diff --git a/drivers/mmc/host/pxamci.c b/drivers/mmc/host/pxamci.c
index a8e18fe..5bbae05 100644
--- a/drivers/mmc/host/pxamci.c
+++ b/drivers/mmc/host/pxamci.c
@@ -36,6 +36,10 @@
 
 #include "pxamci.h"
 
+/* FIXME: For debugging purposes */
+#undef cpu_is_pxa25x
+#define cpu_is_pxa25x cpu_is_pxa27x
+
 #define DRIVER_NAME	"pxa2xx-mci"
 
 #define NR_SG	1
@@ -67,6 +71,10 @@ struct pxamci_host {
 	unsigned int		dma_dir;
 	unsigned int		dma_drcmrrx;
 	unsigned int		dma_drcmrtx;
+
+	dma_addr_t		bounce_dma;
+	void			*bounce_cpu;
+	unsigned int            offset;
 };
 
 static void pxamci_stop_clock(struct pxamci_host *host)
@@ -116,7 +124,7 @@ static void pxamci_setup_data(struct pxamci_host *host, struct mmc_data *data)
 	unsigned int timeout;
 	bool dalgn = 0;
 	u32 dcmd;
-	int i;
+	int i, j;
 
 	host->data = data;
 
@@ -148,22 +156,56 @@ static void pxamci_setup_data(struct pxamci_host *host, struct mmc_data *data)
 	host->dma_len = dma_map_sg(mmc_dev(host->mmc), data->sg, data->sg_len,
 				   host->dma_dir);
 
-	for (i = 0; i < host->dma_len; i++) {
+	for (i = 0, j = 0; i < host->dma_len; i++, j++) {
 		unsigned int length = sg_dma_len(&data->sg[i]);
-		host->sg_cpu[i].dcmd = dcmd | length;
-		if (length & 31 && !(data->flags & MMC_DATA_READ))
-			host->sg_cpu[i].dcmd |= DCMD_ENDIRQEN;
+		unsigned int addr = sg_dma_address(&data->sg[i]);
+		unsigned int offset = addr & 7;
 		/* Not aligned to 8-byte boundary? */
-		if (sg_dma_address(&data->sg[i]) & 0x7)
-			dalgn = 1;
+		if (offset) {
+        		if (cpu_is_pxa21x() || cpu_is_pxa25x()) {
+				/*
+				 * PXA25x doesn't support unaligned DMA
+				 * at all, use a bounce buffer once for
+				 * up to 7 unaligned bytes.
+				 */
+				BUG_ON(host->offset != 0);
+				offset = (8 - offset) & 7;
+				host->sg_cpu[j].dcmd = dcmd | offset;
+				if (length & 31 && !(data->flags & MMC_DATA_READ))
+					host->sg_cpu[j].dcmd |= DCMD_ENDIRQEN;
+				if (data->flags & MMC_DATA_READ) {
+					host->sg_cpu[j].dsadr =
+						host->res->start + MMC_RXFIFO;
+					host->sg_cpu[j].dtadr =
+						host->bounce_dma;
+				} else {
+					memcpy(host->bounce_cpu, X, offset);
+					host->sg_cpu[j].dsadr =
+						host->bounce_dma;
+					host->sg_cpu[j].dtadr =
+						host->res->start + MMC_TXFIFO;
+				}
+				host->offset = offset;
+				host->dma_len++;
+				length -= offset;
+				addr += offset;
+				j++;
+			} else {
+				/* PXA27x, PXA3xx can handle unaligned DMA */
+				dalgn = 1;
+			}
+		}
+		host->sg_cpu[j].dcmd = dcmd | length;
+		if (length & 31 && !(data->flags & MMC_DATA_READ))
+			host->sg_cpu[j].dcmd |= DCMD_ENDIRQEN;
 		if (data->flags & MMC_DATA_READ) {
-			host->sg_cpu[i].dsadr = host->res->start + MMC_RXFIFO;
-			host->sg_cpu[i].dtadr = sg_dma_address(&data->sg[i]);
+			host->sg_cpu[j].dsadr = host->res->start + MMC_RXFIFO;
+			host->sg_cpu[j].dtadr = addr;
 		} else {
-			host->sg_cpu[i].dsadr = sg_dma_address(&data->sg[i]);
-			host->sg_cpu[i].dtadr = host->res->start + MMC_TXFIFO;
+			host->sg_cpu[j].dsadr = addr;
+			host->sg_cpu[j].dtadr = host->res->start + MMC_TXFIFO;
 		}
-		host->sg_cpu[i].ddadr = host->sg_dma + (i + 1) *
+		host->sg_cpu[j].ddadr = host->sg_dma + (j + 1) *
 					sizeof(struct pxa_dma_desc);
 	}
 	host->sg_cpu[host->dma_len - 1].ddadr = DDADR_STOP;
@@ -174,10 +216,12 @@ static void pxamci_setup_data(struct pxamci_host *host, struct mmc_data *data)
 	 * unaligned (to 8-byte boundaries) data, so switch on byte alignment
 	 * mode only if we have unaligned data.
 	 */
-	if (dalgn)
-		DALGN |= (1 << host->dma);
-	else
-		DALGN &= ~(1 << host->dma);
+	if (!cpu_is_pxa21x() && !cpu_is_pxa25x()) {
+		if (dalgn)
+			DALGN |= (1 << host->dma);
+		else
+			DALGN &= ~(1 << host->dma);
+	}
 	DDADR(host->dma) = host->sg_dma;
 	DCSR(host->dma) = DCSR_RUN;
 }
@@ -281,6 +325,12 @@ static int pxamci_data_done(struct pxamci_host *host, unsigned int stat)
 	if (!data)
 		return 0;
 
+	if ((cpu_is_pxa21x() || cpu_is_pxa25x())
+			&& host->offset && data->flags & MMC_DATA_READ) {
+		memcpy(XX, host->bounce_cpu, host->offset);
+		host->offset = 0;
+	}
+
 	DCSR(host->dma) = 0;
 	dma_unmap_sg(mmc_dev(host->mmc), data->sg, host->dma_len,
 		     host->dma_dir);
@@ -562,11 +612,20 @@ static int pxamci_probe(struct platform_device *pdev)
 				     MMC_CAP_SD_HIGHSPEED;
 	}
 
-	host->sg_cpu = dma_alloc_coherent(&pdev->dev, PAGE_SIZE, &host->sg_dma, GFP_KERNEL);
+	host->sg_cpu = dma_alloc_coherent(&pdev->dev, PAGE_SIZE,
+						&host->sg_dma, GFP_KERNEL);
 	if (!host->sg_cpu) {
 		ret = -ENOMEM;
 		goto out;
 	}
+	if (cpu_is_pxa25x()) {
+		host->bounce = dma_alloc_coherent(&pdev->dev, PAGE_SIZE,
+						&host->aligned, GFP_KERNEL);
+		if (!host->bounce) {
+			ret = -ENOMEM;
+			goto out;
+		}
+	}
 
 	spin_lock_init(&host->lock);
 	host->res = r;
@@ -629,7 +688,11 @@ static int pxamci_probe(struct platform_device *pdev)
 		if (host->base)
 			iounmap(host->base);
 		if (host->sg_cpu)
-			dma_free_coherent(&pdev->dev, PAGE_SIZE, host->sg_cpu, host->sg_dma);
+			dma_free_coherent(&pdev->dev, PAGE_SIZE,
+						host->sg_cpu, host->sg_dma);
+		if (cpu_is_pxa25x() && host->bounce_cpu)
+			dma_free_coherent(&pdev->dev, PAGE_SIZE,
+					host->bounce_cpu, host->bounce_dma);
 		if (host->clk)
 			clk_put(host->clk);
 	}
@@ -665,6 +728,9 @@ static int pxamci_remove(struct platform_device *pdev)
 		pxa_free_dma(host->dma);
 		iounmap(host->base);
 		dma_free_coherent(&pdev->dev, PAGE_SIZE, host->sg_cpu, host->sg_dma);
+		if (cpu_is_pxa25x())
+			dma_free_coherent(&pdev->dev, PAGE_SIZE,
+					host->bounce_cpu, host->bounce_dma);
 
 		clk_put(host->clk);