[PATCH v2 0/3] mmc: rtsx: fix bug and support nonblocking request

[PATCH v2 0/3] mmc: rtsx: fix bug and support nonblocking request

[micky_ching]

From: Micky Ching <micky_ching@realsil.com.cn>

First we fix the card poweroff bug: the card power is not shutdown when sd/mmc
card removed, this will make UHS-card failed to running in high speed mode if we
insert the card again.

We offer a concise tuning searching method, it is much easier to read.

At last we add support for nonblocking request, which will improve card read
speed at about 10% write speed about 5% normally.

Micky Ching (3):
  mmc: rtsx: fix card poweroff bug
  mmc: rtsx: modify phase searching method for tuning
  mmc: rtsx: add support for pre_req and post_req

 drivers/mfd/rtsx_pcr.c            |  132 ++++++---
 drivers/mmc/host/rtsx_pci_sdmmc.c |  529 ++++++++++++++++++++++++++-----------
 include/linux/mfd/rtsx_common.h   |    1 +
 include/linux/mfd/rtsx_pci.h      |    8 +-
 4 files changed, 481 insertions(+), 189 deletions(-)

--
1.7.9.5

[PATCH v2 0/3] mmc: rtsx: fix bug and support nonblocking request

[micky_ching]

From: Micky Ching <micky_ching@realsil.com.cn>

First we fix the card poweroff bug: the card power is not shutdown when sd/mmc
card removed, this will make UHS-card failed to running in high speed mode if we
insert the card again.

We offer a concise tuning searching method, it is much easier to read.

At last we add support for nonblocking request, which will improve card read
speed at about 10% write speed about 5% normally.

Micky Ching (3):
  mmc: rtsx: fix card poweroff bug
  mmc: rtsx: modify phase searching method for tuning
  mmc: rtsx: add support for pre_req and post_req

 drivers/mfd/rtsx_pcr.c            |  132 ++++++---
 drivers/mmc/host/rtsx_pci_sdmmc.c |  529 ++++++++++++++++++++++++++-----------
 include/linux/mfd/rtsx_common.h   |    1 +
 include/linux/mfd/rtsx_pci.h      |    8 +-
 4 files changed, 481 insertions(+), 189 deletions(-)

--
1.7.9.5

[PATCH v2 1/3] mmc: rtsx: fix card poweroff bug

[micky_ching]

From: Micky Ching <micky_ching@realsil.com.cn>

If the host driver removed while card in the slot, the host will not
power off card power correctly. This bug is produced because host
eject flag set before the last mmc_set_ios callback, we should set the
eject flag after power off.

Signed-off-by: Micky Ching <micky_ching@realsil.com.cn>
---
 drivers/mmc/host/rtsx_pci_sdmmc.c |    3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)

diff --git a/drivers/mmc/host/rtsx_pci_sdmmc.c b/drivers/mmc/host/rtsx_pci_sdmmc.c
index c46feda..cc80e31 100644
--- a/drivers/mmc/host/rtsx_pci_sdmmc.c
+++ b/drivers/mmc/host/rtsx_pci_sdmmc.c
@@ -1297,7 +1297,6 @@ static int rtsx_pci_sdmmc_drv_remove(struct platform_device *pdev)
 	pcr->slots[RTSX_SD_CARD].p_dev = NULL;
 	pcr->slots[RTSX_SD_CARD].card_event = NULL;
 	mmc = host->mmc;
-	host->eject = true;
 
 	mutex_lock(&host->host_mutex);
 	if (host->mrq) {
@@ -1315,6 +1314,8 @@ static int rtsx_pci_sdmmc_drv_remove(struct platform_device *pdev)
 	mutex_unlock(&host->host_mutex);
 
 	mmc_remove_host(mmc);
+	host->eject = true;
+
 	mmc_free_host(mmc);
 
 	dev_dbg(&(pdev->dev),
-- 
1.7.9.5

[PATCH v2 1/3] mmc: rtsx: fix card poweroff bug

[micky_ching]

From: Micky Ching <micky_ching@realsil.com.cn>

If the host driver removed while card in the slot, the host will not
power off card power correctly. This bug is produced because host
eject flag set before the last mmc_set_ios callback, we should set the
eject flag after power off.

Signed-off-by: Micky Ching <micky_ching@realsil.com.cn>
---
 drivers/mmc/host/rtsx_pci_sdmmc.c |    3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)

diff --git a/drivers/mmc/host/rtsx_pci_sdmmc.c b/drivers/mmc/host/rtsx_pci_sdmmc.c
index c46feda..cc80e31 100644
--- a/drivers/mmc/host/rtsx_pci_sdmmc.c
+++ b/drivers/mmc/host/rtsx_pci_sdmmc.c
@@ -1297,7 +1297,6 @@ static int rtsx_pci_sdmmc_drv_remove(struct platform_device *pdev)
 	pcr->slots[RTSX_SD_CARD].p_dev = NULL;
 	pcr->slots[RTSX_SD_CARD].card_event = NULL;
 	mmc = host->mmc;
-	host->eject = true;
 
 	mutex_lock(&host->host_mutex);
 	if (host->mrq) {
@@ -1315,6 +1314,8 @@ static int rtsx_pci_sdmmc_drv_remove(struct platform_device *pdev)
 	mutex_unlock(&host->host_mutex);
 
 	mmc_remove_host(mmc);
+	host->eject = true;
+
 	mmc_free_host(mmc);
 
 	dev_dbg(&(pdev->dev),
-- 
1.7.9.5

[PATCH v2 2/3] mmc: rtsx: modify phase searching method for tuning

[micky_ching]

From: Micky Ching <micky_ching@realsil.com.cn>

The new phase searching method is more concise and easier to
understand.

Signed-off-by: Micky Ching <micky_ching@realsil.com.cn>
---
 drivers/mmc/host/rtsx_pci_sdmmc.c |  112 +++++++++++--------------------------
 include/linux/mfd/rtsx_pci.h      |    2 +-
 2 files changed, 33 insertions(+), 81 deletions(-)

diff --git a/drivers/mmc/host/rtsx_pci_sdmmc.c b/drivers/mmc/host/rtsx_pci_sdmmc.c
index cc80e31..0b9ded1 100644
--- a/drivers/mmc/host/rtsx_pci_sdmmc.c
+++ b/drivers/mmc/host/rtsx_pci_sdmmc.c
@@ -31,16 +31,6 @@
 #include <linux/mfd/rtsx_pci.h>
 #include <asm/unaligned.h>
 
-/* SD Tuning Data Structure
- * Record continuous timing phase path
- */
-struct timing_phase_path {
-	int start;
-	int end;
-	int mid;
-	int len;
-};
-
 struct realtek_pci_sdmmc {
 	struct platform_device	*pdev;
 	struct rtsx_pcr		*pcr;
@@ -511,85 +501,47 @@ static int sd_change_phase(struct realtek_pci_sdmmc *host,
 	return 0;
 }
 
-static u8 sd_search_final_phase(struct realtek_pci_sdmmc *host, u32 phase_map)
+static inline u32 test_phase_bit(u32 phase_map, unsigned int bit)
 {
-	struct timing_phase_path path[MAX_PHASE + 1];
-	int i, j, cont_path_cnt;
-	int new_block, max_len, final_path_idx;
-	u8 final_phase = 0xFF;
+	bit %= RTSX_PHASE_MAX;
+	return phase_map & (1 << bit);
+}
 
-	/* Parse phase_map, take it as a bit-ring */
-	cont_path_cnt = 0;
-	new_block = 1;
-	j = 0;
-	for (i = 0; i < MAX_PHASE + 1; i++) {
-		if (phase_map & (1 << i)) {
-			if (new_block) {
-				new_block = 0;
-				j = cont_path_cnt++;
-				path[j].start = i;
-				path[j].end = i;
-			} else {
-				path[j].end = i;
-			}
-		} else {
-			new_block = 1;
-			if (cont_path_cnt) {
-				/* Calculate path length and middle point */
-				int idx = cont_path_cnt - 1;
-				path[idx].len =
-					path[idx].end - path[idx].start + 1;
-				path[idx].mid =
-					path[idx].start + path[idx].len / 2;
-			}
-		}
-	}
+static int sd_get_phase_len(u32 phase_map, unsigned int start_bit)
+{
+	int i;
 
-	if (cont_path_cnt == 0) {
-		dev_dbg(sdmmc_dev(host), "No continuous phase path\n");
-		goto finish;
-	} else {
-		/* Calculate last continuous path length and middle point */
-		int idx = cont_path_cnt - 1;
-		path[idx].len = path[idx].end - path[idx].start + 1;
-		path[idx].mid = path[idx].start + path[idx].len / 2;
+	for (i = 0; i < RTSX_PHASE_MAX; i++) {
+		if (test_phase_bit(phase_map, start_bit + i) == 0)
+			return i;
 	}
+	return RTSX_PHASE_MAX;
+}
+
+static u8 sd_search_final_phase(struct realtek_pci_sdmmc *host, u32 phase_map)
+{
+	int start = 0, len = 0;
+	int start_final = 0, len_final = 0;
+	u8 final_phase = 0xFF;
 
-	/* Connect the first and last continuous paths if they are adjacent */
-	if (!path[0].start && (path[cont_path_cnt - 1].end == MAX_PHASE)) {
-		/* Using negative index */
-		path[0].start = path[cont_path_cnt - 1].start - MAX_PHASE - 1;
-		path[0].len += path[cont_path_cnt - 1].len;
-		path[0].mid = path[0].start + path[0].len / 2;
-		/* Convert negative middle point index to positive one */
-		if (path[0].mid < 0)
-			path[0].mid += MAX_PHASE + 1;
-		cont_path_cnt--;
+	if (phase_map == 0) {
+		dev_err(sdmmc_dev(host), "phase error: [map:%x]\n", phase_map);
+		return final_phase;
 	}
 
-	/* Choose the longest continuous phase path */
-	max_len = 0;
-	final_phase = 0;
-	final_path_idx = 0;
-	for (i = 0; i < cont_path_cnt; i++) {
-		if (path[i].len > max_len) {
-			max_len = path[i].len;
-			final_phase = (u8)path[i].mid;
-			final_path_idx = i;
+	while (start < RTSX_PHASE_MAX) {
+		len = sd_get_phase_len(phase_map, start);
+		if (len_final < len) {
+			start_final = start;
+			len_final = len;
 		}
-
-		dev_dbg(sdmmc_dev(host), "path[%d].start = %d\n",
-				i, path[i].start);
-		dev_dbg(sdmmc_dev(host), "path[%d].end = %d\n",
-				i, path[i].end);
-		dev_dbg(sdmmc_dev(host), "path[%d].len = %d\n",
-				i, path[i].len);
-		dev_dbg(sdmmc_dev(host), "path[%d].mid = %d\n",
-				i, path[i].mid);
+		start += len ? len : 1;
 	}
 
-finish:
-	dev_dbg(sdmmc_dev(host), "Final chosen phase: %d\n", final_phase);
+	final_phase = (start_final + len_final / 2) % RTSX_PHASE_MAX;
+	dev_dbg(sdmmc_dev(host), "phase: [map:%x] [maxlen:%d] [final:%d]\n",
+		phase_map, len_final, final_phase);
+
 	return final_phase;
 }
 
@@ -635,7 +587,7 @@ static int sd_tuning_phase(struct realtek_pci_sdmmc *host,
 	int err, i;
 	u32 raw_phase_map = 0;
 
-	for (i = MAX_PHASE; i >= 0; i--) {
+	for (i = 0; i < RTSX_PHASE_MAX; i++) {
 		err = sd_tuning_rx_cmd(host, opcode, (u8)i);
 		if (err == 0)
 			raw_phase_map |= 1 << i;
diff --git a/include/linux/mfd/rtsx_pci.h b/include/linux/mfd/rtsx_pci.h
index 0ce7721..a383597 100644
--- a/include/linux/mfd/rtsx_pci.h
+++ b/include/linux/mfd/rtsx_pci.h
@@ -144,7 +144,7 @@
 #define HOST_TO_DEVICE		0
 #define DEVICE_TO_HOST		1
 
-#define MAX_PHASE		31
+#define RTSX_PHASE_MAX		32
 #define RX_TUNING_CNT		3
 
 /* SG descriptor */
-- 
1.7.9.5

[PATCH v2 2/3] mmc: rtsx: modify phase searching method for tuning

[micky_ching]

From: Micky Ching <micky_ching@realsil.com.cn>

The new phase searching method is more concise and easier to
understand.

Signed-off-by: Micky Ching <micky_ching@realsil.com.cn>
---
 drivers/mmc/host/rtsx_pci_sdmmc.c |  112 +++++++++++--------------------------
 include/linux/mfd/rtsx_pci.h      |    2 +-
 2 files changed, 33 insertions(+), 81 deletions(-)

diff --git a/drivers/mmc/host/rtsx_pci_sdmmc.c b/drivers/mmc/host/rtsx_pci_sdmmc.c
index cc80e31..0b9ded1 100644
--- a/drivers/mmc/host/rtsx_pci_sdmmc.c
+++ b/drivers/mmc/host/rtsx_pci_sdmmc.c
@@ -31,16 +31,6 @@
 #include <linux/mfd/rtsx_pci.h>
 #include <asm/unaligned.h>
 
-/* SD Tuning Data Structure
- * Record continuous timing phase path
- */
-struct timing_phase_path {
-	int start;
-	int end;
-	int mid;
-	int len;
-};
-
 struct realtek_pci_sdmmc {
 	struct platform_device	*pdev;
 	struct rtsx_pcr		*pcr;
@@ -511,85 +501,47 @@ static int sd_change_phase(struct realtek_pci_sdmmc *host,
 	return 0;
 }
 
-static u8 sd_search_final_phase(struct realtek_pci_sdmmc *host, u32 phase_map)
+static inline u32 test_phase_bit(u32 phase_map, unsigned int bit)
 {
-	struct timing_phase_path path[MAX_PHASE + 1];
-	int i, j, cont_path_cnt;
-	int new_block, max_len, final_path_idx;
-	u8 final_phase = 0xFF;
+	bit %= RTSX_PHASE_MAX;
+	return phase_map & (1 << bit);
+}
 
-	/* Parse phase_map, take it as a bit-ring */
-	cont_path_cnt = 0;
-	new_block = 1;
-	j = 0;
-	for (i = 0; i < MAX_PHASE + 1; i++) {
-		if (phase_map & (1 << i)) {
-			if (new_block) {
-				new_block = 0;
-				j = cont_path_cnt++;
-				path[j].start = i;
-				path[j].end = i;
-			} else {
-				path[j].end = i;
-			}
-		} else {
-			new_block = 1;
-			if (cont_path_cnt) {
-				/* Calculate path length and middle point */
-				int idx = cont_path_cnt - 1;
-				path[idx].len =
-					path[idx].end - path[idx].start + 1;
-				path[idx].mid =
-					path[idx].start + path[idx].len / 2;
-			}
-		}
-	}
+static int sd_get_phase_len(u32 phase_map, unsigned int start_bit)
+{
+	int i;
 
-	if (cont_path_cnt == 0) {
-		dev_dbg(sdmmc_dev(host), "No continuous phase path\n");
-		goto finish;
-	} else {
-		/* Calculate last continuous path length and middle point */
-		int idx = cont_path_cnt - 1;
-		path[idx].len = path[idx].end - path[idx].start + 1;
-		path[idx].mid = path[idx].start + path[idx].len / 2;
+	for (i = 0; i < RTSX_PHASE_MAX; i++) {
+		if (test_phase_bit(phase_map, start_bit + i) == 0)
+			return i;
 	}
+	return RTSX_PHASE_MAX;
+}
+
+static u8 sd_search_final_phase(struct realtek_pci_sdmmc *host, u32 phase_map)
+{
+	int start = 0, len = 0;
+	int start_final = 0, len_final = 0;
+	u8 final_phase = 0xFF;
 
-	/* Connect the first and last continuous paths if they are adjacent */
-	if (!path[0].start && (path[cont_path_cnt - 1].end == MAX_PHASE)) {
-		/* Using negative index */
-		path[0].start = path[cont_path_cnt - 1].start - MAX_PHASE - 1;
-		path[0].len += path[cont_path_cnt - 1].len;
-		path[0].mid = path[0].start + path[0].len / 2;
-		/* Convert negative middle point index to positive one */
-		if (path[0].mid < 0)
-			path[0].mid += MAX_PHASE + 1;
-		cont_path_cnt--;
+	if (phase_map == 0) {
+		dev_err(sdmmc_dev(host), "phase error: [map:%x]\n", phase_map);
+		return final_phase;
 	}
 
-	/* Choose the longest continuous phase path */
-	max_len = 0;
-	final_phase = 0;
-	final_path_idx = 0;
-	for (i = 0; i < cont_path_cnt; i++) {
-		if (path[i].len > max_len) {
-			max_len = path[i].len;
-			final_phase = (u8)path[i].mid;
-			final_path_idx = i;
+	while (start < RTSX_PHASE_MAX) {
+		len = sd_get_phase_len(phase_map, start);
+		if (len_final < len) {
+			start_final = start;
+			len_final = len;
 		}
-
-		dev_dbg(sdmmc_dev(host), "path[%d].start = %d\n",
-				i, path[i].start);
-		dev_dbg(sdmmc_dev(host), "path[%d].end = %d\n",
-				i, path[i].end);
-		dev_dbg(sdmmc_dev(host), "path[%d].len = %d\n",
-				i, path[i].len);
-		dev_dbg(sdmmc_dev(host), "path[%d].mid = %d\n",
-				i, path[i].mid);
+		start += len ? len : 1;
 	}
 
-finish:
-	dev_dbg(sdmmc_dev(host), "Final chosen phase: %d\n", final_phase);
+	final_phase = (start_final + len_final / 2) % RTSX_PHASE_MAX;
+	dev_dbg(sdmmc_dev(host), "phase: [map:%x] [maxlen:%d] [final:%d]\n",
+		phase_map, len_final, final_phase);
+
 	return final_phase;
 }
 
@@ -635,7 +587,7 @@ static int sd_tuning_phase(struct realtek_pci_sdmmc *host,
 	int err, i;
 	u32 raw_phase_map = 0;
 
-	for (i = MAX_PHASE; i >= 0; i--) {
+	for (i = 0; i < RTSX_PHASE_MAX; i++) {
 		err = sd_tuning_rx_cmd(host, opcode, (u8)i);
 		if (err == 0)
 			raw_phase_map |= 1 << i;
diff --git a/include/linux/mfd/rtsx_pci.h b/include/linux/mfd/rtsx_pci.h
index 0ce7721..a383597 100644
--- a/include/linux/mfd/rtsx_pci.h
+++ b/include/linux/mfd/rtsx_pci.h
@@ -144,7 +144,7 @@
 #define HOST_TO_DEVICE		0
 #define DEVICE_TO_HOST		1
 
-#define MAX_PHASE		31
+#define RTSX_PHASE_MAX		32
 #define RX_TUNING_CNT		3
 
 /* SG descriptor */
-- 
1.7.9.5

[PATCH v2 3/3] mmc: rtsx: add support for pre_req and post_req

[micky_ching]

From: Micky Ching <micky_ching@realsil.com.cn>

Add support for non-blocking request, pre_req() runs dma_map_sg() and
post_req() runs dma_unmap_sg(). This patch can increase card read/write
speed, especially for high speed card and slow CPU(for some embedded
platform).

Users can get a great benefit from this patch. if CPU frequency is 800MHz,
SDR104 or DDR50 card read/write speed may increase more than 15%.

test results:
intel i3(800MHz - 2.3GHz), SD card clock 208MHz

performance mode(2.3GHz):
Before:
dd if=/dev/mmcblk0p1 of=/dev/null bs=64k count=1024
67108864 bytes (67 MB) copied, 1.18191 s, 56.8 MB/s
After:
 dd if=/dev/mmcblk0p1 of=/dev/null bs=64k count=1024
67108864 bytes (67 MB) copied, 1.09276 s, 61.4 MB/s

powersave mode(800MHz):
Before:
dd if=/dev/mmcblk0p1 of=/dev/null bs=64k count=1024
67108864 bytes (67 MB) copied, 1.29569 s, 51.8 MB/s
After:
dd if=/dev/mmcblk0p1 of=/dev/null bs=64k count=1024
67108864 bytes (67 MB) copied, 1.11218 s, 60.3 MB/s

Signed-off-by: Micky Ching <micky_ching@realsil.com.cn>
---
 drivers/mfd/rtsx_pcr.c            |  132 ++++++++----
 drivers/mmc/host/rtsx_pci_sdmmc.c |  418 +++++++++++++++++++++++++++++++------
 include/linux/mfd/rtsx_common.h   |    1 +
 include/linux/mfd/rtsx_pci.h      |    6 +
 4 files changed, 448 insertions(+), 109 deletions(-)

diff --git a/drivers/mfd/rtsx_pcr.c b/drivers/mfd/rtsx_pcr.c
index 1d15735..c9de3d5 100644
--- a/drivers/mfd/rtsx_pcr.c
+++ b/drivers/mfd/rtsx_pcr.c
@@ -338,58 +338,28 @@ int rtsx_pci_transfer_data(struct rtsx_pcr *pcr, struct scatterlist *sglist,
 		int num_sg, bool read, int timeout)
 {
 	struct completion trans_done;
-	u8 dir;
-	int err = 0, i, count;
+	int err = 0, count;
 	long timeleft;
 	unsigned long flags;
-	struct scatterlist *sg;
-	enum dma_data_direction dma_dir;
-	u32 val;
-	dma_addr_t addr;
-	unsigned int len;
-
-	dev_dbg(&(pcr->pci->dev), "--> %s: num_sg = %d\n", __func__, num_sg);
-
-	/* don't transfer data during abort processing */
-	if (pcr->remove_pci)
-		return -EINVAL;
-
-	if ((sglist == NULL) || (num_sg <= 0))
-		return -EINVAL;
 
-	if (read) {
-		dir = DEVICE_TO_HOST;
-		dma_dir = DMA_FROM_DEVICE;
-	} else {
-		dir = HOST_TO_DEVICE;
-		dma_dir = DMA_TO_DEVICE;
-	}
-
-	count = dma_map_sg(&(pcr->pci->dev), sglist, num_sg, dma_dir);
+	count = rtsx_pci_dma_map_sg(pcr, sglist, num_sg, read);
 	if (count < 1) {
 		dev_err(&(pcr->pci->dev), "scatterlist map failed\n");
 		return -EINVAL;
 	}
 	dev_dbg(&(pcr->pci->dev), "DMA mapping count: %d\n", count);
 
-	val = ((u32)(dir & 0x01) << 29) | TRIG_DMA | ADMA_MODE;
-	pcr->sgi = 0;
-	for_each_sg(sglist, sg, count, i) {
-		addr = sg_dma_address(sg);
-		len = sg_dma_len(sg);
-		rtsx_pci_add_sg_tbl(pcr, addr, len, i == count - 1);
-	}
 
 	spin_lock_irqsave(&pcr->lock, flags);
 
 	pcr->done = &trans_done;
 	pcr->trans_result = TRANS_NOT_READY;
 	init_completion(&trans_done);
-	rtsx_pci_writel(pcr, RTSX_HDBAR, pcr->host_sg_tbl_addr);
-	rtsx_pci_writel(pcr, RTSX_HDBCTLR, val);
 
 	spin_unlock_irqrestore(&pcr->lock, flags);
 
+	rtsx_pci_dma_transfer(pcr, sglist, count, read);
+
 	timeleft = wait_for_completion_interruptible_timeout(
 			&trans_done, msecs_to_jiffies(timeout));
 	if (timeleft <= 0) {
@@ -413,7 +383,7 @@ out:
 	pcr->done = NULL;
 	spin_unlock_irqrestore(&pcr->lock, flags);
 
-	dma_unmap_sg(&(pcr->pci->dev), sglist, num_sg, dma_dir);
+	rtsx_pci_dma_unmap_sg(pcr, sglist, num_sg, read);
 
 	if ((err < 0) && (err != -ENODEV))
 		rtsx_pci_stop_cmd(pcr);
@@ -425,6 +395,73 @@ out:
 }
 EXPORT_SYMBOL_GPL(rtsx_pci_transfer_data);
 
+int rtsx_pci_dma_map_sg(struct rtsx_pcr *pcr, struct scatterlist *sglist,
+		int num_sg, bool read)
+{
+	enum dma_data_direction dir = read ? DMA_FROM_DEVICE : DMA_TO_DEVICE;
+
+	if (pcr->remove_pci)
+		return -EINVAL;
+
+	if ((sglist == NULL) || num_sg < 1)
+		return -EINVAL;
+
+	return dma_map_sg(&(pcr->pci->dev), sglist, num_sg, dir);
+}
+EXPORT_SYMBOL_GPL(rtsx_pci_dma_map_sg);
+
+int rtsx_pci_dma_unmap_sg(struct rtsx_pcr *pcr, struct scatterlist *sglist,
+		int num_sg, bool read)
+{
+	enum dma_data_direction dir = read ? DMA_FROM_DEVICE : DMA_TO_DEVICE;
+
+	if (pcr->remove_pci)
+		return -EINVAL;
+
+	if (sglist == NULL || num_sg < 1)
+		return -EINVAL;
+
+	dma_unmap_sg(&(pcr->pci->dev), sglist, num_sg, dir);
+	return num_sg;
+}
+EXPORT_SYMBOL_GPL(rtsx_pci_dma_unmap_sg);
+
+int rtsx_pci_dma_transfer(struct rtsx_pcr *pcr, struct scatterlist *sglist,
+		int sg_count, bool read)
+{
+	struct scatterlist *sg;
+	dma_addr_t addr;
+	unsigned int len;
+	int i;
+	u32 val;
+	u8 dir = read ? DEVICE_TO_HOST : HOST_TO_DEVICE;
+	unsigned long flags;
+
+	if (pcr->remove_pci)
+		return -EINVAL;
+
+	if ((sglist == NULL) || (sg_count < 1))
+		return -EINVAL;
+
+	val = ((u32)(dir & 0x01) << 29) | TRIG_DMA | ADMA_MODE;
+	pcr->sgi = 0;
+	for_each_sg(sglist, sg, sg_count, i) {
+		addr = sg_dma_address(sg);
+		len = sg_dma_len(sg);
+		rtsx_pci_add_sg_tbl(pcr, addr, len, i == sg_count - 1);
+	}
+
+	spin_lock_irqsave(&pcr->lock, flags);
+
+	rtsx_pci_writel(pcr, RTSX_HDBAR, pcr->host_sg_tbl_addr);
+	rtsx_pci_writel(pcr, RTSX_HDBCTLR, val);
+
+	spin_unlock_irqrestore(&pcr->lock, flags);
+
+	return 0;
+}
+EXPORT_SYMBOL_GPL(rtsx_pci_dma_transfer);
+
 int rtsx_pci_read_ppbuf(struct rtsx_pcr *pcr, u8 *buf, int buf_len)
 {
 	int err;
@@ -836,6 +873,8 @@ static irqreturn_t rtsx_pci_isr(int irq, void *dev_id)
 	int_reg = rtsx_pci_readl(pcr, RTSX_BIPR);
 	/* Clear interrupt flag */
 	rtsx_pci_writel(pcr, RTSX_BIPR, int_reg);
+	dev_dbg(&pcr->pci->dev, "=========== BIPR 0x%8x ==========\n", int_reg);
+
 	if ((int_reg & pcr->bier) == 0) {
 		spin_unlock(&pcr->lock);
 		return IRQ_NONE;
@@ -866,17 +905,28 @@ static irqreturn_t rtsx_pci_isr(int irq, void *dev_id)
 	}
 
 	if (int_reg & (NEED_COMPLETE_INT | DELINK_INT)) {
-		if (int_reg & (TRANS_FAIL_INT | DELINK_INT)) {
+		if (int_reg & (TRANS_FAIL_INT | DELINK_INT))
 			pcr->trans_result = TRANS_RESULT_FAIL;
-			if (pcr->done)
-				complete(pcr->done);
-		} else if (int_reg & TRANS_OK_INT) {
+		else if (int_reg & TRANS_OK_INT)
 			pcr->trans_result = TRANS_RESULT_OK;
-			if (pcr->done)
-				complete(pcr->done);
+
+		if (pcr->done)
+			complete(pcr->done);
+
+		if (int_reg & SD_EXIST) {
+			struct rtsx_slot *slot = &pcr->slots[RTSX_SD_CARD];
+			if (slot && slot->done_transfer)
+				slot->done_transfer(slot->p_dev);
+		}
+
+		if (int_reg & MS_EXIST) {
+			struct rtsx_slot *slot = &pcr->slots[RTSX_SD_CARD];
+			if (slot && slot->done_transfer)
+				slot->done_transfer(slot->p_dev);
 		}
 	}
 
+
 	if (pcr->card_inserted || pcr->card_removed)
 		schedule_delayed_work(&pcr->carddet_work,
 				msecs_to_jiffies(200));
diff --git a/drivers/mmc/host/rtsx_pci_sdmmc.c b/drivers/mmc/host/rtsx_pci_sdmmc.c
index 0b9ded1..5fb994f 100644
--- a/drivers/mmc/host/rtsx_pci_sdmmc.c
+++ b/drivers/mmc/host/rtsx_pci_sdmmc.c
@@ -31,14 +31,28 @@
 #include <linux/mfd/rtsx_pci.h>
 #include <asm/unaligned.h>
 
+struct realtek_next {
+	unsigned int	sg_count;
+	s32		cookie;
+};
+
 struct realtek_pci_sdmmc {
 	struct platform_device	*pdev;
 	struct rtsx_pcr		*pcr;
 	struct mmc_host		*mmc;
 	struct mmc_request	*mrq;
-
-	struct mutex		host_mutex;
-
+	struct mmc_command	*cmd;
+	struct mmc_data		*data;
+
+	spinlock_t		lock;
+	struct timer_list	timer;
+	struct tasklet_struct	cmd_tasklet;
+	struct tasklet_struct	data_tasklet;
+	struct tasklet_struct	finish_tasklet;
+
+	u8			rsp_type;
+	u8			rsp_len;
+	int			sg_count;
 	u8			ssc_depth;
 	unsigned int		clock;
 	bool			vpclk;
@@ -48,8 +62,13 @@ struct realtek_pci_sdmmc {
 	int			power_state;
 #define SDMMC_POWER_ON		1
 #define SDMMC_POWER_OFF		0
+
+	struct realtek_next	next_data;
 };
 
+static int sd_start_multi_rw(struct realtek_pci_sdmmc *host,
+		struct mmc_request *mrq);
+
 static inline struct device *sdmmc_dev(struct realtek_pci_sdmmc *host)
 {
 	return &(host->pdev->dev);
@@ -86,6 +105,95 @@ static void sd_print_debug_regs(struct realtek_pci_sdmmc *host)
 #define sd_print_debug_regs(host)
 #endif /* DEBUG */
 
+static void sd_isr_done_transfer(struct platform_device *pdev)
+{
+	struct realtek_pci_sdmmc *host = platform_get_drvdata(pdev);
+
+	spin_lock(&host->lock);
+	if (host->cmd)
+		tasklet_schedule(&host->cmd_tasklet);
+	if (host->data)
+		tasklet_schedule(&host->data_tasklet);
+	spin_unlock(&host->lock);
+}
+
+static void sd_request_timeout(unsigned long host_addr)
+{
+	struct realtek_pci_sdmmc *host = (struct realtek_pci_sdmmc *)host_addr;
+	unsigned long flags;
+
+	spin_lock_irqsave(&host->lock, flags);
+
+	if (!host->mrq) {
+		dev_err(sdmmc_dev(host), "error: no request exist\n");
+		goto out;
+	}
+
+	if (host->cmd)
+		host->cmd->error = -ETIMEDOUT;
+	if (host->data)
+		host->data->error = -ETIMEDOUT;
+
+	dev_dbg(sdmmc_dev(host), "timeout for request\n");
+
+out:
+	tasklet_schedule(&host->finish_tasklet);
+	spin_unlock_irqrestore(&host->lock, flags);
+}
+
+static void sd_finish_request(unsigned long host_addr)
+{
+	struct realtek_pci_sdmmc *host = (struct realtek_pci_sdmmc *)host_addr;
+	struct rtsx_pcr *pcr = host->pcr;
+	struct mmc_request *mrq;
+	struct mmc_command *cmd;
+	struct mmc_data *data;
+	unsigned long flags;
+	bool any_error;
+
+	spin_lock_irqsave(&host->lock, flags);
+
+	del_timer(&host->timer);
+	mrq = host->mrq;
+	if (!mrq) {
+		dev_err(sdmmc_dev(host), "error: no request need finish\n");
+		goto out;
+	}
+
+	cmd = mrq->cmd;
+	data = mrq->data;
+
+	any_error = (mrq->sbc && mrq->sbc->error) ||
+		(mrq->stop && mrq->stop->error) ||
+		(cmd && cmd->error) || (data && data->error);
+
+	if (any_error) {
+		rtsx_pci_stop_cmd(pcr);
+		sd_clear_error(host);
+	}
+
+	if (data) {
+		if (any_error)
+			data->bytes_xfered = 0;
+		else
+			data->bytes_xfered = data->blocks * data->blksz;
+
+		if (!data->host_cookie)
+			rtsx_pci_dma_unmap_sg(pcr, data->sg, data->sg_len,
+					data->flags & MMC_DATA_READ);
+
+	}
+
+	host->mrq = NULL;
+	host->cmd = NULL;
+	host->data = NULL;
+
+out:
+	spin_unlock_irqrestore(&host->lock, flags);
+	mutex_unlock(&pcr->pcr_mutex);
+	mmc_request_done(host->mmc, mrq);
+}
+
 static int sd_read_data(struct realtek_pci_sdmmc *host, u8 *cmd, u16 byte_cnt,
 		u8 *buf, int buf_len, int timeout)
 {
@@ -203,8 +311,7 @@ static int sd_write_data(struct realtek_pci_sdmmc *host, u8 *cmd, u16 byte_cnt,
 	return 0;
 }
 
-static void sd_send_cmd_get_rsp(struct realtek_pci_sdmmc *host,
-		struct mmc_command *cmd)
+static void sd_send_cmd(struct realtek_pci_sdmmc *host, struct mmc_command *cmd)
 {
 	struct rtsx_pcr *pcr = host->pcr;
 	u8 cmd_idx = (u8)cmd->opcode;
@@ -212,11 +319,14 @@ static void sd_send_cmd_get_rsp(struct realtek_pci_sdmmc *host,
 	int err = 0;
 	int timeout = 100;
 	int i;
-	u8 *ptr;
-	int stat_idx = 0;
 	u8 rsp_type;
 	int rsp_len = 5;
-	bool clock_toggled = false;
+	unsigned long flags;
+
+	if (host->cmd)
+		dev_err(sdmmc_dev(host), "error: cmd already exist\n");
+
+	host->cmd = cmd;
 
 	dev_dbg(sdmmc_dev(host), "%s: SD/MMC CMD %d, arg = 0x%08x\n",
 			__func__, cmd_idx, arg);
@@ -251,6 +361,8 @@ static void sd_send_cmd_get_rsp(struct realtek_pci_sdmmc *host,
 		err = -EINVAL;
 		goto out;
 	}
+	host->rsp_type = rsp_type;
+	host->rsp_len = rsp_len;
 
 	if (rsp_type == SD_RSP_TYPE_R1b)
 		timeout = 3000;
@@ -260,8 +372,6 @@ static void sd_send_cmd_get_rsp(struct realtek_pci_sdmmc *host,
 				0xFF, SD_CLK_TOGGLE_EN);
 		if (err < 0)
 			goto out;
-
-		clock_toggled = true;
 	}
 
 	rtsx_pci_init_cmd(pcr);
@@ -285,25 +395,60 @@ static void sd_send_cmd_get_rsp(struct realtek_pci_sdmmc *host,
 		/* Read data from ping-pong buffer */
 		for (i = PPBUF_BASE2; i < PPBUF_BASE2 + 16; i++)
 			rtsx_pci_add_cmd(pcr, READ_REG_CMD, (u16)i, 0, 0);
-		stat_idx = 16;
 	} else if (rsp_type != SD_RSP_TYPE_R0) {
 		/* Read data from SD_CMDx registers */
 		for (i = SD_CMD0; i <= SD_CMD4; i++)
 			rtsx_pci_add_cmd(pcr, READ_REG_CMD, (u16)i, 0, 0);
-		stat_idx = 5;
 	}
 
 	rtsx_pci_add_cmd(pcr, READ_REG_CMD, SD_STAT1, 0, 0);
 
-	err = rtsx_pci_send_cmd(pcr, timeout);
-	if (err < 0) {
-		sd_print_debug_regs(host);
-		sd_clear_error(host);
-		dev_dbg(sdmmc_dev(host),
-			"rtsx_pci_send_cmd error (err = %d)\n", err);
+	mod_timer(&host->timer, jiffies + msecs_to_jiffies(timeout));
+
+	spin_lock_irqsave(&pcr->lock, flags);
+	pcr->trans_result = TRANS_NOT_READY;
+	rtsx_pci_send_cmd_no_wait(pcr);
+	spin_unlock_irqrestore(&pcr->lock, flags);
+
+	return;
+
+out:
+	cmd->error = err;
+	tasklet_schedule(&host->finish_tasklet);
+}
+
+static void sd_get_rsp(unsigned long host_addr)
+{
+	struct realtek_pci_sdmmc *host = (struct realtek_pci_sdmmc *)host_addr;
+	struct rtsx_pcr *pcr = host->pcr;
+	struct mmc_command *cmd;
+	int i, err = 0, stat_idx;
+	u8 *ptr, rsp_type;
+	unsigned long flags;
+
+	spin_lock_irqsave(&host->lock, flags);
+
+	cmd = host->cmd;
+	host->cmd = NULL;
+
+	if (!cmd) {
+		dev_err(sdmmc_dev(host), "error: cmd not exist\n");
 		goto out;
 	}
 
+	spin_lock(&pcr->lock);
+	if (pcr->trans_result == TRANS_NO_DEVICE)
+		err = -ENODEV;
+	else if (pcr->trans_result != TRANS_RESULT_OK)
+		err = -EINVAL;
+	spin_unlock(&pcr->lock);
+
+	if (err < 0)
+		goto out;
+
+	rsp_type = host->rsp_type;
+	stat_idx = host->rsp_len;
+
 	if (rsp_type == SD_RSP_TYPE_R0) {
 		err = 0;
 		goto out;
@@ -340,26 +485,106 @@ static void sd_send_cmd_get_rsp(struct realtek_pci_sdmmc *host,
 				cmd->resp[0]);
 	}
 
+	if (cmd == host->mrq->sbc) {
+		sd_send_cmd(host, host->mrq->cmd);
+		spin_unlock_irqrestore(&host->lock, flags);
+		return;
+	}
+
+	if (cmd == host->mrq->stop)
+		goto out;
+
+	if (cmd->data) {
+		sd_start_multi_rw(host, host->mrq);
+		spin_unlock_irqrestore(&host->lock, flags);
+		return;
+	}
+
 out:
 	cmd->error = err;
 
-	if (err && clock_toggled)
-		rtsx_pci_write_register(pcr, SD_BUS_STAT,
-				SD_CLK_TOGGLE_EN | SD_CLK_FORCE_STOP, 0);
+	tasklet_schedule(&host->finish_tasklet);
+	spin_unlock_irqrestore(&host->lock, flags);
+}
+
+static int sd_pre_dma_transfer(struct realtek_pci_sdmmc *host,
+			struct mmc_data *data, struct realtek_next *next)
+{
+	struct rtsx_pcr *pcr = host->pcr;
+	int read = data->flags & MMC_DATA_READ;
+	int sg_count = 0;
+
+	if (!next && data->host_cookie &&
+		data->host_cookie != host->next_data.cookie) {
+		dev_err(sdmmc_dev(host),
+			"error: invalid cookie data[%d] host[%d]\n",
+			data->host_cookie, host->next_data.cookie);
+		data->host_cookie = 0;
+	}
+
+	if (next || (!next && data->host_cookie != host->next_data.cookie))
+		sg_count = rtsx_pci_dma_map_sg(pcr,
+				data->sg, data->sg_len, read);
+	else
+		sg_count = host->next_data.sg_count;
+
+	if (next) {
+		next->sg_count = sg_count;
+		if (++next->cookie < 0)
+			next->cookie = 1;
+		data->host_cookie = next->cookie;
+	}
+
+	return sg_count;
+}
+
+static void sdmmc_pre_req(struct mmc_host *mmc, struct mmc_request *mrq,
+		bool is_first_req)
+{
+	struct realtek_pci_sdmmc *host = mmc_priv(mmc);
+	struct mmc_data *data = mrq->data;
+
+	if (data->host_cookie) {
+		dev_err(sdmmc_dev(host),
+			"error: descard already cookie data[%d]\n",
+			data->host_cookie);
+		data->host_cookie = 0;
+	}
+
+	dev_dbg(sdmmc_dev(host), "dma sg prepared: %d\n",
+		sd_pre_dma_transfer(host, data, &host->next_data));
+}
+
+static void sdmmc_post_req(struct mmc_host *mmc, struct mmc_request *mrq,
+		int err)
+{
+	struct realtek_pci_sdmmc *host = mmc_priv(mmc);
+	struct rtsx_pcr *pcr = host->pcr;
+	struct mmc_data *data = mrq->data;
+	int read = data->flags & MMC_DATA_READ;
+
+	rtsx_pci_dma_unmap_sg(pcr, data->sg, data->sg_len, read);
+	data->host_cookie = 0;
 }
 
-static int sd_rw_multi(struct realtek_pci_sdmmc *host, struct mmc_request *mrq)
+static int sd_start_multi_rw(struct realtek_pci_sdmmc *host,
+		struct mmc_request *mrq)
 {
 	struct rtsx_pcr *pcr = host->pcr;
 	struct mmc_host *mmc = host->mmc;
 	struct mmc_card *card = mmc->card;
 	struct mmc_data *data = mrq->data;
 	int uhs = mmc_card_uhs(card);
-	int read = (data->flags & MMC_DATA_READ) ? 1 : 0;
+	int read = data->flags & MMC_DATA_READ;
 	u8 cfg2, trans_mode;
 	int err;
 	size_t data_len = data->blksz * data->blocks;
 
+	if (host->data)
+		dev_err(sdmmc_dev(host), "error: data already exist\n");
+
+	host->data = data;
+
 	if (read) {
 		cfg2 = SD_CALCULATE_CRC7 | SD_CHECK_CRC16 |
 			SD_NO_WAIT_BUSY_END | SD_CHECK_CRC7 | SD_RSP_LEN_0;
@@ -410,17 +635,56 @@ static int sd_rw_multi(struct realtek_pci_sdmmc *host, struct mmc_request *mrq)
 	rtsx_pci_add_cmd(pcr, CHECK_REG_CMD, SD_TRANSFER,
 			SD_TRANSFER_END, SD_TRANSFER_END);
 
+	mod_timer(&host->timer, jiffies + 10 * HZ);
 	rtsx_pci_send_cmd_no_wait(pcr);
 
-	err = rtsx_pci_transfer_data(pcr, data->sg, data->sg_len, read, 10000);
+	err = rtsx_pci_dma_transfer(pcr, data->sg, host->sg_count, read);
 	if (err < 0) {
-		sd_clear_error(host);
-		return err;
+		data->error = err;
+		tasklet_schedule(&host->finish_tasklet);
 	}
-
 	return 0;
 }
 
+static void sd_finish_multi_rw(unsigned long host_addr)
+{
+	struct realtek_pci_sdmmc *host = (struct realtek_pci_sdmmc *)host_addr;
+	struct rtsx_pcr *pcr = host->pcr;
+	struct mmc_data *data;
+	int err = 0;
+	unsigned long flags;
+
+	spin_lock_irqsave(&host->lock, flags);
+
+	if (!host->data) {
+		dev_err(sdmmc_dev(host), "error: no data exist\n");
+		goto out;
+	}
+
+	data = host->data;
+	host->data = NULL;
+
+	if (pcr->trans_result == TRANS_NO_DEVICE)
+		err = -ENODEV;
+	else if (pcr->trans_result != TRANS_RESULT_OK)
+		err = -EINVAL;
+
+	if (err < 0) {
+		data->error = err;
+		goto out;
+	}
+
+	if (!host->mrq->sbc && data->stop) {
+		sd_send_cmd(host, data->stop);
+		spin_unlock_irqrestore(&host->lock, flags);
+		return;
+	}
+
+out:
+	tasklet_schedule(&host->finish_tasklet);
+	spin_unlock_irqrestore(&host->lock, flags);
+}
+
 static inline void sd_enable_initial_mode(struct realtek_pci_sdmmc *host)
 {
 	rtsx_pci_write_register(host->pcr, SD_CFG1,
@@ -637,6 +901,13 @@ static int sd_tuning_rx(struct realtek_pci_sdmmc *host, u8 opcode)
 	return 0;
 }
 
+static inline bool sd_use_muti_rw(struct mmc_command *cmd)
+{
+	return mmc_op_multi(cmd->opcode) ||
+		(cmd->opcode == MMC_READ_SINGLE_BLOCK) ||
+		(cmd->opcode == MMC_WRITE_BLOCK);
+}
+
 static void sdmmc_request(struct mmc_host *mmc, struct mmc_request *mrq)
 {
 	struct realtek_pci_sdmmc *host = mmc_priv(mmc);
@@ -645,6 +916,14 @@ static void sdmmc_request(struct mmc_host *mmc, struct mmc_request *mrq)
 	struct mmc_data *data = mrq->data;
 	unsigned int data_size = 0;
 	int err;
+	unsigned long flags;
+
+	mutex_lock(&pcr->pcr_mutex);
+	spin_lock_irqsave(&host->lock, flags);
+
+	if (host->mrq)
+		dev_err(sdmmc_dev(host), "error: request already exist\n");
+	host->mrq = mrq;
 
 	if (host->eject) {
 		cmd->error = -ENOMEDIUM;
@@ -657,8 +936,6 @@ static void sdmmc_request(struct mmc_host *mmc, struct mmc_request *mrq)
 		goto finish;
 	}
 
-	mutex_lock(&pcr->pcr_mutex);
-
 	rtsx_pci_start_run(pcr);
 
 	rtsx_pci_switch_clock(pcr, host->clock, host->ssc_depth,
@@ -667,46 +944,28 @@ static void sdmmc_request(struct mmc_host *mmc, struct mmc_request *mrq)
 	rtsx_pci_write_register(pcr, CARD_SHARE_MODE,
 			CARD_SHARE_MASK, CARD_SHARE_48_SD);
 
-	mutex_lock(&host->host_mutex);
-	host->mrq = mrq;
-	mutex_unlock(&host->host_mutex);
-
 	if (mrq->data)
 		data_size = data->blocks * data->blksz;
 
-	if (!data_size || mmc_op_multi(cmd->opcode) ||
-			(cmd->opcode == MMC_READ_SINGLE_BLOCK) ||
-			(cmd->opcode == MMC_WRITE_BLOCK)) {
-		sd_send_cmd_get_rsp(host, cmd);
-
-		if (!cmd->error && data_size) {
-			sd_rw_multi(host, mrq);
+	if (sd_use_muti_rw(cmd))
+		host->sg_count = sd_pre_dma_transfer(host, data, NULL);
 
-			if (mmc_op_multi(cmd->opcode) && mrq->stop)
-				sd_send_cmd_get_rsp(host, mrq->stop);
-		}
+	if (!data_size || sd_use_muti_rw(cmd)) {
+		if (mrq->sbc)
+			sd_send_cmd(host, mrq->sbc);
+		else
+			sd_send_cmd(host, cmd);
+		spin_unlock_irqrestore(&host->lock, flags);
 	} else {
+		spin_unlock_irqrestore(&host->lock, flags);
 		sd_normal_rw(host, mrq);
+		tasklet_schedule(&host->finish_tasklet);
 	}
-
-	if (mrq->data) {
-		if (cmd->error || data->error)
-			data->bytes_xfered = 0;
-		else
-			data->bytes_xfered = data->blocks * data->blksz;
-	}
-
-	mutex_unlock(&pcr->pcr_mutex);
+	return;
 
 finish:
-	if (cmd->error)
-		dev_dbg(sdmmc_dev(host), "cmd->error = %d\n", cmd->error);
-
-	mutex_lock(&host->host_mutex);
-	host->mrq = NULL;
-	mutex_unlock(&host->host_mutex);
-
-	mmc_request_done(mmc, mrq);
+	tasklet_schedule(&host->finish_tasklet);
+	spin_unlock_irqrestore(&host->lock, flags);
 }
 
 static int sd_set_bus_width(struct realtek_pci_sdmmc *host,
@@ -1141,6 +1400,8 @@ out:
 }
 
 static const struct mmc_host_ops realtek_pci_sdmmc_ops = {
+	.pre_req = sdmmc_pre_req,
+	.post_req = sdmmc_post_req,
 	.request = sdmmc_request,
 	.set_ios = sdmmc_set_ios,
 	.get_ro = sdmmc_get_ro,
@@ -1204,6 +1465,7 @@ static int rtsx_pci_sdmmc_drv_probe(struct platform_device *pdev)
 	struct realtek_pci_sdmmc *host;
 	struct rtsx_pcr *pcr;
 	struct pcr_handle *handle = pdev->dev.platform_data;
+	unsigned long host_addr;
 
 	if (!handle)
 		return -ENXIO;
@@ -1227,8 +1489,15 @@ static int rtsx_pci_sdmmc_drv_probe(struct platform_device *pdev)
 	pcr->slots[RTSX_SD_CARD].p_dev = pdev;
 	pcr->slots[RTSX_SD_CARD].card_event = rtsx_pci_sdmmc_card_event;
 
-	mutex_init(&host->host_mutex);
+	host_addr = (unsigned long)host;
+	host->next_data.cookie = 1;
+	setup_timer(&host->timer, sd_request_timeout, host_addr);
+	tasklet_init(&host->cmd_tasklet, sd_get_rsp, host_addr);
+	tasklet_init(&host->data_tasklet, sd_finish_multi_rw, host_addr);
+	tasklet_init(&host->finish_tasklet, sd_finish_request, host_addr);
+	spin_lock_init(&host->lock);
 
+	pcr->slots[RTSX_SD_CARD].done_transfer = sd_isr_done_transfer;
 	realtek_init_host(host);
 
 	mmc_add_host(mmc);
@@ -1241,6 +1510,8 @@ static int rtsx_pci_sdmmc_drv_remove(struct platform_device *pdev)
 	struct realtek_pci_sdmmc *host = platform_get_drvdata(pdev);
 	struct rtsx_pcr *pcr;
 	struct mmc_host *mmc;
+	struct mmc_request *mrq;
+	unsigned long flags;
 
 	if (!host)
 		return 0;
@@ -1248,22 +1519,33 @@ static int rtsx_pci_sdmmc_drv_remove(struct platform_device *pdev)
 	pcr = host->pcr;
 	pcr->slots[RTSX_SD_CARD].p_dev = NULL;
 	pcr->slots[RTSX_SD_CARD].card_event = NULL;
+	pcr->slots[RTSX_SD_CARD].done_transfer = NULL;
 	mmc = host->mmc;
+	mrq = host->mrq;
 
-	mutex_lock(&host->host_mutex);
+	spin_lock_irqsave(&host->lock, flags);
 	if (host->mrq) {
 		dev_dbg(&(pdev->dev),
 			"%s: Controller removed during transfer\n",
 			mmc_hostname(mmc));
 
-		rtsx_pci_complete_unfinished_transfer(pcr);
+		if (mrq->sbc)
+			mrq->sbc->error = -ENOMEDIUM;
+		if (mrq->cmd)
+			mrq->cmd->error = -ENOMEDIUM;
+		if (mrq->stop)
+			mrq->stop->error = -ENOMEDIUM;
+		if (mrq->data)
+			mrq->data->error = -ENOMEDIUM;
 
-		host->mrq->cmd->error = -ENOMEDIUM;
-		if (host->mrq->stop)
-			host->mrq->stop->error = -ENOMEDIUM;
-		mmc_request_done(mmc, host->mrq);
+		tasklet_schedule(&host->finish_tasklet);
 	}
-	mutex_unlock(&host->host_mutex);
+	spin_unlock_irqrestore(&host->lock, flags);
+
+	del_timer_sync(&host->timer);
+	tasklet_kill(&host->cmd_tasklet);
+	tasklet_kill(&host->data_tasklet);
+	tasklet_kill(&host->finish_tasklet);
 
 	mmc_remove_host(mmc);
 	host->eject = true;
diff --git a/include/linux/mfd/rtsx_common.h b/include/linux/mfd/rtsx_common.h
index 443176e..7c36cc5 100644
--- a/include/linux/mfd/rtsx_common.h
+++ b/include/linux/mfd/rtsx_common.h
@@ -45,6 +45,7 @@ struct platform_device;
 struct rtsx_slot {
 	struct platform_device	*p_dev;
 	void			(*card_event)(struct platform_device *p_dev);
+	void			(*done_transfer)(struct platform_device *p_dev);
 };
 
 #endif
diff --git a/include/linux/mfd/rtsx_pci.h b/include/linux/mfd/rtsx_pci.h
index a383597..8d6bbd6 100644
--- a/include/linux/mfd/rtsx_pci.h
+++ b/include/linux/mfd/rtsx_pci.h
@@ -943,6 +943,12 @@ void rtsx_pci_send_cmd_no_wait(struct rtsx_pcr *pcr);
 int rtsx_pci_send_cmd(struct rtsx_pcr *pcr, int timeout);
 int rtsx_pci_transfer_data(struct rtsx_pcr *pcr, struct scatterlist *sglist,
 		int num_sg, bool read, int timeout);
+int rtsx_pci_dma_map_sg(struct rtsx_pcr *pcr, struct scatterlist *sglist,
+		int num_sg, bool read);
+int rtsx_pci_dma_unmap_sg(struct rtsx_pcr *pcr, struct scatterlist *sglist,
+		int num_sg, bool read);
+int rtsx_pci_dma_transfer(struct rtsx_pcr *pcr, struct scatterlist *sglist,
+		int sg_count, bool read);
 int rtsx_pci_read_ppbuf(struct rtsx_pcr *pcr, u8 *buf, int buf_len);
 int rtsx_pci_write_ppbuf(struct rtsx_pcr *pcr, u8 *buf, int buf_len);
 int rtsx_pci_card_pull_ctl_enable(struct rtsx_pcr *pcr, int card);
-- 
1.7.9.5

[PATCH v2 3/3] mmc: rtsx: add support for pre_req and post_req

[micky_ching]

From: Micky Ching <micky_ching@realsil.com.cn>

Add support for non-blocking request, pre_req() runs dma_map_sg() and
post_req() runs dma_unmap_sg(). This patch can increase card read/write
speed, especially for high speed card and slow CPU(for some embedded
platform).

Users can get a great benefit from this patch. if CPU frequency is 800MHz,
SDR104 or DDR50 card read/write speed may increase more than 15%.

test results:
intel i3(800MHz - 2.3GHz), SD card clock 208MHz

performance mode(2.3GHz):
Before:
dd if=/dev/mmcblk0p1 of=/dev/null bs=64k count=1024
67108864 bytes (67 MB) copied, 1.18191 s, 56.8 MB/s
After:
 dd if=/dev/mmcblk0p1 of=/dev/null bs=64k count=1024
67108864 bytes (67 MB) copied, 1.09276 s, 61.4 MB/s

powersave mode(800MHz):
Before:
dd if=/dev/mmcblk0p1 of=/dev/null bs=64k count=1024
67108864 bytes (67 MB) copied, 1.29569 s, 51.8 MB/s
After:
dd if=/dev/mmcblk0p1 of=/dev/null bs=64k count=1024
67108864 bytes (67 MB) copied, 1.11218 s, 60.3 MB/s

Signed-off-by: Micky Ching <micky_ching@realsil.com.cn>
---
 drivers/mfd/rtsx_pcr.c            |  132 ++++++++----
 drivers/mmc/host/rtsx_pci_sdmmc.c |  418 +++++++++++++++++++++++++++++++------
 include/linux/mfd/rtsx_common.h   |    1 +
 include/linux/mfd/rtsx_pci.h      |    6 +
 4 files changed, 448 insertions(+), 109 deletions(-)

diff --git a/drivers/mfd/rtsx_pcr.c b/drivers/mfd/rtsx_pcr.c
index 1d15735..c9de3d5 100644
--- a/drivers/mfd/rtsx_pcr.c
+++ b/drivers/mfd/rtsx_pcr.c
@@ -338,58 +338,28 @@ int rtsx_pci_transfer_data(struct rtsx_pcr *pcr, struct scatterlist *sglist,
 		int num_sg, bool read, int timeout)
 {
 	struct completion trans_done;
-	u8 dir;
-	int err = 0, i, count;
+	int err = 0, count;
 	long timeleft;
 	unsigned long flags;
-	struct scatterlist *sg;
-	enum dma_data_direction dma_dir;
-	u32 val;
-	dma_addr_t addr;
-	unsigned int len;
-
-	dev_dbg(&(pcr->pci->dev), "--> %s: num_sg = %d\n", __func__, num_sg);
-
-	/* don't transfer data during abort processing */
-	if (pcr->remove_pci)
-		return -EINVAL;
-
-	if ((sglist == NULL) || (num_sg <= 0))
-		return -EINVAL;
 
-	if (read) {
-		dir = DEVICE_TO_HOST;
-		dma_dir = DMA_FROM_DEVICE;
-	} else {
-		dir = HOST_TO_DEVICE;
-		dma_dir = DMA_TO_DEVICE;
-	}
-
-	count = dma_map_sg(&(pcr->pci->dev), sglist, num_sg, dma_dir);
+	count = rtsx_pci_dma_map_sg(pcr, sglist, num_sg, read);
 	if (count < 1) {
 		dev_err(&(pcr->pci->dev), "scatterlist map failed\n");
 		return -EINVAL;
 	}
 	dev_dbg(&(pcr->pci->dev), "DMA mapping count: %d\n", count);
 
-	val = ((u32)(dir & 0x01) << 29) | TRIG_DMA | ADMA_MODE;
-	pcr->sgi = 0;
-	for_each_sg(sglist, sg, count, i) {
-		addr = sg_dma_address(sg);
-		len = sg_dma_len(sg);
-		rtsx_pci_add_sg_tbl(pcr, addr, len, i == count - 1);
-	}
 
 	spin_lock_irqsave(&pcr->lock, flags);
 
 	pcr->done = &trans_done;
 	pcr->trans_result = TRANS_NOT_READY;
 	init_completion(&trans_done);
-	rtsx_pci_writel(pcr, RTSX_HDBAR, pcr->host_sg_tbl_addr);
-	rtsx_pci_writel(pcr, RTSX_HDBCTLR, val);
 
 	spin_unlock_irqrestore(&pcr->lock, flags);
 
+	rtsx_pci_dma_transfer(pcr, sglist, count, read);
+
 	timeleft = wait_for_completion_interruptible_timeout(
 			&trans_done, msecs_to_jiffies(timeout));
 	if (timeleft <= 0) {
@@ -413,7 +383,7 @@ out:
 	pcr->done = NULL;
 	spin_unlock_irqrestore(&pcr->lock, flags);
 
-	dma_unmap_sg(&(pcr->pci->dev), sglist, num_sg, dma_dir);
+	rtsx_pci_dma_unmap_sg(pcr, sglist, num_sg, read);
 
 	if ((err < 0) && (err != -ENODEV))
 		rtsx_pci_stop_cmd(pcr);
@@ -425,6 +395,73 @@ out:
 }
 EXPORT_SYMBOL_GPL(rtsx_pci_transfer_data);
 
+int rtsx_pci_dma_map_sg(struct rtsx_pcr *pcr, struct scatterlist *sglist,
+		int num_sg, bool read)
+{
+	enum dma_data_direction dir = read ? DMA_FROM_DEVICE : DMA_TO_DEVICE;
+
+	if (pcr->remove_pci)
+		return -EINVAL;
+
+	if ((sglist == NULL) || num_sg < 1)
+		return -EINVAL;
+
+	return dma_map_sg(&(pcr->pci->dev), sglist, num_sg, dir);
+}
+EXPORT_SYMBOL_GPL(rtsx_pci_dma_map_sg);
+
+int rtsx_pci_dma_unmap_sg(struct rtsx_pcr *pcr, struct scatterlist *sglist,
+		int num_sg, bool read)
+{
+	enum dma_data_direction dir = read ? DMA_FROM_DEVICE : DMA_TO_DEVICE;
+
+	if (pcr->remove_pci)
+		return -EINVAL;
+
+	if (sglist == NULL || num_sg < 1)
+		return -EINVAL;
+
+	dma_unmap_sg(&(pcr->pci->dev), sglist, num_sg, dir);
+	return num_sg;
+}
+EXPORT_SYMBOL_GPL(rtsx_pci_dma_unmap_sg);
+
+int rtsx_pci_dma_transfer(struct rtsx_pcr *pcr, struct scatterlist *sglist,
+		int sg_count, bool read)
+{
+	struct scatterlist *sg;
+	dma_addr_t addr;
+	unsigned int len;
+	int i;
+	u32 val;
+	u8 dir = read ? DEVICE_TO_HOST : HOST_TO_DEVICE;
+	unsigned long flags;
+
+	if (pcr->remove_pci)
+		return -EINVAL;
+
+	if ((sglist == NULL) || (sg_count < 1))
+		return -EINVAL;
+
+	val = ((u32)(dir & 0x01) << 29) | TRIG_DMA | ADMA_MODE;
+	pcr->sgi = 0;
+	for_each_sg(sglist, sg, sg_count, i) {
+		addr = sg_dma_address(sg);
+		len = sg_dma_len(sg);
+		rtsx_pci_add_sg_tbl(pcr, addr, len, i == sg_count - 1);
+	}
+
+	spin_lock_irqsave(&pcr->lock, flags);
+
+	rtsx_pci_writel(pcr, RTSX_HDBAR, pcr->host_sg_tbl_addr);
+	rtsx_pci_writel(pcr, RTSX_HDBCTLR, val);
+
+	spin_unlock_irqrestore(&pcr->lock, flags);
+
+	return 0;
+}
+EXPORT_SYMBOL_GPL(rtsx_pci_dma_transfer);
+
 int rtsx_pci_read_ppbuf(struct rtsx_pcr *pcr, u8 *buf, int buf_len)
 {
 	int err;
@@ -836,6 +873,8 @@ static irqreturn_t rtsx_pci_isr(int irq, void *dev_id)
 	int_reg = rtsx_pci_readl(pcr, RTSX_BIPR);
 	/* Clear interrupt flag */
 	rtsx_pci_writel(pcr, RTSX_BIPR, int_reg);
+	dev_dbg(&pcr->pci->dev, "=========== BIPR 0x%8x ==========\n", int_reg);
+
 	if ((int_reg & pcr->bier) == 0) {
 		spin_unlock(&pcr->lock);
 		return IRQ_NONE;
@@ -866,17 +905,28 @@ static irqreturn_t rtsx_pci_isr(int irq, void *dev_id)
 	}
 
 	if (int_reg & (NEED_COMPLETE_INT | DELINK_INT)) {
-		if (int_reg & (TRANS_FAIL_INT | DELINK_INT)) {
+		if (int_reg & (TRANS_FAIL_INT | DELINK_INT))
 			pcr->trans_result = TRANS_RESULT_FAIL;
-			if (pcr->done)
-				complete(pcr->done);
-		} else if (int_reg & TRANS_OK_INT) {
+		else if (int_reg & TRANS_OK_INT)
 			pcr->trans_result = TRANS_RESULT_OK;
-			if (pcr->done)
-				complete(pcr->done);
+
+		if (pcr->done)
+			complete(pcr->done);
+
+		if (int_reg & SD_EXIST) {
+			struct rtsx_slot *slot = &pcr->slots[RTSX_SD_CARD];
+			if (slot && slot->done_transfer)
+				slot->done_transfer(slot->p_dev);
+		}
+
+		if (int_reg & MS_EXIST) {
+			struct rtsx_slot *slot = &pcr->slots[RTSX_SD_CARD];
+			if (slot && slot->done_transfer)
+				slot->done_transfer(slot->p_dev);
 		}
 	}
 
+
 	if (pcr->card_inserted || pcr->card_removed)
 		schedule_delayed_work(&pcr->carddet_work,
 				msecs_to_jiffies(200));
diff --git a/drivers/mmc/host/rtsx_pci_sdmmc.c b/drivers/mmc/host/rtsx_pci_sdmmc.c
index 0b9ded1..5fb994f 100644
--- a/drivers/mmc/host/rtsx_pci_sdmmc.c
+++ b/drivers/mmc/host/rtsx_pci_sdmmc.c
@@ -31,14 +31,28 @@
 #include <linux/mfd/rtsx_pci.h>
 #include <asm/unaligned.h>
 
+struct realtek_next {
+	unsigned int	sg_count;
+	s32		cookie;
+};
+
 struct realtek_pci_sdmmc {
 	struct platform_device	*pdev;
 	struct rtsx_pcr		*pcr;
 	struct mmc_host		*mmc;
 	struct mmc_request	*mrq;
-
-	struct mutex		host_mutex;
-
+	struct mmc_command	*cmd;
+	struct mmc_data		*data;
+
+	spinlock_t		lock;
+	struct timer_list	timer;
+	struct tasklet_struct	cmd_tasklet;
+	struct tasklet_struct	data_tasklet;
+	struct tasklet_struct	finish_tasklet;
+
+	u8			rsp_type;
+	u8			rsp_len;
+	int			sg_count;
 	u8			ssc_depth;
 	unsigned int		clock;
 	bool			vpclk;
@@ -48,8 +62,13 @@ struct realtek_pci_sdmmc {
 	int			power_state;
 #define SDMMC_POWER_ON		1
 #define SDMMC_POWER_OFF		0
+
+	struct realtek_next	next_data;
 };
 
+static int sd_start_multi_rw(struct realtek_pci_sdmmc *host,
+		struct mmc_request *mrq);
+
 static inline struct device *sdmmc_dev(struct realtek_pci_sdmmc *host)
 {
 	return &(host->pdev->dev);
@@ -86,6 +105,95 @@ static void sd_print_debug_regs(struct realtek_pci_sdmmc *host)
 #define sd_print_debug_regs(host)
 #endif /* DEBUG */
 
+static void sd_isr_done_transfer(struct platform_device *pdev)
+{
+	struct realtek_pci_sdmmc *host = platform_get_drvdata(pdev);
+
+	spin_lock(&host->lock);
+	if (host->cmd)
+		tasklet_schedule(&host->cmd_tasklet);
+	if (host->data)
+		tasklet_schedule(&host->data_tasklet);
+	spin_unlock(&host->lock);
+}
+
+static void sd_request_timeout(unsigned long host_addr)
+{
+	struct realtek_pci_sdmmc *host = (struct realtek_pci_sdmmc *)host_addr;
+	unsigned long flags;
+
+	spin_lock_irqsave(&host->lock, flags);
+
+	if (!host->mrq) {
+		dev_err(sdmmc_dev(host), "error: no request exist\n");
+		goto out;
+	}
+
+	if (host->cmd)
+		host->cmd->error = -ETIMEDOUT;
+	if (host->data)
+		host->data->error = -ETIMEDOUT;
+
+	dev_dbg(sdmmc_dev(host), "timeout for request\n");
+
+out:
+	tasklet_schedule(&host->finish_tasklet);
+	spin_unlock_irqrestore(&host->lock, flags);
+}
+
+static void sd_finish_request(unsigned long host_addr)
+{
+	struct realtek_pci_sdmmc *host = (struct realtek_pci_sdmmc *)host_addr;
+	struct rtsx_pcr *pcr = host->pcr;
+	struct mmc_request *mrq;
+	struct mmc_command *cmd;
+	struct mmc_data *data;
+	unsigned long flags;
+	bool any_error;
+
+	spin_lock_irqsave(&host->lock, flags);
+
+	del_timer(&host->timer);
+	mrq = host->mrq;
+	if (!mrq) {
+		dev_err(sdmmc_dev(host), "error: no request need finish\n");
+		goto out;
+	}
+
+	cmd = mrq->cmd;
+	data = mrq->data;
+
+	any_error = (mrq->sbc && mrq->sbc->error) ||
+		(mrq->stop && mrq->stop->error) ||
+		(cmd && cmd->error) || (data && data->error);
+
+	if (any_error) {
+		rtsx_pci_stop_cmd(pcr);
+		sd_clear_error(host);
+	}
+
+	if (data) {
+		if (any_error)
+			data->bytes_xfered = 0;
+		else
+			data->bytes_xfered = data->blocks * data->blksz;
+
+		if (!data->host_cookie)
+			rtsx_pci_dma_unmap_sg(pcr, data->sg, data->sg_len,
+					data->flags & MMC_DATA_READ);
+
+	}
+
+	host->mrq = NULL;
+	host->cmd = NULL;
+	host->data = NULL;
+
+out:
+	spin_unlock_irqrestore(&host->lock, flags);
+	mutex_unlock(&pcr->pcr_mutex);
+	mmc_request_done(host->mmc, mrq);
+}
+
 static int sd_read_data(struct realtek_pci_sdmmc *host, u8 *cmd, u16 byte_cnt,
 		u8 *buf, int buf_len, int timeout)
 {
@@ -203,8 +311,7 @@ static int sd_write_data(struct realtek_pci_sdmmc *host, u8 *cmd, u16 byte_cnt,
 	return 0;
 }
 
-static void sd_send_cmd_get_rsp(struct realtek_pci_sdmmc *host,
-		struct mmc_command *cmd)
+static void sd_send_cmd(struct realtek_pci_sdmmc *host, struct mmc_command *cmd)
 {
 	struct rtsx_pcr *pcr = host->pcr;
 	u8 cmd_idx = (u8)cmd->opcode;
@@ -212,11 +319,14 @@ static void sd_send_cmd_get_rsp(struct realtek_pci_sdmmc *host,
 	int err = 0;
 	int timeout = 100;
 	int i;
-	u8 *ptr;
-	int stat_idx = 0;
 	u8 rsp_type;
 	int rsp_len = 5;
-	bool clock_toggled = false;
+	unsigned long flags;
+
+	if (host->cmd)
+		dev_err(sdmmc_dev(host), "error: cmd already exist\n");
+
+	host->cmd = cmd;
 
 	dev_dbg(sdmmc_dev(host), "%s: SD/MMC CMD %d, arg = 0x%08x\n",
 			__func__, cmd_idx, arg);
@@ -251,6 +361,8 @@ static void sd_send_cmd_get_rsp(struct realtek_pci_sdmmc *host,
 		err = -EINVAL;
 		goto out;
 	}
+	host->rsp_type = rsp_type;
+	host->rsp_len = rsp_len;
 
 	if (rsp_type == SD_RSP_TYPE_R1b)
 		timeout = 3000;
@@ -260,8 +372,6 @@ static void sd_send_cmd_get_rsp(struct realtek_pci_sdmmc *host,
 				0xFF, SD_CLK_TOGGLE_EN);
 		if (err < 0)
 			goto out;
-
-		clock_toggled = true;
 	}
 
 	rtsx_pci_init_cmd(pcr);
@@ -285,25 +395,60 @@ static void sd_send_cmd_get_rsp(struct realtek_pci_sdmmc *host,
 		/* Read data from ping-pong buffer */
 		for (i = PPBUF_BASE2; i < PPBUF_BASE2 + 16; i++)
 			rtsx_pci_add_cmd(pcr, READ_REG_CMD, (u16)i, 0, 0);
-		stat_idx = 16;
 	} else if (rsp_type != SD_RSP_TYPE_R0) {
 		/* Read data from SD_CMDx registers */
 		for (i = SD_CMD0; i <= SD_CMD4; i++)
 			rtsx_pci_add_cmd(pcr, READ_REG_CMD, (u16)i, 0, 0);
-		stat_idx = 5;
 	}
 
 	rtsx_pci_add_cmd(pcr, READ_REG_CMD, SD_STAT1, 0, 0);
 
-	err = rtsx_pci_send_cmd(pcr, timeout);
-	if (err < 0) {
-		sd_print_debug_regs(host);
-		sd_clear_error(host);
-		dev_dbg(sdmmc_dev(host),
-			"rtsx_pci_send_cmd error (err = %d)\n", err);
+	mod_timer(&host->timer, jiffies + msecs_to_jiffies(timeout));
+
+	spin_lock_irqsave(&pcr->lock, flags);
+	pcr->trans_result = TRANS_NOT_READY;
+	rtsx_pci_send_cmd_no_wait(pcr);
+	spin_unlock_irqrestore(&pcr->lock, flags);
+
+	return;
+
+out:
+	cmd->error = err;
+	tasklet_schedule(&host->finish_tasklet);
+}
+
+static void sd_get_rsp(unsigned long host_addr)
+{
+	struct realtek_pci_sdmmc *host = (struct realtek_pci_sdmmc *)host_addr;
+	struct rtsx_pcr *pcr = host->pcr;
+	struct mmc_command *cmd;
+	int i, err = 0, stat_idx;
+	u8 *ptr, rsp_type;
+	unsigned long flags;
+
+	spin_lock_irqsave(&host->lock, flags);
+
+	cmd = host->cmd;
+	host->cmd = NULL;
+
+	if (!cmd) {
+		dev_err(sdmmc_dev(host), "error: cmd not exist\n");
 		goto out;
 	}
 
+	spin_lock(&pcr->lock);
+	if (pcr->trans_result == TRANS_NO_DEVICE)
+		err = -ENODEV;
+	else if (pcr->trans_result != TRANS_RESULT_OK)
+		err = -EINVAL;
+	spin_unlock(&pcr->lock);
+
+	if (err < 0)
+		goto out;
+
+	rsp_type = host->rsp_type;
+	stat_idx = host->rsp_len;
+
 	if (rsp_type == SD_RSP_TYPE_R0) {
 		err = 0;
 		goto out;
@@ -340,26 +485,106 @@ static void sd_send_cmd_get_rsp(struct realtek_pci_sdmmc *host,
 				cmd->resp[0]);
 	}
 
+	if (cmd == host->mrq->sbc) {
+		sd_send_cmd(host, host->mrq->cmd);
+		spin_unlock_irqrestore(&host->lock, flags);
+		return;
+	}
+
+	if (cmd == host->mrq->stop)
+		goto out;
+
+	if (cmd->data) {
+		sd_start_multi_rw(host, host->mrq);
+		spin_unlock_irqrestore(&host->lock, flags);
+		return;
+	}
+
 out:
 	cmd->error = err;
 
-	if (err && clock_toggled)
-		rtsx_pci_write_register(pcr, SD_BUS_STAT,
-				SD_CLK_TOGGLE_EN | SD_CLK_FORCE_STOP, 0);
+	tasklet_schedule(&host->finish_tasklet);
+	spin_unlock_irqrestore(&host->lock, flags);
+}
+
+static int sd_pre_dma_transfer(struct realtek_pci_sdmmc *host,
+			struct mmc_data *data, struct realtek_next *next)
+{
+	struct rtsx_pcr *pcr = host->pcr;
+	int read = data->flags & MMC_DATA_READ;
+	int sg_count = 0;
+
+	if (!next && data->host_cookie &&
+		data->host_cookie != host->next_data.cookie) {
+		dev_err(sdmmc_dev(host),
+			"error: invalid cookie data[%d] host[%d]\n",
+			data->host_cookie, host->next_data.cookie);
+		data->host_cookie = 0;
+	}
+
+	if (next || (!next && data->host_cookie != host->next_data.cookie))
+		sg_count = rtsx_pci_dma_map_sg(pcr,
+				data->sg, data->sg_len, read);
+	else
+		sg_count = host->next_data.sg_count;
+
+	if (next) {
+		next->sg_count = sg_count;
+		if (++next->cookie < 0)
+			next->cookie = 1;
+		data->host_cookie = next->cookie;
+	}
+
+	return sg_count;
+}
+
+static void sdmmc_pre_req(struct mmc_host *mmc, struct mmc_request *mrq,
+		bool is_first_req)
+{
+	struct realtek_pci_sdmmc *host = mmc_priv(mmc);
+	struct mmc_data *data = mrq->data;
+
+	if (data->host_cookie) {
+		dev_err(sdmmc_dev(host),
+			"error: descard already cookie data[%d]\n",
+			data->host_cookie);
+		data->host_cookie = 0;
+	}
+
+	dev_dbg(sdmmc_dev(host), "dma sg prepared: %d\n",
+		sd_pre_dma_transfer(host, data, &host->next_data));
+}
+
+static void sdmmc_post_req(struct mmc_host *mmc, struct mmc_request *mrq,
+		int err)
+{
+	struct realtek_pci_sdmmc *host = mmc_priv(mmc);
+	struct rtsx_pcr *pcr = host->pcr;
+	struct mmc_data *data = mrq->data;
+	int read = data->flags & MMC_DATA_READ;
+
+	rtsx_pci_dma_unmap_sg(pcr, data->sg, data->sg_len, read);
+	data->host_cookie = 0;
 }
 
-static int sd_rw_multi(struct realtek_pci_sdmmc *host, struct mmc_request *mrq)
+static int sd_start_multi_rw(struct realtek_pci_sdmmc *host,
+		struct mmc_request *mrq)
 {
 	struct rtsx_pcr *pcr = host->pcr;
 	struct mmc_host *mmc = host->mmc;
 	struct mmc_card *card = mmc->card;
 	struct mmc_data *data = mrq->data;
 	int uhs = mmc_card_uhs(card);
-	int read = (data->flags & MMC_DATA_READ) ? 1 : 0;
+	int read = data->flags & MMC_DATA_READ;
 	u8 cfg2, trans_mode;
 	int err;
 	size_t data_len = data->blksz * data->blocks;
 
+	if (host->data)
+		dev_err(sdmmc_dev(host), "error: data already exist\n");
+
+	host->data = data;
+
 	if (read) {
 		cfg2 = SD_CALCULATE_CRC7 | SD_CHECK_CRC16 |
 			SD_NO_WAIT_BUSY_END | SD_CHECK_CRC7 | SD_RSP_LEN_0;
@@ -410,17 +635,56 @@ static int sd_rw_multi(struct realtek_pci_sdmmc *host, struct mmc_request *mrq)
 	rtsx_pci_add_cmd(pcr, CHECK_REG_CMD, SD_TRANSFER,
 			SD_TRANSFER_END, SD_TRANSFER_END);
 
+	mod_timer(&host->timer, jiffies + 10 * HZ);
 	rtsx_pci_send_cmd_no_wait(pcr);
 
-	err = rtsx_pci_transfer_data(pcr, data->sg, data->sg_len, read, 10000);
+	err = rtsx_pci_dma_transfer(pcr, data->sg, host->sg_count, read);
 	if (err < 0) {
-		sd_clear_error(host);
-		return err;
+		data->error = err;
+		tasklet_schedule(&host->finish_tasklet);
 	}
-
 	return 0;
 }
 
+static void sd_finish_multi_rw(unsigned long host_addr)
+{
+	struct realtek_pci_sdmmc *host = (struct realtek_pci_sdmmc *)host_addr;
+	struct rtsx_pcr *pcr = host->pcr;
+	struct mmc_data *data;
+	int err = 0;
+	unsigned long flags;
+
+	spin_lock_irqsave(&host->lock, flags);
+
+	if (!host->data) {
+		dev_err(sdmmc_dev(host), "error: no data exist\n");
+		goto out;
+	}
+
+	data = host->data;
+	host->data = NULL;
+
+	if (pcr->trans_result == TRANS_NO_DEVICE)
+		err = -ENODEV;
+	else if (pcr->trans_result != TRANS_RESULT_OK)
+		err = -EINVAL;
+
+	if (err < 0) {
+		data->error = err;
+		goto out;
+	}
+
+	if (!host->mrq->sbc && data->stop) {
+		sd_send_cmd(host, data->stop);
+		spin_unlock_irqrestore(&host->lock, flags);
+		return;
+	}
+
+out:
+	tasklet_schedule(&host->finish_tasklet);
+	spin_unlock_irqrestore(&host->lock, flags);
+}
+
 static inline void sd_enable_initial_mode(struct realtek_pci_sdmmc *host)
 {
 	rtsx_pci_write_register(host->pcr, SD_CFG1,
@@ -637,6 +901,13 @@ static int sd_tuning_rx(struct realtek_pci_sdmmc *host, u8 opcode)
 	return 0;
 }
 
+static inline bool sd_use_muti_rw(struct mmc_command *cmd)
+{
+	return mmc_op_multi(cmd->opcode) ||
+		(cmd->opcode == MMC_READ_SINGLE_BLOCK) ||
+		(cmd->opcode == MMC_WRITE_BLOCK);
+}
+
 static void sdmmc_request(struct mmc_host *mmc, struct mmc_request *mrq)
 {
 	struct realtek_pci_sdmmc *host = mmc_priv(mmc);
@@ -645,6 +916,14 @@ static void sdmmc_request(struct mmc_host *mmc, struct mmc_request *mrq)
 	struct mmc_data *data = mrq->data;
 	unsigned int data_size = 0;
 	int err;
+	unsigned long flags;
+
+	mutex_lock(&pcr->pcr_mutex);
+	spin_lock_irqsave(&host->lock, flags);
+
+	if (host->mrq)
+		dev_err(sdmmc_dev(host), "error: request already exist\n");
+	host->mrq = mrq;
 
 	if (host->eject) {
 		cmd->error = -ENOMEDIUM;
@@ -657,8 +936,6 @@ static void sdmmc_request(struct mmc_host *mmc, struct mmc_request *mrq)
 		goto finish;
 	}
 
-	mutex_lock(&pcr->pcr_mutex);
-
 	rtsx_pci_start_run(pcr);
 
 	rtsx_pci_switch_clock(pcr, host->clock, host->ssc_depth,
@@ -667,46 +944,28 @@ static void sdmmc_request(struct mmc_host *mmc, struct mmc_request *mrq)
 	rtsx_pci_write_register(pcr, CARD_SHARE_MODE,
 			CARD_SHARE_MASK, CARD_SHARE_48_SD);
 
-	mutex_lock(&host->host_mutex);
-	host->mrq = mrq;
-	mutex_unlock(&host->host_mutex);
-
 	if (mrq->data)
 		data_size = data->blocks * data->blksz;
 
-	if (!data_size || mmc_op_multi(cmd->opcode) ||
-			(cmd->opcode == MMC_READ_SINGLE_BLOCK) ||
-			(cmd->opcode == MMC_WRITE_BLOCK)) {
-		sd_send_cmd_get_rsp(host, cmd);
-
-		if (!cmd->error && data_size) {
-			sd_rw_multi(host, mrq);
+	if (sd_use_muti_rw(cmd))
+		host->sg_count = sd_pre_dma_transfer(host, data, NULL);
 
-			if (mmc_op_multi(cmd->opcode) && mrq->stop)
-				sd_send_cmd_get_rsp(host, mrq->stop);
-		}
+	if (!data_size || sd_use_muti_rw(cmd)) {
+		if (mrq->sbc)
+			sd_send_cmd(host, mrq->sbc);
+		else
+			sd_send_cmd(host, cmd);
+		spin_unlock_irqrestore(&host->lock, flags);
 	} else {
+		spin_unlock_irqrestore(&host->lock, flags);
 		sd_normal_rw(host, mrq);
+		tasklet_schedule(&host->finish_tasklet);
 	}
-
-	if (mrq->data) {
-		if (cmd->error || data->error)
-			data->bytes_xfered = 0;
-		else
-			data->bytes_xfered = data->blocks * data->blksz;
-	}
-
-	mutex_unlock(&pcr->pcr_mutex);
+	return;
 
 finish:
-	if (cmd->error)
-		dev_dbg(sdmmc_dev(host), "cmd->error = %d\n", cmd->error);
-
-	mutex_lock(&host->host_mutex);
-	host->mrq = NULL;
-	mutex_unlock(&host->host_mutex);
-
-	mmc_request_done(mmc, mrq);
+	tasklet_schedule(&host->finish_tasklet);
+	spin_unlock_irqrestore(&host->lock, flags);
 }
 
 static int sd_set_bus_width(struct realtek_pci_sdmmc *host,
@@ -1141,6 +1400,8 @@ out:
 }
 
 static const struct mmc_host_ops realtek_pci_sdmmc_ops = {
+	.pre_req = sdmmc_pre_req,
+	.post_req = sdmmc_post_req,
 	.request = sdmmc_request,
 	.set_ios = sdmmc_set_ios,
 	.get_ro = sdmmc_get_ro,
@@ -1204,6 +1465,7 @@ static int rtsx_pci_sdmmc_drv_probe(struct platform_device *pdev)
 	struct realtek_pci_sdmmc *host;
 	struct rtsx_pcr *pcr;
 	struct pcr_handle *handle = pdev->dev.platform_data;
+	unsigned long host_addr;
 
 	if (!handle)
 		return -ENXIO;
@@ -1227,8 +1489,15 @@ static int rtsx_pci_sdmmc_drv_probe(struct platform_device *pdev)
 	pcr->slots[RTSX_SD_CARD].p_dev = pdev;
 	pcr->slots[RTSX_SD_CARD].card_event = rtsx_pci_sdmmc_card_event;
 
-	mutex_init(&host->host_mutex);
+	host_addr = (unsigned long)host;
+	host->next_data.cookie = 1;
+	setup_timer(&host->timer, sd_request_timeout, host_addr);
+	tasklet_init(&host->cmd_tasklet, sd_get_rsp, host_addr);
+	tasklet_init(&host->data_tasklet, sd_finish_multi_rw, host_addr);
+	tasklet_init(&host->finish_tasklet, sd_finish_request, host_addr);
+	spin_lock_init(&host->lock);
 
+	pcr->slots[RTSX_SD_CARD].done_transfer = sd_isr_done_transfer;
 	realtek_init_host(host);
 
 	mmc_add_host(mmc);
@@ -1241,6 +1510,8 @@ static int rtsx_pci_sdmmc_drv_remove(struct platform_device *pdev)
 	struct realtek_pci_sdmmc *host = platform_get_drvdata(pdev);
 	struct rtsx_pcr *pcr;
 	struct mmc_host *mmc;
+	struct mmc_request *mrq;
+	unsigned long flags;
 
 	if (!host)
 		return 0;
@@ -1248,22 +1519,33 @@ static int rtsx_pci_sdmmc_drv_remove(struct platform_device *pdev)
 	pcr = host->pcr;
 	pcr->slots[RTSX_SD_CARD].p_dev = NULL;
 	pcr->slots[RTSX_SD_CARD].card_event = NULL;
+	pcr->slots[RTSX_SD_CARD].done_transfer = NULL;
 	mmc = host->mmc;
+	mrq = host->mrq;
 
-	mutex_lock(&host->host_mutex);
+	spin_lock_irqsave(&host->lock, flags);
 	if (host->mrq) {
 		dev_dbg(&(pdev->dev),
 			"%s: Controller removed during transfer\n",
 			mmc_hostname(mmc));
 
-		rtsx_pci_complete_unfinished_transfer(pcr);
+		if (mrq->sbc)
+			mrq->sbc->error = -ENOMEDIUM;
+		if (mrq->cmd)
+			mrq->cmd->error = -ENOMEDIUM;
+		if (mrq->stop)
+			mrq->stop->error = -ENOMEDIUM;
+		if (mrq->data)
+			mrq->data->error = -ENOMEDIUM;
 
-		host->mrq->cmd->error = -ENOMEDIUM;
-		if (host->mrq->stop)
-			host->mrq->stop->error = -ENOMEDIUM;
-		mmc_request_done(mmc, host->mrq);
+		tasklet_schedule(&host->finish_tasklet);
 	}
-	mutex_unlock(&host->host_mutex);
+	spin_unlock_irqrestore(&host->lock, flags);
+
+	del_timer_sync(&host->timer);
+	tasklet_kill(&host->cmd_tasklet);
+	tasklet_kill(&host->data_tasklet);
+	tasklet_kill(&host->finish_tasklet);
 
 	mmc_remove_host(mmc);
 	host->eject = true;
diff --git a/include/linux/mfd/rtsx_common.h b/include/linux/mfd/rtsx_common.h
index 443176e..7c36cc5 100644
--- a/include/linux/mfd/rtsx_common.h
+++ b/include/linux/mfd/rtsx_common.h
@@ -45,6 +45,7 @@ struct platform_device;
 struct rtsx_slot {
 	struct platform_device	*p_dev;
 	void			(*card_event)(struct platform_device *p_dev);
+	void			(*done_transfer)(struct platform_device *p_dev);
 };
 
 #endif
diff --git a/include/linux/mfd/rtsx_pci.h b/include/linux/mfd/rtsx_pci.h
index a383597..8d6bbd6 100644
--- a/include/linux/mfd/rtsx_pci.h
+++ b/include/linux/mfd/rtsx_pci.h
@@ -943,6 +943,12 @@ void rtsx_pci_send_cmd_no_wait(struct rtsx_pcr *pcr);
 int rtsx_pci_send_cmd(struct rtsx_pcr *pcr, int timeout);
 int rtsx_pci_transfer_data(struct rtsx_pcr *pcr, struct scatterlist *sglist,
 		int num_sg, bool read, int timeout);
+int rtsx_pci_dma_map_sg(struct rtsx_pcr *pcr, struct scatterlist *sglist,
+		int num_sg, bool read);
+int rtsx_pci_dma_unmap_sg(struct rtsx_pcr *pcr, struct scatterlist *sglist,
+		int num_sg, bool read);
+int rtsx_pci_dma_transfer(struct rtsx_pcr *pcr, struct scatterlist *sglist,
+		int sg_count, bool read);
 int rtsx_pci_read_ppbuf(struct rtsx_pcr *pcr, u8 *buf, int buf_len);
 int rtsx_pci_write_ppbuf(struct rtsx_pcr *pcr, u8 *buf, int buf_len);
 int rtsx_pci_card_pull_ctl_enable(struct rtsx_pcr *pcr, int card);
-- 
1.7.9.5

Re: [PATCH v2 0/3] mmc: rtsx: fix bug and support nonblocking request

[micky]

Hi Chris,

Could you please give me some idea about this patch-set?
I send this patch-set many times, but didn't get any reply.

Best Regards.
micky <micky_ching@realsil.com.cn>

On 02/17/2014 04:45 PM, micky_ching@realsil.com.cn wrote:
> From: Micky Ching <micky_ching@realsil.com.cn>
>
> First we fix the card poweroff bug: the card power is not shutdown when sd/mmc
> card removed, this will make UHS-card failed to running in high speed mode if we
> insert the card again.
>
> We offer a concise tuning searching method, it is much easier to read.
>
> At last we add support for nonblocking request, which will improve card read
> speed at about 10% write speed about 5% normally.
>
> Micky Ching (3):
>    mmc: rtsx: fix card poweroff bug
>    mmc: rtsx: modify phase searching method for tuning
>    mmc: rtsx: add support for pre_req and post_req
>
>   drivers/mfd/rtsx_pcr.c            |  132 ++++++---
>   drivers/mmc/host/rtsx_pci_sdmmc.c |  529 ++++++++++++++++++++++++++-----------
>   include/linux/mfd/rtsx_common.h   |    1 +
>   include/linux/mfd/rtsx_pci.h      |    8 +-
>   4 files changed, 481 insertions(+), 189 deletions(-)
>
> --
> 1.7.9.5
> .
>

Re: [PATCH v2 0/3] mmc: rtsx: fix bug and support nonblocking request

[micky]

Hi Chris,

Could you please give me some idea about this patch-set?
I send this patch-set many times, but didn't get any reply.

Best Regards.
micky <micky_ching@realsil.com.cn>

On 02/17/2014 04:45 PM, micky_ching@realsil.com.cn wrote:
> From: Micky Ching <micky_ching@realsil.com.cn>
>
> First we fix the card poweroff bug: the card power is not shutdown when sd/mmc
> card removed, this will make UHS-card failed to running in high speed mode if we
> insert the card again.
>
> We offer a concise tuning searching method, it is much easier to read.
>
> At last we add support for nonblocking request, which will improve card read
> speed at about 10% write speed about 5% normally.
>
> Micky Ching (3):
>    mmc: rtsx: fix card poweroff bug
>    mmc: rtsx: modify phase searching method for tuning
>    mmc: rtsx: add support for pre_req and post_req
>
>   drivers/mfd/rtsx_pcr.c            |  132 ++++++---
>   drivers/mmc/host/rtsx_pci_sdmmc.c |  529 ++++++++++++++++++++++++++-----------
>   include/linux/mfd/rtsx_common.h   |    1 +
>   include/linux/mfd/rtsx_pci.h      |    8 +-
>   4 files changed, 481 insertions(+), 189 deletions(-)
>
> --
> 1.7.9.5
> .
>

Re: [PATCH v2 0/3] mmc: rtsx: fix bug and support nonblocking request

[Chris Ball]

Hi Micky,

On Mon, Feb 17 2014, micky_ching@realsil.com.cn wrote:
> From: Micky Ching <micky_ching@realsil.com.cn>
>
> First we fix the card poweroff bug: the card power is not shutdown when sd/mmc
> card removed, this will make UHS-card failed to running in high speed mode if we
> insert the card again.
>
> We offer a concise tuning searching method, it is much easier to read.
>
> At last we add support for nonblocking request, which will improve card read
> speed at about 10% write speed about 5% normally.

Thanks, all 3 patches pushed to mmc-next for 3.15.

- Chris.
-- 
Chris Ball   <chris@printf.net>   <http://printf.net/>

Re: [PATCH v2 0/3] mmc: rtsx: fix bug and support nonblocking request

[Chris Ball]

Hi Micky,

On Mon, Feb 17 2014, micky_ching@realsil.com.cn wrote:
> From: Micky Ching <micky_ching@realsil.com.cn>
>
> First we fix the card poweroff bug: the card power is not shutdown when sd/mmc
> card removed, this will make UHS-card failed to running in high speed mode if we
> insert the card again.
>
> We offer a concise tuning searching method, it is much easier to read.
>
> At last we add support for nonblocking request, which will improve card read
> speed at about 10% write speed about 5% normally.

Thanks, all 3 patches pushed to mmc-next for 3.15.

- Chris.
-- 
Chris Ball   <chris@printf.net>   <http://printf.net/>