Merge tag 'sunxi-rework-mbus' of ssh://gitolite.kernel.org/pub/scm/linux/kernel/git/mripard/linux into arm/drivers

Allwinner MBUS and DMA-ops rework

The Allwinner SoCs have a number of high-bandwidth devices connected to
a memory bus with a different RAM mapping than the CPU.

This was addressed before through drivers setting the DMA offsets
directly, and subsequently changed to calls to dma_direct_set_offset.
However that wasn't really meant to be exported to modules (and thus
drivers). The duplicated code also led to small inconsistencies across
drivers in how we dealt with DT backward compatibility.

Move all that DMA setup code into a platform bus notifier to share that
code and remove the export on dma_direct_set_offset.

* tag 'sunxi-rework-mbus' of ssh://gitolite.kernel.org/pub/scm/linux/kernel/git/mripard/linux:
  dma-mapping: remove the dma_direct_set_offset export
  media: sun8i-di: Remove the call to of_dma_configure
  media: cedrus: Remove the MBUS quirks
  media: sun6i: Remove the MBUS quirks
  media: sun4i: Remove the MBUS quirks
  drm/sun4i: backend: Remove the MBUS quirks
  soc: sunxi: Deal with the MBUS DMA offsets in a central place
  drm/sun4i: backend: Fix probe failure with multiple backends

Signed-off-by: Arnd Bergmann <[email protected]>

17 files changed, 149 insertions, 95 deletions

diff --git a/arch/arm/mach-keystone/keystone.c b/arch/arm/mach-keystone/keystone.c
index 09a65c2dfd73..cd711bfc591f 100644
--- a/arch/arm/mach-keystone/keystone.c
+++ b/arch/arm/mach-keystone/keystone.c
@@ -8,7 +8,7 @@
  */
 #include <linux/io.h>
 #include <linux/of.h>
-#include <linux/dma-mapping.h>
+#include <linux/dma-map-ops.h>
 #include <linux/init.h>
 #include <linux/of_platform.h>
 #include <linux/of_address.h>
diff --git a/arch/arm/mach-omap1/usb.c b/arch/arm/mach-omap1/usb.c
index ba8566204ea9..86d3b3c157af 100644
--- a/arch/arm/mach-omap1/usb.c
+++ b/arch/arm/mach-omap1/usb.c
@@ -9,7 +9,7 @@
 #include <linux/kernel.h>
 #include <linux/init.h>
 #include <linux/platform_device.h>
-#include <linux/dma-mapping.h>
+#include <linux/dma-map-ops.h>
 #include <linux/io.h>
 
 #include <asm/irq.h>
diff --git a/arch/sh/drivers/pci/pcie-sh7786.c b/arch/sh/drivers/pci/pcie-sh7786.c
index 4468289ab2ca..4d499476c33a 100644
--- a/arch/sh/drivers/pci/pcie-sh7786.c
+++ b/arch/sh/drivers/pci/pcie-sh7786.c
@@ -12,7 +12,7 @@
 #include <linux/io.h>
 #include <linux/async.h>
 #include <linux/delay.h>
-#include <linux/dma-mapping.h>
+#include <linux/dma-map-ops.h>
 #include <linux/slab.h>
 #include <linux/clk.h>
 #include <linux/sh_clk.h>
diff --git a/arch/x86/pci/sta2x11-fixup.c b/arch/x86/pci/sta2x11-fixup.c
index 5701d5ba3df4..7d2525691854 100644
--- a/arch/x86/pci/sta2x11-fixup.c
+++ b/arch/x86/pci/sta2x11-fixup.c
@@ -11,7 +11,8 @@
 #include <linux/pci_ids.h>
 #include <linux/export.h>
 #include <linux/list.h>
-#include <linux/dma-direct.h>
+#include <linux/dma-map-ops.h>
+#include <linux/swiotlb.h>
 #include <asm/iommu.h>
 
 #define STA2X11_SWIOTLB_SIZE (4*1024*1024)
diff --git a/drivers/gpu/drm/sun4i/sun4i_backend.c b/drivers/gpu/drm/sun4i/sun4i_backend.c
index 77497b45f9a2..522e51a404cc 100644
--- a/drivers/gpu/drm/sun4i/sun4i_backend.c
+++ b/drivers/gpu/drm/sun4i/sun4i_backend.c
@@ -805,19 +805,6 @@ static int sun4i_backend_bind(struct device *dev, struct device *master,
 		ret = of_dma_configure(drm->dev, dev->of_node, true);
 		if (ret)
 			return ret;
-	} else {
-		/*
-		 * If we don't have the interconnect property, most likely
-		 * because of an old DT, we need to set the DMA offset by hand
-		 * on our device since the RAM mapping is at 0 for the DMA bus,
-		 * unlike the CPU.
-		 *
-		 * XXX(hch): this has no business in a driver and needs to move
-		 * to the device tree.
-		 */
-		ret = dma_direct_set_offset(drm->dev, PHYS_OFFSET, 0, SZ_4G);
-		if (ret)
-			return ret;
 	}
 
 	backend->engine.node = dev->of_node;
diff --git a/drivers/media/platform/sunxi/sun4i-csi/sun4i_csi.c b/drivers/media/platform/sunxi/sun4i-csi/sun4i_csi.c
index eb15c8c725ca..ec46cff80fdb 100644
--- a/drivers/media/platform/sunxi/sun4i-csi/sun4i_csi.c
+++ b/drivers/media/platform/sunxi/sun4i-csi/sun4i_csi.c
@@ -167,33 +167,6 @@ static int sun4i_csi_probe(struct platform_device *pdev)
 	if (!csi->traits)
 		return -EINVAL;
 
-	/*
-	 * On Allwinner SoCs, some high memory bandwidth devices do DMA
-	 * directly over the memory bus (called MBUS), instead of the
-	 * system bus. The memory bus has a different addressing scheme
-	 * without the DRAM starting offset.
-	 *
-	 * In some cases this can be described by an interconnect in
-	 * the device tree. In other cases where the hardware is not
-	 * fully understood and the interconnect is left out of the
-	 * device tree, fall back to a default offset.
-	 */
-	if (of_find_property(csi->dev->of_node, "interconnects", NULL)) {
-		ret = of_dma_configure(csi->dev, csi->dev->of_node, true);
-		if (ret)
-			return ret;
-	} else {
-		/*
-		 * XXX(hch): this has no business in a driver and needs to move
-		 * to the device tree.
-		 */
-#ifdef PHYS_PFN_OFFSET
-		ret = dma_direct_set_offset(csi->dev, PHYS_OFFSET, 0, SZ_4G);
-		if (ret)
-			return ret;
-#endif
-	}
-
 	csi->mdev.dev = csi->dev;
 	strscpy(csi->mdev.model, "Allwinner Video Capture Device",
 		sizeof(csi->mdev.model));
diff --git a/drivers/media/platform/sunxi/sun6i-csi/sun6i_csi.c b/drivers/media/platform/sunxi/sun6i-csi/sun6i_csi.c
index e69e14379fc6..27935f1e9555 100644
--- a/drivers/media/platform/sunxi/sun6i-csi/sun6i_csi.c
+++ b/drivers/media/platform/sunxi/sun6i-csi/sun6i_csi.c
@@ -881,14 +881,6 @@ static int sun6i_csi_resource_request(struct sun6i_csi_dev *sdev,
 	return 0;
 }
 
-/*
- * PHYS_OFFSET isn't available on all architectures. In order to
- * accommodate for COMPILE_TEST, let's define it to something dumb.
- */
-#if defined(CONFIG_COMPILE_TEST) && !defined(PHYS_OFFSET)
-#define PHYS_OFFSET 0
-#endif
-
 static int sun6i_csi_probe(struct platform_device *pdev)
 {
 	struct sun6i_csi_dev *sdev;
@@ -899,15 +891,6 @@ static int sun6i_csi_probe(struct platform_device *pdev)
 		return -ENOMEM;
 
 	sdev->dev = &pdev->dev;
-	/*
-	 * The DMA bus has the memory mapped at 0.
-	 *
-	 * XXX(hch): this has no business in a driver and needs to move
-	 * to the device tree.
-	 */
-	ret = dma_direct_set_offset(sdev->dev, PHYS_OFFSET, 0, SZ_4G);
-	if (ret)
-		return ret;
 
 	ret = sun6i_csi_resource_request(sdev, pdev);
 	if (ret)
diff --git a/drivers/media/platform/sunxi/sun8i-di/sun8i-di.c b/drivers/media/platform/sunxi/sun8i-di/sun8i-di.c
index ba5d07886607..ed863bf5ea80 100644
--- a/drivers/media/platform/sunxi/sun8i-di/sun8i-di.c
+++ b/drivers/media/platform/sunxi/sun8i-di/sun8i-di.c
@@ -825,10 +825,6 @@ static int deinterlace_probe(struct platform_device *pdev)
 		return ret;
 	}
 
-	ret = of_dma_configure(dev->dev, dev->dev->of_node, true);
-	if (ret)
-		return ret;
-
 	res = platform_get_resource(pdev, IORESOURCE_MEM, 0);
 	dev->base = devm_ioremap_resource(&pdev->dev, res);
 	if (IS_ERR(dev->base))
diff --git a/drivers/soc/sunxi/Kconfig b/drivers/soc/sunxi/Kconfig
index f10fd6cae13e..1fef0e711056 100644
--- a/drivers/soc/sunxi/Kconfig
+++ b/drivers/soc/sunxi/Kconfig
@@ -2,6 +2,14 @@
 #
 # Allwinner sunXi SoC drivers
 #
+
+config SUNXI_MBUS
+	bool
+	default ARCH_SUNXI
+	help
+	  Say y to enable the fixups needed to support the Allwinner
+	  MBUS DMA quirks.
+
 config SUNXI_SRAM
 	bool
 	default ARCH_SUNXI
diff --git a/drivers/soc/sunxi/Makefile b/drivers/soc/sunxi/Makefile
index 7816fbbec387..549159571d4f 100644
--- a/drivers/soc/sunxi/Makefile
+++ b/drivers/soc/sunxi/Makefile
@@ -1,2 +1,3 @@
 # SPDX-License-Identifier: GPL-2.0-only
+obj-$(CONFIG_SUNXI_MBUS) +=	sunxi_mbus.o
 obj-$(CONFIG_SUNXI_SRAM) +=	sunxi_sram.o
diff --git a/drivers/soc/sunxi/sunxi_mbus.c b/drivers/soc/sunxi/sunxi_mbus.c
new file mode 100644
index 000000000000..a9d077f73c3a
--- /dev/null
+++ b/drivers/soc/sunxi/sunxi_mbus.c
@@ -0,0 +1,132 @@
+// SPDX-License-Identifier: GPL-2.0
+/* Copyright (C) 2020 Maxime Ripard <[email protected]> */
+
+#include <linux/device.h>
+#include <linux/dma-mapping.h>
+#include <linux/init.h>
+#include <linux/notifier.h>
+#include <linux/of.h>
+#include <linux/platform_device.h>
+
+static const char * const sunxi_mbus_devices[] = {
+	/*
+	 * The display engine virtual devices are not strictly speaking
+	 * connected to the MBUS, but since DRM will perform all the
+	 * memory allocations and DMA operations through that device, we
+	 * need to have the quirk on those devices too.
+	 */
+	"allwinner,sun4i-a10-display-engine",
+	"allwinner,sun5i-a10s-display-engine",
+	"allwinner,sun5i-a13-display-engine",
+	"allwinner,sun6i-a31-display-engine",
+	"allwinner,sun6i-a31s-display-engine",
+	"allwinner,sun7i-a20-display-engine",
+	"allwinner,sun8i-a23-display-engine",
+	"allwinner,sun8i-a33-display-engine",
+	"allwinner,sun8i-a83t-display-engine",
+	"allwinner,sun8i-h3-display-engine",
+	"allwinner,sun8i-r40-display-engine",
+	"allwinner,sun8i-v3s-display-engine",
+	"allwinner,sun9i-a80-display-engine",
+	"allwinner,sun50i-a64-display-engine",
+
+	/*
+	 * And now we have the regular devices connected to the MBUS
+	 * (that we know of).
+	 */
+	"allwinner,sun4i-a10-csi1",
+	"allwinner,sun4i-a10-display-backend",
+	"allwinner,sun4i-a10-display-frontend",
+	"allwinner,sun4i-a10-video-engine",
+	"allwinner,sun5i-a13-display-backend",
+	"allwinner,sun5i-a13-video-engine",
+	"allwinner,sun6i-a31-csi",
+	"allwinner,sun6i-a31-display-backend",
+	"allwinner,sun7i-a20-csi0",
+	"allwinner,sun7i-a20-display-backend",
+	"allwinner,sun7i-a20-display-frontend",
+	"allwinner,sun7i-a20-video-engine",
+	"allwinner,sun8i-a23-display-backend",
+	"allwinner,sun8i-a23-display-frontend",
+	"allwinner,sun8i-a33-display-backend",
+	"allwinner,sun8i-a33-display-frontend",
+	"allwinner,sun8i-a33-video-engine",
+	"allwinner,sun8i-a83t-csi",
+	"allwinner,sun8i-h3-csi",
+	"allwinner,sun8i-h3-video-engine",
+	"allwinner,sun8i-v3s-csi",
+	"allwinner,sun9i-a80-display-backend",
+	"allwinner,sun50i-a64-csi",
+	"allwinner,sun50i-a64-video-engine",
+	"allwinner,sun50i-h5-video-engine",
+	NULL,
+};
+
+static int sunxi_mbus_notifier(struct notifier_block *nb,
+			       unsigned long event, void *__dev)
+{
+	struct device *dev = __dev;
+	int ret;
+
+	if (event != BUS_NOTIFY_ADD_DEVICE)
+		return NOTIFY_DONE;
+
+	/*
+	 * Only the devices that need a large memory bandwidth do DMA
+	 * directly over the memory bus (called MBUS), instead of going
+	 * through the regular system bus.
+	 */
+	if (!of_device_compatible_match(dev->of_node, sunxi_mbus_devices))
+		return NOTIFY_DONE;
+
+	/*
+	 * Devices with an interconnects property have the MBUS
+	 * relationship described in their DT and dealt with by
+	 * of_dma_configure, so we can just skip them.
+	 *
+	 * Older DTs or SoCs who are not clearly understood need to set
+	 * that DMA offset though.
+	 */
+	if (of_find_property(dev->of_node, "interconnects", NULL))
+		return NOTIFY_DONE;
+
+	ret = dma_direct_set_offset(dev, PHYS_OFFSET, 0, SZ_4G);
+	if (ret)
+		dev_err(dev, "Couldn't setup our DMA offset: %d\n", ret);
+
+	return NOTIFY_DONE;
+}
+
+static struct notifier_block sunxi_mbus_nb = {
+	.notifier_call = sunxi_mbus_notifier,
+};
+
+static const char * const sunxi_mbus_platforms[] __initconst = {
+	"allwinner,sun4i-a10",
+	"allwinner,sun5i-a10s",
+	"allwinner,sun5i-a13",
+	"allwinner,sun6i-a31",
+	"allwinner,sun7i-a20",
+	"allwinner,sun8i-a23",
+	"allwinner,sun8i-a33",
+	"allwinner,sun8i-a83t",
+	"allwinner,sun8i-h3",
+	"allwinner,sun8i-r40",
+	"allwinner,sun8i-v3",
+	"allwinner,sun8i-v3s",
+	"allwinner,sun9i-a80",
+	"allwinner,sun50i-a64",
+	"allwinner,sun50i-h5",
+	"nextthing,gr8",
+	NULL,
+};
+
+static int __init sunxi_mbus_init(void)
+{
+	if (!of_device_compatible_match(of_root, sunxi_mbus_platforms))
+		return 0;
+
+	bus_register_notifier(&platform_bus_type, &sunxi_mbus_nb);
+	return 0;
+}
+arch_initcall(sunxi_mbus_init);
diff --git a/drivers/staging/media/sunxi/cedrus/cedrus.c b/drivers/staging/media/sunxi/cedrus/cedrus.c
index e0e35502e34a..d5fca10ea5b4 100644
--- a/drivers/staging/media/sunxi/cedrus/cedrus.c
+++ b/drivers/staging/media/sunxi/cedrus/cedrus.c
@@ -523,7 +523,6 @@ static const struct cedrus_variant sun50i_h5_cedrus_variant = {
 static const struct cedrus_variant sun50i_h6_cedrus_variant = {
 	.capabilities	= CEDRUS_CAPABILITY_UNTILED |
 			  CEDRUS_CAPABILITY_H265_DEC,
-	.quirks		= CEDRUS_QUIRK_NO_DMA_OFFSET,
 	.mod_rate	= 600000000,
 };
 
diff --git a/drivers/staging/media/sunxi/cedrus/cedrus.h b/drivers/staging/media/sunxi/cedrus/cedrus.h
index 93c843ae14bb..626090a5811c 100644
--- a/drivers/staging/media/sunxi/cedrus/cedrus.h
+++ b/drivers/staging/media/sunxi/cedrus/cedrus.h
@@ -29,8 +29,6 @@
 #define CEDRUS_CAPABILITY_UNTILED	BIT(0)
 #define CEDRUS_CAPABILITY_H265_DEC	BIT(1)
 
-#define CEDRUS_QUIRK_NO_DMA_OFFSET	BIT(0)
-
 enum cedrus_codec {
 	CEDRUS_CODEC_MPEG2,
 	CEDRUS_CODEC_H264,
@@ -150,7 +148,6 @@ struct cedrus_dec_ops {
 
 struct cedrus_variant {
 	unsigned int	capabilities;
-	unsigned int	quirks;
 	unsigned int	mod_rate;
 };
 
diff --git a/drivers/staging/media/sunxi/cedrus/cedrus_hw.c b/drivers/staging/media/sunxi/cedrus/cedrus_hw.c
index bcf050a04ffc..286c7fe844c3 100644
--- a/drivers/staging/media/sunxi/cedrus/cedrus_hw.c
+++ b/drivers/staging/media/sunxi/cedrus/cedrus_hw.c
@@ -222,24 +222,6 @@ int cedrus_hw_probe(struct cedrus_dev *dev)
 		return ret;
 	}
 
-	/*
-	 * The VPU is only able to handle bus addresses so we have to subtract
-	 * the RAM offset to the physcal addresses.
-	 *
-	 * This information will eventually be obtained from device-tree.
-	 *
-	 * XXX(hch): this has no business in a driver and needs to move
-	 * to the device tree.
-	 */
-
-#ifdef PHYS_PFN_OFFSET
-	if (!(variant->quirks & CEDRUS_QUIRK_NO_DMA_OFFSET)) {
-		ret = dma_direct_set_offset(dev->dev, PHYS_OFFSET, 0, SZ_4G);
-		if (ret)
-			return ret;
-	}
-#endif
-
 	ret = of_reserved_mem_device_init(dev->dev);
 	if (ret && ret != -ENODEV) {
 		dev_err(dev->dev, "Failed to reserve memory\n");
diff --git a/include/linux/dma-map-ops.h b/include/linux/dma-map-ops.h
index a5f89fc4d6df..03925e438ec3 100644
--- a/include/linux/dma-map-ops.h
+++ b/include/linux/dma-map-ops.h
@@ -226,6 +226,9 @@ struct page *dma_alloc_from_pool(struct device *dev, size_t size,
 		bool (*phys_addr_ok)(struct device *, phys_addr_t, size_t));
 bool dma_free_from_pool(struct device *dev, void *start, size_t size);
 
+int dma_direct_set_offset(struct device *dev, phys_addr_t cpu_start,
+		dma_addr_t dma_start, u64 size);
+
 #ifdef CONFIG_ARCH_HAS_DMA_COHERENCE_H
 #include <asm/dma-coherence.h>
 #elif defined(CONFIG_ARCH_HAS_SYNC_DMA_FOR_DEVICE) || \
diff --git a/include/linux/dma-mapping.h b/include/linux/dma-mapping.h
index 956151052d45..199d85285246 100644
--- a/include/linux/dma-mapping.h
+++ b/include/linux/dma-mapping.h
@@ -558,13 +558,6 @@ static inline int dma_mmap_wc(struct device *dev,
 #define dma_unmap_len_set(PTR, LEN_NAME, VAL)    do { } while (0)
 #endif
 
-/*
- * Legacy interface to set up the dma offset map.  Drivers really should not
- * actually use it, but we have a few legacy cases left.
- */
-int dma_direct_set_offset(struct device *dev, phys_addr_t cpu_start,
-		dma_addr_t dma_start, u64 size);
-
 extern const struct dma_map_ops dma_virt_ops;
 
 #endif /* _LINUX_DMA_MAPPING_H */
diff --git a/kernel/dma/direct.c b/kernel/dma/direct.c
index 06c111544f61..002268262c9a 100644
--- a/kernel/dma/direct.c
+++ b/kernel/dma/direct.c
@@ -547,4 +547,3 @@ int dma_direct_set_offset(struct device *dev, phys_addr_t cpu_start,
 	dev->dma_range_map = map;
 	return 0;
 }
-EXPORT_SYMBOL_GPL(dma_direct_set_offset);