diff --git a/drivers/mmc/au1xmmc.c b/drivers/mmc/au1xmmc.c
index 914d62b24064..5dc4bee7abeb 100644
--- a/drivers/mmc/au1xmmc.c
+++ b/drivers/mmc/au1xmmc.c
@@ -310,7 +310,7 @@ static void au1xmmc_data_complete(struct au1xmmc_host *host, u32 status)
 		}
 		else
 			data->bytes_xfered =
-				(data->blocks * (1 << data->blksz_bits)) -
+				(data->blocks * data->blksz) -
 				host->pio.len;
 	}
 
@@ -575,7 +575,7 @@ static int
 au1xmmc_prepare_data(struct au1xmmc_host *host, struct mmc_data *data)
 {
 
-	int datalen = data->blocks * (1 << data->blksz_bits);
+	int datalen = data->blocks * data->blksz;
 
 	if (dma != 0)
 		host->flags |= HOST_F_DMA;
@@ -596,7 +596,7 @@ au1xmmc_prepare_data(struct au1xmmc_host *host, struct mmc_data *data)
 	if (host->dma.len == 0)
 		return MMC_ERR_TIMEOUT;
 
-	au_writel((1 << data->blksz_bits) - 1, HOST_BLKSIZE(host));
+	au_writel(data->blksz - 1, HOST_BLKSIZE(host));
 
 	if (host->flags & HOST_F_DMA) {
 		int i;
diff --git a/drivers/mmc/imxmmc.c b/drivers/mmc/imxmmc.c
index 79358e223f57..a4eb1d0e7a71 100644
--- a/drivers/mmc/imxmmc.c
+++ b/drivers/mmc/imxmmc.c
@@ -218,8 +218,10 @@ static int imxmci_busy_wait_for_status(struct imxmci_host *host,
 	if(!loops)
 		return 0;
 
-	dev_info(mmc_dev(host->mmc), "busy wait for %d usec in %s, STATUS = 0x%x (0x%x)\n",
-		loops, where, *pstat, stat_mask);
+	/* The busy-wait is expected there for clock <8MHz due to SDHC hardware flaws */
+	if(!(stat_mask & STATUS_END_CMD_RESP) || (host->mmc->ios.clock>=8000000))
+		dev_info(mmc_dev(host->mmc), "busy wait for %d usec in %s, STATUS = 0x%x (0x%x)\n",
+			loops, where, *pstat, stat_mask);
 	return loops;
 }
 
@@ -333,6 +335,9 @@ static void imxmci_start_cmd(struct imxmci_host *host, struct mmc_command *cmd,
 	WARN_ON(host->cmd != NULL);
 	host->cmd = cmd;
 
+	/* Ensure, that clock are stopped else command programming and start fails */
+	imxmci_stop_clock(host);
+
 	if (cmd->flags & MMC_RSP_BUSY)
 		cmdat |= CMD_DAT_CONT_BUSY;
 
@@ -553,7 +558,7 @@ static int imxmci_cpu_driven_data(struct imxmci_host *host, unsigned int *pstat)
 	int trans_done = 0;
 	unsigned int stat = *pstat;
 
-	if(host->actual_bus_width == MMC_BUS_WIDTH_4)
+	if(host->actual_bus_width != MMC_BUS_WIDTH_4)
 		burst_len = 16;
 	else
 		burst_len = 64;
@@ -591,8 +596,7 @@ static int imxmci_cpu_driven_data(struct imxmci_host *host, unsigned int *pstat)
 			stat = MMC_STATUS;
 
 			/* Flush extra bytes from FIFO */
-			while(flush_len >= 2){
-				flush_len -= 2;
+			while(flush_len && !(stat & STATUS_DATA_TRANS_DONE)){
 				i = MMC_BUFFER_ACCESS;
 				stat = MMC_STATUS;
 				stat &= ~STATUS_CRC_READ_ERR; /* Stupid but required there */
@@ -746,10 +750,6 @@ static void imxmci_tasklet_fnc(unsigned long data)
 			data_dir_mask = STATUS_DATA_TRANS_DONE;
 		}
 
-		imxmci_busy_wait_for_status(host, &stat,
-				data_dir_mask,
-				50, "imxmci_tasklet_fnc data");
-
 		if(stat & data_dir_mask) {
 			clear_bit(IMXMCI_PEND_DMA_END_b, &host->pending_events);
 			imxmci_data_done(host, stat);
@@ -865,7 +865,11 @@ static void imxmci_set_ios(struct mmc_host *mmc, struct mmc_ios *ios)
 
 		imxmci_stop_clock(host);
 		MMC_CLK_RATE = (prescaler<<3) | clk;
-		imxmci_start_clock(host);
+		/*
+		 * Under my understanding, clock should not be started there, because it would
+		 * initiate SDHC sequencer and send last or random command into card
+		 */
+		/*imxmci_start_clock(host);*/
 
 		dev_dbg(mmc_dev(host->mmc), "MMC_CLK_RATE: 0x%08x\n", MMC_CLK_RATE);
 	} else {
diff --git a/drivers/mmc/mmc.c b/drivers/mmc/mmc.c
index 1ca2c8b9c9b5..6201f3086a02 100644
--- a/drivers/mmc/mmc.c
+++ b/drivers/mmc/mmc.c
@@ -951,6 +951,7 @@ static void mmc_read_scrs(struct mmc_host *host)
 		data.timeout_ns = card->csd.tacc_ns * 10;
 		data.timeout_clks = card->csd.tacc_clks * 10;
 		data.blksz_bits = 3;
+		data.blksz = 1 << 3;
 		data.blocks = 1;
 		data.flags = MMC_DATA_READ;
 		data.sg = &sg;
diff --git a/drivers/mmc/mmc_block.c b/drivers/mmc/mmc_block.c
index 06bd1f4cb9b1..e39cc05c64c2 100644
--- a/drivers/mmc/mmc_block.c
+++ b/drivers/mmc/mmc_block.c
@@ -175,6 +175,7 @@ static int mmc_blk_issue_rq(struct mmc_queue *mq, struct request *req)
 		brq.data.timeout_ns = card->csd.tacc_ns * 10;
 		brq.data.timeout_clks = card->csd.tacc_clks * 10;
 		brq.data.blksz_bits = md->block_bits;
+		brq.data.blksz = 1 << md->block_bits;
 		brq.data.blocks = req->nr_sectors >> (md->block_bits - 9);
 		brq.stop.opcode = MMC_STOP_TRANSMISSION;
 		brq.stop.arg = 0;
diff --git a/drivers/mmc/pxamci.c b/drivers/mmc/pxamci.c
index f97b472085cb..b49368fd96b8 100644
--- a/drivers/mmc/pxamci.c
+++ b/drivers/mmc/pxamci.c
@@ -119,7 +119,7 @@ static void pxamci_setup_data(struct pxamci_host *host, struct mmc_data *data)
 		nob = 0xffff;
 
 	writel(nob, host->base + MMC_NOB);
-	writel(1 << data->blksz_bits, host->base + MMC_BLKLEN);
+	writel(data->blksz, host->base + MMC_BLKLEN);
 
 	clks = (unsigned long long)data->timeout_ns * CLOCKRATE;
 	do_div(clks, 1000000000UL);
@@ -283,7 +283,7 @@ static int pxamci_data_done(struct pxamci_host *host, unsigned int stat)
 	 * data blocks as being in error.
 	 */
 	if (data->error == MMC_ERR_NONE)
-		data->bytes_xfered = data->blocks << data->blksz_bits;
+		data->bytes_xfered = data->blocks * data->blksz;
 	else
 		data->bytes_xfered = 0;
 
diff --git a/drivers/mmc/wbsd.c b/drivers/mmc/wbsd.c
index 39b3d97f891e..8167332d4013 100644
--- a/drivers/mmc/wbsd.c
+++ b/drivers/mmc/wbsd.c
@@ -662,14 +662,14 @@ static void wbsd_prepare_data(struct wbsd_host *host, struct mmc_data *data)
 	unsigned long dmaflags;
 
 	DBGF("blksz %04x blks %04x flags %08x\n",
-		1 << data->blksz_bits, data->blocks, data->flags);
+		data->blksz, data->blocks, data->flags);
 	DBGF("tsac %d ms nsac %d clk\n",
 		data->timeout_ns / 1000000, data->timeout_clks);
 
 	/*
 	 * Calculate size.
 	 */
-	host->size = data->blocks << data->blksz_bits;
+	host->size = data->blocks * data->blksz;
 
 	/*
 	 * Check timeout values for overflow.
@@ -696,12 +696,12 @@ static void wbsd_prepare_data(struct wbsd_host *host, struct mmc_data *data)
 	 * Two bytes are needed for each data line.
 	 */
 	if (host->bus_width == MMC_BUS_WIDTH_1) {
-		blksize = (1 << data->blksz_bits) + 2;
+		blksize = data->blksz + 2;
 
 		wbsd_write_index(host, WBSD_IDX_PBSMSB, (blksize >> 4) & 0xF0);
 		wbsd_write_index(host, WBSD_IDX_PBSLSB, blksize & 0xFF);
 	} else if (host->bus_width == MMC_BUS_WIDTH_4) {
-		blksize = (1 << data->blksz_bits) + 2 * 4;
+		blksize = data->blksz + 2 * 4;
 
 		wbsd_write_index(host, WBSD_IDX_PBSMSB,
 			((blksize >> 4) & 0xF0) | WBSD_DATA_WIDTH);
diff --git a/include/linux/mmc/mmc.h b/include/linux/mmc/mmc.h
index bdc556d88498..03a14a30c46a 100644
--- a/include/linux/mmc/mmc.h
+++ b/include/linux/mmc/mmc.h
@@ -69,6 +69,7 @@ struct mmc_data {
 	unsigned int		timeout_ns;	/* data timeout (in ns, max 80ms) */
 	unsigned int		timeout_clks;	/* data timeout (in clocks) */
 	unsigned int		blksz_bits;	/* data block size */
+	unsigned int		blksz;		/* data block size */
 	unsigned int		blocks;		/* number of blocks */
 	unsigned int		error;		/* data error */
 	unsigned int		flags;