The generic API is intended to be separated from the implementation of page table algorithms. It contains only accessors for walking and manipulating the table and helpers that are useful for building an implementation. Memory management is not in the generic API, but part of the implementation.

Using a multi-compilation approach the implementation module would include headers in this order:

common.h defs_FMT.h pt_defs.h FMT.h pt_common.h IMPLEMENTATION.h

Where each compilation unit would have a combination of FMT and IMPLEMENTATION to produce a per-format per-implementation module.

The API is designed so that the format headers have minimal logic, and default implementations are provided if the format doesn't include one.

Generally formats provide their code via an inline function using the pattern:

static inline FMTpt_XX(..) {} #define pt_XX FMTpt_XX

The common code then enforces a function signature so that there is no drift in function arguments, or accidental polymorphic functions (as has been slightly troublesome in mm). Use of function-like #defines are avoided in the format even though many of the functions are small enough.

Provide kdocs for the API surface.

This is enough to implement the 8 initial format variations with all of their features: * Entries comprised of contiguous blocks of IO PTEs for larger page sizes (AMDv1, ARMv8) * Multi-level tables, up to 6 levels. Runtime selected top level * Runtime variable table level size (ARM's concatenated tables) * Expandable top level (AMDv1) * Optional leaf entries at any level * 32 bit/64 bit virtual and output addresses, using every bit * Sign extended addressing (x86) * Dirty tracking

A basic simple format takes about 200 lines to declare the require inline functions.

Tested-by: Alejandro Jimenez alejandro.j.jimenez@oracle.com Signed-off-by: Jason Gunthorpe jgg@nvidia.com --- .clang-format | 1 + drivers/iommu/Kconfig | 2 + drivers/iommu/generic_pt/Kconfig | 20 + drivers/iommu/generic_pt/pt_common.h | 355 ++++++++++++ drivers/iommu/generic_pt/pt_defs.h | 323 +++++++++++ drivers/iommu/generic_pt/pt_fmt_defaults.h | 193 +++++++ drivers/iommu/generic_pt/pt_iter.h | 636 +++++++++++++++++++++ drivers/iommu/generic_pt/pt_log2.h | 130 +++++ include/linux/generic_pt/common.h | 134 +++++ 9 files changed, 1794 insertions(+) create mode 100644 drivers/iommu/generic_pt/Kconfig create mode 100644 drivers/iommu/generic_pt/pt_common.h create mode 100644 drivers/iommu/generic_pt/pt_defs.h create mode 100644 drivers/iommu/generic_pt/pt_fmt_defaults.h create mode 100644 drivers/iommu/generic_pt/pt_iter.h create mode 100644 drivers/iommu/generic_pt/pt_log2.h create mode 100644 include/linux/generic_pt/common.h

diff --git a/.clang-format b/.clang-format index 48405c54ef271e..43fd61dd2e2092 100644 --- a/.clang-format +++ b/.clang-format @@ -416,6 +416,7 @@ ForEachMacros: - 'for_each_prop_dlc_cpus' - 'for_each_prop_dlc_platforms' - 'for_each_property_of_node' + - 'for_each_pt_level_entry' - 'for_each_rdt_resource' - 'for_each_reg' - 'for_each_reg_filtered' diff --git a/drivers/iommu/Kconfig b/drivers/iommu/Kconfig index 70d29b14d85196..c9ae3221cd6f50 100644 --- a/drivers/iommu/Kconfig +++ b/drivers/iommu/Kconfig @@ -384,3 +384,5 @@ config SPRD_IOMMU Say Y here if you want to use the multimedia devices listed above.

endif # IOMMU_SUPPORT + +source "drivers/iommu/generic_pt/Kconfig" diff --git a/drivers/iommu/generic_pt/Kconfig b/drivers/iommu/generic_pt/Kconfig new file mode 100644 index 00000000000000..fb0f431ddba0a8 --- /dev/null +++ b/drivers/iommu/generic_pt/Kconfig @@ -0,0 +1,20 @@ +# SPDX-License-Identifier: GPL-2.0-only + +menuconfig GENERIC_PT + bool "Generic Radix Page Table" + help + Generic library for building radix tree page tables. + + Generic PT provides a set of HW page table formats and a common + set of APIs to work with them. + +if GENERIC_PT +config DEBUG_GENERIC_PT + bool "Extra debugging checks for GENERIC_PT" + help + Enable extra run time debugging checks for GENERIC_PT code. This + incurs a runtime cost and should not be enabled for production + kernels. + + The kunit tests require this to be enabled to get full coverage. +endif diff --git a/drivers/iommu/generic_pt/pt_common.h b/drivers/iommu/generic_pt/pt_common.h new file mode 100644 index 00000000000000..e9a071c93d6c63 --- /dev/null +++ b/drivers/iommu/generic_pt/pt_common.h @@ -0,0 +1,355 @@ +/* SPDX-License-Identifier: GPL-2.0-only */ +/* + * Copyright (c) 2024-2025, NVIDIA CORPORATION & AFFILIATES + * + * This header is included after the format. It contains definitions + * that build on the format definitions to create the basic format API. + * + * The format API is listed here, with kdocs, in alphabetical order. The + * functions without bodies are implemented in the format using the pattern: + * static inline FMTpt_XXX(..) {..} + * #define pt_XXX FMTpt_XXX + * + * The routines marked "@pts: Entry to query" operate on the entire contiguous + * entry and can be called with a pts->index pointing to any sub item that makes + * up that entry. + * + * The header order is: + * pt_defs.h + * fmt_XX.h + * pt_common.h + */ +#ifndef __GENERIC_PT_PT_COMMON_H +#define __GENERIC_PT_PT_COMMON_H + +#include "pt_defs.h" +#include "pt_fmt_defaults.h" + +/** + * pt_attr_from_entry() - Convert the permission bits back to attrs + * @pts: Entry to convert from + * @attrs: Resulting attrs + * + * Fill in the attrs with the permission bits encoded in the current leaf entry. + * The attrs should be usable with pt_install_leaf_entry() to reconstruct the + * same entry. + */ +static inline void pt_attr_from_entry(const struct pt_state *pts, + struct pt_write_attrs *attrs); + +/** + * pt_can_have_leaf() - True if the current level can have an OA entry + * @pts: The current level + * + * True if the current level can support pt_install_leaf_entry(). A leaf + * entry produce an OA. + */ +static inline bool pt_can_have_leaf(const struct pt_state *pts); + +/** + * pt_can_have_table() - True if the current level can have a lower table + * @pts: The current level + * + * Every level except 0 is allowed to have a lower table. + */ +static inline bool pt_can_have_table(const struct pt_state *pts) +{ + /* No further tables at level 0 */ + return pts->level > 0; +} + +/** + * pt_clear_entry() - Make entries empty (non-present) + * @pts: Starting table index + * @num_contig_lg2: Number of contiguous items to clear + * + * Clear a run of entries. A cleared entry will load back as PT_ENTRY_EMPTY + * and does not have any effect on table walking. The starting index must be + * aligned to num_contig_lg2. + */ +static inline void pt_clear_entry(struct pt_state *pts, + unsigned int num_contig_lg2); + +/** + * pt_entry_make_write_dirty() - Make an entry dirty + * @pts: Table index to change + * + * Make pt_entry_write_is_dirty() return true for this entry. This can be called + * asynchronously with any other table manipulation under a RCU lock and must + * not corrupt the table. + */ +static inline bool pt_entry_make_write_dirty(struct pt_state *pts); + +/** + * pt_dirty_supported() - True if the page table supports dirty tracking + * @common: Page table to query + */ +static inline bool pt_dirty_supported(struct pt_common *common); + +/** + * pt_entry_num_contig_lg2() - Number of contiguous items for this leaf entry + * @pts: Entry to query + * + * Return the number of contiguous items this leaf entry spans. If the entry + * is single item it returns ilog2(1). + */ +static inline unsigned int pt_entry_num_contig_lg2(const struct pt_state *pts); + +/** + * pt_entry_oa() - Output Address for this leaf entry + * @pts: Entry to query + * + * Return the output address for the start of the entry. If the entry + * is contiguous this returns the same value for each sub-item. I.e.:: + * + * log2_mod(pt_entry_oa(), pt_entry_oa_lg2sz()) == 0 + * + * See pt_item_oa(). The format should implement one of these two functions + * depending on how it stores the OAs in the table. + */ +static inline pt_oaddr_t pt_entry_oa(const struct pt_state *pts); + +/** + * pt_entry_oa_lg2sz() - Return the size of a OA entry + * @pts: Entry to query + * + * If the entry is not contiguous this returns pt_table_item_lg2sz(), otherwise + * it returns the total VA/OA size of the entire contiguous entry. + */ +static inline unsigned int pt_entry_oa_lg2sz(const struct pt_state *pts) +{ + return pt_entry_num_contig_lg2(pts) + pt_table_item_lg2sz(pts); +} + +/** + * pt_entry_oa_full() - Return the full OA for an entry + * @pts: Entry to query + * + * During iteration the first entry could have a VA with an offset from the + * natural start of the entry. Return the true full OA considering the pts's VA + * offset. + */ +static inline pt_oaddr_t pt_entry_oa_full(const struct pt_state *pts) +{ + return _pt_entry_oa_fast(pts) | + log2_mod(pts->range->va, pt_entry_oa_lg2sz(pts)); +} + +/** + * pt_entry_set_write_clean() - Make the entry write clean + * @pts: Table index to change + * + * Modify the entry so that pt_entry_write_is_dirty() == false. The HW will + * eventually be notified of this change via a TLB flush, which is the point + * that the HW must become synchronized. Any "write dirty" prior to the TLB + * flush can be lost, but once the TLB flush completes all writes must make + * their entries write dirty. + * + * The format should alter the entry in a way that is compatible with any + * concurrent update from HW. The entire contiguous entry is changed. + */ +static inline void pt_entry_set_write_clean(struct pt_state *pts); + +/** + * pt_entry_write_is_dirty() - True if the entry has been written to + * @pts: Entry to query + * + * "write dirty" means that the HW has written to the OA translated + * by this entry. If the entry is contiguous then the consolidated + * "write dirty" for all the items must be returned. + */ +static inline bool pt_entry_write_is_dirty(const struct pt_state *pts); + +/** + * pt_full_va_prefix() - The top bits of the VA + * @common: Page table to query + * + * This is usually 0, but some formats have their VA space going downward from + * PT_VADDR_MAX, and will return that instead. This value must always be + * adjusted by struct pt_common max_vasz_lg2. + */ +static inline pt_vaddr_t pt_full_va_prefix(const struct pt_common *common); + +/** + * pt_has_system_page() - True if level 0 can install a PAGE_SHIFT entry + * @common: Page table to query + * + * If true the caller can use, at level 0, pt_install_leaf_entry(PAGE_SHIFT). + * This is useful to create optimized paths for common cases of PAGE_SIZE + * mappings. + */ +static inline bool pt_has_system_page(const struct pt_common *common); + +/** + * pt_install_leaf_entry() - Write a leaf entry to the table + * @pts: Table index to change + * @oa: Output Address for this leaf + * @oasz_lg2: Size in VA for this leaf + * @attrs: Attributes to modify the entry + * + * A leaf OA entry will return PT_ENTRY_OA from pt_load_entry(). It translates + * the VA indicated by pts to the given OA. + * + * For a single item non-contiguous entry oasz_lg2 is pt_table_item_lg2sz(). + * For contiguous it is pt_table_item_lg2sz() + num_contig_lg2. + * + * This must not be called if pt_can_have_leaf() == false. Contiguous sizes + * not indicated by pt_possible_sizes() must not be specified. + */ +static inline void pt_install_leaf_entry(struct pt_state *pts, pt_oaddr_t oa, + unsigned int oasz_lg2, + const struct pt_write_attrs *attrs); + +/** + * pt_install_table() - Write a table entry to the table + * @pts: Table index to change + * @table_pa: CPU physical address of the lower table's memory + * @attrs: Attributes to modify the table index + * + * A table entry will return PT_ENTRY_TABLE from pt_load_entry(). The table_pa + * is the table at pts->level - 1. This is done by cmpxchg so pts must have the + * current entry loaded. The pts is updated with the installed entry. + * + * This must not be called if pt_can_have_table() == false. + * + * Returns: true if the table was installed successfully. + */ +static inline bool pt_install_table(struct pt_state *pts, pt_oaddr_t table_pa, + const struct pt_write_attrs *attrs); + +/** + * pt_item_oa() - Output Address for this leaf item + * @pts: Item to query + * + * Return the output address for this item. If the item is part of a contiguous + * entry it returns the value of the OA for this individual sub item. + * + * See pt_entry_oa(). The format should implement one of these two functions + * depending on how it stores the OA's in the table. + */ +static inline pt_oaddr_t pt_item_oa(const struct pt_state *pts); + +/** + * pt_load_entry_raw() - Read from the location pts points at into the pts + * @pts: Table index to load + * + * Return the type of entry that was loaded. pts->entry will be filled in with + * the entry's content. See pt_load_entry() + */ +static inline enum pt_entry_type pt_load_entry_raw(struct pt_state *pts); + +/** + * pt_max_output_address_lg2() - Return the maximum OA the table format can hold + * @common: Page table to query + * + * The value oalog2_to_max_int(pt_max_output_address_lg2()) is the MAX for the + * OA. This is the absolute maximum address the table can hold. struct pt_common + * max_oasz_lg2 sets a lower dynamic maximum based on HW capability. + */ +static inline unsigned int +pt_max_output_address_lg2(const struct pt_common *common); + +/** + * pt_num_items_lg2() - Return the number of items in this table level + * @pts: The current level + * + * The number of items in a table level defines the number of bits this level + * decodes from the VA. This function is not called for the top level, + * so it does not need to compute a special value for the top case. The + * result for the top is based on pt_common max_vasz_lg2. + * + * The value is used as part of determining the table indexes via the + * equation:: + * + * log2_mod(log2_div(VA, pt_table_item_lg2sz()), pt_num_items_lg2()) + */ +static inline unsigned int pt_num_items_lg2(const struct pt_state *pts); + +/** + * pt_pgsz_lg2_to_level - Return the level that maps the page size + * @common: Page table to query + * @pgsize_lg2: Log2 page size + * + * Returns the table level that will map the given page size. The page + * size must be part of the pt_possible_sizes() for some level. + */ +static inline unsigned int pt_pgsz_lg2_to_level(struct pt_common *common, + unsigned int pgsize_lg2); + +/** + * pt_possible_sizes() - Return a bitmap of possible output sizes at this level + * @pts: The current level + * + * Each level has a list of possible output sizes that can be installed as + * leaf entries. If pt_can_have_leaf() is false returns zero. + * + * Otherwise the bit in position pt_table_item_lg2sz() should be set indicating + * that a non-contiguous single item leaf entry is supported. The following + * pt_num_items_lg2() number of bits can be set indicating contiguous entries + * are supported. Bit pt_table_item_lg2sz() + pt_num_items_lg2() must not be + * set, contiguous entries cannot span the entire table. + * + * The OR of pt_possible_sizes() of all levels is the typical bitmask of all + * supported sizes in the entire table. + */ +static inline pt_vaddr_t pt_possible_sizes(const struct pt_state *pts); + +/** + * pt_table_item_lg2sz() - Size of a single item entry in this table level + * @pts: The current level + * + * The size of the item specifies how much VA and OA a single item occupies. + * + * See pt_entry_oa_lg2sz() for the same value including the effect of contiguous + * entries. + */ +static inline unsigned int pt_table_item_lg2sz(const struct pt_state *pts); + +/** + * pt_table_oa_lg2sz() - Return the VA/OA size of the entire table + * @pts: The current level + * + * Return the size of VA decoded by the entire table level. + */ +static inline unsigned int pt_table_oa_lg2sz(const struct pt_state *pts) +{ + if (pts->range->top_level == pts->level) + return pts->range->max_vasz_lg2; + return min_t(unsigned int, pts->range->common->max_vasz_lg2, + pt_num_items_lg2(pts) + pt_table_item_lg2sz(pts)); +} + +/** + * pt_table_pa() - Return the CPU physical address of the table entry + * @pts: Entry to query + * + * This is only ever called on PT_ENTRY_TABLE entries. Must return the same + * value passed to pt_install_table(). + */ +static inline pt_oaddr_t pt_table_pa(const struct pt_state *pts); + +/** + * pt_table_ptr() - Return a CPU pointer for a table item + * @pts: Entry to query + * + * Same as pt_table_pa() but returns a CPU pointer. + */ +static inline struct pt_table_p *pt_table_ptr(const struct pt_state *pts) +{ + return __va(pt_table_pa(pts)); +} + +/** + * pt_load_entry() - Read from the location pts points at into the pts + * @pts: Table index to load + * + * Set the type of entry that was loaded. pts->entry and pts->table_lower + * will be filled in with the entry's content. + */ +static inline void pt_load_entry(struct pt_state *pts) +{ + pts->type = pt_load_entry_raw(pts); + if (pts->type == PT_ENTRY_TABLE) + pts->table_lower = pt_table_ptr(pts); +} +#endif diff --git a/drivers/iommu/generic_pt/pt_defs.h b/drivers/iommu/generic_pt/pt_defs.h new file mode 100644 index 00000000000000..fa53b0903ab255 --- /dev/null +++ b/drivers/iommu/generic_pt/pt_defs.h @@ -0,0 +1,323 @@ +/* SPDX-License-Identifier: GPL-2.0-only */ +/* + * Copyright (c) 2024-2025, NVIDIA CORPORATION & AFFILIATES + * + * This header is included before the format. It contains definitions + * that are required to compile the format. The header order is: + * pt_defs.h + * fmt_XX.h + * pt_common.h + */ +#ifndef __GENERIC_PT_DEFS_H +#define __GENERIC_PT_DEFS_H + +#include <linux/generic_pt/common.h> + +#include <linux/types.h> +#include <linux/atomic.h> +#include <linux/bits.h> +#include <linux/limits.h> +#include <linux/bug.h> +#include <linux/kconfig.h> +#include "pt_log2.h" + +/* Header self-compile default defines */ +#ifndef pt_write_attrs +typedef u64 pt_vaddr_t; +typedef u64 pt_oaddr_t; +#endif + +struct pt_table_p; + +enum { + PT_VADDR_MAX = sizeof(pt_vaddr_t) == 8 ? U64_MAX : U32_MAX, + PT_VADDR_MAX_LG2 = sizeof(pt_vaddr_t) == 8 ? 64 : 32, + PT_OADDR_MAX = sizeof(pt_oaddr_t) == 8 ? U64_MAX : U32_MAX, + PT_OADDR_MAX_LG2 = sizeof(pt_oaddr_t) == 8 ? 64 : 32, +}; + +/* + * The format instantiation can have features wired off or on to optimize the + * code gen. Supported features are just a reflection of what the current set of + * kernel users want to use. + */ +#ifndef PT_SUPPORTED_FEATURES +#define PT_SUPPORTED_FEATURES 0 +#endif + +/* + * When in debug mode we compile all formats with all features. This allows the + * kunit to test the full matrix. SIGN_EXTEND can't co-exist with DYNAMIC_TOP or + * FULL_VA. + */ +#if IS_ENABLED(CONFIG_DEBUG_GENERIC_PT) +enum { + PT_ORIG_SUPPORTED_FEATURES = PT_SUPPORTED_FEATURES, + PT_DEBUG_SUPPORTED_FEATURES = + UINT_MAX & + ~((PT_ORIG_SUPPORTED_FEATURES & BIT(PT_FEAT_SIGN_EXTEND)) ? + BIT(PT_FEAT_DYNAMIC_TOP) | BIT(PT_FEAT_FULL_VA) : + BIT(PT_FEAT_SIGN_EXTEND)), +}; +#undef PT_SUPPORTED_FEATURES +#define PT_SUPPORTED_FEATURES PT_DEBUG_SUPPORTED_FEATURES +#endif + +#ifndef PT_FORCE_ENABLED_FEATURES +#define PT_FORCE_ENABLED_FEATURES 0 +#endif + +/** + * DOC: Generic Page Table Language + * + * Language used in Generic Page Table + * VA + * The input address to the page table, often the virtual address. + * OA + * The output address from the page table, often the physical address. + * leaf + * An entry that results in an output address. I.e. a physical memory addr + * start/end + * An half-open range, e.g. [0,0) refers to no VA. + * start/last + * An inclusive closed range, e.g. [0,0] refers to the VA 0 + * common + * The generic page table container struct pt_common + * level + * The number of table hops from the lowest leaf. Level 0 + * is always a table of only leaves of the least significant VA bits. The + * labels used by HW descriptions are never used. + * top_level + * The inclusive highest level of the table. A two-level table + * has a top level of 1. + * table + * A linear array of entries representing the translation items for that + * level. + * index + * The position in a table of an element: item = table[index] + * item + * A single position in a table + * entry + * A single logical element in a table. If contiguous pages are not + * supported then item and entry are the same thing, otherwise entry refers + * to the all the items that comprise a single contiguous translation. + * item/entry_size + * The number of bytes of VA the table translates for. + * If the item is a table entry then the next table covers + * this size. If the entry is an output address then the + * full OA is: OA | (VA % entry_size) + * contig_count + * The number of consecutive items fused into a single entry. + * item_size * contig_count is the size of that entry's translation. + * lg2 + * Indicates the value is encoded as log2, i.e. 1<<x is the actual value. + * Normally the compiler is fine to optimize divide and mod with log2 values + * automatically when inlining, however if the values are not constant + * expressions it can't. So we do it by hand; we want to avoid 64-bit + * divmod. + */ + +/* Returned by pt_load_entry() and for_each_pt_level_entry() */ +enum pt_entry_type { + PT_ENTRY_EMPTY, + PT_ENTRY_TABLE, + /* Entry is valid and returns an output address */ + PT_ENTRY_OA, +}; + +struct pt_range { + struct pt_common *common; + struct pt_table_p *top_table; + pt_vaddr_t va; + pt_vaddr_t last_va; + u8 top_level; + u8 max_vasz_lg2; +}; + +/* + * Similar to xa_state, this records information about an in-progress parse at a + * single level. + */ +struct pt_state { + struct pt_range *range; + struct pt_table_p *table; + struct pt_table_p *table_lower; + u64 entry; + enum pt_entry_type type; + unsigned short index; + unsigned short end_index; + u8 level; +}; + +#define pt_cur_table(pts, type) ((type *)((pts)->table)) + +/* + * Try to install a new table pointer. The locking methodology requires this to + * be atomic (multiple threads can race to install a pointer) the losing threads + * will fail the atomic and return false. They should free any memory and + * reparse the table level again. + */ +#if !IS_ENABLED(CONFIG_GENERIC_ATOMIC64) +static inline bool pt_table_install64(struct pt_state *pts, u64 table_entry) +{ + u64 *entryp = pt_cur_table(pts, u64) + pts->index; + u64 old_entry = pts->entry; + bool ret; + + /* + * Ensure the zero'd table content itself is visible before its PTE can + * be. release is a NOP on !SMP, but the HW is still doing an acquire. + */ + if (!IS_ENABLED(CONFIG_SMP)) + dma_wmb(); + ret = try_cmpxchg64_release(entryp, &old_entry, table_entry); + if (ret) + pts->entry = table_entry; + return ret; +} +#endif + +static inline bool pt_table_install32(struct pt_state *pts, u32 table_entry) +{ + u32 *entryp = pt_cur_table(pts, u32) + pts->index; + u32 old_entry = pts->entry; + bool ret; + + /* + * Ensure the zero'd table content itself is visible before its PTE can + * be. release is a NOP on !SMP, but the HW is still doing an acquire. + */ + if (!IS_ENABLED(CONFIG_SMP)) + dma_wmb(); + ret = try_cmpxchg_release(entryp, &old_entry, table_entry); + if (ret) + pts->entry = table_entry; + return ret; +} + +#define PT_SUPPORTED_FEATURE(feature_nr) (PT_SUPPORTED_FEATURES & BIT(feature_nr)) + +static inline bool pt_feature(const struct pt_common *common, + unsigned int feature_nr) +{ + if (PT_FORCE_ENABLED_FEATURES & BIT(feature_nr)) + return true; + if (!PT_SUPPORTED_FEATURE(feature_nr)) + return false; + return common->features & BIT(feature_nr); +} + +static inline bool pts_feature(const struct pt_state *pts, + unsigned int feature_nr) +{ + return pt_feature(pts->range->common, feature_nr); +} + +/* + * PT_WARN_ON is used for invariants that the kunit should be checking can't + * happen. + */ +#if IS_ENABLED(CONFIG_DEBUG_GENERIC_PT) +#define PT_WARN_ON WARN_ON +#else +static inline bool PT_WARN_ON(bool condition) +{ + return false; +} +#endif + +/* These all work on the VA type */ +#define log2_to_int(a_lg2) log2_to_int_t(pt_vaddr_t, a_lg2) +#define log2_to_max_int(a_lg2) log2_to_max_int_t(pt_vaddr_t, a_lg2) +#define log2_div(a, b_lg2) log2_div_t(pt_vaddr_t, a, b_lg2) +#define log2_div_eq(a, b, c_lg2) log2_div_eq_t(pt_vaddr_t, a, b, c_lg2) +#define log2_mod(a, b_lg2) log2_mod_t(pt_vaddr_t, a, b_lg2) +#define log2_mod_eq_max(a, b_lg2) log2_mod_eq_max_t(pt_vaddr_t, a, b_lg2) +#define log2_set_mod(a, val, b_lg2) log2_set_mod_t(pt_vaddr_t, a, val, b_lg2) +#define log2_set_mod_max(a, b_lg2) log2_set_mod_max_t(pt_vaddr_t, a, b_lg2) +#define log2_mul(a, b_lg2) log2_mul_t(pt_vaddr_t, a, b_lg2) +#define log2_ffs(a) log2_ffs_t(pt_vaddr_t, a) +#define log2_fls(a) log2_fls_t(pt_vaddr_t, a) +#define log2_ffz(a) log2_ffz_t(pt_vaddr_t, a) + +/* + * The full VA (fva) versions permit the lg2 value to be == PT_VADDR_MAX_LG2 and + * generate a useful defined result. The non-fva versions will malfunction at + * this extreme. + */ +static inline pt_vaddr_t fvalog2_div(pt_vaddr_t a, unsigned int b_lg2) +{ + if (PT_SUPPORTED_FEATURE(PT_FEAT_FULL_VA) && b_lg2 == PT_VADDR_MAX_LG2) + return 0; + return log2_div_t(pt_vaddr_t, a, b_lg2); +} + +static inline pt_vaddr_t fvalog2_mod(pt_vaddr_t a, unsigned int b_lg2) +{ + if (PT_SUPPORTED_FEATURE(PT_FEAT_FULL_VA) && b_lg2 == PT_VADDR_MAX_LG2) + return a; + return log2_mod_t(pt_vaddr_t, a, b_lg2); +} + +static inline bool fvalog2_div_eq(pt_vaddr_t a, pt_vaddr_t b, + unsigned int c_lg2) +{ + if (PT_SUPPORTED_FEATURE(PT_FEAT_FULL_VA) && c_lg2 == PT_VADDR_MAX_LG2) + return true; + return log2_div_eq_t(pt_vaddr_t, a, b, c_lg2); +} + +static inline pt_vaddr_t fvalog2_set_mod(pt_vaddr_t a, pt_vaddr_t val, + unsigned int b_lg2) +{ + if (PT_SUPPORTED_FEATURE(PT_FEAT_FULL_VA) && b_lg2 == PT_VADDR_MAX_LG2) + return val; + return log2_set_mod_t(pt_vaddr_t, a, val, b_lg2); +} + +static inline pt_vaddr_t fvalog2_set_mod_max(pt_vaddr_t a, unsigned int b_lg2) +{ + if (PT_SUPPORTED_FEATURE(PT_FEAT_FULL_VA) && b_lg2 == PT_VADDR_MAX_LG2) + return PT_VADDR_MAX; + return log2_set_mod_max_t(pt_vaddr_t, a, b_lg2); +} + +/* These all work on the OA type */ +#define oalog2_to_int(a_lg2) log2_to_int_t(pt_oaddr_t, a_lg2) +#define oalog2_to_max_int(a_lg2) log2_to_max_int_t(pt_oaddr_t, a_lg2) +#define oalog2_div(a, b_lg2) log2_div_t(pt_oaddr_t, a, b_lg2) +#define oalog2_div_eq(a, b, c_lg2) log2_div_eq_t(pt_oaddr_t, a, b, c_lg2) +#define oalog2_mod(a, b_lg2) log2_mod_t(pt_oaddr_t, a, b_lg2) +#define oalog2_mod_eq_max(a, b_lg2) log2_mod_eq_max_t(pt_oaddr_t, a, b_lg2) +#define oalog2_set_mod(a, val, b_lg2) log2_set_mod_t(pt_oaddr_t, a, val, b_lg2) +#define oalog2_set_mod_max(a, b_lg2) log2_set_mod_max_t(pt_oaddr_t, a, b_lg2) +#define oalog2_mul(a, b_lg2) log2_mul_t(pt_oaddr_t, a, b_lg2) +#define oalog2_ffs(a) log2_ffs_t(pt_oaddr_t, a) +#define oalog2_fls(a) log2_fls_t(pt_oaddr_t, a) +#define oalog2_ffz(a) log2_ffz_t(pt_oaddr_t, a) + +static inline uintptr_t _pt_top_set(struct pt_table_p *table_mem, + unsigned int top_level) +{ + return top_level | (uintptr_t)table_mem; +} + +static inline void pt_top_set(struct pt_common *common, + struct pt_table_p *table_mem, + unsigned int top_level) +{ + WRITE_ONCE(common->top_of_table, _pt_top_set(table_mem, top_level)); +} + +static inline void pt_top_set_level(struct pt_common *common, + unsigned int top_level) +{ + pt_top_set(common, NULL, top_level); +} + +static inline unsigned int pt_top_get_level(const struct pt_common *common) +{ + return READ_ONCE(common->top_of_table) % (1 << PT_TOP_LEVEL_BITS); +} + +#endif diff --git a/drivers/iommu/generic_pt/pt_fmt_defaults.h b/drivers/iommu/generic_pt/pt_fmt_defaults.h new file mode 100644 index 00000000000000..19e8f820c1dccf --- /dev/null +++ b/drivers/iommu/generic_pt/pt_fmt_defaults.h @@ -0,0 +1,193 @@ +/* SPDX-License-Identifier: GPL-2.0-only */ +/* + * Copyright (c) 2024-2025, NVIDIA CORPORATION & AFFILIATES + * + * Default definitions for formats that don't define these functions. + */ +#ifndef __GENERIC_PT_PT_FMT_DEFAULTS_H +#define __GENERIC_PT_PT_FMT_DEFAULTS_H + +#include "pt_defs.h" +#include <linux/log2.h> + +/* Header self-compile default defines */ +#ifndef pt_load_entry_raw +#include "fmt/amdv1.h" +#endif + +/* + * The format must provide PT_GRANULE_LG2SZ, PT_TABLEMEM_LG2SZ, and + * PT_ITEM_WORD_SIZE. They must be the same at every level excluding the top. + */ +#ifndef pt_table_item_lg2sz +static inline unsigned int pt_table_item_lg2sz(const struct pt_state *pts) +{ + return PT_GRANULE_LG2SZ + + (PT_TABLEMEM_LG2SZ - ilog2(PT_ITEM_WORD_SIZE)) * pts->level; +} +#endif + +#ifndef pt_pgsz_lg2_to_level +static inline unsigned int pt_pgsz_lg2_to_level(struct pt_common *common, + unsigned int pgsize_lg2) +{ + return (pgsize_lg2 - PT_GRANULE_LG2SZ) / + (PT_TABLEMEM_LG2SZ - ilog2(PT_ITEM_WORD_SIZE)); + return 0; +} +#endif + +/* If not supplied by the format then contiguous pages are not supported */ +#ifndef pt_entry_num_contig_lg2 +static inline unsigned int pt_entry_num_contig_lg2(const struct pt_state *pts) +{ + return ilog2(1); +} + +static inline unsigned short pt_contig_count_lg2(const struct pt_state *pts) +{ + return ilog2(1); +} +#endif + +/* If not supplied by the format then dirty tracking is not supported */ +#ifndef pt_entry_write_is_dirty +static inline bool pt_entry_write_is_dirty(const struct pt_state *pts) +{ + return false; +} + +static inline void pt_entry_set_write_clean(struct pt_state *pts) +{ +} + +static inline bool pt_dirty_supported(struct pt_common *common) +{ + return true; +} +#else +/* If not supplied then dirty tracking is always enabled */ +#ifndef pt_dirty_supported +static inline bool pt_dirty_supported(struct pt_common *common) +{ + return true; +} +#endif +#endif + +#ifndef pt_entry_make_write_dirty +static inline bool pt_entry_make_write_dirty(struct pt_state *pts) +{ + return false; +} +#endif + +/* + * Format supplies either: + * pt_entry_oa - OA is at the start of a contiguous entry + * or + * pt_item_oa - OA is correct for every item in a contiguous entry + * + * Build the missing one + */ +#ifdef pt_entry_oa +static inline pt_oaddr_t pt_item_oa(const struct pt_state *pts) +{ + return pt_entry_oa(pts) | + log2_mul(pts->index, pt_table_item_lg2sz(pts)); +} +#define _pt_entry_oa_fast pt_entry_oa +#endif + +#ifdef pt_item_oa +static inline pt_oaddr_t pt_entry_oa(const struct pt_state *pts) +{ + return log2_set_mod(pt_item_oa(pts), 0, + pt_entry_num_contig_lg2(pts) + + pt_table_item_lg2sz(pts)); +} +#define _pt_entry_oa_fast pt_item_oa +#endif + +/* + * If not supplied by the format then use the constant + * PT_MAX_OUTPUT_ADDRESS_LG2. + */ +#ifndef pt_max_output_address_lg2 +static inline unsigned int +pt_max_output_address_lg2(const struct pt_common *common) +{ + return PT_MAX_OUTPUT_ADDRESS_LG2; +} +#endif + +#ifndef pt_has_system_page +static inline bool pt_has_system_page(const struct pt_common *common) +{ + return PT_GRANULE_LG2SZ == PAGE_SHIFT; +} +#endif + +/* + * If not supplied by the format then assume only one contiguous size determined + * by pt_contig_count_lg2() + */ +#ifndef pt_possible_sizes +static inline unsigned short pt_contig_count_lg2(const struct pt_state *pts); + +/* Return a bitmap of possible leaf page sizes at this level */ +static inline pt_vaddr_t pt_possible_sizes(const struct pt_state *pts) +{ + unsigned int isz_lg2 = pt_table_item_lg2sz(pts); + + if (!pt_can_have_leaf(pts)) + return 0; + return log2_to_int(isz_lg2) | + log2_to_int(pt_contig_count_lg2(pts) + isz_lg2); +} +#endif + +/* If not supplied by the format then use 0. */ +#ifndef pt_full_va_prefix +static inline pt_vaddr_t pt_full_va_prefix(const struct pt_common *common) +{ + return 0; +} +#endif + +/* If not supplied by the format then zero fill using PT_ITEM_WORD_SIZE */ +#ifndef pt_clear_entry +static inline void pt_clear_entry64(struct pt_state *pts, + unsigned int num_contig_lg2) +{ + u64 *tablep = pt_cur_table(pts, u64) + pts->index; + u64 *end = tablep + log2_to_int(num_contig_lg2); + + PT_WARN_ON(log2_mod(pts->index, num_contig_lg2)); + for (; tablep != end; tablep++) + WRITE_ONCE(*tablep, 0); +} + +static inline void pt_clear_entry32(struct pt_state *pts, + unsigned int num_contig_lg2) +{ + u32 *tablep = pt_cur_table(pts, u32) + pts->index; + u32 *end = tablep + log2_to_int(num_contig_lg2); + + PT_WARN_ON(log2_mod(pts->index, num_contig_lg2)); + for (; tablep != end; tablep++) + WRITE_ONCE(*tablep, 0); +} + +static inline void pt_clear_entry(struct pt_state *pts, + unsigned int num_contig_lg2) +{ + if (PT_ITEM_WORD_SIZE == sizeof(u32)) + pt_clear_entry32(pts, num_contig_lg2); + else + pt_clear_entry64(pts, num_contig_lg2); +} +#define pt_clear_entry pt_clear_entry +#endif + +#endif diff --git a/drivers/iommu/generic_pt/pt_iter.h b/drivers/iommu/generic_pt/pt_iter.h new file mode 100644 index 00000000000000..6bd84f96f5ad6e --- /dev/null +++ b/drivers/iommu/generic_pt/pt_iter.h @@ -0,0 +1,636 @@ +/* SPDX-License-Identifier: GPL-2.0-only */ +/* + * Copyright (c) 2024-2025, NVIDIA CORPORATION & AFFILIATES + * + * Iterators for Generic Page Table + */ +#ifndef __GENERIC_PT_PT_ITER_H +#define __GENERIC_PT_PT_ITER_H + +#include "pt_common.h" + +#include <linux/errno.h> + +/* + * Use to mangle symbols so that backtraces and the symbol table are + * understandable. Any non-inlined function should get mangled like this. + */ +#define NS(fn) CONCATENATE(PTPFX, fn) + +/** + * pt_check_range() - Validate the range can be iterated + * @range: Range to validate + * + * Check that va and last_va fall within the permitted range of VAs. If the + * format is using PT_FEAT_SIGN_EXTEND then this also checks the sign extension + * is correct. + */ +static inline int pt_check_range(struct pt_range *range) +{ + pt_vaddr_t prefix; + + PT_WARN_ON(!range->max_vasz_lg2); + + if (pt_feature(range->common, PT_FEAT_SIGN_EXTEND)) { + PT_WARN_ON(range->common->max_vasz_lg2 != range->max_vasz_lg2); + prefix = fvalog2_div(range->va, range->max_vasz_lg2 - 1) ? + PT_VADDR_MAX : + 0; + } else { + prefix = pt_full_va_prefix(range->common); + } + + if (!fvalog2_div_eq(range->va, prefix, range->max_vasz_lg2) || + !fvalog2_div_eq(range->last_va, prefix, range->max_vasz_lg2)) + return -ERANGE; + return 0; +} + +/** + * pt_index_to_va() - Update range->va to the current pts->index + * @pts: Iteration State + * + * Adjust range->va to match the current index. This is done in a lazy manner + * since computing the VA takes several instructions and is rarely required. + */ +static inline void pt_index_to_va(struct pt_state *pts) +{ + pt_vaddr_t lower_va; + + lower_va = log2_mul(pts->index, pt_table_item_lg2sz(pts)); + pts->range->va = fvalog2_set_mod(pts->range->va, lower_va, + pt_table_oa_lg2sz(pts)); +} + +/* + * Add index_count_lg2 number of entries to pts's VA and index. The VA will be + * adjusted to the end of the contiguous block if it is currently in the middle. + */ +static inline void _pt_advance(struct pt_state *pts, + unsigned int index_count_lg2) +{ + pts->index = log2_set_mod(pts->index + log2_to_int(index_count_lg2), 0, + index_count_lg2); +} + +/** + * pt_item_fully_covered() - Check if the item or entry is entirely contained + * within pts->range + * @pts: Iteration State + * @oasz_lg2: The size of the item to check, pt_table_item_lg2sz() or + * pt_entry_oa_lg2sz() + * + * Returns: true if the item is fully enclosed by the pts->range. + */ +static inline bool pt_item_fully_covered(const struct pt_state *pts, + unsigned int oasz_lg2) +{ + struct pt_range *range = pts->range; + + /* Range begins at the start of the entry */ + if (log2_mod(pts->range->va, oasz_lg2)) + return false; + + /* Range ends past the end of the entry */ + if (!log2_div_eq(range->va, range->last_va, oasz_lg2)) + return true; + + /* Range ends at the end of the entry */ + return log2_mod_eq_max(range->last_va, oasz_lg2); +} + +/** + * pt_range_to_index() - Starting index for an iteration + * @pts: Iteration State + * + * Return: the starting index for the iteration in pts. + */ +static inline unsigned int pt_range_to_index(const struct pt_state *pts) +{ + unsigned int isz_lg2 = pt_table_item_lg2sz(pts); + + PT_WARN_ON(pts->level > pts->range->top_level); + if (pts->range->top_level == pts->level) + return log2_div(fvalog2_mod(pts->range->va, + pts->range->max_vasz_lg2), + isz_lg2); + return log2_mod(log2_div(pts->range->va, isz_lg2), + pt_num_items_lg2(pts)); +} + +/** + * pt_range_to_end_index() - Ending index iteration + * @pts: Iteration State + * + * Return: the last index for the iteration in pts. + */ +static inline unsigned int pt_range_to_end_index(const struct pt_state *pts) +{ + unsigned int isz_lg2 = pt_table_item_lg2sz(pts); + struct pt_range *range = pts->range; + unsigned int num_entries_lg2; + + if (range->va == range->last_va) + return pts->index + 1; + + if (pts->range->top_level == pts->level) + return log2_div(fvalog2_mod(pts->range->last_va, + pts->range->max_vasz_lg2), + isz_lg2) + + 1; + + num_entries_lg2 = pt_num_items_lg2(pts); + + /* last_va falls within this table */ + if (log2_div_eq(range->va, range->last_va, num_entries_lg2 + isz_lg2)) + return log2_mod(log2_div(pts->range->last_va, isz_lg2), + num_entries_lg2) + + 1; + + return log2_to_int(num_entries_lg2); +} + +static inline void _pt_iter_first(struct pt_state *pts) +{ + pts->index = pt_range_to_index(pts); + pts->end_index = pt_range_to_end_index(pts); + PT_WARN_ON(pts->index > pts->end_index); +} + +static inline bool _pt_iter_load(struct pt_state *pts) +{ + if (pts->index >= pts->end_index) + return false; + pt_load_entry(pts); + return true; +} + +/** + * pt_next_entry() - Advance pts to the next entry + * @pts: Iteration State + * + * Update pts to go to the next index at this level. If pts is pointing at a + * contiguous entry then the index may advance my more than one. + */ +static inline void pt_next_entry(struct pt_state *pts) +{ + if (pts->type == PT_ENTRY_OA && + !__builtin_constant_p(pt_entry_num_contig_lg2(pts) == 0)) + _pt_advance(pts, pt_entry_num_contig_lg2(pts)); + else + pts->index++; + pt_index_to_va(pts); +} + +/** + * for_each_pt_level_entry() - For loop wrapper over entries in the range + * @pts: Iteration State + * + * This is the basic iteration primitive. It iterates over all the entries in + * pts->range that fall within the pts's current table level. Each step does + * pt_load_entry(pts). + */ +#define for_each_pt_level_entry(pts) \ + for (_pt_iter_first(pts); _pt_iter_load(pts); pt_next_entry(pts)) + +/** + * pt_load_single_entry() - Version of pt_load_entry() usable within a walker + * @pts: Iteration State + * + * Alternative to for_each_pt_level_entry() if the walker function uses only a + * single entry. + */ +static inline enum pt_entry_type pt_load_single_entry(struct pt_state *pts) +{ + pts->index = pt_range_to_index(pts); + pt_load_entry(pts); + return pts->type; +} + +static __always_inline struct pt_range _pt_top_range(struct pt_common *common, + uintptr_t top_of_table) +{ + struct pt_range range = { + .common = common, + .top_table = + (struct pt_table_p *)(top_of_table & + ~(uintptr_t)PT_TOP_LEVEL_MASK), + .top_level = top_of_table % (1 << PT_TOP_LEVEL_BITS), + }; + struct pt_state pts = { .range = &range, .level = range.top_level }; + unsigned int max_vasz_lg2; + + max_vasz_lg2 = common->max_vasz_lg2; + if (pt_feature(common, PT_FEAT_DYNAMIC_TOP) && + pts.level != PT_MAX_TOP_LEVEL) + max_vasz_lg2 = min_t(unsigned int, common->max_vasz_lg2, + pt_num_items_lg2(&pts) + + pt_table_item_lg2sz(&pts)); + + /* + * The top range will default to the lower region only with sign extend. + */ + range.max_vasz_lg2 = max_vasz_lg2; + if (pt_feature(common, PT_FEAT_SIGN_EXTEND)) + max_vasz_lg2--; + + range.va = fvalog2_set_mod(pt_full_va_prefix(common), 0, max_vasz_lg2); + range.last_va = + fvalog2_set_mod_max(pt_full_va_prefix(common), max_vasz_lg2); + return range; +} + +/** + * pt_top_range() - Return a range that spans part of the top level + * @common: Table + * + * For PT_FEAT_SIGN_EXTEND this will return the lower range, and cover half the + * total page table. Otherwise it returns the entire page table. + */ +static __always_inline struct pt_range pt_top_range(struct pt_common *common) +{ + /* + * The top pointer can change without locking. We capture the value and + * it's level here and are safe to walk it so long as both values are + * captured without tearing. + */ + return _pt_top_range(common, READ_ONCE(common->top_of_table)); +} + +/** + * pt_all_range() - Return a range that spans the entire page table + * @common: Table + * + * The returned range spans the whole page table. Due to how PT_FEAT_SIGN_EXTEND + * is supported range->va and range->last_va will be incorrect during the + * iteration and must not be accessed. + */ +static inline struct pt_range pt_all_range(struct pt_common *common) +{ + struct pt_range range = pt_top_range(common); + + if (!pt_feature(common, PT_FEAT_SIGN_EXTEND)) + return range; + + /* + * Pretend the table is linear from 0 without a sign extension. This + * generates the correct indexes for iteration. + */ + range.last_va = fvalog2_set_mod_max(0, range.max_vasz_lg2); + return range; +} + +/** + * pt_upper_range() - Return a range that spans part of the top level + * @common: Table + * + * For PT_FEAT_SIGN_EXTEND this will return the upper range, and cover half the + * total page table. Otherwise it returns the entire page table. + */ +static inline struct pt_range pt_upper_range(struct pt_common *common) +{ + struct pt_range range = pt_top_range(common); + + if (!pt_feature(common, PT_FEAT_SIGN_EXTEND)) + return range; + + range.va = fvalog2_set_mod(PT_VADDR_MAX, 0, range.max_vasz_lg2 - 1); + range.last_va = PT_VADDR_MAX; + return range; +} + +/** + * pt_make_range() - Return a range that spans part of the table + * @common: Table + * @va: Start address + * @last_va: Last address + * + * The caller must validate the range with pt_check_range() before using it. + */ +static __always_inline struct pt_range +pt_make_range(struct pt_common *common, pt_vaddr_t va, pt_vaddr_t last_va) +{ + struct pt_range range = + _pt_top_range(common, READ_ONCE(common->top_of_table)); + + range.va = va; + range.last_va = last_va; + + return range; +} + +/* + * Span a slice of the table starting at a lower table level from an active + * walk. + */ +static __always_inline struct pt_range +pt_make_child_range(const struct pt_range *parent, pt_vaddr_t va, + pt_vaddr_t last_va) +{ + struct pt_range range = *parent; + + range.va = va; + range.last_va = last_va; + + PT_WARN_ON(last_va < va); + PT_WARN_ON(pt_check_range(&range)); + + return range; +} + +/** + * pt_init() - Initialize a pt_state on the stack + * @range: Range pointer to embed in the state + * @level: Table level for the state + * @table: Pointer to the table memory at level + * + * Helper to initialize the on-stack pt_state from walker arguments. + */ +static __always_inline struct pt_state +pt_init(struct pt_range *range, unsigned int level, struct pt_table_p *table) +{ + struct pt_state pts = { + .range = range, + .table = table, + .level = level, + }; + return pts; +} + +/** + * pt_init_top() - Initialize a pt_state on the stack + * @range: Range pointer to embed in the state + * + * The pt_state points to the top most level. + */ +static __always_inline struct pt_state pt_init_top(struct pt_range *range) +{ + return pt_init(range, range->top_level, range->top_table); +} + +typedef int (*pt_level_fn_t)(struct pt_range *range, void *arg, + unsigned int level, struct pt_table_p *table); + +/** + * pt_descend() - Recursively invoke the walker for the lower level + * @pts: Iteration State + * @arg: Value to pass to the function + * @fn: Walker function to call + * + * pts must point to a table item. Invoke fn as a walker on the table + * pts points to. + */ +static __always_inline int pt_descend(struct pt_state *pts, void *arg, + pt_level_fn_t fn) +{ + int ret; + + if (PT_WARN_ON(!pts->table_lower)) + return -EINVAL; + + ret = (*fn)(pts->range, arg, pts->level - 1, pts->table_lower); + return ret; +} + +/** + * pt_walk_range() - Walk over a VA range + * @range: Range pointer + * @fn: Walker function to call + * @arg: Value to pass to the function + * + * Walk over a VA range. The caller should have done a validity check, at + * least calling pt_check_range(), when building range. The walk will + * start at the top most table. + */ +static __always_inline int pt_walk_range(struct pt_range *range, + pt_level_fn_t fn, void *arg) +{ + return fn(range, arg, range->top_level, range->top_table); +} + +/* + * pt_walk_descend() - Recursively invoke the walker for a slice of a lower + * level + * @pts: Iteration State + * @va: Start address + * @last_va: Last address + * @fn: Walker function to call + * @arg: Value to pass to the function + * + * With pts pointing at a table item this will descend and over a slice of the + * lower table. The caller must ensure that va/last_va are within the table + * item. This creates a new walk and does not alter pts or pts->range. + */ +static __always_inline int pt_walk_descend(const struct pt_state *pts, + pt_vaddr_t va, pt_vaddr_t last_va, + pt_level_fn_t fn, void *arg) +{ + struct pt_range range = pt_make_child_range(pts->range, va, last_va); + + if (PT_WARN_ON(!pt_can_have_table(pts)) || + PT_WARN_ON(!pts->table_lower)) + return -EINVAL; + + return fn(&range, arg, pts->level - 1, pts->table_lower); +} + +/* + * pt_walk_descend_all() - Recursively invoke the walker for a table item + * @parent_pts: Iteration State + * @fn: Walker function to call + * @arg: Value to pass to the function + * + * With pts pointing at a table item this will descend and over the entire lower + * table. This creates a new walk and does not alter pts or pts->range. + */ +static __always_inline int +pt_walk_descend_all(const struct pt_state *parent_pts, pt_level_fn_t fn, + void *arg) +{ + unsigned int isz_lg2 = pt_table_item_lg2sz(parent_pts); + + return pt_walk_descend(parent_pts, + log2_set_mod(parent_pts->range->va, 0, isz_lg2), + log2_set_mod_max(parent_pts->range->va, isz_lg2), + fn, arg); +} + +/** + * pt_range_slice() - Return a range that spans indexes + * @pts: Iteration State + * @start_index: Starting index within pts + * @end_index: Ending index within pts + * + * Create a range than spans an index range of the current table level + * pt_state points at. + */ +static inline struct pt_range pt_range_slice(const struct pt_state *pts, + unsigned int start_index, + unsigned int end_index) +{ + unsigned int table_lg2sz = pt_table_oa_lg2sz(pts); + pt_vaddr_t last_va; + pt_vaddr_t va; + + va = fvalog2_set_mod(pts->range->va, + log2_mul(start_index, pt_table_item_lg2sz(pts)), + table_lg2sz); + last_va = fvalog2_set_mod( + pts->range->va, + log2_mul(end_index, pt_table_item_lg2sz(pts)) - 1, table_lg2sz); + return pt_make_child_range(pts->range, va, last_va); +} + +/** + * pt_top_memsize_lg2() + * @common: Table + * @top_of_table: Top of table value from _pt_top_set() + * + * Compute the allocation size of the top table. For PT_FEAT_DYNAMIC_TOP this + * will compute the top size assuming the table will grow. + */ +static inline unsigned int pt_top_memsize_lg2(struct pt_common *common, + uintptr_t top_of_table) +{ + struct pt_range range = _pt_top_range(common, top_of_table); + struct pt_state pts = pt_init_top(&range); + unsigned int num_items_lg2; + + num_items_lg2 = common->max_vasz_lg2 - pt_table_item_lg2sz(&pts); + if (range.top_level != PT_MAX_TOP_LEVEL && + pt_feature(common, PT_FEAT_DYNAMIC_TOP)) + num_items_lg2 = min(num_items_lg2, pt_num_items_lg2(&pts)); + + /* Round up the allocation size to the minimum alignment */ + return max(log2_ffs_t(u64, PT_TOP_PHYS_MASK), + num_items_lg2 + ilog2(PT_ITEM_WORD_SIZE)); +} + +/** + * pt_compute_best_pgsize() - Determine the best page size for leaf entries + * @pgsz_bitmap: Permitted page sizes + * @va: Starting virtual address for the leaf entry + * @last_va: Last virtual address for the leaf entry, sets the max page size + * @oa: Starting output address for the leaf entry + * + * Compute the largest page size for va, last_va, and oa together and return it + * in lg2. The largest page size depends on the format's supported page sizes at + * this level, and the relative alignment of the VA and OA addresses. 0 means + * the OA cannot be stored with the provided pgsz_bitmap. + */ +static inline unsigned int pt_compute_best_pgsize(pt_vaddr_t pgsz_bitmap, + pt_vaddr_t va, + pt_vaddr_t last_va, + pt_oaddr_t oa) +{ + unsigned int best_pgsz_lg2; + unsigned int pgsz_lg2; + pt_vaddr_t len = last_va - va + 1; + pt_vaddr_t mask; + + if (PT_WARN_ON(va >= last_va)) + return 0; + + /* + * Given a VA/OA pair the best page size is the largest page side + * where: + * + * 1) VA and OA start at the page. Bitwise this is the count of least + * significant 0 bits. + * This also implies that last_va/oa has the same prefix as va/oa. + */ + mask = va | oa; + + /* + * 2) The page size is not larger than the last_va (length). Since page + * sizes are always power of two this can't be larger than the + * largest power of two factor of the length. + */ + mask |= log2_to_int(log2_fls(len) - 1); + + best_pgsz_lg2 = log2_ffs(mask); + + /* Choose the higest bit <= best_pgsz_lg2 */ + if (best_pgsz_lg2 < PT_VADDR_MAX_LG2 - 1) + pgsz_bitmap = log2_mod(pgsz_bitmap, best_pgsz_lg2 + 1); + + pgsz_lg2 = log2_fls(pgsz_bitmap); + if (!pgsz_lg2) + return 0; + + pgsz_lg2--; + + PT_WARN_ON(log2_mod(va, pgsz_lg2) != 0); + PT_WARN_ON(oalog2_mod(oa, pgsz_lg2) != 0); + PT_WARN_ON(va + log2_to_int(pgsz_lg2) - 1 > last_va); + PT_WARN_ON(!log2_div_eq(va, va + log2_to_int(pgsz_lg2) - 1, pgsz_lg2)); + PT_WARN_ON( + !oalog2_div_eq(oa, oa + log2_to_int(pgsz_lg2) - 1, pgsz_lg2)); + return pgsz_lg2; +} + +#define _PT_MAKE_CALL_LEVEL(fn) \ + static __always_inline int fn(struct pt_range *range, void *arg, \ + unsigned int level, \ + struct pt_table_p *table) \ + { \ + static_assert(PT_MAX_TOP_LEVEL <= 5); \ + if (level == 0) \ + return CONCATENATE(fn, 0)(range, arg, 0, table); \ + if (level == 1 || PT_MAX_TOP_LEVEL == 1) \ + return CONCATENATE(fn, 1)(range, arg, 1, table); \ + if (level == 2 || PT_MAX_TOP_LEVEL == 2) \ + return CONCATENATE(fn, 2)(range, arg, 2, table); \ + if (level == 3 || PT_MAX_TOP_LEVEL == 3) \ + return CONCATENATE(fn, 3)(range, arg, 3, table); \ + if (level == 4 || PT_MAX_TOP_LEVEL == 4) \ + return CONCATENATE(fn, 4)(range, arg, 4, table); \ + return CONCATENATE(fn, 5)(range, arg, 5, table); \ + } + +static inline int __pt_make_level_fn_err(struct pt_range *range, void *arg, + unsigned int unused_level, + struct pt_table_p *table) +{ + static_assert(PT_MAX_TOP_LEVEL <= 5); + return -EPROTOTYPE; +} + +#define __PT_MAKE_LEVEL_FN(fn, level, descend_fn, do_fn) \ + static inline int fn(struct pt_range *range, void *arg, \ + unsigned int unused_level, \ + struct pt_table_p *table) \ + { \ + return do_fn(range, arg, level, table, descend_fn); \ + } + +/** + * PT_MAKE_LEVELS() - Build an unwound walker + * @fn: Name of the walker function + * @do_fn: Function to call at each level + * + * This builds a function call tree that can be fully inlined. + * The caller must provide a function body in an __always_inline function:: + * + * static __always_inline int do(struct pt_range *range, void *arg, + * unsigned int level, struct pt_table_p *table, + * pt_level_fn_t descend_fn) + * + * An inline function will be created for each table level that calls do_fn with + * a compile time constant for level and a pointer to the next lower function. + * This generates an optimally inlined walk where each of the functions sees a + * constant level and can codegen the exact constants/etc for that level. + * + * Note this can produce a lot of code! + */ +#define PT_MAKE_LEVELS(fn, do_fn) \ + __PT_MAKE_LEVEL_FN(CONCATENATE(fn, 0), 0, __pt_make_level_fn_err, \ + do_fn); \ + __PT_MAKE_LEVEL_FN(CONCATENATE(fn, 1), 1, CONCATENATE(fn, 0), do_fn); \ + __PT_MAKE_LEVEL_FN(CONCATENATE(fn, 2), 2, CONCATENATE(fn, 1), do_fn); \ + __PT_MAKE_LEVEL_FN(CONCATENATE(fn, 3), 3, CONCATENATE(fn, 2), do_fn); \ + __PT_MAKE_LEVEL_FN(CONCATENATE(fn, 4), 4, CONCATENATE(fn, 3), do_fn); \ + __PT_MAKE_LEVEL_FN(CONCATENATE(fn, 5), 5, CONCATENATE(fn, 4), do_fn); \ + _PT_MAKE_CALL_LEVEL(fn) + +#endif diff --git a/drivers/iommu/generic_pt/pt_log2.h b/drivers/iommu/generic_pt/pt_log2.h new file mode 100644 index 00000000000000..70d40fbfd7a7b6 --- /dev/null +++ b/drivers/iommu/generic_pt/pt_log2.h @@ -0,0 +1,130 @@ +/* SPDX-License-Identifier: GPL-2.0-only */ +/* + * Copyright (c) 2024-2025, NVIDIA CORPORATION & AFFILIATES + * + * Helper macros for working with log2 values + * + */ +#ifndef __GENERIC_PT_LOG2_H +#define __GENERIC_PT_LOG2_H +#include <linux/bitops.h> +#include <linux/limits.h> + +/* Compute a */ +#define log2_to_int_t(type, a_lg2) ((type)(((type)1) << (a_lg2))) +static_assert(log2_to_int_t(unsigned int, 0) == 1); + +/* Compute a - 1 (aka all low bits set) */ +#define log2_to_max_int_t(type, a_lg2) ((type)(log2_to_int_t(type, a_lg2) - 1)) + +/* Compute a / b */ +#define log2_div_t(type, a, b_lg2) ((type)(((type)a) >> (b_lg2))) +static_assert(log2_div_t(unsigned int, 4, 2) == 1); + +/* + * Compute: + * a / c == b / c + * aka the high bits are equal + */ +#define log2_div_eq_t(type, a, b, c_lg2) \ + (log2_div_t(type, (a) ^ (b), c_lg2) == 0) +static_assert(log2_div_eq_t(unsigned int, 1, 1, 2)); + +/* Compute a % b */ +#define log2_mod_t(type, a, b_lg2) \ + ((type)(((type)a) & log2_to_max_int_t(type, b_lg2))) +static_assert(log2_mod_t(unsigned int, 1, 2) == 1); + +/* + * Compute: + * a % b == b - 1 + * aka the low bits are all 1s + */ +#define log2_mod_eq_max_t(type, a, b_lg2) \ + (log2_mod_t(type, a, b_lg2) == log2_to_max_int_t(type, b_lg2)) +static_assert(log2_mod_eq_max_t(unsigned int, 3, 2)); + +/* + * Return a value such that: + * a / b == ret / b + * ret % b == val + * aka set the low bits to val. val must be < b + */ +#define log2_set_mod_t(type, a, val, b_lg2) \ + ((((type)(a)) & (~log2_to_max_int_t(type, b_lg2))) | ((type)(val))) +static_assert(log2_set_mod_t(unsigned int, 3, 1, 2) == 1); + +/* Return a value such that: + * a / b == ret / b + * ret % b == b - 1 + * aka set the low bits to all 1s + */ +#define log2_set_mod_max_t(type, a, b_lg2) \ + (((type)(a)) | log2_to_max_int_t(type, b_lg2)) +static_assert(log2_set_mod_max_t(unsigned int, 2, 2) == 3); + +/* Compute a * b */ +#define log2_mul_t(type, a, b_lg2) ((type)(((type)a) << (b_lg2))) +static_assert(log2_mul_t(unsigned int, 2, 2) == 8); + +#define _dispatch_sz(type, fn, a) \ + (sizeof(type) == 4 ? fn##32((u32)a) : fn##64(a)) + +/* + * Return the highest value such that: + * log2_fls(0) == 0 + * log2_fls(1) == 1 + * a >= log2_to_int(ret - 1) + * aka find last set bit + */ +static inline unsigned int log2_fls32(u32 a) +{ + return fls(a); +} +static inline unsigned int log2_fls64(u64 a) +{ + return fls64(a); +} +#define log2_fls_t(type, a) _dispatch_sz(type, log2_fls, a) + +/* + * Return the highest value such that: + * log2_ffs(0) == UNDEFINED + * log2_ffs(1) == 0 + * log_mod(a, ret) == 0 + * aka find first set bit + */ +static inline unsigned int log2_ffs32(u32 a) +{ + return __ffs(a); +} +static inline unsigned int log2_ffs64(u64 a) +{ + return __ffs64(a); +} +#define log2_ffs_t(type, a) _dispatch_sz(type, log2_ffs, a) + +/* + * Return the highest value such that: + * log2_ffz(MAX) == UNDEFINED + * log2_ffz(0) == 0 + * log2_ffz(1) == 1 + * log_mod(a, ret) == log_to_max_int(ret) + * aka find first zero bit + */ +static inline unsigned int log2_ffz32(u32 a) +{ + return ffz(a); +} +static inline unsigned int log2_ffz64(u64 a) +{ + if (sizeof(u64) == sizeof(unsigned long)) + return ffz(a); + + if ((u32)a == U32_MAX) + return log2_ffz32(a >> 32) + 32; + return log2_ffz32(a); +} +#define log2_ffz_t(type, a) _dispatch_sz(type, log2_ffz, a) + +#endif diff --git a/include/linux/generic_pt/common.h b/include/linux/generic_pt/common.h new file mode 100644 index 00000000000000..a29bdd7b244de6 --- /dev/null +++ b/include/linux/generic_pt/common.h @@ -0,0 +1,134 @@ +/* SPDX-License-Identifier: GPL-2.0-only */ +/* + * Copyright (c) 2024-2025, NVIDIA CORPORATION & AFFILIATES + */ +#ifndef __GENERIC_PT_COMMON_H +#define __GENERIC_PT_COMMON_H + +#include <linux/types.h> +#include <linux/build_bug.h> +#include <linux/bits.h> + +/** + * DOC: Generic Radix Page Table + * + * Generic Radix Page Table is a set of functions and helpers to efficiently + * parse radix style page tables typically seen in HW implementations. The + * interface is built to deliver similar code generation as the mm's pte/pmd/etc + * system by fully inlining the exact code required to handle each table level. + * + * Like the mm subsystem each format contributes its parsing implementation + * under common names and the common code implements the required algorithms. + * + * The system is divided into three logical levels: + * + * - The page table format and its manipulation functions + * - Generic helpers to give a consistent API regardless of underlying format + * - An algorithm implementation (e.g. IOMMU/DRM/KVM/MM) + * + * Multiple implementations are supported. The intention is to have the generic + * format code be re-usable for whatever specalized implementation is required. + * The generic code is solely about the format of the radix tree; it does not + * include memory allocation or higher level decisions that are left for the + * implementation. + * + * The generic framework supports a superset of functions across many HW + * implementations: + * + * - Entries comprised of contiguous blocks of IO PTEs for larger page sizes + * - Multi-level tables, up to 6 levels. Runtime selected top level + * - Runtime variable table level size (ARM's concatenated tables) + * - Expandable top level allowing dynamic sizing of table levels + * - Optional leaf entries at any level + * - 32-bit/64-bit virtual and output addresses, using every address bit + * - Dirty tracking + */ + +/** + * struct pt_common - struct for all page table implementations + */ +struct pt_common { + /** + * @top_of_table: Encodes the table top pointer and the top level in a + * single value. Must use READ_ONCE/WRITE_ONCE to access it. The lower + * bits of the aligned table pointer are used for the level. + */ + uintptr_t top_of_table; + /** + * @max_oasz_lg2: Maximum number of bits the OA can contain. Upper bits + * must be zero. This may be less than what the page table format + * supports, but must not be more. + */ + u8 max_oasz_lg2; + /** + * @max_vasz_lg2: Maximum number of bits the VA can contain. Upper bits + * are 0 or 1 depending on pt_full_va_prefix(). This may be less than + * what the page table format supports, but must not be more. When + * PT_FEAT_DYNAMIC_TOP is set this reflects the maximum VA capability. + */ + u8 max_vasz_lg2; + /** + * @features: Bitmap of `enum pt_features` + */ + unsigned int features; +}; + +/* Encoding parameters for top_of_table */ +enum { + PT_TOP_LEVEL_BITS = 3, + PT_TOP_LEVEL_MASK = GENMASK(PT_TOP_LEVEL_BITS - 1, 0), +}; + +/** + * enum pt_features - Features turned on in the table. Each symbol is a bit + * position. + */ +enum pt_features { + /** + * @PT_FEAT_FULL_VA: The table can span the full VA range from 0 to + * PT_VADDR_MAX. + */ + PT_FEAT_FULL_VA, + /** + * @PT_FEAT_DYNAMIC_TOP: The table's top level can be increased + * dynamically during map. This requires HW support for atomically + * setting both the table top pointer and the starting table level. + */ + PT_FEAT_DYNAMIC_TOP, + /** + * @PT_FEAT_SIGN_EXTEND: The top most bit of the valid VA range sign + * extends up to the full pt_vaddr_t. This divides the page table into + * three VA ranges:: + * + * 0 -> 2^N - 1 Lower + * 2^N -> (MAX - 2^N - 1) Non-Canonical + * MAX - 2^N -> MAX Upper + * + * In this mode pt_common::max_vasz_lg2 includes the sign bit and the + * upper bits that don't fall within the translation are just validated. + * + * If not set there is no sign extension and valid VA goes from 0 to 2^N + * - 1. + */ + PT_FEAT_SIGN_EXTEND, + /** + * @PT_FEAT_FLUSH_RANGE: IOTLB maintenance is done by flushing IOVA + * ranges which will clean out any walk cache or any IOPTE fully + * contained by the range. The optimization objective is to minimize the + * number of flushes even if ranges include IOVA gaps that do not need + * to be flushed. + */ + PT_FEAT_FLUSH_RANGE, + /** + * @PT_FEAT_FLUSH_RANGE_NO_GAPS: Like PT_FEAT_FLUSH_RANGE except that + * the optimization objective is to only flush IOVA that has been + * changed. This mode is suitable for cases like hypervisor shadowing + * where flushing unchanged ranges may cause the hypervisor to reparse + * significant amount of page table. + */ + PT_FEAT_FLUSH_RANGE_NO_GAPS, + /* private: */ + PT_FEAT_FMT_START, +}; + +#endif

unmap_pages removes mappings and any fully contained interior tables from the given range. This follows the now-standard iommu_domain API definition where it does not split up larger page sizes into smaller. The caller must perform unmap only on ranges created by map or it must have somehow otherwise determined safe cut points (eg iommufd/vfio use iova_to_phys to scan for them)

A future work will provide 'cut' which explicitly does the page size split if the HW can support it.

unmap is implemented with a recursive descent of the tree. If the caller provides a VA range that spans an entire table item then the table memory can be freed as well.

If an entire table item can be freed then this version will also check the leaf-only level of the tree to ensure that all entries are present to generate -EINVAL. Many of the existing drivers don't do this extra check.

This version sits under the iommu_domain_ops as unmap_pages() but does not require the external page size calculation. The implementation is actually unmap_range() and can do arbitrary ranges, internally handling all the validation and supporting any arrangment of page sizes. A future series can optimize __iommu_unmap() to take advantage of this.

Freed page table memory is batched up in the gather and will be freed in the driver's iotlb_sync() callback after the IOTLB flush completes.

Tested-by: Alejandro Jimenez alejandro.j.jimenez@oracle.com Signed-off-by: Jason Gunthorpe jgg@nvidia.com --- drivers/iommu/generic_pt/iommu_pt.h | 155 ++++++++++++++++++++++++++++ include/linux/generic_pt/iommu.h | 10 +- 2 files changed, 163 insertions(+), 2 deletions(-)

diff --git a/drivers/iommu/generic_pt/iommu_pt.h b/drivers/iommu/generic_pt/iommu_pt.h index 06e1c85340439e..614af5d1e17e50 100644 --- a/drivers/iommu/generic_pt/iommu_pt.h +++ b/drivers/iommu/generic_pt/iommu_pt.h @@ -14,6 +14,29 @@ #include <linux/export.h> #include <linux/iommu.h> #include "../iommu-pages.h" +#include <linux/cleanup.h> +#include <linux/dma-mapping.h> + +static void gather_range_pages(struct iommu_iotlb_gather *iotlb_gather, + struct pt_iommu *iommu_table, pt_vaddr_t iova, + pt_vaddr_t len, + struct iommu_pages_list *free_list) +{ + struct pt_common *common = common_from_iommu(iommu_table); + + if (pt_feature(common, PT_FEAT_FLUSH_RANGE_NO_GAPS) && + iommu_iotlb_gather_is_disjoint(iotlb_gather, iova, len)) { + iommu_iotlb_sync(&iommu_table->domain, iotlb_gather); + /* + * Note that the sync frees the gather's free list, so we must + * not have any pages on that list that are covered by iova/len + */ + } else if (pt_feature(common, PT_FEAT_FLUSH_RANGE)) { + iommu_iotlb_gather_add_range(iotlb_gather, iova, len); + } + + iommu_pages_list_splice(free_list, &iotlb_gather->freelist); +}

#define DOMAIN_NS(op) CONCATENATE(CONCATENATE(pt_iommu_, PTPFX), op)

@@ -167,6 +190,138 @@ static inline struct pt_table_p *table_alloc_top(struct pt_common *common, log2_to_int(pt_top_memsize_lg2(common, top_of_table))); }

+struct pt_unmap_args { + struct iommu_pages_list free_list; + pt_vaddr_t unmapped; +}; + +static __maybe_unused int __unmap_range(struct pt_range *range, void *arg, + unsigned int level, + struct pt_table_p *table) +{ + struct pt_state pts = pt_init(range, level, table); + struct pt_unmap_args *unmap = arg; + unsigned int num_oas = 0; + unsigned int start_index; + int ret = 0; + + _pt_iter_first(&pts); + start_index = pts.index; + pts.type = pt_load_entry_raw(&pts); + /* + * A starting index is in the middle of a contiguous entry + * + * The IOMMU API does not require drivers to support unmapping parts of + * large pages. Long ago VFIO would try to split maps but the current + * version never does. + * + * Instead when unmap reaches a partial unmap of the start of a large + * IOPTE it should remove the entire IOPTE and return that size to the + * caller. + */ + if (pts.type == PT_ENTRY_OA) { + if (log2_mod(range->va, pt_entry_oa_lg2sz(&pts))) + return -EINVAL; + goto start_oa; + } + + do { + if (pts.type != PT_ENTRY_OA) { + bool fully_covered; + + if (pts.type != PT_ENTRY_TABLE) { + ret = -EINVAL; + break; + } + + if (pts.index != start_index) + pt_index_to_va(&pts); + pts.table_lower = pt_table_ptr(&pts); + + fully_covered = pt_item_fully_covered( + &pts, pt_table_item_lg2sz(&pts)); + + ret = pt_descend(&pts, arg, __unmap_range); + if (ret) + break; + + /* + * If the unmapping range fully covers the table then we + * can free it as well. The clear is delayed until we + * succeed in clearing the lower table levels. + */ + if (fully_covered) { + iommu_pages_list_add(&unmap->free_list, + pts.table_lower); + pt_clear_entry(&pts, ilog2(1)); + } + pts.index++; + } else { + unsigned int num_contig_lg2; +start_oa: + /* + * If the caller requested an last that falls within a + * single entry then the entire entry is unmapped and + * the length returned will be larger than requested. + */ + num_contig_lg2 = pt_entry_num_contig_lg2(&pts); + pt_clear_entry(&pts, num_contig_lg2); + num_oas += log2_to_int(num_contig_lg2); + pts.index += log2_to_int(num_contig_lg2); + } + if (pts.index >= pts.end_index) + break; + pts.type = pt_load_entry_raw(&pts); + } while (true); + + unmap->unmapped += log2_mul(num_oas, pt_table_item_lg2sz(&pts)); + return ret; +} + +/** + * unmap_pages() - Make a range of IOVA empty/not present + * @domain: Domain to manipulate + * @iova: IO virtual address to start + * @pgsize: Length of each page + * @pgcount: Length of the range in pgsize units starting from @iova + * @iotlb_gather: Gather struct that must be flushed on return + * + * unmap_pages() will remove a translation created by map_pages(). It cannot + * subdivide a mapping created by map_pages(), so it should be called with IOVA + * ranges that match those passed to map_pages(). The IOVA range can aggregate + * contiguous map_pages() calls so long as no individual range is split. + * + * Context: The caller must hold a write range lock that includes + * the whole range. + * + * Returns: Number of bytes of VA unmapped. iova + res will be the point + * unmapping stopped. + */ +size_t DOMAIN_NS(unmap_pages)(struct iommu_domain *domain, unsigned long iova, + size_t pgsize, size_t pgcount, + struct iommu_iotlb_gather *iotlb_gather) +{ + struct pt_iommu *iommu_table = + container_of(domain, struct pt_iommu, domain); + struct pt_unmap_args unmap = { .free_list = IOMMU_PAGES_LIST_INIT( + unmap.free_list) }; + pt_vaddr_t len = pgsize * pgcount; + struct pt_range range; + int ret; + + ret = make_range(common_from_iommu(iommu_table), &range, iova, len); + if (ret) + return 0; + + pt_walk_range(&range, __unmap_range, &unmap); + + gather_range_pages(iotlb_gather, iommu_table, iova, len, + &unmap.free_list); + + return unmap.unmapped; +} +EXPORT_SYMBOL_NS_GPL(DOMAIN_NS(unmap_pages), "GENERIC_PT_IOMMU"); + static void NS(get_info)(struct pt_iommu *iommu_table, struct pt_iommu_info *info) { diff --git a/include/linux/generic_pt/iommu.h b/include/linux/generic_pt/iommu.h index 5622856e199881..ceb6bc9cea37cd 100644 --- a/include/linux/generic_pt/iommu.h +++ b/include/linux/generic_pt/iommu.h @@ -9,6 +9,7 @@ #include <linux/iommu.h> #include <linux/mm_types.h>

+struct iommu_iotlb_gather; struct pt_iommu_ops;

/** @@ -119,6 +120,10 @@ struct pt_iommu_cfg { #define IOMMU_PROTOTYPES(fmt) \ phys_addr_t pt_iommu_##fmt##_iova_to_phys(struct iommu_domain *domain, \ dma_addr_t iova); \ + size_t pt_iommu_##fmt##_unmap_pages( \ + struct iommu_domain *domain, unsigned long iova, \ + size_t pgsize, size_t pgcount, \ + struct iommu_iotlb_gather *iotlb_gather); \ int pt_iommu_##fmt##_init(struct pt_iommu_##fmt *table, \ const struct pt_iommu_##fmt##_cfg *cfg, \ gfp_t gfp); \ @@ -135,8 +140,9 @@ struct pt_iommu_cfg { * A driver uses IOMMU_PT_DOMAIN_OPS to populate the iommu_domain_ops for the * iommu_pt */ -#define IOMMU_PT_DOMAIN_OPS(fmt) \ - .iova_to_phys = &pt_iommu_##fmt##_iova_to_phys, +#define IOMMU_PT_DOMAIN_OPS(fmt) \ + .iova_to_phys = &pt_iommu_##fmt##_iova_to_phys, \ + .unmap_pages = &pt_iommu_##fmt##_unmap_pages

/* * The driver should setup its domain struct like

map is slightly complicated because it has to handle a number of special edge cases: - Overmapping a previously shared table with an OA - requries validating and freeing the possibly empty tables - Doing the above across an entire to-be-created contiguous entry - Installing a new shared table level concurrently with another thread - Expanding the table by adding more top levels

Table expansion is a unique feature of AMDv1, this version is quite similar except we handle racing concurrent lockless map. The table top pointer and starting level are encoded in a single uintptr_t which ensures we can READ_ONCE() without tearing. Any op will do the READ_ONCE() and use that fixed point as its starting point. Concurrent expansion is handled with a table global spinlock.

When inserting a new table entry map checks that the entire portion of the table is empty. This includes freeing any empty lower tables that will be overwritten by an OA. A separate free list is used while checking and collecting all the empty lower tables so that writing the new entry is uninterrupted, either the new entry fully writes or nothing changes.

A special fast path for PAGE_SIZE is implemented that does a direct walk to the leaf level and installs a single entry. This gives ~15% improvement for iommu_map() when mapping lists of single pages.

This version sits under the iommu_domain_ops as map_pages() but does not require the external page size calculation. The implementation is actually map_range() and can do arbitrary ranges, internally handling all the validation and supporting any arrangment of page sizes. A future series can optimize iommu_map() to take advantage of this.

Tested-by: Alejandro Jimenez alejandro.j.jimenez@oracle.com Signed-off-by: Jason Gunthorpe jgg@nvidia.com --- drivers/iommu/generic_pt/iommu_pt.h | 481 ++++++++++++++++++++++++++++ include/linux/generic_pt/iommu.h | 58 ++++ 2 files changed, 539 insertions(+)

diff --git a/drivers/iommu/generic_pt/iommu_pt.h b/drivers/iommu/generic_pt/iommu_pt.h index 614af5d1e17e50..f15c4ca937c222 100644 --- a/drivers/iommu/generic_pt/iommu_pt.h +++ b/drivers/iommu/generic_pt/iommu_pt.h @@ -91,6 +91,23 @@ static __maybe_unused int make_range_u64(struct pt_common *common, ret; \ })

+static inline unsigned int compute_best_pgsize(struct pt_state *pts, + pt_oaddr_t oa) +{ + struct pt_iommu *iommu_table = iommu_from_common(pts->range->common); + + if (!pt_can_have_leaf(pts)) + return 0; + + /* + * The page size is limited by the domain's bitmap. This allows the core + * code to reduce the supported page sizes by changing the bitmap. + */ + return pt_compute_best_pgsize(pt_possible_sizes(pts) & + iommu_table->domain.pgsize_bitmap, + pts->range->va, pts->range->last_va, oa); +} + static __always_inline int __do_iova_to_phys(struct pt_range *range, void *arg, unsigned int level, struct pt_table_p *table, @@ -190,6 +207,463 @@ static inline struct pt_table_p *table_alloc_top(struct pt_common *common, log2_to_int(pt_top_memsize_lg2(common, top_of_table))); }

+/* Allocate an interior table */ +static inline struct pt_table_p *table_alloc(const struct pt_state *parent_pts, + gfp_t gfp) +{ + struct pt_iommu *iommu_table = + iommu_from_common(parent_pts->range->common); + struct pt_state child_pts = + pt_init(parent_pts->range, parent_pts->level - 1, NULL); + + return iommu_alloc_pages_node_sz( + iommu_table->nid, gfp, + log2_to_int(pt_num_items_lg2(&child_pts) + + ilog2(PT_ITEM_WORD_SIZE))); +} + +static inline int pt_iommu_new_table(struct pt_state *pts, + struct pt_write_attrs *attrs) +{ + struct pt_table_p *table_mem; + phys_addr_t phys; + + /* Given PA/VA/length can't be represented */ + if (unlikely(!pt_can_have_table(pts))) + return -ENXIO; + + table_mem = table_alloc(pts, attrs->gfp); + if (IS_ERR(table_mem)) + return PTR_ERR(table_mem); + + phys = virt_to_phys(table_mem); + if (!pt_install_table(pts, phys, attrs)) { + iommu_free_pages(table_mem); + return -EAGAIN; + } + + if (IS_ENABLED(CONFIG_DEBUG_GENERIC_PT)) { + /* + * The underlying table can't store the physical table address. + * This happens when kunit testing tables outside their normal + * environment where a CPU might be limited. + */ + pt_load_single_entry(pts); + if (PT_WARN_ON(pt_table_pa(pts) != phys)) { + pt_clear_entry(pts, ilog2(1)); + iommu_free_pages(table_mem); + return -EINVAL; + } + } + + pts->table_lower = table_mem; + return 0; +} + +struct pt_iommu_map_args { + struct iommu_iotlb_gather *iotlb_gather; + struct pt_write_attrs attrs; + pt_oaddr_t oa; + unsigned int leaf_pgsize_lg2; + unsigned int leaf_level; +}; + +/* + * Check that the items in a contiguous block are all empty. This will + * recursively check any tables in the block to validate they are empty and then + * free them through the gather. + */ +static int clear_contig(const struct pt_state *start_pts, + struct iommu_iotlb_gather *iotlb_gather, + unsigned int step, unsigned int pgsize_lg2) +{ + struct pt_iommu *iommu_table = + iommu_from_common(start_pts->range->common); + struct pt_range range = *start_pts->range; + struct pt_state pts = + pt_init(&range, start_pts->level, start_pts->table); + struct pt_iommu_collect_args collect = {}; + int ret; + + pts.index = start_pts->index; + pts.end_index = start_pts->index + step; + for (; _pt_iter_load(&pts); pt_next_entry(&pts)) { + if (pts.type == PT_ENTRY_TABLE) { + collect.free_list = + IOMMU_PAGES_LIST_INIT(collect.free_list); + ret = pt_walk_descend_all(&pts, __collect_tables, + &collect); + if (ret) + return ret; + + /* + * The table item must be cleared before we can update + * the gather + */ + pt_clear_entry(&pts, ilog2(1)); + + iommu_pages_list_add(&collect.free_list, + pt_table_ptr(&pts)); + gather_range_pages( + iotlb_gather, iommu_table, range.va, + log2_to_int(pt_table_item_lg2sz(&pts)), + &collect.free_list); + } else if (pts.type != PT_ENTRY_EMPTY) { + return -EADDRINUSE; + } + } + return 0; +} + +static int __map_range_leaf(struct pt_range *range, void *arg, + unsigned int level, struct pt_table_p *table) +{ + struct pt_state pts = pt_init(range, level, table); + struct pt_iommu_map_args *map = arg; + unsigned int leaf_pgsize_lg2 = map->leaf_pgsize_lg2; + unsigned int start_index; + pt_oaddr_t oa = map->oa; + unsigned int step; + bool need_contig; + int ret = 0; + + PT_WARN_ON(map->leaf_level != level); + PT_WARN_ON(!pt_can_have_leaf(&pts)); + + step = log2_to_int_t(unsigned int, + leaf_pgsize_lg2 - pt_table_item_lg2sz(&pts)); + need_contig = leaf_pgsize_lg2 != pt_table_item_lg2sz(&pts); + + _pt_iter_first(&pts); + start_index = pts.index; + do { + pts.type = pt_load_entry_raw(&pts); + if (pts.type != PT_ENTRY_EMPTY || need_contig) { + if (pts.index != start_index) + pt_index_to_va(&pts); + ret = clear_contig(&pts, map->iotlb_gather, step, + leaf_pgsize_lg2); + if (ret) + break; + } + + PT_WARN_ON(compute_best_pgsize(&pts, oa) != leaf_pgsize_lg2); + + pt_install_leaf_entry(&pts, oa, leaf_pgsize_lg2, &map->attrs); + + oa += log2_to_int(leaf_pgsize_lg2); + pts.index += step; + } while (pts.index < pts.end_index); + + map->oa = oa; + return ret; +} + +static int __map_range(struct pt_range *range, void *arg, unsigned int level, + struct pt_table_p *table) +{ + struct pt_state pts = pt_init(range, level, table); + struct pt_iommu_map_args *map = arg; + int ret; + + PT_WARN_ON(map->leaf_level == level); + PT_WARN_ON(!pt_can_have_table(&pts)); + + _pt_iter_first(&pts); + + /* Descend to a child table */ + do { + pts.type = pt_load_entry_raw(&pts); + + if (pts.type != PT_ENTRY_TABLE) { + if (pts.type != PT_ENTRY_EMPTY) + return -EADDRINUSE; + ret = pt_iommu_new_table(&pts, &map->attrs); + if (ret) { + /* + * Racing with another thread installing a table + */ + if (ret == -EAGAIN) + continue; + return ret; + } + } else { + pts.table_lower = pt_table_ptr(&pts); + } + + /* + * The already present table can possibly be shared with another + * concurrent map. + */ + if (map->leaf_level == level - 1) + ret = pt_descend(&pts, arg, __map_range_leaf); + else + ret = pt_descend(&pts, arg, __map_range); + if (ret) + return ret; + + pts.index++; + pt_index_to_va(&pts); + if (pts.index >= pts.end_index) + break; + pts.type = pt_load_entry_raw(&pts); + } while (true); + return 0; +} + +static __always_inline int __do_map_single_page(struct pt_range *range, + void *arg, unsigned int level, + struct pt_table_p *table, + pt_level_fn_t descend_fn) +{ + struct pt_state pts = pt_init(range, level, table); + struct pt_iommu_map_args *map = arg; + + pts.type = pt_load_single_entry(&pts); + if (level == 0) { + if (pts.type != PT_ENTRY_EMPTY) + return -EADDRINUSE; + pt_install_leaf_entry(&pts, map->oa, PAGE_SHIFT, + &map->attrs); + map->oa += PAGE_SIZE; + return 0; + } + if (pts.type != PT_ENTRY_TABLE) + return -EAGAIN; + return pt_descend(&pts, arg, descend_fn); +} +PT_MAKE_LEVELS(__map_single_page, __do_map_single_page); + +/* + * Add a table to the top, increasing the top level as much as necessary to + * encompass range. + */ +static int increase_top(struct pt_iommu *iommu_table, struct pt_range *range, + struct pt_iommu_map_args *map) +{ + struct iommu_pages_list free_list = IOMMU_PAGES_LIST_INIT(free_list); + struct pt_common *common = common_from_iommu(iommu_table); + uintptr_t top_of_table = READ_ONCE(common->top_of_table); + uintptr_t new_top_of_table = top_of_table; + struct pt_table_p *table_mem; + unsigned int new_level; + spinlock_t *domain_lock; + unsigned long flags; + int ret; + + while (true) { + struct pt_range top_range = + _pt_top_range(common, new_top_of_table); + struct pt_state pts = pt_init_top(&top_range); + + top_range.va = range->va; + top_range.last_va = range->last_va; + + if (!pt_check_range(&top_range) && map->leaf_level <= pts.level) + break; + + pts.level++; + if (pts.level > PT_MAX_TOP_LEVEL || + pt_table_item_lg2sz(&pts) >= common->max_vasz_lg2) { + ret = -ERANGE; + goto err_free; + } + + new_level = pts.level; + table_mem = table_alloc_top( + common, _pt_top_set(NULL, pts.level), map->attrs.gfp); + if (IS_ERR(table_mem)) + return PTR_ERR(table_mem); + iommu_pages_list_add(&free_list, table_mem); + + /* The new table links to the lower table always at index 0 */ + top_range.va = 0; + top_range.top_level = new_level; + pts.table_lower = pts.table; + pts.table = table_mem; + pt_load_single_entry(&pts); + PT_WARN_ON(pts.index != 0); + pt_install_table(&pts, virt_to_phys(pts.table_lower), + &map->attrs); + new_top_of_table = _pt_top_set(pts.table, pts.level); + } + + /* + * top_of_table is write locked by the spinlock, but readers can use + * READ_ONCE() to get the value. Since we encode both the level and the + * pointer in one quanta the lockless reader will always see something + * valid. The HW must be updated to the new level under the spinlock + * before top_of_table is updated so that concurrent readers don't map + * into the new level until it is fully functional. If another thread + * already updated it while we were working then throw everything away + * and try again. + */ + domain_lock = iommu_table->hw_flush_ops->get_top_lock(iommu_table); + spin_lock_irqsave(domain_lock, flags); + if (common->top_of_table != top_of_table) { + spin_unlock_irqrestore(domain_lock, flags); + ret = -EAGAIN; + goto err_free; + } + + /* + * We do not issue any flushes for change_top on the expectation that + * any walk cache will not become a problem by adding another layer to + * the tree. Misses will rewalk from the updated top pointer, hits + * continue to be correct. Negative caching is fine too since all the + * new IOVA added by the new top is non-present. + */ + iommu_table->hw_flush_ops->change_top( + iommu_table, virt_to_phys(table_mem), new_level); + WRITE_ONCE(common->top_of_table, new_top_of_table); + spin_unlock_irqrestore(domain_lock, flags); + return 0; + +err_free: + iommu_put_pages_list(&free_list); + return ret; +} + +static int check_map_range(struct pt_iommu *iommu_table, struct pt_range *range, + struct pt_iommu_map_args *map) +{ + struct pt_common *common = common_from_iommu(iommu_table); + int ret; + + do { + ret = pt_check_range(range); + if (!pt_feature(common, PT_FEAT_DYNAMIC_TOP)) + return ret; + + if (!ret && map->leaf_level <= range->top_level) + break; + + ret = increase_top(iommu_table, range, map); + if (ret && ret != -EAGAIN) + return ret; + + /* Reload the new top */ + *range = pt_make_range(common, range->va, range->last_va); + } while (ret); + PT_WARN_ON(pt_check_range(range)); + return 0; +} + +/** + * map_pages() - Install translation for an IOVA range + * @domain: Domain to manipulate + * @iova: IO virtual address to start + * @paddr: Physical/Output address to start + * @pgsize: Length of each page + * @pgcount: Length of the range in pgsize units starting from @iova + * @prot: A bitmap of IOMMU_READ/WRITE/CACHE/NOEXEC/MMIO + * @gfp: GFP flags for any memory allocations + * @mapped: Total bytes successfully mapped + * + * The range starting at IOVA will have paddr installed into it. The caller + * must specify a valid pgsize and pgcount to segment the range into compatible + * blocks. + * + * On error the caller will probably want to invoke unmap on the range from iova + * up to the amount indicated by @mapped to return the table back to an + * unchanged state. + * + * Context: The caller must hold a write range lock that includes the whole + * range. + * + * Returns: -ERRNO on failure, 0 on success. The number of bytes of VA that were + * mapped are added to @mapped, @mapped is not zerod first. + */ +int DOMAIN_NS(map_pages)(struct iommu_domain *domain, unsigned long iova, + phys_addr_t paddr, size_t pgsize, size_t pgcount, + int prot, gfp_t gfp, size_t *mapped) +{ + struct pt_iommu *iommu_table = + container_of(domain, struct pt_iommu, domain); + pt_vaddr_t pgsize_bitmap = iommu_table->domain.pgsize_bitmap; + struct pt_common *common = common_from_iommu(iommu_table); + struct iommu_iotlb_gather iotlb_gather; + pt_vaddr_t len = pgsize * pgcount; + struct pt_iommu_map_args map = { + .iotlb_gather = &iotlb_gather, + .oa = paddr, + .leaf_pgsize_lg2 = log2_ffs(pgsize), + }; + bool single_page = false; + struct pt_range range; + int ret; + + iommu_iotlb_gather_init(&iotlb_gather); + + if (WARN_ON(!(prot & (IOMMU_READ | IOMMU_WRITE)))) + return -EINVAL; + + /* Check the paddr doesn't exceed what the table can store */ + if ((sizeof(pt_oaddr_t) < sizeof(paddr) && + (pt_vaddr_t)paddr > PT_VADDR_MAX) || + (common->max_oasz_lg2 != PT_VADDR_MAX_LG2 && + oalog2_div(paddr, common->max_oasz_lg2))) + return -ERANGE; + + ret = pt_iommu_set_prot(common, &map.attrs, prot); + if (ret) + return ret; + map.attrs.gfp = gfp; + + ret = make_range_no_check(common, &range, iova, len); + if (ret) + return ret; + + /* Calculate target page size and level for the leaves */ + if (pt_has_system_page(common) && pgsize == PAGE_SIZE && pgcount == 1) { + PT_WARN_ON(!(pgsize_bitmap & PAGE_SIZE)); + if (log2_mod(iova | paddr, PAGE_SHIFT)) + return -ENXIO; + map.leaf_pgsize_lg2 = PAGE_SHIFT; + map.leaf_level = 0; + single_page = true; + } else { + map.leaf_pgsize_lg2 = pt_compute_best_pgsize( + pgsize_bitmap, range.va, range.last_va, paddr); + if (!map.leaf_pgsize_lg2) + return -ENXIO; + map.leaf_level = + pt_pgsz_lg2_to_level(common, map.leaf_pgsize_lg2); + } + + ret = check_map_range(iommu_table, &range, &map); + if (ret) + return ret; + + PT_WARN_ON(map.leaf_level > range.top_level); + + do { + if (single_page) { + ret = pt_walk_range(&range, __map_single_page, &map); + if (ret != -EAGAIN) + break; + } + + if (map.leaf_level == range.top_level) + ret = pt_walk_range(&range, __map_range_leaf, &map); + else + ret = pt_walk_range(&range, __map_range, &map); + } while (false); + + /* + * Table levels were freed and replaced with large items, flush any walk + * cache that may refer to the freed levels. + */ + if (!iommu_pages_list_empty(&iotlb_gather.freelist)) + iommu_iotlb_sync(&iommu_table->domain, &iotlb_gather); + + /* Bytes successfully mapped */ + PT_WARN_ON(!ret && map.oa - paddr != len); + *mapped += map.oa - paddr; + return ret; +} +EXPORT_SYMBOL_NS_GPL(DOMAIN_NS(map_pages), "GENERIC_PT_IOMMU"); + struct pt_unmap_args { struct iommu_pages_list free_list; pt_vaddr_t unmapped; @@ -450,6 +924,7 @@ static void pt_iommu_zero(struct pt_iommu_table *fmt_table) memset_after(fmt_table, 0, iommu.domain);

/* The caller can initialize some of these values */ + iommu_table->hw_flush_ops = cfg.hw_flush_ops; iommu_table->nid = cfg.nid; }

@@ -483,6 +958,12 @@ int pt_iommu_init(struct pt_iommu_table *fmt_table, if (ret) return ret;

+ if (pt_feature(common, PT_FEAT_DYNAMIC_TOP) && + WARN_ON(!iommu_table->hw_flush_ops || + !iommu_table->hw_flush_ops->change_top || + !iommu_table->hw_flush_ops->get_top_lock)) + return -EINVAL; + if (pt_feature(common, PT_FEAT_SIGN_EXTEND) && (pt_feature(common, PT_FEAT_FULL_VA) || pt_feature(common, PT_FEAT_DYNAMIC_TOP))) diff --git a/include/linux/generic_pt/iommu.h b/include/linux/generic_pt/iommu.h index ceb6bc9cea37cd..2769689379c26d 100644 --- a/include/linux/generic_pt/iommu.h +++ b/include/linux/generic_pt/iommu.h @@ -11,6 +11,7 @@

struct iommu_iotlb_gather; struct pt_iommu_ops; +struct pt_iommu_flush_ops;

/** * DOC: IOMMU Radix Page Table @@ -43,6 +44,12 @@ struct pt_iommu { */ const struct pt_iommu_ops *ops;

+ /** + * @hw_flush_ops: Function pointers provided by the HW driver to flush + * HW caches after changes to the page table. + */ + const struct pt_iommu_flush_ops *hw_flush_ops; + /** * @nid: Node ID to use for table memory allocations. The IOMMU driver * may want to set the NID to the device's NID, if there are multiple @@ -84,6 +91,52 @@ struct pt_iommu_ops { void (*deinit)(struct pt_iommu *iommu_table); };

+/** + * struct pt_iommu_flush_ops - HW IOTLB cache flushing operations + * + * The IOMMU driver should implement these using container_of(iommu_table) to + * get to it's iommu_domain dervied structure. All ops can be called in atomic + * contexts as they are buried under DMA API calls. + */ +struct pt_iommu_flush_ops { + /** + * @change_top: Update the top of table pointer + * @iommu_table: Table to operate on + * @top_paddr: New CPU physical address of the top pointer + * @top_level: IOMMU PT level of the new top + * + * Called under the get_top_lock() spinlock. The driver must update all + * HW references to this domain with a new top address and + * configuration. On return mappings placed in the new top must be + * reachable by the HW. + * + * top_level encodes the level in IOMMU PT format, level 0 is the + * smallest page size increasing from there. This has to be translated + * to any HW specific format. During this call the new top will not be + * visible to any other API. + * + * This op is only used by PT_FEAT_DYNAMIC_TOP, and is required if + * enabled. + */ + void (*change_top)(struct pt_iommu *iommu_table, phys_addr_t top_paddr, + unsigned int top_level); + + /** + * @get_top_lock: Return a lock to hold when changing the table top + * @iommu_table: Table to operate on + * + * page table from being stored in HW. The lock will be held prior + * to calling change_top() and released once the top is fully visible. + * + * Typically this would be a lock that protects the iommu_domain's + * attachment list. + * + * This op is only used by PT_FEAT_DYNAMIC_TOP, and is required if + * enabled. + */ + spinlock_t *(*get_top_lock)(struct pt_iommu *iommu_table); +}; + static inline void pt_iommu_deinit(struct pt_iommu *iommu_table) { /* @@ -120,6 +173,10 @@ struct pt_iommu_cfg { #define IOMMU_PROTOTYPES(fmt) \ phys_addr_t pt_iommu_##fmt##_iova_to_phys(struct iommu_domain *domain, \ dma_addr_t iova); \ + int pt_iommu_##fmt##_map_pages(struct iommu_domain *domain, \ + unsigned long iova, phys_addr_t paddr, \ + size_t pgsize, size_t pgcount, \ + int prot, gfp_t gfp, size_t *mapped); \ size_t pt_iommu_##fmt##_unmap_pages( \ struct iommu_domain *domain, unsigned long iova, \ size_t pgsize, size_t pgcount, \ @@ -142,6 +199,7 @@ struct pt_iommu_cfg { */ #define IOMMU_PT_DOMAIN_OPS(fmt) \ .iova_to_phys = &pt_iommu_##fmt##_iova_to_phys, \ + .map_pages = &pt_iommu_##fmt##_map_pages, \ .unmap_pages = &pt_iommu_##fmt##_unmap_pages

/*

IOMMU HW now supports updating a dirty bit in an entry when a DMA writes to the entry's VA range. iommufd has a uAPI to read and clear the dirty bits from the tables.

This is a trivial recursive descent algorithm to read and optionally clear the dirty bits. The format needs a function to tell if a contiguous entry is dirty, and a function to clear a contiguous entry back to clean.

Tested-by: Alejandro Jimenez alejandro.j.jimenez@oracle.com Signed-off-by: Jason Gunthorpe jgg@nvidia.com --- drivers/iommu/generic_pt/iommu_pt.h | 104 ++++++++++++++++++++++++++++ include/linux/generic_pt/iommu.h | 6 ++ 2 files changed, 110 insertions(+)

diff --git a/drivers/iommu/generic_pt/iommu_pt.h b/drivers/iommu/generic_pt/iommu_pt.h index f15c4ca937c222..c9062e3bf3efad 100644 --- a/drivers/iommu/generic_pt/iommu_pt.h +++ b/drivers/iommu/generic_pt/iommu_pt.h @@ -162,6 +162,108 @@ phys_addr_t DOMAIN_NS(iova_to_phys)(struct iommu_domain *domain, } EXPORT_SYMBOL_NS_GPL(DOMAIN_NS(iova_to_phys), "GENERIC_PT_IOMMU");

+struct pt_iommu_dirty_args { + struct iommu_dirty_bitmap *dirty; + unsigned int flags; +}; + +static void record_dirty(struct pt_state *pts, + struct pt_iommu_dirty_args *dirty, + unsigned int num_contig_lg2) +{ + pt_vaddr_t dirty_len; + + if (num_contig_lg2 != ilog2(1)) { + unsigned int index = pts->index; + unsigned int end_index = log2_set_mod_max_t( + unsigned int, pts->index, num_contig_lg2); + + /* Adjust for being contained inside a contiguous page */ + end_index = min(end_index, pts->end_index); + dirty_len = (end_index - index) * + log2_to_int(pt_table_item_lg2sz(pts)); + } else { + dirty_len = log2_to_int(pt_table_item_lg2sz(pts)); + } + + if (dirty->dirty->bitmap) + iova_bitmap_set(dirty->dirty->bitmap, pts->range->va, + dirty_len); + + if (!(dirty->flags & IOMMU_DIRTY_NO_CLEAR)) { + pt_entry_set_write_clean(pts); + iommu_iotlb_gather_add_range(dirty->dirty->gather, + pts->range->va, dirty_len); + } +} + +static inline int __read_and_clear_dirty(struct pt_range *range, void *arg, + unsigned int level, + struct pt_table_p *table) +{ + struct pt_state pts = pt_init(range, level, table); + struct pt_iommu_dirty_args *dirty = arg; + int ret; + + for_each_pt_level_entry(&pts) { + if (pts.type == PT_ENTRY_TABLE) { + ret = pt_descend(&pts, arg, __read_and_clear_dirty); + if (ret) + return ret; + continue; + } + if (pts.type == PT_ENTRY_OA && pt_entry_write_is_dirty(&pts)) + record_dirty(&pts, dirty, + pt_entry_num_contig_lg2(&pts)); + } + return 0; +} + +/** + * read_and_clear_dirty() - Manipulate the HW set write dirty state + * @domain: Domain to manipulate + * @iova: IO virtual address to start + * @size: Length of the IOVA + * @flags: A bitmap of IOMMU_DIRTY_NO_CLEAR + * @dirty: Place to store the dirty bits + * + * Iterate over all the entries in the mapped range and record their write dirty + * status in iommu_dirty_bitmap. If IOMMU_DIRTY_NO_CLEAR is not specified then + * the entries will be left dirty, otherwise they are returned to being not + * write dirty. + * + * Context: The caller must hold a read range lock that includes @iova. + * + * Returns: -ERRNO on failure, 0 on success. + */ +int DOMAIN_NS(read_and_clear_dirty)(struct iommu_domain *domain, + unsigned long iova, size_t size, + unsigned long flags, + struct iommu_dirty_bitmap *dirty) +{ + struct pt_iommu *iommu_table = + container_of(domain, struct pt_iommu, domain); + struct pt_iommu_dirty_args dirty_args = { + .dirty = dirty, + .flags = flags, + }; + struct pt_range range; + int ret; + +#if !IS_ENABLED(CONFIG_IOMMUFD_DRIVER) || !defined(pt_entry_write_is_dirty) + return -EOPNOTSUPP; +#endif + + ret = make_range(common_from_iommu(iommu_table), &range, iova, size); + if (ret) + return ret; + + ret = pt_walk_range(&range, __read_and_clear_dirty, &dirty_args); + PT_WARN_ON(ret); + return ret; +} +EXPORT_SYMBOL_NS_GPL(DOMAIN_NS(read_and_clear_dirty), "GENERIC_PT_IOMMU"); + struct pt_iommu_collect_args { struct iommu_pages_list free_list; u8 ignore_mapped : 1; @@ -1002,5 +1104,7 @@ EXPORT_SYMBOL_NS_GPL(pt_iommu_hw_info, "GENERIC_PT_IOMMU"); MODULE_LICENSE("GPL"); MODULE_DESCRIPTION("IOMMU Page table implementation for " __stringify(PTPFX_RAW)); MODULE_IMPORT_NS("GENERIC_PT"); +/* For iommu_dirty_bitmap_record() */ +MODULE_IMPORT_NS("IOMMUFD");

#endif diff --git a/include/linux/generic_pt/iommu.h b/include/linux/generic_pt/iommu.h index 2769689379c26d..21dbe0bf3880b3 100644 --- a/include/linux/generic_pt/iommu.h +++ b/include/linux/generic_pt/iommu.h @@ -12,6 +12,7 @@ struct iommu_iotlb_gather; struct pt_iommu_ops; struct pt_iommu_flush_ops; +struct iommu_dirty_bitmap;

/** * DOC: IOMMU Radix Page Table @@ -181,6 +182,9 @@ struct pt_iommu_cfg { struct iommu_domain *domain, unsigned long iova, \ size_t pgsize, size_t pgcount, \ struct iommu_iotlb_gather *iotlb_gather); \ + int pt_iommu_##fmt##_read_and_clear_dirty( \ + struct iommu_domain *domain, unsigned long iova, size_t size, \ + unsigned long flags, struct iommu_dirty_bitmap *dirty); \ int pt_iommu_##fmt##_init(struct pt_iommu_##fmt *table, \ const struct pt_iommu_##fmt##_cfg *cfg, \ gfp_t gfp); \ @@ -201,6 +205,8 @@ struct pt_iommu_cfg { .iova_to_phys = &pt_iommu_##fmt##_iova_to_phys, \ .map_pages = &pt_iommu_##fmt##_map_pages, \ .unmap_pages = &pt_iommu_##fmt##_unmap_pages +#define IOMMU_PT_DIRTY_OPS(fmt) \ + .read_and_clear_dirty = &pt_iommu_##fmt##_read_and_clear_dirty

/* * The driver should setup its domain struct like

This intends to have high coverage of the page table format functions, it uses the IOMMU implementation to create a tree which it then walks through and directly calls the generic page table functions to test them.

It is a good starting point to test a new format header as it is often able to find typos and inconsistencies much more directly, rather than with an obscure failure in the iommu implementation.

The tests can be run with commands like:

tools/testing/kunit/kunit.py run --build_dir build_kunit_arm64 --arch arm64 --make_options LLVM=-19 --kunitconfig ./drivers/iommu/generic_pt/.kunitconfig tools/testing/kunit/kunit.py run --build_dir build_kunit_uml --kunitconfig ./drivers/iommu/generic_pt/.kunitconfig --kconfig_add CONFIG_WERROR=n tools/testing/kunit/kunit.py run --build_dir build_kunit_x86_64 --arch x86_64 --kunitconfig ./drivers/iommu/generic_pt/.kunitconfig tools/testing/kunit/kunit.py run --build_dir build_kunit_i386 --arch i386 --kunitconfig ./drivers/iommu/generic_pt/.kunitconfig tools/testing/kunit/kunit.py run --build_dir build_kunit_i386pae --arch i386 --kunitconfig ./drivers/iommu/generic_pt/.kunitconfig --kconfig_add CONFIG_X86_PAE=y

There are several interesting corner cases on the 32 bit platforms that need checking.

The format can declare a list of configurations that generate different configurations the initialize the page table, for instance with different top levels or other parameters. The kunit will turn these into "params" which cause each test to run multiple times.

The tests are repeated to run at every table level to check that all the item encoding formats work.

The following are checked: - Basic init works for each configuration - The various log2 functions have the expected behavior at the limits - pt_compute_best_pgsize() works - pt_table_pa() reads back what pt_install_table() writes - range.max_vasz_lg2 works properly - pt_table_oa_lg2sz() and pt_table_item_lg2sz() use a contiguous non-overlapping set of bits from the VA up to the defined max_va - pt_possible_sizes() and pt_can_have_leaf() produces a sensible layout - pt_item_oa(), pt_entry_oa(), and pt_entry_num_contig_lg2() read back what pt_install_leaf_entry() writes - pt_clear_entry() works - pt_attr_from_entry() reads back what pt_iommu_set_prot() & pt_install_leaf_entry() writes - pt_entry_set_write_clean(), pt_entry_make_write_dirty(), and pt_entry_write_is_dirty() work

Tested-by: Alejandro Jimenez alejandro.j.jimenez@oracle.com Signed-off-by: Jason Gunthorpe jgg@nvidia.com --- drivers/iommu/generic_pt/.kunitconfig | 12 + drivers/iommu/generic_pt/Kconfig | 12 + drivers/iommu/generic_pt/fmt/Makefile | 12 + drivers/iommu/generic_pt/fmt/amdv1.h | 10 + drivers/iommu/generic_pt/fmt/iommu_template.h | 8 + drivers/iommu/generic_pt/kunit_generic_pt.h | 717 ++++++++++++++++++ drivers/iommu/generic_pt/kunit_iommu.h | 181 +++++ 7 files changed, 952 insertions(+) create mode 100644 drivers/iommu/generic_pt/.kunitconfig create mode 100644 drivers/iommu/generic_pt/kunit_generic_pt.h create mode 100644 drivers/iommu/generic_pt/kunit_iommu.h

diff --git a/drivers/iommu/generic_pt/.kunitconfig b/drivers/iommu/generic_pt/.kunitconfig new file mode 100644 index 00000000000000..936c327f0661cf --- /dev/null +++ b/drivers/iommu/generic_pt/.kunitconfig @@ -0,0 +1,12 @@ +CONFIG_KUNIT=y +CONFIG_GENERIC_PT=y +CONFIG_DEBUG_GENERIC_PT=y +CONFIG_IOMMU_PT=y +CONFIG_IOMMU_PT_AMDV1=y +CONFIG_IOMMU_PT_KUNIT_TEST=y + +CONFIG_IOMMUFD=y +CONFIG_DEBUG_KERNEL=y +CONFIG_FAULT_INJECTION=y +CONFIG_RUNTIME_TESTING_MENU=y +CONFIG_IOMMUFD_TEST=y diff --git a/drivers/iommu/generic_pt/Kconfig b/drivers/iommu/generic_pt/Kconfig index 208c8178d5dbd2..a117cc23a1c328 100644 --- a/drivers/iommu/generic_pt/Kconfig +++ b/drivers/iommu/generic_pt/Kconfig @@ -40,5 +40,17 @@ config IOMMU_PT_AMDV1 power of 2 and decodes an full 64-bit IOVA space.

Selected automatically by an IOMMU driver that uses this format. + +config IOMMU_PT_KUNIT_TEST + tristate "IOMMU Page Table KUnit Test" if !KUNIT_ALL_TESTS + depends on KUNIT + depends on IOMMU_PT_AMDV1 || !IOMMU_PT_AMDV1 + default KUNIT_ALL_TESTS + help + Enable kunit tests for GENERIC_PT and IOMMU_PT that covers all the + enabled page table formats. The test covers most of the GENERIC_PT + functions provided by the page table format, as well as covering the + iommu_domain related functions. + endif endif diff --git a/drivers/iommu/generic_pt/fmt/Makefile b/drivers/iommu/generic_pt/fmt/Makefile index a4d83b7e0cf691..32f3956c7509f8 100644 --- a/drivers/iommu/generic_pt/fmt/Makefile +++ b/drivers/iommu/generic_pt/fmt/Makefile @@ -2,10 +2,22 @@

iommu_pt_fmt-$(CONFIG_IOMMU_PT_AMDV1) += amdv1

+IOMMU_PT_KUNIT_TEST := define create_format obj-$(2) += iommu_$(1).o +iommu_pt_kunit_test-y += kunit_iommu_$(1).o +CFLAGS_kunit_iommu_$(1).o += -DGENERIC_PT_KUNIT=1 +IOMMU_PT_KUNIT_TEST := iommu_pt_kunit_test.o

endef

$(eval $(foreach fmt,$(iommu_pt_fmt-y),$(call create_format,$(fmt),y))) $(eval $(foreach fmt,$(iommu_pt_fmt-m),$(call create_format,$(fmt),m))) + +# The kunit objects are constructed by compiling the main source +# with -DGENERIC_PT_KUNIT +$(obj)/kunit_iommu_%.o: $(src)/iommu_%.c FORCE + $(call rule_mkdir) + $(call if_changed_dep,cc_o_c) + +obj-$(CONFIG_IOMMU_PT_KUNIT_TEST) += $(IOMMU_PT_KUNIT_TEST) diff --git a/drivers/iommu/generic_pt/fmt/amdv1.h b/drivers/iommu/generic_pt/fmt/amdv1.h index 901fc4a80e9a83..cd72688322c6bd 100644 --- a/drivers/iommu/generic_pt/fmt/amdv1.h +++ b/drivers/iommu/generic_pt/fmt/amdv1.h @@ -382,4 +382,14 @@ amdv1pt_iommu_fmt_hw_info(struct pt_iommu_amdv1 *table, info->mode = top_range->top_level + 1; } #define pt_iommu_fmt_hw_info amdv1pt_iommu_fmt_hw_info + +#if defined(GENERIC_PT_KUNIT) +static const struct pt_iommu_amdv1_cfg amdv1_kunit_fmt_cfgs[] = { + /* Matches what io_pgtable does */ + [0] = { .starting_level = 2 }, +}; +#define kunit_fmt_cfgs amdv1_kunit_fmt_cfgs +enum { KUNIT_FMT_FEATURES = 0 }; +#endif + #endif diff --git a/drivers/iommu/generic_pt/fmt/iommu_template.h b/drivers/iommu/generic_pt/fmt/iommu_template.h index 5b631bc07cbc16..11e85106ae302e 100644 --- a/drivers/iommu/generic_pt/fmt/iommu_template.h +++ b/drivers/iommu/generic_pt/fmt/iommu_template.h @@ -36,4 +36,12 @@ #include PT_FMT_H #include "../pt_common.h"

+#ifndef GENERIC_PT_KUNIT #include "../iommu_pt.h" +#else +/* + * The makefile will compile the .c file twice, once with GENERIC_PT_KUNIT set + * which means we are building the kunit modle. + */ +#include "../kunit_generic_pt.h" +#endif diff --git a/drivers/iommu/generic_pt/kunit_generic_pt.h b/drivers/iommu/generic_pt/kunit_generic_pt.h new file mode 100644 index 00000000000000..3f3eab35e5cb1b --- /dev/null +++ b/drivers/iommu/generic_pt/kunit_generic_pt.h @@ -0,0 +1,717 @@ +/* SPDX-License-Identifier: GPL-2.0-only */ +/* + * Copyright (c) 2024-2025, NVIDIA CORPORATION & AFFILIATES + * + * Test the format API directly. + * + */ +#include "kunit_iommu.h" +#include "pt_iter.h" + +static void do_map(struct kunit *test, pt_vaddr_t va, pt_oaddr_t pa, + pt_vaddr_t len) +{ + struct kunit_iommu_priv *priv = test->priv; + int ret; + + KUNIT_ASSERT_EQ(test, len, (size_t)len); + + ret = iommu_map(&priv->domain, va, pa, len, IOMMU_READ | IOMMU_WRITE, + GFP_KERNEL); + KUNIT_ASSERT_NO_ERRNO_FN(test, "map_pages", ret); +} + +#define KUNIT_ASSERT_PT_LOAD(test, pts, entry) \ + ({ \ + pt_load_entry(pts); \ + KUNIT_ASSERT_EQ(test, (pts)->type, entry); \ + }) + +struct check_levels_arg { + struct kunit *test; + void *fn_arg; + void (*fn)(struct kunit *test, struct pt_state *pts, void *arg); +}; + +static int __check_all_levels(struct pt_range *range, void *arg, + unsigned int level, struct pt_table_p *table) +{ + struct pt_state pts = pt_init(range, level, table); + struct check_levels_arg *chk = arg; + struct kunit *test = chk->test; + int ret; + + _pt_iter_first(&pts); + + + /* + * If we were able to use the full VA space this should always be the + * last index in each table. + */ + if (!(IS_32BIT && range->max_vasz_lg2 > 32)) { + if (pt_feature(range->common, PT_FEAT_SIGN_EXTEND) && + pts.level == pts.range->top_level) + KUNIT_ASSERT_EQ(test, pts.index, + log2_to_int(range->max_vasz_lg2 - 1 - + pt_table_item_lg2sz(&pts)) - + 1); + else + KUNIT_ASSERT_EQ(test, pts.index, + log2_to_int(pt_table_oa_lg2sz(&pts) - + pt_table_item_lg2sz(&pts)) - + 1); + } + + if (pt_can_have_table(&pts)) { + pt_load_single_entry(&pts); + KUNIT_ASSERT_EQ(test, pts.type, PT_ENTRY_TABLE); + ret = pt_descend(&pts, arg, __check_all_levels); + KUNIT_ASSERT_EQ(test, ret, 0); + + /* Index 0 is used by the test */ + if (IS_32BIT && !pts.index) + return 0; + KUNIT_ASSERT_NE(chk->test, pts.index, 0); + } + + /* + * A format should not create a table with only one entry, at least this + * test approach won't work. + */ + KUNIT_ASSERT_GT(chk->test, pts.end_index, 1); + + /* + * For increase top we end up using index 0 for the original top's tree, + * so use index 1 for testing instead. + */ + pts.index = 0; + pt_index_to_va(&pts); + pt_load_single_entry(&pts); + if (pts.type == PT_ENTRY_TABLE && pts.end_index > 2) { + pts.index = 1; + pt_index_to_va(&pts); + } + (*chk->fn)(chk->test, &pts, chk->fn_arg); + return 0; +} + +/* + * Call fn for each level in the table with a pts setup to index 0 in a table + * for that level. This allows writing tests that run on every level. + * The test can use every index in the table except the last one. + */ +static void check_all_levels(struct kunit *test, + void (*fn)(struct kunit *test, + struct pt_state *pts, void *arg), + void *fn_arg) +{ + struct kunit_iommu_priv *priv = test->priv; + struct pt_range range = pt_top_range(priv->common); + struct check_levels_arg chk = { + .test = test, + .fn = fn, + .fn_arg = fn_arg, + }; + int ret; + + if (pt_feature(priv->common, PT_FEAT_DYNAMIC_TOP) && + priv->common->max_vasz_lg2 > range.max_vasz_lg2) + range.last_va = fvalog2_set_mod_max(range.va, + priv->common->max_vasz_lg2); + + /* + * Map a page at the highest VA, this will populate all the levels so we + * can then iterate over them. Index 0 will be used for testing. + */ + if (IS_32BIT && range.max_vasz_lg2 > 32) + range.last_va = (u32)range.last_va; + range.va = range.last_va - (priv->smallest_pgsz - 1); + do_map(test, range.va, 0, priv->smallest_pgsz); + + range = pt_make_range(priv->common, range.va, range.last_va); + ret = pt_walk_range(&range, __check_all_levels, &chk); + KUNIT_ASSERT_EQ(test, ret, 0); +} + +static void test_init(struct kunit *test) +{ + struct kunit_iommu_priv *priv = test->priv; + + /* Fixture does the setup */ + KUNIT_ASSERT_NE(test, priv->info.pgsize_bitmap, 0); +} + +/* + * Basic check that the log2_* functions are working, especially at the integer + * limits. + */ +static void test_bitops(struct kunit *test) +{ + int i; + + KUNIT_ASSERT_EQ(test, log2_fls_t(u32, 0), 0); + KUNIT_ASSERT_EQ(test, log2_fls_t(u32, 1), 1); + KUNIT_ASSERT_EQ(test, log2_fls_t(u32, BIT(2)), 3); + KUNIT_ASSERT_EQ(test, log2_fls_t(u32, U32_MAX), 32); + + KUNIT_ASSERT_EQ(test, log2_fls_t(u64, 0), 0); + KUNIT_ASSERT_EQ(test, log2_fls_t(u64, 1), 1); + KUNIT_ASSERT_EQ(test, log2_fls_t(u64, BIT(2)), 3); + KUNIT_ASSERT_EQ(test, log2_fls_t(u64, U64_MAX), 64); + + KUNIT_ASSERT_EQ(test, log2_ffs_t(u32, 1), 0); + KUNIT_ASSERT_EQ(test, log2_ffs_t(u32, BIT(2)), 2); + KUNIT_ASSERT_EQ(test, log2_ffs_t(u32, BIT(31)), 31); + + KUNIT_ASSERT_EQ(test, log2_ffs_t(u64, 1), 0); + KUNIT_ASSERT_EQ(test, log2_ffs_t(u64, BIT(2)), 2); + KUNIT_ASSERT_EQ(test, log2_ffs_t(u64, BIT_ULL(63)), 63); + + for (i = 0; i != 31; i++) + KUNIT_ASSERT_EQ(test, log2_ffz_t(u64, BIT_ULL(i) - 1), i); + + for (i = 0; i != 63; i++) + KUNIT_ASSERT_EQ(test, log2_ffz_t(u64, BIT_ULL(i) - 1), i); + + for (i = 0; i != 32; i++) { + u64 val = get_random_u64(); + + KUNIT_ASSERT_EQ(test, + log2_mod_t(u32, val, log2_ffs_t(u32, val)), 0); + KUNIT_ASSERT_EQ(test, + log2_mod_t(u64, val, log2_ffs_t(u64, val)), 0); + + KUNIT_ASSERT_EQ(test, + log2_mod_t(u32, val, log2_ffz_t(u32, val)), + log2_to_max_int_t(u32, log2_ffz_t(u32, val))); + KUNIT_ASSERT_EQ(test, + log2_mod_t(u64, val, log2_ffz_t(u64, val)), + log2_to_max_int_t(u64, log2_ffz_t(u64, val))); + } +} + +static unsigned int ref_best_pgsize(pt_vaddr_t pgsz_bitmap, pt_vaddr_t va, + pt_vaddr_t last_va, pt_oaddr_t oa) +{ + pt_vaddr_t pgsz_lg2; + + /* Brute force the constraints described in pt_compute_best_pgsize() */ + for (pgsz_lg2 = PT_VADDR_MAX_LG2 - 1; pgsz_lg2 != 0; pgsz_lg2--) { + if ((pgsz_bitmap & log2_to_int(pgsz_lg2)) && + log2_mod(va, pgsz_lg2) == 0 && + oalog2_mod(oa, pgsz_lg2) == 0 && + va + log2_to_int(pgsz_lg2) - 1 <= last_va && + log2_div_eq(va, va + log2_to_int(pgsz_lg2) - 1, pgsz_lg2) && + oalog2_div_eq(oa, oa + log2_to_int(pgsz_lg2) - 1, pgsz_lg2)) + return pgsz_lg2; + } + return 0; +} + +/* Check that the bit logic in pt_compute_best_pgsize() works. */ +static void test_best_pgsize(struct kunit *test) +{ + unsigned int a_lg2; + unsigned int b_lg2; + unsigned int c_lg2; + + /* Try random prefixes with every suffix combination */ + for (a_lg2 = 1; a_lg2 != 10; a_lg2++) { + for (b_lg2 = 1; b_lg2 != 10; b_lg2++) { + for (c_lg2 = 1; c_lg2 != 10; c_lg2++) { + pt_vaddr_t pgsz_bitmap = get_random_u64(); + pt_vaddr_t va = get_random_u64() << a_lg2; + pt_oaddr_t oa = get_random_u64() << b_lg2; + pt_vaddr_t last_va = log2_set_mod_max( + get_random_u64(), c_lg2); + + if (va > last_va) + swap(va, last_va); + KUNIT_ASSERT_EQ( + test, + pt_compute_best_pgsize(pgsz_bitmap, va, + last_va, oa), + ref_best_pgsize(pgsz_bitmap, va, + last_va, oa)); + } + } + } + + /* 0 prefix, every suffix */ + for (c_lg2 = 1; c_lg2 != PT_VADDR_MAX_LG2 - 1; c_lg2++) { + pt_vaddr_t pgsz_bitmap = get_random_u64(); + pt_vaddr_t va = 0; + pt_oaddr_t oa = 0; + pt_vaddr_t last_va = log2_set_mod_max(0, c_lg2); + + KUNIT_ASSERT_EQ(test, + pt_compute_best_pgsize(pgsz_bitmap, va, last_va, + oa), + ref_best_pgsize(pgsz_bitmap, va, last_va, oa)); + } + + /* 1's prefix, every suffix */ + for (a_lg2 = 1; a_lg2 != 10; a_lg2++) { + for (b_lg2 = 1; b_lg2 != 10; b_lg2++) { + for (c_lg2 = 1; c_lg2 != 10; c_lg2++) { + pt_vaddr_t pgsz_bitmap = get_random_u64(); + pt_vaddr_t va = PT_VADDR_MAX << a_lg2; + pt_oaddr_t oa = PT_VADDR_MAX << b_lg2; + pt_vaddr_t last_va = PT_VADDR_MAX; + + KUNIT_ASSERT_EQ( + test, + pt_compute_best_pgsize(pgsz_bitmap, va, + last_va, oa), + ref_best_pgsize(pgsz_bitmap, va, + last_va, oa)); + } + } + } + + /* pgsize_bitmap is always 0 */ + for (a_lg2 = 1; a_lg2 != 10; a_lg2++) { + for (b_lg2 = 1; b_lg2 != 10; b_lg2++) { + for (c_lg2 = 1; c_lg2 != 10; c_lg2++) { + pt_vaddr_t pgsz_bitmap = 0; + pt_vaddr_t va = get_random_u64() << a_lg2; + pt_oaddr_t oa = get_random_u64() << b_lg2; + pt_vaddr_t last_va = log2_set_mod_max( + get_random_u64(), c_lg2); + + if (va > last_va) + swap(va, last_va); + KUNIT_ASSERT_EQ( + test, + pt_compute_best_pgsize(pgsz_bitmap, va, + last_va, oa), + 0); + } + } + } + + if (sizeof(pt_vaddr_t) <= 4) + return; + + /* over 32 bit page sizes */ + for (a_lg2 = 32; a_lg2 != 42; a_lg2++) { + for (b_lg2 = 32; b_lg2 != 42; b_lg2++) { + for (c_lg2 = 32; c_lg2 != 42; c_lg2++) { + pt_vaddr_t pgsz_bitmap = get_random_u64(); + pt_vaddr_t va = get_random_u64() << a_lg2; + pt_oaddr_t oa = get_random_u64() << b_lg2; + pt_vaddr_t last_va = log2_set_mod_max( + get_random_u64(), c_lg2); + + if (va > last_va) + swap(va, last_va); + KUNIT_ASSERT_EQ( + test, + pt_compute_best_pgsize(pgsz_bitmap, va, + last_va, oa), + ref_best_pgsize(pgsz_bitmap, va, + last_va, oa)); + } + } + } +} + +/* + * Check that pt_install_table() and pt_table_pa() match + */ +static void test_lvl_table_ptr(struct kunit *test, struct pt_state *pts, + void *arg) +{ + struct kunit_iommu_priv *priv = test->priv; + pt_oaddr_t paddr = + log2_set_mod(priv->test_oa, 0, priv->smallest_pgsz_lg2); + struct pt_write_attrs attrs = {}; + + if (!pt_can_have_table(pts)) + return; + + KUNIT_ASSERT_NO_ERRNO_FN(test, "pt_iommu_set_prot", + pt_iommu_set_prot(pts->range->common, &attrs, + IOMMU_READ)); + + pt_load_single_entry(pts); + KUNIT_ASSERT_PT_LOAD(test, pts, PT_ENTRY_EMPTY); + + KUNIT_ASSERT_TRUE(test, pt_install_table(pts, paddr, &attrs)); + + /* A second install should pass because install updates pts->entry. */ + KUNIT_ASSERT_EQ(test, pt_install_table(pts, paddr, &attrs), true); + + KUNIT_ASSERT_PT_LOAD(test, pts, PT_ENTRY_TABLE); + KUNIT_ASSERT_EQ(test, pt_table_pa(pts), paddr); + + pt_clear_entry(pts, ilog2(1)); + KUNIT_ASSERT_PT_LOAD(test, pts, PT_ENTRY_EMPTY); +} + +static void test_table_ptr(struct kunit *test) +{ + check_all_levels(test, test_lvl_table_ptr, NULL); +} + +struct lvl_radix_arg { + pt_vaddr_t vbits; +}; + +/* + * Check pt_table_oa_lg2sz() and pt_table_item_lg2sz() they need to decode a + * continuous list of VA across all the levels that covers the entire advertised + * VA space. + */ +static void test_lvl_radix(struct kunit *test, struct pt_state *pts, void *arg) +{ + unsigned int table_lg2sz = pt_table_oa_lg2sz(pts); + unsigned int isz_lg2 = pt_table_item_lg2sz(pts); + struct lvl_radix_arg *radix = arg; + + /* Every bit below us is decoded */ + KUNIT_ASSERT_EQ(test, log2_set_mod_max(0, isz_lg2), radix->vbits); + + /* We are not decoding bits someone else is */ + KUNIT_ASSERT_EQ(test, log2_div(radix->vbits, isz_lg2), 0); + + /* Can't decode past the pt_vaddr_t size */ + KUNIT_ASSERT_LE(test, table_lg2sz, PT_VADDR_MAX_LG2); + KUNIT_ASSERT_EQ(test, fvalog2_div(table_lg2sz, PT_MAX_VA_ADDRESS_LG2), + 0); + + radix->vbits = fvalog2_set_mod_max(0, table_lg2sz); +} + +static void test_max_va(struct kunit *test) +{ + struct kunit_iommu_priv *priv = test->priv; + struct pt_range range = pt_top_range(priv->common); + + KUNIT_ASSERT_GE(test, priv->common->max_vasz_lg2, range.max_vasz_lg2); +} + +static void test_table_radix(struct kunit *test) +{ + struct kunit_iommu_priv *priv = test->priv; + struct lvl_radix_arg radix = { .vbits = priv->smallest_pgsz - 1 }; + struct pt_range range; + + check_all_levels(test, test_lvl_radix, &radix); + + range = pt_top_range(priv->common); + if (range.max_vasz_lg2 == PT_VADDR_MAX_LG2) { + KUNIT_ASSERT_EQ(test, radix.vbits, PT_VADDR_MAX); + } else { + if (!IS_32BIT) + KUNIT_ASSERT_EQ(test, + log2_set_mod_max(0, range.max_vasz_lg2), + radix.vbits); + KUNIT_ASSERT_EQ(test, log2_div(radix.vbits, range.max_vasz_lg2), + 0); + } +} + +static unsigned int safe_pt_num_items_lg2(const struct pt_state *pts) +{ + struct pt_range top_range = pt_top_range(pts->range->common); + struct pt_state top_pts = pt_init_top(&top_range); + + /* + * Avoid calling pt_num_items_lg2() on the top, instead we can derive + * the size of the top table from the top range. + */ + if (pts->level == top_range.top_level) + return ilog2(pt_range_to_end_index(&top_pts)); + return pt_num_items_lg2(pts); +} + +static void test_lvl_possible_sizes(struct kunit *test, struct pt_state *pts, + void *arg) +{ + unsigned int num_items_lg2 = safe_pt_num_items_lg2(pts); + pt_vaddr_t pgsize_bitmap = pt_possible_sizes(pts); + unsigned int isz_lg2 = pt_table_item_lg2sz(pts); + + if (!pt_can_have_leaf(pts)) { + KUNIT_ASSERT_EQ(test, pgsize_bitmap, 0); + return; + } + + /* No bits for sizes that would be outside this table */ + KUNIT_ASSERT_EQ(test, log2_mod(pgsize_bitmap, isz_lg2), 0); + KUNIT_ASSERT_EQ( + test, fvalog2_div(pgsize_bitmap, num_items_lg2 + isz_lg2), 0); + + /* + * Non contiguous must be supported. AMDv1 has a HW bug where it does + * not support it on one of the levels. + */ + if ((u64)pgsize_bitmap != 0xff0000000000ULL || + strcmp(__stringify(PTPFX_RAW), "amdv1") != 0) + KUNIT_ASSERT_TRUE(test, pgsize_bitmap & log2_to_int(isz_lg2)); + else + KUNIT_ASSERT_NE(test, pgsize_bitmap, 0); + + /* A contiguous entry should not span the whole table */ + if (num_items_lg2 + isz_lg2 != PT_VADDR_MAX_LG2) + KUNIT_ASSERT_FALSE( + test, + pgsize_bitmap & log2_to_int(num_items_lg2 + isz_lg2)); +} + +static void test_entry_possible_sizes(struct kunit *test) +{ + check_all_levels(test, test_lvl_possible_sizes, NULL); +} + +static void sweep_all_pgsizes(struct kunit *test, struct pt_state *pts, + struct pt_write_attrs *attrs, + pt_oaddr_t test_oaddr) +{ + pt_vaddr_t pgsize_bitmap = pt_possible_sizes(pts); + unsigned int isz_lg2 = pt_table_item_lg2sz(pts); + unsigned int len_lg2; + + if (pts->index != 0) + return; + + for (len_lg2 = 0; len_lg2 < PT_VADDR_MAX_LG2 - 1; len_lg2++) { + struct pt_state sub_pts = *pts; + pt_oaddr_t oaddr; + + if (!(pgsize_bitmap & log2_to_int(len_lg2))) + continue; + + oaddr = log2_set_mod(test_oaddr, 0, len_lg2); + pt_install_leaf_entry(pts, oaddr, len_lg2, attrs); + /* Verify that every contiguous item translates correctly */ + for (sub_pts.index = 0; + sub_pts.index != log2_to_int(len_lg2 - isz_lg2); + sub_pts.index++) { + KUNIT_ASSERT_PT_LOAD(test, &sub_pts, PT_ENTRY_OA); + KUNIT_ASSERT_EQ(test, pt_item_oa(&sub_pts), + oaddr + sub_pts.index * + oalog2_mul(1, isz_lg2)); + KUNIT_ASSERT_EQ(test, pt_entry_oa(&sub_pts), oaddr); + KUNIT_ASSERT_EQ(test, pt_entry_num_contig_lg2(&sub_pts), + len_lg2 - isz_lg2); + } + + pt_clear_entry(pts, len_lg2 - isz_lg2); + KUNIT_ASSERT_PT_LOAD(test, pts, PT_ENTRY_EMPTY); + } +} + +/* + * Check that pt_install_leaf_entry() and pt_entry_oa() match. + * Check that pt_clear_entry() works. + */ +static void test_lvl_entry_oa(struct kunit *test, struct pt_state *pts, + void *arg) +{ + unsigned int max_oa_lg2 = pts->range->common->max_oasz_lg2; + struct kunit_iommu_priv *priv = test->priv; + struct pt_write_attrs attrs = {}; + + if (!pt_can_have_leaf(pts)) + return; + + KUNIT_ASSERT_NO_ERRNO_FN(test, "pt_iommu_set_prot", + pt_iommu_set_prot(pts->range->common, &attrs, + IOMMU_READ)); + + sweep_all_pgsizes(test, pts, &attrs, priv->test_oa); + + /* Check that the table can store the boundary OAs */ + sweep_all_pgsizes(test, pts, &attrs, 0); + if (max_oa_lg2 == PT_OADDR_MAX_LG2) + sweep_all_pgsizes(test, pts, &attrs, PT_OADDR_MAX); + else + sweep_all_pgsizes(test, pts, &attrs, + oalog2_to_max_int(max_oa_lg2)); +} + +static void test_entry_oa(struct kunit *test) +{ + check_all_levels(test, test_lvl_entry_oa, NULL); +} + +/* Test pt_attr_from_entry() */ +static void test_lvl_attr_from_entry(struct kunit *test, struct pt_state *pts, + void *arg) +{ + pt_vaddr_t pgsize_bitmap = pt_possible_sizes(pts); + unsigned int isz_lg2 = pt_table_item_lg2sz(pts); + struct kunit_iommu_priv *priv = test->priv; + unsigned int len_lg2; + unsigned int prot; + + if (!pt_can_have_leaf(pts)) + return; + + for (len_lg2 = 0; len_lg2 < PT_VADDR_MAX_LG2; len_lg2++) { + if (!(pgsize_bitmap & log2_to_int(len_lg2))) + continue; + for (prot = 0; prot <= (IOMMU_READ | IOMMU_WRITE | IOMMU_CACHE | + IOMMU_NOEXEC | IOMMU_MMIO); + prot++) { + pt_oaddr_t oaddr; + struct pt_write_attrs attrs = {}; + u64 good_entry; + + /* + * If the format doesn't support this combination of + * prot bits skip it + */ + if (pt_iommu_set_prot(pts->range->common, &attrs, + prot)) { + /* But RW has to be supported */ + KUNIT_ASSERT_NE(test, prot, + IOMMU_READ | IOMMU_WRITE); + continue; + } + + oaddr = log2_set_mod(priv->test_oa, 0, len_lg2); + pt_install_leaf_entry(pts, oaddr, len_lg2, &attrs); + KUNIT_ASSERT_PT_LOAD(test, pts, PT_ENTRY_OA); + + good_entry = pts->entry; + + memset(&attrs, 0, sizeof(attrs)); + pt_attr_from_entry(pts, &attrs); + + pt_clear_entry(pts, len_lg2 - isz_lg2); + KUNIT_ASSERT_PT_LOAD(test, pts, PT_ENTRY_EMPTY); + + pt_install_leaf_entry(pts, oaddr, len_lg2, &attrs); + KUNIT_ASSERT_PT_LOAD(test, pts, PT_ENTRY_OA); + + /* + * The descriptor produced by pt_attr_from_entry() + * produce an identical entry value when re-written + */ + KUNIT_ASSERT_EQ(test, good_entry, pts->entry); + + pt_clear_entry(pts, len_lg2 - isz_lg2); + } + } +} + +static void test_attr_from_entry(struct kunit *test) +{ + check_all_levels(test, test_lvl_attr_from_entry, NULL); +} + +static void test_lvl_dirty(struct kunit *test, struct pt_state *pts, void *arg) +{ + pt_vaddr_t pgsize_bitmap = pt_possible_sizes(pts); + unsigned int isz_lg2 = pt_table_item_lg2sz(pts); + struct kunit_iommu_priv *priv = test->priv; + unsigned int start_idx = pts->index; + struct pt_write_attrs attrs = {}; + unsigned int len_lg2; + + if (!pt_can_have_leaf(pts)) + return; + + KUNIT_ASSERT_NO_ERRNO_FN(test, "pt_iommu_set_prot", + pt_iommu_set_prot(pts->range->common, &attrs, + IOMMU_READ | IOMMU_WRITE)); + + for (len_lg2 = 0; len_lg2 < PT_VADDR_MAX_LG2; len_lg2++) { + pt_oaddr_t oaddr; + unsigned int i; + + if (!(pgsize_bitmap & log2_to_int(len_lg2))) + continue; + + oaddr = log2_set_mod(priv->test_oa, 0, len_lg2); + pt_install_leaf_entry(pts, oaddr, len_lg2, &attrs); + KUNIT_ASSERT_PT_LOAD(test, pts, PT_ENTRY_OA); + + pt_load_entry(pts); + pt_entry_set_write_clean(pts); + pt_load_entry(pts); + KUNIT_ASSERT_FALSE(test, pt_entry_write_is_dirty(pts)); + + for (i = 0; i != log2_to_int(len_lg2 - isz_lg2); i++) { + /* dirty every contiguous entry */ + pts->index = start_idx + i; + pt_load_entry(pts); + KUNIT_ASSERT_TRUE(test, pt_entry_make_write_dirty(pts)); + pts->index = start_idx; + pt_load_entry(pts); + KUNIT_ASSERT_TRUE(test, pt_entry_write_is_dirty(pts)); + + pt_entry_set_write_clean(pts); + pt_load_entry(pts); + KUNIT_ASSERT_FALSE(test, pt_entry_write_is_dirty(pts)); + } + + pt_clear_entry(pts, len_lg2 - isz_lg2); + } +} + +static __maybe_unused void test_dirty(struct kunit *test) +{ + struct kunit_iommu_priv *priv = test->priv; + + if (!pt_dirty_supported(priv->common)) + kunit_skip(test, + "Page table features do not support dirty tracking"); + + check_all_levels(test, test_lvl_dirty, NULL); +} + +static struct kunit_case generic_pt_test_cases[] = { + KUNIT_CASE_FMT(test_init), + KUNIT_CASE_FMT(test_bitops), + KUNIT_CASE_FMT(test_best_pgsize), + KUNIT_CASE_FMT(test_table_ptr), + KUNIT_CASE_FMT(test_max_va), + KUNIT_CASE_FMT(test_table_radix), + KUNIT_CASE_FMT(test_entry_possible_sizes), + KUNIT_CASE_FMT(test_entry_oa), + KUNIT_CASE_FMT(test_attr_from_entry), +#ifdef pt_entry_write_is_dirty + KUNIT_CASE_FMT(test_dirty), +#endif + {}, +}; + +static int pt_kunit_generic_pt_init(struct kunit *test) +{ + struct kunit_iommu_priv *priv; + int ret; + + priv = kunit_kzalloc(test, sizeof(*priv), GFP_KERNEL); + if (!priv) + return -ENOMEM; + ret = pt_kunit_priv_init(test, priv); + if (ret) { + kunit_kfree(test, priv); + return ret; + } + test->priv = priv; + return 0; +} + +static void pt_kunit_generic_pt_exit(struct kunit *test) +{ + struct kunit_iommu_priv *priv = test->priv; + + if (!test->priv) + return; + + pt_iommu_deinit(priv->iommu); + kunit_kfree(test, test->priv); +} + +static struct kunit_suite NS(generic_pt_suite) = { + .name = __stringify(NS(fmt_test)), + .init = pt_kunit_generic_pt_init, + .exit = pt_kunit_generic_pt_exit, + .test_cases = generic_pt_test_cases, +}; +kunit_test_suites(&NS(generic_pt_suite)); diff --git a/drivers/iommu/generic_pt/kunit_iommu.h b/drivers/iommu/generic_pt/kunit_iommu.h new file mode 100644 index 00000000000000..8a53b1d772ca9d --- /dev/null +++ b/drivers/iommu/generic_pt/kunit_iommu.h @@ -0,0 +1,181 @@ +/* SPDX-License-Identifier: GPL-2.0-only */ +/* + * Copyright (c) 2024-2025, NVIDIA CORPORATION & AFFILIATES + */ +#ifndef __GENERIC_PT_KUNIT_IOMMU_H +#define __GENERIC_PT_KUNIT_IOMMU_H + +#define GENERIC_PT_KUNIT 1 +#include <kunit/device.h> +#include <kunit/test.h> +#include <../../iommu-pages.h> +#include "pt_iter.h" + +#define pt_iommu_table_cfg CONCATENATE(pt_iommu_table, _cfg) +#define pt_iommu_init CONCATENATE(CONCATENATE(pt_iommu_, PTPFX), init) +int pt_iommu_init(struct pt_iommu_table *fmt_table, + const struct pt_iommu_table_cfg *cfg, gfp_t gfp); + +/* The format can provide a list of configurations it would like to test */ +#ifdef kunit_fmt_cfgs +static const void *kunit_pt_gen_params_cfg(const void *prev, char *desc) +{ + uintptr_t cfg_id = (uintptr_t)prev; + + cfg_id++; + if (cfg_id >= ARRAY_SIZE(kunit_fmt_cfgs) + 1) + return NULL; + snprintf(desc, KUNIT_PARAM_DESC_SIZE, "%s_cfg_%u", + __stringify(PTPFX_RAW), (unsigned int)(cfg_id - 1)); + return (void *)cfg_id; +} +#define KUNIT_CASE_FMT(test_name) \ + KUNIT_CASE_PARAM(test_name, kunit_pt_gen_params_cfg) +#else +#define KUNIT_CASE_FMT(test_name) KUNIT_CASE(test_name) +#endif + +#define KUNIT_ASSERT_NO_ERRNO(test, ret) \ + KUNIT_ASSERT_EQ_MSG(test, ret, 0, KUNIT_SUBSUBTEST_INDENT "errno %pe", \ + ERR_PTR(ret)) + +#define KUNIT_ASSERT_NO_ERRNO_FN(test, fn, ret) \ + KUNIT_ASSERT_EQ_MSG(test, ret, 0, \ + KUNIT_SUBSUBTEST_INDENT "errno %pe from %s", \ + ERR_PTR(ret), fn) + +/* + * When the test is run on a 32 bit system unsigned long can be 32 bits. This + * cause the iommu op signatures to be restricted to 32 bits. Meaning the test + * has to be mindful not to create any VA's over the 32 bit limit. Reduce the + * scope of the testing as the main purpose of checking on full 32 bit is to + * look for 32bitism in the core code. Run the test on i386 with X86_PAE=y to + * get the full coverage when dma_addr_t & phys_addr_t are 8 bytes + */ +#define IS_32BIT (sizeof(unsigned long) == 4) + +struct kunit_iommu_priv { + union { + struct iommu_domain domain; + struct pt_iommu_table fmt_table; + }; + spinlock_t top_lock; + struct device *dummy_dev; + struct pt_iommu *iommu; + struct pt_common *common; + struct pt_iommu_table_cfg cfg; + struct pt_iommu_info info; + unsigned int smallest_pgsz_lg2; + pt_vaddr_t smallest_pgsz; + unsigned int largest_pgsz_lg2; + pt_oaddr_t test_oa; + pt_vaddr_t safe_pgsize_bitmap; +}; +PT_IOMMU_CHECK_DOMAIN(struct kunit_iommu_priv, fmt_table.iommu, domain); + +static void pt_kunit_iotlb_sync(struct iommu_domain *domain, + struct iommu_iotlb_gather *gather) +{ + iommu_put_pages_list(&gather->freelist); +} + +#define IOMMU_PT_DOMAIN_OPS1(x) IOMMU_PT_DOMAIN_OPS(x) +static const struct iommu_domain_ops kunit_pt_ops = { + IOMMU_PT_DOMAIN_OPS1(PTPFX_RAW), + .iotlb_sync = &pt_kunit_iotlb_sync, +}; + +static void pt_kunit_change_top(struct pt_iommu *iommu_table, + phys_addr_t top_paddr, unsigned int top_level) +{ +} + +static spinlock_t *pt_kunit_get_top_lock(struct pt_iommu *iommu_table) +{ + struct kunit_iommu_priv *priv = container_of( + iommu_table, struct kunit_iommu_priv, fmt_table.iommu); + + return &priv->top_lock; +} + +static const struct pt_iommu_flush_ops pt_kunit_flush_ops = { + .change_top = &pt_kunit_change_top, + .get_top_lock = &pt_kunit_get_top_lock, +}; + +static int pt_kunit_priv_init(struct kunit *test, struct kunit_iommu_priv *priv) +{ + unsigned int va_lg2sz; + int ret; + + /* Enough so the memory allocator works */ + priv->dummy_dev = kunit_device_register(test, "pt_kunit_dev"); + if (IS_ERR(priv->dummy_dev)) + return PTR_ERR(priv->dummy_dev); + set_dev_node(priv->dummy_dev, NUMA_NO_NODE); + + spin_lock_init(&priv->top_lock); + +#ifdef kunit_fmt_cfgs + priv->cfg = kunit_fmt_cfgs[((uintptr_t)test->param_value) - 1]; + /* + * The format can set a list of features that the kunit_fmt_cfgs + * controls, other features are default to on. + */ + priv->cfg.common.features |= PT_SUPPORTED_FEATURES & + (~KUNIT_FMT_FEATURES); +#else + priv->cfg.common.features = PT_SUPPORTED_FEATURES; +#endif + + /* Defaults, for the kunit */ + if (!priv->cfg.common.hw_max_vasz_lg2) + priv->cfg.common.hw_max_vasz_lg2 = PT_MAX_VA_ADDRESS_LG2; + if (!priv->cfg.common.hw_max_oasz_lg2) + priv->cfg.common.hw_max_oasz_lg2 = + pt_max_output_address_lg2(NULL); + + priv->fmt_table.iommu.nid = NUMA_NO_NODE; + priv->fmt_table.iommu.hw_flush_ops = &pt_kunit_flush_ops; + priv->domain.ops = &kunit_pt_ops; + ret = pt_iommu_init(&priv->fmt_table, &priv->cfg, GFP_KERNEL); + if (ret) { + if (ret == -EOVERFLOW) + kunit_skip( + test, + "This configuration cannot be tested on 32 bit"); + return ret; + } + + priv->iommu = &priv->fmt_table.iommu; + priv->common = common_from_iommu(&priv->fmt_table.iommu); + priv->iommu->ops->get_info(priv->iommu, &priv->info); + + /* + * size_t is used to pass the mapping length, it can be 32 bit, truncate + * the pagesizes so we don't use large sizes. + */ + priv->info.pgsize_bitmap = (size_t)priv->info.pgsize_bitmap; + + priv->smallest_pgsz_lg2 = log2_ffs(priv->info.pgsize_bitmap); + priv->smallest_pgsz = log2_to_int(priv->smallest_pgsz_lg2); + priv->largest_pgsz_lg2 = + log2_fls((dma_addr_t)priv->info.pgsize_bitmap) - 1; + + priv->test_oa = + oalog2_mod(0x74a71445deadbeef, priv->common->max_oasz_lg2); + + /* + * We run out of VA space if the mappings get too big, make something + * smaller that can safely pass through dma_addr_t API. + */ + va_lg2sz = priv->common->max_vasz_lg2; + if (IS_32BIT && va_lg2sz > 32) + va_lg2sz = 32; + priv->safe_pgsize_bitmap = + log2_mod(priv->info.pgsize_bitmap, va_lg2sz - 1); + + return 0; +} + +#endif

This is used by x86 CPUs and can be used in AMD/VT-D x86 IOMMUs. When a x86 IOMMU is running SVA it is using this page table format.

This implementation follows the AMD v2 io-pgtable version.

There is nothing remarkable here, the format can have 4 or 5 levels and limited support for different page sizes. No contiguous pages support.

x86 uses a sign extension mechanism where the top bits of the VA must match the sign bit. The core code supports this through PT_FEAT_SIGN_EXTEND which creates and upper and lower VA range. All the new operations will work correctly in both spaces, however currently there is no way to report the upper space to other layers. Future patches can improve that.

In principle this can support 3 page tables levels matching the 32 bit PAE table format, but no iommu driver needs this. The focus is on the modern 64 bit 4 and 5 level formats.

Comparing the performance of several operations to the existing version:

iommu_map() pgsz ,avg new,old ns, min new,old ns , min % (+ve is better) 2^12, 71,61 , 66,58 , -13.13 2^21, 66,60 , 61,55 , -10.10 2^30, 59,56 , 56,54 , -3.03 256*2^12, 392,1360 , 345,1289 , 73.73 256*2^21, 383,1159 , 335,1145 , 70.70 256*2^30, 378,965 , 331,892 , 62.62

iommu_unmap() pgsz ,avg new,old ns, min new,old ns , min % (+ve is better) 2^12, 77,71 , 73,68 , -7.07 2^21, 76,70 , 70,66 , -6.06 2^30, 69,66 , 66,63 , -4.04 256*2^12, 225,899 , 210,870 , 75.75 256*2^21, 262,722 , 248,710 , 65.65 256*2^30, 251,643 , 244,634 , 61.61

The small -ve values in the iommu_unmap() are due to the core code calling iommu_pgsize() before invoking the domain op. This is unncessary with this implementation. Future work optimizes this and gets to 2%, 4%, 3%.

Tested-by: Alejandro Jimenez alejandro.j.jimenez@oracle.com Signed-off-by: Jason Gunthorpe jgg@nvidia.com --- drivers/iommu/generic_pt/.kunitconfig | 1 + drivers/iommu/generic_pt/Kconfig | 11 + drivers/iommu/generic_pt/fmt/Makefile | 2 + drivers/iommu/generic_pt/fmt/defs_x86_64.h | 21 ++ drivers/iommu/generic_pt/fmt/iommu_x86_64.c | 11 + drivers/iommu/generic_pt/fmt/x86_64.h | 248 ++++++++++++++++++++ include/linux/generic_pt/common.h | 13 + include/linux/generic_pt/iommu.h | 11 + 8 files changed, 318 insertions(+) create mode 100644 drivers/iommu/generic_pt/fmt/defs_x86_64.h create mode 100644 drivers/iommu/generic_pt/fmt/iommu_x86_64.c create mode 100644 drivers/iommu/generic_pt/fmt/x86_64.h

diff --git a/drivers/iommu/generic_pt/.kunitconfig b/drivers/iommu/generic_pt/.kunitconfig index 936c327f0661cf..2016c5e5ac0fe9 100644 --- a/drivers/iommu/generic_pt/.kunitconfig +++ b/drivers/iommu/generic_pt/.kunitconfig @@ -3,6 +3,7 @@ CONFIG_GENERIC_PT=y CONFIG_DEBUG_GENERIC_PT=y CONFIG_IOMMU_PT=y CONFIG_IOMMU_PT_AMDV1=y +CONFIG_IOMMU_PT_X86_64=y CONFIG_IOMMU_PT_KUNIT_TEST=y

CONFIG_IOMMUFD=y diff --git a/drivers/iommu/generic_pt/Kconfig b/drivers/iommu/generic_pt/Kconfig index a117cc23a1c328..5a93df987e8728 100644 --- a/drivers/iommu/generic_pt/Kconfig +++ b/drivers/iommu/generic_pt/Kconfig @@ -41,10 +41,21 @@ config IOMMU_PT_AMDV1

Selected automatically by an IOMMU driver that uses this format.

+config IOMMU_PT_X86_64 + tristate "IOMMU page table for x86 64-bit, 4/5 levels" + depends on !GENERIC_ATOMIC64 # for cmpxchg64 + help + iommu_domain implementation for the x86 64-bit 4/5 level page table. + It supports 4K/2M/1G page sizes and can decode a sign-extended + portion of the 64-bit IOVA space. + + Selected automatically by an IOMMU driver that uses this format. + config IOMMU_PT_KUNIT_TEST tristate "IOMMU Page Table KUnit Test" if !KUNIT_ALL_TESTS depends on KUNIT depends on IOMMU_PT_AMDV1 || !IOMMU_PT_AMDV1 + depends on IOMMU_PT_X86_64 || !IOMMU_PT_X86_64 default KUNIT_ALL_TESTS help Enable kunit tests for GENERIC_PT and IOMMU_PT that covers all the diff --git a/drivers/iommu/generic_pt/fmt/Makefile b/drivers/iommu/generic_pt/fmt/Makefile index f0c22cf5f7bee6..5a3379107999f5 100644 --- a/drivers/iommu/generic_pt/fmt/Makefile +++ b/drivers/iommu/generic_pt/fmt/Makefile @@ -3,6 +3,8 @@ iommu_pt_fmt-$(CONFIG_IOMMU_PT_AMDV1) += amdv1 iommu_pt_fmt-$(CONFIG_IOMMUFD_TEST) += mock

+iommu_pt_fmt-$(CONFIG_IOMMU_PT_X86_64) += x86_64 + IOMMU_PT_KUNIT_TEST := define create_format obj-$(2) += iommu_$(1).o diff --git a/drivers/iommu/generic_pt/fmt/defs_x86_64.h b/drivers/iommu/generic_pt/fmt/defs_x86_64.h new file mode 100644 index 00000000000000..6f589e1f55d35a --- /dev/null +++ b/drivers/iommu/generic_pt/fmt/defs_x86_64.h @@ -0,0 +1,21 @@ +/* SPDX-License-Identifier: GPL-2.0-only */ +/* + * Copyright (c) 2024-2025, NVIDIA CORPORATION & AFFILIATES + * + */ +#ifndef __GENERIC_PT_FMT_DEFS_X86_64_H +#define __GENERIC_PT_FMT_DEFS_X86_64_H + +#include <linux/generic_pt/common.h> +#include <linux/types.h> + +typedef u64 pt_vaddr_t; +typedef u64 pt_oaddr_t; + +struct x86_64_pt_write_attrs { + u64 descriptor_bits; + gfp_t gfp; +}; +#define pt_write_attrs x86_64_pt_write_attrs + +#endif diff --git a/drivers/iommu/generic_pt/fmt/iommu_x86_64.c b/drivers/iommu/generic_pt/fmt/iommu_x86_64.c new file mode 100644 index 00000000000000..5c5960d871a32f --- /dev/null +++ b/drivers/iommu/generic_pt/fmt/iommu_x86_64.c @@ -0,0 +1,11 @@ +// SPDX-License-Identifier: GPL-2.0-only +/* + * Copyright (c) 2024-2025, NVIDIA CORPORATION & AFFILIATES + */ +#define PT_FMT x86_64 +#define PT_SUPPORTED_FEATURES \ + (BIT(PT_FEAT_SIGN_EXTEND) | BIT(PT_FEAT_FLUSH_RANGE) | \ + BIT(PT_FEAT_FLUSH_RANGE_NO_GAPS) | \ + BIT(PT_FEAT_X86_64_AMD_ENCRYPT_TABLES)) + +#include "iommu_template.h" diff --git a/drivers/iommu/generic_pt/fmt/x86_64.h b/drivers/iommu/generic_pt/fmt/x86_64.h new file mode 100644 index 00000000000000..c01815b6229cce --- /dev/null +++ b/drivers/iommu/generic_pt/fmt/x86_64.h @@ -0,0 +1,248 @@ +/* SPDX-License-Identifier: GPL-2.0-only */ +/* + * Copyright (c) 2024-2025, NVIDIA CORPORATION & AFFILIATES + * + * x86 page table. Supports the 4 and 5 level variations. + * + * The 4 and 5 level version is described in: + * Section "4.4 4-Level Paging and 5-Level Paging" of the Intel Software + * Developer's Manual Volume 3 + * + * Section "9.7 First-Stage Paging Entries" of the "Intel Virtualization + * Technology for Directed I/O Architecture Specification" + * + * Section "2.2.6 I/O Page Tables for Guest Translations" of the "AMD I/O + * Virtualization Technology (IOMMU) Specification" + * + * It is used by x86 CPUs, AMD and VT-D IOMMU HW. + * + * Note the 3 level format is very similar and almost implemented here. The + * reserved/ignored layout is different and there are functional bit + * differences. + * + * This format uses PT_FEAT_SIGN_EXTEND to have a upper/non-canonical/lower + * split. PT_FEAT_SIGN_EXTEND is optional as AMD IOMMU sometimes uses non-sign + * extended addressing with this page table format. + * + * The named levels in the spec map to the pts->level as: + * Table/PTE - 0 + * Directory/PDE - 1 + * Directory Ptr/PDPTE - 2 + * PML4/PML4E - 3 + * PML5/PML5E - 4 + */ +#ifndef __GENERIC_PT_FMT_X86_64_H +#define __GENERIC_PT_FMT_X86_64_H + +#include "defs_x86_64.h" +#include "../pt_defs.h" + +#include <linux/bitfield.h> +#include <linux/container_of.h> +#include <linux/log2.h> +#include <linux/mem_encrypt.h> + +enum { + PT_MAX_OUTPUT_ADDRESS_LG2 = 52, + PT_MAX_VA_ADDRESS_LG2 = 57, + PT_ITEM_WORD_SIZE = sizeof(u64), + PT_MAX_TOP_LEVEL = 4, + PT_GRANULE_LG2SZ = 12, + PT_TABLEMEM_LG2SZ = 12, + + /* + * For AMD the GCR3 Base only has these bits. For VT-D FSPTPTR is 4k + * aligned and is limited by the architected HAW + */ + PT_TOP_PHYS_MASK = GENMASK_ULL(51, 12), +}; + +/* Shared descriptor bits */ +enum { + X86_64_FMT_P = BIT(0), + X86_64_FMT_RW = BIT(1), + X86_64_FMT_U = BIT(2), + X86_64_FMT_A = BIT(5), + X86_64_FMT_D = BIT(6), + X86_64_FMT_OA = GENMASK_ULL(51, 12), + X86_64_FMT_XD = BIT_ULL(63), +}; + +/* PDPTE/PDE */ +enum { + X86_64_FMT_PS = BIT(7), +}; + +#define common_to_x86_64_pt(common_ptr) \ + container_of_const(common_ptr, struct pt_x86_64, common) +#define to_x86_64_pt(pts) common_to_x86_64_pt((pts)->range->common) + +static inline pt_oaddr_t x86_64_pt_table_pa(const struct pt_state *pts) +{ + return oalog2_mul(FIELD_GET(X86_64_FMT_OA, pts->entry), + PT_TABLEMEM_LG2SZ); +} +#define pt_table_pa x86_64_pt_table_pa + +static inline pt_oaddr_t x86_64_pt_entry_oa(const struct pt_state *pts) +{ + return oalog2_mul(FIELD_GET(X86_64_FMT_OA, pts->entry), + PT_GRANULE_LG2SZ); +} +#define pt_entry_oa x86_64_pt_entry_oa + +static inline bool x86_64_pt_can_have_leaf(const struct pt_state *pts) +{ + return pts->level <= 2; +} +#define pt_can_have_leaf x86_64_pt_can_have_leaf + +static inline unsigned int x86_64_pt_num_items_lg2(const struct pt_state *pts) +{ + return PT_TABLEMEM_LG2SZ - ilog2(sizeof(u64)); +} +#define pt_num_items_lg2 x86_64_pt_num_items_lg2 + +static inline enum pt_entry_type x86_64_pt_load_entry_raw(struct pt_state *pts) +{ + const u64 *tablep = pt_cur_table(pts, u64); + u64 entry; + + pts->entry = entry = READ_ONCE(tablep[pts->index]); + if (!(entry & X86_64_FMT_P)) + return PT_ENTRY_EMPTY; + if (pts->level == 0 || + (x86_64_pt_can_have_leaf(pts) && (entry & X86_64_FMT_PS))) + return PT_ENTRY_OA; + return PT_ENTRY_TABLE; +} +#define pt_load_entry_raw x86_64_pt_load_entry_raw + +static inline void +x86_64_pt_install_leaf_entry(struct pt_state *pts, pt_oaddr_t oa, + unsigned int oasz_lg2, + const struct pt_write_attrs *attrs) +{ + u64 *tablep = pt_cur_table(pts, u64); + u64 entry; + + entry = X86_64_FMT_P | + FIELD_PREP(X86_64_FMT_OA, log2_div(oa, PT_GRANULE_LG2SZ)) | + attrs->descriptor_bits; + if (pts->level != 0) + entry |= X86_64_FMT_PS; + + WRITE_ONCE(tablep[pts->index], entry); + pts->entry = entry; +} +#define pt_install_leaf_entry x86_64_pt_install_leaf_entry + +static inline bool x86_64_pt_install_table(struct pt_state *pts, + pt_oaddr_t table_pa, + const struct pt_write_attrs *attrs) +{ + u64 entry; + + entry = X86_64_FMT_P | X86_64_FMT_RW | X86_64_FMT_U | X86_64_FMT_A | + FIELD_PREP(X86_64_FMT_OA, log2_div(table_pa, PT_GRANULE_LG2SZ)); + if (pts_feature(pts, PT_FEAT_X86_64_AMD_ENCRYPT_TABLES)) + entry = __sme_set(entry); + return pt_table_install64(pts, entry); +} +#define pt_install_table x86_64_pt_install_table + +static inline void x86_64_pt_attr_from_entry(const struct pt_state *pts, + struct pt_write_attrs *attrs) +{ + attrs->descriptor_bits = pts->entry & + (X86_64_FMT_RW | X86_64_FMT_U | X86_64_FMT_A | + X86_64_FMT_D | X86_64_FMT_XD); +} +#define pt_attr_from_entry x86_64_pt_attr_from_entry + +/* --- iommu */ +#include <linux/generic_pt/iommu.h> +#include <linux/iommu.h> + +#define pt_iommu_table pt_iommu_x86_64 + +/* The common struct is in the per-format common struct */ +static inline struct pt_common *common_from_iommu(struct pt_iommu *iommu_table) +{ + return &container_of(iommu_table, struct pt_iommu_table, iommu) + ->x86_64_pt.common; +} + +static inline struct pt_iommu *iommu_from_common(struct pt_common *common) +{ + return &container_of(common, struct pt_iommu_table, x86_64_pt.common) + ->iommu; +} + +static inline int x86_64_pt_iommu_set_prot(struct pt_common *common, + struct pt_write_attrs *attrs, + unsigned int iommu_prot) +{ + u64 pte; + + pte = X86_64_FMT_U | X86_64_FMT_A | X86_64_FMT_D; + if (iommu_prot & IOMMU_WRITE) + pte |= X86_64_FMT_RW; + + /* + * Ideally we'd have an IOMMU_ENCRYPTED flag set by higher levels to + * control this. For now if the tables use sme_set then so do the ptes. + */ + if (pt_feature(common, PT_FEAT_X86_64_AMD_ENCRYPT_TABLES)) + pte = __sme_set(pte); + + attrs->descriptor_bits = pte; + return 0; +} +#define pt_iommu_set_prot x86_64_pt_iommu_set_prot + +static inline int +x86_64_pt_iommu_fmt_init(struct pt_iommu_x86_64 *iommu_table, + const struct pt_iommu_x86_64_cfg *cfg) +{ + struct pt_x86_64 *table = &iommu_table->x86_64_pt; + + if (cfg->common.hw_max_vasz_lg2 < 31 || + cfg->common.hw_max_vasz_lg2 > 57) + return -EINVAL; + + /* Top of 2, 3, 4 */ + pt_top_set_level(&table->common, + (cfg->common.hw_max_vasz_lg2 - 31) / 9 + 2); + + table->common.max_oasz_lg2 = + min(PT_MAX_OUTPUT_ADDRESS_LG2, cfg->common.hw_max_oasz_lg2); + return 0; +} +#define pt_iommu_fmt_init x86_64_pt_iommu_fmt_init + +static inline void +x86_64_pt_iommu_fmt_hw_info(struct pt_iommu_x86_64 *table, + const struct pt_range *top_range, + struct pt_iommu_x86_64_hw_info *info) +{ + info->gcr3_pt = virt_to_phys(top_range->top_table); + PT_WARN_ON(info->gcr3_pt & ~PT_TOP_PHYS_MASK); + info->levels = top_range->top_level + 1; +} +#define pt_iommu_fmt_hw_info x86_64_pt_iommu_fmt_hw_info + +#if defined(GENERIC_PT_KUNIT) +static const struct pt_iommu_x86_64_cfg x86_64_kunit_fmt_cfgs[] = { + [0] = { .common.features = BIT(PT_FEAT_SIGN_EXTEND), + .common.hw_max_vasz_lg2 = 48 }, + [1] = { .common.features = BIT(PT_FEAT_SIGN_EXTEND), + .common.hw_max_vasz_lg2 = 57 }, + /* AMD IOMMU PASID 0 formats with no SIGN_EXTEND */ + [2] = { .common.hw_max_vasz_lg2 = 47 }, + [3] = { .common.hw_max_vasz_lg2 = 56 }, +}; +#define kunit_fmt_cfgs x86_64_kunit_fmt_cfgs +enum { KUNIT_FMT_FEATURES = BIT(PT_FEAT_SIGN_EXTEND)}; +#endif +#endif diff --git a/include/linux/generic_pt/common.h b/include/linux/generic_pt/common.h index ff4da1ae95087b..1cad7b16013a60 100644 --- a/include/linux/generic_pt/common.h +++ b/include/linux/generic_pt/common.h @@ -150,4 +150,17 @@ enum { PT_FEAT_AMDV1_FORCE_COHERENCE, };

+struct pt_x86_64 { + struct pt_common common; +}; + +enum { + /* + * The memory backing the tables is encrypted. Use __sme_set() to adjust + * the page table pointers in the tree. This only works with + * CONFIG_AMD_MEM_ENCRYPT. + */ + PT_FEAT_X86_64_AMD_ENCRYPT_TABLES = PT_FEAT_FMT_START, +}; + #endif diff --git a/include/linux/generic_pt/iommu.h b/include/linux/generic_pt/iommu.h index 8b48e374d983f9..3e0bc7938d954d 100644 --- a/include/linux/generic_pt/iommu.h +++ b/include/linux/generic_pt/iommu.h @@ -254,6 +254,17 @@ IOMMU_FORMAT(amdv1, amdpt); struct pt_iommu_amdv1_mock_hw_info; IOMMU_PROTOTYPES(amdv1_mock);

+struct pt_iommu_x86_64_cfg { + struct pt_iommu_cfg common; +}; + +struct pt_iommu_x86_64_hw_info { + u64 gcr3_pt; + u8 levels; +}; + +IOMMU_FORMAT(x86_64, x86_64_pt); + #undef IOMMU_PROTOTYPES #undef IOMMU_FORMAT #endif

From: Alejandro Jimenez alejandro.j.jimenez@oracle.com

Replace the io_pgtable versions with pt_iommu versions. The v2 page table uses the x86 implementation that will be eventually shared with VT-D.

This supports the same special features as the original code: - increase_top for the v1 format to allow scaling from 3 to 6 levels - non-present flushing - Dirty tracking for v1 only - __sme_set() to adjust the PTEs for CC - Optimization for flushing with virtualization to minimize the range - amd_iommu_pgsize_bitmap override of the native page sizes - page tables allocate from the device's NUMA node

Rework the domain ops so that v1/v2 get their own ops. Make dedicated allocation functions for v1 and v2. Hook up invalidation for a top change to struct pt_iommu_flush_ops. Delete some of the iopgtable related code that becomes unused in this patch. The next patch will delete the rest of it.

This fixes a race bug in AMD's increase_address_space() implementation. It stores the top level and top pointer in different memory, which prevents other threads from reading a coherent version:

increase_address_space() alloc_pte() level = pgtable->mode - 1; pgtable->root = pte; pgtable->mode += 1; pte = &pgtable->root[PM_LEVEL_INDEX(level, address)];

The iommupt version is careful to put mode and root under a single READ_ONCE and then is careful to only READ_ONCE a single time per walk.

Signed-off-by: Alejandro Jimenez alejandro.j.jimenez@oracle.com Tested-by: Alejandro Jimenez alejandro.j.jimenez@oracle.com Signed-off-by: Jason Gunthorpe jgg@nvidia.com --- drivers/iommu/amd/Kconfig | 5 +- drivers/iommu/amd/amd_iommu.h | 1 - drivers/iommu/amd/amd_iommu_types.h | 12 +- drivers/iommu/amd/io_pgtable.c | 1 - drivers/iommu/amd/iommu.c | 538 ++++++++++++++-------------- 5 files changed, 282 insertions(+), 275 deletions(-)

diff --git a/drivers/iommu/amd/Kconfig b/drivers/iommu/amd/Kconfig index ecef69c11144db..f2acf471cb5d9f 100644 --- a/drivers/iommu/amd/Kconfig +++ b/drivers/iommu/amd/Kconfig @@ -11,10 +11,13 @@ config AMD_IOMMU select MMU_NOTIFIER select IOMMU_API select IOMMU_IOVA - select IOMMU_IO_PGTABLE select IOMMU_SVA select IOMMU_IOPF select IOMMUFD_DRIVER if IOMMUFD + select GENERIC_PT + select IOMMU_PT + select IOMMU_PT_AMDV1 + select IOMMU_PT_X86_64 depends on X86_64 && PCI && ACPI && HAVE_CMPXCHG_DOUBLE help With this option you can enable support for AMD IOMMU hardware in diff --git a/drivers/iommu/amd/amd_iommu.h b/drivers/iommu/amd/amd_iommu.h index 9b4b589a54b57e..25044d28f28a8d 100644 --- a/drivers/iommu/amd/amd_iommu.h +++ b/drivers/iommu/amd/amd_iommu.h @@ -88,7 +88,6 @@ int amd_iommu_complete_ppr(struct device *dev, u32 pasid, int status, int tag); * the IOMMU used by this driver. */ void amd_iommu_flush_all_caches(struct amd_iommu *iommu); -void amd_iommu_update_and_flush_device_table(struct protection_domain *domain); void amd_iommu_domain_flush_pages(struct protection_domain *domain, u64 address, size_t size); void amd_iommu_dev_flush_pasid_pages(struct iommu_dev_data *dev_data, diff --git a/drivers/iommu/amd/amd_iommu_types.h b/drivers/iommu/amd/amd_iommu_types.h index 5219d7ddfdaa8b..90cf2bc4e14528 100644 --- a/drivers/iommu/amd/amd_iommu_types.h +++ b/drivers/iommu/amd/amd_iommu_types.h @@ -19,6 +19,7 @@ #include <linux/pci.h> #include <linux/irqreturn.h> #include <linux/io-pgtable.h> +#include <linux/generic_pt/iommu.h>

/* * Maximum number of IOMMUs supported @@ -588,9 +589,13 @@ struct pdom_iommu_info { * independent of their use. */ struct protection_domain { + union { + struct iommu_domain domain; + struct pt_iommu iommu; + struct pt_iommu_amdv1 amdv1; + struct pt_iommu_x86_64 amdv2; + }; struct list_head dev_list; /* List of all devices in this domain */ - struct iommu_domain domain; /* generic domain handle used by - iommu core code */ struct amd_io_pgtable iop; spinlock_t lock; /* mostly used to lock the page table*/ u16 id; /* the domain id written to the device table */ @@ -601,6 +606,9 @@ struct protection_domain { struct mmu_notifier mn; /* mmu notifier for the SVA domain */ struct list_head dev_data_list; /* List of pdom_dev_data */ }; +PT_IOMMU_CHECK_DOMAIN(struct protection_domain, iommu, domain); +PT_IOMMU_CHECK_DOMAIN(struct protection_domain, amdv1.iommu, domain); +PT_IOMMU_CHECK_DOMAIN(struct protection_domain, amdv2.iommu, domain);

/* * This structure contains information about one PCI segment in the system. diff --git a/drivers/iommu/amd/io_pgtable.c b/drivers/iommu/amd/io_pgtable.c index a91e71f981efb9..ba61869b23857a 100644 --- a/drivers/iommu/amd/io_pgtable.c +++ b/drivers/iommu/amd/io_pgtable.c @@ -132,7 +132,6 @@ static bool increase_address_space(struct amd_io_pgtable *pgtable,

pgtable->root = pte; pgtable->mode += 1; - amd_iommu_update_and_flush_device_table(domain);

pte = NULL; ret = true; diff --git a/drivers/iommu/amd/iommu.c b/drivers/iommu/amd/iommu.c index eb348c63a8d096..97e9b4df340a08 100644 --- a/drivers/iommu/amd/iommu.c +++ b/drivers/iommu/amd/iommu.c @@ -29,7 +29,6 @@ #include <linux/msi.h> #include <linux/irqdomain.h> #include <linux/percpu.h> -#include <linux/io-pgtable.h> #include <linux/cc_platform.h> #include <asm/irq_remapping.h> #include <asm/io_apic.h> @@ -40,9 +39,9 @@ #include <asm/gart.h> #include <asm/dma.h> #include <uapi/linux/iommufd.h> +#include <linux/generic_pt/iommu.h>

#include "amd_iommu.h" -#include "../dma-iommu.h" #include "../irq_remapping.h" #include "../iommu-pages.h"

@@ -59,7 +58,6 @@ LIST_HEAD(hpet_map); LIST_HEAD(acpihid_map);

const struct iommu_ops amd_iommu_ops; -static const struct iommu_dirty_ops amd_dirty_ops;

int amd_iommu_max_glx_val = -1;

@@ -73,11 +71,18 @@ static int amd_iommu_attach_device(struct iommu_domain *dom, struct device *dev);

static void set_dte_entry(struct amd_iommu *iommu, - struct iommu_dev_data *dev_data); + struct iommu_dev_data *dev_data, + phys_addr_t top_paddr, unsigned int top_level); + +static void amd_iommu_change_top(struct pt_iommu *iommu_table, + phys_addr_t top_paddr, unsigned int top_level);

static void iommu_flush_dte_sync(struct amd_iommu *iommu, u16 devid);

static struct iommu_dev_data *find_dev_data(struct amd_iommu *iommu, u16 devid); +static bool amd_iommu_enforce_cache_coherency(struct iommu_domain *domain); +static int amd_iommu_set_dirty_tracking(struct iommu_domain *domain, + bool enable);

/**************************************************************************** * @@ -1755,42 +1760,6 @@ static void dev_flush_pasid_all(struct iommu_dev_data *dev_data, CMD_INV_IOMMU_ALL_PAGES_ADDRESS); }

-/* Flush the not present cache if it exists */ -static void domain_flush_np_cache(struct protection_domain *domain, - dma_addr_t iova, size_t size) -{ - if (unlikely(amd_iommu_np_cache)) { - unsigned long flags; - - spin_lock_irqsave(&domain->lock, flags); - amd_iommu_domain_flush_pages(domain, iova, size); - spin_unlock_irqrestore(&domain->lock, flags); - } -} - - -/* - * This function flushes the DTEs for all devices in domain - */ -void amd_iommu_update_and_flush_device_table(struct protection_domain *domain) -{ - struct iommu_dev_data *dev_data; - - lockdep_assert_held(&domain->lock); - - list_for_each_entry(dev_data, &domain->dev_list, list) { - struct amd_iommu *iommu = rlookup_amd_iommu(dev_data->dev); - - set_dte_entry(iommu, dev_data); - clone_aliases(iommu, dev_data->dev); - } - - list_for_each_entry(dev_data, &domain->dev_list, list) - device_flush_dte(dev_data); - - domain_flush_complete(domain); -} - int amd_iommu_complete_ppr(struct device *dev, u32 pasid, int status, int tag) { struct iommu_dev_data *dev_data; @@ -2050,7 +2019,8 @@ static void set_dte_gcr3_table(struct amd_iommu *iommu, }

static void set_dte_entry(struct amd_iommu *iommu, - struct iommu_dev_data *dev_data) + struct iommu_dev_data *dev_data, + phys_addr_t top_paddr, unsigned int top_level) { u16 domid; u32 old_domid; @@ -2059,19 +2029,36 @@ static void set_dte_entry(struct amd_iommu *iommu, struct protection_domain *domain = dev_data->domain; struct gcr3_tbl_info *gcr3_info = &dev_data->gcr3_info; struct dev_table_entry *dte = &get_dev_table(iommu)[dev_data->devid]; - - if (gcr3_info && gcr3_info->gcr3_tbl) - domid = dev_data->gcr3_info.domid; - else - domid = domain->id; + struct pt_iommu_amdv1_hw_info pt_info;

make_clear_dte(dev_data, dte, &new);

- if (domain->iop.mode != PAGE_MODE_NONE) - new.data[0] |= iommu_virt_to_phys(domain->iop.root); + if (gcr3_info && gcr3_info->gcr3_tbl) + domid = dev_data->gcr3_info.domid; + else { + domid = domain->id;

- new.data[0] |= (domain->iop.mode & DEV_ENTRY_MODE_MASK) - << DEV_ENTRY_MODE_SHIFT; + if (domain->domain.type & __IOMMU_DOMAIN_PAGING) { + /* + * When updating the IO pagetable, the new top and level + * are provided as parameters. For other operations i.e. + * device attach, retrieve the current pagetable info + * via the IOMMU PT API. + */ + if (top_paddr) { + pt_info.host_pt_root = top_paddr; + pt_info.mode = top_level + 1; + } else { + WARN_ON(top_paddr || top_level); + pt_iommu_amdv1_hw_info(&domain->amdv1, + &pt_info); + } + + new.data[0] |= pt_info.host_pt_root | + (pt_info.mode & DEV_ENTRY_MODE_MASK) + << DEV_ENTRY_MODE_SHIFT; + } + }

new.data[0] |= DTE_FLAG_IR | DTE_FLAG_IW;

@@ -2137,7 +2124,7 @@ static void dev_update_dte(struct iommu_dev_data *dev_data, bool set) struct amd_iommu *iommu = get_amd_iommu_from_dev(dev_data->dev);

if (set) - set_dte_entry(iommu, dev_data); + set_dte_entry(iommu, dev_data, 0, 0); else clear_dte_entry(iommu, dev_data);

@@ -2155,6 +2142,7 @@ static int init_gcr3_table(struct iommu_dev_data *dev_data, { struct amd_iommu *iommu = get_amd_iommu_from_dev_data(dev_data); int max_pasids = dev_data->max_pasids; + struct pt_iommu_x86_64_hw_info pt_info; int ret = 0;

/* @@ -2177,7 +2165,8 @@ static int init_gcr3_table(struct iommu_dev_data *dev_data, if (!pdom_is_v2_pgtbl_mode(pdom)) return ret;

- ret = update_gcr3(dev_data, 0, iommu_virt_to_phys(pdom->iop.pgd), true); + pt_iommu_x86_64_hw_info(&pdom->amdv2, &pt_info); + ret = update_gcr3(dev_data, 0, pt_info.gcr3_pt, true); if (ret) free_gcr3_table(&dev_data->gcr3_info);

@@ -2499,54 +2488,6 @@ struct protection_domain *protection_domain_alloc(void) return domain; }

-static int pdom_setup_pgtable(struct protection_domain *domain, - struct device *dev) -{ - struct io_pgtable_ops *pgtbl_ops; - enum io_pgtable_fmt fmt; - - switch (domain->pd_mode) { - case PD_MODE_V1: - fmt = AMD_IOMMU_V1; - break; - case PD_MODE_V2: - fmt = AMD_IOMMU_V2; - break; - case PD_MODE_NONE: - WARN_ON_ONCE(1); - return -EPERM; - } - - domain->iop.pgtbl.cfg.amd.nid = dev_to_node(dev); - pgtbl_ops = alloc_io_pgtable_ops(fmt, &domain->iop.pgtbl.cfg, domain); - if (!pgtbl_ops) - return -ENOMEM; - - return 0; -} - -static inline u64 dma_max_address(enum protection_domain_mode pgtable) -{ - if (pgtable == PD_MODE_V1) - return PM_LEVEL_SIZE(amd_iommu_hpt_level); - - /* - * V2 with 4/5 level page table. Note that "2.2.6.5 AMD64 4-Kbyte Page - * Translation" shows that the V2 table sign extends the top of the - * address space creating a reserved region in the middle of the - * translation, just like the CPU does. Further Vasant says the docs are - * incomplete and this only applies to non-zero PASIDs. If the AMDv2 - * page table is assigned to the 0 PASID then there is no sign extension - * check. - * - * Since the IOMMU must have a fixed geometry, and the core code does - * not understand sign extended addressing, we have to chop off the high - * bit to get consistent behavior with attachments of the domain to any - * PASID. - */ - return ((1ULL << (PM_LEVEL_SHIFT(amd_iommu_gpt_level) - 1)) - 1); -} - static bool amd_iommu_hd_support(struct amd_iommu *iommu) { if (amd_iommu_hatdis) @@ -2555,38 +2496,229 @@ static bool amd_iommu_hd_support(struct amd_iommu *iommu) return iommu && (iommu->features & FEATURE_HDSUP); }

-static struct iommu_domain * -do_iommu_domain_alloc(struct device *dev, u32 flags, - enum protection_domain_mode pgtable) +static spinlock_t *amd_iommu_get_top_lock(struct pt_iommu *iommupt) { - bool dirty_tracking = flags & IOMMU_HWPT_ALLOC_DIRTY_TRACKING; - struct amd_iommu *iommu = get_amd_iommu_from_dev(dev); + struct protection_domain *pdom = + container_of(iommupt, struct protection_domain, iommu); + + return &pdom->lock; +} + +/* + * Update all HW references to the domain with a new pgtable configuration. + */ +static void amd_iommu_change_top(struct pt_iommu *iommu_table, + phys_addr_t top_paddr, unsigned int top_level) +{ + struct protection_domain *pdom = + container_of(iommu_table, struct protection_domain, iommu); + struct iommu_dev_data *dev_data; + + lockdep_assert_held(&pdom->lock); + + /* Update the DTE for all devices attached to this domain */ + list_for_each_entry(dev_data, &pdom->dev_list, list) { + struct amd_iommu *iommu = rlookup_amd_iommu(dev_data->dev); + + /* Update the HW references with the new level and top ptr */ + set_dte_entry(iommu, dev_data, top_paddr, top_level); + clone_aliases(iommu, dev_data->dev); + } + + list_for_each_entry(dev_data, &pdom->dev_list, list) + device_flush_dte(dev_data); + + domain_flush_complete(pdom); +} + +/* + * amd_iommu_iotlb_sync_map() is used to generate flushes for non-present to + * present (ie mapping) operations. It is a NOP if the IOMMU doesn't have non + * present caching (like hypervisor shadowing). + */ +static int amd_iommu_iotlb_sync_map(struct iommu_domain *dom, + unsigned long iova, size_t size) +{ + struct protection_domain *domain = to_pdomain(dom); + unsigned long flags; + + if (likely(!amd_iommu_np_cache)) + return 0; + + spin_lock_irqsave(&domain->lock, flags); + amd_iommu_domain_flush_pages(domain, iova, size); + spin_unlock_irqrestore(&domain->lock, flags); + return 0; +} + +static void amd_iommu_flush_iotlb_all(struct iommu_domain *domain) +{ + struct protection_domain *dom = to_pdomain(domain); + unsigned long flags; + + spin_lock_irqsave(&dom->lock, flags); + amd_iommu_domain_flush_all(dom); + spin_unlock_irqrestore(&dom->lock, flags); +} + +static void amd_iommu_iotlb_sync(struct iommu_domain *domain, + struct iommu_iotlb_gather *gather) +{ + struct protection_domain *dom = to_pdomain(domain); + unsigned long flags; + + spin_lock_irqsave(&dom->lock, flags); + amd_iommu_domain_flush_pages(dom, gather->start, + gather->end - gather->start + 1); + spin_unlock_irqrestore(&dom->lock, flags); + iommu_put_pages_list(&gather->freelist); +} + +static const struct pt_iommu_flush_ops amd_hw_flush_ops_v1 = { + .get_top_lock = amd_iommu_get_top_lock, + .change_top = amd_iommu_change_top, +}; + +static const struct iommu_domain_ops amdv1_ops = { + IOMMU_PT_DOMAIN_OPS(amdv1), + .iotlb_sync_map = amd_iommu_iotlb_sync_map, + .flush_iotlb_all = amd_iommu_flush_iotlb_all, + .iotlb_sync = amd_iommu_iotlb_sync, + .attach_dev = amd_iommu_attach_device, + .free = amd_iommu_domain_free, + .enforce_cache_coherency = amd_iommu_enforce_cache_coherency, +}; + +static const struct iommu_dirty_ops amdv1_dirty_ops = { + IOMMU_PT_DIRTY_OPS(amdv1), + .set_dirty_tracking = amd_iommu_set_dirty_tracking, +}; + +static struct iommu_domain *amd_iommu_domain_alloc_paging_v1(struct device *dev, + u32 flags) +{ + struct pt_iommu_amdv1_cfg cfg = {}; struct protection_domain *domain; int ret;

+ if (amd_iommu_hatdis) + return ERR_PTR(-EOPNOTSUPP); + domain = protection_domain_alloc(); if (!domain) return ERR_PTR(-ENOMEM);

- domain->pd_mode = pgtable; - ret = pdom_setup_pgtable(domain, dev); + domain->pd_mode = PD_MODE_V1; + domain->iommu.hw_flush_ops = &amd_hw_flush_ops_v1; + domain->iommu.nid = dev_to_node(dev); + if (flags & IOMMU_HWPT_ALLOC_DIRTY_TRACKING) + domain->domain.dirty_ops = &amdv1_dirty_ops; + + /* + * Someday FORCE_COHERENCE should be set by + * amd_iommu_enforce_cache_coherency() like VT-D does. + */ + cfg.common.features = BIT(PT_FEAT_DYNAMIC_TOP) | + BIT(PT_FEAT_AMDV1_ENCRYPT_TABLES) | + BIT(PT_FEAT_AMDV1_FORCE_COHERENCE); + + /* + * AMD's IOMMU can flush as many pages as necessary in a single flush. + * Unless we run in a virtual machine, which can be inferred according + * to whether "non-present cache" is on, it is probably best to prefer + * (potentially) too extensive TLB flushing (i.e., more misses) over + * multiple TLB flushes (i.e., more flushes). For virtual machines the + * hypervisor needs to synchronize the host IOMMU PTEs with those of + * the guest, and the trade-off is different: unnecessary TLB flushes + * should be avoided. + */ + if (amd_iommu_np_cache) + cfg.common.features |= BIT(PT_FEAT_FLUSH_RANGE_NO_GAPS); + else + cfg.common.features |= BIT(PT_FEAT_FLUSH_RANGE); + + cfg.common.hw_max_vasz_lg2 = + max(64, (amd_iommu_hpt_level - 1) * 9 + 21); + cfg.common.hw_max_oasz_lg2 = 52; + cfg.starting_level = 2; + domain->domain.ops = &amdv1_ops; + + ret = pt_iommu_amdv1_init(&domain->amdv1, &cfg, GFP_KERNEL); if (ret) { - pdom_id_free(domain->id); - kfree(domain); + amd_iommu_domain_free(&domain->domain); return ERR_PTR(ret); }

- domain->domain.geometry.aperture_start = 0; - domain->domain.geometry.aperture_end = dma_max_address(pgtable); - domain->domain.geometry.force_aperture = true; - domain->domain.pgsize_bitmap = domain->iop.pgtbl.cfg.pgsize_bitmap; + /* + * Narrow the supported page sizes to those selected by the kernel + * command line. + */ + domain->domain.pgsize_bitmap &= amd_iommu_pgsize_bitmap; + return &domain->domain; +}

- domain->domain.type = IOMMU_DOMAIN_UNMANAGED; - domain->domain.ops = iommu->iommu.ops->default_domain_ops; +static const struct iommu_domain_ops amdv2_ops = { + IOMMU_PT_DOMAIN_OPS(x86_64), + .iotlb_sync_map = amd_iommu_iotlb_sync_map, + .flush_iotlb_all = amd_iommu_flush_iotlb_all, + .iotlb_sync = amd_iommu_iotlb_sync, + .attach_dev = amd_iommu_attach_device, + .free = amd_iommu_domain_free, + /* + * Note the AMDv2 page table format does not support a Force Coherency + * bit, so enforce_cache_coherency should not be set. However VFIO is + * not prepared to handle a case where some domains will support + * enforcement and others do not. VFIO and iommufd will have to be fixed + * before it can fully use the V2 page table. See the comment in + * iommufd_hwpt_paging_alloc(). For now leave things as they have + * historically been and lie about enforce_cache_coherencey. + */ + .enforce_cache_coherency = amd_iommu_enforce_cache_coherency, +};

- if (dirty_tracking) - domain->domain.dirty_ops = &amd_dirty_ops; +static struct iommu_domain *amd_iommu_domain_alloc_paging_v2(struct device *dev, + u32 flags) +{ + struct pt_iommu_x86_64_cfg cfg = {}; + struct protection_domain *domain; + int ret;

+ if (!amd_iommu_v2_pgtbl_supported()) + return ERR_PTR(-EOPNOTSUPP); + + domain = protection_domain_alloc(); + if (!domain) + return ERR_PTR(-ENOMEM); + + domain->pd_mode = PD_MODE_V2; + domain->iommu.nid = dev_to_node(dev); + + cfg.common.features = BIT(PT_FEAT_X86_64_AMD_ENCRYPT_TABLES); + if (amd_iommu_np_cache) + cfg.common.features |= BIT(PT_FEAT_FLUSH_RANGE_NO_GAPS); + else + cfg.common.features |= BIT(PT_FEAT_FLUSH_RANGE); + + /* + * The v2 table behaves differently if it is attached to PASID 0 vs a + * non-zero PASID. On PASID 0 it has no sign extension and the full + * 57/48 bits decode the lower addresses. Otherwise it behaves like a + * normal sign extended x86 page table. Since we want the domain to work + * in both modes the top bit is removed and PT_FEAT_SIGN_EXTEND is not + * set which creates a table that is compatible in both modes. + */ + if (amd_iommu_gpt_level == PAGE_MODE_5_LEVEL) + cfg.common.hw_max_vasz_lg2 = 56; + else + cfg.common.hw_max_vasz_lg2 = 47; + cfg.common.hw_max_oasz_lg2 = 52; + domain->domain.ops = &amdv2_ops; + + ret = pt_iommu_x86_64_init(&domain->amdv2, &cfg, GFP_KERNEL); + if (ret) { + amd_iommu_domain_free(&domain->domain); + return ERR_PTR(ret); + } return &domain->domain; }

@@ -2607,15 +2739,27 @@ amd_iommu_domain_alloc_paging_flags(struct device *dev, u32 flags, /* Allocate domain with v1 page table for dirty tracking */ if (!amd_iommu_hd_support(iommu)) break; - return do_iommu_domain_alloc(dev, flags, PD_MODE_V1); + return amd_iommu_domain_alloc_paging_v1(dev, flags); case IOMMU_HWPT_ALLOC_PASID: /* Allocate domain with v2 page table if IOMMU supports PASID. */ if (!amd_iommu_pasid_supported()) break; - return do_iommu_domain_alloc(dev, flags, PD_MODE_V2); - case 0: + return amd_iommu_domain_alloc_paging_v2(dev, flags); + case 0: { + struct iommu_domain *ret; + /* If nothing specific is required use the kernel commandline default */ - return do_iommu_domain_alloc(dev, 0, amd_iommu_pgtable); + if (amd_iommu_pgtable == PD_MODE_V1) { + ret = amd_iommu_domain_alloc_paging_v1(dev, flags); + if (ret != ERR_PTR(-EOPNOTSUPP)) + return ret; + return amd_iommu_domain_alloc_paging_v2(dev, flags); + } + ret = amd_iommu_domain_alloc_paging_v2(dev, flags); + if (ret != ERR_PTR(-EOPNOTSUPP)) + return ret; + return amd_iommu_domain_alloc_paging_v1(dev, flags); + } default: break; } @@ -2627,8 +2771,7 @@ void amd_iommu_domain_free(struct iommu_domain *dom) struct protection_domain *domain = to_pdomain(dom);

WARN_ON(!list_empty(&domain->dev_list)); - if (domain->domain.type & __IOMMU_DOMAIN_PAGING) - free_io_pgtable_ops(&domain->iop.pgtbl.ops); + pt_iommu_deinit(&domain->iommu); pdom_id_free(domain->id); kfree(domain); } @@ -2733,93 +2876,6 @@ static int amd_iommu_attach_device(struct iommu_domain *dom, return ret; }

-static int amd_iommu_iotlb_sync_map(struct iommu_domain *dom, - unsigned long iova, size_t size) -{ - struct protection_domain *domain = to_pdomain(dom); - struct io_pgtable_ops *ops = &domain->iop.pgtbl.ops; - - if (ops->map_pages) - domain_flush_np_cache(domain, iova, size); - return 0; -} - -static int amd_iommu_map_pages(struct iommu_domain *dom, unsigned long iova, - phys_addr_t paddr, size_t pgsize, size_t pgcount, - int iommu_prot, gfp_t gfp, size_t *mapped) -{ - struct protection_domain *domain = to_pdomain(dom); - struct io_pgtable_ops *ops = &domain->iop.pgtbl.ops; - int prot = 0; - int ret = -EINVAL; - - if ((domain->pd_mode == PD_MODE_V1) && - (domain->iop.mode == PAGE_MODE_NONE)) - return -EINVAL; - - if (iommu_prot & IOMMU_READ) - prot |= IOMMU_PROT_IR; - if (iommu_prot & IOMMU_WRITE) - prot |= IOMMU_PROT_IW; - - if (ops->map_pages) { - ret = ops->map_pages(ops, iova, paddr, pgsize, - pgcount, prot, gfp, mapped); - } - - return ret; -} - -static void amd_iommu_iotlb_gather_add_page(struct iommu_domain *domain, - struct iommu_iotlb_gather *gather, - unsigned long iova, size_t size) -{ - /* - * AMD's IOMMU can flush as many pages as necessary in a single flush. - * Unless we run in a virtual machine, which can be inferred according - * to whether "non-present cache" is on, it is probably best to prefer - * (potentially) too extensive TLB flushing (i.e., more misses) over - * mutliple TLB flushes (i.e., more flushes). For virtual machines the - * hypervisor needs to synchronize the host IOMMU PTEs with those of - * the guest, and the trade-off is different: unnecessary TLB flushes - * should be avoided. - */ - if (amd_iommu_np_cache && - iommu_iotlb_gather_is_disjoint(gather, iova, size)) - iommu_iotlb_sync(domain, gather); - - iommu_iotlb_gather_add_range(gather, iova, size); -} - -static size_t amd_iommu_unmap_pages(struct iommu_domain *dom, unsigned long iova, - size_t pgsize, size_t pgcount, - struct iommu_iotlb_gather *gather) -{ - struct protection_domain *domain = to_pdomain(dom); - struct io_pgtable_ops *ops = &domain->iop.pgtbl.ops; - size_t r; - - if ((domain->pd_mode == PD_MODE_V1) && - (domain->iop.mode == PAGE_MODE_NONE)) - return 0; - - r = (ops->unmap_pages) ? ops->unmap_pages(ops, iova, pgsize, pgcount, NULL) : 0; - - if (r) - amd_iommu_iotlb_gather_add_page(dom, gather, iova, r); - - return r; -} - -static phys_addr_t amd_iommu_iova_to_phys(struct iommu_domain *dom, - dma_addr_t iova) -{ - struct protection_domain *domain = to_pdomain(dom); - struct io_pgtable_ops *ops = &domain->iop.pgtbl.ops; - - return ops->iova_to_phys(ops, iova); -} - static bool amd_iommu_capable(struct device *dev, enum iommu_cap cap) { switch (cap) { @@ -2886,28 +2942,6 @@ static int amd_iommu_set_dirty_tracking(struct iommu_domain *domain, return 0; }

-static int amd_iommu_read_and_clear_dirty(struct iommu_domain *domain, - unsigned long iova, size_t size, - unsigned long flags, - struct iommu_dirty_bitmap *dirty) -{ - struct protection_domain *pdomain = to_pdomain(domain); - struct io_pgtable_ops *ops = &pdomain->iop.pgtbl.ops; - unsigned long lflags; - - if (!ops || !ops->read_and_clear_dirty) - return -EOPNOTSUPP; - - spin_lock_irqsave(&pdomain->lock, lflags); - if (!pdomain->dirty_tracking && dirty->bitmap) { - spin_unlock_irqrestore(&pdomain->lock, lflags); - return -EINVAL; - } - spin_unlock_irqrestore(&pdomain->lock, lflags); - - return ops->read_and_clear_dirty(ops, iova, size, flags, dirty); -} - static void amd_iommu_get_resv_regions(struct device *dev, struct list_head *head) { @@ -2977,28 +3011,6 @@ static bool amd_iommu_is_attach_deferred(struct device *dev) return dev_data->defer_attach; }

-static void amd_iommu_flush_iotlb_all(struct iommu_domain *domain) -{ - struct protection_domain *dom = to_pdomain(domain); - unsigned long flags; - - spin_lock_irqsave(&dom->lock, flags); - amd_iommu_domain_flush_all(dom); - spin_unlock_irqrestore(&dom->lock, flags); -} - -static void amd_iommu_iotlb_sync(struct iommu_domain *domain, - struct iommu_iotlb_gather *gather) -{ - struct protection_domain *dom = to_pdomain(domain); - unsigned long flags; - - spin_lock_irqsave(&dom->lock, flags); - amd_iommu_domain_flush_pages(dom, gather->start, - gather->end - gather->start + 1); - spin_unlock_irqrestore(&dom->lock, flags); -} - static int amd_iommu_def_domain_type(struct device *dev) { struct iommu_dev_data *dev_data; @@ -3033,11 +3045,6 @@ static bool amd_iommu_enforce_cache_coherency(struct iommu_domain *domain) return true; }

-static const struct iommu_dirty_ops amd_dirty_ops = { - .set_dirty_tracking = amd_iommu_set_dirty_tracking, - .read_and_clear_dirty = amd_iommu_read_and_clear_dirty, -}; - const struct iommu_ops amd_iommu_ops = { .capable = amd_iommu_capable, .blocked_domain = &blocked_domain, @@ -3052,17 +3059,6 @@ const struct iommu_ops amd_iommu_ops = { .is_attach_deferred = amd_iommu_is_attach_deferred, .def_domain_type = amd_iommu_def_domain_type, .page_response = amd_iommu_page_response, - .default_domain_ops = &(const struct iommu_domain_ops) { - .attach_dev = amd_iommu_attach_device, - .map_pages = amd_iommu_map_pages, - .unmap_pages = amd_iommu_unmap_pages, - .iotlb_sync_map = amd_iommu_iotlb_sync_map, - .iova_to_phys = amd_iommu_iova_to_phys, - .flush_iotlb_all = amd_iommu_flush_iotlb_all, - .iotlb_sync = amd_iommu_iotlb_sync, - .free = amd_iommu_domain_free, - .enforce_cache_coherency = amd_iommu_enforce_cache_coherency, - } };

#ifdef CONFIG_IRQ_REMAP @@ -4071,3 +4067,5 @@ int amd_iommu_create_irq_domain(struct amd_iommu *iommu) return 0; } #endif + +MODULE_IMPORT_NS("GENERIC_PT_IOMMU");

This intends to have high coverage of the page table format functions and the IOMMU implementation itself, exercising the various corner cases.

The kunit tests can be run in the kunit framework, using commands like:

tools/testing/kunit/kunit.py run --build_dir build_kunit_arm64 --arch arm64 --make_options LLVM=-19 --kunitconfig ./drivers/iommu/generic_pt/.kunitconfig tools/testing/kunit/kunit.py run --build_dir build_kunit_uml --kunitconfig ./drivers/iommu/generic_pt/.kunitconfig --kconfig_add CONFIG_WERROR=n --kconfig_add CONFIG_UML_PCI_OVER_VIRTIO_DEVICE_ID=100 tools/testing/kunit/kunit.py run --build_dir build_kunit_x86_64 --arch x86_64 --kunitconfig ./drivers/iommu/generic_pt/.kunitconfig tools/testing/kunit/kunit.py run --build_dir build_kunit_i386 --arch i386 --kunitconfig ./drivers/iommu/generic_pt/.kunitconfig tools/testing/kunit/kunit.py run --build_dir build_kunit_i386pae --arch i386 --kunitconfig ./drivers/iommu/generic_pt/.kunitconfig --kconfig_add CONFIG_X86_PAE=y

There are several interesting corner cases on the 32 bit platforms that need checking.

Like the generic tests, these are run on the format's configuration list using kunit "params". This also checks the core iommu parts of the page table code as it enters the logic through a mock iommu_domain.

The following are checked: - PT_FEAT_DYNAMIC_TOP properly adds levels one by one - Every page size can be iommu_map()'d, and mapping creates that size - iommu_iova_to_phys() works with every page size - Test converting OA -> non present -> OA when the two OAs overlap and free table levels - Test that unmap stops at holes, unmap doesn't split, and unmap returns the right values for partial unmap requests - Randomly map/unmap. Checks map with random sizes, that map fails when hitting collisions doing nothing, unmap/map with random intersections and full unmap of random sizes. Also checks iommu_iova_to_phys() with random sizes - Check for memory leaks by monitoring NR_SECONDARY_PAGETABLE

Tested-by: Alejandro Jimenez alejandro.j.jimenez@oracle.com Signed-off-by: Jason Gunthorpe jgg@nvidia.com --- drivers/iommu/generic_pt/fmt/iommu_template.h | 1 + drivers/iommu/generic_pt/kunit_iommu.h | 2 + drivers/iommu/generic_pt/kunit_iommu_pt.h | 451 ++++++++++++++++++ 3 files changed, 454 insertions(+) create mode 100644 drivers/iommu/generic_pt/kunit_iommu_pt.h

diff --git a/drivers/iommu/generic_pt/fmt/iommu_template.h b/drivers/iommu/generic_pt/fmt/iommu_template.h index 11e85106ae302e..d28e86abdf2e74 100644 --- a/drivers/iommu/generic_pt/fmt/iommu_template.h +++ b/drivers/iommu/generic_pt/fmt/iommu_template.h @@ -44,4 +44,5 @@ * which means we are building the kunit modle. */ #include "../kunit_generic_pt.h" +#include "../kunit_iommu_pt.h" #endif diff --git a/drivers/iommu/generic_pt/kunit_iommu.h b/drivers/iommu/generic_pt/kunit_iommu.h index 8a53b1d772ca9d..cca4e72efcaa04 100644 --- a/drivers/iommu/generic_pt/kunit_iommu.h +++ b/drivers/iommu/generic_pt/kunit_iommu.h @@ -70,6 +70,8 @@ struct kunit_iommu_priv { unsigned int largest_pgsz_lg2; pt_oaddr_t test_oa; pt_vaddr_t safe_pgsize_bitmap; + unsigned long orig_nr_secondary_pagetable; + }; PT_IOMMU_CHECK_DOMAIN(struct kunit_iommu_priv, fmt_table.iommu, domain);

diff --git a/drivers/iommu/generic_pt/kunit_iommu_pt.h b/drivers/iommu/generic_pt/kunit_iommu_pt.h new file mode 100644 index 00000000000000..5e25d698450783 --- /dev/null +++ b/drivers/iommu/generic_pt/kunit_iommu_pt.h @@ -0,0 +1,451 @@ +/* SPDX-License-Identifier: GPL-2.0-only */ +/* + * Copyright (c) 2024, NVIDIA CORPORATION & AFFILIATES + */ +#include "kunit_iommu.h" +#include "pt_iter.h" +#include <linux/generic_pt/iommu.h> +#include <linux/iommu.h> + +static void do_map(struct kunit *test, pt_vaddr_t va, pt_oaddr_t pa, + pt_vaddr_t len); + +struct count_valids { + u64 per_size[PT_VADDR_MAX_LG2]; +}; + +static int __count_valids(struct pt_range *range, void *arg, unsigned int level, + struct pt_table_p *table) +{ + struct pt_state pts = pt_init(range, level, table); + struct count_valids *valids = arg; + + for_each_pt_level_entry(&pts) { + if (pts.type == PT_ENTRY_TABLE) { + pt_descend(&pts, arg, __count_valids); + continue; + } + if (pts.type == PT_ENTRY_OA) { + valids->per_size[pt_entry_oa_lg2sz(&pts)]++; + continue; + } + } + return 0; +} + +/* + * Number of valid table entries. This counts contiguous entries as a single + * valid. + */ +static unsigned int count_valids(struct kunit *test) +{ + struct kunit_iommu_priv *priv = test->priv; + struct pt_range range = pt_top_range(priv->common); + struct count_valids valids = {}; + u64 total = 0; + unsigned int i; + + KUNIT_ASSERT_NO_ERRNO(test, + pt_walk_range(&range, __count_valids, &valids)); + + for (i = 0; i != ARRAY_SIZE(valids.per_size); i++) + total += valids.per_size[i]; + return total; +} + +/* Only a single page size is present, count the number of valid entries */ +static unsigned int count_valids_single(struct kunit *test, pt_vaddr_t pgsz) +{ + struct kunit_iommu_priv *priv = test->priv; + struct pt_range range = pt_top_range(priv->common); + struct count_valids valids = {}; + u64 total = 0; + unsigned int i; + + KUNIT_ASSERT_NO_ERRNO(test, + pt_walk_range(&range, __count_valids, &valids)); + + for (i = 0; i != ARRAY_SIZE(valids.per_size); i++) { + if ((1ULL << i) == pgsz) + total = valids.per_size[i]; + else + KUNIT_ASSERT_EQ(test, valids.per_size[i], 0); + } + return total; +} + +static void do_unmap(struct kunit *test, pt_vaddr_t va, pt_vaddr_t len) +{ + struct kunit_iommu_priv *priv = test->priv; + size_t ret; + + ret = iommu_unmap(&priv->domain, va, len); + KUNIT_ASSERT_EQ(test, ret, len); +} + +static void check_iova(struct kunit *test, pt_vaddr_t va, pt_oaddr_t pa, + pt_vaddr_t len) +{ + struct kunit_iommu_priv *priv = test->priv; + pt_vaddr_t pfn = log2_div(va, priv->smallest_pgsz_lg2); + pt_vaddr_t end_pfn = pfn + log2_div(len, priv->smallest_pgsz_lg2); + + for (; pfn != end_pfn; pfn++) { + phys_addr_t res = iommu_iova_to_phys(&priv->domain, + pfn * priv->smallest_pgsz); + + KUNIT_ASSERT_EQ(test, res, (phys_addr_t)pa); + if (res != pa) + break; + pa += priv->smallest_pgsz; + } +} + +static void test_increase_level(struct kunit *test) +{ + struct kunit_iommu_priv *priv = test->priv; + struct pt_common *common = priv->common; + + if (!pt_feature(common, PT_FEAT_DYNAMIC_TOP)) + kunit_skip(test, "PT_FEAT_DYNAMIC_TOP not set for this format"); + + if (IS_32BIT) + kunit_skip(test, "Unable to test on 32bit"); + + KUNIT_ASSERT_GT(test, common->max_vasz_lg2, + pt_top_range(common).max_vasz_lg2); + + /* Add every possible level to the max */ + while (common->max_vasz_lg2 != pt_top_range(common).max_vasz_lg2) { + struct pt_range top_range = pt_top_range(common); + + if (top_range.va == 0) + do_map(test, top_range.last_va + 1, 0, + priv->smallest_pgsz); + else + do_map(test, top_range.va - priv->smallest_pgsz, 0, + priv->smallest_pgsz); + + KUNIT_ASSERT_EQ(test, pt_top_range(common).top_level, + top_range.top_level + 1); + KUNIT_ASSERT_GE(test, common->max_vasz_lg2, + pt_top_range(common).max_vasz_lg2); + } +} + +static void test_map_simple(struct kunit *test) +{ + struct kunit_iommu_priv *priv = test->priv; + struct pt_range range = pt_top_range(priv->common); + struct count_valids valids = {}; + pt_vaddr_t pgsize_bitmap = priv->safe_pgsize_bitmap; + unsigned int pgsz_lg2; + pt_vaddr_t cur_va; + + /* Map every reported page size */ + cur_va = range.va + priv->smallest_pgsz * 256; + for (pgsz_lg2 = 0; pgsz_lg2 != PT_VADDR_MAX_LG2; pgsz_lg2++) { + pt_oaddr_t paddr = log2_set_mod(priv->test_oa, 0, pgsz_lg2); + u64 len = log2_to_int(pgsz_lg2); + + if (!(pgsize_bitmap & len)) + continue; + + cur_va = ALIGN(cur_va, len); + do_map(test, cur_va, paddr, len); + if (len <= SZ_2G) + check_iova(test, cur_va, paddr, len); + cur_va += len; + } + + /* The read interface reports that every page size was created */ + range = pt_top_range(priv->common); + KUNIT_ASSERT_NO_ERRNO(test, + pt_walk_range(&range, __count_valids, &valids)); + for (pgsz_lg2 = 0; pgsz_lg2 != PT_VADDR_MAX_LG2; pgsz_lg2++) { + if (pgsize_bitmap & (1ULL << pgsz_lg2)) + KUNIT_ASSERT_EQ(test, valids.per_size[pgsz_lg2], 1); + else + KUNIT_ASSERT_EQ(test, valids.per_size[pgsz_lg2], 0); + } + + /* Unmap works */ + range = pt_top_range(priv->common); + cur_va = range.va + priv->smallest_pgsz * 256; + for (pgsz_lg2 = 0; pgsz_lg2 != PT_VADDR_MAX_LG2; pgsz_lg2++) { + u64 len = log2_to_int(pgsz_lg2); + + if (!(pgsize_bitmap & len)) + continue; + cur_va = ALIGN(cur_va, len); + do_unmap(test, cur_va, len); + cur_va += len; + } + KUNIT_ASSERT_EQ(test, count_valids(test), 0); +} + +/* + * Test to convert a table pointer into an OA by mapping something small, + * unmapping it so as to leave behind a table pointer, then mapping something + * larger that will convert the table into an OA. + */ +static void test_map_table_to_oa(struct kunit *test) +{ + struct kunit_iommu_priv *priv = test->priv; + pt_vaddr_t limited_pgbitmap = + priv->info.pgsize_bitmap % (IS_32BIT ? SZ_2G : SZ_16G); + struct pt_range range = pt_top_range(priv->common); + unsigned int pgsz_lg2; + pt_vaddr_t max_pgsize; + pt_vaddr_t cur_va; + + max_pgsize = 1ULL << (log2_fls(limited_pgbitmap) - 1); + KUNIT_ASSERT_TRUE(test, priv->info.pgsize_bitmap & max_pgsize); + + for (pgsz_lg2 = 0; pgsz_lg2 != PT_VADDR_MAX_LG2; pgsz_lg2++) { + pt_oaddr_t paddr = log2_set_mod(priv->test_oa, 0, pgsz_lg2); + u64 len = log2_to_int(pgsz_lg2); + pt_vaddr_t offset; + + if (!(priv->info.pgsize_bitmap & len)) + continue; + if (len > max_pgsize) + break; + + cur_va = ALIGN(range.va + priv->smallest_pgsz * 256, + max_pgsize); + for (offset = 0; offset != max_pgsize; offset += len) + do_map(test, cur_va + offset, paddr + offset, len); + check_iova(test, cur_va, paddr, max_pgsize); + KUNIT_ASSERT_EQ(test, count_valids_single(test, len), + max_pgsize / len); + + if (len == max_pgsize) { + do_unmap(test, cur_va, max_pgsize); + } else { + do_unmap(test, cur_va, max_pgsize / 2); + for (offset = max_pgsize / 2; offset != max_pgsize; + offset += len) + do_unmap(test, cur_va + offset, len); + } + + KUNIT_ASSERT_EQ(test, count_valids(test), 0); + } +} + +/* + * Test unmapping a small page at the start of a large page. This always unmaps + * the large page. + */ +static void test_unmap_split(struct kunit *test) +{ + struct kunit_iommu_priv *priv = test->priv; + struct pt_range top_range = pt_top_range(priv->common); + pt_vaddr_t pgsize_bitmap = priv->safe_pgsize_bitmap; + unsigned int pgsz_lg2; + unsigned int count = 0; + + for (pgsz_lg2 = 0; pgsz_lg2 != PT_VADDR_MAX_LG2; pgsz_lg2++) { + pt_vaddr_t base_len = log2_to_int(pgsz_lg2); + unsigned int next_pgsz_lg2; + + if (!(pgsize_bitmap & base_len)) + continue; + + for (next_pgsz_lg2 = pgsz_lg2 + 1; + next_pgsz_lg2 != PT_VADDR_MAX_LG2; next_pgsz_lg2++) { + pt_vaddr_t next_len = log2_to_int(next_pgsz_lg2); + pt_vaddr_t vaddr = top_range.va; + pt_oaddr_t paddr = 0; + size_t gnmapped; + + if (!(pgsize_bitmap & next_len)) + continue; + + do_map(test, vaddr, paddr, next_len); + gnmapped = iommu_unmap(&priv->domain, vaddr, base_len); + KUNIT_ASSERT_EQ(test, gnmapped, next_len); + + /* Make sure unmap doesn't keep going */ + do_map(test, vaddr, paddr, next_len); + do_map(test, vaddr + next_len, paddr, next_len); + gnmapped = iommu_unmap(&priv->domain, vaddr, base_len); + KUNIT_ASSERT_EQ(test, gnmapped, next_len); + gnmapped = iommu_unmap(&priv->domain, vaddr + next_len, + next_len); + KUNIT_ASSERT_EQ(test, gnmapped, next_len); + + count++; + } + } + + if (count == 0) + kunit_skip(test, "Test needs two page sizes"); +} + +static void unmap_collisions(struct kunit *test, struct maple_tree *mt, + pt_vaddr_t start, pt_vaddr_t last) +{ + struct kunit_iommu_priv *priv = test->priv; + MA_STATE(mas, mt, start, last); + void *entry; + + mtree_lock(mt); + mas_for_each(&mas, entry, last) { + pt_vaddr_t mas_start = mas.index; + pt_vaddr_t len = (mas.last - mas_start) + 1; + pt_oaddr_t paddr; + + mas_erase(&mas); + mas_pause(&mas); + mtree_unlock(mt); + + paddr = oalog2_mod(mas_start, priv->common->max_oasz_lg2); + check_iova(test, mas_start, paddr, len); + do_unmap(test, mas_start, len); + mtree_lock(mt); + } + mtree_unlock(mt); +} + +static void clamp_range(struct kunit *test, struct pt_range *range) +{ + struct kunit_iommu_priv *priv = test->priv; + + if (range->last_va - range->va > SZ_1G) + range->last_va = range->va + SZ_1G; + KUNIT_ASSERT_NE(test, range->last_va, PT_VADDR_MAX); + if (range->va <= MAPLE_RESERVED_RANGE) + range->va = + ALIGN(MAPLE_RESERVED_RANGE, priv->smallest_pgsz); +} + +/* + * Randomly map and unmap ranges that can large physical pages. If a random + * range overlaps with existing ranges then unmap them. This hits all the + * special cases. + */ +static void test_random_map(struct kunit *test) +{ + struct kunit_iommu_priv *priv = test->priv; + struct pt_range upper_range = pt_upper_range(priv->common); + struct pt_range top_range = pt_top_range(priv->common); + struct maple_tree mt; + unsigned int iter; + + mt_init(&mt); + + /* + * Shrink the range so randomization is more likely to have + * intersections + */ + clamp_range(test, &top_range); + clamp_range(test, &upper_range); + + for (iter = 0; iter != 1000; iter++) { + struct pt_range *range = &top_range; + pt_oaddr_t paddr; + pt_vaddr_t start; + pt_vaddr_t end; + int ret; + + if (pt_feature(priv->common, PT_FEAT_SIGN_EXTEND) && + ULONG_MAX >= PT_VADDR_MAX && get_random_u32_inclusive(0, 1)) + range = &upper_range; + + start = get_random_u32_below( + min(U32_MAX, range->last_va - range->va)); + end = get_random_u32_below( + min(U32_MAX, range->last_va - start)); + + start = ALIGN_DOWN(start, priv->smallest_pgsz); + end = ALIGN(end, priv->smallest_pgsz); + start += range->va; + end += start; + if (start < range->va || end > range->last_va + 1 || + start >= end) + continue; + + /* Try overmapping to test the failure handling */ + paddr = oalog2_mod(start, priv->common->max_oasz_lg2); + ret = iommu_map(&priv->domain, start, paddr, end - start, + IOMMU_READ | IOMMU_WRITE, GFP_KERNEL); + if (ret) { + KUNIT_ASSERT_EQ(test, ret, -EADDRINUSE); + unmap_collisions(test, &mt, start, end - 1); + do_map(test, start, paddr, end - start); + } + + KUNIT_ASSERT_NO_ERRNO_FN(test, "mtree_insert_range", + mtree_insert_range(&mt, start, end - 1, + XA_ZERO_ENTRY, + GFP_KERNEL)); + + check_iova(test, start, paddr, end - start); + if (iter % 100) + cond_resched(); + } + + unmap_collisions(test, &mt, 0, PT_VADDR_MAX); + KUNIT_ASSERT_EQ(test, count_valids(test), 0); + + mtree_destroy(&mt); +} + +static struct kunit_case iommu_test_cases[] = { + KUNIT_CASE_FMT(test_increase_level), + KUNIT_CASE_FMT(test_map_simple), + KUNIT_CASE_FMT(test_map_table_to_oa), + KUNIT_CASE_FMT(test_unmap_split), + KUNIT_CASE_FMT(test_random_map), + {}, +}; + +static int pt_kunit_iommu_init(struct kunit *test) +{ + struct kunit_iommu_priv *priv; + int ret; + + priv = kunit_kzalloc(test, sizeof(*priv), GFP_KERNEL); + if (!priv) + return -ENOMEM; + + priv->orig_nr_secondary_pagetable = + global_node_page_state(NR_SECONDARY_PAGETABLE); + ret = pt_kunit_priv_init(test, priv); + if (ret) { + kunit_kfree(test, priv); + return ret; + } + test->priv = priv; + return 0; +} + +static void pt_kunit_iommu_exit(struct kunit *test) +{ + struct kunit_iommu_priv *priv = test->priv; + + if (!test->priv) + return; + + pt_iommu_deinit(priv->iommu); + /* + * Look for memory leaks, assumes kunit is running isolated and nothing + * else is using secondary page tables. + */ + KUNIT_ASSERT_EQ(test, priv->orig_nr_secondary_pagetable, + global_node_page_state(NR_SECONDARY_PAGETABLE)); + kunit_kfree(test, test->priv); +} + +static struct kunit_suite NS(iommu_suite) = { + .name = __stringify(NS(iommu_test)), + .init = pt_kunit_iommu_init, + .exit = pt_kunit_iommu_exit, + .test_cases = iommu_test_cases, +}; +kunit_test_suites(&NS(iommu_suite)); + +MODULE_LICENSE("GPL"); +MODULE_DESCRIPTION("Kunit for generic page table"); +MODULE_IMPORT_NS("GENERIC_PT_IOMMU");