iommu: Add driver for Renesas VMSA-compatible IPMMU
authorLaurent Pinchart <laurent.pinchart+renesas@ideasonboard.com>
Wed, 2 Apr 2014 10:47:37 +0000 (12:47 +0200)
committerJoerg Roedel <jroedel@suse.de>
Mon, 26 May 2014 09:18:21 +0000 (11:18 +0200)
Signed-off-by: Laurent Pinchart <laurent.pinchart+renesas@ideasonboard.com>
Signed-off-by: Joerg Roedel <jroedel@suse.de>
drivers/iommu/Kconfig
drivers/iommu/Makefile
drivers/iommu/ipmmu-vmsa.c [new file with mode: 0644]
include/linux/platform_data/ipmmu-vmsa.h [new file with mode: 0644]

index df56e4c74a7e450aae8820d4b7d6837c897c781b..a22b537caacd88f8794094c32f8be6e6e0bc24b0 100644 (file)
@@ -272,6 +272,18 @@ config SHMOBILE_IOMMU_L1SIZE
        default 256 if SHMOBILE_IOMMU_ADDRSIZE_64MB
        default 128 if SHMOBILE_IOMMU_ADDRSIZE_32MB
 
+config IPMMU_VMSA
+       bool "Renesas VMSA-compatible IPMMU"
+       depends on ARM_LPAE
+       depends on ARCH_SHMOBILE || COMPILE_TEST
+       select IOMMU_API
+       select ARM_DMA_USE_IOMMU
+       help
+         Support for the Renesas VMSA-compatible IPMMU Renesas found in the
+         R-Mobile APE6 and R-Car H2/M2 SoCs.
+
+         If unsure, say N.
+
 config SPAPR_TCE_IOMMU
        bool "sPAPR TCE IOMMU Support"
        depends on PPC_POWERNV || PPC_PSERIES
index 5d58bf16e9e3c15d013c3cb071019be5403223df..8893bad048e0e7efdb7eafef3d7fbd8c7e051701 100644 (file)
@@ -7,6 +7,7 @@ obj-$(CONFIG_AMD_IOMMU_V2) += amd_iommu_v2.o
 obj-$(CONFIG_ARM_SMMU) += arm-smmu.o
 obj-$(CONFIG_DMAR_TABLE) += dmar.o
 obj-$(CONFIG_INTEL_IOMMU) += iova.o intel-iommu.o
+obj-$(CONFIG_IPMMU_VMSA) += ipmmu-vmsa.o
 obj-$(CONFIG_IRQ_REMAP) += intel_irq_remapping.o irq_remapping.o
 obj-$(CONFIG_OMAP_IOMMU) += omap-iommu.o
 obj-$(CONFIG_OMAP_IOMMU) += omap-iommu2.o
diff --git a/drivers/iommu/ipmmu-vmsa.c b/drivers/iommu/ipmmu-vmsa.c
new file mode 100644 (file)
index 0000000..b084530
--- /dev/null
@@ -0,0 +1,1070 @@
+/*
+ * IPMMU VMSA
+ *
+ * Copyright (C) 2014 Renesas Electronics Corporation
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License as published by
+ * the Free Software Foundation; version 2 of the License.
+ */
+
+#include <linux/delay.h>
+#include <linux/dma-mapping.h>
+#include <linux/err.h>
+#include <linux/export.h>
+#include <linux/interrupt.h>
+#include <linux/io.h>
+#include <linux/iommu.h>
+#include <linux/module.h>
+#include <linux/platform_data/ipmmu-vmsa.h>
+#include <linux/platform_device.h>
+#include <linux/sizes.h>
+#include <linux/slab.h>
+
+#include <asm/dma-iommu.h>
+#include <asm/pgalloc.h>
+
+struct ipmmu_vmsa_device {
+       struct device *dev;
+       void __iomem *base;
+       struct list_head list;
+
+       const struct ipmmu_vmsa_platform_data *pdata;
+       unsigned int num_utlbs;
+
+       struct dma_iommu_mapping *mapping;
+};
+
+struct ipmmu_vmsa_domain {
+       struct ipmmu_vmsa_device *mmu;
+       struct iommu_domain *io_domain;
+
+       unsigned int context_id;
+       spinlock_t lock;                        /* Protects mappings */
+       pgd_t *pgd;
+};
+
+static DEFINE_SPINLOCK(ipmmu_devices_lock);
+static LIST_HEAD(ipmmu_devices);
+
+#define TLB_LOOP_TIMEOUT               100     /* 100us */
+
+/* -----------------------------------------------------------------------------
+ * Registers Definition
+ */
+
+#define IM_CTX_SIZE                    0x40
+
+#define IMCTR                          0x0000
+#define IMCTR_TRE                      (1 << 17)
+#define IMCTR_AFE                      (1 << 16)
+#define IMCTR_RTSEL_MASK               (3 << 4)
+#define IMCTR_RTSEL_SHIFT              4
+#define IMCTR_TREN                     (1 << 3)
+#define IMCTR_INTEN                    (1 << 2)
+#define IMCTR_FLUSH                    (1 << 1)
+#define IMCTR_MMUEN                    (1 << 0)
+
+#define IMCAAR                         0x0004
+
+#define IMTTBCR                                0x0008
+#define IMTTBCR_EAE                    (1 << 31)
+#define IMTTBCR_PMB                    (1 << 30)
+#define IMTTBCR_SH1_NON_SHAREABLE      (0 << 28)
+#define IMTTBCR_SH1_OUTER_SHAREABLE    (2 << 28)
+#define IMTTBCR_SH1_INNER_SHAREABLE    (3 << 28)
+#define IMTTBCR_SH1_MASK               (3 << 28)
+#define IMTTBCR_ORGN1_NC               (0 << 26)
+#define IMTTBCR_ORGN1_WB_WA            (1 << 26)
+#define IMTTBCR_ORGN1_WT               (2 << 26)
+#define IMTTBCR_ORGN1_WB               (3 << 26)
+#define IMTTBCR_ORGN1_MASK             (3 << 26)
+#define IMTTBCR_IRGN1_NC               (0 << 24)
+#define IMTTBCR_IRGN1_WB_WA            (1 << 24)
+#define IMTTBCR_IRGN1_WT               (2 << 24)
+#define IMTTBCR_IRGN1_WB               (3 << 24)
+#define IMTTBCR_IRGN1_MASK             (3 << 24)
+#define IMTTBCR_TSZ1_MASK              (7 << 16)
+#define IMTTBCR_TSZ1_SHIFT             16
+#define IMTTBCR_SH0_NON_SHAREABLE      (0 << 12)
+#define IMTTBCR_SH0_OUTER_SHAREABLE    (2 << 12)
+#define IMTTBCR_SH0_INNER_SHAREABLE    (3 << 12)
+#define IMTTBCR_SH0_MASK               (3 << 12)
+#define IMTTBCR_ORGN0_NC               (0 << 10)
+#define IMTTBCR_ORGN0_WB_WA            (1 << 10)
+#define IMTTBCR_ORGN0_WT               (2 << 10)
+#define IMTTBCR_ORGN0_WB               (3 << 10)
+#define IMTTBCR_ORGN0_MASK             (3 << 10)
+#define IMTTBCR_IRGN0_NC               (0 << 8)
+#define IMTTBCR_IRGN0_WB_WA            (1 << 8)
+#define IMTTBCR_IRGN0_WT               (2 << 8)
+#define IMTTBCR_IRGN0_WB               (3 << 8)
+#define IMTTBCR_IRGN0_MASK             (3 << 8)
+#define IMTTBCR_SL0_LVL_2              (0 << 4)
+#define IMTTBCR_SL0_LVL_1              (1 << 4)
+#define IMTTBCR_TSZ0_MASK              (7 << 0)
+#define IMTTBCR_TSZ0_SHIFT             O
+
+#define IMBUSCR                                0x000c
+#define IMBUSCR_DVM                    (1 << 2)
+#define IMBUSCR_BUSSEL_SYS             (0 << 0)
+#define IMBUSCR_BUSSEL_CCI             (1 << 0)
+#define IMBUSCR_BUSSEL_IMCAAR          (2 << 0)
+#define IMBUSCR_BUSSEL_CCI_IMCAAR      (3 << 0)
+#define IMBUSCR_BUSSEL_MASK            (3 << 0)
+
+#define IMTTLBR0                       0x0010
+#define IMTTUBR0                       0x0014
+#define IMTTLBR1                       0x0018
+#define IMTTUBR1                       0x001c
+
+#define IMSTR                          0x0020
+#define IMSTR_ERRLVL_MASK              (3 << 12)
+#define IMSTR_ERRLVL_SHIFT             12
+#define IMSTR_ERRCODE_TLB_FORMAT       (1 << 8)
+#define IMSTR_ERRCODE_ACCESS_PERM      (4 << 8)
+#define IMSTR_ERRCODE_SECURE_ACCESS    (5 << 8)
+#define IMSTR_ERRCODE_MASK             (7 << 8)
+#define IMSTR_MHIT                     (1 << 4)
+#define IMSTR_ABORT                    (1 << 2)
+#define IMSTR_PF                       (1 << 1)
+#define IMSTR_TF                       (1 << 0)
+
+#define IMMAIR0                                0x0028
+#define IMMAIR1                                0x002c
+#define IMMAIR_ATTR_MASK               0xff
+#define IMMAIR_ATTR_DEVICE             0x04
+#define IMMAIR_ATTR_NC                 0x44
+#define IMMAIR_ATTR_WBRWA              0xff
+#define IMMAIR_ATTR_SHIFT(n)           ((n) << 3)
+#define IMMAIR_ATTR_IDX_NC             0
+#define IMMAIR_ATTR_IDX_WBRWA          1
+#define IMMAIR_ATTR_IDX_DEV            2
+
+#define IMEAR                          0x0030
+
+#define IMPCTR                         0x0200
+#define IMPSTR                         0x0208
+#define IMPEAR                         0x020c
+#define IMPMBA(n)                      (0x0280 + ((n) * 4))
+#define IMPMBD(n)                      (0x02c0 + ((n) * 4))
+
+#define IMUCTR(n)                      (0x0300 + ((n) * 16))
+#define IMUCTR_FIXADDEN                        (1 << 31)
+#define IMUCTR_FIXADD_MASK             (0xff << 16)
+#define IMUCTR_FIXADD_SHIFT            16
+#define IMUCTR_TTSEL_MMU(n)            ((n) << 4)
+#define IMUCTR_TTSEL_PMB               (8 << 4)
+#define IMUCTR_TTSEL_MASK              (15 << 4)
+#define IMUCTR_FLUSH                   (1 << 1)
+#define IMUCTR_MMUEN                   (1 << 0)
+
+#define IMUASID(n)                     (0x0308 + ((n) * 16))
+#define IMUASID_ASID8_MASK             (0xff << 8)
+#define IMUASID_ASID8_SHIFT            8
+#define IMUASID_ASID0_MASK             (0xff << 0)
+#define IMUASID_ASID0_SHIFT            0
+
+/* -----------------------------------------------------------------------------
+ * Page Table Bits
+ */
+
+/*
+ * VMSA states in section B3.6.3 "Control of Secure or Non-secure memory access,
+ * Long-descriptor format" that the NStable bit being set in a table descriptor
+ * will result in the NStable and NS bits of all child entries being ignored and
+ * considered as being set. The IPMMU seems not to comply with this, as it
+ * generates a secure access page fault if any of the NStable and NS bits isn't
+ * set when running in non-secure mode.
+ */
+#ifndef PMD_NSTABLE
+#define PMD_NSTABLE                    (_AT(pmdval_t, 1) << 63)
+#endif
+
+#define ARM_VMSA_PTE_XN                        (((pteval_t)3) << 53)
+#define ARM_VMSA_PTE_CONT              (((pteval_t)1) << 52)
+#define ARM_VMSA_PTE_AF                        (((pteval_t)1) << 10)
+#define ARM_VMSA_PTE_SH_NS             (((pteval_t)0) << 8)
+#define ARM_VMSA_PTE_SH_OS             (((pteval_t)2) << 8)
+#define ARM_VMSA_PTE_SH_IS             (((pteval_t)3) << 8)
+#define ARM_VMSA_PTE_NS                        (((pteval_t)1) << 5)
+#define ARM_VMSA_PTE_PAGE              (((pteval_t)3) << 0)
+
+/* Stage-1 PTE */
+#define ARM_VMSA_PTE_AP_UNPRIV         (((pteval_t)1) << 6)
+#define ARM_VMSA_PTE_AP_RDONLY         (((pteval_t)2) << 6)
+#define ARM_VMSA_PTE_ATTRINDX_SHIFT    2
+#define ARM_VMSA_PTE_nG                        (((pteval_t)1) << 11)
+
+/* Stage-2 PTE */
+#define ARM_VMSA_PTE_HAP_FAULT         (((pteval_t)0) << 6)
+#define ARM_VMSA_PTE_HAP_READ          (((pteval_t)1) << 6)
+#define ARM_VMSA_PTE_HAP_WRITE         (((pteval_t)2) << 6)
+#define ARM_VMSA_PTE_MEMATTR_OIWB      (((pteval_t)0xf) << 2)
+#define ARM_VMSA_PTE_MEMATTR_NC                (((pteval_t)0x5) << 2)
+#define ARM_VMSA_PTE_MEMATTR_DEV       (((pteval_t)0x1) << 2)
+
+/* -----------------------------------------------------------------------------
+ * Read/Write Access
+ */
+
+static u32 ipmmu_read(struct ipmmu_vmsa_device *mmu, unsigned int offset)
+{
+       return ioread32(mmu->base + offset);
+}
+
+static void ipmmu_write(struct ipmmu_vmsa_device *mmu, unsigned int offset,
+                       u32 data)
+{
+       iowrite32(data, mmu->base + offset);
+}
+
+static u32 ipmmu_ctx_read(struct ipmmu_vmsa_domain *domain, unsigned int reg)
+{
+       return ipmmu_read(domain->mmu, domain->context_id * IM_CTX_SIZE + reg);
+}
+
+static void ipmmu_ctx_write(struct ipmmu_vmsa_domain *domain, unsigned int reg,
+                           u32 data)
+{
+       ipmmu_write(domain->mmu, domain->context_id * IM_CTX_SIZE + reg, data);
+}
+
+/* -----------------------------------------------------------------------------
+ * TLB and microTLB Management
+ */
+
+/* Wait for any pending TLB invalidations to complete */
+static void ipmmu_tlb_sync(struct ipmmu_vmsa_domain *domain)
+{
+       unsigned int count = 0;
+
+       while (ipmmu_ctx_read(domain, IMCTR) & IMCTR_FLUSH) {
+               cpu_relax();
+               if (++count == TLB_LOOP_TIMEOUT) {
+                       dev_err_ratelimited(domain->mmu->dev,
+                       "TLB sync timed out -- MMU may be deadlocked\n");
+                       return;
+               }
+               udelay(1);
+       }
+}
+
+static void ipmmu_tlb_invalidate(struct ipmmu_vmsa_domain *domain)
+{
+       u32 reg;
+
+       reg = ipmmu_ctx_read(domain, IMCTR);
+       reg |= IMCTR_FLUSH;
+       ipmmu_ctx_write(domain, IMCTR, reg);
+
+       ipmmu_tlb_sync(domain);
+}
+
+/*
+ * Enable MMU translation for the microTLB.
+ */
+static void ipmmu_utlb_enable(struct ipmmu_vmsa_domain *domain,
+                             const struct ipmmu_vmsa_master *master)
+{
+       struct ipmmu_vmsa_device *mmu = domain->mmu;
+
+       /* TODO: What should we set the ASID to ? */
+       ipmmu_write(mmu, IMUASID(master->utlb), 0);
+       /* TODO: Do we need to flush the microTLB ? */
+       ipmmu_write(mmu, IMUCTR(master->utlb),
+                   IMUCTR_TTSEL_MMU(domain->context_id) | IMUCTR_FLUSH |
+                   IMUCTR_MMUEN);
+}
+
+/*
+ * Disable MMU translation for the microTLB.
+ */
+static void ipmmu_utlb_disable(struct ipmmu_vmsa_domain *domain,
+                              const struct ipmmu_vmsa_master *master)
+{
+       struct ipmmu_vmsa_device *mmu = domain->mmu;
+
+       ipmmu_write(mmu, IMUCTR(master->utlb), 0);
+}
+
+static void ipmmu_flush_pgtable(struct ipmmu_vmsa_device *mmu, void *addr,
+                               size_t size)
+{
+       unsigned long offset = (unsigned long)addr & ~PAGE_MASK;
+
+       /*
+        * TODO: Add support for coherent walk through CCI with DVM and remove
+        * cache handling.
+        */
+       dma_map_page(mmu->dev, virt_to_page(addr), offset, size, DMA_TO_DEVICE);
+}
+
+/* -----------------------------------------------------------------------------
+ * Domain/Context Management
+ */
+
+static int ipmmu_domain_init_context(struct ipmmu_vmsa_domain *domain)
+{
+       phys_addr_t ttbr;
+       u32 reg;
+
+       /*
+        * TODO: When adding support for multiple contexts, find an unused
+        * context.
+        */
+       domain->context_id = 0;
+
+       /* TTBR0 */
+       ipmmu_flush_pgtable(domain->mmu, domain->pgd,
+                           PTRS_PER_PGD * sizeof(*domain->pgd));
+       ttbr = __pa(domain->pgd);
+       ipmmu_ctx_write(domain, IMTTLBR0, ttbr);
+       ipmmu_ctx_write(domain, IMTTUBR0, ttbr >> 32);
+
+       /*
+        * TTBCR
+        * We use long descriptors with inner-shareable WBWA tables and allocate
+        * the whole 32-bit VA space to TTBR0.
+        */
+       ipmmu_ctx_write(domain, IMTTBCR, IMTTBCR_EAE |
+                       IMTTBCR_SH0_INNER_SHAREABLE | IMTTBCR_ORGN0_WB_WA |
+                       IMTTBCR_IRGN0_WB_WA | IMTTBCR_SL0_LVL_1);
+
+       /*
+        * MAIR0
+        * We need three attributes only, non-cacheable, write-back read/write
+        * allocate and device memory.
+        */
+       reg = (IMMAIR_ATTR_NC << IMMAIR_ATTR_SHIFT(IMMAIR_ATTR_IDX_NC))
+           | (IMMAIR_ATTR_WBRWA << IMMAIR_ATTR_SHIFT(IMMAIR_ATTR_IDX_WBRWA))
+           | (IMMAIR_ATTR_DEVICE << IMMAIR_ATTR_SHIFT(IMMAIR_ATTR_IDX_DEV));
+       ipmmu_ctx_write(domain, IMMAIR0, reg);
+
+       /* IMBUSCR */
+       ipmmu_ctx_write(domain, IMBUSCR,
+                       ipmmu_ctx_read(domain, IMBUSCR) &
+                       ~(IMBUSCR_DVM | IMBUSCR_BUSSEL_MASK));
+
+       /*
+        * IMSTR
+        * Clear all interrupt flags.
+        */
+       ipmmu_ctx_write(domain, IMSTR, ipmmu_ctx_read(domain, IMSTR));
+
+       /*
+        * IMCTR
+        * Enable the MMU and interrupt generation. The long-descriptor
+        * translation table format doesn't use TEX remapping. Don't enable AF
+        * software management as we have no use for it. Flush the TLB as
+        * required when modifying the context registers.
+        */
+       ipmmu_ctx_write(domain, IMCTR, IMCTR_INTEN | IMCTR_FLUSH | IMCTR_MMUEN);
+
+       return 0;
+}
+
+static void ipmmu_domain_destroy_context(struct ipmmu_vmsa_domain *domain)
+{
+       /*
+        * Disable the context. Flush the TLB as required when modifying the
+        * context registers.
+        *
+        * TODO: Is TLB flush really needed ?
+        */
+       ipmmu_ctx_write(domain, IMCTR, IMCTR_FLUSH);
+       ipmmu_tlb_sync(domain);
+}
+
+/* -----------------------------------------------------------------------------
+ * Fault Handling
+ */
+
+static irqreturn_t ipmmu_domain_irq(struct ipmmu_vmsa_domain *domain)
+{
+       const u32 err_mask = IMSTR_MHIT | IMSTR_ABORT | IMSTR_PF | IMSTR_TF;
+       struct ipmmu_vmsa_device *mmu = domain->mmu;
+       u32 status;
+       u32 iova;
+
+       status = ipmmu_ctx_read(domain, IMSTR);
+       if (!(status & err_mask))
+               return IRQ_NONE;
+
+       iova = ipmmu_ctx_read(domain, IMEAR);
+
+       /*
+        * Clear the error status flags. Unlike traditional interrupt flag
+        * registers that must be cleared by writing 1, this status register
+        * seems to require 0. The error address register must be read before,
+        * otherwise its value will be 0.
+        */
+       ipmmu_ctx_write(domain, IMSTR, 0);
+
+       /* Log fatal errors. */
+       if (status & IMSTR_MHIT)
+               dev_err_ratelimited(mmu->dev, "Multiple TLB hits @0x%08x\n",
+                                   iova);
+       if (status & IMSTR_ABORT)
+               dev_err_ratelimited(mmu->dev, "Page Table Walk Abort @0x%08x\n",
+                                   iova);
+
+       if (!(status & (IMSTR_PF | IMSTR_TF)))
+               return IRQ_NONE;
+
+       /*
+        * Try to handle page faults and translation faults.
+        *
+        * TODO: We need to look up the faulty device based on the I/O VA. Use
+        * the IOMMU device for now.
+        */
+       if (!report_iommu_fault(domain->io_domain, mmu->dev, iova, 0))
+               return IRQ_HANDLED;
+
+       dev_err_ratelimited(mmu->dev,
+                           "Unhandled fault: status 0x%08x iova 0x%08x\n",
+                           status, iova);
+
+       return IRQ_HANDLED;
+}
+
+static irqreturn_t ipmmu_irq(int irq, void *dev)
+{
+       struct ipmmu_vmsa_device *mmu = dev;
+       struct iommu_domain *io_domain;
+       struct ipmmu_vmsa_domain *domain;
+
+       if (!mmu->mapping)
+               return IRQ_NONE;
+
+       io_domain = mmu->mapping->domain;
+       domain = io_domain->priv;
+
+       return ipmmu_domain_irq(domain);
+}
+
+/* -----------------------------------------------------------------------------
+ * Page Table Management
+ */
+
+static void ipmmu_free_ptes(pmd_t *pmd)
+{
+       pgtable_t table = pmd_pgtable(*pmd);
+       __free_page(table);
+}
+
+static void ipmmu_free_pmds(pud_t *pud)
+{
+       pmd_t *pmd, *pmd_base = pmd_offset(pud, 0);
+       unsigned int i;
+
+       pmd = pmd_base;
+       for (i = 0; i < PTRS_PER_PMD; ++i) {
+               if (pmd_none(*pmd))
+                       continue;
+
+               ipmmu_free_ptes(pmd);
+               pmd++;
+       }
+
+       pmd_free(NULL, pmd_base);
+}
+
+static void ipmmu_free_puds(pgd_t *pgd)
+{
+       pud_t *pud, *pud_base = pud_offset(pgd, 0);
+       unsigned int i;
+
+       pud = pud_base;
+       for (i = 0; i < PTRS_PER_PUD; ++i) {
+               if (pud_none(*pud))
+                       continue;
+
+               ipmmu_free_pmds(pud);
+               pud++;
+       }
+
+       pud_free(NULL, pud_base);
+}
+
+static void ipmmu_free_pgtables(struct ipmmu_vmsa_domain *domain)
+{
+       pgd_t *pgd, *pgd_base = domain->pgd;
+       unsigned int i;
+
+       /*
+        * Recursively free the page tables for this domain. We don't care about
+        * speculative TLB filling, because the TLB will be nuked next time this
+        * context bank is re-allocated and no devices currently map to these
+        * tables.
+        */
+       pgd = pgd_base;
+       for (i = 0; i < PTRS_PER_PGD; ++i) {
+               if (pgd_none(*pgd))
+                       continue;
+               ipmmu_free_puds(pgd);
+               pgd++;
+       }
+
+       kfree(pgd_base);
+}
+
+/*
+ * We can't use the (pgd|pud|pmd|pte)_populate or the set_(pgd|pud|pmd|pte)
+ * functions as they would flush the CPU TLB.
+ */
+
+static int ipmmu_alloc_init_pte(struct ipmmu_vmsa_device *mmu, pmd_t *pmd,
+                               unsigned long addr, unsigned long end,
+                               phys_addr_t phys, int prot)
+{
+       unsigned long pfn = __phys_to_pfn(phys);
+       pteval_t pteval = ARM_VMSA_PTE_PAGE | ARM_VMSA_PTE_NS | ARM_VMSA_PTE_AF
+                       | ARM_VMSA_PTE_XN;
+       pte_t *pte, *start;
+
+       if (pmd_none(*pmd)) {
+               /* Allocate a new set of tables */
+               pte = (pte_t *)get_zeroed_page(GFP_ATOMIC);
+               if (!pte)
+                       return -ENOMEM;
+
+               ipmmu_flush_pgtable(mmu, pte, PAGE_SIZE);
+               *pmd = __pmd(__pa(pte) | PMD_NSTABLE | PMD_TYPE_TABLE);
+               ipmmu_flush_pgtable(mmu, pmd, sizeof(*pmd));
+
+               pte += pte_index(addr);
+       } else
+               pte = pte_offset_kernel(pmd, addr);
+
+       pteval |= ARM_VMSA_PTE_AP_UNPRIV | ARM_VMSA_PTE_nG;
+       if (!(prot & IOMMU_WRITE) && (prot & IOMMU_READ))
+               pteval |= ARM_VMSA_PTE_AP_RDONLY;
+
+       if (prot & IOMMU_CACHE)
+               pteval |= (IMMAIR_ATTR_IDX_WBRWA <<
+                          ARM_VMSA_PTE_ATTRINDX_SHIFT);
+
+       /* If no access, create a faulting entry to avoid TLB fills */
+       if (prot & IOMMU_EXEC)
+               pteval &= ~ARM_VMSA_PTE_XN;
+       else if (!(prot & (IOMMU_READ | IOMMU_WRITE)))
+               pteval &= ~ARM_VMSA_PTE_PAGE;
+
+       pteval |= ARM_VMSA_PTE_SH_IS;
+       start = pte;
+
+       /* Install the page table entries. */
+       do {
+               *pte++ = pfn_pte(pfn++, __pgprot(pteval));
+               addr += PAGE_SIZE;
+       } while (addr != end);
+
+       ipmmu_flush_pgtable(mmu, start, sizeof(*pte) * (pte - start));
+       return 0;
+}
+
+static int ipmmu_alloc_init_pmd(struct ipmmu_vmsa_device *mmu, pud_t *pud,
+                               unsigned long addr, unsigned long end,
+                               phys_addr_t phys, int prot)
+{
+       unsigned long next;
+       pmd_t *pmd;
+       int ret;
+
+#ifndef __PAGETABLE_PMD_FOLDED
+       if (pud_none(*pud)) {
+               pmd = (pmd_t *)get_zeroed_page(GFP_ATOMIC);
+               if (!pmd)
+                       return -ENOMEM;
+
+               ipmmu_flush_pgtable(mmu, pmd, PAGE_SIZE);
+               *pud = __pud(__pa(pmd) | PMD_NSTABLE | PMD_TYPE_TABLE);
+               ipmmu_flush_pgtable(mmu, pud, sizeof(*pud));
+
+               pmd += pmd_index(addr);
+       } else
+#endif
+               pmd = pmd_offset(pud, addr);
+
+       do {
+               next = pmd_addr_end(addr, end);
+               ret = ipmmu_alloc_init_pte(mmu, pmd, addr, end, phys, prot);
+               phys += next - addr;
+       } while (pmd++, addr = next, addr < end);
+
+       return ret;
+}
+
+static int ipmmu_alloc_init_pud(struct ipmmu_vmsa_device *mmu, pgd_t *pgd,
+                               unsigned long addr, unsigned long end,
+                               phys_addr_t phys, int prot)
+{
+       unsigned long next;
+       pud_t *pud;
+       int ret;
+
+#ifndef __PAGETABLE_PUD_FOLDED
+       if (pgd_none(*pgd)) {
+               pud = (pud_t *)get_zeroed_page(GFP_ATOMIC);
+               if (!pud)
+                       return -ENOMEM;
+
+               ipmmu_flush_pgtable(mmu, pud, PAGE_SIZE);
+               *pgd = __pgd(__pa(pud) | PMD_NSTABLE | PMD_TYPE_TABLE);
+               ipmmu_flush_pgtable(mmu, pgd, sizeof(*pgd));
+
+               pud += pud_index(addr);
+       } else
+#endif
+               pud = pud_offset(pgd, addr);
+
+       do {
+               next = pud_addr_end(addr, end);
+               ret = ipmmu_alloc_init_pmd(mmu, pud, addr, next, phys, prot);
+               phys += next - addr;
+       } while (pud++, addr = next, addr < end);
+
+       return ret;
+}
+
+static int ipmmu_handle_mapping(struct ipmmu_vmsa_domain *domain,
+                               unsigned long iova, phys_addr_t paddr,
+                               size_t size, int prot)
+{
+       struct ipmmu_vmsa_device *mmu = domain->mmu;
+       pgd_t *pgd = domain->pgd;
+       unsigned long flags;
+       unsigned long end;
+       int ret;
+
+       if (!pgd)
+               return -EINVAL;
+
+       if (size & ~PAGE_MASK)
+               return -EINVAL;
+
+       if (paddr & ~((1ULL << 40) - 1))
+               return -ERANGE;
+
+       spin_lock_irqsave(&domain->lock, flags);
+
+       pgd += pgd_index(iova);
+       end = iova + size;
+
+       do {
+               unsigned long next = pgd_addr_end(iova, end);
+
+               ret = ipmmu_alloc_init_pud(mmu, pgd, iova, next, paddr, prot);
+               if (ret)
+                       break;
+
+               paddr += next - iova;
+               iova = next;
+       } while (pgd++, iova != end);
+
+       spin_unlock_irqrestore(&domain->lock, flags);
+
+       ipmmu_tlb_invalidate(domain);
+
+       return ret;
+}
+
+/* -----------------------------------------------------------------------------
+ * IOMMU Operations
+ */
+
+static const struct ipmmu_vmsa_master *
+ipmmu_find_master(struct ipmmu_vmsa_device *ipmmu, struct device *dev)
+{
+       const struct ipmmu_vmsa_master *master = ipmmu->pdata->masters;
+       const char *devname = dev_name(dev);
+       unsigned int i;
+
+       for (i = 0; i < ipmmu->pdata->num_masters; ++i, ++master) {
+               if (strcmp(master->name, devname) == 0)
+                       return master;
+       }
+
+       return NULL;
+}
+
+static int ipmmu_domain_init(struct iommu_domain *io_domain)
+{
+       struct ipmmu_vmsa_domain *domain;
+
+       domain = kzalloc(sizeof(*domain), GFP_KERNEL);
+       if (!domain)
+               return -ENOMEM;
+
+       spin_lock_init(&domain->lock);
+
+       domain->pgd = kzalloc(PTRS_PER_PGD * sizeof(pgd_t), GFP_KERNEL);
+       if (!domain->pgd) {
+               kfree(domain);
+               return -ENOMEM;
+       }
+
+       io_domain->priv = domain;
+       domain->io_domain = io_domain;
+
+       return 0;
+}
+
+static void ipmmu_domain_destroy(struct iommu_domain *io_domain)
+{
+       struct ipmmu_vmsa_domain *domain = io_domain->priv;
+
+       /*
+        * Free the domain resources. We assume that all devices have already
+        * been detached.
+        */
+       ipmmu_domain_destroy_context(domain);
+       ipmmu_free_pgtables(domain);
+       kfree(domain);
+}
+
+static int ipmmu_attach_device(struct iommu_domain *io_domain,
+                              struct device *dev)
+{
+       struct ipmmu_vmsa_device *mmu = dev->archdata.iommu;
+       struct ipmmu_vmsa_domain *domain = io_domain->priv;
+       const struct ipmmu_vmsa_master *master;
+       unsigned long flags;
+       int ret = 0;
+
+       if (!mmu) {
+               dev_err(dev, "Cannot attach to IPMMU\n");
+               return -ENXIO;
+       }
+
+       spin_lock_irqsave(&domain->lock, flags);
+
+       if (!domain->mmu) {
+               /* The domain hasn't been used yet, initialize it. */
+               domain->mmu = mmu;
+               ret = ipmmu_domain_init_context(domain);
+       } else if (domain->mmu != mmu) {
+               /*
+                * Something is wrong, we can't attach two devices using
+                * different IOMMUs to the same domain.
+                */
+               dev_err(dev, "Can't attach IPMMU %s to domain on IPMMU %s\n",
+                       dev_name(mmu->dev), dev_name(domain->mmu->dev));
+               ret = -EINVAL;
+       }
+
+       spin_unlock_irqrestore(&domain->lock, flags);
+
+       if (ret < 0)
+               return ret;
+
+       master = ipmmu_find_master(mmu, dev);
+       if (!master)
+               return -EINVAL;
+
+       ipmmu_utlb_enable(domain, master);
+
+       return 0;
+}
+
+static void ipmmu_detach_device(struct iommu_domain *io_domain,
+                               struct device *dev)
+{
+       struct ipmmu_vmsa_domain *domain = io_domain->priv;
+       const struct ipmmu_vmsa_master *master;
+
+       master = ipmmu_find_master(domain->mmu, dev);
+       if (!master)
+               return;
+
+       ipmmu_utlb_disable(domain, master);
+
+       /*
+        * TODO: Optimize by disabling the context when no device is attached.
+        */
+}
+
+static int ipmmu_map(struct iommu_domain *io_domain, unsigned long iova,
+                    phys_addr_t paddr, size_t size, int prot)
+{
+       struct ipmmu_vmsa_domain *domain = io_domain->priv;
+
+       if (!domain)
+               return -ENODEV;
+
+       return ipmmu_handle_mapping(domain, iova, paddr, size, prot);
+}
+
+static size_t ipmmu_unmap(struct iommu_domain *io_domain, unsigned long iova,
+                         size_t size)
+{
+       struct ipmmu_vmsa_domain *domain = io_domain->priv;
+       int ret;
+
+       ret = ipmmu_handle_mapping(domain, iova, 0, size, 0);
+       return ret ? 0 : size;
+}
+
+static phys_addr_t ipmmu_iova_to_phys(struct iommu_domain *io_domain,
+                                     dma_addr_t iova)
+{
+       struct ipmmu_vmsa_domain *domain = io_domain->priv;
+       pgd_t pgd;
+       pud_t pud;
+       pmd_t pmd;
+       pte_t pte;
+
+       /* TODO: Is locking needed ? */
+
+       if (!domain->pgd)
+               return 0;
+
+       pgd = *(domain->pgd + pgd_index(iova));
+       if (pgd_none(pgd))
+               return 0;
+
+       pud = *pud_offset(&pgd, iova);
+       if (pud_none(pud))
+               return 0;
+
+       pmd = *pmd_offset(&pud, iova);
+       if (pmd_none(pmd))
+               return 0;
+
+       pte = *(pmd_page_vaddr(pmd) + pte_index(iova));
+       if (pte_none(pte))
+               return 0;
+
+       return __pfn_to_phys(pte_pfn(pte)) | (iova & ~PAGE_MASK);
+}
+
+static int ipmmu_add_device(struct device *dev)
+{
+       const struct ipmmu_vmsa_master *master = NULL;
+       struct ipmmu_vmsa_device *mmu;
+       struct iommu_group *group;
+       int ret;
+
+       if (dev->archdata.iommu) {
+               dev_warn(dev, "IOMMU driver already assigned to device %s\n",
+                        dev_name(dev));
+               return -EINVAL;
+       }
+
+       /* Find the master corresponding to the device. */
+       spin_lock(&ipmmu_devices_lock);
+
+       list_for_each_entry(mmu, &ipmmu_devices, list) {
+               master = ipmmu_find_master(mmu, dev);
+               if (master) {
+                       /*
+                        * TODO Take a reference to the master to protect
+                        * against device removal.
+                        */
+                       break;
+               }
+       }
+
+       spin_unlock(&ipmmu_devices_lock);
+
+       if (!master)
+               return -ENODEV;
+
+       if (!master->utlb >= mmu->num_utlbs)
+               return -EINVAL;
+
+       /* Create a device group and add the device to it. */
+       group = iommu_group_alloc();
+       if (IS_ERR(group)) {
+               dev_err(dev, "Failed to allocate IOMMU group\n");
+               return PTR_ERR(group);
+       }
+
+       ret = iommu_group_add_device(group, dev);
+       iommu_group_put(group);
+
+       if (ret < 0) {
+               dev_err(dev, "Failed to add device to IPMMU group\n");
+               return ret;
+       }
+
+       dev->archdata.iommu = mmu;
+
+       /*
+        * Create the ARM mapping, used by the ARM DMA mapping core to allocate
+        * VAs. This will allocate a corresponding IOMMU domain.
+        *
+        * TODO:
+        * - Create one mapping per context (TLB).
+        * - Make the mapping size configurable ? We currently use a 2GB mapping
+        *   at a 1GB offset to ensure that NULL VAs will fault.
+        */
+       if (!mmu->mapping) {
+               struct dma_iommu_mapping *mapping;
+
+               mapping = arm_iommu_create_mapping(&platform_bus_type,
+                                                       SZ_1G, SZ_2G, 0);
+               if (IS_ERR(mapping)) {
+                       dev_err(mmu->dev, "failed to create ARM IOMMU mapping\n");
+                       return PTR_ERR(mapping);
+               }
+
+               mmu->mapping = mapping;
+       }
+
+       /* Attach the ARM VA mapping to the device. */
+       ret = arm_iommu_attach_device(dev, mmu->mapping);
+       if (ret < 0) {
+               dev_err(dev, "Failed to attach device to VA mapping\n");
+               goto error;
+       }
+
+       return 0;
+
+error:
+       dev->archdata.iommu = NULL;
+       iommu_group_remove_device(dev);
+       return ret;
+}
+
+static void ipmmu_remove_device(struct device *dev)
+{
+       arm_iommu_detach_device(dev);
+       iommu_group_remove_device(dev);
+       dev->archdata.iommu = NULL;
+}
+
+static struct iommu_ops ipmmu_ops = {
+       .domain_init = ipmmu_domain_init,
+       .domain_destroy = ipmmu_domain_destroy,
+       .attach_dev = ipmmu_attach_device,
+       .detach_dev = ipmmu_detach_device,
+       .map = ipmmu_map,
+       .unmap = ipmmu_unmap,
+       .iova_to_phys = ipmmu_iova_to_phys,
+       .add_device = ipmmu_add_device,
+       .remove_device = ipmmu_remove_device,
+       .pgsize_bitmap = SZ_1M | SZ_64K | SZ_4K,
+};
+
+/* -----------------------------------------------------------------------------
+ * Probe/remove and init
+ */
+
+static void ipmmu_device_reset(struct ipmmu_vmsa_device *mmu)
+{
+       unsigned int i;
+
+       /* Disable all contexts. */
+       for (i = 0; i < 4; ++i)
+               ipmmu_write(mmu, i * IM_CTX_SIZE + IMCTR, 0);
+}
+
+static int ipmmu_probe(struct platform_device *pdev)
+{
+       struct ipmmu_vmsa_device *mmu;
+       struct resource *res;
+       int irq;
+       int ret;
+
+       if (!pdev->dev.platform_data) {
+               dev_err(&pdev->dev, "missing platform data\n");
+               return -EINVAL;
+       }
+
+       mmu = devm_kzalloc(&pdev->dev, sizeof(*mmu), GFP_KERNEL);
+       if (!mmu) {
+               dev_err(&pdev->dev, "cannot allocate device data\n");
+               return -ENOMEM;
+       }
+
+       mmu->dev = &pdev->dev;
+       mmu->pdata = pdev->dev.platform_data;
+       mmu->num_utlbs = 32;
+
+       /* Map I/O memory and request IRQ. */
+       res = platform_get_resource(pdev, IORESOURCE_MEM, 0);
+       mmu->base = devm_ioremap_resource(&pdev->dev, res);
+       if (IS_ERR(mmu->base))
+               return PTR_ERR(mmu->base);
+
+       irq = platform_get_irq(pdev, 0);
+       if (irq < 0) {
+               dev_err(&pdev->dev, "no IRQ found\n");
+               return irq;
+       }
+
+       ret = devm_request_irq(&pdev->dev, irq, ipmmu_irq, 0,
+                              dev_name(&pdev->dev), mmu);
+       if (ret < 0) {
+               dev_err(&pdev->dev, "failed to request IRQ %d\n", irq);
+               return irq;
+       }
+
+       ipmmu_device_reset(mmu);
+
+       /*
+        * We can't create the ARM mapping here as it requires the bus to have
+        * an IOMMU, which only happens when bus_set_iommu() is called in
+        * ipmmu_init() after the probe function returns.
+        */
+
+       spin_lock(&ipmmu_devices_lock);
+       list_add(&mmu->list, &ipmmu_devices);
+       spin_unlock(&ipmmu_devices_lock);
+
+       platform_set_drvdata(pdev, mmu);
+
+       return 0;
+}
+
+static int ipmmu_remove(struct platform_device *pdev)
+{
+       struct ipmmu_vmsa_device *mmu = platform_get_drvdata(pdev);
+
+       spin_lock(&ipmmu_devices_lock);
+       list_del(&mmu->list);
+       spin_unlock(&ipmmu_devices_lock);
+
+       arm_iommu_release_mapping(mmu->mapping);
+
+       ipmmu_device_reset(mmu);
+
+       return 0;
+}
+
+static struct platform_driver ipmmu_driver = {
+       .driver = {
+               .owner = THIS_MODULE,
+               .name = "ipmmu-vmsa",
+       },
+       .probe = ipmmu_probe,
+       .remove = ipmmu_remove,
+};
+
+static int __init ipmmu_init(void)
+{
+       int ret;
+
+       ret = platform_driver_register(&ipmmu_driver);
+       if (ret < 0)
+               return ret;
+
+       if (!iommu_present(&platform_bus_type))
+               bus_set_iommu(&platform_bus_type, &ipmmu_ops);
+
+       return 0;
+}
+
+static void __exit ipmmu_exit(void)
+{
+       return platform_driver_unregister(&ipmmu_driver);
+}
+
+subsys_initcall(ipmmu_init);
+module_exit(ipmmu_exit);
+
+MODULE_DESCRIPTION("IOMMU API for Renesas VMSA-compatible IPMMU");
+MODULE_AUTHOR("Laurent Pinchart <laurent.pinchart@ideasonboard.com>");
+MODULE_LICENSE("GPL v2");
diff --git a/include/linux/platform_data/ipmmu-vmsa.h b/include/linux/platform_data/ipmmu-vmsa.h
new file mode 100644 (file)
index 0000000..5275b3a
--- /dev/null
@@ -0,0 +1,24 @@
+/*
+ * IPMMU VMSA Platform Data
+ *
+ * Copyright (C) 2014 Renesas Electronics Corporation
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License as published by
+ * the Free Software Foundation; version 2 of the License.
+ */
+
+#ifndef __IPMMU_VMSA_H__
+#define __IPMMU_VMSA_H__
+
+struct ipmmu_vmsa_master {
+       const char *name;
+       unsigned int utlb;
+};
+
+struct ipmmu_vmsa_platform_data {
+       const struct ipmmu_vmsa_master *masters;
+       unsigned int num_masters;
+};
+
+#endif /* __IPMMU_VMSA_H__ */