Skip to contentSkip to conversation

mmc: tegra: Use proper DMA mask

PATCH Superseded Thierry Reding 3 participants 7 messages Sep 20, 2019 – Sep 23, 2019

Subsystems (3)

v1 of 2 Superseded Latest is v2 →

lore
Share & tools
lore
#0 lore raw

cover mmc: tegra: Use proper DMA mask

From: Thierry Reding <[email protected]>

Hi,

this puts together the two patches from this thread:

	http://patchwork.ozlabs.org/patch/1146705/

there hasn't been any activity there for a couple of weeks, so I thought I'd just throw them together and update the second patch to work on top of the first one.

The goal here is to avoid the use of bounce buffers and fix IOVA space issues when an IOMMU gets involved. The first patch introduces the ->set_dma_mask() host operation as proposed by Adrian and the second patch is just a rebase of Nicolin's Tegra SDHCI patch that uses the new host operation rather than doing it as part of ->enable_dma().

Thierry

Adrian Hunter (1):
mmc: sdhci: Let drivers define their DMA mask

Nicolin Chen (1):
mmc: tegra: Implement ->set_dma_mask()

 drivers/mmc/host/sdhci-tegra.c | 48 ++++++++++++++++++++--------------
 drivers/mmc/host/sdhci.c       | 12 +++------
 drivers/mmc/host/sdhci.h       |  1 +
 3 files changed, 33 insertions(+), 28 deletions(-)
Signature · 1 line
2.23.0
#1 lore raw

PATCH 1/2 mmc: sdhci: Let drivers define their DMA mask

From: Adrian Hunter <[email protected]>

Add host operation ->set_dma_mask() so that drivers can define their own DMA masks.

---
 drivers/mmc/host/sdhci.c | 12 ++++--------
 drivers/mmc/host/sdhci.h |  1 +
 2 files changed, 5 insertions(+), 8 deletions(-)
drivers/mmc/host/sdhci.c +4 -8
index a5dc5aae973e..bc04c3180477 100644
@@ -3756,18 +3756,14 @@ int sdhci_setup_host(struct sdhci_host *host)
 		host->flags &= ~SDHCI_USE_ADMA;
 	}
 
-	/*
-	 * It is assumed that a 64-bit capable device has set a 64-bit DMA mask
-	 * and *must* do 64-bit DMA.  A driver has the opportunity to change
-	 * that during the first call to ->enable_dma().  Similarly
-	 * SDHCI_QUIRK2_BROKEN_64_BIT_DMA must be left to the drivers to
-	 * implement.
-	 */
 	if (sdhci_can_64bit_dma(host))
 		host->flags |= SDHCI_USE_64_BIT_DMA;
 
 	if (host->flags & (SDHCI_USE_SDMA | SDHCI_USE_ADMA)) {
-		ret = sdhci_set_dma_mask(host);
+		if (host->ops->set_dma_mask)
+			ret = host->ops->set_dma_mask(host);
+		else
+			ret = sdhci_set_dma_mask(host);
 
 		if (!ret && host->ops->enable_dma)
 			ret = host->ops->enable_dma(host);
drivers/mmc/host/sdhci.h +1 -0
index 902f855efe8f..8285498c0d8a 100644
@@ -622,6 +622,7 @@ struct sdhci_ops {
 
 	u32		(*irq)(struct sdhci_host *host, u32 intmask);
 
+	int		(*set_dma_mask)(struct sdhci_host *host);
 	int		(*enable_dma)(struct sdhci_host *host);
 	unsigned int	(*get_max_clock)(struct sdhci_host *host);
 	unsigned int	(*get_min_clock)(struct sdhci_host *host);
Signature · 1 line
2.23.0
Tested-by on v1 1/2
#2 lore raw

Re: [PATCH 1/2] mmc: sdhci: Let drivers define their DMA mask

Quoted 7 lines · Thierry Reding

From: Adrian Hunter <[email protected]>

Add host operation ->set_dma_mask() so that drivers can define their own DMA masks.

Ran a boot test with both patches on a Tegra186 board.

Thanks!

Quoted 47 lines
---
 drivers/mmc/host/sdhci.c | 12 ++++--------
 drivers/mmc/host/sdhci.h |  1 +
 2 files changed, 5 insertions(+), 8 deletions(-)
drivers/mmc/host/sdhci.c +4 -8
index a5dc5aae973e..bc04c3180477 100644
@@ -3756,18 +3756,14 @@ int sdhci_setup_host(struct sdhci_host *host)
 		host->flags &= ~SDHCI_USE_ADMA;
 	}
 
-	/*
-	 * It is assumed that a 64-bit capable device has set a 64-bit DMA mask
-	 * and *must* do 64-bit DMA.  A driver has the opportunity to change
-	 * that during the first call to ->enable_dma().  Similarly
-	 * SDHCI_QUIRK2_BROKEN_64_BIT_DMA must be left to the drivers to
-	 * implement.
-	 */
 	if (sdhci_can_64bit_dma(host))
 		host->flags |= SDHCI_USE_64_BIT_DMA;
 
 	if (host->flags & (SDHCI_USE_SDMA | SDHCI_USE_ADMA)) {
-		ret = sdhci_set_dma_mask(host);
+		if (host->ops->set_dma_mask)
+			ret = host->ops->set_dma_mask(host);
+		else
+			ret = sdhci_set_dma_mask(host);
 
 		if (!ret && host->ops->enable_dma)
 			ret = host->ops->enable_dma(host);
drivers/mmc/host/sdhci.h +1 -0
index 902f855efe8f..8285498c0d8a 100644
@@ -622,6 +622,7 @@ struct sdhci_ops {
 
 	u32		(*irq)(struct sdhci_host *host, u32 intmask);
 
+	int		(*set_dma_mask)(struct sdhci_host *host);
 	int		(*enable_dma)(struct sdhci_host *host);
 	unsigned int	(*get_max_clock)(struct sdhci_host *host);
 	unsigned int	(*get_min_clock)(struct sdhci_host *host);
Signature · 1 line
2.23.0
#3 lore raw

PATCH 2/2 mmc: tegra: Implement ->set_dma_mask()

In reply to Thierry Reding

From: Nicolin Chen <[email protected]>

The SDHCI controller on Tegra186 supports 40-bit addressing, which is usually enough to address all of system memory. However, if the SDHCI controller is behind an IOMMU, the address space can go beyond. This happens on Tegra186 and later where the ARM SMMU has an input address space of 48 bits. If the DMA API is backed by this ARM SMMU, the top- down IOVA allocator will cause IOV addresses to be returned that the SDHCI controller cannot access.

Unfortunately, prior to the introduction of the ->set_dma_mask() host operation, the SDHCI core would set either a 64-bit DMA mask if the controller claimed to support 64-bit addressing, or a 32-bit DMA mask otherwise.

Since the full 64 bits cannot be addressed on Tegra, this had to be worked around in commit 68481a7e1c84 ("mmc: tegra: Mark 64 bit dma broken on Tegra186") by setting the SDHCI_QUIRK2_BROKEN_64_BIT_DMA quirk, which effectively restricts the DMA mask to 32 bits.

One disadvantage of this is that dma_map_*() APIs will now try to use the swiotlb to bounce DMA to addresses beyond of the controller's DMA mask. This in turn caused degraded performance and can lead to situations where the swiotlb buffer is exhausted, which in turn leads to DMA transfers to fail.

With the recent introduction of the ->set_dma_mask() host operation, this can now be properly fixed. For each generation of Tegra, the exact supported DMA mask can be configured. This kills two birds with one
stone: it avoids the use of bounce buffers because system memory never exceeds the addressable memory range of the SDHCI controllers on these devices, and at the same time when an IOMMU is involved, it prevents IOV addresses from being allocated beyond the addressible range of the controllers.

Since the DMA mask is now properly handled, the 64-bit DMA quirk can be removed.

[[email protected]: provide more background in commit message]
---
 drivers/mmc/host/sdhci-tegra.c | 48 ++++++++++++++++++++--------------
 1 file changed, 28 insertions(+), 20 deletions(-)
drivers/mmc/host/sdhci-tegra.c +28 -20
index 02d8f524bb9e..7bc950520fd9 100644
@@ -4,6 +4,7 @@
  */
 
 #include <linux/delay.h>
+#include <linux/dma-mapping.h>
 #include <linux/err.h>
 #include <linux/module.h>
 #include <linux/init.h>
@@ -104,6 +105,7 @@
 
 struct sdhci_tegra_soc_data {
 	const struct sdhci_pltfm_data *pdata;
+	u64 dma_mask;
 	u32 nvquirks;
 	u8 min_tap_delay;
 	u8 max_tap_delay;
@@ -1233,11 +1235,25 @@ static const struct cqhci_host_ops sdhci_tegra_cqhci_ops = {
 	.update_dcmd_desc = sdhci_tegra_update_dcmd_desc,
 };
 
+static int tegra_sdhci_set_dma_mask(struct sdhci_host *host)
+{
+	struct sdhci_pltfm_host *platform = sdhci_priv(host);
+	struct sdhci_tegra *tegra = sdhci_pltfm_priv(platform);
+	const struct sdhci_tegra_soc_data *soc = tegra->soc_data;
+	struct device *dev = mmc_dev(host->mmc);
+
+	if (soc->dma_mask)
+		return dma_set_mask_and_coherent(dev, soc->dma_mask);
+
+	return 0;
+}
+
 static const struct sdhci_ops tegra_sdhci_ops = {
 	.get_ro     = tegra_sdhci_get_ro,
 	.read_w     = tegra_sdhci_readw,
 	.write_l    = tegra_sdhci_writel,
 	.set_clock  = tegra_sdhci_set_clock,
+	.set_dma_mask = tegra_sdhci_set_dma_mask,
 	.set_bus_width = sdhci_set_bus_width,
 	.reset      = tegra_sdhci_reset,
 	.platform_execute_tuning = tegra_sdhci_execute_tuning,
@@ -1257,6 +1273,7 @@ static const struct sdhci_pltfm_data sdhci_tegra20_pdata = {
 
 static const struct sdhci_tegra_soc_data soc_data_tegra20 = {
 	.pdata = &sdhci_tegra20_pdata,
+	.dma_mask = DMA_BIT_MASK(32),
 	.nvquirks = NVQUIRK_FORCE_SDHCI_SPEC_200 |
 		    NVQUIRK_ENABLE_BLOCK_GAP_DET,
 };
@@ -1283,6 +1300,7 @@ static const struct sdhci_pltfm_data sdhci_tegra30_pdata = {
 
 static const struct sdhci_tegra_soc_data soc_data_tegra30 = {
 	.pdata = &sdhci_tegra30_pdata,
+	.dma_mask = DMA_BIT_MASK(32),
 	.nvquirks = NVQUIRK_ENABLE_SDHCI_SPEC_300 |
 		    NVQUIRK_ENABLE_SDR50 |
 		    NVQUIRK_ENABLE_SDR104 |
@@ -1295,6 +1313,7 @@ static const struct sdhci_ops tegra114_sdhci_ops = {
 	.write_w    = tegra_sdhci_writew,
 	.write_l    = tegra_sdhci_writel,
 	.set_clock  = tegra_sdhci_set_clock,
+	.set_dma_mask = tegra_sdhci_set_dma_mask,
 	.set_bus_width = sdhci_set_bus_width,
 	.reset      = tegra_sdhci_reset,
 	.platform_execute_tuning = tegra_sdhci_execute_tuning,
@@ -1316,6 +1335,7 @@ static const struct sdhci_pltfm_data sdhci_tegra114_pdata = {
 
 static const struct sdhci_tegra_soc_data soc_data_tegra114 = {
 	.pdata = &sdhci_tegra114_pdata,
+	.dma_mask = DMA_BIT_MASK(32),
 };
 
 static const struct sdhci_pltfm_data sdhci_tegra124_pdata = {
@@ -1325,22 +1345,13 @@ static const struct sdhci_pltfm_data sdhci_tegra124_pdata = {
 		  SDHCI_QUIRK_NO_HISPD_BIT |
 		  SDHCI_QUIRK_BROKEN_ADMA_ZEROLEN_DESC |
 		  SDHCI_QUIRK_CAP_CLOCK_BASE_BROKEN,
-	.quirks2 = SDHCI_QUIRK2_PRESET_VALUE_BROKEN |
-		   /*
-		    * The TRM states that the SD/MMC controller found on
-		    * Tegra124 can address 34 bits (the maximum supported by
-		    * the Tegra memory controller), but tests show that DMA
-		    * to or from above 4 GiB doesn't work. This is possibly
-		    * caused by missing programming, though it's not obvious
-		    * what sequence is required. Mark 64-bit DMA broken for
-		    * now to fix this for existing users (e.g. Nyan boards).
-		    */
-		   SDHCI_QUIRK2_BROKEN_64_BIT_DMA,
+	.quirks2 = SDHCI_QUIRK2_PRESET_VALUE_BROKEN,
 	.ops  = &tegra114_sdhci_ops,
 };
 
 static const struct sdhci_tegra_soc_data soc_data_tegra124 = {
 	.pdata = &sdhci_tegra124_pdata,
+	.dma_mask = DMA_BIT_MASK(34),
 };
 
 static const struct sdhci_ops tegra210_sdhci_ops = {
@@ -1349,6 +1360,7 @@ static const struct sdhci_ops tegra210_sdhci_ops = {
 	.write_w    = tegra210_sdhci_writew,
 	.write_l    = tegra_sdhci_writel,
 	.set_clock  = tegra_sdhci_set_clock,
+	.set_dma_mask = tegra_sdhci_set_dma_mask,
 	.set_bus_width = sdhci_set_bus_width,
 	.reset      = tegra_sdhci_reset,
 	.set_uhs_signaling = tegra_sdhci_set_uhs_signaling,
@@ -1369,6 +1381,7 @@ static const struct sdhci_pltfm_data sdhci_tegra210_pdata = {
 
 static const struct sdhci_tegra_soc_data soc_data_tegra210 = {
 	.pdata = &sdhci_tegra210_pdata,
+	.dma_mask = DMA_BIT_MASK(34),
 	.nvquirks = NVQUIRK_NEEDS_PAD_CONTROL |
 		    NVQUIRK_HAS_PADCALIB |
 		    NVQUIRK_DIS_CARD_CLK_CONFIG_TAP |
@@ -1383,6 +1396,7 @@ static const struct sdhci_ops tegra186_sdhci_ops = {
 	.read_w     = tegra_sdhci_readw,
 	.write_l    = tegra_sdhci_writel,
 	.set_clock  = tegra_sdhci_set_clock,
+	.set_dma_mask = tegra_sdhci_set_dma_mask,
 	.set_bus_width = sdhci_set_bus_width,
 	.reset      = tegra_sdhci_reset,
 	.set_uhs_signaling = tegra_sdhci_set_uhs_signaling,
@@ -1398,20 +1412,13 @@ static const struct sdhci_pltfm_data sdhci_tegra186_pdata = {
 		  SDHCI_QUIRK_NO_HISPD_BIT |
 		  SDHCI_QUIRK_BROKEN_ADMA_ZEROLEN_DESC |
 		  SDHCI_QUIRK_CAP_CLOCK_BASE_BROKEN,
-	.quirks2 = SDHCI_QUIRK2_PRESET_VALUE_BROKEN |
-		   /* SDHCI controllers on Tegra186 support 40-bit addressing.
-		    * IOVA addresses are 48-bit wide on Tegra186.
-		    * With 64-bit dma mask used for SDHCI, accesses can
-		    * be broken. Disable 64-bit dma, which would fall back
-		    * to 32-bit dma mask. Ideally 40-bit dma mask would work,
-		    * But it is not supported as of now.
-		    */
-		   SDHCI_QUIRK2_BROKEN_64_BIT_DMA,
+	.quirks2 = SDHCI_QUIRK2_PRESET_VALUE_BROKEN,
 	.ops  = &tegra186_sdhci_ops,
 };
 
 static const struct sdhci_tegra_soc_data soc_data_tegra186 = {
 	.pdata = &sdhci_tegra186_pdata,
+	.dma_mask = DMA_BIT_MASK(40),
 	.nvquirks = NVQUIRK_NEEDS_PAD_CONTROL |
 		    NVQUIRK_HAS_PADCALIB |
 		    NVQUIRK_DIS_CARD_CLK_CONFIG_TAP |
@@ -1424,6 +1431,7 @@ static const struct sdhci_tegra_soc_data soc_data_tegra186 = {
 
 static const struct sdhci_tegra_soc_data soc_data_tegra194 = {
 	.pdata = &sdhci_tegra186_pdata,
+	.dma_mask = DMA_BIT_MASK(39),
 	.nvquirks = NVQUIRK_NEEDS_PAD_CONTROL |
 		    NVQUIRK_HAS_PADCALIB |
 		    NVQUIRK_DIS_CARD_CLK_CONFIG_TAP |
Signature · 1 line
2.23.0
maintainer Acked-by on v1 2/2
#4 lore raw

Re: [PATCH 2/2] mmc: tegra: Implement ->set_dma_mask()

Quoted 35 lines · Thierry Reding

From: Nicolin Chen <[email protected]>

The SDHCI controller on Tegra186 supports 40-bit addressing, which is usually enough to address all of system memory. However, if the SDHCI controller is behind an IOMMU, the address space can go beyond. This happens on Tegra186 and later where the ARM SMMU has an input address space of 48 bits. If the DMA API is backed by this ARM SMMU, the top- down IOVA allocator will cause IOV addresses to be returned that the SDHCI controller cannot access.

Unfortunately, prior to the introduction of the ->set_dma_mask() host operation, the SDHCI core would set either a 64-bit DMA mask if the controller claimed to support 64-bit addressing, or a 32-bit DMA mask otherwise.

Since the full 64 bits cannot be addressed on Tegra, this had to be worked around in commit 68481a7e1c84 ("mmc: tegra: Mark 64 bit dma broken on Tegra186") by setting the SDHCI_QUIRK2_BROKEN_64_BIT_DMA quirk, which effectively restricts the DMA mask to 32 bits.

One disadvantage of this is that dma_map_*() APIs will now try to use the swiotlb to bounce DMA to addresses beyond of the controller's DMA mask. This in turn caused degraded performance and can lead to situations where the swiotlb buffer is exhausted, which in turn leads to DMA transfers to fail.

With the recent introduction of the ->set_dma_mask() host operation, this can now be properly fixed. For each generation of Tegra, the exact supported DMA mask can be configured. This kills two birds with one
stone: it avoids the use of bounce buffers because system memory never exceeds the addressable memory range of the SDHCI controllers on these devices, and at the same time when an IOMMU is involved, it prevents IOV addresses from being allocated beyond the addressible range of the controllers.

Quoted 6 lines

Since the DMA mask is now properly handled, the 64-bit DMA quirk can be removed.

[[email protected]: provide more background in commit message]
Quoted 163 lines
---
 drivers/mmc/host/sdhci-tegra.c | 48 ++++++++++++++++++++--------------
 1 file changed, 28 insertions(+), 20 deletions(-)
drivers/mmc/host/sdhci-tegra.c +28 -20
index 02d8f524bb9e..7bc950520fd9 100644
@@ -4,6 +4,7 @@
  */
 
 #include <linux/delay.h>
+#include <linux/dma-mapping.h>
 #include <linux/err.h>
 #include <linux/module.h>
 #include <linux/init.h>
@@ -104,6 +105,7 @@
 
 struct sdhci_tegra_soc_data {
 	const struct sdhci_pltfm_data *pdata;
+	u64 dma_mask;
 	u32 nvquirks;
 	u8 min_tap_delay;
 	u8 max_tap_delay;
@@ -1233,11 +1235,25 @@ static const struct cqhci_host_ops sdhci_tegra_cqhci_ops = {
 	.update_dcmd_desc = sdhci_tegra_update_dcmd_desc,
 };
 
+static int tegra_sdhci_set_dma_mask(struct sdhci_host *host)
+{
+	struct sdhci_pltfm_host *platform = sdhci_priv(host);
+	struct sdhci_tegra *tegra = sdhci_pltfm_priv(platform);
+	const struct sdhci_tegra_soc_data *soc = tegra->soc_data;
+	struct device *dev = mmc_dev(host->mmc);
+
+	if (soc->dma_mask)
+		return dma_set_mask_and_coherent(dev, soc->dma_mask);
+
+	return 0;
+}
+
 static const struct sdhci_ops tegra_sdhci_ops = {
 	.get_ro     = tegra_sdhci_get_ro,
 	.read_w     = tegra_sdhci_readw,
 	.write_l    = tegra_sdhci_writel,
 	.set_clock  = tegra_sdhci_set_clock,
+	.set_dma_mask = tegra_sdhci_set_dma_mask,
 	.set_bus_width = sdhci_set_bus_width,
 	.reset      = tegra_sdhci_reset,
 	.platform_execute_tuning = tegra_sdhci_execute_tuning,
@@ -1257,6 +1273,7 @@ static const struct sdhci_pltfm_data sdhci_tegra20_pdata = {
 
 static const struct sdhci_tegra_soc_data soc_data_tegra20 = {
 	.pdata = &sdhci_tegra20_pdata,
+	.dma_mask = DMA_BIT_MASK(32),
 	.nvquirks = NVQUIRK_FORCE_SDHCI_SPEC_200 |
 		    NVQUIRK_ENABLE_BLOCK_GAP_DET,
 };
@@ -1283,6 +1300,7 @@ static const struct sdhci_pltfm_data sdhci_tegra30_pdata = {
 
 static const struct sdhci_tegra_soc_data soc_data_tegra30 = {
 	.pdata = &sdhci_tegra30_pdata,
+	.dma_mask = DMA_BIT_MASK(32),
 	.nvquirks = NVQUIRK_ENABLE_SDHCI_SPEC_300 |
 		    NVQUIRK_ENABLE_SDR50 |
 		    NVQUIRK_ENABLE_SDR104 |
@@ -1295,6 +1313,7 @@ static const struct sdhci_ops tegra114_sdhci_ops = {
 	.write_w    = tegra_sdhci_writew,
 	.write_l    = tegra_sdhci_writel,
 	.set_clock  = tegra_sdhci_set_clock,
+	.set_dma_mask = tegra_sdhci_set_dma_mask,
 	.set_bus_width = sdhci_set_bus_width,
 	.reset      = tegra_sdhci_reset,
 	.platform_execute_tuning = tegra_sdhci_execute_tuning,
@@ -1316,6 +1335,7 @@ static const struct sdhci_pltfm_data sdhci_tegra114_pdata = {
 
 static const struct sdhci_tegra_soc_data soc_data_tegra114 = {
 	.pdata = &sdhci_tegra114_pdata,
+	.dma_mask = DMA_BIT_MASK(32),
 };
 
 static const struct sdhci_pltfm_data sdhci_tegra124_pdata = {
@@ -1325,22 +1345,13 @@ static const struct sdhci_pltfm_data sdhci_tegra124_pdata = {
 		  SDHCI_QUIRK_NO_HISPD_BIT |
 		  SDHCI_QUIRK_BROKEN_ADMA_ZEROLEN_DESC |
 		  SDHCI_QUIRK_CAP_CLOCK_BASE_BROKEN,
-	.quirks2 = SDHCI_QUIRK2_PRESET_VALUE_BROKEN |
-		   /*
-		    * The TRM states that the SD/MMC controller found on
-		    * Tegra124 can address 34 bits (the maximum supported by
-		    * the Tegra memory controller), but tests show that DMA
-		    * to or from above 4 GiB doesn't work. This is possibly
-		    * caused by missing programming, though it's not obvious
-		    * what sequence is required. Mark 64-bit DMA broken for
-		    * now to fix this for existing users (e.g. Nyan boards).
-		    */
-		   SDHCI_QUIRK2_BROKEN_64_BIT_DMA,
+	.quirks2 = SDHCI_QUIRK2_PRESET_VALUE_BROKEN,
 	.ops  = &tegra114_sdhci_ops,
 };
 
 static const struct sdhci_tegra_soc_data soc_data_tegra124 = {
 	.pdata = &sdhci_tegra124_pdata,
+	.dma_mask = DMA_BIT_MASK(34),
 };
 
 static const struct sdhci_ops tegra210_sdhci_ops = {
@@ -1349,6 +1360,7 @@ static const struct sdhci_ops tegra210_sdhci_ops = {
 	.write_w    = tegra210_sdhci_writew,
 	.write_l    = tegra_sdhci_writel,
 	.set_clock  = tegra_sdhci_set_clock,
+	.set_dma_mask = tegra_sdhci_set_dma_mask,
 	.set_bus_width = sdhci_set_bus_width,
 	.reset      = tegra_sdhci_reset,
 	.set_uhs_signaling = tegra_sdhci_set_uhs_signaling,
@@ -1369,6 +1381,7 @@ static const struct sdhci_pltfm_data sdhci_tegra210_pdata = {
 
 static const struct sdhci_tegra_soc_data soc_data_tegra210 = {
 	.pdata = &sdhci_tegra210_pdata,
+	.dma_mask = DMA_BIT_MASK(34),
 	.nvquirks = NVQUIRK_NEEDS_PAD_CONTROL |
 		    NVQUIRK_HAS_PADCALIB |
 		    NVQUIRK_DIS_CARD_CLK_CONFIG_TAP |
@@ -1383,6 +1396,7 @@ static const struct sdhci_ops tegra186_sdhci_ops = {
 	.read_w     = tegra_sdhci_readw,
 	.write_l    = tegra_sdhci_writel,
 	.set_clock  = tegra_sdhci_set_clock,
+	.set_dma_mask = tegra_sdhci_set_dma_mask,
 	.set_bus_width = sdhci_set_bus_width,
 	.reset      = tegra_sdhci_reset,
 	.set_uhs_signaling = tegra_sdhci_set_uhs_signaling,
@@ -1398,20 +1412,13 @@ static const struct sdhci_pltfm_data sdhci_tegra186_pdata = {
 		  SDHCI_QUIRK_NO_HISPD_BIT |
 		  SDHCI_QUIRK_BROKEN_ADMA_ZEROLEN_DESC |
 		  SDHCI_QUIRK_CAP_CLOCK_BASE_BROKEN,
-	.quirks2 = SDHCI_QUIRK2_PRESET_VALUE_BROKEN |
-		   /* SDHCI controllers on Tegra186 support 40-bit addressing.
-		    * IOVA addresses are 48-bit wide on Tegra186.
-		    * With 64-bit dma mask used for SDHCI, accesses can
-		    * be broken. Disable 64-bit dma, which would fall back
-		    * to 32-bit dma mask. Ideally 40-bit dma mask would work,
-		    * But it is not supported as of now.
-		    */
-		   SDHCI_QUIRK2_BROKEN_64_BIT_DMA,
+	.quirks2 = SDHCI_QUIRK2_PRESET_VALUE_BROKEN,
 	.ops  = &tegra186_sdhci_ops,
 };
 
 static const struct sdhci_tegra_soc_data soc_data_tegra186 = {
 	.pdata = &sdhci_tegra186_pdata,
+	.dma_mask = DMA_BIT_MASK(40),
 	.nvquirks = NVQUIRK_NEEDS_PAD_CONTROL |
 		    NVQUIRK_HAS_PADCALIB |
 		    NVQUIRK_DIS_CARD_CLK_CONFIG_TAP |
@@ -1424,6 +1431,7 @@ static const struct sdhci_tegra_soc_data soc_data_tegra186 = {
 
 static const struct sdhci_tegra_soc_data soc_data_tegra194 = {
 	.pdata = &sdhci_tegra186_pdata,
+	.dma_mask = DMA_BIT_MASK(39),
 	.nvquirks = NVQUIRK_NEEDS_PAD_CONTROL |
 		    NVQUIRK_HAS_PADCALIB |
 		    NVQUIRK_DIS_CARD_CLK_CONFIG_TAP |
maintainer Acked-by on v1 2/2
#5 lore raw

Re: [PATCH 2/2] mmc: tegra: Implement ->set_dma_mask()

In reply to Thierry Reding
Quoted 35 lines · Thierry Reding

From: Nicolin Chen <[email protected]>

The SDHCI controller on Tegra186 supports 40-bit addressing, which is usually enough to address all of system memory. However, if the SDHCI controller is behind an IOMMU, the address space can go beyond. This happens on Tegra186 and later where the ARM SMMU has an input address space of 48 bits. If the DMA API is backed by this ARM SMMU, the top- down IOVA allocator will cause IOV addresses to be returned that the SDHCI controller cannot access.

Unfortunately, prior to the introduction of the ->set_dma_mask() host operation, the SDHCI core would set either a 64-bit DMA mask if the controller claimed to support 64-bit addressing, or a 32-bit DMA mask otherwise.

Since the full 64 bits cannot be addressed on Tegra, this had to be worked around in commit 68481a7e1c84 ("mmc: tegra: Mark 64 bit dma broken on Tegra186") by setting the SDHCI_QUIRK2_BROKEN_64_BIT_DMA quirk, which effectively restricts the DMA mask to 32 bits.

One disadvantage of this is that dma_map_*() APIs will now try to use the swiotlb to bounce DMA to addresses beyond of the controller's DMA mask. This in turn caused degraded performance and can lead to situations where the swiotlb buffer is exhausted, which in turn leads to DMA transfers to fail.

With the recent introduction of the ->set_dma_mask() host operation, this can now be properly fixed. For each generation of Tegra, the exact supported DMA mask can be configured. This kills two birds with one
stone: it avoids the use of bounce buffers because system memory never exceeds the addressable memory range of the SDHCI controllers on these devices, and at the same time when an IOMMU is involved, it prevents IOV addresses from being allocated beyond the addressible range of the controllers.

Quoted 6 lines

Since the DMA mask is now properly handled, the 64-bit DMA quirk can be removed.

[[email protected]: provide more background in commit message]

And again with inux->mmc -> linux-mmc

Quoted 163 lines
---
 drivers/mmc/host/sdhci-tegra.c | 48 ++++++++++++++++++++--------------
 1 file changed, 28 insertions(+), 20 deletions(-)
drivers/mmc/host/sdhci-tegra.c +28 -20
index 02d8f524bb9e..7bc950520fd9 100644
@@ -4,6 +4,7 @@
  */
 
 #include <linux/delay.h>
+#include <linux/dma-mapping.h>
 #include <linux/err.h>
 #include <linux/module.h>
 #include <linux/init.h>
@@ -104,6 +105,7 @@
 
 struct sdhci_tegra_soc_data {
 	const struct sdhci_pltfm_data *pdata;
+	u64 dma_mask;
 	u32 nvquirks;
 	u8 min_tap_delay;
 	u8 max_tap_delay;
@@ -1233,11 +1235,25 @@ static const struct cqhci_host_ops sdhci_tegra_cqhci_ops = {
 	.update_dcmd_desc = sdhci_tegra_update_dcmd_desc,
 };
 
+static int tegra_sdhci_set_dma_mask(struct sdhci_host *host)
+{
+	struct sdhci_pltfm_host *platform = sdhci_priv(host);
+	struct sdhci_tegra *tegra = sdhci_pltfm_priv(platform);
+	const struct sdhci_tegra_soc_data *soc = tegra->soc_data;
+	struct device *dev = mmc_dev(host->mmc);
+
+	if (soc->dma_mask)
+		return dma_set_mask_and_coherent(dev, soc->dma_mask);
+
+	return 0;
+}
+
 static const struct sdhci_ops tegra_sdhci_ops = {
 	.get_ro     = tegra_sdhci_get_ro,
 	.read_w     = tegra_sdhci_readw,
 	.write_l    = tegra_sdhci_writel,
 	.set_clock  = tegra_sdhci_set_clock,
+	.set_dma_mask = tegra_sdhci_set_dma_mask,
 	.set_bus_width = sdhci_set_bus_width,
 	.reset      = tegra_sdhci_reset,
 	.platform_execute_tuning = tegra_sdhci_execute_tuning,
@@ -1257,6 +1273,7 @@ static const struct sdhci_pltfm_data sdhci_tegra20_pdata = {
 
 static const struct sdhci_tegra_soc_data soc_data_tegra20 = {
 	.pdata = &sdhci_tegra20_pdata,
+	.dma_mask = DMA_BIT_MASK(32),
 	.nvquirks = NVQUIRK_FORCE_SDHCI_SPEC_200 |
 		    NVQUIRK_ENABLE_BLOCK_GAP_DET,
 };
@@ -1283,6 +1300,7 @@ static const struct sdhci_pltfm_data sdhci_tegra30_pdata = {
 
 static const struct sdhci_tegra_soc_data soc_data_tegra30 = {
 	.pdata = &sdhci_tegra30_pdata,
+	.dma_mask = DMA_BIT_MASK(32),
 	.nvquirks = NVQUIRK_ENABLE_SDHCI_SPEC_300 |
 		    NVQUIRK_ENABLE_SDR50 |
 		    NVQUIRK_ENABLE_SDR104 |
@@ -1295,6 +1313,7 @@ static const struct sdhci_ops tegra114_sdhci_ops = {
 	.write_w    = tegra_sdhci_writew,
 	.write_l    = tegra_sdhci_writel,
 	.set_clock  = tegra_sdhci_set_clock,
+	.set_dma_mask = tegra_sdhci_set_dma_mask,
 	.set_bus_width = sdhci_set_bus_width,
 	.reset      = tegra_sdhci_reset,
 	.platform_execute_tuning = tegra_sdhci_execute_tuning,
@@ -1316,6 +1335,7 @@ static const struct sdhci_pltfm_data sdhci_tegra114_pdata = {
 
 static const struct sdhci_tegra_soc_data soc_data_tegra114 = {
 	.pdata = &sdhci_tegra114_pdata,
+	.dma_mask = DMA_BIT_MASK(32),
 };
 
 static const struct sdhci_pltfm_data sdhci_tegra124_pdata = {
@@ -1325,22 +1345,13 @@ static const struct sdhci_pltfm_data sdhci_tegra124_pdata = {
 		  SDHCI_QUIRK_NO_HISPD_BIT |
 		  SDHCI_QUIRK_BROKEN_ADMA_ZEROLEN_DESC |
 		  SDHCI_QUIRK_CAP_CLOCK_BASE_BROKEN,
-	.quirks2 = SDHCI_QUIRK2_PRESET_VALUE_BROKEN |
-		   /*
-		    * The TRM states that the SD/MMC controller found on
-		    * Tegra124 can address 34 bits (the maximum supported by
-		    * the Tegra memory controller), but tests show that DMA
-		    * to or from above 4 GiB doesn't work. This is possibly
-		    * caused by missing programming, though it's not obvious
-		    * what sequence is required. Mark 64-bit DMA broken for
-		    * now to fix this for existing users (e.g. Nyan boards).
-		    */
-		   SDHCI_QUIRK2_BROKEN_64_BIT_DMA,
+	.quirks2 = SDHCI_QUIRK2_PRESET_VALUE_BROKEN,
 	.ops  = &tegra114_sdhci_ops,
 };
 
 static const struct sdhci_tegra_soc_data soc_data_tegra124 = {
 	.pdata = &sdhci_tegra124_pdata,
+	.dma_mask = DMA_BIT_MASK(34),
 };
 
 static const struct sdhci_ops tegra210_sdhci_ops = {
@@ -1349,6 +1360,7 @@ static const struct sdhci_ops tegra210_sdhci_ops = {
 	.write_w    = tegra210_sdhci_writew,
 	.write_l    = tegra_sdhci_writel,
 	.set_clock  = tegra_sdhci_set_clock,
+	.set_dma_mask = tegra_sdhci_set_dma_mask,
 	.set_bus_width = sdhci_set_bus_width,
 	.reset      = tegra_sdhci_reset,
 	.set_uhs_signaling = tegra_sdhci_set_uhs_signaling,
@@ -1369,6 +1381,7 @@ static const struct sdhci_pltfm_data sdhci_tegra210_pdata = {
 
 static const struct sdhci_tegra_soc_data soc_data_tegra210 = {
 	.pdata = &sdhci_tegra210_pdata,
+	.dma_mask = DMA_BIT_MASK(34),
 	.nvquirks = NVQUIRK_NEEDS_PAD_CONTROL |
 		    NVQUIRK_HAS_PADCALIB |
 		    NVQUIRK_DIS_CARD_CLK_CONFIG_TAP |
@@ -1383,6 +1396,7 @@ static const struct sdhci_ops tegra186_sdhci_ops = {
 	.read_w     = tegra_sdhci_readw,
 	.write_l    = tegra_sdhci_writel,
 	.set_clock  = tegra_sdhci_set_clock,
+	.set_dma_mask = tegra_sdhci_set_dma_mask,
 	.set_bus_width = sdhci_set_bus_width,
 	.reset      = tegra_sdhci_reset,
 	.set_uhs_signaling = tegra_sdhci_set_uhs_signaling,
@@ -1398,20 +1412,13 @@ static const struct sdhci_pltfm_data sdhci_tegra186_pdata = {
 		  SDHCI_QUIRK_NO_HISPD_BIT |
 		  SDHCI_QUIRK_BROKEN_ADMA_ZEROLEN_DESC |
 		  SDHCI_QUIRK_CAP_CLOCK_BASE_BROKEN,
-	.quirks2 = SDHCI_QUIRK2_PRESET_VALUE_BROKEN |
-		   /* SDHCI controllers on Tegra186 support 40-bit addressing.
-		    * IOVA addresses are 48-bit wide on Tegra186.
-		    * With 64-bit dma mask used for SDHCI, accesses can
-		    * be broken. Disable 64-bit dma, which would fall back
-		    * to 32-bit dma mask. Ideally 40-bit dma mask would work,
-		    * But it is not supported as of now.
-		    */
-		   SDHCI_QUIRK2_BROKEN_64_BIT_DMA,
+	.quirks2 = SDHCI_QUIRK2_PRESET_VALUE_BROKEN,
 	.ops  = &tegra186_sdhci_ops,
 };
 
 static const struct sdhci_tegra_soc_data soc_data_tegra186 = {
 	.pdata = &sdhci_tegra186_pdata,
+	.dma_mask = DMA_BIT_MASK(40),
 	.nvquirks = NVQUIRK_NEEDS_PAD_CONTROL |
 		    NVQUIRK_HAS_PADCALIB |
 		    NVQUIRK_DIS_CARD_CLK_CONFIG_TAP |
@@ -1424,6 +1431,7 @@ static const struct sdhci_tegra_soc_data soc_data_tegra186 = {
 
 static const struct sdhci_tegra_soc_data soc_data_tegra194 = {
 	.pdata = &sdhci_tegra186_pdata,
+	.dma_mask = DMA_BIT_MASK(39),
 	.nvquirks = NVQUIRK_NEEDS_PAD_CONTROL |
 		    NVQUIRK_HAS_PADCALIB |
 		    NVQUIRK_DIS_CARD_CLK_CONFIG_TAP |
#6 lore raw
In reply to Thierry Reding
Quoted 24 lines · Thierry Reding

From: Thierry Reding <[email protected]>

Hi,

this puts together the two patches from this thread:

	http://patchwork.ozlabs.org/patch/1146705/

there hasn't been any activity there for a couple of weeks, so I thought I'd just throw them together and update the second patch to work on top of the first one.

The goal here is to avoid the use of bounce buffers and fix IOVA space issues when an IOMMU gets involved. The first patch introduces the ->set_dma_mask() host operation as proposed by Adrian and the second patch is just a rebase of Nicolin's Tegra SDHCI patch that uses the new host operation rather than doing it as part of ->enable_dma().

Thierry

Adrian Hunter (1):
mmc: sdhci: Let drivers define their DMA mask

Quoted 7 lines

Nicolin Chen (1):
mmc: tegra: Implement ->set_dma_mask()

 drivers/mmc/host/sdhci-tegra.c | 48 ++++++++++++++++++++--------------
 drivers/mmc/host/sdhci.c       | 12 +++------
 drivers/mmc/host/sdhci.h       |  1 +
 3 files changed, 33 insertions(+), 28 deletions(-)

Ugh, great. I misspelled [email protected] in my aliases file...

Ulf, Adrian, do you want me to resend for archival purposes?

Thierry