From a9dcad5e375800fcb07f7617dba23b3aade8f09d Mon Sep 17 00:00:00 2001 From: Hiroshi DOYU Date: Mon, 26 Jan 2009 15:13:40 +0200 Subject: [PATCH 1/7] omap iommu: tlb and pagetable primitives This patch provides: - iotlb_*() : iommu tlb operations - iopgtable_*() : iommu pagetable(twl) operations - iommu_*() : the other generic operations and the entry points to register and acquire iommu object. Signed-off-by: Hiroshi DOYU --- arch/arm/plat-omap/include/mach/iommu.h | 168 ++++ arch/arm/plat-omap/iommu.c | 996 ++++++++++++++++++++++++ arch/arm/plat-omap/iopgtable.h | 72 ++ 3 files changed, 1236 insertions(+) create mode 100644 arch/arm/plat-omap/include/mach/iommu.h create mode 100644 arch/arm/plat-omap/iommu.c create mode 100644 arch/arm/plat-omap/iopgtable.h diff --git a/arch/arm/plat-omap/include/mach/iommu.h b/arch/arm/plat-omap/include/mach/iommu.h new file mode 100644 index 000000000000..769b00b4c34a --- /dev/null +++ b/arch/arm/plat-omap/include/mach/iommu.h @@ -0,0 +1,168 @@ +/* + * omap iommu: main structures + * + * Copyright (C) 2008-2009 Nokia Corporation + * + * Written by Hiroshi DOYU + * + * 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. + */ + +#ifndef __MACH_IOMMU_H +#define __MACH_IOMMU_H + +struct iotlb_entry { + u32 da; + u32 pa; + u32 pgsz, prsvd, valid; + union { + u16 ap; + struct { + u32 endian, elsz, mixed; + }; + }; +}; + +struct iommu { + const char *name; + struct module *owner; + struct clk *clk; + void __iomem *regbase; + struct device *dev; + + unsigned int refcount; + struct mutex iommu_lock; /* global for this whole object */ + + /* + * We don't change iopgd for a situation like pgd for a task, + * but share it globally for each iommu. + */ + u32 *iopgd; + spinlock_t page_table_lock; /* protect iopgd */ + + int nr_tlb_entries; + + struct list_head mmap; + struct mutex mmap_lock; /* protect mmap */ + + int (*isr)(struct iommu *obj); + + void *ctx; /* iommu context: registres saved area */ +}; + +struct cr_regs { + union { + struct { + u16 cam_l; + u16 cam_h; + }; + u32 cam; + }; + union { + struct { + u16 ram_l; + u16 ram_h; + }; + u32 ram; + }; +}; + +struct iotlb_lock { + short base; + short vict; +}; + +/* architecture specific functions */ +struct iommu_functions { + unsigned long version; + + int (*enable)(struct iommu *obj); + void (*disable)(struct iommu *obj); + u32 (*fault_isr)(struct iommu *obj, u32 *ra); + + void (*tlb_read_cr)(struct iommu *obj, struct cr_regs *cr); + void (*tlb_load_cr)(struct iommu *obj, struct cr_regs *cr); + + struct cr_regs *(*alloc_cr)(struct iommu *obj, struct iotlb_entry *e); + int (*cr_valid)(struct cr_regs *cr); + u32 (*cr_to_virt)(struct cr_regs *cr); + void (*cr_to_e)(struct cr_regs *cr, struct iotlb_entry *e); + ssize_t (*dump_cr)(struct iommu *obj, struct cr_regs *cr, char *buf); + + u32 (*get_pte_attr)(struct iotlb_entry *e); + + void (*save_ctx)(struct iommu *obj); + void (*restore_ctx)(struct iommu *obj); + ssize_t (*dump_ctx)(struct iommu *obj, char *buf); +}; + +struct iommu_platform_data { + const char *name; + const char *clk_name; + const int nr_tlb_entries; +}; + +#if defined(CONFIG_ARCH_OMAP1) +#error "iommu for this processor not implemented yet" +#else +#include +#endif + +/* + * utilities for super page(16MB, 1MB, 64KB and 4KB) + */ + +#define iopgsz_max(bytes) \ + (((bytes) >= SZ_16M) ? SZ_16M : \ + ((bytes) >= SZ_1M) ? SZ_1M : \ + ((bytes) >= SZ_64K) ? SZ_64K : \ + ((bytes) >= SZ_4K) ? SZ_4K : 0) + +#define bytes_to_iopgsz(bytes) \ + (((bytes) == SZ_16M) ? MMU_CAM_PGSZ_16M : \ + ((bytes) == SZ_1M) ? MMU_CAM_PGSZ_1M : \ + ((bytes) == SZ_64K) ? MMU_CAM_PGSZ_64K : \ + ((bytes) == SZ_4K) ? MMU_CAM_PGSZ_4K : -1) + +#define iopgsz_to_bytes(iopgsz) \ + (((iopgsz) == MMU_CAM_PGSZ_16M) ? SZ_16M : \ + ((iopgsz) == MMU_CAM_PGSZ_1M) ? SZ_1M : \ + ((iopgsz) == MMU_CAM_PGSZ_64K) ? SZ_64K : \ + ((iopgsz) == MMU_CAM_PGSZ_4K) ? SZ_4K : 0) + +#define iopgsz_ok(bytes) (bytes_to_iopgsz(bytes) >= 0) + +/* + * global functions + */ +extern u32 iommu_arch_version(void); + +extern void iotlb_cr_to_e(struct cr_regs *cr, struct iotlb_entry *e); +extern u32 iotlb_cr_to_virt(struct cr_regs *cr); + +extern int load_iotlb_entry(struct iommu *obj, struct iotlb_entry *e); +extern void flush_iotlb_page(struct iommu *obj, u32 da); +extern void flush_iotlb_range(struct iommu *obj, u32 start, u32 end); +extern void flush_iotlb_all(struct iommu *obj); + +extern int iopgtable_store_entry(struct iommu *obj, struct iotlb_entry *e); +extern size_t iopgtable_clear_entry(struct iommu *obj, u32 iova); + +extern struct iommu *iommu_get(const char *name); +extern void iommu_put(struct iommu *obj); + +extern void iommu_save_ctx(struct iommu *obj); +extern void iommu_restore_ctx(struct iommu *obj); + +extern int install_iommu_arch(const struct iommu_functions *ops); +extern void uninstall_iommu_arch(const struct iommu_functions *ops); + +extern int foreach_iommu_device(void *data, + int (*fn)(struct device *, void *)); + +extern ssize_t iommu_dump_ctx(struct iommu *obj, char *buf); +extern size_t dump_tlb_entries(struct iommu *obj, char *buf); + +#endif /* __MACH_IOMMU_H */ diff --git a/arch/arm/plat-omap/iommu.c b/arch/arm/plat-omap/iommu.c new file mode 100644 index 000000000000..4cf449fa2cb5 --- /dev/null +++ b/arch/arm/plat-omap/iommu.c @@ -0,0 +1,996 @@ +/* + * omap iommu: tlb and pagetable primitives + * + * Copyright (C) 2008-2009 Nokia Corporation + * + * Written by Hiroshi DOYU , + * Paul Mundt and Toshihiro Kobayashi + * + * 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 "iopgtable.h" + +/* accommodate the difference between omap1 and omap2/3 */ +static const struct iommu_functions *arch_iommu; + +static struct platform_driver omap_iommu_driver; +static struct kmem_cache *iopte_cachep; + +/** + * install_iommu_arch - Install archtecure specific iommu functions + * @ops: a pointer to architecture specific iommu functions + * + * There are several kind of iommu algorithm(tlb, pagetable) among + * omap series. This interface installs such an iommu algorighm. + **/ +int install_iommu_arch(const struct iommu_functions *ops) +{ + if (arch_iommu) + return -EBUSY; + + arch_iommu = ops; + return 0; +} +EXPORT_SYMBOL_GPL(install_iommu_arch); + +/** + * uninstall_iommu_arch - Uninstall archtecure specific iommu functions + * @ops: a pointer to architecture specific iommu functions + * + * This interface uninstalls the iommu algorighm installed previously. + **/ +void uninstall_iommu_arch(const struct iommu_functions *ops) +{ + if (arch_iommu != ops) + pr_err("%s: not your arch\n", __func__); + + arch_iommu = NULL; +} +EXPORT_SYMBOL_GPL(uninstall_iommu_arch); + +/** + * iommu_save_ctx - Save registers for pm off-mode support + * @obj: target iommu + **/ +void iommu_save_ctx(struct iommu *obj) +{ + arch_iommu->save_ctx(obj); +} +EXPORT_SYMBOL_GPL(iommu_save_ctx); + +/** + * iommu_restore_ctx - Restore registers for pm off-mode support + * @obj: target iommu + **/ +void iommu_restore_ctx(struct iommu *obj) +{ + arch_iommu->restore_ctx(obj); +} +EXPORT_SYMBOL_GPL(iommu_restore_ctx); + +/** + * iommu_arch_version - Return running iommu arch version + **/ +u32 iommu_arch_version(void) +{ + return arch_iommu->version; +} +EXPORT_SYMBOL_GPL(iommu_arch_version); + +static int iommu_enable(struct iommu *obj) +{ + int err; + + if (!obj) + return -EINVAL; + + clk_enable(obj->clk); + + err = arch_iommu->enable(obj); + + clk_disable(obj->clk); + return err; +} + +static void iommu_disable(struct iommu *obj) +{ + if (!obj) + return; + + clk_enable(obj->clk); + + arch_iommu->disable(obj); + + clk_disable(obj->clk); +} + +/* + * TLB operations + */ +void iotlb_cr_to_e(struct cr_regs *cr, struct iotlb_entry *e) +{ + BUG_ON(!cr || !e); + + arch_iommu->cr_to_e(cr, e); +} +EXPORT_SYMBOL_GPL(iotlb_cr_to_e); + +static inline int iotlb_cr_valid(struct cr_regs *cr) +{ + if (!cr) + return -EINVAL; + + return arch_iommu->cr_valid(cr); +} + +static inline struct cr_regs *iotlb_alloc_cr(struct iommu *obj, + struct iotlb_entry *e) +{ + if (!e) + return NULL; + + return arch_iommu->alloc_cr(obj, e); +} + +u32 iotlb_cr_to_virt(struct cr_regs *cr) +{ + return arch_iommu->cr_to_virt(cr); +} +EXPORT_SYMBOL_GPL(iotlb_cr_to_virt); + +static u32 get_iopte_attr(struct iotlb_entry *e) +{ + return arch_iommu->get_pte_attr(e); +} + +static u32 iommu_report_fault(struct iommu *obj, u32 *da) +{ + return arch_iommu->fault_isr(obj, da); +} + +static void iotlb_lock_get(struct iommu *obj, struct iotlb_lock *l) +{ + u32 val; + + val = iommu_read_reg(obj, MMU_LOCK); + + l->base = MMU_LOCK_BASE(val); + l->vict = MMU_LOCK_VICT(val); + + BUG_ON(l->base != 0); /* Currently no preservation is used */ +} + +static void iotlb_lock_set(struct iommu *obj, struct iotlb_lock *l) +{ + u32 val; + + BUG_ON(l->base != 0); /* Currently no preservation is used */ + + val = (l->base << MMU_LOCK_BASE_SHIFT); + val |= (l->vict << MMU_LOCK_VICT_SHIFT); + + iommu_write_reg(obj, val, MMU_LOCK); +} + +static void iotlb_read_cr(struct iommu *obj, struct cr_regs *cr) +{ + arch_iommu->tlb_read_cr(obj, cr); +} + +static void iotlb_load_cr(struct iommu *obj, struct cr_regs *cr) +{ + arch_iommu->tlb_load_cr(obj, cr); + + iommu_write_reg(obj, 1, MMU_FLUSH_ENTRY); + iommu_write_reg(obj, 1, MMU_LD_TLB); +} + +/** + * iotlb_dump_cr - Dump an iommu tlb entry into buf + * @obj: target iommu + * @cr: contents of cam and ram register + * @buf: output buffer + **/ +static inline ssize_t iotlb_dump_cr(struct iommu *obj, struct cr_regs *cr, + char *buf) +{ + BUG_ON(!cr || !buf); + + return arch_iommu->dump_cr(obj, cr, buf); +} + +/** + * load_iotlb_entry - Set an iommu tlb entry + * @obj: target iommu + * @e: an iommu tlb entry info + **/ +int load_iotlb_entry(struct iommu *obj, struct iotlb_entry *e) +{ + int i; + int err = 0; + struct iotlb_lock l; + struct cr_regs *cr; + + if (!obj || !obj->nr_tlb_entries || !e) + return -EINVAL; + + clk_enable(obj->clk); + + for (i = 0; i < obj->nr_tlb_entries; i++) { + struct cr_regs tmp; + + iotlb_lock_get(obj, &l); + l.vict = i; + iotlb_lock_set(obj, &l); + iotlb_read_cr(obj, &tmp); + if (!iotlb_cr_valid(&tmp)) + break; + } + + if (i == obj->nr_tlb_entries) { + dev_dbg(obj->dev, "%s: full: no entry\n", __func__); + err = -EBUSY; + goto out; + } + + cr = iotlb_alloc_cr(obj, e); + if (IS_ERR(cr)) { + clk_disable(obj->clk); + return PTR_ERR(cr); + } + + iotlb_load_cr(obj, cr); + kfree(cr); + + /* increment victim for next tlb load */ + if (++l.vict == obj->nr_tlb_entries) + l.vict = 0; + iotlb_lock_set(obj, &l); +out: + clk_disable(obj->clk); + return err; +} +EXPORT_SYMBOL_GPL(load_iotlb_entry); + +/** + * flush_iotlb_page - Clear an iommu tlb entry + * @obj: target iommu + * @da: iommu device virtual address + * + * Clear an iommu tlb entry which includes 'da' address. + **/ +void flush_iotlb_page(struct iommu *obj, u32 da) +{ + struct iotlb_lock l; + int i; + + clk_enable(obj->clk); + + for (i = 0; i < obj->nr_tlb_entries; i++) { + struct cr_regs cr; + u32 start; + size_t bytes; + + iotlb_lock_get(obj, &l); + l.vict = i; + iotlb_lock_set(obj, &l); + iotlb_read_cr(obj, &cr); + if (!iotlb_cr_valid(&cr)) + continue; + + start = iotlb_cr_to_virt(&cr); + bytes = iopgsz_to_bytes(cr.cam & 3); + + if ((start <= da) && (da < start + bytes)) { + dev_dbg(obj->dev, "%s: %08x<=%08x(%x)\n", + __func__, start, da, bytes); + + iommu_write_reg(obj, 1, MMU_FLUSH_ENTRY); + } + } + clk_disable(obj->clk); + + if (i == obj->nr_tlb_entries) + dev_dbg(obj->dev, "%s: no page for %08x\n", __func__, da); +} +EXPORT_SYMBOL_GPL(flush_iotlb_page); + +/** + * flush_iotlb_range - Clear an iommu tlb entries + * @obj: target iommu + * @start: iommu device virtual address(start) + * @end: iommu device virtual address(end) + * + * Clear an iommu tlb entry which includes 'da' address. + **/ +void flush_iotlb_range(struct iommu *obj, u32 start, u32 end) +{ + u32 da = start; + + while (da < end) { + flush_iotlb_page(obj, da); + /* FIXME: Optimize for multiple page size */ + da += IOPTE_SIZE; + } +} +EXPORT_SYMBOL_GPL(flush_iotlb_range); + +/** + * flush_iotlb_all - Clear all iommu tlb entries + * @obj: target iommu + **/ +void flush_iotlb_all(struct iommu *obj) +{ + struct iotlb_lock l; + + clk_enable(obj->clk); + + l.base = 0; + l.vict = 0; + iotlb_lock_set(obj, &l); + + iommu_write_reg(obj, 1, MMU_GFLUSH); + + clk_disable(obj->clk); +} +EXPORT_SYMBOL_GPL(flush_iotlb_all); + +#if defined(CONFIG_OMAP_IOMMU_DEBUG_MODULE) + +ssize_t iommu_dump_ctx(struct iommu *obj, char *buf) +{ + ssize_t bytes; + + if (!obj || !buf) + return -EINVAL; + + clk_enable(obj->clk); + + bytes = arch_iommu->dump_ctx(obj, buf); + + clk_disable(obj->clk); + + return bytes; +} +EXPORT_SYMBOL_GPL(iommu_dump_ctx); + +static int __dump_tlb_entries(struct iommu *obj, struct cr_regs *crs) +{ + int i; + struct iotlb_lock saved, l; + struct cr_regs *p = crs; + + clk_enable(obj->clk); + + iotlb_lock_get(obj, &saved); + memcpy(&l, &saved, sizeof(saved)); + + for (i = 0; i < obj->nr_tlb_entries; i++) { + struct cr_regs tmp; + + iotlb_lock_get(obj, &l); + l.vict = i; + iotlb_lock_set(obj, &l); + iotlb_read_cr(obj, &tmp); + if (!iotlb_cr_valid(&tmp)) + continue; + + *p++ = tmp; + } + iotlb_lock_set(obj, &saved); + clk_disable(obj->clk); + + return p - crs; +} + +/** + * dump_tlb_entries - dump cr arrays to given buffer + * @obj: target iommu + * @buf: output buffer + **/ +size_t dump_tlb_entries(struct iommu *obj, char *buf) +{ + int i, n; + struct cr_regs *cr; + char *p = buf; + + cr = kcalloc(obj->nr_tlb_entries, sizeof(*cr), GFP_KERNEL); + if (!cr) + return 0; + + n = __dump_tlb_entries(obj, cr); + for (i = 0; i < n; i++) + p += iotlb_dump_cr(obj, cr + i, p); + kfree(cr); + + return p - buf; +} +EXPORT_SYMBOL_GPL(dump_tlb_entries); + +int foreach_iommu_device(void *data, int (*fn)(struct device *, void *)) +{ + return driver_for_each_device(&omap_iommu_driver.driver, + NULL, data, fn); +} +EXPORT_SYMBOL_GPL(foreach_iommu_device); + +#endif /* CONFIG_OMAP_IOMMU_DEBUG_MODULE */ + +/* + * H/W pagetable operations + */ +static void flush_iopgd_range(u32 *first, u32 *last) +{ + /* FIXME: L2 cache should be taken care of if it exists */ + do { + asm("mcr p15, 0, %0, c7, c10, 1 @ flush_pgd" + : : "r" (first)); + first += L1_CACHE_BYTES / sizeof(*first); + } while (first <= last); +} + +static void flush_iopte_range(u32 *first, u32 *last) +{ + /* FIXME: L2 cache should be taken care of if it exists */ + do { + asm("mcr p15, 0, %0, c7, c10, 1 @ flush_pte" + : : "r" (first)); + first += L1_CACHE_BYTES / sizeof(*first); + } while (first <= last); +} + +static void iopte_free(u32 *iopte) +{ + /* Note: freed iopte's must be clean ready for re-use */ + kmem_cache_free(iopte_cachep, iopte); +} + +static u32 *iopte_alloc(struct iommu *obj, u32 *iopgd, u32 da) +{ + u32 *iopte; + + /* a table has already existed */ + if (*iopgd) + goto pte_ready; + + /* + * do the allocation outside the page table lock + */ + spin_unlock(&obj->page_table_lock); + iopte = kmem_cache_zalloc(iopte_cachep, GFP_KERNEL); + spin_lock(&obj->page_table_lock); + + if (!*iopgd) { + if (!iopte) + return ERR_PTR(-ENOMEM); + + *iopgd = virt_to_phys(iopte) | IOPGD_TABLE; + flush_iopgd_range(iopgd, iopgd); + + dev_vdbg(obj->dev, "%s: a new pte:%p\n", __func__, iopte); + } else { + /* We raced, free the reduniovant table */ + iopte_free(iopte); + } + +pte_ready: + iopte = iopte_offset(iopgd, da); + + dev_vdbg(obj->dev, + "%s: da:%08x pgd:%p *pgd:%08x pte:%p *pte:%08x\n", + __func__, da, iopgd, *iopgd, iopte, *iopte); + + return iopte; +} + +static int iopgd_alloc_section(struct iommu *obj, u32 da, u32 pa, u32 prot) +{ + u32 *iopgd = iopgd_offset(obj, da); + + *iopgd = (pa & IOSECTION_MASK) | prot | IOPGD_SECTION; + flush_iopgd_range(iopgd, iopgd); + return 0; +} + +static int iopgd_alloc_super(struct iommu *obj, u32 da, u32 pa, u32 prot) +{ + u32 *iopgd = iopgd_offset(obj, da); + int i; + + for (i = 0; i < 16; i++) + *(iopgd + i) = (pa & IOSUPER_MASK) | prot | IOPGD_SUPER; + flush_iopgd_range(iopgd, iopgd + 15); + return 0; +} + +static int iopte_alloc_page(struct iommu *obj, u32 da, u32 pa, u32 prot) +{ + u32 *iopgd = iopgd_offset(obj, da); + u32 *iopte = iopte_alloc(obj, iopgd, da); + + if (IS_ERR(iopte)) + return PTR_ERR(iopte); + + *iopte = (pa & IOPAGE_MASK) | prot | IOPTE_SMALL; + flush_iopte_range(iopte, iopte); + + dev_vdbg(obj->dev, "%s: da:%08x pa:%08x pte:%p *pte:%08x\n", + __func__, da, pa, iopte, *iopte); + + return 0; +} + +static int iopte_alloc_large(struct iommu *obj, u32 da, u32 pa, u32 prot) +{ + u32 *iopgd = iopgd_offset(obj, da); + u32 *iopte = iopte_alloc(obj, iopgd, da); + int i; + + if (IS_ERR(iopte)) + return PTR_ERR(iopte); + + for (i = 0; i < 16; i++) + *(iopte + i) = (pa & IOLARGE_MASK) | prot | IOPTE_LARGE; + flush_iopte_range(iopte, iopte + 15); + return 0; +} + +static int iopgtable_store_entry_core(struct iommu *obj, struct iotlb_entry *e) +{ + int (*fn)(struct iommu *, u32, u32, u32); + u32 prot; + int err; + + if (!obj || !e) + return -EINVAL; + + switch (e->pgsz) { + case MMU_CAM_PGSZ_16M: + fn = iopgd_alloc_super; + break; + case MMU_CAM_PGSZ_1M: + fn = iopgd_alloc_section; + break; + case MMU_CAM_PGSZ_64K: + fn = iopte_alloc_large; + break; + case MMU_CAM_PGSZ_4K: + fn = iopte_alloc_page; + break; + default: + fn = NULL; + BUG(); + break; + } + + prot = get_iopte_attr(e); + + spin_lock(&obj->page_table_lock); + err = fn(obj, e->da, e->pa, prot); + spin_unlock(&obj->page_table_lock); + + return err; +} + +/** + * iopgtable_store_entry - Make an iommu pte entry + * @obj: target iommu + * @e: an iommu tlb entry info + **/ +int iopgtable_store_entry(struct iommu *obj, struct iotlb_entry *e) +{ + int err; + + flush_iotlb_page(obj, e->da); + err = iopgtable_store_entry_core(obj, e); +#ifdef PREFETCH_IOTLB + if (!err) + load_iotlb_entry(obj, e); +#endif + return err; +} +EXPORT_SYMBOL_GPL(iopgtable_store_entry); + +/** + * iopgtable_lookup_entry - Lookup an iommu pte entry + * @obj: target iommu + * @da: iommu device virtual address + * @ppgd: iommu pgd entry pointer to be returned + * @ppte: iommu pte entry pointer to be returned + **/ +void iopgtable_lookup_entry(struct iommu *obj, u32 da, u32 **ppgd, u32 **ppte) +{ + u32 *iopgd, *iopte = NULL; + + iopgd = iopgd_offset(obj, da); + if (!*iopgd) + goto out; + + if (*iopgd & IOPGD_TABLE) + iopte = iopte_offset(iopgd, da); +out: + *ppgd = iopgd; + *ppte = iopte; +} +EXPORT_SYMBOL_GPL(iopgtable_lookup_entry); + +static size_t iopgtable_clear_entry_core(struct iommu *obj, u32 da) +{ + size_t bytes; + u32 *iopgd = iopgd_offset(obj, da); + int nent = 1; + + if (!*iopgd) + return 0; + + if (*iopgd & IOPGD_TABLE) { + int i; + u32 *iopte = iopte_offset(iopgd, da); + + bytes = IOPTE_SIZE; + if (*iopte & IOPTE_LARGE) { + nent *= 16; + /* rewind to the 1st entry */ + iopte = (u32 *)((u32)iopte & IOLARGE_MASK); + } + bytes *= nent; + memset(iopte, 0, nent * sizeof(*iopte)); + flush_iopte_range(iopte, iopte + (nent - 1) * sizeof(*iopte)); + + /* + * do table walk to check if this table is necessary or not + */ + iopte = iopte_offset(iopgd, 0); + for (i = 0; i < PTRS_PER_IOPTE; i++) + if (iopte[i]) + goto out; + + iopte_free(iopte); + nent = 1; /* for the next L1 entry */ + } else { + bytes = IOPGD_SIZE; + if (*iopgd & IOPGD_SUPER) { + nent *= 16; + /* rewind to the 1st entry */ + iopgd = (u32 *)((u32)iopgd & IOSUPER_MASK); + } + bytes *= nent; + } + memset(iopgd, 0, nent * sizeof(*iopgd)); + flush_iopgd_range(iopgd, iopgd + (nent - 1) * sizeof(*iopgd)); +out: + return bytes; +} + +/** + * iopgtable_clear_entry - Remove an iommu pte entry + * @obj: target iommu + * @da: iommu device virtual address + **/ +size_t iopgtable_clear_entry(struct iommu *obj, u32 da) +{ + size_t bytes; + + spin_lock(&obj->page_table_lock); + + bytes = iopgtable_clear_entry_core(obj, da); + flush_iotlb_page(obj, da); + + spin_unlock(&obj->page_table_lock); + + return bytes; +} +EXPORT_SYMBOL_GPL(iopgtable_clear_entry); + +static void iopgtable_clear_entry_all(struct iommu *obj) +{ + int i; + + spin_lock(&obj->page_table_lock); + + for (i = 0; i < PTRS_PER_IOPGD; i++) { + u32 da; + u32 *iopgd; + + da = i << IOPGD_SHIFT; + iopgd = iopgd_offset(obj, da); + + if (!*iopgd) + continue; + + if (*iopgd & IOPGD_TABLE) + iopte_free(iopte_offset(iopgd, 0)); + + *iopgd = 0; + flush_iopgd_range(iopgd, iopgd); + } + + flush_iotlb_all(obj); + + spin_unlock(&obj->page_table_lock); +} + +/* + * Device IOMMU generic operations + */ +static irqreturn_t iommu_fault_handler(int irq, void *data) +{ + u32 stat, da; + u32 *iopgd, *iopte; + int err = -EIO; + struct iommu *obj = data; + + if (!obj->refcount) + return IRQ_NONE; + + /* Dynamic loading TLB or PTE */ + if (obj->isr) + err = obj->isr(obj); + + if (!err) + return IRQ_HANDLED; + + clk_enable(obj->clk); + stat = iommu_report_fault(obj, &da); + clk_disable(obj->clk); + if (!stat) + return IRQ_HANDLED; + + iopgd = iopgd_offset(obj, da); + + if (!(*iopgd & IOPGD_TABLE)) { + dev_err(obj->dev, "%s: da:%08x pgd:%p *pgd:%08x\n", __func__, + da, iopgd, *iopgd); + return IRQ_NONE; + } + + iopte = iopte_offset(iopgd, da); + + dev_err(obj->dev, "%s: da:%08x pgd:%p *pgd:%08x pte:%p *pte:%08x\n", + __func__, da, iopgd, *iopgd, iopte, *iopte); + + return IRQ_NONE; +} + +static int device_match_by_alias(struct device *dev, void *data) +{ + struct iommu *obj = to_iommu(dev); + const char *name = data; + + pr_debug("%s: %s %s\n", __func__, obj->name, name); + + return strcmp(obj->name, name) == 0; +} + +/** + * iommu_get - Get iommu handler + * @name: target iommu name + **/ +struct iommu *iommu_get(const char *name) +{ + int err = -ENOMEM; + struct device *dev; + struct iommu *obj; + + dev = driver_find_device(&omap_iommu_driver.driver, NULL, (void *)name, + device_match_by_alias); + if (!dev) + return ERR_PTR(-ENODEV); + + obj = to_iommu(dev); + + mutex_lock(&obj->iommu_lock); + + if (obj->refcount++ == 0) { + err = iommu_enable(obj); + if (err) + goto err_enable; + flush_iotlb_all(obj); + } + + if (!try_module_get(obj->owner)) + goto err_module; + + mutex_unlock(&obj->iommu_lock); + + dev_dbg(obj->dev, "%s: %s\n", __func__, obj->name); + return obj; + +err_module: + if (obj->refcount == 1) + iommu_disable(obj); +err_enable: + obj->refcount--; + mutex_unlock(&obj->iommu_lock); + return ERR_PTR(err); +} +EXPORT_SYMBOL_GPL(iommu_get); + +/** + * iommu_put - Put back iommu handler + * @obj: target iommu + **/ +void iommu_put(struct iommu *obj) +{ + if (!obj && IS_ERR(obj)) + return; + + mutex_lock(&obj->iommu_lock); + + if (--obj->refcount == 0) + iommu_disable(obj); + + module_put(obj->owner); + + mutex_unlock(&obj->iommu_lock); + + dev_dbg(obj->dev, "%s: %s\n", __func__, obj->name); +} +EXPORT_SYMBOL_GPL(iommu_put); + +/* + * OMAP Device MMU(IOMMU) detection + */ +static int __devinit omap_iommu_probe(struct platform_device *pdev) +{ + int err = -ENODEV; + void *p; + int irq; + struct iommu *obj; + struct resource *res; + struct iommu_platform_data *pdata = pdev->dev.platform_data; + + if (pdev->num_resources != 2) + return -EINVAL; + + obj = kzalloc(sizeof(*obj) + MMU_REG_SIZE, GFP_KERNEL); + if (!obj) + return -ENOMEM; + + obj->clk = clk_get(&pdev->dev, pdata->clk_name); + if (IS_ERR(obj->clk)) + goto err_clk; + + obj->nr_tlb_entries = pdata->nr_tlb_entries; + obj->name = pdata->name; + obj->dev = &pdev->dev; + obj->ctx = (void *)obj + sizeof(*obj); + + mutex_init(&obj->iommu_lock); + mutex_init(&obj->mmap_lock); + spin_lock_init(&obj->page_table_lock); + INIT_LIST_HEAD(&obj->mmap); + + res = platform_get_resource(pdev, IORESOURCE_MEM, 0); + if (!res) { + err = -ENODEV; + goto err_mem; + } + obj->regbase = ioremap(res->start, resource_size(res)); + if (!obj->regbase) { + err = -ENOMEM; + goto err_mem; + } + + res = request_mem_region(res->start, resource_size(res), + dev_name(&pdev->dev)); + if (!res) { + err = -EIO; + goto err_mem; + } + + irq = platform_get_irq(pdev, 0); + if (irq < 0) { + err = -ENODEV; + goto err_irq; + } + err = request_irq(irq, iommu_fault_handler, IRQF_SHARED, + dev_name(&pdev->dev), obj); + if (err < 0) + goto err_irq; + platform_set_drvdata(pdev, obj); + + p = (void *)__get_free_pages(GFP_KERNEL, get_order(IOPGD_TABLE_SIZE)); + if (!p) { + err = -ENOMEM; + goto err_pgd; + } + memset(p, 0, IOPGD_TABLE_SIZE); + clean_dcache_area(p, IOPGD_TABLE_SIZE); + obj->iopgd = p; + + BUG_ON(!IS_ALIGNED((unsigned long)obj->iopgd, IOPGD_TABLE_SIZE)); + + dev_info(&pdev->dev, "%s registered\n", obj->name); + return 0; + +err_pgd: + free_irq(irq, obj); +err_irq: + release_mem_region(res->start, resource_size(res)); + iounmap(obj->regbase); +err_mem: + clk_put(obj->clk); +err_clk: + kfree(obj); + return err; +} + +static int __devexit omap_iommu_remove(struct platform_device *pdev) +{ + int irq; + struct resource *res; + struct iommu *obj = platform_get_drvdata(pdev); + + platform_set_drvdata(pdev, NULL); + + iopgtable_clear_entry_all(obj); + free_pages((unsigned long)obj->iopgd, get_order(IOPGD_TABLE_SIZE)); + + irq = platform_get_irq(pdev, 0); + free_irq(irq, obj); + res = platform_get_resource(pdev, IORESOURCE_MEM, 0); + release_mem_region(res->start, resource_size(res)); + iounmap(obj->regbase); + + clk_put(obj->clk); + dev_info(&pdev->dev, "%s removed\n", obj->name); + kfree(obj); + return 0; +} + +static struct platform_driver omap_iommu_driver = { + .probe = omap_iommu_probe, + .remove = __devexit_p(omap_iommu_remove), + .driver = { + .name = "omap-iommu", + }, +}; + +static void iopte_cachep_ctor(void *iopte) +{ + clean_dcache_area(iopte, IOPTE_TABLE_SIZE); +} + +static int __init omap_iommu_init(void) +{ + struct kmem_cache *p; + const unsigned long flags = SLAB_HWCACHE_ALIGN; + size_t align = 1 << 10; /* L2 pagetable alignement */ + + p = kmem_cache_create("iopte_cache", IOPTE_TABLE_SIZE, align, flags, + iopte_cachep_ctor); + if (!p) + return -ENOMEM; + iopte_cachep = p; + + return platform_driver_register(&omap_iommu_driver); +} +module_init(omap_iommu_init); + +static void __exit omap_iommu_exit(void) +{ + kmem_cache_destroy(iopte_cachep); + + platform_driver_unregister(&omap_iommu_driver); +} +module_exit(omap_iommu_exit); + +MODULE_DESCRIPTION("omap iommu: tlb and pagetable primitives"); +MODULE_ALIAS("platform:omap-iommu"); +MODULE_AUTHOR("Hiroshi DOYU, Paul Mundt and Toshihiro Kobayashi"); +MODULE_LICENSE("GPL v2"); diff --git a/arch/arm/plat-omap/iopgtable.h b/arch/arm/plat-omap/iopgtable.h new file mode 100644 index 000000000000..37dac434c7a1 --- /dev/null +++ b/arch/arm/plat-omap/iopgtable.h @@ -0,0 +1,72 @@ +/* + * omap iommu: pagetable definitions + * + * Copyright (C) 2008-2009 Nokia Corporation + * + * Written by Hiroshi DOYU + * + * 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. + */ + +#ifndef __PLAT_OMAP_IOMMU_H +#define __PLAT_OMAP_IOMMU_H + +#define IOPGD_SHIFT 20 +#define IOPGD_SIZE (1 << IOPGD_SHIFT) +#define IOPGD_MASK (~(IOPGD_SIZE - 1)) +#define IOSECTION_MASK IOPGD_MASK +#define PTRS_PER_IOPGD (1 << (32 - IOPGD_SHIFT)) +#define IOPGD_TABLE_SIZE (PTRS_PER_IOPGD * sizeof(u32)) + +#define IOSUPER_SIZE (IOPGD_SIZE << 4) +#define IOSUPER_MASK (~(IOSUPER_SIZE - 1)) + +#define IOPTE_SHIFT 12 +#define IOPTE_SIZE (1 << IOPTE_SHIFT) +#define IOPTE_MASK (~(IOPTE_SIZE - 1)) +#define IOPAGE_MASK IOPTE_MASK +#define PTRS_PER_IOPTE (1 << (IOPGD_SHIFT - IOPTE_SHIFT)) +#define IOPTE_TABLE_SIZE (PTRS_PER_IOPTE * sizeof(u32)) + +#define IOLARGE_SIZE (IOPTE_SIZE << 4) +#define IOLARGE_MASK (~(IOLARGE_SIZE - 1)) + +#define IOPGD_TABLE (1 << 0) +#define IOPGD_SECTION (2 << 0) +#define IOPGD_SUPER (1 << 18 | 2 << 0) + +#define IOPTE_SMALL (2 << 0) +#define IOPTE_LARGE (1 << 0) + +#define iopgd_index(da) (((da) >> IOPGD_SHIFT) & (PTRS_PER_IOPGD - 1)) +#define iopgd_offset(obj, da) ((obj)->iopgd + iopgd_index(da)) + +#define iopte_paddr(iopgd) (*iopgd & ~((1 << 10) - 1)) +#define iopte_vaddr(iopgd) ((u32 *)phys_to_virt(iopte_paddr(iopgd))) + +#define iopte_index(da) (((da) >> IOPTE_SHIFT) & (PTRS_PER_IOPTE - 1)) +#define iopte_offset(iopgd, da) (iopte_vaddr(iopgd) + iopte_index(da)) + +static inline u32 iotlb_init_entry(struct iotlb_entry *e, u32 da, u32 pa, + u32 flags) +{ + memset(e, 0, sizeof(*e)); + + e->da = da; + e->pa = pa; + e->valid = 1; + /* FIXME: add OMAP1 support */ + e->pgsz = flags & MMU_CAM_PGSZ_MASK; + e->endian = flags & MMU_RAM_ENDIAN_MASK; + e->elsz = flags & MMU_RAM_ELSZ_MASK; + e->mixed = flags & MMU_RAM_MIXED_MASK; + + return iopgsz_to_bytes(e->pgsz); +} + +#define to_iommu(dev) \ + (struct iommu *)platform_get_drvdata(to_platform_device(dev)) + +#endif /* __PLAT_OMAP_IOMMU_H */ From 2bcb573343dbd7d913cb62b81463960644a3737f Mon Sep 17 00:00:00 2001 From: Hiroshi DOYU Date: Mon, 26 Jan 2009 15:13:45 +0200 Subject: [PATCH 2/7] omap iommu: omap2 architecture specific functions The structure 'arch_mmu' accommodates the difference between omap1 and omap2/3. This patch provides omap2/3 specific functions Signed-off-by: Hiroshi DOYU --- arch/arm/mach-omap2/iommu2.c | 323 +++++++++++++++++++++++ arch/arm/plat-omap/include/mach/iommu2.h | 96 +++++++ 2 files changed, 419 insertions(+) create mode 100644 arch/arm/mach-omap2/iommu2.c create mode 100644 arch/arm/plat-omap/include/mach/iommu2.h diff --git a/arch/arm/mach-omap2/iommu2.c b/arch/arm/mach-omap2/iommu2.c new file mode 100644 index 000000000000..015f22a53ead --- /dev/null +++ b/arch/arm/mach-omap2/iommu2.c @@ -0,0 +1,323 @@ +/* + * omap iommu: omap2/3 architecture specific functions + * + * Copyright (C) 2008-2009 Nokia Corporation + * + * Written by Hiroshi DOYU , + * Paul Mundt and Toshihiro Kobayashi + * + * 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 + +/* + * omap2 architecture specific register bit definitions + */ +#define IOMMU_ARCH_VERSION 0x00000011 + +/* SYSCONF */ +#define MMU_SYS_IDLE_SHIFT 3 +#define MMU_SYS_IDLE_FORCE (0 << MMU_SYS_IDLE_SHIFT) +#define MMU_SYS_IDLE_NONE (1 << MMU_SYS_IDLE_SHIFT) +#define MMU_SYS_IDLE_SMART (2 << MMU_SYS_IDLE_SHIFT) +#define MMU_SYS_IDLE_MASK (3 << MMU_SYS_IDLE_SHIFT) + +#define MMU_SYS_SOFTRESET (1 << 1) +#define MMU_SYS_AUTOIDLE 1 + +/* SYSSTATUS */ +#define MMU_SYS_RESETDONE 1 + +/* IRQSTATUS & IRQENABLE */ +#define MMU_IRQ_MULTIHITFAULT (1 << 4) +#define MMU_IRQ_TABLEWALKFAULT (1 << 3) +#define MMU_IRQ_EMUMISS (1 << 2) +#define MMU_IRQ_TRANSLATIONFAULT (1 << 1) +#define MMU_IRQ_TLBMISS (1 << 0) +#define MMU_IRQ_MASK \ + (MMU_IRQ_MULTIHITFAULT | MMU_IRQ_TABLEWALKFAULT | MMU_IRQ_EMUMISS | \ + MMU_IRQ_TRANSLATIONFAULT) + +/* MMU_CNTL */ +#define MMU_CNTL_SHIFT 1 +#define MMU_CNTL_MASK (7 << MMU_CNTL_SHIFT) +#define MMU_CNTL_EML_TLB (1 << 3) +#define MMU_CNTL_TWL_EN (1 << 2) +#define MMU_CNTL_MMU_EN (1 << 1) + +#define get_cam_va_mask(pgsz) \ + (((pgsz) == MMU_CAM_PGSZ_16M) ? 0xff000000 : \ + ((pgsz) == MMU_CAM_PGSZ_1M) ? 0xfff00000 : \ + ((pgsz) == MMU_CAM_PGSZ_64K) ? 0xffff0000 : \ + ((pgsz) == MMU_CAM_PGSZ_4K) ? 0xfffff000 : 0) + +static int omap2_iommu_enable(struct iommu *obj) +{ + u32 l, pa; + unsigned long timeout; + + if (!obj->iopgd || !IS_ALIGNED((u32)obj->iopgd, SZ_16K)) + return -EINVAL; + + pa = virt_to_phys(obj->iopgd); + if (!IS_ALIGNED(pa, SZ_16K)) + return -EINVAL; + + iommu_write_reg(obj, MMU_SYS_SOFTRESET, MMU_SYSCONFIG); + + timeout = jiffies + msecs_to_jiffies(20); + do { + l = iommu_read_reg(obj, MMU_SYSSTATUS); + if (l & MMU_SYS_RESETDONE) + break; + } while (time_after(jiffies, timeout)); + + if (!(l & MMU_SYS_RESETDONE)) { + dev_err(obj->dev, "can't take mmu out of reset\n"); + return -ENODEV; + } + + l = iommu_read_reg(obj, MMU_REVISION); + dev_info(obj->dev, "%s: version %d.%d\n", obj->name, + (l >> 4) & 0xf, l & 0xf); + + l = iommu_read_reg(obj, MMU_SYSCONFIG); + l &= ~MMU_SYS_IDLE_MASK; + l |= (MMU_SYS_IDLE_SMART | MMU_SYS_AUTOIDLE); + iommu_write_reg(obj, l, MMU_SYSCONFIG); + + iommu_write_reg(obj, MMU_IRQ_MASK, MMU_IRQENABLE); + iommu_write_reg(obj, pa, MMU_TTB); + + l = iommu_read_reg(obj, MMU_CNTL); + l &= ~MMU_CNTL_MASK; + l |= (MMU_CNTL_MMU_EN | MMU_CNTL_TWL_EN); + iommu_write_reg(obj, l, MMU_CNTL); + + return 0; +} + +static void omap2_iommu_disable(struct iommu *obj) +{ + u32 l = iommu_read_reg(obj, MMU_CNTL); + + l &= ~MMU_CNTL_MASK; + iommu_write_reg(obj, l, MMU_CNTL); + iommu_write_reg(obj, MMU_SYS_IDLE_FORCE, MMU_SYSCONFIG); + + dev_dbg(obj->dev, "%s is shutting down\n", obj->name); +} + +static u32 omap2_iommu_fault_isr(struct iommu *obj, u32 *ra) +{ + int i; + u32 stat, da; + const char *err_msg[] = { + "tlb miss", + "translation fault", + "emulation miss", + "table walk fault", + "multi hit fault", + }; + + stat = iommu_read_reg(obj, MMU_IRQSTATUS); + stat &= MMU_IRQ_MASK; + if (!stat) + return 0; + + da = iommu_read_reg(obj, MMU_FAULT_AD); + *ra = da; + + dev_err(obj->dev, "%s:\tda:%08x ", __func__, da); + + for (i = 0; i < ARRAY_SIZE(err_msg); i++) { + if (stat & (1 << i)) + printk("%s ", err_msg[i]); + } + printk("\n"); + + iommu_write_reg(obj, stat, MMU_IRQSTATUS); + return stat; +} + +static void omap2_tlb_read_cr(struct iommu *obj, struct cr_regs *cr) +{ + cr->cam = iommu_read_reg(obj, MMU_READ_CAM); + cr->ram = iommu_read_reg(obj, MMU_READ_RAM); +} + +static void omap2_tlb_load_cr(struct iommu *obj, struct cr_regs *cr) +{ + iommu_write_reg(obj, cr->cam | MMU_CAM_V, MMU_CAM); + iommu_write_reg(obj, cr->ram, MMU_RAM); +} + +static u32 omap2_cr_to_virt(struct cr_regs *cr) +{ + u32 page_size = cr->cam & MMU_CAM_PGSZ_MASK; + u32 mask = get_cam_va_mask(cr->cam & page_size); + + return cr->cam & mask; +} + +static struct cr_regs *omap2_alloc_cr(struct iommu *obj, struct iotlb_entry *e) +{ + struct cr_regs *cr; + + if (e->da & ~(get_cam_va_mask(e->pgsz))) { + dev_err(obj->dev, "%s:\twrong alignment: %08x\n", __func__, + e->da); + return ERR_PTR(-EINVAL); + } + + cr = kmalloc(sizeof(*cr), GFP_KERNEL); + if (!cr) + return ERR_PTR(-ENOMEM); + + cr->cam = (e->da & MMU_CAM_VATAG_MASK) | e->prsvd | e->pgsz; + cr->ram = e->pa | e->endian | e->elsz | e->mixed; + + return cr; +} + +static inline int omap2_cr_valid(struct cr_regs *cr) +{ + return cr->cam & MMU_CAM_V; +} + +static u32 omap2_get_pte_attr(struct iotlb_entry *e) +{ + u32 attr; + + attr = e->mixed << 5; + attr |= e->endian; + attr |= e->elsz >> 3; + attr <<= ((e->pgsz & MMU_CAM_PGSZ_4K) ? 0 : 6); + + return attr; +} + +static ssize_t omap2_dump_cr(struct iommu *obj, struct cr_regs *cr, char *buf) +{ + char *p = buf; + + /* FIXME: Need more detail analysis of cam/ram */ + p += sprintf(p, "%08x %08x\n", cr->cam, cr->ram); + + return p - buf; +} + +#define pr_reg(name) \ + p += sprintf(p, "%20s: %08x\n", \ + __stringify(name), iommu_read_reg(obj, MMU_##name)); + +static ssize_t omap2_iommu_dump_ctx(struct iommu *obj, char *buf) +{ + char *p = buf; + + pr_reg(REVISION); + pr_reg(SYSCONFIG); + pr_reg(SYSSTATUS); + pr_reg(IRQSTATUS); + pr_reg(IRQENABLE); + pr_reg(WALKING_ST); + pr_reg(CNTL); + pr_reg(FAULT_AD); + pr_reg(TTB); + pr_reg(LOCK); + pr_reg(LD_TLB); + pr_reg(CAM); + pr_reg(RAM); + pr_reg(GFLUSH); + pr_reg(FLUSH_ENTRY); + pr_reg(READ_CAM); + pr_reg(READ_RAM); + pr_reg(EMU_FAULT_AD); + + return p - buf; +} + +static void omap2_iommu_save_ctx(struct iommu *obj) +{ + int i; + u32 *p = obj->ctx; + + for (i = 0; i < (MMU_REG_SIZE / sizeof(u32)); i++) { + p[i] = iommu_read_reg(obj, i * sizeof(u32)); + dev_dbg(obj->dev, "%s\t[%02d] %08x\n", __func__, i, p[i]); + } + + BUG_ON(p[0] != IOMMU_ARCH_VERSION); +} + +static void omap2_iommu_restore_ctx(struct iommu *obj) +{ + int i; + u32 *p = obj->ctx; + + for (i = 0; i < (MMU_REG_SIZE / sizeof(u32)); i++) { + iommu_write_reg(obj, p[i], i * sizeof(u32)); + dev_dbg(obj->dev, "%s\t[%02d] %08x\n", __func__, i, p[i]); + } + + BUG_ON(p[0] != IOMMU_ARCH_VERSION); +} + +static void omap2_cr_to_e(struct cr_regs *cr, struct iotlb_entry *e) +{ + e->da = cr->cam & MMU_CAM_VATAG_MASK; + e->pa = cr->ram & MMU_RAM_PADDR_MASK; + e->valid = cr->cam & MMU_CAM_V; + e->pgsz = cr->cam & MMU_CAM_PGSZ_MASK; + e->endian = cr->ram & MMU_RAM_ENDIAN_MASK; + e->elsz = cr->ram & MMU_RAM_ELSZ_MASK; + e->mixed = cr->ram & MMU_RAM_MIXED; +} + +static const struct iommu_functions omap2_iommu_ops = { + .version = IOMMU_ARCH_VERSION, + + .enable = omap2_iommu_enable, + .disable = omap2_iommu_disable, + .fault_isr = omap2_iommu_fault_isr, + + .tlb_read_cr = omap2_tlb_read_cr, + .tlb_load_cr = omap2_tlb_load_cr, + + .cr_to_e = omap2_cr_to_e, + .cr_to_virt = omap2_cr_to_virt, + .alloc_cr = omap2_alloc_cr, + .cr_valid = omap2_cr_valid, + .dump_cr = omap2_dump_cr, + + .get_pte_attr = omap2_get_pte_attr, + + .save_ctx = omap2_iommu_save_ctx, + .restore_ctx = omap2_iommu_restore_ctx, + .dump_ctx = omap2_iommu_dump_ctx, +}; + +static int __init omap2_iommu_init(void) +{ + return install_iommu_arch(&omap2_iommu_ops); +} +module_init(omap2_iommu_init); + +static void __exit omap2_iommu_exit(void) +{ + uninstall_iommu_arch(&omap2_iommu_ops); +} +module_exit(omap2_iommu_exit); + +MODULE_AUTHOR("Hiroshi DOYU, Paul Mundt and Toshihiro Kobayashi"); +MODULE_DESCRIPTION("omap iommu: omap2/3 architecture specific functions"); +MODULE_LICENSE("GPL v2"); diff --git a/arch/arm/plat-omap/include/mach/iommu2.h b/arch/arm/plat-omap/include/mach/iommu2.h new file mode 100644 index 000000000000..10ad05f410e9 --- /dev/null +++ b/arch/arm/plat-omap/include/mach/iommu2.h @@ -0,0 +1,96 @@ +/* + * omap iommu: omap2 architecture specific definitions + * + * Copyright (C) 2008-2009 Nokia Corporation + * + * Written by Hiroshi DOYU + * + * 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. + */ + +#ifndef __MACH_IOMMU2_H +#define __MACH_IOMMU2_H + +#include + +/* + * MMU Register offsets + */ +#define MMU_REVISION 0x00 +#define MMU_SYSCONFIG 0x10 +#define MMU_SYSSTATUS 0x14 +#define MMU_IRQSTATUS 0x18 +#define MMU_IRQENABLE 0x1c +#define MMU_WALKING_ST 0x40 +#define MMU_CNTL 0x44 +#define MMU_FAULT_AD 0x48 +#define MMU_TTB 0x4c +#define MMU_LOCK 0x50 +#define MMU_LD_TLB 0x54 +#define MMU_CAM 0x58 +#define MMU_RAM 0x5c +#define MMU_GFLUSH 0x60 +#define MMU_FLUSH_ENTRY 0x64 +#define MMU_READ_CAM 0x68 +#define MMU_READ_RAM 0x6c +#define MMU_EMU_FAULT_AD 0x70 + +#define MMU_REG_SIZE 256 + +/* + * MMU Register bit definitions + */ +#define MMU_LOCK_BASE_SHIFT 10 +#define MMU_LOCK_BASE_MASK (0x1f << MMU_LOCK_BASE_SHIFT) +#define MMU_LOCK_BASE(x) \ + ((x & MMU_LOCK_BASE_MASK) >> MMU_LOCK_BASE_SHIFT) + +#define MMU_LOCK_VICT_SHIFT 4 +#define MMU_LOCK_VICT_MASK (0x1f << MMU_LOCK_VICT_SHIFT) +#define MMU_LOCK_VICT(x) \ + ((x & MMU_LOCK_VICT_MASK) >> MMU_LOCK_VICT_SHIFT) + +#define MMU_CAM_VATAG_SHIFT 12 +#define MMU_CAM_VATAG_MASK \ + ((~0UL >> MMU_CAM_VATAG_SHIFT) << MMU_CAM_VATAG_SHIFT) +#define MMU_CAM_P (1 << 3) +#define MMU_CAM_V (1 << 2) +#define MMU_CAM_PGSZ_MASK 3 +#define MMU_CAM_PGSZ_1M (0 << 0) +#define MMU_CAM_PGSZ_64K (1 << 0) +#define MMU_CAM_PGSZ_4K (2 << 0) +#define MMU_CAM_PGSZ_16M (3 << 0) + +#define MMU_RAM_PADDR_SHIFT 12 +#define MMU_RAM_PADDR_MASK \ + ((~0UL >> MMU_RAM_PADDR_SHIFT) << MMU_RAM_PADDR_SHIFT) +#define MMU_RAM_ENDIAN_SHIFT 9 +#define MMU_RAM_ENDIAN_MASK (1 << MMU_RAM_ENDIAN_SHIFT) +#define MMU_RAM_ENDIAN_BIG (1 << MMU_RAM_ENDIAN_SHIFT) +#define MMU_RAM_ENDIAN_LITTLE (0 << MMU_RAM_ENDIAN_SHIFT) +#define MMU_RAM_ELSZ_SHIFT 7 +#define MMU_RAM_ELSZ_MASK (3 << MMU_RAM_ELSZ_SHIFT) +#define MMU_RAM_ELSZ_8 (0 << MMU_RAM_ELSZ_SHIFT) +#define MMU_RAM_ELSZ_16 (1 << MMU_RAM_ELSZ_SHIFT) +#define MMU_RAM_ELSZ_32 (2 << MMU_RAM_ELSZ_SHIFT) +#define MMU_RAM_ELSZ_NONE (3 << MMU_RAM_ELSZ_SHIFT) +#define MMU_RAM_MIXED_SHIFT 6 +#define MMU_RAM_MIXED_MASK (1 << MMU_RAM_MIXED_SHIFT) +#define MMU_RAM_MIXED MMU_RAM_MIXED_MASK + +/* + * register accessors + */ +static inline u32 iommu_read_reg(struct iommu *obj, size_t offs) +{ + return __raw_readl(obj->regbase + offs); +} + +static inline void iommu_write_reg(struct iommu *obj, u32 val, size_t offs) +{ + __raw_writel(val, obj->regbase + offs); +} + +#endif /* __MACH_IOMMU2_H */ From 066aa9c1e3d0af52619fe26f05810990ff381d8c Mon Sep 17 00:00:00 2001 From: Hiroshi DOYU Date: Wed, 28 Jan 2009 21:32:04 +0200 Subject: [PATCH 3/7] omap iommu: omap3 iommu device registration Signed-off-by: Hiroshi DOYU --- arch/arm/mach-omap2/omap3-iommu.c | 103 ++++++++++++++++++++++++++++++ 1 file changed, 103 insertions(+) create mode 100644 arch/arm/mach-omap2/omap3-iommu.c diff --git a/arch/arm/mach-omap2/omap3-iommu.c b/arch/arm/mach-omap2/omap3-iommu.c new file mode 100644 index 000000000000..91ee38a485a4 --- /dev/null +++ b/arch/arm/mach-omap2/omap3-iommu.c @@ -0,0 +1,103 @@ +/* + * omap iommu: omap3 device registration + * + * Copyright (C) 2008-2009 Nokia Corporation + * + * Written by Hiroshi DOYU + * + * 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 + +#define OMAP3_MMU1_BASE 0x480bd400 +#define OMAP3_MMU2_BASE 0x5d000000 +#define OMAP3_MMU1_IRQ 24 +#define OMAP3_MMU2_IRQ 28 + + +static unsigned long iommu_base[] __initdata = { + OMAP3_MMU1_BASE, + OMAP3_MMU2_BASE, +}; + +static int iommu_irq[] __initdata = { + OMAP3_MMU1_IRQ, + OMAP3_MMU2_IRQ, +}; + +static const struct iommu_platform_data omap3_iommu_pdata[] __initconst = { + { + .name = "isp", + .nr_tlb_entries = 8, + .clk_name = "cam_ick", + }, + { + .name = "iva2", + .nr_tlb_entries = 32, + .clk_name = "iva2_ck", + }, +}; +#define NR_IOMMU_DEVICES ARRAY_SIZE(omap3_iommu_pdata) + +static struct platform_device *omap3_iommu_pdev[NR_IOMMU_DEVICES]; + +static int __init omap3_iommu_init(void) +{ + int i, err; + + for (i = 0; i < NR_IOMMU_DEVICES; i++) { + struct platform_device *pdev; + struct resource res[2]; + + pdev = platform_device_alloc("omap-iommu", i); + if (!pdev) { + err = -ENOMEM; + goto err_out; + } + + memset(res, 0, sizeof(res)); + res[0].start = iommu_base[i]; + res[0].end = iommu_base[i] + MMU_REG_SIZE - 1; + res[0].flags = IORESOURCE_MEM; + res[1].start = res[1].end = iommu_irq[i]; + res[1].flags = IORESOURCE_IRQ; + + err = platform_device_add_resources(pdev, res, + ARRAY_SIZE(res)); + if (err) + goto err_out; + err = platform_device_add_data(pdev, &omap3_iommu_pdata[i], + sizeof(omap3_iommu_pdata[0])); + if (err) + goto err_out; + err = platform_device_add(pdev); + if (err) + goto err_out; + omap3_iommu_pdev[i] = pdev; + } + return 0; + +err_out: + while (i--) + platform_device_put(omap3_iommu_pdev[i]); + return err; +} +module_init(omap3_iommu_init); + +static void __exit omap3_iommu_exit(void) +{ + int i; + + for (i = 0; i < NR_IOMMU_DEVICES; i++) + platform_device_unregister(omap3_iommu_pdev[i]); +} +module_exit(omap3_iommu_exit); + +MODULE_AUTHOR("Hiroshi DOYU"); +MODULE_DESCRIPTION("omap iommu: omap3 device registration"); +MODULE_LICENSE("GPL v2"); From 69d3a84a646d6ad6cd693a7a3d5b9af414113d2c Mon Sep 17 00:00:00 2001 From: Hiroshi DOYU Date: Wed, 28 Jan 2009 21:32:08 +0200 Subject: [PATCH 4/7] omap iommu: simple virtual address space management This patch provides a device drivers, which has a omap iommu, with address mapping APIs between device virtual address(iommu), physical address and MPU virtual address. There are 4 possible patterns for iommu virtual address(iova/da) mapping. |iova/ mapping iommu_ page | da pa va (d)-(p)-(v) function type --------------------------------------------------------------------------- 1 | c c c 1 - 1 - 1 _kmap() / _kunmap() s 2 | c c,a c 1 - 1 - 1 _kmalloc()/ _kfree() s 3 | c d c 1 - n - 1 _vmap() / _vunmap() s 4 | c d,a c 1 - n - 1 _vmalloc()/ _vfree() n* 'iova': device iommu virtual address 'da': alias of 'iova' 'pa': physical address 'va': mpu virtual address 'c': contiguous memory area 'd': dicontiguous memory area 'a': anonymous memory allocation '()': optional feature 'n': a normal page(4KB) size is used. 's': multiple iommu superpage(16MB, 1MB, 64KB, 4KB) size is used. '*': not yet, but feasible. Signed-off-by: Hiroshi DOYU --- arch/arm/include/asm/mach/map.h | 8 + arch/arm/mm/ioremap.c | 6 + arch/arm/mm/mmu.c | 1 + arch/arm/plat-omap/include/mach/iovmm.h | 94 +++ arch/arm/plat-omap/iovmm.c | 896 ++++++++++++++++++++++++ 5 files changed, 1005 insertions(+) create mode 100644 arch/arm/plat-omap/include/mach/iovmm.h create mode 100644 arch/arm/plat-omap/iovmm.c diff --git a/arch/arm/include/asm/mach/map.h b/arch/arm/include/asm/mach/map.h index 58cf91f38e6f..742c2aaeb020 100644 --- a/arch/arm/include/asm/mach/map.h +++ b/arch/arm/include/asm/mach/map.h @@ -30,6 +30,14 @@ struct map_desc { #ifdef CONFIG_MMU extern void iotable_init(struct map_desc *, int); + +struct mem_type; +extern const struct mem_type *get_mem_type(unsigned int type); +/* + * external interface to remap single page with appropriate type + */ +extern int ioremap_page(unsigned long virt, unsigned long phys, + const struct mem_type *mtype); #else #define iotable_init(map,num) do { } while (0) #endif diff --git a/arch/arm/mm/ioremap.c b/arch/arm/mm/ioremap.c index 9f88dd3be601..0ab75c60f7cf 100644 --- a/arch/arm/mm/ioremap.c +++ b/arch/arm/mm/ioremap.c @@ -110,6 +110,12 @@ static int remap_area_pages(unsigned long start, unsigned long pfn, return err; } +int ioremap_page(unsigned long virt, unsigned long phys, + const struct mem_type *mtype) +{ + return remap_area_pages(virt, __phys_to_pfn(phys), PAGE_SIZE, mtype); +} +EXPORT_SYMBOL(ioremap_page); void __check_kvm_seq(struct mm_struct *mm) { diff --git a/arch/arm/mm/mmu.c b/arch/arm/mm/mmu.c index e6344ece00ce..70974d75a075 100644 --- a/arch/arm/mm/mmu.c +++ b/arch/arm/mm/mmu.c @@ -255,6 +255,7 @@ const struct mem_type *get_mem_type(unsigned int type) { return type < ARRAY_SIZE(mem_types) ? &mem_types[type] : NULL; } +EXPORT_SYMBOL(get_mem_type); /* * Adjust the PMD section entries according to the CPU in use. diff --git a/arch/arm/plat-omap/include/mach/iovmm.h b/arch/arm/plat-omap/include/mach/iovmm.h new file mode 100644 index 000000000000..bdc7ce5d7a4a --- /dev/null +++ b/arch/arm/plat-omap/include/mach/iovmm.h @@ -0,0 +1,94 @@ +/* + * omap iommu: simple virtual address space management + * + * Copyright (C) 2008-2009 Nokia Corporation + * + * Written by Hiroshi DOYU + * + * 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. + */ + +#ifndef __IOMMU_MMAP_H +#define __IOMMU_MMAP_H + +struct iovm_struct { + struct iommu *iommu; /* iommu object which this belongs to */ + u32 da_start; /* area definition */ + u32 da_end; + u32 flags; /* IOVMF_: see below */ + struct list_head list; /* linked in ascending order */ + const struct sg_table *sgt; /* keep 'page' <-> 'da' mapping */ + void *va; /* mpu side mapped address */ +}; + +/* + * IOVMF_FLAGS: attribute for iommu virtual memory area(iovma) + * + * lower 16 bit is used for h/w and upper 16 bit is for s/w. + */ +#define IOVMF_SW_SHIFT 16 +#define IOVMF_HW_SIZE (1 << IOVMF_SW_SHIFT) +#define IOVMF_HW_MASK (IOVMF_HW_SIZE - 1) +#define IOVMF_SW_MASK (~IOVMF_HW_MASK)UL + +/* + * iovma: h/w flags derived from cam and ram attribute + */ +#define IOVMF_CAM_MASK (~((1 << 10) - 1)) +#define IOVMF_RAM_MASK (~IOVMF_CAM_MASK) + +#define IOVMF_PGSZ_MASK (3 << 0) +#define IOVMF_PGSZ_1M MMU_CAM_PGSZ_1M +#define IOVMF_PGSZ_64K MMU_CAM_PGSZ_64K +#define IOVMF_PGSZ_4K MMU_CAM_PGSZ_4K +#define IOVMF_PGSZ_16M MMU_CAM_PGSZ_16M + +#define IOVMF_ENDIAN_MASK (1 << 9) +#define IOVMF_ENDIAN_BIG MMU_RAM_ENDIAN_BIG +#define IOVMF_ENDIAN_LITTLE MMU_RAM_ENDIAN_LITTLE + +#define IOVMF_ELSZ_MASK (3 << 7) +#define IOVMF_ELSZ_8 MMU_RAM_ELSZ_8 +#define IOVMF_ELSZ_16 MMU_RAM_ELSZ_16 +#define IOVMF_ELSZ_32 MMU_RAM_ELSZ_32 +#define IOVMF_ELSZ_NONE MMU_RAM_ELSZ_NONE + +#define IOVMF_MIXED_MASK (1 << 6) +#define IOVMF_MIXED MMU_RAM_MIXED + +/* + * iovma: s/w flags, used for mapping and umapping internally. + */ +#define IOVMF_MMIO (1 << IOVMF_SW_SHIFT) +#define IOVMF_ALLOC (2 << IOVMF_SW_SHIFT) +#define IOVMF_ALLOC_MASK (3 << IOVMF_SW_SHIFT) + +/* "superpages" is supported just with physically linear pages */ +#define IOVMF_DISCONT (1 << (2 + IOVMF_SW_SHIFT)) +#define IOVMF_LINEAR (2 << (2 + IOVMF_SW_SHIFT)) +#define IOVMF_LINEAR_MASK (3 << (2 + IOVMF_SW_SHIFT)) + +#define IOVMF_DA_FIXED (1 << (4 + IOVMF_SW_SHIFT)) +#define IOVMF_DA_ANON (2 << (4 + IOVMF_SW_SHIFT)) +#define IOVMF_DA_MASK (3 << (4 + IOVMF_SW_SHIFT)) + + +extern struct iovm_struct *find_iovm_area(struct iommu *obj, u32 da); +extern u32 iommu_vmap(struct iommu *obj, u32 da, + const struct sg_table *sgt, u32 flags); +extern struct sg_table *iommu_vunmap(struct iommu *obj, u32 da); +extern u32 iommu_vmalloc(struct iommu *obj, u32 da, size_t bytes, + u32 flags); +extern void iommu_vfree(struct iommu *obj, const u32 da); +extern u32 iommu_kmap(struct iommu *obj, u32 da, u32 pa, size_t bytes, + u32 flags); +extern void iommu_kunmap(struct iommu *obj, u32 da); +extern u32 iommu_kmalloc(struct iommu *obj, u32 da, size_t bytes, + u32 flags); +extern void iommu_kfree(struct iommu *obj, u32 da); + +extern void *da_to_va(struct iommu *obj, u32 da); + +#endif /* __IOMMU_MMAP_H */ diff --git a/arch/arm/plat-omap/iovmm.c b/arch/arm/plat-omap/iovmm.c new file mode 100644 index 000000000000..2fce2c151a95 --- /dev/null +++ b/arch/arm/plat-omap/iovmm.c @@ -0,0 +1,896 @@ +/* + * omap iommu: simple virtual address space management + * + * Copyright (C) 2008-2009 Nokia Corporation + * + * Written by Hiroshi DOYU + * + * 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 "iopgtable.h" + +/* + * A device driver needs to create address mappings between: + * + * - iommu/device address + * - physical address + * - mpu virtual address + * + * There are 4 possible patterns for them: + * + * |iova/ mapping iommu_ page + * | da pa va (d)-(p)-(v) function type + * --------------------------------------------------------------------------- + * 1 | c c c 1 - 1 - 1 _kmap() / _kunmap() s + * 2 | c c,a c 1 - 1 - 1 _kmalloc()/ _kfree() s + * 3 | c d c 1 - n - 1 _vmap() / _vunmap() s + * 4 | c d,a c 1 - n - 1 _vmalloc()/ _vfree() n* + * + * + * 'iova': device iommu virtual address + * 'da': alias of 'iova' + * 'pa': physical address + * 'va': mpu virtual address + * + * 'c': contiguous memory area + * 'd': dicontiguous memory area + * 'a': anonymous memory allocation + * '()': optional feature + * + * 'n': a normal page(4KB) size is used. + * 's': multiple iommu superpage(16MB, 1MB, 64KB, 4KB) size is used. + * + * '*': not yet, but feasible. + */ + +static struct kmem_cache *iovm_area_cachep; + +/* return total bytes of sg buffers */ +static size_t sgtable_len(const struct sg_table *sgt) +{ + unsigned int i, total = 0; + struct scatterlist *sg; + + if (!sgt) + return 0; + + for_each_sg(sgt->sgl, sg, sgt->nents, i) { + size_t bytes; + + bytes = sg_dma_len(sg); + + if (!iopgsz_ok(bytes)) { + pr_err("%s: sg[%d] not iommu pagesize(%x)\n", + __func__, i, bytes); + return 0; + } + + total += bytes; + } + + return total; +} +#define sgtable_ok(x) (!!sgtable_len(x)) + +/* + * calculate the optimal number sg elements from total bytes based on + * iommu superpages + */ +static unsigned int sgtable_nents(size_t bytes) +{ + int i; + unsigned int nr_entries; + const unsigned long pagesize[] = { SZ_16M, SZ_1M, SZ_64K, SZ_4K, }; + + if (!IS_ALIGNED(bytes, PAGE_SIZE)) { + pr_err("%s: wrong size %08x\n", __func__, bytes); + return 0; + } + + nr_entries = 0; + for (i = 0; i < ARRAY_SIZE(pagesize); i++) { + if (bytes >= pagesize[i]) { + nr_entries += (bytes / pagesize[i]); + bytes %= pagesize[i]; + } + } + BUG_ON(bytes); + + return nr_entries; +} + +/* allocate and initialize sg_table header(a kind of 'superblock') */ +static struct sg_table *sgtable_alloc(const size_t bytes, u32 flags) +{ + unsigned int nr_entries; + int err; + struct sg_table *sgt; + + if (!bytes) + return ERR_PTR(-EINVAL); + + if (!IS_ALIGNED(bytes, PAGE_SIZE)) + return ERR_PTR(-EINVAL); + + /* FIXME: IOVMF_DA_FIXED should support 'superpages' */ + if ((flags & IOVMF_LINEAR) && (flags & IOVMF_DA_ANON)) { + nr_entries = sgtable_nents(bytes); + if (!nr_entries) + return ERR_PTR(-EINVAL); + } else + nr_entries = bytes / PAGE_SIZE; + + sgt = kzalloc(sizeof(*sgt), GFP_KERNEL); + if (!sgt) + return ERR_PTR(-ENOMEM); + + err = sg_alloc_table(sgt, nr_entries, GFP_KERNEL); + if (err) + return ERR_PTR(err); + + pr_debug("%s: sgt:%p(%d entries)\n", __func__, sgt, nr_entries); + + return sgt; +} + +/* free sg_table header(a kind of superblock) */ +static void sgtable_free(struct sg_table *sgt) +{ + if (!sgt) + return; + + sg_free_table(sgt); + kfree(sgt); + + pr_debug("%s: sgt:%p\n", __func__, sgt); +} + +/* map 'sglist' to a contiguous mpu virtual area and return 'va' */ +static void *vmap_sg(const struct sg_table *sgt) +{ + u32 va; + size_t total; + unsigned int i; + struct scatterlist *sg; + struct vm_struct *new; + const struct mem_type *mtype; + + mtype = get_mem_type(MT_DEVICE); + if (!mtype) + return ERR_PTR(-EINVAL); + + total = sgtable_len(sgt); + if (!total) + return ERR_PTR(-EINVAL); + + new = __get_vm_area(total, VM_IOREMAP, VMALLOC_START, VMALLOC_END); + if (!new) + return ERR_PTR(-ENOMEM); + va = (u32)new->addr; + + for_each_sg(sgt->sgl, sg, sgt->nents, i) { + size_t bytes; + u32 pa; + int err; + + pa = sg_phys(sg); + bytes = sg_dma_len(sg); + + BUG_ON(bytes != PAGE_SIZE); + + err = ioremap_page(va, pa, mtype); + if (err) + goto err_out; + + va += bytes; + } + + flush_cache_vmap(new->addr, total); + return new->addr; + +err_out: + WARN_ON(1); /* FIXME: cleanup some mpu mappings */ + vunmap(new->addr); + return ERR_PTR(-EAGAIN); +} + +static inline void vunmap_sg(const void *va) +{ + vunmap(va); +} + +static struct iovm_struct *__find_iovm_area(struct iommu *obj, const u32 da) +{ + struct iovm_struct *tmp; + + list_for_each_entry(tmp, &obj->mmap, list) { + if ((da >= tmp->da_start) && (da < tmp->da_end)) { + size_t len; + + len = tmp->da_end - tmp->da_start; + + dev_dbg(obj->dev, "%s: %08x-%08x-%08x(%x) %08x\n", + __func__, tmp->da_start, da, tmp->da_end, len, + tmp->flags); + + return tmp; + } + } + + return NULL; +} + +/** + * find_iovm_area - find iovma which includes @da + * @da: iommu device virtual address + * + * Find the existing iovma starting at @da + */ +struct iovm_struct *find_iovm_area(struct iommu *obj, u32 da) +{ + struct iovm_struct *area; + + mutex_lock(&obj->mmap_lock); + area = __find_iovm_area(obj, da); + mutex_unlock(&obj->mmap_lock); + + return area; +} +EXPORT_SYMBOL_GPL(find_iovm_area); + +/* + * This finds the hole(area) which fits the requested address and len + * in iovmas mmap, and returns the new allocated iovma. + */ +static struct iovm_struct *alloc_iovm_area(struct iommu *obj, u32 da, + size_t bytes, u32 flags) +{ + struct iovm_struct *new, *tmp; + u32 start, prev_end, alignement; + + if (!obj || !bytes) + return ERR_PTR(-EINVAL); + + start = da; + alignement = PAGE_SIZE; + + if (flags & IOVMF_DA_ANON) { + /* + * Reserve the first page for NULL + */ + start = PAGE_SIZE; + if (flags & IOVMF_LINEAR) + alignement = iopgsz_max(bytes); + start = roundup(start, alignement); + } + + tmp = NULL; + if (list_empty(&obj->mmap)) + goto found; + + prev_end = 0; + list_for_each_entry(tmp, &obj->mmap, list) { + + if ((prev_end <= start) && (start + bytes < tmp->da_start)) + goto found; + + if (flags & IOVMF_DA_ANON) + start = roundup(tmp->da_end, alignement); + + prev_end = tmp->da_end; + } + + if ((start >= prev_end) && (ULONG_MAX - start >= bytes)) + goto found; + + dev_dbg(obj->dev, "%s: no space to fit %08x(%x) flags: %08x\n", + __func__, da, bytes, flags); + + return ERR_PTR(-EINVAL); + +found: + new = kmem_cache_zalloc(iovm_area_cachep, GFP_KERNEL); + if (!new) + return ERR_PTR(-ENOMEM); + + new->iommu = obj; + new->da_start = start; + new->da_end = start + bytes; + new->flags = flags; + + /* + * keep ascending order of iovmas + */ + if (tmp) + list_add_tail(&new->list, &tmp->list); + else + list_add(&new->list, &obj->mmap); + + dev_dbg(obj->dev, "%s: found %08x-%08x-%08x(%x) %08x\n", + __func__, new->da_start, start, new->da_end, bytes, flags); + + return new; +} + +static void free_iovm_area(struct iommu *obj, struct iovm_struct *area) +{ + size_t bytes; + + BUG_ON(!obj || !area); + + bytes = area->da_end - area->da_start; + + dev_dbg(obj->dev, "%s: %08x-%08x(%x) %08x\n", + __func__, area->da_start, area->da_end, bytes, area->flags); + + list_del(&area->list); + kmem_cache_free(iovm_area_cachep, area); +} + +/** + * da_to_va - convert (d) to (v) + * @obj: objective iommu + * @da: iommu device virtual address + * @va: mpu virtual address + * + * Returns mpu virtual addr which corresponds to a given device virtual addr + */ +void *da_to_va(struct iommu *obj, u32 da) +{ + void *va = NULL; + struct iovm_struct *area; + + mutex_lock(&obj->mmap_lock); + + area = __find_iovm_area(obj, da); + if (!area) { + dev_dbg(obj->dev, "%s: no da area(%08x)\n", __func__, da); + goto out; + } + va = area->va; + mutex_unlock(&obj->mmap_lock); +out: + return va; +} +EXPORT_SYMBOL_GPL(da_to_va); + +static void sgtable_fill_vmalloc(struct sg_table *sgt, void *_va) +{ + unsigned int i; + struct scatterlist *sg; + void *va = _va; + void *va_end; + + for_each_sg(sgt->sgl, sg, sgt->nents, i) { + struct page *pg; + const size_t bytes = PAGE_SIZE; + + /* + * iommu 'superpage' isn't supported with 'iommu_vmalloc()' + */ + pg = vmalloc_to_page(va); + BUG_ON(!pg); + sg_set_page(sg, pg, bytes, 0); + + va += bytes; + } + + va_end = _va + PAGE_SIZE * i; + flush_cache_vmap(_va, va_end); +} + +static inline void sgtable_drain_vmalloc(struct sg_table *sgt) +{ + /* + * Actually this is not necessary at all, just exists for + * consistency of the code readibility. + */ + BUG_ON(!sgt); +} + +static void sgtable_fill_kmalloc(struct sg_table *sgt, u32 pa, size_t len) +{ + unsigned int i; + struct scatterlist *sg; + void *va; + + va = phys_to_virt(pa); + + for_each_sg(sgt->sgl, sg, sgt->nents, i) { + size_t bytes; + + bytes = iopgsz_max(len); + + BUG_ON(!iopgsz_ok(bytes)); + + sg_set_buf(sg, phys_to_virt(pa), bytes); + /* + * 'pa' is cotinuous(linear). + */ + pa += bytes; + len -= bytes; + } + BUG_ON(len); + + clean_dcache_area(va, len); +} + +static inline void sgtable_drain_kmalloc(struct sg_table *sgt) +{ + /* + * Actually this is not necessary at all, just exists for + * consistency of the code readibility + */ + BUG_ON(!sgt); +} + +/* create 'da' <-> 'pa' mapping from 'sgt' */ +static int map_iovm_area(struct iommu *obj, struct iovm_struct *new, + const struct sg_table *sgt, u32 flags) +{ + int err; + unsigned int i, j; + struct scatterlist *sg; + u32 da = new->da_start; + + if (!obj || !new || !sgt) + return -EINVAL; + + BUG_ON(!sgtable_ok(sgt)); + + for_each_sg(sgt->sgl, sg, sgt->nents, i) { + u32 pa; + int pgsz; + size_t bytes; + struct iotlb_entry e; + + pa = sg_phys(sg); + bytes = sg_dma_len(sg); + + flags &= ~IOVMF_PGSZ_MASK; + pgsz = bytes_to_iopgsz(bytes); + if (pgsz < 0) + goto err_out; + flags |= pgsz; + + pr_debug("%s: [%d] %08x %08x(%x)\n", __func__, + i, da, pa, bytes); + + iotlb_init_entry(&e, da, pa, flags); + err = iopgtable_store_entry(obj, &e); + if (err) + goto err_out; + + da += bytes; + } + return 0; + +err_out: + da = new->da_start; + + for_each_sg(sgt->sgl, sg, i, j) { + size_t bytes; + + bytes = iopgtable_clear_entry(obj, da); + + BUG_ON(!iopgsz_ok(bytes)); + + da += bytes; + } + return err; +} + +/* release 'da' <-> 'pa' mapping */ +static void unmap_iovm_area(struct iommu *obj, struct iovm_struct *area) +{ + u32 start; + size_t total = area->da_end - area->da_start; + + BUG_ON((!total) || !IS_ALIGNED(total, PAGE_SIZE)); + + start = area->da_start; + while (total > 0) { + size_t bytes; + + bytes = iopgtable_clear_entry(obj, start); + if (bytes == 0) + bytes = PAGE_SIZE; + else + dev_dbg(obj->dev, "%s: unmap %08x(%x) %08x\n", + __func__, start, bytes, area->flags); + + BUG_ON(!IS_ALIGNED(bytes, PAGE_SIZE)); + + total -= bytes; + start += bytes; + } + BUG_ON(total); +} + +/* template function for all unmapping */ +static struct sg_table *unmap_vm_area(struct iommu *obj, const u32 da, + void (*fn)(const void *), u32 flags) +{ + struct sg_table *sgt = NULL; + struct iovm_struct *area; + + if (!IS_ALIGNED(da, PAGE_SIZE)) { + dev_err(obj->dev, "%s: alignment err(%08x)\n", __func__, da); + return NULL; + } + + mutex_lock(&obj->mmap_lock); + + area = __find_iovm_area(obj, da); + if (!area) { + dev_dbg(obj->dev, "%s: no da area(%08x)\n", __func__, da); + goto out; + } + + if ((area->flags & flags) != flags) { + dev_err(obj->dev, "%s: wrong flags(%08x)\n", __func__, + area->flags); + goto out; + } + sgt = (struct sg_table *)area->sgt; + + unmap_iovm_area(obj, area); + + fn(area->va); + + dev_dbg(obj->dev, "%s: %08x-%08x-%08x(%x) %08x\n", __func__, + area->da_start, da, area->da_end, + area->da_end - area->da_start, area->flags); + + free_iovm_area(obj, area); +out: + mutex_unlock(&obj->mmap_lock); + + return sgt; +} + +static u32 map_iommu_region(struct iommu *obj, u32 da, + const struct sg_table *sgt, void *va, size_t bytes, u32 flags) +{ + int err = -ENOMEM; + struct iovm_struct *new; + + mutex_lock(&obj->mmap_lock); + + new = alloc_iovm_area(obj, da, bytes, flags); + if (IS_ERR(new)) { + err = PTR_ERR(new); + goto err_alloc_iovma; + } + new->va = va; + new->sgt = sgt; + + if (map_iovm_area(obj, new, sgt, new->flags)) + goto err_map; + + mutex_unlock(&obj->mmap_lock); + + dev_dbg(obj->dev, "%s: da:%08x(%x) flags:%08x va:%p\n", + __func__, new->da_start, bytes, new->flags, va); + + return new->da_start; + +err_map: + free_iovm_area(obj, new); +err_alloc_iovma: + mutex_unlock(&obj->mmap_lock); + return err; +} + +static inline u32 __iommu_vmap(struct iommu *obj, u32 da, + const struct sg_table *sgt, void *va, size_t bytes, u32 flags) +{ + return map_iommu_region(obj, da, sgt, va, bytes, flags); +} + +/** + * iommu_vmap - (d)-(p)-(v) address mapper + * @obj: objective iommu + * @sgt: address of scatter gather table + * @flags: iovma and page property + * + * Creates 1-n-1 mapping with given @sgt and returns @da. + * All @sgt element must be io page size aligned. + */ +u32 iommu_vmap(struct iommu *obj, u32 da, const struct sg_table *sgt, + u32 flags) +{ + size_t bytes; + void *va; + + if (!obj || !obj->dev || !sgt) + return -EINVAL; + + bytes = sgtable_len(sgt); + if (!bytes) + return -EINVAL; + bytes = PAGE_ALIGN(bytes); + + va = vmap_sg(sgt); + if (IS_ERR(va)) + return PTR_ERR(va); + + flags &= IOVMF_HW_MASK; + flags |= IOVMF_DISCONT; + flags |= IOVMF_MMIO; + flags |= (da ? IOVMF_DA_FIXED : IOVMF_DA_ANON); + + da = __iommu_vmap(obj, da, sgt, va, bytes, flags); + if (IS_ERR_VALUE(da)) + vunmap_sg(va); + + return da; +} +EXPORT_SYMBOL_GPL(iommu_vmap); + +/** + * iommu_vunmap - release virtual mapping obtained by 'iommu_vmap()' + * @obj: objective iommu + * @da: iommu device virtual address + * + * Free the iommu virtually contiguous memory area starting at + * @da, which was returned by 'iommu_vmap()'. + */ +struct sg_table *iommu_vunmap(struct iommu *obj, u32 da) +{ + struct sg_table *sgt; + /* + * 'sgt' is allocated before 'iommu_vmalloc()' is called. + * Just returns 'sgt' to the caller to free + */ + sgt = unmap_vm_area(obj, da, vunmap_sg, IOVMF_DISCONT | IOVMF_MMIO); + if (!sgt) + dev_dbg(obj->dev, "%s: No sgt\n", __func__); + return sgt; +} +EXPORT_SYMBOL_GPL(iommu_vunmap); + +/** + * iommu_vmalloc - (d)-(p)-(v) address allocator and mapper + * @obj: objective iommu + * @da: contiguous iommu virtual memory + * @bytes: allocation size + * @flags: iovma and page property + * + * Allocate @bytes linearly and creates 1-n-1 mapping and returns + * @da again, which might be adjusted if 'IOVMF_DA_ANON' is set. + */ +u32 iommu_vmalloc(struct iommu *obj, u32 da, size_t bytes, u32 flags) +{ + void *va; + struct sg_table *sgt; + + if (!obj || !obj->dev || !bytes) + return -EINVAL; + + bytes = PAGE_ALIGN(bytes); + + va = vmalloc(bytes); + if (!va) + return -ENOMEM; + + sgt = sgtable_alloc(bytes, flags); + if (IS_ERR(sgt)) { + da = PTR_ERR(sgt); + goto err_sgt_alloc; + } + sgtable_fill_vmalloc(sgt, va); + + flags &= IOVMF_HW_MASK; + flags |= IOVMF_DISCONT; + flags |= IOVMF_ALLOC; + flags |= (da ? IOVMF_DA_FIXED : IOVMF_DA_ANON); + + da = __iommu_vmap(obj, da, sgt, va, bytes, flags); + if (IS_ERR_VALUE(da)) + goto err_iommu_vmap; + + return da; + +err_iommu_vmap: + sgtable_drain_vmalloc(sgt); + sgtable_free(sgt); +err_sgt_alloc: + vfree(va); + return da; +} +EXPORT_SYMBOL_GPL(iommu_vmalloc); + +/** + * iommu_vfree - release memory allocated by 'iommu_vmalloc()' + * @obj: objective iommu + * @da: iommu device virtual address + * + * Frees the iommu virtually continuous memory area starting at + * @da, as obtained from 'iommu_vmalloc()'. + */ +void iommu_vfree(struct iommu *obj, const u32 da) +{ + struct sg_table *sgt; + + sgt = unmap_vm_area(obj, da, vfree, IOVMF_DISCONT | IOVMF_ALLOC); + if (!sgt) + dev_dbg(obj->dev, "%s: No sgt\n", __func__); + sgtable_free(sgt); +} +EXPORT_SYMBOL_GPL(iommu_vfree); + +static u32 __iommu_kmap(struct iommu *obj, u32 da, u32 pa, void *va, + size_t bytes, u32 flags) +{ + struct sg_table *sgt; + + sgt = sgtable_alloc(bytes, flags); + if (IS_ERR(sgt)) + return PTR_ERR(sgt); + + sgtable_fill_kmalloc(sgt, pa, bytes); + + da = map_iommu_region(obj, da, sgt, va, bytes, flags); + if (IS_ERR_VALUE(da)) { + sgtable_drain_kmalloc(sgt); + sgtable_free(sgt); + } + + return da; +} + +/** + * iommu_kmap - (d)-(p)-(v) address mapper + * @obj: objective iommu + * @da: contiguous iommu virtual memory + * @pa: contiguous physical memory + * @flags: iovma and page property + * + * Creates 1-1-1 mapping and returns @da again, which can be + * adjusted if 'IOVMF_DA_ANON' is set. + */ +u32 iommu_kmap(struct iommu *obj, u32 da, u32 pa, size_t bytes, + u32 flags) +{ + void *va; + + if (!obj || !obj->dev || !bytes) + return -EINVAL; + + bytes = PAGE_ALIGN(bytes); + + va = ioremap(pa, bytes); + if (!va) + return -ENOMEM; + + flags &= IOVMF_HW_MASK; + flags |= IOVMF_LINEAR; + flags |= IOVMF_MMIO; + flags |= (da ? IOVMF_DA_FIXED : IOVMF_DA_ANON); + + da = __iommu_kmap(obj, da, pa, va, bytes, flags); + if (IS_ERR_VALUE(da)) + iounmap(va); + + return da; +} +EXPORT_SYMBOL_GPL(iommu_kmap); + +/** + * iommu_kunmap - release virtual mapping obtained by 'iommu_kmap()' + * @obj: objective iommu + * @da: iommu device virtual address + * + * Frees the iommu virtually contiguous memory area starting at + * @da, which was passed to and was returned by'iommu_kmap()'. + */ +void iommu_kunmap(struct iommu *obj, u32 da) +{ + struct sg_table *sgt; + typedef void (*func_t)(const void *); + + sgt = unmap_vm_area(obj, da, (func_t)__iounmap, + IOVMF_LINEAR | IOVMF_MMIO); + if (!sgt) + dev_dbg(obj->dev, "%s: No sgt\n", __func__); + sgtable_free(sgt); +} +EXPORT_SYMBOL_GPL(iommu_kunmap); + +/** + * iommu_kmalloc - (d)-(p)-(v) address allocator and mapper + * @obj: objective iommu + * @da: contiguous iommu virtual memory + * @bytes: bytes for allocation + * @flags: iovma and page property + * + * Allocate @bytes linearly and creates 1-1-1 mapping and returns + * @da again, which might be adjusted if 'IOVMF_DA_ANON' is set. + */ +u32 iommu_kmalloc(struct iommu *obj, u32 da, size_t bytes, u32 flags) +{ + void *va; + u32 pa; + + if (!obj || !obj->dev || !bytes) + return -EINVAL; + + bytes = PAGE_ALIGN(bytes); + + va = kmalloc(bytes, GFP_KERNEL | GFP_DMA); + if (!va) + return -ENOMEM; + pa = virt_to_phys(va); + + flags &= IOVMF_HW_MASK; + flags |= IOVMF_LINEAR; + flags |= IOVMF_ALLOC; + flags |= (da ? IOVMF_DA_FIXED : IOVMF_DA_ANON); + + da = __iommu_kmap(obj, da, pa, va, bytes, flags); + if (IS_ERR_VALUE(da)) + kfree(va); + + return da; +} +EXPORT_SYMBOL_GPL(iommu_kmalloc); + +/** + * iommu_kfree - release virtual mapping obtained by 'iommu_kmalloc()' + * @obj: objective iommu + * @da: iommu device virtual address + * + * Frees the iommu virtually contiguous memory area starting at + * @da, which was passed to and was returned by'iommu_kmalloc()'. + */ +void iommu_kfree(struct iommu *obj, u32 da) +{ + struct sg_table *sgt; + + sgt = unmap_vm_area(obj, da, kfree, IOVMF_LINEAR | IOVMF_ALLOC); + if (!sgt) + dev_dbg(obj->dev, "%s: No sgt\n", __func__); + sgtable_free(sgt); +} +EXPORT_SYMBOL_GPL(iommu_kfree); + + +static int __init iovmm_init(void) +{ + const unsigned long flags = SLAB_HWCACHE_ALIGN; + struct kmem_cache *p; + + p = kmem_cache_create("iovm_area_cache", sizeof(struct iovm_struct), 0, + flags, NULL); + if (!p) + return -ENOMEM; + iovm_area_cachep = p; + + return 0; +} +module_init(iovmm_init); + +static void __exit iovmm_exit(void) +{ + kmem_cache_destroy(iovm_area_cachep); +} +module_exit(iovmm_exit); + +MODULE_DESCRIPTION("omap iommu: simple virtual address space management"); +MODULE_AUTHOR("Hiroshi DOYU "); +MODULE_LICENSE("GPL v2"); From e769421de22422461f0becae0ec6e792ec6f4476 Mon Sep 17 00:00:00 2001 From: Hiroshi DOYU Date: Wed, 28 Jan 2009 21:32:08 +0200 Subject: [PATCH 5/7] omap iommu: entries for Kconfig and Makefile Signed-off-by: Hiroshi DOYU --- arch/arm/plat-omap/Kconfig | 3 +++ arch/arm/plat-omap/Makefile | 1 + 2 files changed, 4 insertions(+) diff --git a/arch/arm/plat-omap/Kconfig b/arch/arm/plat-omap/Kconfig index 9dd68fafb374..d483b2592ec8 100644 --- a/arch/arm/plat-omap/Kconfig +++ b/arch/arm/plat-omap/Kconfig @@ -115,6 +115,9 @@ config OMAP_MBOX_FWK Say Y here if you want to use OMAP Mailbox framework support for DSP, IVA1.0 and IVA2 in OMAP1/2/3. +config OMAP_IOMMU + tristate + choice prompt "System timer" default OMAP_MPU_TIMER diff --git a/arch/arm/plat-omap/Makefile b/arch/arm/plat-omap/Makefile index 04a100cfb8e5..a83279523958 100644 --- a/arch/arm/plat-omap/Makefile +++ b/arch/arm/plat-omap/Makefile @@ -13,6 +13,7 @@ obj- := obj-$(CONFIG_ARCH_OMAP16XX) += ocpi.o obj-$(CONFIG_OMAP_MCBSP) += mcbsp.o +obj-$(CONFIG_OMAP_IOMMU) += iommu.o iovmm.o obj-$(CONFIG_CPU_FREQ) += cpu-omap.o obj-$(CONFIG_OMAP_DM_TIMER) += dmtimer.o From caf60779a6c5795340767ccf2f73ed7d5c7e0486 Mon Sep 17 00:00:00 2001 From: Hiroshi DOYU Date: Wed, 28 Jan 2009 21:32:09 +0200 Subject: [PATCH 6/7] omap2 iommu: entries for Kconfig and Makefile Signed-off-by: Hiroshi DOYU --- arch/arm/mach-omap2/Makefile | 5 +++++ 1 file changed, 5 insertions(+) diff --git a/arch/arm/mach-omap2/Makefile b/arch/arm/mach-omap2/Makefile index c49d9bfa3abd..88629a7455a6 100644 --- a/arch/arm/mach-omap2/Makefile +++ b/arch/arm/mach-omap2/Makefile @@ -28,6 +28,11 @@ endif obj-$(CONFIG_ARCH_OMAP2) += clock24xx.o obj-$(CONFIG_ARCH_OMAP3) += clock34xx.o +iommu-y += iommu2.o +iommu-$(CONFIG_ARCH_OMAP3) += omap3-iommu.o + +obj-$(CONFIG_OMAP_IOMMU) += $(iommu-y) + # Specific board support obj-$(CONFIG_MACH_OMAP_GENERIC) += board-generic.o obj-$(CONFIG_MACH_OMAP_H4) += board-h4.o From 5c651ffaee6f07aa1f5e6d0763845a2ee606b6bd Mon Sep 17 00:00:00 2001 From: Hiroshi DOYU Date: Tue, 19 May 2009 09:07:55 +0300 Subject: [PATCH 7/7] omap iommu: add MPU_BRIDGE_IOMMU for tidspbridge migration Currently "tidspbridge" driver uses its own mmu implementation and will migrate to use this "omap iommu" eventually. This config is provided to make this migration happen smoothly. Signed-off-by: Hiroshi DOYU --- arch/arm/mach-omap2/omap3-iommu.c | 2 ++ 1 file changed, 2 insertions(+) diff --git a/arch/arm/mach-omap2/omap3-iommu.c b/arch/arm/mach-omap2/omap3-iommu.c index 91ee38a485a4..194189c746c2 100644 --- a/arch/arm/mach-omap2/omap3-iommu.c +++ b/arch/arm/mach-omap2/omap3-iommu.c @@ -36,11 +36,13 @@ static const struct iommu_platform_data omap3_iommu_pdata[] __initconst = { .nr_tlb_entries = 8, .clk_name = "cam_ick", }, +#if defined(CONFIG_MPU_BRIDGE_IOMMU) { .name = "iva2", .nr_tlb_entries = 32, .clk_name = "iva2_ck", }, +#endif }; #define NR_IOMMU_DEVICES ARRAY_SIZE(omap3_iommu_pdata)