[PATCH v3 0/2] ARM: IOMMU: tegra: Add iommu_ops for GART/SMMU driver

[PATCH v3 0/2] ARM: IOMMU: tegra: Add iommu_ops for GART/SMMU driver

[Hiroshi DOYU]

This patchset adds support for IOMMU API for Tegra20(GART) and
Tegra30(SMMU). "struct iommu_ops" are implemented for both H/W IOMMU
modules.

Tested with "Ion memory manager" and "DMA mapping API"(*1).

This was originally developed with Tegra specific IOMMU
API(tegra_iovmm_*()) in:

git://nv-tegra.nvidia.com/linux-2.6.git.

v2:
https://lkml.org/lkml/2011/12/15/203

v1:
https://lkml.org/lkml/2011/11/17/111

*1: Reworking the DMA mapping code (especially on ARM)
    http://lwn.net/Articles/467509/

Hiroshi DOYU (2):
  ARM: IOMMU: Tegra20: Add iommu_ops for GART driver
  ARM: IOMMU: Tegra30: Add iommu_ops for SMMU driver

 arch/arm/mach-tegra/include/mach/smmu.h |   63 ++
 drivers/iommu/Kconfig                   |   22 +
 drivers/iommu/Makefile                  |    2 +
 drivers/iommu/tegra-gart.c              |  436 +++++++++++++
 drivers/iommu/tegra-smmu.c              | 1026 +++++++++++++++++++++++++++++++
 5 files changed, 1549 insertions(+), 0 deletions(-)
 create mode 100644 arch/arm/mach-tegra/include/mach/smmu.h
 create mode 100644 drivers/iommu/tegra-gart.c
 create mode 100644 drivers/iommu/tegra-smmu.c

-- 
1.7.5.4

[PATCH 1/2] ARM: IOMMU: Tegra20: Add iommu_ops for GART driver

[Hiroshi DOYU]

Tegra 20 IOMMU H/W, GART (Graphics Address Relocation Table). This
patch implements struct iommu_ops for GART for the upper IOMMU API.

This H/W module supports only single virtual address space(domain),
and manages a single level 1-to-1 mapping H/W translation page table.

Signed-off-by: Hiroshi DOYU <hdoyu@nvidia.com>
---
 drivers/iommu/Kconfig      |   11 +
 drivers/iommu/Makefile     |    1 +
 drivers/iommu/tegra-gart.c |  436 ++++++++++++++++++++++++++++++++++++++++++++
 3 files changed, 448 insertions(+), 0 deletions(-)
 create mode 100644 drivers/iommu/tegra-gart.c

diff --git a/drivers/iommu/Kconfig b/drivers/iommu/Kconfig
index 6bea696..4562698 100644
--- a/drivers/iommu/Kconfig
+++ b/drivers/iommu/Kconfig
@@ -142,4 +142,15 @@ config OMAP_IOMMU_DEBUG
 
          Say N unless you know you need this.
 
+config TEGRA_IOMMU_GART
+	bool "Tegra GART IOMMU Support"
+	depends on ARCH_TEGRA_2x_SOC
+	default y
+	select IOMMU_API
+	help
+	  Enables support for remapping discontiguous physical memory
+	  shared with the operating system into contiguous I/O virtual
+	  space through the GART (Graphics Address Relocation Table)
+	  hardware included on Tegra SoCs.
+
 endif # IOMMU_SUPPORT
diff --git a/drivers/iommu/Makefile b/drivers/iommu/Makefile
index 0e36b49..3238a31 100644
--- a/drivers/iommu/Makefile
+++ b/drivers/iommu/Makefile
@@ -8,3 +8,4 @@ obj-$(CONFIG_IRQ_REMAP) += intr_remapping.o
 obj-$(CONFIG_OMAP_IOMMU) += omap-iommu.o
 obj-$(CONFIG_OMAP_IOVMM) += omap-iovmm.o
 obj-$(CONFIG_OMAP_IOMMU_DEBUG) += omap-iommu-debug.o
+obj-$(CONFIG_TEGRA_IOMMU_GART) += tegra-gart.o
diff --git a/drivers/iommu/tegra-gart.c b/drivers/iommu/tegra-gart.c
new file mode 100644
index 0000000..feaccd4
--- /dev/null
+++ b/drivers/iommu/tegra-gart.c
@@ -0,0 +1,436 @@
+/*
+ * IOMMU API for GART in Tegra20
+ *
+ * Copyright (c) 2010-2012, NVIDIA CORPORATION.  All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or modify it
+ * under the terms and conditions of the GNU General Public License,
+ * version 2, as published by the Free Software Foundation.
+ *
+ * This program is distributed in the hope it will be useful, but WITHOUT
+ * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
+ * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
+ * more details.
+ *
+ * You should have received a copy of the GNU General Public License along with
+ * this program; if not, write to the Free Software Foundation, Inc.,
+ * 51 Franklin St - Fifth Floor, Boston, MA 02110-1301 USA.
+ */
+
+#define pr_fmt(fmt)	"%s(): " fmt, __func__
+
+#include <linux/module.h>
+#include <linux/platform_device.h>
+#include <linux/spinlock.h>
+#include <linux/slab.h>
+#include <linux/vmalloc.h>
+#include <linux/mm.h>
+#include <linux/list.h>
+#include <linux/device.h>
+#include <linux/io.h>
+#include <linux/iommu.h>
+
+#include <asm/cacheflush.h>
+
+/* bitmap of the page sizes currently supported */
+#define GART_IOMMU_PGSIZES	(SZ_4K)
+
+#define GART_CONFIG		0x24
+#define GART_ENTRY_ADDR		0x28
+#define GART_ENTRY_DATA		0x2c
+#define GART_ENTRY_PHYS_ADDR_VALID	(1 << 31)
+
+#define GART_PAGE_SHIFT		12
+#define GART_PAGE_SIZE		(1 << GART_PAGE_SHIFT)
+#define GART_PAGE_MASK						\
+	(~(GART_PAGE_SIZE - 1) & ~GART_ENTRY_PHYS_ADDR_VALID)
+
+struct gart_client {
+	struct device		*dev;
+	struct list_head	list;
+};
+
+struct gart_device {
+	void __iomem		*regs;
+	u32			*savedata;
+	u32			page_count; /* total remappable size */
+	dma_addr_t		iovmm_base; /* offset to apply to vmm_area */
+	spinlock_t		pte_lock; /* for pagetable */
+	struct list_head	client;
+	spinlock_t		client_lock; /* for client list */
+	struct device		*dev;
+};
+
+#define GART_PTE(_pfn)						\
+	(GART_ENTRY_PHYS_ADDR_VALID | ((_pfn) << PAGE_SHIFT))
+
+/*
+ * Any interaction between any block on PPSB and a block on APB or AHB
+ * must have these read-back to ensure the APB/AHB bus transaction is
+ * complete before initiating activity on the PPSB block.
+ */
+#define FLUSH_GART_REGS(gart)	((void)readl((gart)->regs + GART_CONFIG))
+
+#define for_each_gart_pte(gart, iova)					\
+	for (iova = gart->iovmm_base;					\
+	     iova < gart->iovmm_base + GART_PAGE_SIZE * gart->page_count; \
+	     iova += GART_PAGE_SIZE)
+
+static inline void gart_set_pte(struct gart_device *gart,
+				unsigned long offs, u32 pte)
+{
+	writel(offs, gart->regs + GART_ENTRY_ADDR);
+	writel(pte, gart->regs + GART_ENTRY_DATA);
+
+	dev_dbg(gart->dev, "%s %08lx:%08x\n",
+		 pte ? "map" : "unmap", offs, pte & GART_PAGE_MASK);
+}
+
+static inline unsigned long gart_read_pte(struct gart_device *gart,
+					  unsigned long offs)
+{
+	unsigned long pte;
+
+	writel(offs, gart->regs + GART_ENTRY_ADDR);
+	pte = readl(gart->regs + GART_ENTRY_DATA);
+
+	return pte;
+}
+
+static void do_gart_setup(struct gart_device *gart, const u32 *data)
+{
+	unsigned long iova;
+
+	for_each_gart_pte(gart, iova)
+		gart_set_pte(gart, iova, data ? *(data++) : 0);
+
+	writel(1, gart->regs + GART_CONFIG);
+	FLUSH_GART_REGS(gart);
+}
+
+#ifdef DEBUG
+static void gart_dump_table(struct gart_device *gart)
+{
+	unsigned long iova;
+	unsigned long flags;
+
+	spin_lock_irqsave(&gart->pte_lock, flags);
+	for_each_gart_pte(gart, iova) {
+		unsigned long pte;
+
+		pte = gart_read_pte(gart, iova);
+
+		dev_dbg(gart->dev, "%s %08lx:%08lx\n",
+			(GART_ENTRY_PHYS_ADDR_VALID & pte) ? "v" : " ",
+			iova, pte & GART_PAGE_MASK);
+	}
+	spin_unlock_irqrestore(&gart->pte_lock, flags);
+}
+#else
+static inline void gart_dump_table(struct gart_device *gart)
+{
+}
+#endif
+
+static inline bool gart_iova_range_valid(struct gart_device *gart,
+					 unsigned long iova, size_t bytes)
+{
+	unsigned long iova_start, iova_end, gart_start, gart_end;
+
+	iova_start = iova;
+	iova_end = iova_start + bytes - 1;
+	gart_start = gart->iovmm_base;
+	gart_end = gart_start + gart->page_count * GART_PAGE_SIZE - 1;
+
+	if (iova_start < gart_start)
+		return false;
+	if (iova_end > gart_end)
+		return false;
+	return true;
+}
+
+static int gart_iommu_domain_init(struct iommu_domain *domain)
+{
+	return 0;
+}
+
+static void gart_iommu_domain_destroy(struct iommu_domain *domain)
+{
+	struct gart_device *gart = domain->priv;
+
+	spin_lock(&gart->client_lock);
+	if (!list_empty(&gart->client)) {
+		struct gart_client *c;
+
+		list_for_each_entry(c, &gart->client, list)
+			dev_err(gart->dev,
+				"%s is still attached\n", dev_name(c->dev));
+	}
+	spin_unlock(&gart->client_lock);
+	domain->priv = NULL;
+}
+
+static int gart_iommu_attach_dev(struct iommu_domain *domain,
+				 struct device *dev)
+{
+	struct gart_device *gart;
+	struct gart_client *client, *c;
+	int err = 0;
+
+	gart = dev_get_drvdata(dev->parent);
+	if (!gart)
+		return -EINVAL;
+	domain->priv = gart;
+	client = devm_kzalloc(dev, sizeof(*c), GFP_KERNEL);
+	if (!client)
+		return -ENOMEM;
+	client->dev = dev;
+
+	spin_lock(&gart->client_lock);
+	list_for_each_entry(c, &gart->client, list) {
+		if (c->dev == dev) {
+			dev_err(gart->dev,
+				"%s is already attached\n", dev_name(dev));
+			err = -EINVAL;
+			goto fail;
+		}
+	}
+	list_add(&client->list, &gart->client);
+	spin_unlock(&gart->client_lock);
+	dev_dbg(gart->dev, "Attached %s\n", dev_name(dev));
+	return 0;
+
+fail:
+	devm_kfree(dev, client);
+	spin_unlock(&gart->client_lock);
+	return err;
+}
+
+static void gart_iommu_detach_dev(struct iommu_domain *domain,
+				  struct device *dev)
+{
+	struct gart_device *gart = domain->priv;
+	struct gart_client *c;
+
+	spin_lock(&gart->client_lock);
+
+	list_for_each_entry(c, &gart->client, list) {
+		if (c->dev == dev) {
+			list_del(&c->list);
+			devm_kfree(dev, c);
+			dev_dbg(gart->dev, "Detached %s\n", dev_name(dev));
+			goto out;
+		}
+	}
+	dev_err(gart->dev, "Couldn't find\n");
+out:
+	spin_unlock(&gart->client_lock);
+}
+
+static int gart_iommu_map(struct iommu_domain *domain, unsigned long iova,
+			  phys_addr_t pa, size_t bytes, int prot)
+{
+	struct gart_device *gart = domain->priv;
+	unsigned long flags;
+	unsigned long pfn;
+
+	if (!gart_iova_range_valid(gart, iova, bytes))
+		return -EINVAL;
+
+	spin_lock_irqsave(&gart->pte_lock, flags);
+	pfn = __phys_to_pfn(pa);
+	if (!pfn_valid(pfn)) {
+		dev_err(gart->dev, "Invalid page: %08x\n", pa);
+		spin_unlock(&gart->pte_lock);
+		return -EINVAL;
+	}
+	gart_set_pte(gart, iova, GART_PTE(pfn));
+	FLUSH_GART_REGS(gart);
+	spin_unlock_irqrestore(&gart->pte_lock, flags);
+	return 0;
+}
+
+static size_t gart_iommu_unmap(struct iommu_domain *domain, unsigned long iova,
+			       size_t bytes)
+{
+	struct gart_device *gart = domain->priv;
+	unsigned long flags;
+
+	if (!gart_iova_range_valid(gart, iova, bytes))
+		return -EINVAL;
+
+	spin_lock_irqsave(&gart->pte_lock, flags);
+	gart_set_pte(gart, iova, 0);
+	FLUSH_GART_REGS(gart);
+	spin_unlock_irqrestore(&gart->pte_lock, flags);
+	return 0;
+}
+
+static phys_addr_t gart_iommu_iova_to_phys(struct iommu_domain *domain,
+					   unsigned long iova)
+{
+	struct gart_device *gart = domain->priv;
+	unsigned long pte;
+	phys_addr_t pa;
+	unsigned long flags;
+
+	if (!gart_iova_range_valid(gart, iova, 0))
+		return -EINVAL;
+
+	spin_lock_irqsave(&gart->pte_lock, flags);
+	pte = gart_read_pte(gart, iova);
+	spin_unlock_irqrestore(&gart->pte_lock, flags);
+
+	pa = (pte & GART_PAGE_MASK);
+	if (!pfn_valid(__phys_to_pfn(pa))) {
+		dev_err(gart->dev, "No entry for %08lx:%08x\n", iova, pa);
+		gart_dump_table(gart);
+		return -EINVAL;
+	}
+	return pa;
+}
+
+static int gart_iommu_domain_has_cap(struct iommu_domain *domain,
+				     unsigned long cap)
+{
+	return 0;
+}
+
+static struct iommu_ops gart_iommu_ops = {
+	.domain_init	= gart_iommu_domain_init,
+	.domain_destroy	= gart_iommu_domain_destroy,
+	.attach_dev	= gart_iommu_attach_dev,
+	.detach_dev	= gart_iommu_detach_dev,
+	.map		= gart_iommu_map,
+	.unmap		= gart_iommu_unmap,
+	.iova_to_phys	= gart_iommu_iova_to_phys,
+	.domain_has_cap	= gart_iommu_domain_has_cap,
+	.pgsize_bitmap	= GART_IOMMU_PGSIZES,
+};
+
+static int tegra_gart_suspend(struct device *dev)
+{
+	struct gart_device *gart = dev_get_drvdata(dev);
+	unsigned long iova;
+	u32 *data = gart->savedata;
+	unsigned long flags;
+
+	spin_lock_irqsave(&gart->pte_lock, flags);
+	for_each_gart_pte(gart, iova)
+		*(data++) = gart_read_pte(gart, iova);
+	spin_unlock_irqrestore(&gart->pte_lock, flags);
+	return 0;
+}
+
+static int tegra_gart_resume(struct device *dev)
+{
+	struct gart_device *gart = dev_get_drvdata(dev);
+	unsigned long flags;
+
+	spin_lock_irqsave(&gart->pte_lock, flags);
+	do_gart_setup(gart, gart->savedata);
+	spin_unlock_irqrestore(&gart->pte_lock, flags);
+	return 0;
+}
+
+static int tegra_gart_probe(struct platform_device *pdev)
+{
+	struct gart_device *gart;
+	struct resource *res, *res_remap;
+	void __iomem *gart_regs;
+	int err;
+	struct device *dev = &pdev->dev;
+
+	BUILD_BUG_ON(PAGE_SHIFT != GART_PAGE_SHIFT);
+
+	/* the GART memory aperture is required */
+	res = platform_get_resource(pdev, IORESOURCE_MEM, 0);
+	res_remap = platform_get_resource(pdev, IORESOURCE_MEM, 1);
+	if (!res || !res_remap) {
+		dev_err(dev, "GART memory aperture expected\n");
+		return -ENXIO;
+	}
+
+	gart = devm_kzalloc(dev, sizeof(*gart), GFP_KERNEL);
+	if (!gart) {
+		dev_err(dev, "failed to allocate gart_device\n");
+		return -ENOMEM;
+	}
+
+	gart_regs = devm_ioremap(dev, res->start, resource_size(res));
+	if (!gart_regs) {
+		dev_err(dev, "failed to remap GART registers\n");
+		err = -ENXIO;
+		goto fail;
+	}
+
+	gart->dev = &pdev->dev;
+	spin_lock_init(&gart->pte_lock);
+	spin_lock_init(&gart->client_lock);
+	INIT_LIST_HEAD(&gart->client);
+	gart->regs = gart_regs;
+	gart->iovmm_base = (dma_addr_t)res_remap->start;
+	gart->page_count = (resource_size(res_remap) >> GART_PAGE_SHIFT);
+
+	gart->savedata = vmalloc(sizeof(u32) * gart->page_count);
+	if (!gart->savedata) {
+		dev_err(dev, "failed to allocate context save area\n");
+		err = -ENOMEM;
+		goto fail;
+	}
+
+	platform_set_drvdata(pdev, gart);
+	do_gart_setup(gart, NULL);
+	return 0;
+
+fail:
+	if (gart_regs)
+		devm_iounmap(dev, gart_regs);
+	if (gart && gart->savedata)
+		vfree(gart->savedata);
+	devm_kfree(dev, gart);
+	return err;
+}
+
+static int tegra_gart_remove(struct platform_device *pdev)
+{
+	struct gart_device *gart = platform_get_drvdata(pdev);
+	struct device *dev = gart->dev;
+
+	writel(0, gart->regs + GART_CONFIG);
+	if (gart->savedata)
+		vfree(gart->savedata);
+	if (gart->regs)
+		devm_iounmap(dev, gart->regs);
+	devm_kfree(dev, gart);
+	return 0;
+}
+
+const struct dev_pm_ops tegra_gart_pm_ops = {
+	.suspend	= tegra_gart_suspend,
+	.resume		= tegra_gart_resume,
+};
+
+static struct platform_driver tegra_gart_driver = {
+	.probe		= tegra_gart_probe,
+	.remove		= tegra_gart_remove,
+	.driver = {
+		.owner	= THIS_MODULE,
+		.name	= "tegra-gart",
+		.pm	= &tegra_gart_pm_ops,
+	},
+};
+
+static int __devinit tegra_gart_init(void)
+{
+	bus_set_iommu(&platform_bus_type, &gart_iommu_ops);
+	return platform_driver_register(&tegra_gart_driver);
+}
+
+static void __exit tegra_gart_exit(void)
+{
+	platform_driver_unregister(&tegra_gart_driver);
+}
+
+subsys_initcall(tegra_gart_init);
+module_exit(tegra_gart_exit);
-- 
1.7.5.4

Re: [PATCH 1/2] ARM: IOMMU: Tegra20: Add iommu_ops for GART driver

[Joerg Roedel]

Hi,

first, some questions about the GART on your platform.

1) How many GARTs are usually implemented? One GART per device or is
   there a single GART covering all devices, or a mix of both?
2) Are devices allowed to DMA outside of the remappable range or will
   this fail?

Besides that I think IOMMU-API is not yet fully ready for GART-like
drivers like this one. But I will merge it anyway to get things moving.
But please answer or fix my objections first.

On Thu, Jan 05, 2012 at 09:11:48AM +0200, Hiroshi DOYU wrote:
> +static void gart_iommu_domain_destroy(struct iommu_domain *domain)
> +{
> +	struct gart_device *gart = domain->priv;
> +
> +	spin_lock(&gart->client_lock);
> +	if (!list_empty(&gart->client)) {
> +		struct gart_client *c;
> +
> +		list_for_each_entry(c, &gart->client, list)
> +			dev_err(gart->dev,
> +				"%s is still attached\n", dev_name(c->dev));
> +	}
> +	spin_unlock(&gart->client_lock);

gart needs a NULL check. When you create a domain and immediatly destroy
it without ever attaching a device this code will dereference a NULL
pointer.
As a general improvement I suggest that you introduce a gart_domain
structure and store it in domain->priv (assigned in domain_init) instead
of using the hardware descriptor.

> +static int gart_iommu_attach_dev(struct iommu_domain *domain,
> +				 struct device *dev)
> +{
> +	struct gart_device *gart;
> +	struct gart_client *client, *c;
> +	int err = 0;
> +
> +	gart = dev_get_drvdata(dev->parent);
> +	if (!gart)
> +		return -EINVAL;
> +	domain->priv = gart;

What happens when devices behind different GARTs are assigned to the
same domain? domain->priv can only hold a pointer to one hardware GART.
This can be solved by a 'struct gart_domain' holding a linked list to
all gart_devices in this domain.


	Joerg

Re: [PATCH 1/2] ARM: IOMMU: Tegra20: Add iommu_ops for GART driver

[Hiroshi Doyu]

[-- Attachment #1: Type: text/plain, Size: 2433 bytes --]

Hi Joerg,

From: Joerg Roedel <joro-zLv9SwRftAIdnm+yROfE0A@public.gmane.org>
Subject: Re: [PATCH 1/2] ARM: IOMMU: Tegra20: Add iommu_ops for GART driver
Date: Mon, 23 Jan 2012 16:00:48 +0100
Message-ID: <20120123150048.GB6269-zLv9SwRftAIdnm+yROfE0A@public.gmane.org>

> Hi,
> 
> first, some questions about the GART on your platform.
> 
> 1) How many GARTs are usually implemented? One GART per device or is
>    there a single GART covering all devices, or a mix of both?

There's a single GART in the system which takes care of all devices.

> 2) Are devices allowed to DMA outside of the remappable range or will
>    this fail?

Fail.

> 
> Besides that I think IOMMU-API is not yet fully ready for GART-like
> drivers like this one. But I will merge it anyway to get things moving.
> But please answer or fix my objections first.
> 
> On Thu, Jan 05, 2012 at 09:11:48AM +0200, Hiroshi DOYU wrote:
> > +static void gart_iommu_domain_destroy(struct iommu_domain *domain)
> > +{
> > +	struct gart_device *gart = domain->priv;
> > +
> > +	spin_lock(&gart->client_lock);
> > +	if (!list_empty(&gart->client)) {
> > +		struct gart_client *c;
> > +
> > +		list_for_each_entry(c, &gart->client, list)
> > +			dev_err(gart->dev,
> > +				"%s is still attached\n", dev_name(c->dev));
> > +	}
> > +	spin_unlock(&gart->client_lock);
> 
> gart needs a NULL check. When you create a domain and immediatly destroy
> it without ever attaching a device this code will dereference a NULL
> pointer.

Fixed.

> As a general improvement I suggest that you introduce a gart_domain
> structure and store it in domain->priv (assigned in domain_init) instead
> of using the hardware descriptor.
> 
> > +static int gart_iommu_attach_dev(struct iommu_domain *domain,
> > +				 struct device *dev)
> > +{
> > +	struct gart_device *gart;
> > +	struct gart_client *client, *c;
> > +	int err = 0;
> > +
> > +	gart = dev_get_drvdata(dev->parent);
> > +	if (!gart)
> > +		return -EINVAL;
> > +	domain->priv = gart;
> 
> What happens when devices behind different GARTs are assigned to the
> same domain? domain->priv can only hold a pointer to one hardware GART.
> This can be solved by a 'struct gart_domain' holding a linked list to
> all gart_devices in this domain.

As discussed in SMMU thread, I'll revisit the above multiple iommu
device support later.

Attached the update patch.

[-- Warning: decoded text below may be mangled, UTF-8 assumed --]
[-- Attachment #2: 0001-ARM-IOMMU-Tegra20-Add-iommu_ops-for-GART-driver.patch --]
[-- Type: text/x-patch; name="0001-ARM-IOMMU-Tegra20-Add-iommu_ops-for-GART-driver.patch", Size: 14203 bytes --]

From f4c1dbb0dfee79ecb4443777780470f6698044e9 Mon Sep 17 00:00:00 2001
From: Hiroshi DOYU <hdoyu@nvidia.com>
Date: Wed, 16 Nov 2011 17:36:37 +0200
Subject: [PATCH 1/2] ARM: IOMMU: Tegra20: Add iommu_ops for GART driver

Tegra 20 IOMMU H/W, GART (Graphics Address Relocation Table). This
patch implements struct iommu_ops for GART for the upper IOMMU API.

This H/W module supports only single virtual address space(domain),
and manages a single level 1-to-1 mapping H/W translation page table.

Signed-off-by: Hiroshi DOYU <hdoyu@nvidia.com>
---
 drivers/iommu/Kconfig      |   11 +
 drivers/iommu/Makefile     |    1 +
 drivers/iommu/tegra-gart.c |  451 ++++++++++++++++++++++++++++++++++++++++++++
 3 files changed, 463 insertions(+), 0 deletions(-)
 create mode 100644 drivers/iommu/tegra-gart.c

diff --git a/drivers/iommu/Kconfig b/drivers/iommu/Kconfig
index 6bea696..4562698 100644
--- a/drivers/iommu/Kconfig
+++ b/drivers/iommu/Kconfig
@@ -142,4 +142,15 @@ config OMAP_IOMMU_DEBUG
 
          Say N unless you know you need this.
 
+config TEGRA_IOMMU_GART
+	bool "Tegra GART IOMMU Support"
+	depends on ARCH_TEGRA_2x_SOC
+	default y
+	select IOMMU_API
+	help
+	  Enables support for remapping discontiguous physical memory
+	  shared with the operating system into contiguous I/O virtual
+	  space through the GART (Graphics Address Relocation Table)
+	  hardware included on Tegra SoCs.
+
 endif # IOMMU_SUPPORT
diff --git a/drivers/iommu/Makefile b/drivers/iommu/Makefile
index 0e36b49..3238a31 100644
--- a/drivers/iommu/Makefile
+++ b/drivers/iommu/Makefile
@@ -8,3 +8,4 @@ obj-$(CONFIG_IRQ_REMAP) += intr_remapping.o
 obj-$(CONFIG_OMAP_IOMMU) += omap-iommu.o
 obj-$(CONFIG_OMAP_IOVMM) += omap-iovmm.o
 obj-$(CONFIG_OMAP_IOMMU_DEBUG) += omap-iommu-debug.o
+obj-$(CONFIG_TEGRA_IOMMU_GART) += tegra-gart.o
diff --git a/drivers/iommu/tegra-gart.c b/drivers/iommu/tegra-gart.c
new file mode 100644
index 0000000..f6be902
--- /dev/null
+++ b/drivers/iommu/tegra-gart.c
@@ -0,0 +1,451 @@
+/*
+ * IOMMU API for GART in Tegra20
+ *
+ * Copyright (c) 2010-2012, NVIDIA CORPORATION.  All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or modify it
+ * under the terms and conditions of the GNU General Public License,
+ * version 2, as published by the Free Software Foundation.
+ *
+ * This program is distributed in the hope it will be useful, but WITHOUT
+ * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
+ * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
+ * more details.
+ *
+ * You should have received a copy of the GNU General Public License along with
+ * this program; if not, write to the Free Software Foundation, Inc.,
+ * 51 Franklin St - Fifth Floor, Boston, MA 02110-1301 USA.
+ */
+
+#define pr_fmt(fmt)	"%s(): " fmt, __func__
+
+#include <linux/module.h>
+#include <linux/platform_device.h>
+#include <linux/spinlock.h>
+#include <linux/slab.h>
+#include <linux/vmalloc.h>
+#include <linux/mm.h>
+#include <linux/list.h>
+#include <linux/device.h>
+#include <linux/io.h>
+#include <linux/iommu.h>
+
+#include <asm/cacheflush.h>
+
+/* bitmap of the page sizes currently supported */
+#define GART_IOMMU_PGSIZES	(SZ_4K)
+
+#define GART_CONFIG		0x24
+#define GART_ENTRY_ADDR		0x28
+#define GART_ENTRY_DATA		0x2c
+#define GART_ENTRY_PHYS_ADDR_VALID	(1 << 31)
+
+#define GART_PAGE_SHIFT		12
+#define GART_PAGE_SIZE		(1 << GART_PAGE_SHIFT)
+#define GART_PAGE_MASK						\
+	(~(GART_PAGE_SIZE - 1) & ~GART_ENTRY_PHYS_ADDR_VALID)
+
+struct gart_client {
+	struct device		*dev;
+	struct list_head	list;
+};
+
+struct gart_device {
+	void __iomem		*regs;
+	u32			*savedata;
+	u32			page_count;	/* total remappable size */
+	dma_addr_t		iovmm_base;	/* offset to vmm_area */
+	spinlock_t		pte_lock;	/* for pagetable */
+	struct list_head	client;
+	spinlock_t		client_lock;	/* for client list */
+	struct device		*dev;
+};
+
+static struct gart_device *gart_handle; /* unique for a system */
+
+#define GART_PTE(_pfn)						\
+	(GART_ENTRY_PHYS_ADDR_VALID | ((_pfn) << PAGE_SHIFT))
+
+/*
+ * Any interaction between any block on PPSB and a block on APB or AHB
+ * must have these read-back to ensure the APB/AHB bus transaction is
+ * complete before initiating activity on the PPSB block.
+ */
+#define FLUSH_GART_REGS(gart)	((void)readl((gart)->regs + GART_CONFIG))
+
+#define for_each_gart_pte(gart, iova)					\
+	for (iova = gart->iovmm_base;					\
+	     iova < gart->iovmm_base + GART_PAGE_SIZE * gart->page_count; \
+	     iova += GART_PAGE_SIZE)
+
+static inline void gart_set_pte(struct gart_device *gart,
+				unsigned long offs, u32 pte)
+{
+	writel(offs, gart->regs + GART_ENTRY_ADDR);
+	writel(pte, gart->regs + GART_ENTRY_DATA);
+
+	dev_dbg(gart->dev, "%s %08lx:%08x\n",
+		 pte ? "map" : "unmap", offs, pte & GART_PAGE_MASK);
+}
+
+static inline unsigned long gart_read_pte(struct gart_device *gart,
+					  unsigned long offs)
+{
+	unsigned long pte;
+
+	writel(offs, gart->regs + GART_ENTRY_ADDR);
+	pte = readl(gart->regs + GART_ENTRY_DATA);
+
+	return pte;
+}
+
+static void do_gart_setup(struct gart_device *gart, const u32 *data)
+{
+	unsigned long iova;
+
+	for_each_gart_pte(gart, iova)
+		gart_set_pte(gart, iova, data ? *(data++) : 0);
+
+	writel(1, gart->regs + GART_CONFIG);
+	FLUSH_GART_REGS(gart);
+}
+
+#ifdef DEBUG
+static void gart_dump_table(struct gart_device *gart)
+{
+	unsigned long iova;
+	unsigned long flags;
+
+	spin_lock_irqsave(&gart->pte_lock, flags);
+	for_each_gart_pte(gart, iova) {
+		unsigned long pte;
+
+		pte = gart_read_pte(gart, iova);
+
+		dev_dbg(gart->dev, "%s %08lx:%08lx\n",
+			(GART_ENTRY_PHYS_ADDR_VALID & pte) ? "v" : " ",
+			iova, pte & GART_PAGE_MASK);
+	}
+	spin_unlock_irqrestore(&gart->pte_lock, flags);
+}
+#else
+static inline void gart_dump_table(struct gart_device *gart)
+{
+}
+#endif
+
+static inline bool gart_iova_range_valid(struct gart_device *gart,
+					 unsigned long iova, size_t bytes)
+{
+	unsigned long iova_start, iova_end, gart_start, gart_end;
+
+	iova_start = iova;
+	iova_end = iova_start + bytes - 1;
+	gart_start = gart->iovmm_base;
+	gart_end = gart_start + gart->page_count * GART_PAGE_SIZE - 1;
+
+	if (iova_start < gart_start)
+		return false;
+	if (iova_end > gart_end)
+		return false;
+	return true;
+}
+
+static int gart_iommu_attach_dev(struct iommu_domain *domain,
+				 struct device *dev)
+{
+	struct gart_device *gart;
+	struct gart_client *client, *c;
+	int err = 0;
+
+	gart = dev_get_drvdata(dev->parent);
+	if (!gart)
+		return -EINVAL;
+	domain->priv = gart;
+
+	client = devm_kzalloc(gart->dev, sizeof(*c), GFP_KERNEL);
+	if (!client)
+		return -ENOMEM;
+	client->dev = dev;
+
+	spin_lock(&gart->client_lock);
+	list_for_each_entry(c, &gart->client, list) {
+		if (c->dev == dev) {
+			dev_err(gart->dev,
+				"%s is already attached\n", dev_name(dev));
+			err = -EINVAL;
+			goto fail;
+		}
+	}
+	list_add(&client->list, &gart->client);
+	spin_unlock(&gart->client_lock);
+	dev_dbg(gart->dev, "Attached %s\n", dev_name(dev));
+	return 0;
+
+fail:
+	devm_kfree(gart->dev, client);
+	spin_unlock(&gart->client_lock);
+	return err;
+}
+
+static void gart_iommu_detach_dev(struct iommu_domain *domain,
+				  struct device *dev)
+{
+	struct gart_device *gart = domain->priv;
+	struct gart_client *c;
+
+	spin_lock(&gart->client_lock);
+
+	list_for_each_entry(c, &gart->client, list) {
+		if (c->dev == dev) {
+			list_del(&c->list);
+			devm_kfree(gart->dev, c);
+			dev_dbg(gart->dev, "Detached %s\n", dev_name(dev));
+			goto out;
+		}
+	}
+	dev_err(gart->dev, "Couldn't find\n");
+out:
+	spin_unlock(&gart->client_lock);
+}
+
+static int gart_iommu_domain_init(struct iommu_domain *domain)
+{
+	return 0;
+}
+
+static void gart_iommu_domain_destroy(struct iommu_domain *domain)
+{
+	struct gart_device *gart = domain->priv;
+
+	if (!gart)
+		return;
+
+	spin_lock(&gart->client_lock);
+	if (!list_empty(&gart->client)) {
+		struct gart_client *c;
+
+		list_for_each_entry(c, &gart->client, list)
+			gart_iommu_detach_dev(domain, c->dev);
+	}
+	spin_unlock(&gart->client_lock);
+	domain->priv = NULL;
+}
+
+static int gart_iommu_map(struct iommu_domain *domain, unsigned long iova,
+			  phys_addr_t pa, size_t bytes, int prot)
+{
+	struct gart_device *gart = domain->priv;
+	unsigned long flags;
+	unsigned long pfn;
+
+	if (!gart_iova_range_valid(gart, iova, bytes))
+		return -EINVAL;
+
+	spin_lock_irqsave(&gart->pte_lock, flags);
+	pfn = __phys_to_pfn(pa);
+	if (!pfn_valid(pfn)) {
+		dev_err(gart->dev, "Invalid page: %08x\n", pa);
+		spin_unlock(&gart->pte_lock);
+		return -EINVAL;
+	}
+	gart_set_pte(gart, iova, GART_PTE(pfn));
+	FLUSH_GART_REGS(gart);
+	spin_unlock_irqrestore(&gart->pte_lock, flags);
+	return 0;
+}
+
+static size_t gart_iommu_unmap(struct iommu_domain *domain, unsigned long iova,
+			       size_t bytes)
+{
+	struct gart_device *gart = domain->priv;
+	unsigned long flags;
+
+	if (!gart_iova_range_valid(gart, iova, bytes))
+		return -EINVAL;
+
+	spin_lock_irqsave(&gart->pte_lock, flags);
+	gart_set_pte(gart, iova, 0);
+	FLUSH_GART_REGS(gart);
+	spin_unlock_irqrestore(&gart->pte_lock, flags);
+	return 0;
+}
+
+static phys_addr_t gart_iommu_iova_to_phys(struct iommu_domain *domain,
+					   unsigned long iova)
+{
+	struct gart_device *gart = domain->priv;
+	unsigned long pte;
+	phys_addr_t pa;
+	unsigned long flags;
+
+	if (!gart_iova_range_valid(gart, iova, 0))
+		return -EINVAL;
+
+	spin_lock_irqsave(&gart->pte_lock, flags);
+	pte = gart_read_pte(gart, iova);
+	spin_unlock_irqrestore(&gart->pte_lock, flags);
+
+	pa = (pte & GART_PAGE_MASK);
+	if (!pfn_valid(__phys_to_pfn(pa))) {
+		dev_err(gart->dev, "No entry for %08lx:%08x\n", iova, pa);
+		gart_dump_table(gart);
+		return -EINVAL;
+	}
+	return pa;
+}
+
+static int gart_iommu_domain_has_cap(struct iommu_domain *domain,
+				     unsigned long cap)
+{
+	return 0;
+}
+
+static struct iommu_ops gart_iommu_ops = {
+	.domain_init	= gart_iommu_domain_init,
+	.domain_destroy	= gart_iommu_domain_destroy,
+	.attach_dev	= gart_iommu_attach_dev,
+	.detach_dev	= gart_iommu_detach_dev,
+	.map		= gart_iommu_map,
+	.unmap		= gart_iommu_unmap,
+	.iova_to_phys	= gart_iommu_iova_to_phys,
+	.domain_has_cap	= gart_iommu_domain_has_cap,
+	.pgsize_bitmap	= GART_IOMMU_PGSIZES,
+};
+
+static int tegra_gart_suspend(struct device *dev)
+{
+	struct gart_device *gart = dev_get_drvdata(dev);
+	unsigned long iova;
+	u32 *data = gart->savedata;
+	unsigned long flags;
+
+	spin_lock_irqsave(&gart->pte_lock, flags);
+	for_each_gart_pte(gart, iova)
+		*(data++) = gart_read_pte(gart, iova);
+	spin_unlock_irqrestore(&gart->pte_lock, flags);
+	return 0;
+}
+
+static int tegra_gart_resume(struct device *dev)
+{
+	struct gart_device *gart = dev_get_drvdata(dev);
+	unsigned long flags;
+
+	spin_lock_irqsave(&gart->pte_lock, flags);
+	do_gart_setup(gart, gart->savedata);
+	spin_unlock_irqrestore(&gart->pte_lock, flags);
+	return 0;
+}
+
+static int tegra_gart_probe(struct platform_device *pdev)
+{
+	struct gart_device *gart;
+	struct resource *res, *res_remap;
+	void __iomem *gart_regs;
+	int err;
+	struct device *dev = &pdev->dev;
+
+	if (gart_handle)
+		return -EIO;
+
+	BUILD_BUG_ON(PAGE_SHIFT != GART_PAGE_SHIFT);
+
+	/* the GART memory aperture is required */
+	res = platform_get_resource(pdev, IORESOURCE_MEM, 0);
+	res_remap = platform_get_resource(pdev, IORESOURCE_MEM, 1);
+	if (!res || !res_remap) {
+		dev_err(dev, "GART memory aperture expected\n");
+		return -ENXIO;
+	}
+
+	gart = devm_kzalloc(dev, sizeof(*gart), GFP_KERNEL);
+	if (!gart) {
+		dev_err(dev, "failed to allocate gart_device\n");
+		return -ENOMEM;
+	}
+
+	gart_regs = devm_ioremap(dev, res->start, resource_size(res));
+	if (!gart_regs) {
+		dev_err(dev, "failed to remap GART registers\n");
+		err = -ENXIO;
+		goto fail;
+	}
+
+	gart->dev = &pdev->dev;
+	spin_lock_init(&gart->pte_lock);
+	spin_lock_init(&gart->client_lock);
+	INIT_LIST_HEAD(&gart->client);
+	gart->regs = gart_regs;
+	gart->iovmm_base = (dma_addr_t)res_remap->start;
+	gart->page_count = (resource_size(res_remap) >> GART_PAGE_SHIFT);
+
+	gart->savedata = vmalloc(sizeof(u32) * gart->page_count);
+	if (!gart->savedata) {
+		dev_err(dev, "failed to allocate context save area\n");
+		err = -ENOMEM;
+		goto fail;
+	}
+
+	platform_set_drvdata(pdev, gart);
+	do_gart_setup(gart, NULL);
+
+	gart_handle = gart;
+	return 0;
+
+fail:
+	if (gart_regs)
+		devm_iounmap(dev, gart_regs);
+	if (gart && gart->savedata)
+		vfree(gart->savedata);
+	devm_kfree(dev, gart);
+	return err;
+}
+
+static int tegra_gart_remove(struct platform_device *pdev)
+{
+	struct gart_device *gart = platform_get_drvdata(pdev);
+	struct device *dev = gart->dev;
+
+	writel(0, gart->regs + GART_CONFIG);
+	if (gart->savedata)
+		vfree(gart->savedata);
+	if (gart->regs)
+		devm_iounmap(dev, gart->regs);
+	devm_kfree(dev, gart);
+	gart_handle = NULL;
+	return 0;
+}
+
+const struct dev_pm_ops tegra_gart_pm_ops = {
+	.suspend	= tegra_gart_suspend,
+	.resume		= tegra_gart_resume,
+};
+
+static struct platform_driver tegra_gart_driver = {
+	.probe		= tegra_gart_probe,
+	.remove		= tegra_gart_remove,
+	.driver = {
+		.owner	= THIS_MODULE,
+		.name	= "tegra-gart",
+		.pm	= &tegra_gart_pm_ops,
+	},
+};
+
+static int __devinit tegra_gart_init(void)
+{
+	bus_set_iommu(&platform_bus_type, &gart_iommu_ops);
+	return platform_driver_register(&tegra_gart_driver);
+}
+
+static void __exit tegra_gart_exit(void)
+{
+	platform_driver_unregister(&tegra_gart_driver);
+}
+
+subsys_initcall(tegra_gart_init);
+module_exit(tegra_gart_exit);
+
+MODULE_DESCRIPTION("IOMMU API for GART in Tegra20");
+MODULE_AUTHOR("Hiroshi DOYU <hdoyu@nvidia.com>");
+MODULE_LICENSE("GPL v2");
-- 
1.7.5.4

Re: [PATCH 1/2] ARM: IOMMU: Tegra20: Add iommu_ops for GART driver

[joro-zLv9SwRftAIdnm+yROfE0A]

On Wed, Jan 25, 2012 at 08:40:20AM +0100, Hiroshi Doyu wrote:
> From: Hiroshi DOYU <hdoyu-DDmLM1+adcrQT0dZR+AlfA@public.gmane.org>
> Date: Wed, 16 Nov 2011 17:36:37 +0200
> Subject: [PATCH 1/2] ARM: IOMMU: Tegra20: Add iommu_ops for GART driver
> 
> Tegra 20 IOMMU H/W, GART (Graphics Address Relocation Table). This
> patch implements struct iommu_ops for GART for the upper IOMMU API.
> 
> This H/W module supports only single virtual address space(domain),
> and manages a single level 1-to-1 mapping H/W translation page table.

Thanks. Applied with a few minor changes and fixes. Please see below.

> +config TEGRA_IOMMU_GART
> +	bool "Tegra GART IOMMU Support"
> +	depends on ARCH_TEGRA_2x_SOC
> +	default y

I removed 'default y'. New drivers shouldn't be selected by default.

> +static int gart_iommu_map(struct iommu_domain *domain, unsigned long iova,
> +			  phys_addr_t pa, size_t bytes, int prot)
> +{
> +	struct gart_device *gart = domain->priv;
> +	unsigned long flags;
> +	unsigned long pfn;
> +
> +	if (!gart_iova_range_valid(gart, iova, bytes))
> +		return -EINVAL;
> +
> +	spin_lock_irqsave(&gart->pte_lock, flags);
> +	pfn = __phys_to_pfn(pa);
> +	if (!pfn_valid(pfn)) {
> +		dev_err(gart->dev, "Invalid page: %08x\n", pa);
> +		spin_unlock(&gart->pte_lock);

Changed this to spin_unlock_irqrestore().

> +		return -EINVAL;
> +	}
> +	gart_set_pte(gart, iova, GART_PTE(pfn));
> +	FLUSH_GART_REGS(gart);
> +	spin_unlock_irqrestore(&gart->pte_lock, flags);
> +	return 0;
> +}
> +
> +static size_t gart_iommu_unmap(struct iommu_domain *domain, unsigned long iova,
> +			       size_t bytes)
> +{
> +	struct gart_device *gart = domain->priv;
> +	unsigned long flags;
> +
> +	if (!gart_iova_range_valid(gart, iova, bytes))
> +		return -EINVAL;

Return 0 here instead of -EINVAL. Size_t is unsigned and the unmap path
returns 0 on failure.

Re: [PATCH 1/2] ARM: IOMMU: Tegra20: Add iommu_ops for GART driver

[Hiroshi Doyu]

On Thu, 26 Jan 2012 12:58:13 +0100
"joro-zLv9SwRftAIdnm+yROfE0A@public.gmane.org" <joro-zLv9SwRftAIdnm+yROfE0A@public.gmane.org> wrote:

> On Wed, Jan 25, 2012 at 08:40:20AM +0100, Hiroshi Doyu wrote:
> > From: Hiroshi DOYU <hdoyu-DDmLM1+adcrQT0dZR+AlfA@public.gmane.org>
> > Date: Wed, 16 Nov 2011 17:36:37 +0200
> > Subject: [PATCH 1/2] ARM: IOMMU: Tegra20: Add iommu_ops for GART driver
> > 
> > Tegra 20 IOMMU H/W, GART (Graphics Address Relocation Table). This
> > patch implements struct iommu_ops for GART for the upper IOMMU API.
> > 
> > This H/W module supports only single virtual address space(domain),
> > and manages a single level 1-to-1 mapping H/W translation page table.
> 
> Thanks. Applied with a few minor changes and fixes. Please see below.

All fixes are valid. Thanks.

> > +config TEGRA_IOMMU_GART
> > +	bool "Tegra GART IOMMU Support"
> > +	depends on ARCH_TEGRA_2x_SOC
> > +	default y
> 
> I removed 'default y'. New drivers shouldn't be selected by default.
> 
> > +static int gart_iommu_map(struct iommu_domain *domain, unsigned long iova,
> > +			  phys_addr_t pa, size_t bytes, int prot)
> > +{
> > +	struct gart_device *gart = domain->priv;
> > +	unsigned long flags;
> > +	unsigned long pfn;
> > +
> > +	if (!gart_iova_range_valid(gart, iova, bytes))
> > +		return -EINVAL;
> > +
> > +	spin_lock_irqsave(&gart->pte_lock, flags);
> > +	pfn = __phys_to_pfn(pa);
> > +	if (!pfn_valid(pfn)) {
> > +		dev_err(gart->dev, "Invalid page: %08x\n", pa);
> > +		spin_unlock(&gart->pte_lock);
> 
> Changed this to spin_unlock_irqrestore().
> 
> > +		return -EINVAL;
> > +	}
> > +	gart_set_pte(gart, iova, GART_PTE(pfn));
> > +	FLUSH_GART_REGS(gart);
> > +	spin_unlock_irqrestore(&gart->pte_lock, flags);
> > +	return 0;
> > +}
> > +
> > +static size_t gart_iommu_unmap(struct iommu_domain *domain, unsigned long iova,
> > +			       size_t bytes)
> > +{
> > +	struct gart_device *gart = domain->priv;
> > +	unsigned long flags;
> > +
> > +	if (!gart_iova_range_valid(gart, iova, bytes))
> > +		return -EINVAL;
> 
> Return 0 here instead of -EINVAL. Size_t is unsigned and the unmap path
> returns 0 on failure.
> 

[PATCH 2/2] ARM: IOMMU: Tegra30: Add iommu_ops for SMMU driver

[Hiroshi DOYU]

Tegra 30 IOMMU H/W, SMMU (System Memory Management Unit). This patch
implements struct iommu_ops for SMMU for the upper IOMMU API.

This H/W module supports multiple virtual address spaces(domain x4),
and manages 2 level H/W translation pagetable.

Signed-off-by: Hiroshi DOYU <hdoyu@nvidia.com>
---
 arch/arm/mach-tegra/include/mach/smmu.h |   63 ++
 drivers/iommu/Kconfig                   |   11 +
 drivers/iommu/Makefile                  |    1 +
 drivers/iommu/tegra-smmu.c              | 1026 +++++++++++++++++++++++++++++++
 4 files changed, 1101 insertions(+), 0 deletions(-)
 create mode 100644 arch/arm/mach-tegra/include/mach/smmu.h
 create mode 100644 drivers/iommu/tegra-smmu.c

diff --git a/arch/arm/mach-tegra/include/mach/smmu.h b/arch/arm/mach-tegra/include/mach/smmu.h
new file mode 100644
index 0000000..dad403a
--- /dev/null
+++ b/arch/arm/mach-tegra/include/mach/smmu.h
@@ -0,0 +1,63 @@
+/*
+ * IOMMU API for SMMU in Tegra30
+ *
+ * Copyright (c) 2012, NVIDIA CORPORATION.  All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or modify it
+ * under the terms and conditions of the GNU General Public License,
+ * version 2, as published by the Free Software Foundation.
+ *
+ * This program is distributed in the hope it will be useful, but WITHOUT
+ * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
+ * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
+ * more details.
+ *
+ * You should have received a copy of the GNU General Public License along with
+ * this program; if not, write to the Free Software Foundation, Inc.,
+ * 51 Franklin St - Fifth Floor, Boston, MA 02110-1301 USA.
+ */
+
+#ifndef	MACH_SMMU_H
+#define	MACH_SMMU_H
+
+enum smmu_hwgrp {
+	HWGRP_AFI,
+	HWGRP_AVPC,
+	HWGRP_DC,
+	HWGRP_DCB,
+	HWGRP_EPP,
+	HWGRP_G2,
+	HWGRP_HC,
+	HWGRP_HDA,
+	HWGRP_ISP,
+	HWGRP_MPE,
+	HWGRP_NV,
+	HWGRP_NV2,
+	HWGRP_PPCS,
+	HWGRP_SATA,
+	HWGRP_VDE,
+	HWGRP_VI,
+
+	HWGRP_COUNT,
+
+	HWGRP_END = ~0,
+};
+
+#define HWG_AFI		(1 << HWGRP_AFI)
+#define HWG_AVPC	(1 << HWGRP_AVPC)
+#define HWG_DC		(1 << HWGRP_DC)
+#define HWG_DCB		(1 << HWGRP_DCB)
+#define HWG_EPP		(1 << HWGRP_EPP)
+#define HWG_G2		(1 << HWGRP_G2)
+#define HWG_HC		(1 << HWGRP_HC)
+#define HWG_HDA		(1 << HWGRP_HDA)
+#define HWG_ISP		(1 << HWGRP_ISP)
+#define HWG_MPE		(1 << HWGRP_MPE)
+#define HWG_NV		(1 << HWGRP_NV)
+#define HWG_NV2		(1 << HWGRP_NV2)
+#define HWG_PPCS	(1 << HWGRP_PPCS)
+#define HWG_SATA	(1 << HWGRP_SATA)
+#define HWG_VDE		(1 << HWGRP_VDE)
+#define HWG_VI		(1 << HWGRP_VI)
+
+#endif	/* MACH_SMMU_H */
diff --git a/drivers/iommu/Kconfig b/drivers/iommu/Kconfig
index 4562698..4b3d616 100644
--- a/drivers/iommu/Kconfig
+++ b/drivers/iommu/Kconfig
@@ -153,4 +153,15 @@ config TEGRA_IOMMU_GART
 	  space through the GART (Graphics Address Relocation Table)
 	  hardware included on Tegra SoCs.
 
+config TEGRA_IOMMU_SMMU
+	bool "Tegra SMMU IOMMU Support"
+	depends on ARCH_TEGRA_3x_SOC
+	default y
+	select IOMMU_API
+	help
+	  Enables support for remapping discontiguous physical memory
+	  shared with the operating system into contiguous I/O virtual
+	  space through the SMMU (System Memory Management Unit)
+	  hardware included on Tegra SoCs.
+
 endif # IOMMU_SUPPORT
diff --git a/drivers/iommu/Makefile b/drivers/iommu/Makefile
index 3238a31..7ad7a3b 100644
--- a/drivers/iommu/Makefile
+++ b/drivers/iommu/Makefile
@@ -9,3 +9,4 @@ obj-$(CONFIG_OMAP_IOMMU) += omap-iommu.o
 obj-$(CONFIG_OMAP_IOVMM) += omap-iovmm.o
 obj-$(CONFIG_OMAP_IOMMU_DEBUG) += omap-iommu-debug.o
 obj-$(CONFIG_TEGRA_IOMMU_GART) += tegra-gart.o
+obj-$(CONFIG_TEGRA_IOMMU_SMMU) += tegra-smmu.o
diff --git a/drivers/iommu/tegra-smmu.c b/drivers/iommu/tegra-smmu.c
new file mode 100644
index 0000000..7deea32
--- /dev/null
+++ b/drivers/iommu/tegra-smmu.c
@@ -0,0 +1,1026 @@
+/*
+ * IOMMU API for SMMU in Tegra30
+ *
+ * Copyright (c) 2011-2012, NVIDIA CORPORATION.  All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or modify it
+ * under the terms and conditions of the GNU General Public License,
+ * version 2, as published by the Free Software Foundation.
+ *
+ * This program is distributed in the hope it will be useful, but WITHOUT
+ * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
+ * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
+ * more details.
+ *
+ * You should have received a copy of the GNU General Public License along with
+ * this program; if not, write to the Free Software Foundation, Inc.,
+ * 51 Franklin St - Fifth Floor, Boston, MA 02110-1301 USA.
+ */
+
+#define pr_fmt(fmt)	"%s(): " fmt, __func__
+
+#include <linux/module.h>
+#include <linux/platform_device.h>
+#include <linux/spinlock.h>
+#include <linux/slab.h>
+#include <linux/vmalloc.h>
+#include <linux/mm.h>
+#include <linux/pagemap.h>
+#include <linux/device.h>
+#include <linux/sched.h>
+#include <linux/iommu.h>
+#include <linux/io.h>
+
+#include <asm/page.h>
+#include <asm/cacheflush.h>
+
+#include <mach/iomap.h>
+#include <mach/smmu.h>
+
+/* bitmap of the page sizes currently supported */
+#define SMMU_IOMMU_PGSIZES	(SZ_4K | SZ_4M)
+
+#define SMMU_CONFIG				0x10
+#define SMMU_CONFIG_DISABLE			0
+#define SMMU_CONFIG_ENABLE			1
+
+#define SMMU_TLB_CONFIG				0x14
+#define SMMU_TLB_CONFIG_STATS__MASK		(1 << 31)
+#define SMMU_TLB_CONFIG_STATS__ENABLE		(1 << 31)
+#define SMMU_TLB_CONFIG_HIT_UNDER_MISS__ENABLE	(1 << 29)
+#define SMMU_TLB_CONFIG_ACTIVE_LINES__VALUE	0x10
+#define SMMU_TLB_CONFIG_RESET_VAL		0x20000010
+
+#define SMMU_PTC_CONFIG				0x18
+#define SMMU_PTC_CONFIG_STATS__MASK		(1 << 31)
+#define SMMU_PTC_CONFIG_STATS__ENABLE		(1 << 31)
+#define SMMU_PTC_CONFIG_CACHE__ENABLE		(1 << 29)
+#define SMMU_PTC_CONFIG_INDEX_MAP__PATTERN	0x3f
+#define SMMU_PTC_CONFIG_RESET_VAL		0x2000003f
+
+#define SMMU_PTB_ASID				0x1c
+#define SMMU_PTB_ASID_CURRENT_SHIFT		0
+
+#define SMMU_PTB_DATA				0x20
+#define SMMU_PTB_DATA_RESET_VAL			0
+#define SMMU_PTB_DATA_ASID_NONSECURE_SHIFT	29
+#define SMMU_PTB_DATA_ASID_WRITABLE_SHIFT	30
+#define SMMU_PTB_DATA_ASID_READABLE_SHIFT	31
+
+#define SMMU_TLB_FLUSH				0x30
+#define SMMU_TLB_FLUSH_VA_MATCH_ALL		0
+#define SMMU_TLB_FLUSH_VA_MATCH_SECTION		2
+#define SMMU_TLB_FLUSH_VA_MATCH_GROUP		3
+#define SMMU_TLB_FLUSH_ASID_SHIFT		29
+#define SMMU_TLB_FLUSH_ASID_MATCH_DISABLE	0
+#define SMMU_TLB_FLUSH_ASID_MATCH_ENABLE	1
+#define SMMU_TLB_FLUSH_ASID_MATCH_SHIFT		31
+
+#define SMMU_PTC_FLUSH				0x34
+#define SMMU_PTC_FLUSH_TYPE_ALL			0
+#define SMMU_PTC_FLUSH_TYPE_ADR			1
+#define SMMU_PTC_FLUSH_ADR_SHIFT		4
+
+#define SMMU_ASID_SECURITY			0x38
+
+#define SMMU_STATS_TLB_HIT_COUNT		0x1f0
+#define SMMU_STATS_TLB_MISS_COUNT		0x1f4
+#define SMMU_STATS_PTC_HIT_COUNT		0x1f8
+#define SMMU_STATS_PTC_MISS_COUNT		0x1fc
+
+#define SMMU_TRANSLATION_ENABLE_0		0x228
+#define SMMU_TRANSLATION_ENABLE_1		0x22c
+#define SMMU_TRANSLATION_ENABLE_2		0x230
+
+#define SMMU_AFI_ASID	0x238   /* PCIE */
+#define SMMU_AVPC_ASID	0x23c   /* AVP */
+#define SMMU_DC_ASID	0x240   /* Display controller */
+#define SMMU_DCB_ASID	0x244   /* Display controller B */
+#define SMMU_EPP_ASID	0x248   /* Encoder pre-processor */
+#define SMMU_G2_ASID	0x24c   /* 2D engine */
+#define SMMU_HC_ASID	0x250   /* Host1x */
+#define SMMU_HDA_ASID	0x254   /* High-def audio */
+#define SMMU_ISP_ASID	0x258   /* Image signal processor */
+#define SMMU_MPE_ASID	0x264   /* MPEG encoder */
+#define SMMU_NV_ASID	0x268   /* (3D) */
+#define SMMU_NV2_ASID	0x26c   /* (3D) */
+#define SMMU_PPCS_ASID	0x270   /* AHB */
+#define SMMU_SATA_ASID	0x278   /* SATA */
+#define SMMU_VDE_ASID	0x27c   /* Video decoder */
+#define SMMU_VI_ASID	0x280   /* Video input */
+
+#define SMMU_PDE_NEXT_SHIFT		28
+
+/* AHB Arbiter Registers */
+#define AHB_XBAR_CTRL				0xe0
+#define AHB_XBAR_CTRL_SMMU_INIT_DONE_DONE	1
+#define AHB_XBAR_CTRL_SMMU_INIT_DONE_SHIFT	17
+
+#define SMMU_NUM_ASIDS				4
+#define SMMU_TLB_FLUSH_VA_SECTION__MASK		0xffc00000
+#define SMMU_TLB_FLUSH_VA_SECTION__SHIFT	12 /* right shift */
+#define SMMU_TLB_FLUSH_VA_GROUP__MASK		0xffffc000
+#define SMMU_TLB_FLUSH_VA_GROUP__SHIFT		12 /* right shift */
+#define SMMU_TLB_FLUSH_VA(iova, which)	\
+	((((iova) & SMMU_TLB_FLUSH_VA_##which##__MASK) >> \
+		SMMU_TLB_FLUSH_VA_##which##__SHIFT) |	\
+	SMMU_TLB_FLUSH_VA_MATCH_##which)
+#define SMMU_PTB_ASID_CUR(n)	\
+		((n) << SMMU_PTB_ASID_CURRENT_SHIFT)
+#define SMMU_TLB_FLUSH_ASID_MATCH_disable		\
+		(SMMU_TLB_FLUSH_ASID_MATCH_DISABLE <<	\
+			SMMU_TLB_FLUSH_ASID_MATCH_SHIFT)
+#define SMMU_TLB_FLUSH_ASID_MATCH__ENABLE		\
+		(SMMU_TLB_FLUSH_ASID_MATCH_ENABLE <<	\
+			SMMU_TLB_FLUSH_ASID_MATCH_SHIFT)
+
+#define SMMU_PAGE_SHIFT 12
+#define SMMU_PAGE_SIZE	(1 << SMMU_PAGE_SHIFT)
+
+#define SMMU_PDIR_COUNT	1024
+#define SMMU_PDIR_SIZE	(sizeof(unsigned long) * SMMU_PDIR_COUNT)
+#define SMMU_PTBL_COUNT	1024
+#define SMMU_PTBL_SIZE	(sizeof(unsigned long) * SMMU_PTBL_COUNT)
+#define SMMU_PDIR_SHIFT	12
+#define SMMU_PDE_SHIFT	12
+#define SMMU_PTE_SHIFT	12
+#define SMMU_PFN_MASK	0x000fffff
+
+#define SMMU_ADDR_TO_PFN(addr)	((addr) >> 12)
+#define SMMU_ADDR_TO_PDN(addr)	((addr) >> 22)
+#define SMMU_PDN_TO_ADDR(addr)	((pdn) << 22)
+
+#define _READABLE	(1 << SMMU_PTB_DATA_ASID_READABLE_SHIFT)
+#define _WRITABLE	(1 << SMMU_PTB_DATA_ASID_WRITABLE_SHIFT)
+#define _NONSECURE	(1 << SMMU_PTB_DATA_ASID_NONSECURE_SHIFT)
+#define _PDE_NEXT	(1 << SMMU_PDE_NEXT_SHIFT)
+#define _MASK_ATTR	(_READABLE | _WRITABLE | _NONSECURE)
+
+#define _PDIR_ATTR	(_READABLE | _WRITABLE | _NONSECURE)
+
+#define _PDE_ATTR	(_READABLE | _WRITABLE | _NONSECURE)
+#define _PDE_ATTR_N	(_PDE_ATTR | _PDE_NEXT)
+#define _PDE_VACANT(pdn)	(((pdn) << 10) | _PDE_ATTR)
+
+#define _PTE_ATTR	(_READABLE | _WRITABLE | _NONSECURE)
+#define _PTE_VACANT(addr)	(((addr) >> SMMU_PAGE_SHIFT) | _PTE_ATTR)
+
+#define SMMU_MK_PDIR(page, attr)	\
+		((page_to_phys(page) >> SMMU_PDIR_SHIFT) | (attr))
+#define SMMU_MK_PDE(page, attr)		\
+		(unsigned long)((page_to_phys(page) >> SMMU_PDE_SHIFT) | (attr))
+#define SMMU_EX_PTBL_PAGE(pde)		\
+		pfn_to_page((unsigned long)(pde) & SMMU_PFN_MASK)
+#define SMMU_PFN_TO_PTE(pfn, attr)	(unsigned long)((pfn) | (attr))
+
+#define SMMU_ASID_ENABLE(asid)	((asid) | (1 << 31))
+#define SMMU_ASID_DISABLE	0
+#define SMMU_ASID_ASID(n)	((n) & ~SMMU_ASID_ENABLE(0))
+
+#define smmu_client_enable_hwgrp(c, m)	smmu_client_set_hwgrp(c, m, 1)
+#define smmu_client_disable_hwgrp(c)	smmu_client_set_hwgrp(c, 0, 0)
+#define __smmu_client_enable_hwgrp(c, m) __smmu_client_set_hwgrp(c, m, 1)
+#define __smmu_client_disable_hwgrp(c)	__smmu_client_set_hwgrp(c, 0, 0)
+
+#define HWGRP_INIT(client) [HWGRP_##client] = SMMU_##client##_ASID
+
+static const u32 smmu_hwgrp_asid_reg[] = {
+	HWGRP_INIT(AFI),
+	HWGRP_INIT(AVPC),
+	HWGRP_INIT(DC),
+	HWGRP_INIT(DCB),
+	HWGRP_INIT(EPP),
+	HWGRP_INIT(G2),
+	HWGRP_INIT(HC),
+	HWGRP_INIT(HDA),
+	HWGRP_INIT(ISP),
+	HWGRP_INIT(MPE),
+	HWGRP_INIT(NV),
+	HWGRP_INIT(NV2),
+	HWGRP_INIT(PPCS),
+	HWGRP_INIT(SATA),
+	HWGRP_INIT(VDE),
+	HWGRP_INIT(VI),
+};
+#define HWGRP_ASID_REG(x) (smmu_hwgrp_asid_reg[x])
+
+/*
+ * Per client for address space
+ */
+struct smmu_client {
+	struct device		*dev;
+	struct list_head	list;
+	struct smmu_as		*as;
+	u32			hwgrp;
+};
+
+/*
+ * Per address space
+ */
+struct smmu_as {
+	struct smmu_device	*smmu;	/* back pointer to container */
+	unsigned int		asid;
+	spinlock_t		lock;	/* for pagetable */
+	struct page		*pdir_page;
+	unsigned long		pdir_attr;
+	unsigned long		pde_attr;
+	unsigned long		pte_attr;
+	unsigned int		*pte_count;
+
+	struct list_head	client;
+	spinlock_t		client_lock; /* for client list */
+};
+
+/*
+ * Per SMMU device - IOMMU device
+ */
+struct smmu_device {
+	void __iomem	*regs, *regs_ahbarb;
+	unsigned long	iovmm_base;	/* remappable base address */
+	unsigned long	page_count;	/* total remappable size */
+	spinlock_t	lock;
+	char		*name;
+	struct device	*dev;
+	int		num_as;
+	struct smmu_as	*as;		/* Run-time allocated array */
+	struct page *avp_vector_page;	/* dummy page shared by all AS's */
+
+	/*
+	 * Register image savers for suspend/resume
+	 */
+	unsigned long translation_enable_0;
+	unsigned long translation_enable_1;
+	unsigned long translation_enable_2;
+	unsigned long asid_security;
+};
+
+static struct smmu_device *smmu_handle; /* unique for a system */
+
+/*
+ *	SMMU/AHB register accessors
+ */
+static inline u32 smmu_read(struct smmu_device *smmu, size_t offs)
+{
+	return readl(smmu->regs + offs);
+}
+static inline void smmu_write(struct smmu_device *smmu, u32 val, size_t offs)
+{
+	writel(val, smmu->regs + offs);
+}
+
+static inline u32 ahb_read(struct smmu_device *smmu, size_t offs)
+{
+	return readl(smmu->regs_ahbarb + offs);
+}
+static inline void ahb_write(struct smmu_device *smmu, u32 val, size_t offs)
+{
+	writel(val, smmu->regs_ahbarb + offs);
+}
+
+#define VA_PAGE_TO_PA(va, page)	\
+	(page_to_phys(page) + ((unsigned long)(va) & ~PAGE_MASK))
+
+#define FLUSH_CPU_DCACHE(va, page, size)	\
+	do {	\
+		unsigned long _pa_ = VA_PAGE_TO_PA(va, page);		\
+		__cpuc_flush_dcache_area((void *)(va), (size_t)(size));	\
+		outer_flush_range(_pa_, _pa_+(size_t)(size));		\
+	} while (0)
+
+/*
+ * Any interaction between any block on PPSB and a block on APB or AHB
+ * must have these read-back barriers to ensure the APB/AHB bus
+ * transaction is complete before initiating activity on the PPSB
+ * block.
+ */
+#define FLUSH_SMMU_REGS(smmu)	smmu_read(smmu, SMMU_CONFIG)
+
+#define smmu_client_hwgrp(c) (u32)((c)->dev->platform_data)
+
+static int __smmu_client_set_hwgrp(struct smmu_client *c,
+				   unsigned long map, int on)
+{
+	int i;
+	struct smmu_as *as = c->as;
+	u32 val, offs, mask = SMMU_ASID_ENABLE(as->asid);
+	struct smmu_device *smmu = as->smmu;
+
+	WARN_ON(!on && map);
+	if (on && !map)
+		return -EINVAL;
+	if (!on)
+		map = smmu_client_hwgrp(c);
+
+	for_each_set_bit(i, &map, HWGRP_COUNT) {
+		offs = HWGRP_ASID_REG(i);
+		val = smmu_read(smmu, offs);
+		if (on) {
+			if (WARN_ON(val & mask))
+				goto err_hw_busy;
+			val |= mask;
+		} else {
+			WARN_ON((val & mask) == mask);
+			val &= ~mask;
+		}
+		smmu_write(smmu, val, offs);
+	}
+	FLUSH_SMMU_REGS(smmu);
+	c->hwgrp = map;
+	return 0;
+
+err_hw_busy:
+	for_each_set_bit(i, &map, HWGRP_COUNT) {
+		offs = HWGRP_ASID_REG(i);
+		val = smmu_read(smmu, offs);
+		val &= ~mask;
+		smmu_write(smmu, val, offs);
+	}
+	return -EBUSY;
+}
+
+static int smmu_client_set_hwgrp(struct smmu_client *c, u32 map, int on)
+{
+	u32 val;
+	unsigned long flags;
+	struct smmu_as *as = c->as;
+	struct smmu_device *smmu = as->smmu;
+
+	spin_lock_irqsave(&smmu->lock, flags);
+	val = __smmu_client_set_hwgrp(c, map, on);
+	spin_unlock_irqrestore(&smmu->lock, flags);
+	return val;
+}
+
+/*
+ * Flush all TLB entries and all PTC entries
+ * Caller must lock smmu
+ */
+static void smmu_flush_regs(struct smmu_device *smmu, int enable)
+{
+	u32 val;
+
+	smmu_write(smmu, SMMU_PTC_FLUSH_TYPE_ALL, SMMU_PTC_FLUSH);
+	FLUSH_SMMU_REGS(smmu);
+	val = SMMU_TLB_FLUSH_VA_MATCH_ALL |
+		SMMU_TLB_FLUSH_ASID_MATCH_disable;
+	smmu_write(smmu, val, SMMU_TLB_FLUSH);
+
+	if (enable)
+		smmu_write(smmu, SMMU_CONFIG_ENABLE, SMMU_CONFIG);
+	FLUSH_SMMU_REGS(smmu);
+}
+
+static void smmu_setup_regs(struct smmu_device *smmu)
+{
+	int i;
+	u32 val;
+
+	for (i = 0; i < smmu->num_as; i++) {
+		struct smmu_as *as = &smmu->as[i];
+		struct smmu_client *c;
+
+		smmu_write(smmu, SMMU_PTB_ASID_CUR(as->asid), SMMU_PTB_ASID);
+		val = as->pdir_page ?
+			SMMU_MK_PDIR(as->pdir_page, as->pdir_attr) :
+			SMMU_PTB_DATA_RESET_VAL;
+		smmu_write(smmu, val, SMMU_PTB_DATA);
+
+		list_for_each_entry(c, &as->client, list)
+			__smmu_client_set_hwgrp(c, c->hwgrp, 1);
+	}
+
+	smmu_write(smmu, smmu->translation_enable_0, SMMU_TRANSLATION_ENABLE_0);
+	smmu_write(smmu, smmu->translation_enable_1, SMMU_TRANSLATION_ENABLE_1);
+	smmu_write(smmu, smmu->translation_enable_2, SMMU_TRANSLATION_ENABLE_2);
+	smmu_write(smmu, smmu->asid_security, SMMU_ASID_SECURITY);
+	smmu_write(smmu, SMMU_TLB_CONFIG_RESET_VAL, SMMU_TLB_CONFIG);
+	smmu_write(smmu, SMMU_PTC_CONFIG_RESET_VAL, SMMU_PTC_CONFIG);
+
+	smmu_flush_regs(smmu, 1);
+
+	val = ahb_read(smmu, AHB_XBAR_CTRL);
+	val |= AHB_XBAR_CTRL_SMMU_INIT_DONE_DONE <<
+		AHB_XBAR_CTRL_SMMU_INIT_DONE_SHIFT;
+	ahb_write(smmu, val, AHB_XBAR_CTRL);
+}
+
+static void flush_ptc_and_tlb(struct smmu_device *smmu,
+		      struct smmu_as *as, dma_addr_t iova,
+		      unsigned long *pte, struct page *page, int is_pde)
+{
+	u32 val;
+	unsigned long tlb_flush_va = is_pde
+		?  SMMU_TLB_FLUSH_VA(iova, SECTION)
+		:  SMMU_TLB_FLUSH_VA(iova, GROUP);
+
+	val = SMMU_PTC_FLUSH_TYPE_ADR | VA_PAGE_TO_PA(pte, page);
+	smmu_write(smmu, val, SMMU_PTC_FLUSH);
+	FLUSH_SMMU_REGS(smmu);
+	val = tlb_flush_va |
+		SMMU_TLB_FLUSH_ASID_MATCH__ENABLE |
+		(as->asid << SMMU_TLB_FLUSH_ASID_SHIFT);
+	smmu_write(smmu, val, SMMU_TLB_FLUSH);
+	FLUSH_SMMU_REGS(smmu);
+}
+
+static void free_ptbl(struct smmu_as *as, dma_addr_t iova)
+{
+	unsigned long pdn = SMMU_ADDR_TO_PDN(iova);
+	unsigned long *pdir = (unsigned long *)page_address(as->pdir_page);
+
+	if (pdir[pdn] != _PDE_VACANT(pdn)) {
+		dev_dbg(as->smmu->dev, "pdn: %lx\n", pdn);
+
+		ClearPageReserved(SMMU_EX_PTBL_PAGE(pdir[pdn]));
+		__free_page(SMMU_EX_PTBL_PAGE(pdir[pdn]));
+		pdir[pdn] = _PDE_VACANT(pdn);
+		FLUSH_CPU_DCACHE(&pdir[pdn], as->pdir_page, sizeof pdir[pdn]);
+		flush_ptc_and_tlb(as->smmu, as, iova, &pdir[pdn],
+				  as->pdir_page, 1);
+	}
+}
+
+static void free_pdir(struct smmu_as *as)
+{
+	unsigned addr;
+	int count;
+	struct device *dev = as->smmu->dev;
+
+	if (!as->pdir_page)
+		return;
+
+	addr = as->smmu->iovmm_base;
+	count = as->smmu->page_count;
+	while (count-- > 0) {
+		free_ptbl(as, addr);
+		addr += SMMU_PAGE_SIZE * SMMU_PTBL_COUNT;
+	}
+	ClearPageReserved(as->pdir_page);
+	__free_page(as->pdir_page);
+	as->pdir_page = NULL;
+	devm_kfree(dev, as->pte_count);
+	as->pte_count = NULL;
+}
+
+/*
+ * Maps PTBL for given iova and returns the PTE address
+ * Caller must unmap the mapped PTBL returned in *ptbl_page_p
+ */
+static unsigned long *locate_pte(struct smmu_as *as,
+				 dma_addr_t iova, bool allocate,
+				 struct page **ptbl_page_p,
+				 unsigned int **count)
+{
+	unsigned long ptn = SMMU_ADDR_TO_PFN(iova);
+	unsigned long pdn = SMMU_ADDR_TO_PDN(iova);
+	unsigned long *pdir = page_address(as->pdir_page);
+	unsigned long *ptbl;
+
+	if (pdir[pdn] != _PDE_VACANT(pdn)) {
+		/* Mapped entry table already exists */
+		*ptbl_page_p = SMMU_EX_PTBL_PAGE(pdir[pdn]);
+		ptbl = page_address(*ptbl_page_p);
+	} else if (!allocate) {
+		return NULL;
+	} else {
+		int pn;
+		unsigned long addr = SMMU_PDN_TO_ADDR(pdn);
+
+		/* Vacant - allocate a new page table */
+		dev_dbg(as->smmu->dev, "New PTBL pdn: %lx\n", pdn);
+
+		*ptbl_page_p = alloc_page(GFP_ATOMIC);
+		if (!*ptbl_page_p) {
+			dev_err(as->smmu->dev,
+				"failed to allocate smmu_device page table\n");
+			return NULL;
+		}
+		SetPageReserved(*ptbl_page_p);
+		ptbl = (unsigned long *)page_address(*ptbl_page_p);
+		for (pn = 0; pn < SMMU_PTBL_COUNT;
+		     pn++, addr += SMMU_PAGE_SIZE) {
+			ptbl[pn] = _PTE_VACANT(addr);
+		}
+		FLUSH_CPU_DCACHE(ptbl, *ptbl_page_p, SMMU_PTBL_SIZE);
+		pdir[pdn] = SMMU_MK_PDE(*ptbl_page_p,
+					as->pde_attr | _PDE_NEXT);
+		FLUSH_CPU_DCACHE(&pdir[pdn], as->pdir_page, sizeof pdir[pdn]);
+		flush_ptc_and_tlb(as->smmu, as, iova, &pdir[pdn],
+				  as->pdir_page, 1);
+	}
+	*count = &as->pte_count[pdn];
+
+	return &ptbl[ptn % SMMU_PTBL_COUNT];
+}
+
+#ifdef CONFIG_SMMU_SIG_DEBUG
+static void put_signature(struct smmu_as *as,
+			  dma_addr_t iova, unsigned long pfn)
+{
+	struct page *page;
+	unsigned long *vaddr;
+
+	page = pfn_to_page(pfn);
+	vaddr = page_address(page);
+	if (!vaddr)
+		return;
+
+	vaddr[0] = iova;
+	vaddr[1] = pfn << PAGE_SHIFT;
+	FLUSH_CPU_DCACHE(vaddr, page, sizeof(vaddr[0]) * 2);
+}
+#else
+static inline void put_signature(struct smmu_as *as,
+				 unsigned long addr, unsigned long pfn)
+{
+}
+#endif
+
+/*
+ * Caller must lock/unlock as
+ */
+static int alloc_pdir(struct smmu_as *as)
+{
+	unsigned long *pdir;
+	int pdn;
+	u32 val;
+	struct smmu_device *smmu = as->smmu;
+
+	if (as->pdir_page)
+		return 0;
+
+	as->pte_count = devm_kzalloc(smmu->dev,
+		     sizeof(as->pte_count[0]) * SMMU_PDIR_COUNT, GFP_KERNEL);
+	if (!as->pte_count) {
+		dev_err(smmu->dev,
+			"failed to allocate smmu_device PTE cunters\n");
+		return -ENOMEM;
+	}
+	as->pdir_page = alloc_page(GFP_KERNEL | __GFP_DMA);
+	if (!as->pdir_page) {
+		dev_err(smmu->dev,
+			"failed to allocate smmu_device page directory\n");
+		devm_kfree(smmu->dev, as->pte_count);
+		as->pte_count = NULL;
+		return -ENOMEM;
+	}
+	SetPageReserved(as->pdir_page);
+	pdir = page_address(as->pdir_page);
+
+	for (pdn = 0; pdn < SMMU_PDIR_COUNT; pdn++)
+		pdir[pdn] = _PDE_VACANT(pdn);
+	FLUSH_CPU_DCACHE(pdir, as->pdir_page, SMMU_PDIR_SIZE);
+	val = SMMU_PTC_FLUSH_TYPE_ADR | VA_PAGE_TO_PA(pdir, as->pdir_page);
+	smmu_write(smmu, val, SMMU_PTC_FLUSH);
+	FLUSH_SMMU_REGS(as->smmu);
+	val = SMMU_TLB_FLUSH_VA_MATCH_ALL |
+		SMMU_TLB_FLUSH_ASID_MATCH__ENABLE |
+		(as->asid << SMMU_TLB_FLUSH_ASID_SHIFT);
+	smmu_write(smmu, val, SMMU_TLB_FLUSH);
+	FLUSH_SMMU_REGS(as->smmu);
+
+	return 0;
+}
+
+static void __smmu_iommu_unmap(struct smmu_as *as, dma_addr_t iova)
+{
+	unsigned long *pte;
+	struct page *page;
+	unsigned int *count;
+
+	pte = locate_pte(as, iova, false, &page, &count);
+	if (WARN_ON(!pte))
+		return;
+
+	if (WARN_ON(*pte == _PTE_VACANT(iova)))
+		return;
+
+	*pte = _PTE_VACANT(iova);
+	FLUSH_CPU_DCACHE(pte, page, sizeof(*pte));
+	flush_ptc_and_tlb(as->smmu, as, iova, pte, page, 0);
+	if (!--(*count)) {
+		free_ptbl(as, iova);
+		smmu_flush_regs(as->smmu, 0);
+	}
+}
+
+static void __smmu_iommu_map_pfn(struct smmu_as *as, dma_addr_t iova,
+				 unsigned long pfn)
+{
+	struct smmu_device *smmu = as->smmu;
+	unsigned long *pte;
+	unsigned int *count;
+	struct page *page;
+
+	pte = locate_pte(as, iova, true, &page, &count);
+	if (WARN_ON(!pte))
+		return;
+
+	if (*pte == _PTE_VACANT(iova))
+		(*count)++;
+	*pte = SMMU_PFN_TO_PTE(pfn, as->pte_attr);
+	if (unlikely((*pte == _PTE_VACANT(iova))))
+		(*count)--;
+	FLUSH_CPU_DCACHE(pte, page, sizeof(*pte));
+	flush_ptc_and_tlb(smmu, as, iova, pte, page, 0);
+	put_signature(as, iova, pfn);
+}
+
+static int smmu_iommu_map(struct iommu_domain *domain, unsigned long iova,
+			  phys_addr_t pa, size_t bytes, int prot)
+{
+	struct smmu_as *as = domain->priv;
+	unsigned long pfn = __phys_to_pfn(pa);
+	unsigned long flags;
+
+	dev_dbg(as->smmu->dev, "[%d] %08lx:%08x\n", as->asid, iova, pa);
+
+	if (!pfn_valid(pfn))
+		return -ENOMEM;
+
+	spin_lock_irqsave(&as->lock, flags);
+	__smmu_iommu_map_pfn(as, iova, pfn);
+	spin_unlock_irqrestore(&as->lock, flags);
+	return 0;
+}
+
+static size_t smmu_iommu_unmap(struct iommu_domain *domain, unsigned long iova,
+			       size_t bytes)
+{
+	struct smmu_as *as = domain->priv;
+	unsigned long flags;
+
+	dev_dbg(as->smmu->dev, "[%d] %08lx\n", as->asid, iova);
+
+	spin_lock_irqsave(&as->lock, flags);
+	__smmu_iommu_unmap(as, iova);
+	spin_unlock_irqrestore(&as->lock, flags);
+	return SMMU_PAGE_SIZE;
+}
+
+static phys_addr_t smmu_iommu_iova_to_phys(struct iommu_domain *domain,
+					   unsigned long iova)
+{
+	struct smmu_as *as = domain->priv;
+	unsigned long *pte;
+	unsigned int *count;
+	struct page *page;
+	unsigned long pfn;
+	unsigned long flags;
+
+	spin_lock_irqsave(&as->lock, flags);
+
+	pte = locate_pte(as, iova, true, &page, &count);
+	pfn = *pte & SMMU_PFN_MASK;
+	WARN_ON(!pfn_valid(pfn));
+	dev_dbg(as->smmu->dev,
+		"iova:%08lx pfn:%08lx asid:%d\n", iova, pfn, as->asid);
+
+	spin_unlock_irqrestore(&as->lock, flags);
+	return PFN_PHYS(pfn);
+}
+
+static int smmu_iommu_domain_has_cap(struct iommu_domain *domain,
+				     unsigned long cap)
+{
+	return 0;
+}
+
+static int smmu_iommu_domain_init(struct iommu_domain *domain)
+{
+	int i;
+	unsigned long flags;
+	struct smmu_as *as;
+	struct smmu_device *smmu = smmu_handle;
+
+	/* Look for a free AS with lock held */
+	for  (i = 0; i < smmu->num_as; i++) {
+		struct smmu_as *tmp = &smmu->as[i];
+
+		spin_lock_irqsave(&tmp->lock, flags);
+		if (!tmp->pdir_page) {
+			as = tmp;
+			goto found;
+		}
+		spin_unlock_irqrestore(&tmp->lock, flags);
+	}
+	dev_err(smmu->dev, "no free AS\n");
+	return -ENODEV;
+
+found:
+	if (alloc_pdir(as) < 0)
+		goto err_alloc_pdir;
+
+	spin_lock(&smmu->lock);
+
+	/* Update PDIR register */
+	smmu_write(smmu, SMMU_PTB_ASID_CUR(as->asid), SMMU_PTB_ASID);
+	smmu_write(smmu,
+		   SMMU_MK_PDIR(as->pdir_page, as->pdir_attr), SMMU_PTB_DATA);
+	FLUSH_SMMU_REGS(smmu);
+
+	spin_unlock(&smmu->lock);
+
+	spin_unlock_irqrestore(&as->lock, flags);
+	domain->priv = as;
+	dev_dbg(smmu->dev, "smmu_as@%p\n", as);
+	return 0;
+
+err_alloc_pdir:
+	spin_unlock_irqrestore(&as->lock, flags);
+	return -ENODEV;
+}
+
+static void smmu_iommu_domain_destroy(struct iommu_domain *domain)
+{
+	struct smmu_as *as = domain->priv;
+	struct smmu_device *smmu = as->smmu;
+	unsigned long flags;
+
+	spin_lock_irqsave(&as->lock, flags);
+
+	if (as->pdir_page) {
+		spin_lock(&smmu->lock);
+		smmu_write(smmu, SMMU_PTB_ASID_CUR(as->asid), SMMU_PTB_ASID);
+		smmu_write(smmu, SMMU_PTB_DATA_RESET_VAL, SMMU_PTB_DATA);
+		FLUSH_SMMU_REGS(smmu);
+		spin_unlock(&smmu->lock);
+
+		free_pdir(as);
+	}
+
+	if (!list_empty(&as->client)) {
+		struct smmu_client *c;
+
+		list_for_each_entry(c, &as->client, list)
+			dev_err(smmu->dev,
+				"%s is still attached\n", dev_name(c->dev));
+	}
+
+	spin_unlock_irqrestore(&as->lock, flags);
+
+	domain->priv = NULL;
+	dev_dbg(smmu->dev, "smmu_as@%p\n", as);
+}
+
+static int smmu_iommu_attach_dev(struct iommu_domain *domain,
+				 struct device *dev)
+{
+	struct smmu_as *as = domain->priv;
+	struct smmu_device *smmu = as->smmu;
+	struct smmu_client *client, *c;
+	u32 map;
+	int err;
+
+	client = devm_kzalloc(smmu->dev, sizeof(*c), GFP_KERNEL);
+	if (!client)
+		return -ENOMEM;
+	client->dev = dev;
+	client->as = as;
+	map = (unsigned long)dev->platform_data;
+	if (!map)
+		return -EINVAL;
+
+	err = smmu_client_enable_hwgrp(client, map);
+	if (err)
+		goto err_hwgrp;
+
+	spin_lock(&as->client_lock);
+	list_for_each_entry(c, &as->client, list) {
+		if (c->dev == dev) {
+			dev_err(smmu->dev,
+				"%s is already attached\n", dev_name(c->dev));
+			err = -EINVAL;
+			goto err_client;
+		}
+	}
+	list_add(&client->list, &as->client);
+	spin_unlock(&as->client_lock);
+
+	/*
+	 * Reserve "page zero" for AVP vectors using a common dummy
+	 * page.
+	 */
+	if (map & HWG_AVPC) {
+		struct page *page;
+
+		page = as->smmu->avp_vector_page;
+		__smmu_iommu_map_pfn(as, 0, page_to_pfn(page));
+
+		pr_info("Reserve \"page zero\" for AVP vectors using a common dummy\n");
+	}
+
+	dev_dbg(smmu->dev, "%s is attached\n", dev_name(c->dev));
+	return 0;
+
+err_client:
+	smmu_client_disable_hwgrp(client);
+	spin_unlock(&as->client_lock);
+err_hwgrp:
+	devm_kfree(smmu->dev, client);
+	return err;
+}
+
+static void smmu_iommu_detach_dev(struct iommu_domain *domain,
+				  struct device *dev)
+{
+	struct smmu_as *as = domain->priv;
+	struct smmu_device *smmu = as->smmu;
+	struct smmu_client *c;
+
+	spin_lock(&as->client_lock);
+
+	list_for_each_entry(c, &as->client, list) {
+		if (c->dev == dev) {
+			smmu_client_disable_hwgrp(c);
+			list_del(&c->list);
+			devm_kfree(smmu->dev, c);
+			c->as = NULL;
+			dev_dbg(smmu->dev,
+				"%s is detached\n", dev_name(c->dev));
+			goto out;
+		}
+	}
+	dev_err(smmu->dev, "Couldn't find %s\n", dev_name(c->dev));
+out:
+	spin_unlock(&as->client_lock);
+}
+
+static struct iommu_ops smmu_iommu_ops = {
+	.domain_init	= smmu_iommu_domain_init,
+	.domain_destroy	= smmu_iommu_domain_destroy,
+	.attach_dev	= smmu_iommu_attach_dev,
+	.detach_dev	= smmu_iommu_detach_dev,
+	.map		= smmu_iommu_map,
+	.unmap		= smmu_iommu_unmap,
+	.iova_to_phys	= smmu_iommu_iova_to_phys,
+	.domain_has_cap	= smmu_iommu_domain_has_cap,
+	.pgsize_bitmap	= SMMU_IOMMU_PGSIZES,
+};
+
+static int tegra_smmu_suspend(struct device *dev)
+{
+	struct smmu_device *smmu = dev_get_drvdata(dev);
+
+	smmu->translation_enable_0 = smmu_read(smmu, SMMU_TRANSLATION_ENABLE_0);
+	smmu->translation_enable_1 = smmu_read(smmu, SMMU_TRANSLATION_ENABLE_1);
+	smmu->translation_enable_2 = smmu_read(smmu, SMMU_TRANSLATION_ENABLE_2);
+	smmu->asid_security = smmu_read(smmu, SMMU_ASID_SECURITY);
+	return 0;
+}
+
+static int tegra_smmu_resume(struct device *dev)
+{
+	struct smmu_device *smmu = dev_get_drvdata(dev);
+	unsigned long flags;
+
+	spin_lock_irqsave(&smmu->lock, flags);
+	smmu_setup_regs(smmu);
+	spin_unlock_irqrestore(&smmu->lock, flags);
+	return 0;
+}
+
+static int tegra_smmu_probe(struct platform_device *pdev)
+{
+	struct smmu_device *smmu;
+	struct resource *regs, *regs2, *window;
+	struct device *dev = &pdev->dev;
+	int i, err = 0;
+
+	BUILD_BUG_ON(PAGE_SHIFT != SMMU_PAGE_SHIFT);
+
+	regs = platform_get_resource(pdev, IORESOURCE_MEM, 0);
+	regs2 = platform_get_resource(pdev, IORESOURCE_MEM, 1);
+	window = platform_get_resource(pdev, IORESOURCE_MEM, 2);
+	if (!regs || !regs2 || !window) {
+		dev_err(dev, "No SMMU resources\n");
+		return -ENODEV;
+	}
+
+	smmu = devm_kzalloc(dev, sizeof(*smmu), GFP_KERNEL);
+	if (!smmu) {
+		dev_err(dev, "failed to allocate smmu_device\n");
+		return -ENOMEM;
+	}
+
+	smmu->dev = dev;
+	smmu->num_as = SMMU_NUM_ASIDS;
+	smmu->iovmm_base = (unsigned long)window->start;
+	smmu->page_count = resource_size(window) >> SMMU_PAGE_SHIFT;
+	smmu->regs = devm_ioremap(dev, regs->start, resource_size(regs));
+	smmu->regs_ahbarb = devm_ioremap(dev, regs2->start,
+					 resource_size(regs2));
+	if (!smmu->regs || !smmu->regs_ahbarb) {
+		dev_err(dev, "failed to remap SMMU registers\n");
+		err = -ENXIO;
+		goto fail;
+	}
+
+	smmu->translation_enable_0 = ~0;
+	smmu->translation_enable_1 = ~0;
+	smmu->translation_enable_2 = ~0;
+	smmu->asid_security = 0;
+
+	smmu->as = devm_kzalloc(dev,
+			sizeof(smmu->as[0]) * smmu->num_as, GFP_KERNEL);
+	if (!smmu->as) {
+		dev_err(dev, "failed to allocate smmu_as\n");
+		err = -ENOMEM;
+		goto fail;
+	}
+
+	for (i = 0; i < smmu->num_as; i++) {
+		struct smmu_as *as = &smmu->as[i];
+
+		as->smmu = smmu;
+		as->asid = i;
+		as->pdir_attr = _PDIR_ATTR;
+		as->pde_attr = _PDE_ATTR;
+		as->pte_attr = _PTE_ATTR;
+
+		spin_lock_init(&as->lock);
+		INIT_LIST_HEAD(&as->client);
+	}
+	spin_lock_init(&smmu->lock);
+	smmu_setup_regs(smmu);
+	platform_set_drvdata(pdev, smmu);
+	smmu_handle = smmu;
+
+	smmu->avp_vector_page = alloc_page(GFP_KERNEL);
+	if (!smmu->avp_vector_page)
+		goto fail;
+	return 0;
+
+fail:
+	if (smmu->avp_vector_page)
+		__free_page(smmu->avp_vector_page);
+	if (smmu->regs)
+		devm_iounmap(dev, smmu->regs);
+	if (smmu->regs_ahbarb)
+		devm_iounmap(dev, smmu->regs_ahbarb);
+	if (smmu && smmu->as) {
+		for (i = 0; i < smmu->num_as; i++) {
+			if (smmu->as[i].pdir_page) {
+				ClearPageReserved(smmu->as[i].pdir_page);
+				__free_page(smmu->as[i].pdir_page);
+			}
+		}
+		devm_kfree(dev, smmu->as);
+	}
+	devm_kfree(dev, smmu);
+	return err;
+}
+
+static int tegra_smmu_remove(struct platform_device *pdev)
+{
+	struct smmu_device *smmu = platform_get_drvdata(pdev);
+	struct device *dev = smmu->dev;
+
+	smmu_write(smmu, SMMU_CONFIG_DISABLE, SMMU_CONFIG);
+	platform_set_drvdata(pdev, NULL);
+	if (smmu->as) {
+		int i;
+
+		for (i = 0; i < smmu->num_as; i++)
+			free_pdir(&smmu->as[i]);
+		devm_kfree(dev, smmu->as);
+	}
+	if (smmu->avp_vector_page)
+		__free_page(smmu->avp_vector_page);
+	if (smmu->regs)
+		devm_iounmap(dev, smmu->regs);
+	if (smmu->regs_ahbarb)
+		devm_iounmap(dev, smmu->regs_ahbarb);
+	devm_kfree(dev, smmu);
+	smmu_handle = NULL;
+	return 0;
+}
+
+const struct dev_pm_ops tegra_smmu_pm_ops = {
+	.suspend	= tegra_smmu_suspend,
+	.resume		= tegra_smmu_resume,
+};
+
+static struct platform_driver tegra_smmu_driver = {
+	.probe		= tegra_smmu_probe,
+	.remove		= tegra_smmu_remove,
+	.driver = {
+		.owner	= THIS_MODULE,
+		.name	= "tegra-smmu",
+		.pm	= &tegra_smmu_pm_ops,
+	},
+};
+
+static int __devinit tegra_smmu_init(void)
+{
+	bus_set_iommu(&platform_bus_type, &smmu_iommu_ops);
+	return platform_driver_register(&tegra_smmu_driver);
+}
+
+static void __exit tegra_smmu_exit(void)
+{
+	platform_driver_unregister(&tegra_smmu_driver);
+}
+
+subsys_initcall(tegra_smmu_init);
+module_exit(tegra_smmu_exit);
-- 
1.7.5.4

Re: [PATCH 2/2] ARM: IOMMU: Tegra30: Add iommu_ops for SMMU driver

[Joerg Roedel]

Hi,

please see my comments inline. When you fix these issues I think the
driver is ready for merging.

On Thu, Jan 05, 2012 at 09:11:49AM +0200, Hiroshi DOYU wrote:
> +static int smmu_iommu_map(struct iommu_domain *domain, unsigned long iova,
> +			  phys_addr_t pa, size_t bytes, int prot)
> +{
> +	struct smmu_as *as = domain->priv;
> +	unsigned long pfn = __phys_to_pfn(pa);
> +	unsigned long flags;
> +
> +	dev_dbg(as->smmu->dev, "[%d] %08lx:%08x\n", as->asid, iova, pa);
> +
> +	if (!pfn_valid(pfn))
> +		return -ENOMEM;
> +
> +	spin_lock_irqsave(&as->lock, flags);
> +	__smmu_iommu_map_pfn(as, iova, pfn);
> +	spin_unlock_irqrestore(&as->lock, flags);
> +	return 0;

Why do you completly ignore the size parameter in this function (and
in the unmap part below)?
According to the page-sizes you export to the generic layer size can be
4k or 4M. You need to take care of that in this function.

> +static void smmu_iommu_domain_destroy(struct iommu_domain *domain)
> +{
> +	struct smmu_as *as = domain->priv;
> +	struct smmu_device *smmu = as->smmu;
> +	unsigned long flags;
> +
> +	spin_lock_irqsave(&as->lock, flags);
> +
> +	if (as->pdir_page) {
> +		spin_lock(&smmu->lock);
> +		smmu_write(smmu, SMMU_PTB_ASID_CUR(as->asid), SMMU_PTB_ASID);
> +		smmu_write(smmu, SMMU_PTB_DATA_RESET_VAL, SMMU_PTB_DATA);
> +		FLUSH_SMMU_REGS(smmu);
> +		spin_unlock(&smmu->lock);
> +
> +		free_pdir(as);
> +	}
> +
> +	if (!list_empty(&as->client)) {
> +		struct smmu_client *c;
> +
> +		list_for_each_entry(c, &as->client, list)
> +			dev_err(smmu->dev,
> +				"%s is still attached\n", dev_name(c->dev));

This is not an error. Just detach the devices when they are still
attached to the domain.

> +	}
> +
> +	spin_unlock_irqrestore(&as->lock, flags);
> +
> +	domain->priv = NULL;
> +	dev_dbg(smmu->dev, "smmu_as@%p\n", as);
> +}
> +
> +static int smmu_iommu_attach_dev(struct iommu_domain *domain,
> +				 struct device *dev)
> +{
> +	struct smmu_as *as = domain->priv;
> +	struct smmu_device *smmu = as->smmu;

Hmm, this looks like there is a 1-1 mapping between hardware SMMU
devices and domains. This is not consistent with IOMMU-API semantics
where a domain can contain devices behind different SMMUs. Please fix
that.


Thanks,

	Joerg

Re: [PATCH 2/2] ARM: IOMMU: Tegra30: Add iommu_ops for SMMU driver

[Hiroshi Doyu]

Hi Joerg,

From: Joerg Roedel <joro-zLv9SwRftAIdnm+yROfE0A@public.gmane.org>
Subject: Re: [PATCH 2/2] ARM: IOMMU: Tegra30: Add iommu_ops for SMMU driver
Date: Mon, 23 Jan 2012 16:43:10 +0100
Message-ID: <20120123154310.GC6269-zLv9SwRftAIdnm+yROfE0A@public.gmane.org>

> Hi,
> 
> please see my comments inline. When you fix these issues I think the
> driver is ready for merging.
>
> On Thu, Jan 05, 2012 at 09:11:49AM +0200, Hiroshi DOYU wrote:
> > +static int smmu_iommu_map(struct iommu_domain *domain, unsigned long iova,
> > +			  phys_addr_t pa, size_t bytes, int prot)
> > +{
> > +	struct smmu_as *as = domain->priv;
> > +	unsigned long pfn = __phys_to_pfn(pa);
> > +	unsigned long flags;
> > +
> > +	dev_dbg(as->smmu->dev, "[%d] %08lx:%08x\n", as->asid, iova, pa);
> > +
> > +	if (!pfn_valid(pfn))
> > +		return -ENOMEM;
> > +
> > +	spin_lock_irqsave(&as->lock, flags);
> > +	__smmu_iommu_map_pfn(as, iova, pfn);
> > +	spin_unlock_irqrestore(&as->lock, flags);
> > +	return 0;
> 
> Why do you completly ignore the size parameter in this function (and
> in the unmap part below)?
> According to the page-sizes you export to the generic layer size can be
> 4k or 4M. You need to take care of that in this function.

I'll drop 4MB support here once. I'll make another patch for 4MB page
support later.

> > +static void smmu_iommu_domain_destroy(struct iommu_domain *domain)
> > +{
> > +	struct smmu_as *as = domain->priv;
> > +	struct smmu_device *smmu = as->smmu;
> > +	unsigned long flags;
> > +
> > +	spin_lock_irqsave(&as->lock, flags);
> > +
> > +	if (as->pdir_page) {
> > +		spin_lock(&smmu->lock);
> > +		smmu_write(smmu, SMMU_PTB_ASID_CUR(as->asid), SMMU_PTB_ASID);
> > +		smmu_write(smmu, SMMU_PTB_DATA_RESET_VAL, SMMU_PTB_DATA);
> > +		FLUSH_SMMU_REGS(smmu);
> > +		spin_unlock(&smmu->lock);
> > +
> > +		free_pdir(as);
> > +	}
> > +
> > +	if (!list_empty(&as->client)) {
> > +		struct smmu_client *c;
> > +
> > +		list_for_each_entry(c, &as->client, list)
> > +			dev_err(smmu->dev,
> > +				"%s is still attached\n", dev_name(c->dev));
> 
> This is not an error. Just detach the devices when they are still
> attached to the domain.

Ok

> > +	}
> > +
> > +	spin_unlock_irqrestore(&as->lock, flags);
> > +
> > +	domain->priv = NULL;
> > +	dev_dbg(smmu->dev, "smmu_as@%p\n", as);
> > +}
> > +
> > +static int smmu_iommu_attach_dev(struct iommu_domain *domain,
> > +				 struct device *dev)
> > +{
> > +	struct smmu_as *as = domain->priv;
> > +	struct smmu_device *smmu = as->smmu;
> 
> Hmm, this looks like there is a 1-1 mapping between hardware SMMU
> devices and domains. This is not consistent with IOMMU-API semantics
> where a domain can contain devices behind different SMMUs. Please fix
> that.

I'm a bit confused with the concept of "domain". I thought that
"domain" is equivalent to a "virtual address space". Usually a IOMMU
device provides a virtual address space for multiple client
devices. IOW, a IOMMU device provides a virtual address space, which
can be shared with multiple client devices.

Actually Tegra SMMU case, a single IOMMU device has 4 different
virtual address speace("smmu_as"). Each "smmu_as" has its own virtual
address space. "smmu_as[i]" has mutiple "smmu_client" devices.

  smmu_as[i] == domain[i]

I don't understand why "a domain can contain devices behind different
SMMUs" because those client devices belong to different virtual
address spaces, and they should belong to different "domains".

Could you please explain a bit more about "domain"?


> 
> 
> Thanks,
> 
> 	Joerg
> 

Re: [PATCH 2/2] ARM: IOMMU: Tegra30: Add iommu_ops for SMMU driver

[Joerg Roedel]

On Tue, Jan 24, 2012 at 10:57:01AM +0100, Hiroshi Doyu wrote:
> > Why do you completly ignore the size parameter in this function (and
> > in the unmap part below)?
> > According to the page-sizes you export to the generic layer size can be
> > 4k or 4M. You need to take care of that in this function.
> 
> I'll drop 4MB support here once. I'll make another patch for 4MB page
> support later.

Okay, so when you only export 4k everything should be fine.

> > Hmm, this looks like there is a 1-1 mapping between hardware SMMU
> > devices and domains. This is not consistent with IOMMU-API semantics
> > where a domain can contain devices behind different SMMUs. Please fix
> > that.
> 
> I'm a bit confused with the concept of "domain". I thought that
> "domain" is equivalent to a "virtual address space". Usually a IOMMU
> device provides a virtual address space for multiple client
> devices. IOW, a IOMMU device provides a virtual address space, which
> can be shared with multiple client devices.
> 
> Actually Tegra SMMU case, a single IOMMU device has 4 different
> virtual address speace("smmu_as"). Each "smmu_as" has its own virtual
> address space. "smmu_as[i]" has mutiple "smmu_client" devices.
> 
>   smmu_as[i] == domain[i]
> 
> I don't understand why "a domain can contain devices behind different
> SMMUs" because those client devices belong to different virtual
> address spaces, and they should belong to different "domains".
> 
> Could you please explain a bit more about "domain"?

A domain is, as you said, a virtual address space for IO devices. But
the important point is, an arbitrary number of devices can be part of a
domain. This also means that the devices can be behind different
hardware SMMUs. In this case your driver needs to program the page-table
pointer into more than one SMMU to give devices behind different SMMUs
the same address space.


	Joerg

-- 
AMD Operating System Research Center

Advanced Micro Devices GmbH Einsteinring 24 85609 Dornach
General Managers: Alberto Bozzo
Registration: Dornach, Landkr. Muenchen; Registerger. Muenchen, HRB Nr. 43632

Re: [PATCH 2/2] ARM: IOMMU: Tegra30: Add iommu_ops for SMMU driver

[Hiroshi Doyu]

From: Joerg Roedel <joerg.roedel-5C7GfCeVMHo@public.gmane.org>
Subject: Re: [PATCH 2/2] ARM: IOMMU: Tegra30: Add iommu_ops for SMMU driver
Date: Tue, 24 Jan 2012 12:04:44 +0100
Message-ID: <20120124110444.GB19255-5C7GfCeVMHo@public.gmane.org>

> > > Hmm, this looks like there is a 1-1 mapping between hardware SMMU
> > > devices and domains. This is not consistent with IOMMU-API semantics
> > > where a domain can contain devices behind different SMMUs. Please fix
> > > that.
> > 
> > I'm a bit confused with the concept of "domain". I thought that
> > "domain" is equivalent to a "virtual address space". Usually a IOMMU
> > device provides a virtual address space for multiple client
> > devices. IOW, a IOMMU device provides a virtual address space, which
> > can be shared with multiple client devices.
> > 
> > Actually Tegra SMMU case, a single IOMMU device has 4 different
> > virtual address speace("smmu_as"). Each "smmu_as" has its own virtual
> > address space. "smmu_as[i]" has mutiple "smmu_client" devices.
> > 
> >   smmu_as[i] == domain[i]
> > 
> > I don't understand why "a domain can contain devices behind different
> > SMMUs" because those client devices belong to different virtual
> > address spaces, and they should belong to different "domains".
> > 
> > Could you please explain a bit more about "domain"?
> 
> A domain is, as you said, a virtual address space for IO devices. But
> the important point is, an arbitrary number of devices can be part of a
> domain. This also means that the devices can be behind different
> hardware SMMUs. In this case your driver needs to program the page-table
> pointer into more than one SMMU to give devices behind different SMMUs
> the same address space.

Thank you for explaining.

Does the above mean that a buffer can be shared with different devices
which belong to different IOMMU devices(virtual address spaces)?

For example, assuming the following:

- We have "struct iommu_domain *domain1".
- "domain1" has iommu device "iommu_dev1" and "iommu_dev2".
- "iommu_dev1" has "client_dev1" and "client_dev2".
- "iommu_dev2" has "client_dev3" and "client_dev4".

"iommu_map(domain1, iova, pa, ...)" will create the following mapping
___at once___:

- (iova)-(pa) mapping in iommu_dev1(iommmu_dev1's virtual address space)
- (iova)-(pa) mapping in iommu_dev2(iommmu_dev2's virtual address space)

Is the above correct?

It seems that the same (iova) is used for different virtual address
spaces. What kind of case is this beneficial most in?

Re: [PATCH 2/2] ARM: IOMMU: Tegra30: Add iommu_ops for SMMU driver

[joerg.roedel-5C7GfCeVMHo]

On Tue, Jan 24, 2012 at 12:36:14PM +0100, Hiroshi Doyu wrote:
> > A domain is, as you said, a virtual address space for IO devices. But
> > the important point is, an arbitrary number of devices can be part of a
> > domain. This also means that the devices can be behind different
> > hardware SMMUs. In this case your driver needs to program the page-table
> > pointer into more than one SMMU to give devices behind different SMMUs
> > the same address space.
> 
> Thank you for explaining.
> 
> Does the above mean that a buffer can be shared with different devices
> which belong to different IOMMU devices(virtual address spaces)?
> 
> For example, assuming the following:
> 
> - We have "struct iommu_domain *domain1".
> - "domain1" has iommu device "iommu_dev1" and "iommu_dev2".
> - "iommu_dev1" has "client_dev1" and "client_dev2".
> - "iommu_dev2" has "client_dev3" and "client_dev4".
> 
> "iommu_map(domain1, iova, pa, ...)" will create the following mapping
> ___at once___:
> 
> - (iova)-(pa) mapping in iommu_dev1(iommmu_dev1's virtual address space)
> - (iova)-(pa) mapping in iommu_dev2(iommmu_dev2's virtual address space)
> 
> Is the above correct?

Yes, this is correct.

> It seems that the same (iova) is used for different virtual address
> spaces. What kind of case is this beneficial most in?

It is actually the _same_ virtual address space which is used by
iommu_dev1 and iommu_dev2. Think of it like multiple threads of a single
process. They also share the address space.
This is a requirement of the iommu-api which is beneficial for
virtualization and simplifies the usage of the api in general. The user
does not need to care which devices can be assigned to which domain
because of underlying hardware constraints. The goal of the iommu-api is
to hide such contraints.


	Joerg

-- 
AMD Operating System Research Center

Advanced Micro Devices GmbH Einsteinring 24 85609 Dornach
General Managers: Alberto Bozzo
Registration: Dornach, Landkr. Muenchen; Registerger. Muenchen, HRB Nr. 43632

Re: [PATCH 2/2] ARM: IOMMU: Tegra30: Add iommu_ops for SMMU driver

[Hiroshi Doyu]

From: "joerg.roedel-5C7GfCeVMHo@public.gmane.org" <joerg.roedel-5C7GfCeVMHo@public.gmane.org>
Subject: Re: [PATCH 2/2] ARM: IOMMU: Tegra30: Add iommu_ops for SMMU driver
Date: Tue, 24 Jan 2012 12:57:23 +0100
Message-ID: <20120124115723.GC19255-5C7GfCeVMHo@public.gmane.org>

> On Tue, Jan 24, 2012 at 12:36:14PM +0100, Hiroshi Doyu wrote:
> > > A domain is, as you said, a virtual address space for IO devices. But
> > > the important point is, an arbitrary number of devices can be part of a
> > > domain. This also means that the devices can be behind different
> > > hardware SMMUs. In this case your driver needs to program the page-table
> > > pointer into more than one SMMU to give devices behind different SMMUs
> > > the same address space.
> > 
> > Thank you for explaining.
> > 
> > Does the above mean that a buffer can be shared with different devices
> > which belong to different IOMMU devices(virtual address spaces)?
> > 
> > For example, assuming the following:
> > 
> > - We have "struct iommu_domain *domain1".
> > - "domain1" has iommu device "iommu_dev1" and "iommu_dev2".
> > - "iommu_dev1" has "client_dev1" and "client_dev2".
> > - "iommu_dev2" has "client_dev3" and "client_dev4".
> > 
> > "iommu_map(domain1, iova, pa, ...)" will create the following mapping
> > ___at once___:
> > 
> > - (iova)-(pa) mapping in iommu_dev1(iommmu_dev1's virtual address space)
> > - (iova)-(pa) mapping in iommu_dev2(iommmu_dev2's virtual address space)
> > 
> > Is the above correct?
> 
> Yes, this is correct.
> 
> > It seems that the same (iova) is used for different virtual address
> > spaces. What kind of case is this beneficial most in?
> 
> It is actually the _same_ virtual address space which is used by
> iommu_dev1 and iommu_dev2. Think of it like multiple threads of a single
> process. They also share the address space.

I see. The above makes more sense.

> This is a requirement of the iommu-api which is beneficial for
> virtualization and simplifies the usage of the api in general. The user
> does not need to care which devices can be assigned to which domain
> because of underlying hardware constraints. The goal of the iommu-api is
> to hide such contraints.

Now it's clear to me. I'll send the update patch soon.
Thank you for explaining all of this.

> 
> 
> 	Joerg
> 
> -- 
> AMD Operating System Research Center
> 
> Advanced Micro Devices GmbH Einsteinring 24 85609 Dornach
> General Managers: Alberto Bozzo
> Registration: Dornach, Landkr. Muenchen; Registerger. Muenchen, HRB Nr. 43632
> 

Re: [PATCH 2/2] ARM: IOMMU: Tegra30: Add iommu_ops for SMMU driver

[Hiroshi Doyu]

From: Joerg Roedel <joro-zLv9SwRftAIdnm+yROfE0A@public.gmane.org>
Subject: Re: [PATCH 2/2] ARM: IOMMU: Tegra30: Add iommu_ops for SMMU driver
Date: Mon, 23 Jan 2012 16:43:10 +0100
Message-ID: <20120123154310.GC6269-zLv9SwRftAIdnm+yROfE0A@public.gmane.org>

> > +	}
> > +
> > +	spin_unlock_irqrestore(&as->lock, flags);
> > +
> > +	domain->priv = NULL;
> > +	dev_dbg(smmu->dev, "smmu_as@%p\n", as);
> > +}
> > +
> > +static int smmu_iommu_attach_dev(struct iommu_domain *domain,
> > +				 struct device *dev)
> > +{
> > +	struct smmu_as *as = domain->priv;
> > +	struct smmu_device *smmu = as->smmu;
> 
> Hmm, this looks like there is a 1-1 mapping between hardware SMMU
> devices and domains. This is not consistent with IOMMU-API semantics
> where a domain can contain devices behind different SMMUs. Please fix
> that.

Actually I really like the concept of this "domain" now, which hides
the H/W hierarchy from users.

But in Tegra SMMU/GART case, there's a single one IOMMU device in the
system. Keeping a iommu device list in a domain and iterating iommu
device list in each iommu_ops seem to be so nice, but I'm afraid that
this may be a bit too much when one already knows that there's only
one IOMMU device in the system.

If there's no actual problem for 1-1 mapping between IOMMU H/Ws and
domains, I think that it may not so bad to keep the original code(1-1)
for GART and SMMU. What do you think?

Re: [PATCH 2/2] ARM: IOMMU: Tegra30: Add iommu_ops for SMMU driver

[Felipe Balbi]

[-- Attachment #1: Type: text/plain, Size: 1988 bytes --]

Hi,

On Tue, Jan 24, 2012 at 02:41:21PM +0100, Hiroshi Doyu wrote:
> From: Joerg Roedel <joro-zLv9SwRftAIdnm+yROfE0A@public.gmane.org>
> Subject: Re: [PATCH 2/2] ARM: IOMMU: Tegra30: Add iommu_ops for SMMU driver
> Date: Mon, 23 Jan 2012 16:43:10 +0100
> Message-ID: <20120123154310.GC6269-zLv9SwRftAIdnm+yROfE0A@public.gmane.org>
> 
> > > +	}
> > > +
> > > +	spin_unlock_irqrestore(&as->lock, flags);
> > > +
> > > +	domain->priv = NULL;
> > > +	dev_dbg(smmu->dev, "smmu_as@%p\n", as);
> > > +}
> > > +
> > > +static int smmu_iommu_attach_dev(struct iommu_domain *domain,
> > > +				 struct device *dev)
> > > +{
> > > +	struct smmu_as *as = domain->priv;
> > > +	struct smmu_device *smmu = as->smmu;
> > 
> > Hmm, this looks like there is a 1-1 mapping between hardware SMMU
> > devices and domains. This is not consistent with IOMMU-API semantics
> > where a domain can contain devices behind different SMMUs. Please fix
> > that.
> 
> Actually I really like the concept of this "domain" now, which hides
> the H/W hierarchy from users.
> 
> But in Tegra SMMU/GART case, there's a single one IOMMU device in the
> system. Keeping a iommu device list in a domain and iterating iommu
> device list in each iommu_ops seem to be so nice, but I'm afraid that
> this may be a bit too much when one already knows that there's only
> one IOMMU device in the system.
>
> If there's no actual problem for 1-1 mapping between IOMMU H/Ws and
> domains, I think that it may not so bad to keep the original code(1-1)
> for GART and SMMU. What do you think?

I think it boils down to "extensability". If you can truly/fully
guarantee that there will *always* be a single IOMMU on all upcoming
Tegras, then it's really overkill.

But if there's even a remote possibility of the HW being changed and you
end up with more IOMMUs, things start to feel necessary for the sake of
making it easy to extend.

my 2 cents, feel free to ignore ;-)

-- 
balbi

[-- Attachment #2: Digital signature --]
[-- Type: application/pgp-signature, Size: 836 bytes --]

Re: [PATCH 2/2] ARM: IOMMU: Tegra30: Add iommu_ops for SMMU driver

[joro-zLv9SwRftAIdnm+yROfE0A]

On Tue, Jan 24, 2012 at 03:46:01PM +0200, Felipe Balbi wrote:
> On Tue, Jan 24, 2012 at 02:41:21PM +0100, Hiroshi Doyu wrote:
> > Actually I really like the concept of this "domain" now, which hides
> > the H/W hierarchy from users.
> > 
> > But in Tegra SMMU/GART case, there's a single one IOMMU device in the
> > system. Keeping a iommu device list in a domain and iterating iommu
> > device list in each iommu_ops seem to be so nice, but I'm afraid that
> > this may be a bit too much when one already knows that there's only
> > one IOMMU device in the system.
> >
> > If there's no actual problem for 1-1 mapping between IOMMU H/Ws and
> > domains, I think that it may not so bad to keep the original code(1-1)
> > for GART and SMMU. What do you think?
> 
> I think it boils down to "extensability". If you can truly/fully
> guarantee that there will *always* be a single IOMMU on all upcoming
> Tegras, then it's really overkill.
> 
> But if there's even a remote possibility of the HW being changed and you
> end up with more IOMMUs, things start to feel necessary for the sake of
> making it easy to extend.

Right. But I am fine with the logic as-is when there is only one SMMU in
the system. But please also change the IOMMU driver so that it really
only initializes a single SMMU. When boards pop up with more than one
you we notice that assumption in the code again and are reminded to
change it.


	Joerg

Re: [PATCH 2/2] ARM: IOMMU: Tegra30: Add iommu_ops for SMMU driver

[Hiroshi Doyu]

[-- Attachment #1: Type: text/plain, Size: 1931 bytes --]

From: "joro-zLv9SwRftAIdnm+yROfE0A@public.gmane.org" <joro-zLv9SwRftAIdnm+yROfE0A@public.gmane.org>
Subject: Re: [PATCH 2/2] ARM: IOMMU: Tegra30: Add iommu_ops for SMMU driver
Date: Tue, 24 Jan 2012 15:25:21 +0100
Message-ID: <20120124142521.GE6269-zLv9SwRftAIdnm+yROfE0A@public.gmane.org>

> On Tue, Jan 24, 2012 at 03:46:01PM +0200, Felipe Balbi wrote:
> > On Tue, Jan 24, 2012 at 02:41:21PM +0100, Hiroshi Doyu wrote:
> > > Actually I really like the concept of this "domain" now, which hides
> > > the H/W hierarchy from users.
> > > 
> > > But in Tegra SMMU/GART case, there's a single one IOMMU device in the
> > > system. Keeping a iommu device list in a domain and iterating iommu
> > > device list in each iommu_ops seem to be so nice, but I'm afraid that
> > > this may be a bit too much when one already knows that there's only
> > > one IOMMU device in the system.
> > >
> > > If there's no actual problem for 1-1 mapping between IOMMU H/Ws and
> > > domains, I think that it may not so bad to keep the original code(1-1)
> > > for GART and SMMU. What do you think?
> > 
> > I think it boils down to "extensability". If you can truly/fully
> > guarantee that there will *always* be a single IOMMU on all upcoming
> > Tegras, then it's really overkill.
> > 
> > But if there's even a remote possibility of the HW being changed and you
> > end up with more IOMMUs, things start to feel necessary for the sake of
> > making it easy to extend.
> 
> Right. But I am fine with the logic as-is when there is only one SMMU in
> the system. But please also change the IOMMU driver so that it really
> only initializes a single SMMU. When boards pop up with more than one
> you we notice that assumption in the code again and are reminded to
> change it.

Fixed.

I'll revisit 4MB pagesize support and the above multiple iommu device
support in a domain later.

Attached the update patch.

[-- Warning: decoded text below may be mangled, UTF-8 assumed --]
[-- Attachment #2: 0002-ARM-IOMMU-Tegra30-Add-iommu_ops-for-SMMU-driver.patch --]
[-- Type: text/x-patch; name="0002-ARM-IOMMU-Tegra30-Add-iommu_ops-for-SMMU-driver.patch", Size: 33838 bytes --]

From 18448424ce1f8ae554cd3b801d61b6a064e4bcd8 Mon Sep 17 00:00:00 2001
From: Hiroshi DOYU <hdoyu@nvidia.com>
Date: Thu, 17 Nov 2011 07:31:31 +0200
Subject: [PATCH 2/2] ARM: IOMMU: Tegra30: Add iommu_ops for SMMU driver

Tegra 30 IOMMU H/W, SMMU (System Memory Management Unit). This patch
implements struct iommu_ops for SMMU for the upper IOMMU API.

This H/W module supports multiple virtual address spaces(domain x4),
and manages 2 level H/W translation pagetable.

Signed-off-by: Hiroshi DOYU <hdoyu@nvidia.com>
---
 arch/arm/mach-tegra/include/mach/smmu.h |   63 ++
 drivers/iommu/Kconfig                   |   11 +
 drivers/iommu/Makefile                  |    1 +
 drivers/iommu/tegra-smmu.c              | 1034 +++++++++++++++++++++++++++++++
 4 files changed, 1109 insertions(+), 0 deletions(-)
 create mode 100644 arch/arm/mach-tegra/include/mach/smmu.h
 create mode 100644 drivers/iommu/tegra-smmu.c

diff --git a/arch/arm/mach-tegra/include/mach/smmu.h b/arch/arm/mach-tegra/include/mach/smmu.h
new file mode 100644
index 0000000..dad403a
--- /dev/null
+++ b/arch/arm/mach-tegra/include/mach/smmu.h
@@ -0,0 +1,63 @@
+/*
+ * IOMMU API for SMMU in Tegra30
+ *
+ * Copyright (c) 2012, NVIDIA CORPORATION.  All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or modify it
+ * under the terms and conditions of the GNU General Public License,
+ * version 2, as published by the Free Software Foundation.
+ *
+ * This program is distributed in the hope it will be useful, but WITHOUT
+ * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
+ * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
+ * more details.
+ *
+ * You should have received a copy of the GNU General Public License along with
+ * this program; if not, write to the Free Software Foundation, Inc.,
+ * 51 Franklin St - Fifth Floor, Boston, MA 02110-1301 USA.
+ */
+
+#ifndef	MACH_SMMU_H
+#define	MACH_SMMU_H
+
+enum smmu_hwgrp {
+	HWGRP_AFI,
+	HWGRP_AVPC,
+	HWGRP_DC,
+	HWGRP_DCB,
+	HWGRP_EPP,
+	HWGRP_G2,
+	HWGRP_HC,
+	HWGRP_HDA,
+	HWGRP_ISP,
+	HWGRP_MPE,
+	HWGRP_NV,
+	HWGRP_NV2,
+	HWGRP_PPCS,
+	HWGRP_SATA,
+	HWGRP_VDE,
+	HWGRP_VI,
+
+	HWGRP_COUNT,
+
+	HWGRP_END = ~0,
+};
+
+#define HWG_AFI		(1 << HWGRP_AFI)
+#define HWG_AVPC	(1 << HWGRP_AVPC)
+#define HWG_DC		(1 << HWGRP_DC)
+#define HWG_DCB		(1 << HWGRP_DCB)
+#define HWG_EPP		(1 << HWGRP_EPP)
+#define HWG_G2		(1 << HWGRP_G2)
+#define HWG_HC		(1 << HWGRP_HC)
+#define HWG_HDA		(1 << HWGRP_HDA)
+#define HWG_ISP		(1 << HWGRP_ISP)
+#define HWG_MPE		(1 << HWGRP_MPE)
+#define HWG_NV		(1 << HWGRP_NV)
+#define HWG_NV2		(1 << HWGRP_NV2)
+#define HWG_PPCS	(1 << HWGRP_PPCS)
+#define HWG_SATA	(1 << HWGRP_SATA)
+#define HWG_VDE		(1 << HWGRP_VDE)
+#define HWG_VI		(1 << HWGRP_VI)
+
+#endif	/* MACH_SMMU_H */
diff --git a/drivers/iommu/Kconfig b/drivers/iommu/Kconfig
index 4562698..4b3d616 100644
--- a/drivers/iommu/Kconfig
+++ b/drivers/iommu/Kconfig
@@ -153,4 +153,15 @@ config TEGRA_IOMMU_GART
 	  space through the GART (Graphics Address Relocation Table)
 	  hardware included on Tegra SoCs.
 
+config TEGRA_IOMMU_SMMU
+	bool "Tegra SMMU IOMMU Support"
+	depends on ARCH_TEGRA_3x_SOC
+	default y
+	select IOMMU_API
+	help
+	  Enables support for remapping discontiguous physical memory
+	  shared with the operating system into contiguous I/O virtual
+	  space through the SMMU (System Memory Management Unit)
+	  hardware included on Tegra SoCs.
+
 endif # IOMMU_SUPPORT
diff --git a/drivers/iommu/Makefile b/drivers/iommu/Makefile
index 3238a31..7ad7a3b 100644
--- a/drivers/iommu/Makefile
+++ b/drivers/iommu/Makefile
@@ -9,3 +9,4 @@ obj-$(CONFIG_OMAP_IOMMU) += omap-iommu.o
 obj-$(CONFIG_OMAP_IOVMM) += omap-iovmm.o
 obj-$(CONFIG_OMAP_IOMMU_DEBUG) += omap-iommu-debug.o
 obj-$(CONFIG_TEGRA_IOMMU_GART) += tegra-gart.o
+obj-$(CONFIG_TEGRA_IOMMU_SMMU) += tegra-smmu.o
diff --git a/drivers/iommu/tegra-smmu.c b/drivers/iommu/tegra-smmu.c
new file mode 100644
index 0000000..eb93c82
--- /dev/null
+++ b/drivers/iommu/tegra-smmu.c
@@ -0,0 +1,1034 @@
+/*
+ * IOMMU API for SMMU in Tegra30
+ *
+ * Copyright (c) 2011-2012, NVIDIA CORPORATION.  All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or modify it
+ * under the terms and conditions of the GNU General Public License,
+ * version 2, as published by the Free Software Foundation.
+ *
+ * This program is distributed in the hope it will be useful, but WITHOUT
+ * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
+ * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
+ * more details.
+ *
+ * You should have received a copy of the GNU General Public License along with
+ * this program; if not, write to the Free Software Foundation, Inc.,
+ * 51 Franklin St - Fifth Floor, Boston, MA 02110-1301 USA.
+ */
+
+#define pr_fmt(fmt)	"%s(): " fmt, __func__
+
+#include <linux/module.h>
+#include <linux/platform_device.h>
+#include <linux/spinlock.h>
+#include <linux/slab.h>
+#include <linux/vmalloc.h>
+#include <linux/mm.h>
+#include <linux/pagemap.h>
+#include <linux/device.h>
+#include <linux/sched.h>
+#include <linux/iommu.h>
+#include <linux/io.h>
+
+#include <asm/page.h>
+#include <asm/cacheflush.h>
+
+#include <mach/iomap.h>
+#include <mach/smmu.h>
+
+/* bitmap of the page sizes currently supported */
+#define SMMU_IOMMU_PGSIZES	(SZ_4K)
+
+#define SMMU_CONFIG				0x10
+#define SMMU_CONFIG_DISABLE			0
+#define SMMU_CONFIG_ENABLE			1
+
+#define SMMU_TLB_CONFIG				0x14
+#define SMMU_TLB_CONFIG_STATS__MASK		(1 << 31)
+#define SMMU_TLB_CONFIG_STATS__ENABLE		(1 << 31)
+#define SMMU_TLB_CONFIG_HIT_UNDER_MISS__ENABLE	(1 << 29)
+#define SMMU_TLB_CONFIG_ACTIVE_LINES__VALUE	0x10
+#define SMMU_TLB_CONFIG_RESET_VAL		0x20000010
+
+#define SMMU_PTC_CONFIG				0x18
+#define SMMU_PTC_CONFIG_STATS__MASK		(1 << 31)
+#define SMMU_PTC_CONFIG_STATS__ENABLE		(1 << 31)
+#define SMMU_PTC_CONFIG_CACHE__ENABLE		(1 << 29)
+#define SMMU_PTC_CONFIG_INDEX_MAP__PATTERN	0x3f
+#define SMMU_PTC_CONFIG_RESET_VAL		0x2000003f
+
+#define SMMU_PTB_ASID				0x1c
+#define SMMU_PTB_ASID_CURRENT_SHIFT		0
+
+#define SMMU_PTB_DATA				0x20
+#define SMMU_PTB_DATA_RESET_VAL			0
+#define SMMU_PTB_DATA_ASID_NONSECURE_SHIFT	29
+#define SMMU_PTB_DATA_ASID_WRITABLE_SHIFT	30
+#define SMMU_PTB_DATA_ASID_READABLE_SHIFT	31
+
+#define SMMU_TLB_FLUSH				0x30
+#define SMMU_TLB_FLUSH_VA_MATCH_ALL		0
+#define SMMU_TLB_FLUSH_VA_MATCH_SECTION		2
+#define SMMU_TLB_FLUSH_VA_MATCH_GROUP		3
+#define SMMU_TLB_FLUSH_ASID_SHIFT		29
+#define SMMU_TLB_FLUSH_ASID_MATCH_DISABLE	0
+#define SMMU_TLB_FLUSH_ASID_MATCH_ENABLE	1
+#define SMMU_TLB_FLUSH_ASID_MATCH_SHIFT		31
+
+#define SMMU_PTC_FLUSH				0x34
+#define SMMU_PTC_FLUSH_TYPE_ALL			0
+#define SMMU_PTC_FLUSH_TYPE_ADR			1
+#define SMMU_PTC_FLUSH_ADR_SHIFT		4
+
+#define SMMU_ASID_SECURITY			0x38
+
+#define SMMU_STATS_TLB_HIT_COUNT		0x1f0
+#define SMMU_STATS_TLB_MISS_COUNT		0x1f4
+#define SMMU_STATS_PTC_HIT_COUNT		0x1f8
+#define SMMU_STATS_PTC_MISS_COUNT		0x1fc
+
+#define SMMU_TRANSLATION_ENABLE_0		0x228
+#define SMMU_TRANSLATION_ENABLE_1		0x22c
+#define SMMU_TRANSLATION_ENABLE_2		0x230
+
+#define SMMU_AFI_ASID	0x238   /* PCIE */
+#define SMMU_AVPC_ASID	0x23c   /* AVP */
+#define SMMU_DC_ASID	0x240   /* Display controller */
+#define SMMU_DCB_ASID	0x244   /* Display controller B */
+#define SMMU_EPP_ASID	0x248   /* Encoder pre-processor */
+#define SMMU_G2_ASID	0x24c   /* 2D engine */
+#define SMMU_HC_ASID	0x250   /* Host1x */
+#define SMMU_HDA_ASID	0x254   /* High-def audio */
+#define SMMU_ISP_ASID	0x258   /* Image signal processor */
+#define SMMU_MPE_ASID	0x264   /* MPEG encoder */
+#define SMMU_NV_ASID	0x268   /* (3D) */
+#define SMMU_NV2_ASID	0x26c   /* (3D) */
+#define SMMU_PPCS_ASID	0x270   /* AHB */
+#define SMMU_SATA_ASID	0x278   /* SATA */
+#define SMMU_VDE_ASID	0x27c   /* Video decoder */
+#define SMMU_VI_ASID	0x280   /* Video input */
+
+#define SMMU_PDE_NEXT_SHIFT		28
+
+/* AHB Arbiter Registers */
+#define AHB_XBAR_CTRL				0xe0
+#define AHB_XBAR_CTRL_SMMU_INIT_DONE_DONE	1
+#define AHB_XBAR_CTRL_SMMU_INIT_DONE_SHIFT	17
+
+#define SMMU_NUM_ASIDS				4
+#define SMMU_TLB_FLUSH_VA_SECTION__MASK		0xffc00000
+#define SMMU_TLB_FLUSH_VA_SECTION__SHIFT	12 /* right shift */
+#define SMMU_TLB_FLUSH_VA_GROUP__MASK		0xffffc000
+#define SMMU_TLB_FLUSH_VA_GROUP__SHIFT		12 /* right shift */
+#define SMMU_TLB_FLUSH_VA(iova, which)	\
+	((((iova) & SMMU_TLB_FLUSH_VA_##which##__MASK) >> \
+		SMMU_TLB_FLUSH_VA_##which##__SHIFT) |	\
+	SMMU_TLB_FLUSH_VA_MATCH_##which)
+#define SMMU_PTB_ASID_CUR(n)	\
+		((n) << SMMU_PTB_ASID_CURRENT_SHIFT)
+#define SMMU_TLB_FLUSH_ASID_MATCH_disable		\
+		(SMMU_TLB_FLUSH_ASID_MATCH_DISABLE <<	\
+			SMMU_TLB_FLUSH_ASID_MATCH_SHIFT)
+#define SMMU_TLB_FLUSH_ASID_MATCH__ENABLE		\
+		(SMMU_TLB_FLUSH_ASID_MATCH_ENABLE <<	\
+			SMMU_TLB_FLUSH_ASID_MATCH_SHIFT)
+
+#define SMMU_PAGE_SHIFT 12
+#define SMMU_PAGE_SIZE	(1 << SMMU_PAGE_SHIFT)
+
+#define SMMU_PDIR_COUNT	1024
+#define SMMU_PDIR_SIZE	(sizeof(unsigned long) * SMMU_PDIR_COUNT)
+#define SMMU_PTBL_COUNT	1024
+#define SMMU_PTBL_SIZE	(sizeof(unsigned long) * SMMU_PTBL_COUNT)
+#define SMMU_PDIR_SHIFT	12
+#define SMMU_PDE_SHIFT	12
+#define SMMU_PTE_SHIFT	12
+#define SMMU_PFN_MASK	0x000fffff
+
+#define SMMU_ADDR_TO_PFN(addr)	((addr) >> 12)
+#define SMMU_ADDR_TO_PDN(addr)	((addr) >> 22)
+#define SMMU_PDN_TO_ADDR(addr)	((pdn) << 22)
+
+#define _READABLE	(1 << SMMU_PTB_DATA_ASID_READABLE_SHIFT)
+#define _WRITABLE	(1 << SMMU_PTB_DATA_ASID_WRITABLE_SHIFT)
+#define _NONSECURE	(1 << SMMU_PTB_DATA_ASID_NONSECURE_SHIFT)
+#define _PDE_NEXT	(1 << SMMU_PDE_NEXT_SHIFT)
+#define _MASK_ATTR	(_READABLE | _WRITABLE | _NONSECURE)
+
+#define _PDIR_ATTR	(_READABLE | _WRITABLE | _NONSECURE)
+
+#define _PDE_ATTR	(_READABLE | _WRITABLE | _NONSECURE)
+#define _PDE_ATTR_N	(_PDE_ATTR | _PDE_NEXT)
+#define _PDE_VACANT(pdn)	(((pdn) << 10) | _PDE_ATTR)
+
+#define _PTE_ATTR	(_READABLE | _WRITABLE | _NONSECURE)
+#define _PTE_VACANT(addr)	(((addr) >> SMMU_PAGE_SHIFT) | _PTE_ATTR)
+
+#define SMMU_MK_PDIR(page, attr)	\
+		((page_to_phys(page) >> SMMU_PDIR_SHIFT) | (attr))
+#define SMMU_MK_PDE(page, attr)		\
+		(unsigned long)((page_to_phys(page) >> SMMU_PDE_SHIFT) | (attr))
+#define SMMU_EX_PTBL_PAGE(pde)		\
+		pfn_to_page((unsigned long)(pde) & SMMU_PFN_MASK)
+#define SMMU_PFN_TO_PTE(pfn, attr)	(unsigned long)((pfn) | (attr))
+
+#define SMMU_ASID_ENABLE(asid)	((asid) | (1 << 31))
+#define SMMU_ASID_DISABLE	0
+#define SMMU_ASID_ASID(n)	((n) & ~SMMU_ASID_ENABLE(0))
+
+#define smmu_client_enable_hwgrp(c, m)	smmu_client_set_hwgrp(c, m, 1)
+#define smmu_client_disable_hwgrp(c)	smmu_client_set_hwgrp(c, 0, 0)
+#define __smmu_client_enable_hwgrp(c, m) __smmu_client_set_hwgrp(c, m, 1)
+#define __smmu_client_disable_hwgrp(c)	__smmu_client_set_hwgrp(c, 0, 0)
+
+#define HWGRP_INIT(client) [HWGRP_##client] = SMMU_##client##_ASID
+
+static const u32 smmu_hwgrp_asid_reg[] = {
+	HWGRP_INIT(AFI),
+	HWGRP_INIT(AVPC),
+	HWGRP_INIT(DC),
+	HWGRP_INIT(DCB),
+	HWGRP_INIT(EPP),
+	HWGRP_INIT(G2),
+	HWGRP_INIT(HC),
+	HWGRP_INIT(HDA),
+	HWGRP_INIT(ISP),
+	HWGRP_INIT(MPE),
+	HWGRP_INIT(NV),
+	HWGRP_INIT(NV2),
+	HWGRP_INIT(PPCS),
+	HWGRP_INIT(SATA),
+	HWGRP_INIT(VDE),
+	HWGRP_INIT(VI),
+};
+#define HWGRP_ASID_REG(x) (smmu_hwgrp_asid_reg[x])
+
+/*
+ * Per client for address space
+ */
+struct smmu_client {
+	struct device		*dev;
+	struct list_head	list;
+	struct smmu_as		*as;
+	u32			hwgrp;
+};
+
+/*
+ * Per address space
+ */
+struct smmu_as {
+	struct smmu_device	*smmu;	/* back pointer to container */
+	unsigned int		asid;
+	spinlock_t		lock;	/* for pagetable */
+	struct page		*pdir_page;
+	unsigned long		pdir_attr;
+	unsigned long		pde_attr;
+	unsigned long		pte_attr;
+	unsigned int		*pte_count;
+
+	struct list_head	client;
+	spinlock_t		client_lock; /* for client list */
+};
+
+/*
+ * Per SMMU device - IOMMU device
+ */
+struct smmu_device {
+	void __iomem	*regs, *regs_ahbarb;
+	unsigned long	iovmm_base;	/* remappable base address */
+	unsigned long	page_count;	/* total remappable size */
+	spinlock_t	lock;
+	char		*name;
+	struct device	*dev;
+	int		num_as;
+	struct smmu_as	*as;		/* Run-time allocated array */
+	struct page *avp_vector_page;	/* dummy page shared by all AS's */
+
+	/*
+	 * Register image savers for suspend/resume
+	 */
+	unsigned long translation_enable_0;
+	unsigned long translation_enable_1;
+	unsigned long translation_enable_2;
+	unsigned long asid_security;
+};
+
+static struct smmu_device *smmu_handle; /* unique for a system */
+
+/*
+ *	SMMU/AHB register accessors
+ */
+static inline u32 smmu_read(struct smmu_device *smmu, size_t offs)
+{
+	return readl(smmu->regs + offs);
+}
+static inline void smmu_write(struct smmu_device *smmu, u32 val, size_t offs)
+{
+	writel(val, smmu->regs + offs);
+}
+
+static inline u32 ahb_read(struct smmu_device *smmu, size_t offs)
+{
+	return readl(smmu->regs_ahbarb + offs);
+}
+static inline void ahb_write(struct smmu_device *smmu, u32 val, size_t offs)
+{
+	writel(val, smmu->regs_ahbarb + offs);
+}
+
+#define VA_PAGE_TO_PA(va, page)	\
+	(page_to_phys(page) + ((unsigned long)(va) & ~PAGE_MASK))
+
+#define FLUSH_CPU_DCACHE(va, page, size)	\
+	do {	\
+		unsigned long _pa_ = VA_PAGE_TO_PA(va, page);		\
+		__cpuc_flush_dcache_area((void *)(va), (size_t)(size));	\
+		outer_flush_range(_pa_, _pa_+(size_t)(size));		\
+	} while (0)
+
+/*
+ * Any interaction between any block on PPSB and a block on APB or AHB
+ * must have these read-back barriers to ensure the APB/AHB bus
+ * transaction is complete before initiating activity on the PPSB
+ * block.
+ */
+#define FLUSH_SMMU_REGS(smmu)	smmu_read(smmu, SMMU_CONFIG)
+
+#define smmu_client_hwgrp(c) (u32)((c)->dev->platform_data)
+
+static int __smmu_client_set_hwgrp(struct smmu_client *c,
+				   unsigned long map, int on)
+{
+	int i;
+	struct smmu_as *as = c->as;
+	u32 val, offs, mask = SMMU_ASID_ENABLE(as->asid);
+	struct smmu_device *smmu = as->smmu;
+
+	WARN_ON(!on && map);
+	if (on && !map)
+		return -EINVAL;
+	if (!on)
+		map = smmu_client_hwgrp(c);
+
+	for_each_set_bit(i, &map, HWGRP_COUNT) {
+		offs = HWGRP_ASID_REG(i);
+		val = smmu_read(smmu, offs);
+		if (on) {
+			if (WARN_ON(val & mask))
+				goto err_hw_busy;
+			val |= mask;
+		} else {
+			WARN_ON((val & mask) == mask);
+			val &= ~mask;
+		}
+		smmu_write(smmu, val, offs);
+	}
+	FLUSH_SMMU_REGS(smmu);
+	c->hwgrp = map;
+	return 0;
+
+err_hw_busy:
+	for_each_set_bit(i, &map, HWGRP_COUNT) {
+		offs = HWGRP_ASID_REG(i);
+		val = smmu_read(smmu, offs);
+		val &= ~mask;
+		smmu_write(smmu, val, offs);
+	}
+	return -EBUSY;
+}
+
+static int smmu_client_set_hwgrp(struct smmu_client *c, u32 map, int on)
+{
+	u32 val;
+	unsigned long flags;
+	struct smmu_as *as = c->as;
+	struct smmu_device *smmu = as->smmu;
+
+	spin_lock_irqsave(&smmu->lock, flags);
+	val = __smmu_client_set_hwgrp(c, map, on);
+	spin_unlock_irqrestore(&smmu->lock, flags);
+	return val;
+}
+
+/*
+ * Flush all TLB entries and all PTC entries
+ * Caller must lock smmu
+ */
+static void smmu_flush_regs(struct smmu_device *smmu, int enable)
+{
+	u32 val;
+
+	smmu_write(smmu, SMMU_PTC_FLUSH_TYPE_ALL, SMMU_PTC_FLUSH);
+	FLUSH_SMMU_REGS(smmu);
+	val = SMMU_TLB_FLUSH_VA_MATCH_ALL |
+		SMMU_TLB_FLUSH_ASID_MATCH_disable;
+	smmu_write(smmu, val, SMMU_TLB_FLUSH);
+
+	if (enable)
+		smmu_write(smmu, SMMU_CONFIG_ENABLE, SMMU_CONFIG);
+	FLUSH_SMMU_REGS(smmu);
+}
+
+static void smmu_setup_regs(struct smmu_device *smmu)
+{
+	int i;
+	u32 val;
+
+	for (i = 0; i < smmu->num_as; i++) {
+		struct smmu_as *as = &smmu->as[i];
+		struct smmu_client *c;
+
+		smmu_write(smmu, SMMU_PTB_ASID_CUR(as->asid), SMMU_PTB_ASID);
+		val = as->pdir_page ?
+			SMMU_MK_PDIR(as->pdir_page, as->pdir_attr) :
+			SMMU_PTB_DATA_RESET_VAL;
+		smmu_write(smmu, val, SMMU_PTB_DATA);
+
+		list_for_each_entry(c, &as->client, list)
+			__smmu_client_set_hwgrp(c, c->hwgrp, 1);
+	}
+
+	smmu_write(smmu, smmu->translation_enable_0, SMMU_TRANSLATION_ENABLE_0);
+	smmu_write(smmu, smmu->translation_enable_1, SMMU_TRANSLATION_ENABLE_1);
+	smmu_write(smmu, smmu->translation_enable_2, SMMU_TRANSLATION_ENABLE_2);
+	smmu_write(smmu, smmu->asid_security, SMMU_ASID_SECURITY);
+	smmu_write(smmu, SMMU_TLB_CONFIG_RESET_VAL, SMMU_TLB_CONFIG);
+	smmu_write(smmu, SMMU_PTC_CONFIG_RESET_VAL, SMMU_PTC_CONFIG);
+
+	smmu_flush_regs(smmu, 1);
+
+	val = ahb_read(smmu, AHB_XBAR_CTRL);
+	val |= AHB_XBAR_CTRL_SMMU_INIT_DONE_DONE <<
+		AHB_XBAR_CTRL_SMMU_INIT_DONE_SHIFT;
+	ahb_write(smmu, val, AHB_XBAR_CTRL);
+}
+
+static void flush_ptc_and_tlb(struct smmu_device *smmu,
+		      struct smmu_as *as, dma_addr_t iova,
+		      unsigned long *pte, struct page *page, int is_pde)
+{
+	u32 val;
+	unsigned long tlb_flush_va = is_pde
+		?  SMMU_TLB_FLUSH_VA(iova, SECTION)
+		:  SMMU_TLB_FLUSH_VA(iova, GROUP);
+
+	val = SMMU_PTC_FLUSH_TYPE_ADR | VA_PAGE_TO_PA(pte, page);
+	smmu_write(smmu, val, SMMU_PTC_FLUSH);
+	FLUSH_SMMU_REGS(smmu);
+	val = tlb_flush_va |
+		SMMU_TLB_FLUSH_ASID_MATCH__ENABLE |
+		(as->asid << SMMU_TLB_FLUSH_ASID_SHIFT);
+	smmu_write(smmu, val, SMMU_TLB_FLUSH);
+	FLUSH_SMMU_REGS(smmu);
+}
+
+static void free_ptbl(struct smmu_as *as, dma_addr_t iova)
+{
+	unsigned long pdn = SMMU_ADDR_TO_PDN(iova);
+	unsigned long *pdir = (unsigned long *)page_address(as->pdir_page);
+
+	if (pdir[pdn] != _PDE_VACANT(pdn)) {
+		dev_dbg(as->smmu->dev, "pdn: %lx\n", pdn);
+
+		ClearPageReserved(SMMU_EX_PTBL_PAGE(pdir[pdn]));
+		__free_page(SMMU_EX_PTBL_PAGE(pdir[pdn]));
+		pdir[pdn] = _PDE_VACANT(pdn);
+		FLUSH_CPU_DCACHE(&pdir[pdn], as->pdir_page, sizeof pdir[pdn]);
+		flush_ptc_and_tlb(as->smmu, as, iova, &pdir[pdn],
+				  as->pdir_page, 1);
+	}
+}
+
+static void free_pdir(struct smmu_as *as)
+{
+	unsigned addr;
+	int count;
+	struct device *dev = as->smmu->dev;
+
+	if (!as->pdir_page)
+		return;
+
+	addr = as->smmu->iovmm_base;
+	count = as->smmu->page_count;
+	while (count-- > 0) {
+		free_ptbl(as, addr);
+		addr += SMMU_PAGE_SIZE * SMMU_PTBL_COUNT;
+	}
+	ClearPageReserved(as->pdir_page);
+	__free_page(as->pdir_page);
+	as->pdir_page = NULL;
+	devm_kfree(dev, as->pte_count);
+	as->pte_count = NULL;
+}
+
+/*
+ * Maps PTBL for given iova and returns the PTE address
+ * Caller must unmap the mapped PTBL returned in *ptbl_page_p
+ */
+static unsigned long *locate_pte(struct smmu_as *as,
+				 dma_addr_t iova, bool allocate,
+				 struct page **ptbl_page_p,
+				 unsigned int **count)
+{
+	unsigned long ptn = SMMU_ADDR_TO_PFN(iova);
+	unsigned long pdn = SMMU_ADDR_TO_PDN(iova);
+	unsigned long *pdir = page_address(as->pdir_page);
+	unsigned long *ptbl;
+
+	if (pdir[pdn] != _PDE_VACANT(pdn)) {
+		/* Mapped entry table already exists */
+		*ptbl_page_p = SMMU_EX_PTBL_PAGE(pdir[pdn]);
+		ptbl = page_address(*ptbl_page_p);
+	} else if (!allocate) {
+		return NULL;
+	} else {
+		int pn;
+		unsigned long addr = SMMU_PDN_TO_ADDR(pdn);
+
+		/* Vacant - allocate a new page table */
+		dev_dbg(as->smmu->dev, "New PTBL pdn: %lx\n", pdn);
+
+		*ptbl_page_p = alloc_page(GFP_ATOMIC);
+		if (!*ptbl_page_p) {
+			dev_err(as->smmu->dev,
+				"failed to allocate smmu_device page table\n");
+			return NULL;
+		}
+		SetPageReserved(*ptbl_page_p);
+		ptbl = (unsigned long *)page_address(*ptbl_page_p);
+		for (pn = 0; pn < SMMU_PTBL_COUNT;
+		     pn++, addr += SMMU_PAGE_SIZE) {
+			ptbl[pn] = _PTE_VACANT(addr);
+		}
+		FLUSH_CPU_DCACHE(ptbl, *ptbl_page_p, SMMU_PTBL_SIZE);
+		pdir[pdn] = SMMU_MK_PDE(*ptbl_page_p,
+					as->pde_attr | _PDE_NEXT);
+		FLUSH_CPU_DCACHE(&pdir[pdn], as->pdir_page, sizeof pdir[pdn]);
+		flush_ptc_and_tlb(as->smmu, as, iova, &pdir[pdn],
+				  as->pdir_page, 1);
+	}
+	*count = &as->pte_count[pdn];
+
+	return &ptbl[ptn % SMMU_PTBL_COUNT];
+}
+
+#ifdef CONFIG_SMMU_SIG_DEBUG
+static void put_signature(struct smmu_as *as,
+			  dma_addr_t iova, unsigned long pfn)
+{
+	struct page *page;
+	unsigned long *vaddr;
+
+	page = pfn_to_page(pfn);
+	vaddr = page_address(page);
+	if (!vaddr)
+		return;
+
+	vaddr[0] = iova;
+	vaddr[1] = pfn << PAGE_SHIFT;
+	FLUSH_CPU_DCACHE(vaddr, page, sizeof(vaddr[0]) * 2);
+}
+#else
+static inline void put_signature(struct smmu_as *as,
+				 unsigned long addr, unsigned long pfn)
+{
+}
+#endif
+
+/*
+ * Caller must lock/unlock as
+ */
+static int alloc_pdir(struct smmu_as *as)
+{
+	unsigned long *pdir;
+	int pdn;
+	u32 val;
+	struct smmu_device *smmu = as->smmu;
+
+	if (as->pdir_page)
+		return 0;
+
+	as->pte_count = devm_kzalloc(smmu->dev,
+		     sizeof(as->pte_count[0]) * SMMU_PDIR_COUNT, GFP_KERNEL);
+	if (!as->pte_count) {
+		dev_err(smmu->dev,
+			"failed to allocate smmu_device PTE cunters\n");
+		return -ENOMEM;
+	}
+	as->pdir_page = alloc_page(GFP_KERNEL | __GFP_DMA);
+	if (!as->pdir_page) {
+		dev_err(smmu->dev,
+			"failed to allocate smmu_device page directory\n");
+		devm_kfree(smmu->dev, as->pte_count);
+		as->pte_count = NULL;
+		return -ENOMEM;
+	}
+	SetPageReserved(as->pdir_page);
+	pdir = page_address(as->pdir_page);
+
+	for (pdn = 0; pdn < SMMU_PDIR_COUNT; pdn++)
+		pdir[pdn] = _PDE_VACANT(pdn);
+	FLUSH_CPU_DCACHE(pdir, as->pdir_page, SMMU_PDIR_SIZE);
+	val = SMMU_PTC_FLUSH_TYPE_ADR | VA_PAGE_TO_PA(pdir, as->pdir_page);
+	smmu_write(smmu, val, SMMU_PTC_FLUSH);
+	FLUSH_SMMU_REGS(as->smmu);
+	val = SMMU_TLB_FLUSH_VA_MATCH_ALL |
+		SMMU_TLB_FLUSH_ASID_MATCH__ENABLE |
+		(as->asid << SMMU_TLB_FLUSH_ASID_SHIFT);
+	smmu_write(smmu, val, SMMU_TLB_FLUSH);
+	FLUSH_SMMU_REGS(as->smmu);
+
+	return 0;
+}
+
+static void __smmu_iommu_unmap(struct smmu_as *as, dma_addr_t iova)
+{
+	unsigned long *pte;
+	struct page *page;
+	unsigned int *count;
+
+	pte = locate_pte(as, iova, false, &page, &count);
+	if (WARN_ON(!pte))
+		return;
+
+	if (WARN_ON(*pte == _PTE_VACANT(iova)))
+		return;
+
+	*pte = _PTE_VACANT(iova);
+	FLUSH_CPU_DCACHE(pte, page, sizeof(*pte));
+	flush_ptc_and_tlb(as->smmu, as, iova, pte, page, 0);
+	if (!--(*count)) {
+		free_ptbl(as, iova);
+		smmu_flush_regs(as->smmu, 0);
+	}
+}
+
+static void __smmu_iommu_map_pfn(struct smmu_as *as, dma_addr_t iova,
+				 unsigned long pfn)
+{
+	struct smmu_device *smmu = as->smmu;
+	unsigned long *pte;
+	unsigned int *count;
+	struct page *page;
+
+	pte = locate_pte(as, iova, true, &page, &count);
+	if (WARN_ON(!pte))
+		return;
+
+	if (*pte == _PTE_VACANT(iova))
+		(*count)++;
+	*pte = SMMU_PFN_TO_PTE(pfn, as->pte_attr);
+	if (unlikely((*pte == _PTE_VACANT(iova))))
+		(*count)--;
+	FLUSH_CPU_DCACHE(pte, page, sizeof(*pte));
+	flush_ptc_and_tlb(smmu, as, iova, pte, page, 0);
+	put_signature(as, iova, pfn);
+}
+
+static int smmu_iommu_map(struct iommu_domain *domain, unsigned long iova,
+			  phys_addr_t pa, size_t bytes, int prot)
+{
+	struct smmu_as *as = domain->priv;
+	unsigned long pfn = __phys_to_pfn(pa);
+	unsigned long flags;
+
+	dev_dbg(as->smmu->dev, "[%d] %08lx:%08x\n", as->asid, iova, pa);
+
+	if (!pfn_valid(pfn))
+		return -ENOMEM;
+
+	spin_lock_irqsave(&as->lock, flags);
+	__smmu_iommu_map_pfn(as, iova, pfn);
+	spin_unlock_irqrestore(&as->lock, flags);
+	return 0;
+}
+
+static size_t smmu_iommu_unmap(struct iommu_domain *domain, unsigned long iova,
+			       size_t bytes)
+{
+	struct smmu_as *as = domain->priv;
+	unsigned long flags;
+
+	dev_dbg(as->smmu->dev, "[%d] %08lx\n", as->asid, iova);
+
+	spin_lock_irqsave(&as->lock, flags);
+	__smmu_iommu_unmap(as, iova);
+	spin_unlock_irqrestore(&as->lock, flags);
+	return SMMU_PAGE_SIZE;
+}
+
+static phys_addr_t smmu_iommu_iova_to_phys(struct iommu_domain *domain,
+					   unsigned long iova)
+{
+	struct smmu_as *as = domain->priv;
+	unsigned long *pte;
+	unsigned int *count;
+	struct page *page;
+	unsigned long pfn;
+	unsigned long flags;
+
+	spin_lock_irqsave(&as->lock, flags);
+
+	pte = locate_pte(as, iova, true, &page, &count);
+	pfn = *pte & SMMU_PFN_MASK;
+	WARN_ON(!pfn_valid(pfn));
+	dev_dbg(as->smmu->dev,
+		"iova:%08lx pfn:%08lx asid:%d\n", iova, pfn, as->asid);
+
+	spin_unlock_irqrestore(&as->lock, flags);
+	return PFN_PHYS(pfn);
+}
+
+static int smmu_iommu_domain_has_cap(struct iommu_domain *domain,
+				     unsigned long cap)
+{
+	return 0;
+}
+
+static int smmu_iommu_attach_dev(struct iommu_domain *domain,
+				 struct device *dev)
+{
+	struct smmu_as *as = domain->priv;
+	struct smmu_device *smmu = as->smmu;
+	struct smmu_client *client, *c;
+	u32 map;
+	int err;
+
+	client = devm_kzalloc(smmu->dev, sizeof(*c), GFP_KERNEL);
+	if (!client)
+		return -ENOMEM;
+	client->dev = dev;
+	client->as = as;
+	map = (unsigned long)dev->platform_data;
+	if (!map)
+		return -EINVAL;
+
+	err = smmu_client_enable_hwgrp(client, map);
+	if (err)
+		goto err_hwgrp;
+
+	spin_lock(&as->client_lock);
+	list_for_each_entry(c, &as->client, list) {
+		if (c->dev == dev) {
+			dev_err(smmu->dev,
+				"%s is already attached\n", dev_name(c->dev));
+			err = -EINVAL;
+			goto err_client;
+		}
+	}
+	list_add(&client->list, &as->client);
+	spin_unlock(&as->client_lock);
+
+	/*
+	 * Reserve "page zero" for AVP vectors using a common dummy
+	 * page.
+	 */
+	if (map & HWG_AVPC) {
+		struct page *page;
+
+		page = as->smmu->avp_vector_page;
+		__smmu_iommu_map_pfn(as, 0, page_to_pfn(page));
+
+		pr_info("Reserve \"page zero\" for AVP vectors using a common dummy\n");
+	}
+
+	dev_dbg(smmu->dev, "%s is attached\n", dev_name(c->dev));
+	return 0;
+
+err_client:
+	smmu_client_disable_hwgrp(client);
+	spin_unlock(&as->client_lock);
+err_hwgrp:
+	devm_kfree(smmu->dev, client);
+	return err;
+}
+
+static void smmu_iommu_detach_dev(struct iommu_domain *domain,
+				  struct device *dev)
+{
+	struct smmu_as *as = domain->priv;
+	struct smmu_device *smmu = as->smmu;
+	struct smmu_client *c;
+
+	spin_lock(&as->client_lock);
+
+	list_for_each_entry(c, &as->client, list) {
+		if (c->dev == dev) {
+			smmu_client_disable_hwgrp(c);
+			list_del(&c->list);
+			devm_kfree(smmu->dev, c);
+			c->as = NULL;
+			dev_dbg(smmu->dev,
+				"%s is detached\n", dev_name(c->dev));
+			goto out;
+		}
+	}
+	dev_err(smmu->dev, "Couldn't find %s\n", dev_name(c->dev));
+out:
+	spin_unlock(&as->client_lock);
+}
+
+static int smmu_iommu_domain_init(struct iommu_domain *domain)
+{
+	int i;
+	unsigned long flags;
+	struct smmu_as *as;
+	struct smmu_device *smmu = smmu_handle;
+
+	/* Look for a free AS with lock held */
+	for  (i = 0; i < smmu->num_as; i++) {
+		struct smmu_as *tmp = &smmu->as[i];
+
+		spin_lock_irqsave(&tmp->lock, flags);
+		if (!tmp->pdir_page) {
+			as = tmp;
+			goto found;
+		}
+		spin_unlock_irqrestore(&tmp->lock, flags);
+	}
+	dev_err(smmu->dev, "no free AS\n");
+	return -ENODEV;
+
+found:
+	if (alloc_pdir(as) < 0)
+		goto err_alloc_pdir;
+
+	spin_lock(&smmu->lock);
+
+	/* Update PDIR register */
+	smmu_write(smmu, SMMU_PTB_ASID_CUR(as->asid), SMMU_PTB_ASID);
+	smmu_write(smmu,
+		   SMMU_MK_PDIR(as->pdir_page, as->pdir_attr), SMMU_PTB_DATA);
+	FLUSH_SMMU_REGS(smmu);
+
+	spin_unlock(&smmu->lock);
+
+	spin_unlock_irqrestore(&as->lock, flags);
+	domain->priv = as;
+
+	dev_dbg(smmu->dev, "smmu_as@%p\n", as);
+	return 0;
+
+err_alloc_pdir:
+	spin_unlock_irqrestore(&as->lock, flags);
+	return -ENODEV;
+}
+
+static void smmu_iommu_domain_destroy(struct iommu_domain *domain)
+{
+	struct smmu_as *as = domain->priv;
+	struct smmu_device *smmu = as->smmu;
+	unsigned long flags;
+
+	spin_lock_irqsave(&as->lock, flags);
+
+	if (as->pdir_page) {
+		spin_lock(&smmu->lock);
+		smmu_write(smmu, SMMU_PTB_ASID_CUR(as->asid), SMMU_PTB_ASID);
+		smmu_write(smmu, SMMU_PTB_DATA_RESET_VAL, SMMU_PTB_DATA);
+		FLUSH_SMMU_REGS(smmu);
+		spin_unlock(&smmu->lock);
+
+		free_pdir(as);
+	}
+
+	if (!list_empty(&as->client)) {
+		struct smmu_client *c;
+
+		list_for_each_entry(c, &as->client, list)
+			smmu_iommu_detach_dev(domain, c->dev);
+	}
+
+	spin_unlock_irqrestore(&as->lock, flags);
+
+	domain->priv = NULL;
+	dev_dbg(smmu->dev, "smmu_as@%p\n", as);
+}
+
+static struct iommu_ops smmu_iommu_ops = {
+	.domain_init	= smmu_iommu_domain_init,
+	.domain_destroy	= smmu_iommu_domain_destroy,
+	.attach_dev	= smmu_iommu_attach_dev,
+	.detach_dev	= smmu_iommu_detach_dev,
+	.map		= smmu_iommu_map,
+	.unmap		= smmu_iommu_unmap,
+	.iova_to_phys	= smmu_iommu_iova_to_phys,
+	.domain_has_cap	= smmu_iommu_domain_has_cap,
+	.pgsize_bitmap	= SMMU_IOMMU_PGSIZES,
+};
+
+static int tegra_smmu_suspend(struct device *dev)
+{
+	struct smmu_device *smmu = dev_get_drvdata(dev);
+
+	smmu->translation_enable_0 = smmu_read(smmu, SMMU_TRANSLATION_ENABLE_0);
+	smmu->translation_enable_1 = smmu_read(smmu, SMMU_TRANSLATION_ENABLE_1);
+	smmu->translation_enable_2 = smmu_read(smmu, SMMU_TRANSLATION_ENABLE_2);
+	smmu->asid_security = smmu_read(smmu, SMMU_ASID_SECURITY);
+	return 0;
+}
+
+static int tegra_smmu_resume(struct device *dev)
+{
+	struct smmu_device *smmu = dev_get_drvdata(dev);
+	unsigned long flags;
+
+	spin_lock_irqsave(&smmu->lock, flags);
+	smmu_setup_regs(smmu);
+	spin_unlock_irqrestore(&smmu->lock, flags);
+	return 0;
+}
+
+static int tegra_smmu_probe(struct platform_device *pdev)
+{
+	struct smmu_device *smmu;
+	struct resource *regs, *regs2, *window;
+	struct device *dev = &pdev->dev;
+	int i, err = 0;
+
+	if (smmu_handle)
+		return -EIO;
+
+	BUILD_BUG_ON(PAGE_SHIFT != SMMU_PAGE_SHIFT);
+
+	regs = platform_get_resource(pdev, IORESOURCE_MEM, 0);
+	regs2 = platform_get_resource(pdev, IORESOURCE_MEM, 1);
+	window = platform_get_resource(pdev, IORESOURCE_MEM, 2);
+	if (!regs || !regs2 || !window) {
+		dev_err(dev, "No SMMU resources\n");
+		return -ENODEV;
+	}
+
+	smmu = devm_kzalloc(dev, sizeof(*smmu), GFP_KERNEL);
+	if (!smmu) {
+		dev_err(dev, "failed to allocate smmu_device\n");
+		return -ENOMEM;
+	}
+
+	smmu->dev = dev;
+	smmu->num_as = SMMU_NUM_ASIDS;
+	smmu->iovmm_base = (unsigned long)window->start;
+	smmu->page_count = resource_size(window) >> SMMU_PAGE_SHIFT;
+	smmu->regs = devm_ioremap(dev, regs->start, resource_size(regs));
+	smmu->regs_ahbarb = devm_ioremap(dev, regs2->start,
+					 resource_size(regs2));
+	if (!smmu->regs || !smmu->regs_ahbarb) {
+		dev_err(dev, "failed to remap SMMU registers\n");
+		err = -ENXIO;
+		goto fail;
+	}
+
+	smmu->translation_enable_0 = ~0;
+	smmu->translation_enable_1 = ~0;
+	smmu->translation_enable_2 = ~0;
+	smmu->asid_security = 0;
+
+	smmu->as = devm_kzalloc(dev,
+			sizeof(smmu->as[0]) * smmu->num_as, GFP_KERNEL);
+	if (!smmu->as) {
+		dev_err(dev, "failed to allocate smmu_as\n");
+		err = -ENOMEM;
+		goto fail;
+	}
+
+	for (i = 0; i < smmu->num_as; i++) {
+		struct smmu_as *as = &smmu->as[i];
+
+		as->smmu = smmu;
+		as->asid = i;
+		as->pdir_attr = _PDIR_ATTR;
+		as->pde_attr = _PDE_ATTR;
+		as->pte_attr = _PTE_ATTR;
+
+		spin_lock_init(&as->lock);
+		INIT_LIST_HEAD(&as->client);
+	}
+	spin_lock_init(&smmu->lock);
+	smmu_setup_regs(smmu);
+	platform_set_drvdata(pdev, smmu);
+
+	smmu->avp_vector_page = alloc_page(GFP_KERNEL);
+	if (!smmu->avp_vector_page)
+		goto fail;
+
+	smmu_handle = smmu;
+	return 0;
+
+fail:
+	if (smmu->avp_vector_page)
+		__free_page(smmu->avp_vector_page);
+	if (smmu->regs)
+		devm_iounmap(dev, smmu->regs);
+	if (smmu->regs_ahbarb)
+		devm_iounmap(dev, smmu->regs_ahbarb);
+	if (smmu && smmu->as) {
+		for (i = 0; i < smmu->num_as; i++) {
+			if (smmu->as[i].pdir_page) {
+				ClearPageReserved(smmu->as[i].pdir_page);
+				__free_page(smmu->as[i].pdir_page);
+			}
+		}
+		devm_kfree(dev, smmu->as);
+	}
+	devm_kfree(dev, smmu);
+	return err;
+}
+
+static int tegra_smmu_remove(struct platform_device *pdev)
+{
+	struct smmu_device *smmu = platform_get_drvdata(pdev);
+	struct device *dev = smmu->dev;
+
+	smmu_write(smmu, SMMU_CONFIG_DISABLE, SMMU_CONFIG);
+	platform_set_drvdata(pdev, NULL);
+	if (smmu->as) {
+		int i;
+
+		for (i = 0; i < smmu->num_as; i++)
+			free_pdir(&smmu->as[i]);
+		devm_kfree(dev, smmu->as);
+	}
+	if (smmu->avp_vector_page)
+		__free_page(smmu->avp_vector_page);
+	if (smmu->regs)
+		devm_iounmap(dev, smmu->regs);
+	if (smmu->regs_ahbarb)
+		devm_iounmap(dev, smmu->regs_ahbarb);
+	devm_kfree(dev, smmu);
+	smmu_handle = NULL;
+	return 0;
+}
+
+const struct dev_pm_ops tegra_smmu_pm_ops = {
+	.suspend	= tegra_smmu_suspend,
+	.resume		= tegra_smmu_resume,
+};
+
+static struct platform_driver tegra_smmu_driver = {
+	.probe		= tegra_smmu_probe,
+	.remove		= tegra_smmu_remove,
+	.driver = {
+		.owner	= THIS_MODULE,
+		.name	= "tegra-smmu",
+		.pm	= &tegra_smmu_pm_ops,
+	},
+};
+
+static int __devinit tegra_smmu_init(void)
+{
+	bus_set_iommu(&platform_bus_type, &smmu_iommu_ops);
+	return platform_driver_register(&tegra_smmu_driver);
+}
+
+static void __exit tegra_smmu_exit(void)
+{
+	platform_driver_unregister(&tegra_smmu_driver);
+}
+
+subsys_initcall(tegra_smmu_init);
+module_exit(tegra_smmu_exit);
+
+MODULE_DESCRIPTION("IOMMU API for SMMU in Tegra30");
+MODULE_AUTHOR("Hiroshi DOYU <hdoyu@nvidia.com>");
+MODULE_LICENSE("GPL v2");
-- 
1.7.5.4

Re: [PATCH 2/2] ARM: IOMMU: Tegra30: Add iommu_ops for SMMU driver

[joro-zLv9SwRftAIdnm+yROfE0A]

On Wed, Jan 25, 2012 at 08:39:32AM +0100, Hiroshi Doyu wrote:
> From: Hiroshi DOYU <hdoyu-DDmLM1+adcrQT0dZR+AlfA@public.gmane.org>
> Date: Thu, 17 Nov 2011 07:31:31 +0200
> Subject: [PATCH 2/2] ARM: IOMMU: Tegra30: Add iommu_ops for SMMU driver
> 
> Tegra 30 IOMMU H/W, SMMU (System Memory Management Unit). This patch
> implements struct iommu_ops for SMMU for the upper IOMMU API.
> 
> This H/W module supports multiple virtual address spaces(domain x4),
> and manages 2 level H/W translation pagetable.
> 
> Signed-off-by: Hiroshi DOYU <hdoyu-DDmLM1+adcrQT0dZR+AlfA@public.gmane.org>
> ---
>  arch/arm/mach-tegra/include/mach/smmu.h |   63 ++
>  drivers/iommu/Kconfig                   |   11 +
>  drivers/iommu/Makefile                  |    1 +
>  drivers/iommu/tegra-smmu.c              | 1034 +++++++++++++++++++++++++++++++
>  4 files changed, 1109 insertions(+), 0 deletions(-)
>  create mode 100644 arch/arm/mach-tegra/include/mach/smmu.h
>  create mode 100644 drivers/iommu/tegra-smmu.c

Removed 'default y' and applied to arm/tegra as well. Thanks.


	Joerg

Re: [PATCH v3 0/2] ARM: IOMMU: tegra: Add iommu_ops for GART/SMMU driver

[Hiroshi Doyu]

On Thu, 5 Jan 2012 08:11:47 +0100
Hiroshi Doyu <hdoyu-DDmLM1+adcrQT0dZR+AlfA@public.gmane.org> wrote:

> This patchset adds support for IOMMU API for Tegra20(GART) and
> Tegra30(SMMU). "struct iommu_ops" are implemented for both H/W IOMMU
> modules.
> 
> Tested with "Ion memory manager" and "DMA mapping API"(*1).

Just for DMA mapping test from MPU side, the following one is used.

From a7a163c29694c056e020ba9200648120498e1764 Mon Sep 17 00:00:00 2001
From: Hiroshi DOYU <hdoyu-DDmLM1+adcrQT0dZR+AlfA@public.gmane.org>
Date: Thu, 15 Dec 2011 13:24:24 +0200
Subject: [PATCH 1/1] [NOT-FOR-MERGE] TEST: Simple dma-iommu-mapping API TEST
 module

This is not posted to merge but this is necessary to understand how
SMMU/GART works with DMA(-iommu-mapping-)API.

This is a test to verify DMA API(DMA iommu mapping API), where SoC
specific iommu_ops is used internally.

This does alloc/(un)map/free, but there's no actual access from device
side since it requires device specific communications.

Signed-off-by: Hiroshi DOYU <hdoyu-DDmLM1+adcrQT0dZR+AlfA@public.gmane.org>
---
 drivers/iommu/dmaapi-test.c |  193 +++++++++++++++++++++++++++++++++++++++++++
 1 files changed, 193 insertions(+), 0 deletions(-)
 create mode 100644 drivers/iommu/dmaapi-test.c

diff --git a/drivers/iommu/dmaapi-test.c b/drivers/iommu/dmaapi-test.c
new file mode 100644
index 0000000..f386249
--- /dev/null
+++ b/drivers/iommu/dmaapi-test.c
@@ -0,0 +1,193 @@
+/*
+ * DMA IOMMU mapping API test module
+ *
+ * Copyright (c) 2012, NVIDIA CORPORATION.  All rights reserved.
+ *
+ * Author: Krishna Reddy <vdumpa-DDmLM1+adcrQT0dZR+AlfA@public.gmane.org>,
+ *	Hiroshi DOYU <hdoyu-DDmLM1+adcrQT0dZR+AlfA@public.gmane.org>
+ *
+ * This program is free software; you can redistribute it and/or modify it
+ * under the terms and conditions of the GNU General Public License,
+ * version 2, as published by the Free Software Foundation.
+ *
+ * This program is distributed in the hope it will be useful, but WITHOUT
+ * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
+ * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
+ * more details.
+ *
+ * You should have received a copy of the GNU General Public License along with
+ * this program; if not, write to the Free Software Foundation, Inc.,
+ * 51 Franklin St - Fifth Floor, Boston, MA 02110-1301 USA.
+ */
+
+#define DEBUG
+#define pr_fmt(fmt)	KBUILD_MODNAME ":" fmt
+
+#include <linux/module.h>
+#include <linux/dma-mapping.h>
+#include <linux/kthread.h>
+#include <linux/delay.h>
+#include <linux/highmem.h>
+#include <linux/pfn.h>
+#include <linux/dma-mapping.h>
+#include <linux/platform_device.h>
+
+#include <mach/iomap.h>
+#include <mach/smmu.h>
+
+#include <asm/dma-iommu.h>
+
+#include "devices.h"
+
+/* FIXME: 2x and 3x should be supported at once */
+#ifdef CONFIG_ARCH_TEGRA_2x_SOC
+#define IOVA_START	TEGRA_GART_BASE
+#define IOVA_SIZE	TEGRA_GART_SIZE
+static struct platform_device *tegra_iommu_device = &tegra_gart_device;
+#elif CONFIG_ARCH_TEGRA_3x_SOC
+#define IOVA_START	TEGRA_SMMU_BASE
+#define IOVA_SIZE	TEGRA_SMMU_SIZE
+static struct platform_device *tegra_iommu_device = &tegra_smmu_device;
+#else
+#error Unsupported device
+#endif
+
+#define NUM_TEST	3
+#define MAP_SIZE	(4 * PAGE_SIZE)
+
+struct dmaapi_test_case {
+	char *name;
+	void (*fn)(struct device *);
+};
+
+static void dmaapi_test_map_page(struct device *dev)
+{
+	struct page *page;
+	dma_addr_t dma_addr;
+	void *cpu_addr;
+
+	page = alloc_page(GFP_KERNEL);
+	BUG_ON(!page);
+
+	dma_addr = dma_map_page(dev, page, 0, PAGE_SIZE, DMA_TO_DEVICE);
+	BUG_ON(!dma_addr);
+
+	cpu_addr = kmap(page);
+	BUG_ON(!cpu_addr);
+	memset(cpu_addr, 0xa5, PAGE_SIZE);
+	kunmap(cpu_addr);
+
+	pr_debug("pid:%d,%s mapped\t%08x:%08x\n",
+		 current->pid, dev_name(dev), dma_addr, page_to_phys(page));
+
+	dma_unmap_page(dev, dma_addr, PAGE_SIZE, DMA_TO_DEVICE);
+	__free_page(page);
+}
+
+static void dmaapi_test_alloc_coherent(struct device *dev)
+{
+	dma_addr_t da[NUM_TEST];
+	void *va[NUM_TEST];
+	int i;
+
+	for (i = 0; i < NUM_TEST; i++) {
+		void *cpu_addr;
+		dma_addr_t dma_addr;
+
+		cpu_addr = dma_alloc_coherent(dev, MAP_SIZE,
+					      &dma_addr, GFP_KERNEL);
+		BUG_ON(!cpu_addr);
+		memset(cpu_addr, 0xa5, MAP_SIZE);
+
+		pr_debug("pid:%d,%s,[%d] mapped\t%08x:%08x\n",
+			 current->pid, dev_name(dev), i,
+			 dma_addr, virt_to_phys(cpu_addr));
+
+		da[i] = dma_addr;
+		va[i] = cpu_addr;
+	}
+
+	while (--i >= 0) {
+		pr_debug("pid:%d,%s,[%d] unmapping\t%08x:%08x\n",
+			 current->pid, dev_name(dev), i,
+			 da[i], virt_to_phys(va[i]));
+		dma_free_coherent(dev, MAP_SIZE, va[i], da[i]);
+	}
+}
+
+static struct dmaapi_test_case test[] = {
+	{
+		.name = "dmaapi/map page",
+		.fn = dmaapi_test_map_page,
+	},
+	{
+		.name = "dmaapi/alloc coherent",
+		.fn = dmaapi_test_alloc_coherent,
+	},
+};
+
+static u32 dummy_hwgrp_map[] = {
+	HWG_DC | HWG_AFI | HWG_AVPC | HWG_DCB,
+	HWG_EPP | HWG_HC | HWG_G2 | HWG_MPE | HWG_HDA | HWG_ISP,
+	HWG_NV | HWG_PPCS | HWG_SATA | HWG_NV2 | HWG_VI | HWG_VDE,
+};
+
+/* FIXME: Need driver for iommu context? */
+static struct platform_device dmaapi_dummy_device[] = {
+	{ .name = "hwgrp@a", .id = -1, },
+	{ .name = "hwgrp@b", .id = -1, },
+	{ .name = "hwgrp@c", .id = -1, },
+};
+
+static int dmaapi_test_thread(void *data)
+{
+	int i;
+	struct dmaapi_test_case *c = data;
+
+	for (i = 0; true; i++) {
+		struct device *dev;
+		int interval[] = {7, 3, 5,};
+		int n;
+
+		n = i % ARRAY_SIZE(dmaapi_dummy_device);
+		ssleep(interval[n]);
+		dev = &dmaapi_dummy_device[n].dev;
+		c->fn(dev);
+	}
+	return 0;
+}
+
+static int __init dmaapi_test_init(void)
+{
+	int i;
+	struct dma_iommu_mapping *map;
+
+	map = arm_iommu_create_mapping(IOVA_START, IOVA_SIZE, 0);
+	BUG_ON(!map);
+	pr_debug("Allocate IOVA: %08x-%08x\n", map->base, map->base + IOVA_SIZE);
+
+	for (i = 0; i < ARRAY_SIZE(dmaapi_dummy_device); i++) {
+		int err;
+		struct platform_device *pdev = &dmaapi_dummy_device[i];
+
+		pdev->dev.platform_data = (void *)dummy_hwgrp_map[i];
+		pdev->dev.parent = &tegra_iommu_device->dev;
+		err = platform_device_register(pdev);
+		BUG_ON(err);
+
+		err = arm_iommu_attach_device(&pdev->dev, map);
+		BUG_ON(err);
+		pr_debug("IOMMU API: Attached to %s\n", dev_name(&pdev->dev));
+	}
+
+	for (i = 0; i < ARRAY_SIZE(test); i++)
+		kthread_run(dmaapi_test_thread, &test[i], test[i].name);
+
+	return 0;
+}
+module_init(dmaapi_test_init);
+
+MODULE_AUTHOR("Krishna Reddy <vdumpa-DDmLM1+adcrQT0dZR+AlfA@public.gmane.org>");
+MODULE_AUTHOR("Hiroshi DOYU <hdoyu-DDmLM1+adcrQT0dZR+AlfA@public.gmane.org>");
+MODULE_DESCRIPTION("DMA IOMMU mapping API test");
+MODULE_LICENSE("GPL v2");
-- 
1.7.5.4

Re: [PATCH v3 0/2] ARM: IOMMU: tegra: Add iommu_ops for GART/SMMU driver

[Russell King - ARM Linux]

On Thu, Jan 05, 2012 at 09:17:18AM +0200, Hiroshi Doyu wrote:
> Just for DMA mapping test from MPU side, the following one is used.

This patch is buggy.

> +static void dmaapi_test_map_page(struct device *dev)
> +{
> +	struct page *page;
> +	dma_addr_t dma_addr;
> +	void *cpu_addr;
> +
> +	page = alloc_page(GFP_KERNEL);
> +	BUG_ON(!page);
> +
> +	dma_addr = dma_map_page(dev, page, 0, PAGE_SIZE, DMA_TO_DEVICE);
> +	BUG_ON(!dma_addr);
> +
> +	cpu_addr = kmap(page);
> +	BUG_ON(!cpu_addr);
> +	memset(cpu_addr, 0xa5, PAGE_SIZE);
> +	kunmap(cpu_addr);

The DMA API works like this:

- The CPU owns the page or buffer and can access it.
- You map the page or buffer.
- The device owns the page or buffer; the CPU must explicitly access it.
- You unmap the page or buffer.
- The CPU again owns the page/buffer and can access it.

Please respect the DMA API rules.

So.  Once dma_map_page() has returned, you must not kmap() or otherwise
access the data contained in that page until after you have unmapped it.

Re: [PATCH v3 0/2] ARM: IOMMU: tegra: Add iommu_ops for GART/SMMU driver

[Hiroshi Doyu]

Hi Russell,

From: Russell King - ARM Linux <linux-lFZ/pmaqli7XmaaqVzeoHQ@public.gmane.org>
Subject: Re: [PATCH v3 0/2] ARM: IOMMU: tegra: Add iommu_ops for GART/SMMU driver
Date: Thu, 5 Jan 2012 13:53:26 +0100
Message-ID: <20120105125326.GT11810-l+eeeJia6m9vn6HldHNs0ANdhmdF6hFW@public.gmane.org>

> On Thu, Jan 05, 2012 at 09:17:18AM +0200, Hiroshi Doyu wrote:
> > Just for DMA mapping test from MPU side, the following one is used.
>
> This patch is buggy.
>
> > +static void dmaapi_test_map_page(struct device *dev)
> > +{
> > +	struct page *page;
> > +	dma_addr_t dma_addr;
> > +	void *cpu_addr;
> > +
> > +	page = alloc_page(GFP_KERNEL);
> > +	BUG_ON(!page);
> > +
> > +	dma_addr = dma_map_page(dev, page, 0, PAGE_SIZE, DMA_TO_DEVICE);
> > +	BUG_ON(!dma_addr);
> > +
> > +	cpu_addr = kmap(page);
> > +	BUG_ON(!cpu_addr);
> > +	memset(cpu_addr, 0xa5, PAGE_SIZE);
> > +	kunmap(cpu_addr);
>
> The DMA API works like this:
>
> - The CPU owns the page or buffer and can access it.
> - You map the page or buffer.
> - The device owns the page or buffer; the CPU must explicitly access it.
> - You unmap the page or buffer.
> - The CPU again owns the page/buffer and can access it.
>
> Please respect the DMA API rules.

Right, I do.

> So.  Once dma_map_page() has returned, you must not kmap() or otherwise
> access the data contained in that page until after you have unmapped it.

Ok, the above function order should be as below?

  page = alloc_page(GFP_KERNEL);
      cpu_addr = kmap(page);
      memset(cpu_addr, 0xa5, PAGE_SIZE);
  	dma_addr = dma_map_page(dev, page, 0, PAGE_SIZE, DMA_TO_DEVICE);
  	    < expect GPU does something >
  	dma_unmap_page(dev, dma_addr, PAGE_SIZE, DMA_TO_DEVICE);
      < CPU can access data here again>
      kunmap(cpu_addr);
  __free_page(page);

Re: [PATCH v3 0/2] ARM: IOMMU: tegra: Add iommu_ops for GART/SMMU driver

[Russell King - ARM Linux]

On Thu, Jan 05, 2012 at 03:29:30PM +0100, Hiroshi Doyu wrote:
> Ok, the above function order should be as below?
> 
>   page = alloc_page(GFP_KERNEL);
>       cpu_addr = kmap(page);
>       memset(cpu_addr, 0xa5, PAGE_SIZE);
>   	dma_addr = dma_map_page(dev, page, 0, PAGE_SIZE, DMA_TO_DEVICE);
>   	    < expect GPU does something >
>   	dma_unmap_page(dev, dma_addr, PAGE_SIZE, DMA_TO_DEVICE);
>       < CPU can access data here again>
>       kunmap(cpu_addr);
>   __free_page(page);

That should work, but it's not particularly nice to wrap kmap around
the DMA API.

Bear in mind that alloc_page(GFP_KERNEL) will always give you a lowmem
page.  So, you might as well do:

	page = alloc_page(GFP_KERNEL);
	cpu_addr = page_address(page);
	memset()
	dma_addr = dma_map_single(dev, cpu_addr, PAGE_SIZE, DMA_TO_DEVICE);
		< GPU access >
	dma_unmap_single(dev, dma_ddr, PAGE_SIZE, DMA_TO_DEVICE);
	< CPU access >
	__free_page(page);

and this doesn't raise any issues with kmap vs DMA API.

Re: [PATCH v3 0/2] ARM: IOMMU: tegra: Add iommu_ops for GART/SMMU driver

[Hiroshi Doyu]

[-- Attachment #1: Type: text/plain, Size: 1447 bytes --]

From: Russell King - ARM Linux <linux-lFZ/pmaqli7XmaaqVzeoHQ@public.gmane.org>
Subject: Re: [PATCH v3 0/2] ARM: IOMMU: tegra: Add iommu_ops for GART/SMMU driver
Date: Thu, 5 Jan 2012 15:46:31 +0100
Message-ID: <20120105144631.GV11810-l+eeeJia6m9vn6HldHNs0ANdhmdF6hFW@public.gmane.org>

> On Thu, Jan 05, 2012 at 03:29:30PM +0100, Hiroshi Doyu wrote:
> > Ok, the above function order should be as below?
> > 
> >   page = alloc_page(GFP_KERNEL);
> >       cpu_addr = kmap(page);
> >       memset(cpu_addr, 0xa5, PAGE_SIZE);
> >   	dma_addr = dma_map_page(dev, page, 0, PAGE_SIZE, DMA_TO_DEVICE);
> >   	    < expect GPU does something >
> >   	dma_unmap_page(dev, dma_addr, PAGE_SIZE, DMA_TO_DEVICE);
> >       < CPU can access data here again>
> >       kunmap(cpu_addr);
> >   __free_page(page);
> 
> That should work, but it's not particularly nice to wrap kmap around
> the DMA API.
> 
> Bear in mind that alloc_page(GFP_KERNEL) will always give you a lowmem
> page.  So, you might as well do:
> 
> 	page = alloc_page(GFP_KERNEL);
> 	cpu_addr = page_address(page);
> 	memset()
> 	dma_addr = dma_map_single(dev, cpu_addr, PAGE_SIZE, DMA_TO_DEVICE);
> 		< GPU access >
> 	dma_unmap_single(dev, dma_ddr, PAGE_SIZE, DMA_TO_DEVICE);
> 	< CPU access >
> 	__free_page(page);
> 
> and this doesn't raise any issues with kmap vs DMA API.

Attached the update one just in case that someone wants to use it for theirs.

[-- Warning: decoded text below may be mangled, UTF-8 assumed --]
[-- Attachment #2: 0001-NOT-FOR-MERGE-TEST-Simple-dma-iommu-mapping-API-TEST.patch --]
[-- Type: text/x-patch; name="0001-NOT-FOR-MERGE-TEST-Simple-dma-iommu-mapping-API-TEST.patch", Size: 6451 bytes --]

From 76aefea1c7ad004590295dbe63c27f0c6f482f7a Mon Sep 17 00:00:00 2001
From: Hiroshi DOYU <hdoyu@nvidia.com>
Date: Thu, 15 Dec 2011 13:24:24 +0200
Subject: [PATCH 1/1] [NOT-FOR-MERGE] TEST: Simple dma-iommu-mapping API TEST
 module

This is not posted to merge but this is necessary to understand how
SMMU/GART works with DMA(-iommu-mapping-)API.

This is a test to verify DMA API(DMA iommu mapping API), where SoC
specific iommu_ops is used internally.

This does alloc/(un)map/free, but there's no actual access from device
side since it requires device specific communications.

Signed-off-by: Hiroshi DOYU <hdoyu@nvidia.com>
---
 drivers/iommu/dmaapi-test.c |  202 +++++++++++++++++++++++++++++++++++++++++++
 1 files changed, 202 insertions(+), 0 deletions(-)
 create mode 100644 drivers/iommu/dmaapi-test.c

diff --git a/drivers/iommu/dmaapi-test.c b/drivers/iommu/dmaapi-test.c
new file mode 100644
index 0000000..3199386
--- /dev/null
+++ b/drivers/iommu/dmaapi-test.c
@@ -0,0 +1,202 @@
+/*
+ * DMA IOMMU mapping API test module
+ *
+ * Copyright (c) 2012, NVIDIA CORPORATION.  All rights reserved.
+ *
+ * Author: Krishna Reddy <vdumpa@nvidia.com>,
+ *	Hiroshi DOYU <hdoyu@nvidia.com>
+ *
+ * This program is free software; you can redistribute it and/or modify it
+ * under the terms and conditions of the GNU General Public License,
+ * version 2, as published by the Free Software Foundation.
+ *
+ * This program is distributed in the hope it will be useful, but WITHOUT
+ * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
+ * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
+ * more details.
+ *
+ * You should have received a copy of the GNU General Public License along with
+ * this program; if not, write to the Free Software Foundation, Inc.,
+ * 51 Franklin St - Fifth Floor, Boston, MA 02110-1301 USA.
+ */
+
+#define DEBUG
+#define pr_fmt(fmt)	KBUILD_MODNAME ":" fmt
+
+#include <linux/module.h>
+#include <linux/dma-mapping.h>
+#include <linux/kthread.h>
+#include <linux/delay.h>
+#include <linux/highmem.h>
+#include <linux/pfn.h>
+#include <linux/dma-mapping.h>
+#include <linux/platform_device.h>
+
+#include <mach/iomap.h>
+#include <mach/smmu.h>
+
+#include <asm/dma-iommu.h>
+
+#include "devices.h"
+
+/* FIXME: 2x and 3x should be supported at once */
+#ifdef CONFIG_ARCH_TEGRA_2x_SOC
+#define IOVA_START	TEGRA_GART_BASE
+#define IOVA_SIZE	TEGRA_GART_SIZE
+static struct platform_device *tegra_iommu_device = &tegra_gart_device;
+#elif CONFIG_ARCH_TEGRA_3x_SOC
+#define IOVA_START	TEGRA_SMMU_BASE
+#define IOVA_SIZE	TEGRA_SMMU_SIZE
+static struct platform_device *tegra_iommu_device = &tegra_smmu_device;
+#else
+#error Unsupported device
+#endif
+
+#define NUM_TEST	3
+#define MAP_SIZE	(4 * PAGE_SIZE)
+
+struct dmaapi_test_case {
+	char *name;
+	void (*fn)(struct device *);
+};
+
+static void dmaapi_test_map_page(struct device *dev)
+{
+	struct page *page;
+	dma_addr_t dma_addr;
+	void *cpu_addr;
+
+	page = alloc_page(GFP_KERNEL);
+	BUG_ON(!page);
+
+	cpu_addr = page_address(page);
+	BUG_ON(!cpu_addr);
+	memset(cpu_addr, 0xa5, PAGE_SIZE);
+
+	dma_addr = dma_map_page(dev, page, 0, PAGE_SIZE, DMA_TO_DEVICE);
+	BUG_ON(!dma_addr);
+
+	pr_debug("pid:%d,%s mapped\t%08x:%08x\n",
+		 current->pid, dev_name(dev), dma_addr, page_to_phys(page));
+
+	/*
+	 * Expect GPU access
+	 */
+
+	dma_unmap_page(dev, dma_addr, PAGE_SIZE, DMA_TO_DEVICE);
+
+	/*
+	 * CPU access
+	 */
+	BUG_ON(*(char *)cpu_addr != 0xa5);
+
+	__free_page(page);
+}
+
+static void dmaapi_test_alloc_coherent(struct device *dev)
+{
+	dma_addr_t da[NUM_TEST];
+	void *va[NUM_TEST];
+	int i;
+
+	for (i = 0; i < NUM_TEST; i++) {
+		void *cpu_addr;
+		dma_addr_t dma_addr;
+
+		cpu_addr = dma_alloc_coherent(dev, MAP_SIZE,
+					      &dma_addr, GFP_KERNEL);
+		BUG_ON(!cpu_addr);
+		memset(cpu_addr, 0xa5, MAP_SIZE);
+
+		pr_debug("pid:%d,%s,[%d] mapped\t%08x:%08x\n",
+			 current->pid, dev_name(dev), i,
+			 dma_addr, virt_to_phys(cpu_addr));
+
+		da[i] = dma_addr;
+		va[i] = cpu_addr;
+	}
+
+	while (--i >= 0) {
+		pr_debug("pid:%d,%s,[%d] unmapping\t%08x:%08x\n",
+			 current->pid, dev_name(dev), i,
+			 da[i], virt_to_phys(va[i]));
+		dma_free_coherent(dev, MAP_SIZE, va[i], da[i]);
+	}
+}
+
+static struct dmaapi_test_case test[] = {
+	{
+		.name = "dmaapi/map page",
+		.fn = dmaapi_test_map_page,
+	},
+	{
+		.name = "dmaapi/alloc coherent",
+		.fn = dmaapi_test_alloc_coherent,
+	},
+};
+
+static u32 dummy_hwgrp_map[] = {
+	HWG_DC | HWG_AFI | HWG_AVPC | HWG_DCB,
+	HWG_EPP | HWG_HC | HWG_G2 | HWG_MPE | HWG_HDA | HWG_ISP,
+	HWG_NV | HWG_PPCS | HWG_SATA | HWG_NV2 | HWG_VI | HWG_VDE,
+};
+
+/* FIXME: Need driver for iommu context? */
+static struct platform_device dmaapi_dummy_device[] = {
+	{ .name = "hwgrp@a", .id = -1, },
+	{ .name = "hwgrp@b", .id = -1, },
+	{ .name = "hwgrp@c", .id = -1, },
+};
+
+static int dmaapi_test_thread(void *data)
+{
+	int i;
+	struct dmaapi_test_case *c = data;
+
+	for (i = 0; true; i++) {
+		struct device *dev;
+		int interval[] = {7, 3, 5,};
+		int n;
+
+		n = i % ARRAY_SIZE(dmaapi_dummy_device);
+		ssleep(interval[n]);
+		dev = &dmaapi_dummy_device[n].dev;
+		c->fn(dev);
+	}
+	return 0;
+}
+
+static int __init dmaapi_test_init(void)
+{
+	int i;
+	struct dma_iommu_mapping *map;
+
+	map = arm_iommu_create_mapping(IOVA_START, IOVA_SIZE, 0);
+	BUG_ON(!map);
+	pr_debug("Allocate IOVA: %08x-%08x\n", map->base, map->base + IOVA_SIZE);
+
+	for (i = 0; i < ARRAY_SIZE(dmaapi_dummy_device); i++) {
+		int err;
+		struct platform_device *pdev = &dmaapi_dummy_device[i];
+
+		pdev->dev.platform_data = (void *)dummy_hwgrp_map[i];
+		pdev->dev.parent = &tegra_iommu_device->dev;
+		err = platform_device_register(pdev);
+		BUG_ON(err);
+
+		err = arm_iommu_attach_device(&pdev->dev, map);
+		BUG_ON(err);
+		pr_debug("IOMMU API: Attached to %s\n", dev_name(&pdev->dev));
+	}
+
+	for (i = 0; i < ARRAY_SIZE(test); i++)
+		kthread_run(dmaapi_test_thread, &test[i], test[i].name);
+
+	return 0;
+}
+module_init(dmaapi_test_init);
+
+MODULE_AUTHOR("Krishna Reddy <vdumpa@nvidia.com>");
+MODULE_AUTHOR("Hiroshi DOYU <hdoyu@nvidia.com>");
+MODULE_DESCRIPTION("DMA IOMMU mapping API test");
+MODULE_LICENSE("GPL v2");
-- 
1.7.5.4

Re: [PATCH v3 0/2] ARM: IOMMU: tegra: Add iommu_ops for GART/SMMU driver

[KyongHo Cho]

On Thu, Jan 5, 2012 at 9:53 PM, Russell King - ARM Linux
<linux@arm.linux.org.uk> wrote:
> On Thu, Jan 05, 2012 at 09:17:18AM +0200, Hiroshi Doyu wrote:
>> Just for DMA mapping test from MPU side, the following one is used.
>
> This patch is buggy.
>
>> +static void dmaapi_test_map_page(struct device *dev)
>> +{
>> +     struct page *page;
>> +     dma_addr_t dma_addr;
>> +     void *cpu_addr;
>> +
>> +     page = alloc_page(GFP_KERNEL);
>> +     BUG_ON(!page);
>> +
>> +     dma_addr = dma_map_page(dev, page, 0, PAGE_SIZE, DMA_TO_DEVICE);
>> +     BUG_ON(!dma_addr);
>> +
>> +     cpu_addr = kmap(page);
>> +     BUG_ON(!cpu_addr);
>> +     memset(cpu_addr, 0xa5, PAGE_SIZE);
>> +     kunmap(cpu_addr);
>
> The DMA API works like this:
>
> - The CPU owns the page or buffer and can access it.
> - You map the page or buffer.
> - The device owns the page or buffer; the CPU must explicitly access it.

I don't understand what  "CPU must __explicitly__ access it" means.

Do you mean that CPU must access the mapped buffer when
it __explicitly__ knows that the access does not cause any side effect?


> - You unmap the page or buffer.
> - The CPU again owns the page/buffer and can access it.
>
> Please respect the DMA API rules.
>
> So.  Once dma_map_page() has returned, you must not kmap() or otherwise
> access the data contained in that page until after you have unmapped it.
> _______________________________________________
> iommu mailing list
> iommu@lists.linux-foundation.org
> https://lists.linuxfoundation.org/mailman/listinfo/iommu

Re: [PATCH v3 0/2] ARM: IOMMU: tegra: Add iommu_ops for GART/SMMU driver

[Russell King - ARM Linux]

On Mon, Jan 09, 2012 at 09:39:48AM +0900, KyongHo Cho wrote:
> On Thu, Jan 5, 2012 at 9:53 PM, Russell King - ARM Linux
> <linux@arm.linux.org.uk> wrote:
> > On Thu, Jan 05, 2012 at 09:17:18AM +0200, Hiroshi Doyu wrote:
> >> Just for DMA mapping test from MPU side, the following one is used.
> >
> > This patch is buggy.
> >
> >> +static void dmaapi_test_map_page(struct device *dev)
> >> +{
> >> +     struct page *page;
> >> +     dma_addr_t dma_addr;
> >> +     void *cpu_addr;
> >> +
> >> +     page = alloc_page(GFP_KERNEL);
> >> +     BUG_ON(!page);
> >> +
> >> +     dma_addr = dma_map_page(dev, page, 0, PAGE_SIZE, DMA_TO_DEVICE);
> >> +     BUG_ON(!dma_addr);
> >> +
> >> +     cpu_addr = kmap(page);
> >> +     BUG_ON(!cpu_addr);
> >> +     memset(cpu_addr, 0xa5, PAGE_SIZE);
> >> +     kunmap(cpu_addr);
> >
> > The DMA API works like this:
> >
> > - The CPU owns the page or buffer and can access it.
> > - You map the page or buffer.
> > - The device owns the page or buffer; the CPU must explicitly access it.

Sorry, that 'must' should have been a 'must not'.

> I don't understand what  "CPU must __explicitly__ access it" means.
> 
> Do you mean that CPU must access the mapped buffer when
> it __explicitly__ knows that the access does not cause any side effect?

No - I mean that the program must not attempt to read or write the
buffer.