) >> 8) & 0xff);
-	st->unmap_param[3] = (unsigned char)  ((16 * st->unmap_range_count) & 0xff);
+	hdr->dxfer_len = cnt + 8;
+	sgio_set_be16(cnt + 8, &hdr->cmdp[7]);
+	sgio_set_be16(cnt + 6, st->unmap_param);
+	sgio_set_be16(cnt, &st->unmap_param[2]);
 
 	return;
 }
@@ -597,8 +638,10 @@ static enum fio_q_status fio_sgio_queue(struct thread_data *td,
 
 			/* finish cdb setup for unmap because we are
 			** doing unmap commands synchronously */
+#ifdef FIO_SGIO_DEBUG
 			assert(st->unmap_range_count == 1);
 			assert(io_u == st->trim_io_us[0]);
+#endif
 			hdr = &io_u->hdr;
 
 			fio_sgio_unmap_setup(hdr, st);
@@ -656,38 +699,39 @@ static int fio_sgio_commit(struct thread_data *td)
 
 	sd->current_queue = -1;
 
-	ret = fio_sgio_rw_doio(io_u->file, io_u, 0);
+	ret = fio_sgio_rw_doio(td, io_u->file, io_u, 0);
 
-	if (ret < 0)
-		for (i = 0; i < st->unmap_range_count; i++)
-			st->trim_io_us[i]->error = errno;
-	else if (hdr->status)
-		for (i = 0; i < st->unmap_range_count; i++) {
-			st->trim_io_us[i]->resid = hdr->resid;
-			st->trim_io_us[i]->error = EIO;
+	if (ret < 0 || hdr->status) {
+		int error;
+
+		if (ret < 0)
+			error = errno;
+		else {
+			error = EIO;
+			ret = -EIO;
 		}
-	else {
-		if (fio_fill_issue_time(td)) {
-			fio_gettime(&now, NULL);
-			for (i = 0; i < st->unmap_range_count; i++) {
-				struct io_u *io_u = st->trim_io_us[i];
-
-				memcpy(&io_u->issue_time, &now, sizeof(now));
-				io_u_queued(td, io_u);
-			}
+
+		for (i = 0; i < st->unmap_range_count; i++) {
+			st->trim_io_us[i]->error = error;
+			clear_io_u(td, st->trim_io_us[i]);
+			if (hdr->status)
+				st->trim_io_us[i]->resid = hdr->resid;
 		}
-		io_u_mark_submit(td, st->unmap_range_count);
+
+		td_verror(td, error, "xfer");
+		return ret;
 	}
 
-	if (io_u->error) {
-		td_verror(td, io_u->error, "xfer");
-		return 0;
+	if (fio_fill_issue_time(td)) {
+		fio_gettime(&now, NULL);
+		for (i = 0; i < st->unmap_range_count; i++) {
+			memcpy(&st->trim_io_us[i]->issue_time, &now, sizeof(now));
+			io_u_queued(td, io_u);
+		}
 	}
+	io_u_mark_submit(td, st->unmap_range_count);
 
-	if (ret == FIO_Q_QUEUED)
-		return 0;
-	else
-		return ret;
+	return 0;
 }
 
 static struct io_u *fio_sgio_event(struct thread_data *td, int event)
@@ -705,6 +749,8 @@ static int fio_sgio_read_capacity(struct thread_data *td, unsigned int *bs,
 	 * io_u structures, which are not initialized until later.
 	 */
 	struct sg_io_hdr hdr;
+	unsigned long long hlba;
+	unsigned int blksz = 0;
 	unsigned char cmd[16];
 	unsigned char sb[64];
 	unsigned char buf[32];  // read capacity return
@@ -741,23 +787,23 @@ static int fio_sgio_read_capacity(struct thread_data *td, unsigned int *bs,
 		return ret;
 	}
 
-	*bs	 = ((unsigned long) buf[4] << 24) | ((unsigned long) buf[5] << 16) |
-		   ((unsigned long) buf[6] << 8) | (unsigned long) buf[7];
-	*max_lba = ((unsigned long) buf[0] << 24) | ((unsigned long) buf[1] << 16) |
-		   ((unsigned long) buf[2] << 8) | (unsigned long) buf[3];
+	if (hdr.info & SG_INFO_CHECK) {
+		/* RCAP(10) might be unsupported by device. Force RCAP(16) */
+		hlba = MAX_10B_LBA;
+	} else {
+		blksz = sgio_get_be32(&buf[4]);
+		hlba = sgio_get_be32(buf);
+	}
 
 	/*
 	 * If max lba masked by MAX_10B_LBA equals MAX_10B_LBA,
 	 * then need to retry with 16 byte Read Capacity command.
 	 */
-	if (*max_lba == MAX_10B_LBA) {
+	if (hlba == MAX_10B_LBA) {
 		hdr.cmd_len = 16;
 		hdr.cmdp[0] = 0x9e; // service action
 		hdr.cmdp[1] = 0x10; // Read Capacity(16)
-		hdr.cmdp[10] = (unsigned char) ((sizeof(buf) >> 24) & 0xff);
-		hdr.cmdp[11] = (unsigned char) ((sizeof(buf) >> 16) & 0xff);
-		hdr.cmdp[12] = (unsigned char) ((sizeof(buf) >> 8) & 0xff);
-		hdr.cmdp[13] = (unsigned char) (sizeof(buf) & 0xff);
+		sgio_set_be32(sizeof(buf), &hdr.cmdp[10]);
 
 		hdr.dxfer_direction = SG_DXFER_FROM_DEV;
 		hdr.dxferp = buf;
@@ -773,19 +819,20 @@ static int fio_sgio_read_capacity(struct thread_data *td, unsigned int *bs,
 		if (hdr.info & SG_INFO_CHECK)
 			td_verror(td, EIO, "fio_sgio_read_capacity");
 
-		*bs = (buf[8] << 24) | (buf[9] << 16) | (buf[10] << 8) | buf[11];
-		*max_lba = ((unsigned long long)buf[0] << 56) |
-				((unsigned long long)buf[1] << 48) |
-				((unsigned long long)buf[2] << 40) |
-				((unsigned long long)buf[3] << 32) |
-				((unsigned long long)buf[4] << 24) |
-				((unsigned long long)buf[5] << 16) |
-				((unsigned long long)buf[6] << 8) |
-				(unsigned long long)buf[7];
+		blksz = sgio_get_be32(&buf[8]);
+		hlba = sgio_get_be64(buf);
+	}
+
+	if (blksz) {
+		*bs = blksz;
+		*max_lba = hlba;
+		ret = 0;
+	} else {
+		ret = EIO;
 	}
 
 	close(fd);
-	return 0;
+	return ret;
 }
 
 static void fio_sgio_cleanup(struct thread_data *td)
@@ -799,7 +846,9 @@ static void fio_sgio_cleanup(struct thread_data *td)
 		free(sd->fd_flags);
 		free(sd->pfds);
 		free(sd->sgbuf);
+#ifdef FIO_SGIO_DEBUG
 		free(sd->trim_queue_map);
+#endif
 
 		for (i = 0; i < td->o.iodepth; i++) {
 			free(sd->trim_queues[i]->unmap_param);
@@ -828,7 +877,9 @@ static int fio_sgio_init(struct thread_data *td)
 
 	sd->trim_queues = calloc(td->o.iodepth, sizeof(struct sgio_trim *));
 	sd->current_queue = -1;
+#ifdef FIO_SGIO_DEBUG
 	sd->trim_queue_map = calloc(td->o.iodepth, sizeof(int));
+#endif
 	for (i = 0; i < td->o.iodepth; i++) {
 		sd->trim_queues[i] = calloc(1, sizeof(struct sgio_trim));
 		st = sd->trim_queues[i];