[PATCH 3/4] iommu/exynos: Add iommu driver for Exynos4 Platforms

[Russell King - ARM Linux]

On Sat, Sep 24, 2011 at 07:35:45AM +0000, 조경호 wrote:
> +# EXYNOS IOMMU support
> +config EXYNOS_IOMMU
> + bool "Exynos IOMMU Support"
> + depends on ARCH_EXYNOS4
> + select IOMMU_API
> + select EXYNOS4_DEV_SYSMMU
> + help

Looks like your mailer converted tabs to one space.  This makes it more
difficult to read this patch as a whole.  Don't expect good reviews as
long as your mailer breaks stuff in this way.

> diff --git a/drivers/iommu/Makefile b/drivers/iommu/Makefile
> index 4d4d77d..1eb924f 100644
> --- a/drivers/iommu/Makefile
> +++ b/drivers/iommu/Makefile
> @@ -3,3 +3,4 @@ obj-$(CONFIG_MSM_IOMMU) += msm_iommu.o msm_iommu_dev.o
> obj-$(CONFIG_AMD_IOMMU) += amd_iommu.o amd_iommu_init.o
> obj-$(CONFIG_DMAR) += dmar.o iova.o intel-iommu.o
> obj-$(CONFIG_INTR_REMAP) += dmar.o intr_remapping.o

Looks like your mailer eliminated spaces at the start of these lines.

> +obj-$(CONFIG_EXYNOS_IOMMU) += exynos_iommu.o
> diff --git a/drivers/iommu/exynos_iommu.c b/drivers/iommu/exynos_iommu.c
> new file mode 100644
> index 0000000..fe5b5d8
> --- /dev/null
> +++ b/drivers/iommu/exynos_iommu.c
> @@ -0,0 +1,859 @@
> +/* linux/drivers/iommu/exynos_iommu.c
> + *
> + * Copyright (c) 2011 Samsung Electronics Co., Ltd.
> + * http://www.samsung.com
> + *
> + * This program is free software; you can redistribute it and/or modify
> + * it under the terms of the GNU General Public License version 2 as
> + * published by the Free Software Foundation.
> + */
> +
> +#include 
> +#include 
> +#include 
> +#include 
> +#include 
> +#include 
> +#include 
> +#include 
> +#include 
> +#include 
> +#include 
> +
> +#include 
> +#include 
> +
> +#include 
> +#include 
> +#include 

Looks like your mailer... well... is totally broken beyond belief.

> +static inline struct sysmmu_platdata *get_platdata(struct device *dev)
> +{
> + return dev_get_platdata(dev);
> +}
> +
> +static inline bool set_sysmmu_active(struct sysmmu_platdata *mmudata)
> +{
> + /* return true if the System MMU was not active previously
> +    and it needs to be initialized */
> +
> + return atomic_inc_return(&mmudata->activations) == 1;
> +}
> +
> +static inline bool set_sysmmu_inactive(struct sysmmu_platdata *mmudata)
> +{
> + /* return true if the System MMU is needed to be disabled */
> + int ref;
> +
> + ref = atomic_dec_return(&mmudata->activations);
> +
> + if (ref == 0)
> + return true;
> +
> + if (WARN_ON(ref < 0)) {
> + /* System MMU is already disabled */
> + atomic_set(&mmudata->activations, 0);
> + ref = 0;
> + }
> +
> + return false;
> +}
> +
> +static inline bool is_sysmmu_active(struct sysmmu_platdata *mmudata)
> +{
> + return atomic_read(&mmudata->activations) != 0;
> +}

This use of an atomic type is total rubbish.  There's nothing magical
about 'atomic_*' stuff which suddenly makes the rest of your code
magically correct.  Think about this:

	Thread0			Thread1
	atomic_inc_return(&v)
		== 1
				atomic_dec_return(&v)
					== 0
				disables MMU
	enables MMU

Now you have the situation where the atomic value is '0' yet the MMU is
enabled.

Repeat after me: atomic types do not make code atomic.  atomic types
just make modifications to the type itself atomic.  Never use atomic
types for code synchronization.

> +static irqreturn_t exynos_sysmmu_irq(int irq, void *dev_id)
> +{
> + /* SYSMMU is in blocked when interrupt occurred. */
> + unsigned long addr;
> + struct sysmmu_platdata *mmudata = dev_id;
> + enum S5P_SYSMMU_INTERRUPT_TYPE itype;
> +
> + WARN_ON(!is_sysmmu_active(mmudata));
> +
> + itype = (enum S5P_SYSMMU_INTERRUPT_TYPE)
> + __ffs(__raw_readl(mmudata->sfrbase + S5P_INT_STATUS));
> +
> + BUG_ON(!((itype >= 0) && (itype < 8)));
> +
> + dev_alert(mmudata->dev, "SYSTEM MMU FAULT!!!.\n");

Do the exclaimation marks do anything for this error message?

> +
> + if (!mmudata->domain)
> + return IRQ_NONE;
> +
> + addr = __raw_readl(mmudata->sfrbase + fault_reg_offset[itype]);
> +
> + if (!report_iommu_fault(mmudata->domain, mmudata->owner, addr, itype)) {
> + __raw_writel(1 << itype, mmudata->sfrbase + S5P_INT_CLEAR);
> + dev_notice(mmudata->dev,
> + "%s is resolved. Retrying translation.\n",

So, this is a notice severity, yet we've printed at alert level.  Either
this condition satisfies being an alert or not.

> + sysmmu_fault_name[itype]);
> + sysmmu_unblock(mmudata->sfrbase);
> + } else {
> + dev_notice(mmudata->dev, "%s is not handled.\n",
> + sysmmu_fault_name[itype]);
> + }
> +
> + return IRQ_HANDLED;
> +}
> +
> +void exynos_sysmmu_set_tablebase_pgd(struct device *owner, unsigned long pgd)
> +{
> + struct sysmmu_platdata *mmudata = NULL;
> +
> + while ((mmudata = get_sysmmu_data(owner, mmudata))) {
> + if (is_sysmmu_active(mmudata)) {
> + sysmmu_block(mmudata->sfrbase);
> + __sysmmu_set_ptbase(mmudata->sfrbase, pgd);
> + sysmmu_unblock(mmudata->sfrbase);
> + dev_dbg(mmudata->dev, "New page table base is %p\n",
> + (void *)pgd);

If it's unsigned long use %08lx and don't cast.

> +static int exynos_sysmmu_probe(struct platform_device *pdev)
> +{
> + struct resource *res, *ioarea;
> + int ret;
> + int irq;
> + struct device *dev;
> + void *sfr;

void __iomem *sfr;

> +
> + dev = &pdev->dev;
> + if (!get_platdata(dev) || (get_platdata(dev)->owner == NULL)) {
> + dev_err(dev, "Failed to probing system MMU: "
> + "Owner device is not set.");
> + return -ENXIO;
> + }
> +
> + res = platform_get_resource(pdev, IORESOURCE_MEM, 0);
> + if (!res) {
> + dev_err(dev,
> + "Failed probing system MMU: failed to get resource.");
> + return -ENOENT;
> + }
> +
> + ioarea = request_mem_region(res->start, resource_size(res), pdev->name);
> + if (ioarea == NULL) {
> + dev_err(dev, "Failed probing system MMU: "
> + "failed to request memory region.");
> + return -ENOMEM;
> + }
> +
> + sfr = ioremap(res->start, resource_size(res));
> + if (!sfr) {
> + dev_err(dev, "Failed probing system MMU: "
> + "failed to call ioremap().");
> + ret = -ENOENT;
> + goto err_ioremap;
> + }
> +
> + irq = platform_get_irq(pdev, 0);
> + if (irq <= 0) {
> + dev_err(dev, "Failed probing system MMU: "
> + "failed to get irq resource.");
> + ret = irq;
> + goto err_irq;
> + }
> +
> + if (request_irq(irq, exynos_sysmmu_irq, 0, dev_name(&pdev->dev),
> + dev_get_platdata(dev))) {
> + dev_err(dev, "Failed probing system MMU: "
> + "failed to request irq.");
> + ret = -ENOENT;
> + goto err_irq;
> + }
> +
> + get_platdata(dev)->clk = clk_get(dev, "sysmmu");

It's rather horrible to see drivers modifying their platform data, rather
than drivers using the driver data pointer in struct device.  As far as
drivers are concerned, the platform data should only ever be read.

> +
> + if (IS_ERR_OR_NULL(get_platdata(dev)->clk)) {

IS_ERR() only please.  The *only* allowable failure value for drivers to
be concerned about here is if the return value is an error pointer.
Drivers have no business rejecting a NULL pointer here.

> +static int exynos_sysmmu_remove(struct platform_device *pdev)
> +{
> + return 0;
> +}

That doesn't look good.  If you can't remove it don't provide the
function at all.

> +
> +int exynos_sysmmu_runtime_suspend(struct device *dev)
> +{
> + if (WARN_ON(is_sysmmu_active(dev_get_platdata(dev))))
> + return -EFAULT;
> +
> + return 0;
> +}
> +
> +int exynos_sysmmu_runtime_resume(struct device *dev)
> +{
> + if (WARN_ON(is_sysmmu_active(dev_get_platdata(dev))))
> + return -EFAULT;
> +
> + return 0;
> +}

Is there much point to having runtime PM support in this driver?

> +static int exynos_iommu_fault_handler(struct iommu_domain *domain,
> + struct device *dev, unsigned long iova, int flags)
> +{
> + struct exynos_iommu_domain *priv = domain->priv;
> +
> + dev_err(priv->dev, "%s occured at %p(Page table base: %p)\n",
> + sysmmu_fault_name[flags], (void *)iova,
> + (void *)(__pa(priv->pgtable)));

Again, you want to get rid of these casts and use a proper format string
for what these are.

> +static int exynos_iommu_attach_device(struct iommu_domain *domain,
> +    struct device *dev)
> +{
> + struct exynos_iommu_domain *priv = domain->priv;
> + int ret;
> +
> + spin_lock(&priv->lock);
> +
> + priv->dev = dev;
> +
> + ret = exynos_sysmmu_enable(domain);
> + if (ret)
> + return ret;
> +
> + spin_unlock(&priv->lock);
> +
> + return 0;

This function has an indeterminant return state for the spinlock - it
sometimes returns with it locked and sometimes with it unlocked.  That's
bad practice.

Getting rid of the 'if (ret) return ret;' and changing this 'return 0;'
to 'return ret;' is a simple way of solving it - and results in less
complex code.

> +static int exynos_iommu_map(struct iommu_domain *domain, unsigned long iova,
> + phys_addr_t paddr, int gfp_order, int prot)
> +{
> + struct exynos_iommu_domain *priv = domain->priv;
> + unsigned long *start_entry, *entry, *end_entry;
> + int num_entry;
> + int ret = 0;
> + unsigned long flags;
> +
> + BUG_ON(priv->pgtable == NULL);
> +
> + spin_lock_irqsave(&priv->pgtablelock, flags);
> +
> + start_entry = entry = priv->pgtable + (iova >> S5P_SECTION_SHIFT);
> +
> + if (gfp_order >= S5P_SECTION_ORDER) {
> + BUG_ON((paddr | iova) & ~S5P_SECTION_MASK);
> + /* 1MiB mapping */
> +
> + num_entry = 1 << (gfp_order - S5P_SECTION_ORDER);
> + end_entry = entry + num_entry;
> +
> + while (entry != end_entry) {
> + if (!section_available(domain, entry))
> + break;
> +
> + MAKE_SECTION_ENTRY(*entry, paddr);
> +
> + paddr += S5P_SECTION_SIZE;
> + entry++;
> + }
> +
> + if (entry != end_entry)
> + goto mapping_error;
> +
> + pgtable_flush(start_entry, entry);
> + goto mapping_done;
> + }
> +
> + if (S5P_FAULT_LV1_ENTRY(*entry)) {
> + unsigned long *l2table;
> +
> + l2table = kmem_cache_zalloc(l2table_cachep, GFP_KERNEL);
> + if (!l2table) {
> + ret = -ENOMEM;
> + goto nomem_error;
> + }
> +
> + pgtable_flush(entry, entry + S5P_LV2TABLE_ENTRIES);
> +
> + MAKE_LV2TABLE_ENTRY(*entry, virt_to_phys(l2table));
> + pgtable_flush(entry, entry + 1);
> + }
> +
> + /* 'entry' points level 2 entries, hereafter */
> + entry = GET_LV2ENTRY(*entry, iova);
> +
> + start_entry = entry;
> + num_entry = 1 << gfp_order;
> + end_entry = entry + num_entry;
> +
> + if (gfp_order >= S5P_LPAGE_ORDER) {
> + /* large page(64KiB) mapping */
> + BUG_ON((paddr | iova) & ~S5P_LPAGE_MASK);
> +
> + while (entry != end_entry) {
> + if (!write_lpage(entry, paddr)) {
> + pr_err("%s: Failed to allocate large page"
> + " entry.\n", __func__);

Don't wrap error messages, irrespective of what checkpatch may tell you.
They make them impossible to grep for.