mmc: sdhci: Use pre-allocated DMA buffers

Use pre-allocated DMA buffers for ADMA descriptor and Bounce buffer
instead of dynamic DMA mapping.
This improves SDHCI driver performance by reducing dynamic DMA mapping
overhead.

Bug 1486735

Change-Id: Ic9c646437be047d33304339eccc48a825f0a8bcc
Reviewed-on: http://git-master/r/380885
Cherry-picked from commit 7ffcc4cf1a1cec42610c1b55c30b3ec28547a11e

Signed-off-by: Jinyoung Park <jinyoungp@nvidia.com>
Signed-off-by: Naveen Kumar Arepalli <naveenk@nvidia.com>
Change-Id: If850a534ba9fbfd169b4fbefd35ca5922b1d1254
Reviewed-on: http://git-master/r/416955
Reviewed-by: Automatic_Commit_Validation_User
GVS: Gerrit_Virtual_Submit
Reviewed-by: Venu Byravarasu <vbyravarasu@nvidia.com>
Reviewed-by: R Raj Kumar <rrajk@nvidia.com>
Reviewed-by: Laxman Dewangan <ldewangan@nvidia.com>
(cherry picked from commit 9e58888afe4e66e83eece0a8332c8e7440bd1bcf)
Reviewed-on: http://git-master/r/419444

1 files changed, 60 insertions, 18 deletions

diff --git a/drivers/mmc/host/sdhci.c b/drivers/mmc/host/sdhci.c
index 9168f676ec2c..8950f641deab 100644
--- a/drivers/mmc/host/sdhci.c
+++ b/drivers/mmc/host/sdhci.c
@@ -556,11 +556,13 @@ static int sdhci_adma_table_pre(struct sdhci_host *host,
 	 * need to fill it with data first.
 	 */
 
-	host->align_addr = dma_map_single(mmc_dev(host->mmc),
-		host->align_buffer, 128 * 8, direction);
-	if (dma_mapping_error(mmc_dev(host->mmc), host->align_addr))
-		goto fail;
-	BUG_ON(host->align_addr & 0x3);
+	if (!host->use_dma_alloc) {
+		host->align_addr = dma_map_single(mmc_dev(host->mmc),
+			host->align_buffer, 128 * 8, direction);
+		if (dma_mapping_error(mmc_dev(host->mmc), host->align_addr))
+			goto fail;
+		BUG_ON(host->align_addr & 0x3);
+	}
 
 	host->sg_count = dma_map_sg(mmc_dev(host->mmc),
 		data->sg, data->sg_len, direction);
@@ -656,11 +658,13 @@ static int sdhci_adma_table_pre(struct sdhci_host *host,
 			host->align_addr, 128 * 8, direction);
 	}
 
-	host->adma_addr = dma_map_single(mmc_dev(host->mmc),
-		host->adma_desc, (128 * 2 + 1) * 8, DMA_TO_DEVICE);
-	if (dma_mapping_error(mmc_dev(host->mmc), host->adma_addr))
-		goto unmap_entries;
-	BUG_ON(host->adma_addr & 0x3);
+	if (!host->use_dma_alloc) {
+		host->adma_addr = dma_map_single(mmc_dev(host->mmc),
+			host->adma_desc, (128 * 2 + 1) * 8, DMA_TO_DEVICE);
+		if (dma_mapping_error(mmc_dev(host->mmc), host->adma_addr))
+			goto unmap_entries;
+		BUG_ON(host->adma_addr & 0x3);
+	}
 
 	return 0;
 
@@ -668,8 +672,9 @@ unmap_entries:
 	dma_unmap_sg(mmc_dev(host->mmc), data->sg,
 		data->sg_len, direction);
 unmap_align:
-	dma_unmap_single(mmc_dev(host->mmc), host->align_addr,
-		128 * 8, direction);
+	if (!host->use_dma_alloc)
+		dma_unmap_single(mmc_dev(host->mmc), host->align_addr,
+				128 * 8, direction);
 fail:
 	return -EINVAL;
 }
@@ -690,11 +695,13 @@ static void sdhci_adma_table_post(struct sdhci_host *host,
 	else
 		direction = DMA_TO_DEVICE;
 
-	dma_unmap_single(mmc_dev(host->mmc), host->adma_addr,
-		(128 * 2 + 1) * 8, DMA_TO_DEVICE);
+	if (!host->use_dma_alloc) {
+		dma_unmap_single(mmc_dev(host->mmc), host->adma_addr,
+			(128 * 2 + 1) * 8, DMA_TO_DEVICE);
 
-	dma_unmap_single(mmc_dev(host->mmc), host->align_addr,
-		128 * 8, direction);
+		dma_unmap_single(mmc_dev(host->mmc), host->align_addr,
+			128 * 8, direction);
+	}
 
 	if (data->flags & MMC_DATA_READ) {
 		dma_sync_sg_for_cpu(mmc_dev(host->mmc), data->sg,
@@ -3535,6 +3542,33 @@ int sdhci_add_host(struct sdhci_host *host)
 		 * each of those entries. Simply allocating 128 bits
 		 * for each entry
 		 */
+		if (mmc_dev(host->mmc)->dma_mask &&
+				mmc_dev(host->mmc)->coherent_dma_mask) {
+			host->adma_desc = dma_alloc_coherent(
+					mmc_dev(host->mmc), (128 * 2 + 1) * 8,
+					&host->adma_addr, GFP_KERNEL);
+			if (!host->adma_desc)
+				goto err_dma_alloc;
+
+			host->align_buffer = dma_alloc_coherent(
+					mmc_dev(host->mmc), 128 * 8,
+					&host->align_addr, GFP_KERNEL);
+			if (!host->align_buffer) {
+				dma_free_coherent(mmc_dev(host->mmc),
+						(128 * 2 + 1) * 8,
+						host->adma_desc,
+						host->adma_addr);
+				host->adma_desc = NULL;
+				goto err_dma_alloc;
+			}
+
+			host->use_dma_alloc = true;
+
+			BUG_ON(host->adma_addr & 0x3);
+			BUG_ON(host->align_addr & 0x3);
+			goto out_dma_alloc;
+		}
+err_dma_alloc:
 		host->adma_desc = kmalloc((128 * 2 + 1) * 8, GFP_KERNEL);
 		host->align_buffer = kmalloc(128 * 8, GFP_KERNEL);
 		if (!host->adma_desc || !host->align_buffer) {
@@ -3546,6 +3580,7 @@ int sdhci_add_host(struct sdhci_host *host)
 			host->flags &= ~SDHCI_USE_ADMA;
 		}
 	}
+out_dma_alloc:
 
 	/*
 	 * If we use DMA, then it's up to the caller to set the DMA
@@ -4031,8 +4066,15 @@ void sdhci_remove_host(struct sdhci_host *host, int dead)
 		regulator_put(host->vqmmc);
 	}
 
-	kfree(host->adma_desc);
-	kfree(host->align_buffer);
+	if (host->use_dma_alloc) {
+		dma_free_coherent(mmc_dev(host->mmc), (128 * 2 + 1) * 8,
+				host->adma_desc, host->adma_addr);
+		dma_free_coherent(mmc_dev(host->mmc), 128 * 8,
+				host->align_buffer, host->align_addr);
+	} else {
+		kfree(host->adma_desc);
+		kfree(host->align_buffer);
+	}
 
 	host->adma_desc = NULL;
 	host->align_buffer = NULL;