[PATCH RFC 0/2] scatterlist: add I/O variant of sg_pcopy & sg_copy and use them

[PATCH RFC 0/2] scatterlist: add I/O variant of sg_pcopy & sg_copy and use them

[Neil Armstrong]

A local variant of sg_copy_buffer has been introduced in the meson-gx mmc driver [1] after
a mempcy optimization, fixing the iomem buffer manipulation and the reported system crash [2].

But, the fix is suboptimal in terms of performace/ugliness [3] and a proper I/O variant of
sg_copy_buffer should be added and used instead.

[1] https://lore.kernel.org/r/20210609150230.9291-1-narmstrong@baylibre.com
[2] https://lore.kernel.org/r/acb244ad-0759-5a96-c659-5c23003d3dcd@samsung.com
[3] https://lore.kernel.org/r/CAPDyKFrLSMpPJOgd5e4B1x3Vwfg4q23zgy4ESc8EmFL2MnyK7g@mail.gmail.com

Neil Armstrong (2):
  scatterlist: add I/O variant of sg_pcopy & sg_copy
  mmc: meson-gx: use sg_copy_to/from_io instead of local version

 drivers/mmc/host/meson-gx-mmc.c |  53 +++-----------
 include/linux/scatterlist.h     |  14 ++++
 lib/scatterlist.c               | 119 ++++++++++++++++++++++++++++++++
 3 files changed, 143 insertions(+), 43 deletions(-)

-- 
2.25.1

[PATCH RFC 1/2] scatterlist: add I/O variant of sg_pcopy & sg_copy

[Neil Armstrong]

When copying from/to an iomem mapped memory, the current sg_copy & sg_pcopy can't
be used and lead to local variants in drivers like in [1] & [2].

This introduces an I/O variant to be used instead of the local variants.

[1] mv_cesa_sg_copy in drivers/crypto/marvell/cesa/tdma.c
[2] meson_mmc_copy_buffer in drivers/mmc/host/meson-gx-mmc.c

Cc: Robin Murphy <robin.murphy@arm.com>
Signed-off-by: Neil Armstrong <narmstrong@baylibre.com>
---
 include/linux/scatterlist.h |  14 +++++
 lib/scatterlist.c           | 119 ++++++++++++++++++++++++++++++++++++
 2 files changed, 133 insertions(+)

diff --git a/include/linux/scatterlist.h b/include/linux/scatterlist.h
index 6f70572b2938..6ef339ba5290 100644
--- a/include/linux/scatterlist.h
+++ b/include/linux/scatterlist.h
@@ -308,15 +308,29 @@ void sgl_free(struct scatterlist *sgl);
 size_t sg_copy_buffer(struct scatterlist *sgl, unsigned int nents, void *buf,
 		      size_t buflen, off_t skip, bool to_buffer);
 
+size_t sg_copy_io(struct scatterlist *sgl, unsigned int nents, void __iomem *buf,
+		  size_t buflen, off_t skip, bool to_buffer);
+
 size_t sg_copy_from_buffer(struct scatterlist *sgl, unsigned int nents,
 			   const void *buf, size_t buflen);
 size_t sg_copy_to_buffer(struct scatterlist *sgl, unsigned int nents,
 			 void *buf, size_t buflen);
 
+size_t sg_copy_from_io(struct scatterlist *sgl, unsigned int nents,
+		       const void __iomem *buf, size_t buflen);
+size_t sg_copy_to_io(struct scatterlist *sgl, unsigned int nents,
+		     void __iomem *buf, size_t buflen);
+
 size_t sg_pcopy_from_buffer(struct scatterlist *sgl, unsigned int nents,
 			    const void *buf, size_t buflen, off_t skip);
 size_t sg_pcopy_to_buffer(struct scatterlist *sgl, unsigned int nents,
 			  void *buf, size_t buflen, off_t skip);
+
+size_t sg_pcopy_from_io(struct scatterlist *sgl, unsigned int nents,
+			const void __iomem *buf, size_t buflen, off_t skip);
+size_t sg_pcopy_to_io(struct scatterlist *sgl, unsigned int nents,
+		      void __iomem *buf, size_t buflen, off_t skip);
+
 size_t sg_zero_buffer(struct scatterlist *sgl, unsigned int nents,
 		       size_t buflen, off_t skip);
 
diff --git a/lib/scatterlist.c b/lib/scatterlist.c
index a59778946404..e52f37b181fa 100644
--- a/lib/scatterlist.c
+++ b/lib/scatterlist.c
@@ -954,6 +954,55 @@ size_t sg_copy_buffer(struct scatterlist *sgl, unsigned int nents, void *buf,
 }
 EXPORT_SYMBOL(sg_copy_buffer);
 
+/**
+ * sg_copy_io - Copy data between an I/O mapped buffer and an SG list
+ * @sgl:		 The SG list
+ * @nents:		 Number of SG entries
+ * @buf:		 Where to copy from
+ * @buflen:		 The number of bytes to copy
+ * @skip:		 Number of bytes to skip before copying
+ * @to_buffer:		 transfer direction (true == from an sg list to a
+ *			 buffer, false == from a buffer to an sg list)
+ *
+ * Returns the number of copied bytes.
+ *
+ **/
+size_t sg_copy_io(struct scatterlist *sgl, unsigned int nents, void __iomem *buf,
+		  size_t buflen, off_t skip, bool to_buffer)
+{
+	unsigned int offset = 0;
+	struct sg_mapping_iter miter;
+	unsigned int sg_flags = SG_MITER_ATOMIC;
+
+	if (to_buffer)
+		sg_flags |= SG_MITER_FROM_SG;
+	else
+		sg_flags |= SG_MITER_TO_SG;
+
+	sg_miter_start(&miter, sgl, nents, sg_flags);
+
+	if (!sg_miter_skip(&miter, skip))
+		return 0;
+
+	while ((offset < buflen) && sg_miter_next(&miter)) {
+		unsigned int len;
+
+		len = min(miter.length, buflen - offset);
+
+		if (to_buffer)
+			memcpy_toio(buf + offset, miter.addr, len);
+		else
+			memcpy_fromio(miter.addr, buf + offset, len);
+
+		offset += len;
+	}
+
+	sg_miter_stop(&miter);
+
+	return offset;
+}
+EXPORT_SYMBOL(sg_copy_io);
+
 /**
  * sg_copy_from_buffer - Copy from a linear buffer to an SG list
  * @sgl:		 The SG list
@@ -988,6 +1037,40 @@ size_t sg_copy_to_buffer(struct scatterlist *sgl, unsigned int nents,
 }
 EXPORT_SYMBOL(sg_copy_to_buffer);
 
+/**
+ * sg_copy_from_io - Copy from an I/O mapped buffer to an SG list
+ * @sgl:		 The SG list
+ * @nents:		 Number of SG entries
+ * @buf:		 Where to copy from
+ * @buflen:		 The number of bytes to copy
+ *
+ * Returns the number of copied bytes.
+ *
+ **/
+size_t sg_copy_from_io(struct scatterlist *sgl, unsigned int nents,
+		       const void *buf, size_t buflen)
+{
+	return sg_copy_io(sgl, nents, (void __iomem *)buf, buflen, 0, false);
+}
+EXPORT_SYMBOL(sg_copy_from_io);
+
+/**
+ * sg_copy_to_io - Copy from an SG list to an I/O mapped buffer
+ * @sgl:		 The SG list
+ * @nents:		 Number of SG entries
+ * @buf:		 Where to copy to
+ * @buflen:		 The number of bytes to copy
+ *
+ * Returns the number of copied bytes.
+ *
+ **/
+size_t sg_copy_to_io(struct scatterlist *sgl, unsigned int nents,
+		     void __iomem *buf, size_t buflen)
+{
+	return sg_copy_io(sgl, nents, buf, buflen, 0, true);
+}
+EXPORT_SYMBOL(sg_copy_to_io);
+
 /**
  * sg_pcopy_from_buffer - Copy from a linear buffer to an SG list
  * @sgl:		 The SG list
@@ -1024,6 +1107,42 @@ size_t sg_pcopy_to_buffer(struct scatterlist *sgl, unsigned int nents,
 }
 EXPORT_SYMBOL(sg_pcopy_to_buffer);
 
+/**
+ * sg_pcopy_from_io - Copy from an I/O mapped buffer to an SG list
+ * @sgl:		 The SG list
+ * @nents:		 Number of SG entries
+ * @buf:		 Where to copy from
+ * @buflen:		 The number of bytes to copy
+ * @skip:		 Number of bytes to skip before copying
+ *
+ * Returns the number of copied bytes.
+ *
+ **/
+size_t sg_pcopy_from_io(struct scatterlist *sgl, unsigned int nents,
+			const void __iomem *buf, size_t buflen, off_t skip)
+{
+	return sg_copy_io(sgl, nents, (void __iomem *)buf, buflen, skip, false);
+}
+EXPORT_SYMBOL(sg_pcopy_from_io);
+
+/**
+ * sg_pcopy_to_io - Copy from an SG list to an I/O mapped buffer
+ * @sgl:		 The SG list
+ * @nents:		 Number of SG entries
+ * @buf:		 Where to copy to
+ * @buflen:		 The number of bytes to copy
+ * @skip:		 Number of bytes to skip before copying
+ *
+ * Returns the number of copied bytes.
+ *
+ **/
+size_t sg_pcopy_to_io(struct scatterlist *sgl, unsigned int nents,
+		      void __iomem *buf, size_t buflen, off_t skip)
+{
+	return sg_copy_io(sgl, nents, buf, buflen, skip, true);
+}
+EXPORT_SYMBOL(sg_pcopy_to_io);
+
 /**
  * sg_zero_buffer - Zero-out a part of a SG list
  * @sgl:		 The SG list
-- 
2.25.1

Re: [PATCH RFC 1/2] scatterlist: add I/O variant of sg_pcopy & sg_copy

[Robin Murphy]

On 2021-06-28 13:34, Neil Armstrong wrote:
> When copying from/to an iomem mapped memory, the current sg_copy & sg_pcopy can't
> be used and lead to local variants in drivers like in [1] & [2].
> 
> This introduces an I/O variant to be used instead of the local variants.
> 
> [1] mv_cesa_sg_copy in drivers/crypto/marvell/cesa/tdma.c
> [2] meson_mmc_copy_buffer in drivers/mmc/host/meson-gx-mmc.c

Other than one apparent typo below, this looks like what I imagined, 
thanks for putting it together!

> Cc: Robin Murphy <robin.murphy@arm.com>
> Signed-off-by: Neil Armstrong <narmstrong@baylibre.com>
> ---
>   include/linux/scatterlist.h |  14 +++++
>   lib/scatterlist.c           | 119 ++++++++++++++++++++++++++++++++++++
>   2 files changed, 133 insertions(+)
> 
> diff --git a/include/linux/scatterlist.h b/include/linux/scatterlist.h
> index 6f70572b2938..6ef339ba5290 100644
> --- a/include/linux/scatterlist.h
> +++ b/include/linux/scatterlist.h
> @@ -308,15 +308,29 @@ void sgl_free(struct scatterlist *sgl);
>   size_t sg_copy_buffer(struct scatterlist *sgl, unsigned int nents, void *buf,
>   		      size_t buflen, off_t skip, bool to_buffer);
>   
> +size_t sg_copy_io(struct scatterlist *sgl, unsigned int nents, void __iomem *buf,
> +		  size_t buflen, off_t skip, bool to_buffer);
> +
>   size_t sg_copy_from_buffer(struct scatterlist *sgl, unsigned int nents,
>   			   const void *buf, size_t buflen);
>   size_t sg_copy_to_buffer(struct scatterlist *sgl, unsigned int nents,
>   			 void *buf, size_t buflen);
>   
> +size_t sg_copy_from_io(struct scatterlist *sgl, unsigned int nents,
> +		       const void __iomem *buf, size_t buflen);
> +size_t sg_copy_to_io(struct scatterlist *sgl, unsigned int nents,
> +		     void __iomem *buf, size_t buflen);
> +
>   size_t sg_pcopy_from_buffer(struct scatterlist *sgl, unsigned int nents,
>   			    const void *buf, size_t buflen, off_t skip);
>   size_t sg_pcopy_to_buffer(struct scatterlist *sgl, unsigned int nents,
>   			  void *buf, size_t buflen, off_t skip);
> +
> +size_t sg_pcopy_from_io(struct scatterlist *sgl, unsigned int nents,
> +			const void __iomem *buf, size_t buflen, off_t skip);
> +size_t sg_pcopy_to_io(struct scatterlist *sgl, unsigned int nents,
> +		      void __iomem *buf, size_t buflen, off_t skip);
> +
>   size_t sg_zero_buffer(struct scatterlist *sgl, unsigned int nents,
>   		       size_t buflen, off_t skip);
>   
> diff --git a/lib/scatterlist.c b/lib/scatterlist.c
> index a59778946404..e52f37b181fa 100644
> --- a/lib/scatterlist.c
> +++ b/lib/scatterlist.c
> @@ -954,6 +954,55 @@ size_t sg_copy_buffer(struct scatterlist *sgl, unsigned int nents, void *buf,
>   }
>   EXPORT_SYMBOL(sg_copy_buffer);
>   
> +/**
> + * sg_copy_io - Copy data between an I/O mapped buffer and an SG list
> + * @sgl:		 The SG list
> + * @nents:		 Number of SG entries
> + * @buf:		 Where to copy from
> + * @buflen:		 The number of bytes to copy
> + * @skip:		 Number of bytes to skip before copying
> + * @to_buffer:		 transfer direction (true == from an sg list to a
> + *			 buffer, false == from a buffer to an sg list)
> + *
> + * Returns the number of copied bytes.
> + *
> + **/
> +size_t sg_copy_io(struct scatterlist *sgl, unsigned int nents, void __iomem *buf,
> +		  size_t buflen, off_t skip, bool to_buffer)
> +{
> +	unsigned int offset = 0;
> +	struct sg_mapping_iter miter;
> +	unsigned int sg_flags = SG_MITER_ATOMIC;
> +
> +	if (to_buffer)
> +		sg_flags |= SG_MITER_FROM_SG;
> +	else
> +		sg_flags |= SG_MITER_TO_SG;
> +
> +	sg_miter_start(&miter, sgl, nents, sg_flags);
> +
> +	if (!sg_miter_skip(&miter, skip))
> +		return 0;
> +
> +	while ((offset < buflen) && sg_miter_next(&miter)) {
> +		unsigned int len;
> +
> +		len = min(miter.length, buflen - offset);
> +
> +		if (to_buffer)
> +			memcpy_toio(buf + offset, miter.addr, len);
> +		else
> +			memcpy_fromio(miter.addr, buf + offset, len);
> +
> +		offset += len;
> +	}
> +
> +	sg_miter_stop(&miter);
> +
> +	return offset;
> +}
> +EXPORT_SYMBOL(sg_copy_io);
> +
>   /**
>    * sg_copy_from_buffer - Copy from a linear buffer to an SG list
>    * @sgl:		 The SG list
> @@ -988,6 +1037,40 @@ size_t sg_copy_to_buffer(struct scatterlist *sgl, unsigned int nents,
>   }
>   EXPORT_SYMBOL(sg_copy_to_buffer);
>   
> +/**
> + * sg_copy_from_io - Copy from an I/O mapped buffer to an SG list
> + * @sgl:		 The SG list
> + * @nents:		 Number of SG entries
> + * @buf:		 Where to copy from
> + * @buflen:		 The number of bytes to copy
> + *
> + * Returns the number of copied bytes.
> + *
> + **/
> +size_t sg_copy_from_io(struct scatterlist *sgl, unsigned int nents,
> +		       const void *buf, size_t buflen)

The __iomem annotation wants to be on buf here raher than cast in below, 
to match the prototype (and everything else).

Cheers,
Robin.

> +{
> +	return sg_copy_io(sgl, nents, (void __iomem *)buf, buflen, 0, false);
> +}
> +EXPORT_SYMBOL(sg_copy_from_io);
> +
> +/**
> + * sg_copy_to_io - Copy from an SG list to an I/O mapped buffer
> + * @sgl:		 The SG list
> + * @nents:		 Number of SG entries
> + * @buf:		 Where to copy to
> + * @buflen:		 The number of bytes to copy
> + *
> + * Returns the number of copied bytes.
> + *
> + **/
> +size_t sg_copy_to_io(struct scatterlist *sgl, unsigned int nents,
> +		     void __iomem *buf, size_t buflen)
> +{
> +	return sg_copy_io(sgl, nents, buf, buflen, 0, true);
> +}
> +EXPORT_SYMBOL(sg_copy_to_io);
> +
>   /**
>    * sg_pcopy_from_buffer - Copy from a linear buffer to an SG list
>    * @sgl:		 The SG list
> @@ -1024,6 +1107,42 @@ size_t sg_pcopy_to_buffer(struct scatterlist *sgl, unsigned int nents,
>   }
>   EXPORT_SYMBOL(sg_pcopy_to_buffer);
>   
> +/**
> + * sg_pcopy_from_io - Copy from an I/O mapped buffer to an SG list
> + * @sgl:		 The SG list
> + * @nents:		 Number of SG entries
> + * @buf:		 Where to copy from
> + * @buflen:		 The number of bytes to copy
> + * @skip:		 Number of bytes to skip before copying
> + *
> + * Returns the number of copied bytes.
> + *
> + **/
> +size_t sg_pcopy_from_io(struct scatterlist *sgl, unsigned int nents,
> +			const void __iomem *buf, size_t buflen, off_t skip)
> +{
> +	return sg_copy_io(sgl, nents, (void __iomem *)buf, buflen, skip, false);
> +}
> +EXPORT_SYMBOL(sg_pcopy_from_io);
> +
> +/**
> + * sg_pcopy_to_io - Copy from an SG list to an I/O mapped buffer
> + * @sgl:		 The SG list
> + * @nents:		 Number of SG entries
> + * @buf:		 Where to copy to
> + * @buflen:		 The number of bytes to copy
> + * @skip:		 Number of bytes to skip before copying
> + *
> + * Returns the number of copied bytes.
> + *
> + **/
> +size_t sg_pcopy_to_io(struct scatterlist *sgl, unsigned int nents,
> +		      void __iomem *buf, size_t buflen, off_t skip)
> +{
> +	return sg_copy_io(sgl, nents, buf, buflen, skip, true);
> +}
> +EXPORT_SYMBOL(sg_pcopy_to_io);
> +
>   /**
>    * sg_zero_buffer - Zero-out a part of a SG list
>    * @sgl:		 The SG list
> 

[PATCH RFC 2/2] mmc: meson-gx: use sg_copy_to/from_io instead of local version

[Neil Armstrong]

Use the proper sg_copy_to_io & sg_copy_from_io instead of having a local
sg_copy_buffer variant to handle the I/O mapped buffer case.

Cc: Robin Murphy <robin.murphy@arm.com>
Signed-off-by: Neil Armstrong <narmstrong@baylibre.com>
---
 drivers/mmc/host/meson-gx-mmc.c | 53 +++++++--------------------------
 1 file changed, 10 insertions(+), 43 deletions(-)

diff --git a/drivers/mmc/host/meson-gx-mmc.c b/drivers/mmc/host/meson-gx-mmc.c
index 3f28eb4d17fe..c13436efb414 100644
--- a/drivers/mmc/host/meson-gx-mmc.c
+++ b/drivers/mmc/host/meson-gx-mmc.c
@@ -746,47 +746,6 @@ static void meson_mmc_desc_chain_transfer(struct mmc_host *mmc, u32 cmd_cfg)
 	writel(start, host->regs + SD_EMMC_START);
 }
 
-/* local sg copy to buffer version with _to/fromio usage for dram_access_quirk */
-static void meson_mmc_copy_buffer(struct meson_host *host, struct mmc_data *data,
-				  size_t buflen, bool to_buffer)
-{
-	unsigned int sg_flags = SG_MITER_ATOMIC;
-	struct scatterlist *sgl = data->sg;
-	unsigned int nents = data->sg_len;
-	struct sg_mapping_iter miter;
-	unsigned int offset = 0;
-
-	if (to_buffer)
-		sg_flags |= SG_MITER_FROM_SG;
-	else
-		sg_flags |= SG_MITER_TO_SG;
-
-	sg_miter_start(&miter, sgl, nents, sg_flags);
-
-	while ((offset < buflen) && sg_miter_next(&miter)) {
-		unsigned int len;
-
-		len = min(miter.length, buflen - offset);
-
-		/* When dram_access_quirk, the bounce buffer is a iomem mapping */
-		if (host->dram_access_quirk) {
-			if (to_buffer)
-				memcpy_toio(host->bounce_iomem_buf + offset, miter.addr, len);
-			else
-				memcpy_fromio(miter.addr, host->bounce_iomem_buf + offset, len);
-		} else {
-			if (to_buffer)
-				memcpy(host->bounce_buf + offset, miter.addr, len);
-			else
-				memcpy(miter.addr, host->bounce_buf + offset, len);
-		}
-
-		offset += len;
-	}
-
-	sg_miter_stop(&miter);
-}
-
 static void meson_mmc_start_cmd(struct mmc_host *mmc, struct mmc_command *cmd)
 {
 	struct meson_host *host = mmc_priv(mmc);
@@ -830,7 +789,12 @@ static void meson_mmc_start_cmd(struct mmc_host *mmc, struct mmc_command *cmd)
 		if (data->flags & MMC_DATA_WRITE) {
 			cmd_cfg |= CMD_CFG_DATA_WR;
 			WARN_ON(xfer_bytes > host->bounce_buf_size);
-			meson_mmc_copy_buffer(host, data, xfer_bytes, true);
+			if (host->dram_access_quirk)
+				sg_copy_to_io(data->sg, data->sg_len,
+					      host->bounce_iomem_buf, xfer_bytes);
+			else
+				sg_copy_to_buffer(data->sg, data->sg_len,
+						  host->bounce_buf, xfer_bytes);
 			dma_wmb();
 		}
 
@@ -999,7 +963,10 @@ static irqreturn_t meson_mmc_irq_thread(int irq, void *dev_id)
 	if (meson_mmc_bounce_buf_read(data)) {
 		xfer_bytes = data->blksz * data->blocks;
 		WARN_ON(xfer_bytes > host->bounce_buf_size);
-		meson_mmc_copy_buffer(host, data, xfer_bytes, false);
+		if (host->dram_access_quirk)
+			sg_copy_from_io(data->sg, data->sg_len, host->bounce_iomem_buf, xfer_bytes);
+		else
+			sg_copy_from_buffer(data->sg, data->sg_len, host->bounce_buf, xfer_bytes);
 	}
 
 	next_cmd = meson_mmc_get_next_command(cmd);
-- 
2.25.1

Re: [PATCH RFC 2/2] mmc: meson-gx: use sg_copy_to/from_io instead of local version

[Robin Murphy]

On 2021-06-28 13:34, Neil Armstrong wrote:
> Use the proper sg_copy_to_io & sg_copy_from_io instead of having a local
> sg_copy_buffer variant to handle the I/O mapped buffer case.
> 
> Cc: Robin Murphy <robin.murphy@arm.com>
> Signed-off-by: Neil Armstrong <narmstrong@baylibre.com>
> ---
>   drivers/mmc/host/meson-gx-mmc.c | 53 +++++++--------------------------
>   1 file changed, 10 insertions(+), 43 deletions(-)
> 
> diff --git a/drivers/mmc/host/meson-gx-mmc.c b/drivers/mmc/host/meson-gx-mmc.c
> index 3f28eb4d17fe..c13436efb414 100644
> --- a/drivers/mmc/host/meson-gx-mmc.c
> +++ b/drivers/mmc/host/meson-gx-mmc.c
> @@ -746,47 +746,6 @@ static void meson_mmc_desc_chain_transfer(struct mmc_host *mmc, u32 cmd_cfg)
>   	writel(start, host->regs + SD_EMMC_START);
>   }
>   
> -/* local sg copy to buffer version with _to/fromio usage for dram_access_quirk */
> -static void meson_mmc_copy_buffer(struct meson_host *host, struct mmc_data *data,
> -				  size_t buflen, bool to_buffer)
> -{
> -	unsigned int sg_flags = SG_MITER_ATOMIC;
> -	struct scatterlist *sgl = data->sg;
> -	unsigned int nents = data->sg_len;
> -	struct sg_mapping_iter miter;
> -	unsigned int offset = 0;
> -
> -	if (to_buffer)
> -		sg_flags |= SG_MITER_FROM_SG;
> -	else
> -		sg_flags |= SG_MITER_TO_SG;
> -
> -	sg_miter_start(&miter, sgl, nents, sg_flags);
> -
> -	while ((offset < buflen) && sg_miter_next(&miter)) {
> -		unsigned int len;
> -
> -		len = min(miter.length, buflen - offset);
> -
> -		/* When dram_access_quirk, the bounce buffer is a iomem mapping */
> -		if (host->dram_access_quirk) {
> -			if (to_buffer)
> -				memcpy_toio(host->bounce_iomem_buf + offset, miter.addr, len);
> -			else
> -				memcpy_fromio(miter.addr, host->bounce_iomem_buf + offset, len);
> -		} else {
> -			if (to_buffer)
> -				memcpy(host->bounce_buf + offset, miter.addr, len);
> -			else
> -				memcpy(miter.addr, host->bounce_buf + offset, len);
> -		}
> -
> -		offset += len;
> -	}
> -
> -	sg_miter_stop(&miter);
> -}
> -
>   static void meson_mmc_start_cmd(struct mmc_host *mmc, struct mmc_command *cmd)
>   {
>   	struct meson_host *host = mmc_priv(mmc);
> @@ -830,7 +789,12 @@ static void meson_mmc_start_cmd(struct mmc_host *mmc, struct mmc_command *cmd)
>   		if (data->flags & MMC_DATA_WRITE) {
>   			cmd_cfg |= CMD_CFG_DATA_WR;
>   			WARN_ON(xfer_bytes > host->bounce_buf_size);
> -			meson_mmc_copy_buffer(host, data, xfer_bytes, true);
> +			if (host->dram_access_quirk)
> +				sg_copy_to_io(data->sg, data->sg_len,
> +					      host->bounce_iomem_buf, xfer_bytes);

Maybe you could just use host->regs + SD_EMMC_SRAM_DATA_BUF_OFF directly 
here (and below) and save carrying host->bounce_iomem_buf around?

Robin.

> +			else
> +				sg_copy_to_buffer(data->sg, data->sg_len,
> +						  host->bounce_buf, xfer_bytes);
>   			dma_wmb();
>   		}
>   
> @@ -999,7 +963,10 @@ static irqreturn_t meson_mmc_irq_thread(int irq, void *dev_id)
>   	if (meson_mmc_bounce_buf_read(data)) {
>   		xfer_bytes = data->blksz * data->blocks;
>   		WARN_ON(xfer_bytes > host->bounce_buf_size);
> -		meson_mmc_copy_buffer(host, data, xfer_bytes, false);
> +		if (host->dram_access_quirk)
> +			sg_copy_from_io(data->sg, data->sg_len, host->bounce_iomem_buf, xfer_bytes);
> +		else
> +			sg_copy_from_buffer(data->sg, data->sg_len, host->bounce_buf, xfer_bytes);
>   	}
>   
>   	next_cmd = meson_mmc_get_next_command(cmd);
>