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 * The size of the top table level can be selected at runtime (ARM's concatenated tables) * The number of levels in the table can optionally increase dynamically during map (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.

Reviewed-by: Kevin Tian kevin.tian@intel.com Reviewed-by: Pasha Tatashin pasha.tatashin@soleen.com Reviewed-by: Samiullah Khawaja skhawaja@google.com Tested-by: Alejandro Jimenez alejandro.j.jimenez@oracle.com Tested-by: Pasha Tatashin pasha.tatashin@soleen.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 | 358 ++++++++++++ drivers/iommu/generic_pt/pt_defs.h | 329 +++++++++++ drivers/iommu/generic_pt/pt_fmt_defaults.h | 233 ++++++++ drivers/iommu/generic_pt/pt_iter.h | 636 +++++++++++++++++++++ drivers/iommu/generic_pt/pt_log2.h | 122 ++++ include/linux/generic_pt/common.h | 135 +++++ 9 files changed, 1836 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 f371a13b4d192d..9e6a9177f8fb32 100644 --- a/.clang-format +++ b/.clang-format @@ -415,6 +415,7 @@ 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..f64f800725dbb7 --- /dev/null +++ b/drivers/iommu/generic_pt/pt_common.h @@ -0,0 +1,358 @@ +/* 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. The functions without bodies are + * implemented in the format using the pattern: + * static inline FMTpt_XXX(..) {..} + * #define pt_XXX FMTpt_XXX + * + * If the format doesn't implement a function then pt_fmt_defaults.h can provide + * a generic version. + * + * 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.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_entries() - 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_entries(struct pt_state *pts, + unsigned int num_contig_lg2); + +/** + * pt_entry_make_write_dirty() - Make an entry dirty + * @pts: Table entry to change + * + * Make pt_entry_is_write_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_entry_make_write_clean() - Make the entry write clean + * @pts: Table entry to change + * + * Modify the entry so that pt_entry_is_write_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_make_write_clean(struct pt_state *pts); + +/** + * pt_entry_is_write_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_is_write_dirty(const 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 an 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_exact() - Return the complete 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 exact OA including the pts's VA + * offset. + */ +static inline pt_oaddr_t pt_entry_oa_exact(const struct pt_state *pts) +{ + return _pt_entry_oa_fast(pts) | + log2_mod(pts->range->va, pt_entry_oa_lg2sz(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_size() - 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_size(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/OA 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_oa_lg2() - Return the maximum OA the table format can hold + * @common: Page table to query + * + * The value oalog2_to_max_int(pt_max_oa_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_oa_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..819057de50d82c --- /dev/null +++ b/drivers/iommu/generic_pt/pt_defs.h @@ -0,0 +1,329 @@ +/* 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. + * 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 + * Level 0 is always a table of only leaves with no futher table pointers. + * Increasing levels increase the size of the table items. The least + * significant VA bits used to index page tables are used to index the Level + * 0 table. The various labels for table levels used by HW descriptions are + * not 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 translation items for that level. + * index + * The position in a table of an element: item = table[index] + * item + * A single index 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 all the items that comprise a single contiguous translation. + * item/entry_size + * The number of bytes of VA the table index translates for. + * If the item is a table entry then the next table covers + * this size. If the entry translates to 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, + /* Entry is valid and points to a lower table level */ + 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 vaffs(a) ffs_t(pt_vaddr_t, a) +#define vafls(a) fls_t(pt_vaddr_t, a) +#define vaffz(a) 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 oaffs(a) ffs_t(pt_oaddr_t, a) +#define oafls(a) fls_t(pt_oaddr_t, a) +#define oaffz(a) 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); +} + +static inline bool pt_check_install_leaf_args(struct pt_state *pts, + pt_oaddr_t oa, + unsigned int oasz_lg2); + +#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..60d594bbb1063e --- /dev/null +++ b/drivers/iommu/generic_pt/pt_fmt_defaults.h @@ -0,0 +1,233 @@ +/* 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 ((unsigned int)(pgsize_lg2 - PT_GRANULE_LG2SZ)) / + (PT_TABLEMEM_LG2SZ - ilog2(PT_ITEM_WORD_SIZE)); +} +#endif + +/* + * If not supplied by the format then contiguous pages are not supported. + * + * If contiguous pages are supported then the format must also provide + * pt_contig_count_lg2() if it supports a single contiguous size per level, + * or pt_possible_sizes() if it supports multiple sizes per level. + */ +#ifndef pt_entry_num_contig_lg2 +static inline unsigned int pt_entry_num_contig_lg2(const struct pt_state *pts) +{ + return ilog2(1); +} + +/* + * Return the number of contiguous OA items forming an entry at this table level + */ +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_is_write_dirty +static inline bool pt_entry_is_write_dirty(const struct pt_state *pts) +{ + return false; +} + +static inline void pt_entry_make_write_clean(struct pt_state *pts) +{ +} + +static inline bool pt_dirty_supported(struct pt_common *common) +{ + return false; +} +#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 adjusted for every item in a contiguous entry + * + * Build the missing one + * + * The internal helper _pt_entry_oa_fast() allows generating + * an efficient pt_entry_oa_exact(), it doesn't care which + * option is selected. + */ +#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_oa_lg2 +static inline unsigned int +pt_max_oa_lg2(const struct pt_common *common) +{ + return PT_MAX_OUTPUT_ADDRESS_LG2; +} +#endif + +#ifndef pt_has_system_page_size +static inline bool pt_has_system_page_size(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_entries +static inline void pt_clear_entries64(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_entries32(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_entries(struct pt_state *pts, + unsigned int num_contig_lg2) +{ + if (PT_ITEM_WORD_SIZE == sizeof(u32)) + pt_clear_entries32(pts, num_contig_lg2); + else + pt_clear_entries64(pts, num_contig_lg2); +} +#define pt_clear_entries pt_clear_entries +#endif + +/* + * Format can call in the pt_install_leaf_entry() to check the arguments are all + * aligned correctly. + */ +static inline bool pt_check_install_leaf_args(struct pt_state *pts, + pt_oaddr_t oa, + unsigned int oasz_lg2) +{ + unsigned int isz_lg2 = pt_table_item_lg2sz(pts); + + if (PT_WARN_ON(oalog2_mod(oa, oasz_lg2))) + return false; + +#ifdef pt_possible_sizes + if (PT_WARN_ON(isz_lg2 > oasz_lg2 || + oasz_lg2 > isz_lg2 + pt_num_items_lg2(pts))) + return false; +#else + if (PT_WARN_ON(oasz_lg2 != isz_lg2 && + oasz_lg2 != isz_lg2 + pt_contig_count_lg2(pts))) + return false; +#endif + + if (PT_WARN_ON(oalog2_mod(pts->index, oasz_lg2 - isz_lg2))) + return false; + return true; +} + +#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..87f4a26c1a417a --- /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_entry_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_entry_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(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 size + * 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(vafls(len) - 1); + + best_pgsz_lg2 = vaffs(mask); + + /* Choose the highest 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 = vafls(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..6dbbed11923853 --- /dev/null +++ b/drivers/iommu/generic_pt/pt_log2.h @@ -0,0 +1,122 @@ +/* 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: + * fls_t(u32, 0) == 0 + * fls_t(u3, 1) == 1 + * a >= log2_to_int(ret - 1) + * aka find last set bit + */ +static inline unsigned int fls32(u32 a) +{ + return fls(a); +} +#define fls_t(type, a) _dispatch_sz(type, fls, a) + +/* + * Return the highest value such that: + * ffs_t(u32, 0) == UNDEFINED + * ffs_t(u32, 1) == 0 + * log_mod(a, ret) == 0 + * aka find first set bit + */ +static inline unsigned int __ffs32(u32 a) +{ + return __ffs(a); +} +#define ffs_t(type, a) _dispatch_sz(type, __ffs, a) + +/* + * Return the highest value such that: + * ffz_t(u32, U32_MAX) == UNDEFINED + * ffz_t(u32, 0) == 0 + * ffz_t(u32, 1) == 1 + * log_mod(a, ret) == log_to_max_int(ret) + * aka find first zero bit + */ +static inline unsigned int ffz32(u32 a) +{ + return ffz(a); +} +static inline unsigned int ffz64(u64 a) +{ + if (sizeof(u64) == sizeof(unsigned long)) + return ffz(a); + + if ((u32)a == U32_MAX) + return ffz32(a >> 32) + 32; + return ffz32(a); +} +#define ffz_t(type, a) _dispatch_sz(type, 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..e69a75511313cb --- /dev/null +++ b/include/linux/generic_pt/common.h @@ -0,0 +1,135 @@ +/* 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 specialized 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 + * - Sign extended addressing + */ + +/** + * 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

The existing IOMMU page table implementations duplicate all of the working algorithms for each format. By using the generic page table API a single C version of the IOMMU algorithms can be created and re-used for all of the different formats used in the drivers. The implementation will provide a single C version of the iommu domain operations: iova_to_phys, map, unmap, and read_and_clear_dirty.

Further, adding new algorithms and techniques becomes easy to do across the entire fleet of drivers and formats.

The C functions are drop in compatible with the existing iommu_domain_ops using the IOMMU_PT_DOMAIN_OPS() macro. Each per-format implementation compilation unit will produce exported symbols following the pattern pt_iommu_FMT_map_pages() which the macro directly maps to the iommu_domain_ops members. This avoids the additional function pointer indirection like io-pgtable has.

The top level struct used by the drivers is pt_iommu_table_FMT. It contains the other structs to allow container_of() to move between the driver, iommu page table, generic page table, and generic format layers.

struct pt_iommu_table_amdv1 { struct pt_iommu { struct iommu_domain domain; } iommu; struct pt_amdv1 { struct pt_common common; } amdpt; };

The driver is expected to union the pt_iommu_table_FMT with its own existing domain struct:

struct driver_domain { union { struct iommu_domain domain; struct pt_iommu_table_amdv1 amdv1; }; }; PT_IOMMU_CHECK_DOMAIN(struct driver_domain, amdv1, domain);

To create an alias to avoid renaming 'domain' in a lot of driver code.

This allows all the layers to access all the necessary functions to implement their different roles with no change to any of the existing iommu core code.

Implement the basic starting point: pt_iommu_init(), get_info() and deinit().

Reviewed-by: Kevin Tian kevin.tian@intel.com Reviewed-by: Pasha Tatashin pasha.tatashin@soleen.com Reviewed-by: Samiullah Khawaja skhawaja@google.com Tested-by: Alejandro Jimenez alejandro.j.jimenez@oracle.com Tested-by: Pasha Tatashin pasha.tatashin@soleen.com Signed-off-by: Jason Gunthorpe jgg@nvidia.com --- drivers/iommu/generic_pt/Kconfig | 13 + drivers/iommu/generic_pt/fmt/iommu_template.h | 39 +++ drivers/iommu/generic_pt/iommu_pt.h | 259 ++++++++++++++++++ include/linux/generic_pt/iommu.h | 150 ++++++++++ 4 files changed, 461 insertions(+) create mode 100644 drivers/iommu/generic_pt/fmt/iommu_template.h create mode 100644 drivers/iommu/generic_pt/iommu_pt.h create mode 100644 include/linux/generic_pt/iommu.h

diff --git a/drivers/iommu/generic_pt/Kconfig b/drivers/iommu/generic_pt/Kconfig index fb0f431ddba0a8..a81dfdd72ca016 100644 --- a/drivers/iommu/generic_pt/Kconfig +++ b/drivers/iommu/generic_pt/Kconfig @@ -17,4 +17,17 @@ config DEBUG_GENERIC_PT kernels.

The kunit tests require this to be enabled to get full coverage. + +config IOMMU_PT + tristate "IOMMU Page Tables" + select IOMMU_API + depends on IOMMU_SUPPORT + depends on GENERIC_PT + help + Generic library for building IOMMU page tables + + IOMMU_PT provides an implementation of the page table operations + related to struct iommu_domain using GENERIC_PT. It provides a single + implementation of the page table operations that can be shared by + multiple drivers. endif diff --git a/drivers/iommu/generic_pt/fmt/iommu_template.h b/drivers/iommu/generic_pt/fmt/iommu_template.h new file mode 100644 index 00000000000000..5b631bc07cbc16 --- /dev/null +++ b/drivers/iommu/generic_pt/fmt/iommu_template.h @@ -0,0 +1,39 @@ +/* SPDX-License-Identifier: GPL-2.0-only */ +/* + * Copyright (c) 2024-2025, NVIDIA CORPORATION & AFFILIATES + * + * Template to build the iommu module and kunit from the format and + * implementation headers. + * + * The format should have: + * #define PT_FMT <name> + * #define PT_SUPPORTED_FEATURES (BIT(PT_FEAT_xx) | BIT(PT_FEAT_yy)) + * And optionally: + * #define PT_FORCE_ENABLED_FEATURES .. + * #define PT_FMT_VARIANT <suffix> + */ +#include <linux/args.h> +#include <linux/stringify.h> + +#ifdef PT_FMT_VARIANT +#define PTPFX_RAW \ + CONCATENATE(CONCATENATE(PT_FMT, _), PT_FMT_VARIANT) +#else +#define PTPFX_RAW PT_FMT +#endif + +#define PTPFX CONCATENATE(PTPFX_RAW, _) + +#define _PT_FMT_H PT_FMT.h +#define PT_FMT_H __stringify(_PT_FMT_H) + +#define _PT_DEFS_H CONCATENATE(defs_, _PT_FMT_H) +#define PT_DEFS_H __stringify(_PT_DEFS_H) + +#include <linux/generic_pt/common.h> +#include PT_DEFS_H +#include "../pt_defs.h" +#include PT_FMT_H +#include "../pt_common.h" + +#include "../iommu_pt.h" diff --git a/drivers/iommu/generic_pt/iommu_pt.h b/drivers/iommu/generic_pt/iommu_pt.h new file mode 100644 index 00000000000000..564f2d3a6e11e1 --- /dev/null +++ b/drivers/iommu/generic_pt/iommu_pt.h @@ -0,0 +1,259 @@ +/* SPDX-License-Identifier: GPL-2.0-only */ +/* + * Copyright (c) 2024-2025, NVIDIA CORPORATION & AFFILIATES + * + * "Templated C code" for implementing the iommu operations for page tables. + * This is compiled multiple times, over all the page table formats to pick up + * the per-format definitions. + */ +#ifndef __GENERIC_PT_IOMMU_PT_H +#define __GENERIC_PT_IOMMU_PT_H + +#include "pt_iter.h" + +#include <linux/export.h> +#include <linux/iommu.h> +#include "../iommu-pages.h" + +#define DOMAIN_NS(op) CONCATENATE(CONCATENATE(pt_iommu_, PTPFX), op) + +struct pt_iommu_collect_args { + struct iommu_pages_list free_list; +}; + +static int __collect_tables(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_collect_args *collect = arg; + int ret; + + if (!pt_can_have_table(&pts)) + return 0; + + for_each_pt_level_entry(&pts) { + if (pts.type == PT_ENTRY_TABLE) { + iommu_pages_list_add(&collect->free_list, pts.table_lower); + ret = pt_descend(&pts, arg, __collect_tables); + if (ret) + return ret; + continue; + } + } + return 0; +} + +static inline struct pt_table_p *table_alloc_top(struct pt_common *common, + uintptr_t top_of_table, + gfp_t gfp) +{ + struct pt_iommu *iommu_table = iommu_from_common(common); + + /* + * Top doesn't need the free list or otherwise, so it technically + * doesn't need to use iommu pages. Use the API anyhow as the top is + * usually not smaller than PAGE_SIZE to keep things simple. + */ + return iommu_alloc_pages_node_sz( + iommu_table->nid, gfp, + log2_to_int(pt_top_memsize_lg2(common, top_of_table))); +} + +static void NS(get_info)(struct pt_iommu *iommu_table, + struct pt_iommu_info *info) +{ + struct pt_common *common = common_from_iommu(iommu_table); + struct pt_range range = pt_top_range(common); + struct pt_state pts = pt_init_top(&range); + pt_vaddr_t pgsize_bitmap = 0; + + if (pt_feature(common, PT_FEAT_DYNAMIC_TOP)) { + for (pts.level = 0; pts.level <= PT_MAX_TOP_LEVEL; + pts.level++) { + if (pt_table_item_lg2sz(&pts) >= common->max_vasz_lg2) + break; + pgsize_bitmap |= pt_possible_sizes(&pts); + } + } else { + for (pts.level = 0; pts.level <= range.top_level; pts.level++) + pgsize_bitmap |= pt_possible_sizes(&pts); + } + + /* Hide page sizes larger than the maximum OA */ + info->pgsize_bitmap = oalog2_mod(pgsize_bitmap, common->max_oasz_lg2); +} + +static void NS(deinit)(struct pt_iommu *iommu_table) +{ + struct pt_common *common = common_from_iommu(iommu_table); + struct pt_range range = pt_all_range(common); + struct pt_iommu_collect_args collect = { + .free_list = IOMMU_PAGES_LIST_INIT(collect.free_list), + }; + + iommu_pages_list_add(&collect.free_list, range.top_table); + pt_walk_range(&range, __collect_tables, &collect); + + /* + * The driver has to already have fenced the HW access to the page table + * and invalidated any caching referring to this memory. + */ + iommu_put_pages_list(&collect.free_list); +} + +static const struct pt_iommu_ops NS(ops) = { + .get_info = NS(get_info), + .deinit = NS(deinit), +}; + +static int pt_init_common(struct pt_common *common) +{ + struct pt_range top_range = pt_top_range(common); + + if (PT_WARN_ON(top_range.top_level > PT_MAX_TOP_LEVEL)) + return -EINVAL; + + if (top_range.top_level == PT_MAX_TOP_LEVEL || + common->max_vasz_lg2 == top_range.max_vasz_lg2) + common->features &= ~BIT(PT_FEAT_DYNAMIC_TOP); + + if (top_range.max_vasz_lg2 == PT_VADDR_MAX_LG2) + common->features |= BIT(PT_FEAT_FULL_VA); + + /* Requested features must match features compiled into this format */ + if ((common->features & ~(unsigned int)PT_SUPPORTED_FEATURES) || + (!IS_ENABLED(CONFIG_DEBUG_GENERIC_PT) && + (common->features & PT_FORCE_ENABLED_FEATURES) != + PT_FORCE_ENABLED_FEATURES)) + return -EOPNOTSUPP; + + if (common->max_oasz_lg2 == 0) + common->max_oasz_lg2 = pt_max_oa_lg2(common); + else + common->max_oasz_lg2 = min(common->max_oasz_lg2, + pt_max_oa_lg2(common)); + return 0; +} + +static int pt_iommu_init_domain(struct pt_iommu *iommu_table, + struct iommu_domain *domain) +{ + struct pt_common *common = common_from_iommu(iommu_table); + struct pt_iommu_info info; + struct pt_range range; + + NS(get_info)(iommu_table, &info); + + domain->type = __IOMMU_DOMAIN_PAGING; + domain->pgsize_bitmap = info.pgsize_bitmap; + + if (pt_feature(common, PT_FEAT_DYNAMIC_TOP)) + range = _pt_top_range(common, + _pt_top_set(NULL, PT_MAX_TOP_LEVEL)); + else + range = pt_top_range(common); + + /* A 64-bit high address space table on a 32-bit system cannot work. */ + domain->geometry.aperture_start = (unsigned long)range.va; + if ((pt_vaddr_t)domain->geometry.aperture_start != range.va) + return -EOVERFLOW; + + /* + * The aperture is limited to what the API can do after considering all + * the different types dma_addr_t/unsigned long/pt_vaddr_t that are used + * to store a VA. Set the aperture to something that is valid for all + * cases. Saturate instead of truncate the end if the types are smaller + * than the top range. aperture_end should be called aperture_last. + */ + domain->geometry.aperture_end = (unsigned long)range.last_va; + if ((pt_vaddr_t)domain->geometry.aperture_end != range.last_va) { + domain->geometry.aperture_end = ULONG_MAX; + domain->pgsize_bitmap &= ULONG_MAX; + } + domain->geometry.force_aperture = true; + + return 0; +} + +static void pt_iommu_zero(struct pt_iommu_table *fmt_table) +{ + struct pt_iommu *iommu_table = &fmt_table->iommu; + struct pt_iommu cfg = *iommu_table; + + static_assert(offsetof(struct pt_iommu_table, iommu.domain) == 0); + memset_after(fmt_table, 0, iommu.domain); + + /* The caller can initialize some of these values */ + iommu_table->nid = cfg.nid; +} + +#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) +{ + struct pt_iommu *iommu_table = &fmt_table->iommu; + struct pt_common *common = common_from_iommu(iommu_table); + struct pt_table_p *table_mem; + int ret; + + if (cfg->common.hw_max_vasz_lg2 > PT_MAX_VA_ADDRESS_LG2 || + !cfg->common.hw_max_vasz_lg2 || !cfg->common.hw_max_oasz_lg2) + return -EINVAL; + + pt_iommu_zero(fmt_table); + common->features = cfg->common.features; + common->max_vasz_lg2 = cfg->common.hw_max_vasz_lg2; + common->max_oasz_lg2 = cfg->common.hw_max_oasz_lg2; + ret = pt_iommu_fmt_init(fmt_table, cfg); + if (ret) + return ret; + + if (cfg->common.hw_max_oasz_lg2 > pt_max_oa_lg2(common)) + return -EINVAL; + + ret = pt_init_common(common); + if (ret) + return ret; + + if (pt_feature(common, PT_FEAT_SIGN_EXTEND) && + (pt_feature(common, PT_FEAT_FULL_VA) || + pt_feature(common, PT_FEAT_DYNAMIC_TOP))) + return -EINVAL; + + ret = pt_iommu_init_domain(iommu_table, &iommu_table->domain); + if (ret) + return ret; + + table_mem = table_alloc_top(common, common->top_of_table, gfp); + if (IS_ERR(table_mem)) + return PTR_ERR(table_mem); + pt_top_set(common, table_mem, pt_top_get_level(common)); + + /* Must be last, see pt_iommu_deinit() */ + iommu_table->ops = &NS(ops); + return 0; +} +EXPORT_SYMBOL_NS_GPL(pt_iommu_init, "GENERIC_PT_IOMMU"); + +#ifdef pt_iommu_fmt_hw_info +#define pt_iommu_table_hw_info CONCATENATE(pt_iommu_table, _hw_info) +#define pt_iommu_hw_info CONCATENATE(CONCATENATE(pt_iommu_, PTPFX), hw_info) +void pt_iommu_hw_info(struct pt_iommu_table *fmt_table, + struct pt_iommu_table_hw_info *info) +{ + struct pt_iommu *iommu_table = &fmt_table->iommu; + struct pt_common *common = common_from_iommu(iommu_table); + struct pt_range top_range = pt_top_range(common); + + pt_iommu_fmt_hw_info(fmt_table, &top_range, info); +} +EXPORT_SYMBOL_NS_GPL(pt_iommu_hw_info, "GENERIC_PT_IOMMU"); +#endif + +MODULE_LICENSE("GPL"); +MODULE_DESCRIPTION("IOMMU Page table implementation for " __stringify(PTPFX_RAW)); +MODULE_IMPORT_NS("GENERIC_PT"); + +#endif /* __GENERIC_PT_IOMMU_PT_H */ diff --git a/include/linux/generic_pt/iommu.h b/include/linux/generic_pt/iommu.h new file mode 100644 index 00000000000000..defa96abc49781 --- /dev/null +++ b/include/linux/generic_pt/iommu.h @@ -0,0 +1,150 @@ +/* SPDX-License-Identifier: GPL-2.0-only */ +/* + * Copyright (c) 2024-2025, NVIDIA CORPORATION & AFFILIATES + */ +#ifndef __GENERIC_PT_IOMMU_H +#define __GENERIC_PT_IOMMU_H + +#include <linux/generic_pt/common.h> +#include <linux/iommu.h> +#include <linux/mm_types.h> + +struct pt_iommu_ops; + +/** + * DOC: IOMMU Radix Page Table + * + * The IOMMU implementation of the Generic Page Table provides an ops struct + * that is useful to go with an iommu_domain to serve the DMA API, IOMMUFD and + * the generic map/unmap interface. + * + * This interface uses a caller provided locking approach. The caller must have + * a VA range lock concept that prevents concurrent threads from calling ops on + * the same VA. Generally the range lock must be at least as large as a single + * map call. + */ + +/** + * struct pt_iommu - Base structure for IOMMU page tables + * + * The format-specific struct will include this as the first member. + */ +struct pt_iommu { + /** + * @domain: The core IOMMU domain. The driver should use a union to + * overlay this memory with its previously existing domain struct to + * create an alias. + */ + struct iommu_domain domain; + + /** + * @ops: Function pointers to access the API + */ + const struct pt_iommu_ops *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 + * table walkers. + */ + int nid; +}; + +/** + * struct pt_iommu_info - Details about the IOMMU page table + * + * Returned from pt_iommu_ops->get_info() + */ +struct pt_iommu_info { + /** + * @pgsize_bitmap: A bitmask where each set bit indicates + * a page size that can be natively stored in the page table. + */ + u64 pgsize_bitmap; +}; + +struct pt_iommu_ops { + /** + * @get_info: Return the pt_iommu_info structure + * @iommu_table: Table to query + * + * Return some basic static information about the page table. + */ + void (*get_info)(struct pt_iommu *iommu_table, + struct pt_iommu_info *info); + + /** + * @deinit: Undo a format specific init operation + * @iommu_table: Table to destroy + * + * Release all of the memory. The caller must have already removed the + * table from all HW access and all caches. + */ + void (*deinit)(struct pt_iommu *iommu_table); +}; + +static inline void pt_iommu_deinit(struct pt_iommu *iommu_table) +{ + /* + * It is safe to call pt_iommu_deinit() before an init, or if init + * fails. The ops pointer will only become non-NULL if deinit needs to be + * run. + */ + if (iommu_table->ops) + iommu_table->ops->deinit(iommu_table); +} + +/** + * struct pt_iommu_cfg - Common configuration values for all formats + */ +struct pt_iommu_cfg { + /** + * @features: Features required. Only these features will be turned on. + * The feature list should reflect what the IOMMU HW is capable of. + */ + unsigned int features; + /** + * @hw_max_vasz_lg2: Maximum VA the IOMMU HW can support. This will + * imply the top level of the table. + */ + u8 hw_max_vasz_lg2; + /** + * @hw_max_oasz_lg2: Maximum OA the IOMMU HW can support. The format + * might select a lower maximum OA. + */ + u8 hw_max_oasz_lg2; +}; + +/* Generate the exported function signatures from iommu_pt.h */ +#define IOMMU_PROTOTYPES(fmt) \ + int pt_iommu_##fmt##_init(struct pt_iommu_##fmt *table, \ + const struct pt_iommu_##fmt##_cfg *cfg, \ + gfp_t gfp); \ + void pt_iommu_##fmt##_hw_info(struct pt_iommu_##fmt *table, \ + struct pt_iommu_##fmt##_hw_info *info) +#define IOMMU_FORMAT(fmt, member) \ + struct pt_iommu_##fmt { \ + struct pt_iommu iommu; \ + struct pt_##fmt member; \ + }; \ + IOMMU_PROTOTYPES(fmt) + +/* + * The driver should setup its domain struct like + * union { + * struct iommu_domain domain; + * struct pt_iommu_xxx xx; + * }; + * PT_IOMMU_CHECK_DOMAIN(struct mock_iommu_domain, xx.iommu, domain); + * + * Which creates an alias between driver_domain.domain and + * driver_domain.xx.iommu.domain. This is to avoid a mass rename of existing + * driver_domain.domain users. + */ +#define PT_IOMMU_CHECK_DOMAIN(s, pt_iommu_memb, domain_memb) \ + static_assert(offsetof(s, pt_iommu_memb.domain) == \ + offsetof(s, domain_memb)) + +#undef IOMMU_PROTOTYPES +#undef IOMMU_FORMAT +#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.

Reviewed-by: Kevin Tian kevin.tian@intel.com Reviewed-by: Pasha Tatashin pasha.tatashin@soleen.com Reviewed-by: Samiullah Khawaja skhawaja@google.com Tested-by: Alejandro Jimenez alejandro.j.jimenez@oracle.com Tested-by: Pasha Tatashin pasha.tatashin@soleen.com Signed-off-by: Jason Gunthorpe jgg@nvidia.com --- drivers/iommu/generic_pt/iommu_pt.h | 156 ++++++++++++++++++++++++++++ include/linux/generic_pt/iommu.h | 10 +- 2 files changed, 164 insertions(+), 2 deletions(-)

diff --git a/drivers/iommu/generic_pt/iommu_pt.h b/drivers/iommu/generic_pt/iommu_pt.h index 5ff1b887928a46..e3d1b272723db0 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)

@@ -164,6 +187,139 @@ 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; + /* Micro optimization */ + 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_entry_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_entries(&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_entries(&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

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.

Reviewed-by: Kevin Tian kevin.tian@intel.com Reviewed-by: Samiullah Khawaja skhawaja@google.com Tested-by: Alejandro Jimenez alejandro.j.jimenez@oracle.com Tested-by: Pasha Tatashin pasha.tatashin@soleen.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 f32e81509f4f09..448c5796d4a861 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_make_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_is_write_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_is_write_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; /* Fail if any OAs are within the range */ @@ -1015,5 +1117,7 @@ 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 /* __GENERIC_PT_IOMMU_PT_H */ diff --git a/include/linux/generic_pt/iommu.h b/include/linux/generic_pt/iommu.h index 0d59423024d57f..03a906fbe12a83 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_driver_ops; +struct iommu_dirty_bitmap;

/** * DOC: IOMMU Radix Page Table @@ -182,6 +183,9 @@ 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); \ @@ -202,6 +206,8 @@ 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

Reviewed-by: Kevin Tian kevin.tian@intel.com Tested-by: Alejandro Jimenez alejandro.j.jimenez@oracle.com Tested-by: Pasha Tatashin pasha.tatashin@soleen.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 | 713 ++++++++++++++++++ drivers/iommu/generic_pt/kunit_iommu.h | 181 +++++ 7 files changed, 948 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 cbdad222923b26..81652cd9c69fe8 100644 --- a/drivers/iommu/generic_pt/Kconfig +++ b/drivers/iommu/generic_pt/Kconfig @@ -41,5 +41,17 @@ 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 7423ed71417dff..aaf76bfd21dadc 100644 --- a/drivers/iommu/generic_pt/fmt/amdv1.h +++ b/drivers/iommu/generic_pt/fmt/amdv1.h @@ -384,4 +384,14 @@ 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..ae8c160966302c --- /dev/null +++ b/drivers/iommu/generic_pt/kunit_generic_pt.h @@ -0,0 +1,713 @@ +/* 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, fls_t(u32, 0), 0); + KUNIT_ASSERT_EQ(test, fls_t(u32, 1), 1); + KUNIT_ASSERT_EQ(test, fls_t(u32, BIT(2)), 3); + KUNIT_ASSERT_EQ(test, fls_t(u32, U32_MAX), 32); + + KUNIT_ASSERT_EQ(test, fls_t(u64, 0), 0); + KUNIT_ASSERT_EQ(test, fls_t(u64, 1), 1); + KUNIT_ASSERT_EQ(test, fls_t(u64, BIT(2)), 3); + KUNIT_ASSERT_EQ(test, fls_t(u64, U64_MAX), 64); + + KUNIT_ASSERT_EQ(test, ffs_t(u32, 1), 0); + KUNIT_ASSERT_EQ(test, ffs_t(u32, BIT(2)), 2); + KUNIT_ASSERT_EQ(test, ffs_t(u32, BIT(31)), 31); + + KUNIT_ASSERT_EQ(test, ffs_t(u64, 1), 0); + KUNIT_ASSERT_EQ(test, ffs_t(u64, BIT(2)), 2); + KUNIT_ASSERT_EQ(test, ffs_t(u64, BIT_ULL(63)), 63); + + for (i = 0; i != 31; i++) + KUNIT_ASSERT_EQ(test, ffz_t(u64, BIT_ULL(i) - 1), i); + + for (i = 0; i != 63; i++) + KUNIT_ASSERT_EQ(test, 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, ffs_t(u32, val)), 0); + KUNIT_ASSERT_EQ(test, log2_mod_t(u64, val, ffs_t(u64, val)), 0); + + KUNIT_ASSERT_EQ(test, log2_mod_t(u32, val, ffz_t(u32, val)), + log2_to_max_int_t(u32, ffz_t(u32, val))); + KUNIT_ASSERT_EQ(test, log2_mod_t(u64, val, ffz_t(u64, val)), + log2_to_max_int_t(u64, 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_entries(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_entries(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_entries() 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_entries(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_entries(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_make_write_clean(pts); + pt_load_entry(pts); + KUNIT_ASSERT_FALSE(test, pt_entry_is_write_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_is_write_dirty(pts)); + + pt_entry_make_write_clean(pts); + pt_load_entry(pts); + KUNIT_ASSERT_FALSE(test, pt_entry_is_write_dirty(pts)); + } + + pt_clear_entries(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_is_write_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..28ec313f151e70 --- /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(struct kunit *test, 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_driver_ops pt_kunit_driver_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_oa_lg2(NULL); + + priv->fmt_table.iommu.nid = NUMA_NO_NODE; + priv->fmt_table.iommu.driver_ops = &pt_kunit_driver_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 = vaffs(priv->info.pgsize_bitmap); + priv->smallest_pgsz = log2_to_int(priv->smallest_pgsz_lg2); + priv->largest_pgsz_lg2 = + vafls((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

The iommufd self test uses an xarray to store the pfns and their orders to emulate a page table. Slightly modify the amdv1 page table to create a real page table that has similar properties:

- 2k base granule to simulate something like a 4k page table on a 64K PAGE_SIZE ARM system - Contiguous page support for every PFN order - Dirty tracking

AMDv1 is the closest format, as it is the only one that already supports every page size. Tweak it to have only 5 levels and an 11 bit base granule and compile it separately as a format variant.

Reviewed-by: Kevin Tian kevin.tian@intel.com Reviewed-by: Samiullah Khawaja skhawaja@google.com Tested-by: Alejandro Jimenez alejandro.j.jimenez@oracle.com Tested-by: Pasha Tatashin pasha.tatashin@soleen.com Signed-off-by: Jason Gunthorpe jgg@nvidia.com --- drivers/iommu/generic_pt/fmt/Makefile | 1 + drivers/iommu/generic_pt/fmt/amdv1.h | 18 ++++++++++++++++-- drivers/iommu/generic_pt/fmt/iommu_mock.c | 10 ++++++++++ include/linux/generic_pt/iommu.h | 6 ++++++ 4 files changed, 33 insertions(+), 2 deletions(-) create mode 100644 drivers/iommu/generic_pt/fmt/iommu_mock.c

diff --git a/drivers/iommu/generic_pt/fmt/Makefile b/drivers/iommu/generic_pt/fmt/Makefile index 32f3956c7509f8..f0c22cf5f7bee6 100644 --- a/drivers/iommu/generic_pt/fmt/Makefile +++ b/drivers/iommu/generic_pt/fmt/Makefile @@ -1,6 +1,7 @@ # SPDX-License-Identifier: GPL-2.0

iommu_pt_fmt-$(CONFIG_IOMMU_PT_AMDV1) += amdv1 +iommu_pt_fmt-$(CONFIG_IOMMUFD_TEST) += mock

IOMMU_PT_KUNIT_TEST := define create_format diff --git a/drivers/iommu/generic_pt/fmt/amdv1.h b/drivers/iommu/generic_pt/fmt/amdv1.h index aaf76bfd21dadc..aa8e1a8ec95fde 100644 --- a/drivers/iommu/generic_pt/fmt/amdv1.h +++ b/drivers/iommu/generic_pt/fmt/amdv1.h @@ -26,11 +26,23 @@ #include <linux/string.h>

enum { - PT_MAX_OUTPUT_ADDRESS_LG2 = 52, - PT_MAX_VA_ADDRESS_LG2 = 64, PT_ITEM_WORD_SIZE = sizeof(u64), + /* + * The IOMMUFD selftest uses the AMDv1 format with some alterations It + * uses a 2k page size to test cases where the CPU page size is not the + * same. + */ +#ifdef AMDV1_IOMMUFD_SELFTEST + PT_MAX_VA_ADDRESS_LG2 = 56, + PT_MAX_OUTPUT_ADDRESS_LG2 = 51, + PT_MAX_TOP_LEVEL = 4, + PT_GRANULE_LG2SZ = 11, +#else + PT_MAX_VA_ADDRESS_LG2 = 64, + PT_MAX_OUTPUT_ADDRESS_LG2 = 52, PT_MAX_TOP_LEVEL = 5, PT_GRANULE_LG2SZ = 12, +#endif PT_TABLEMEM_LG2SZ = 12,

/* The DTE only has these bits for the top phyiscal address */ @@ -374,6 +386,7 @@ static inline int amdv1pt_iommu_fmt_init(struct pt_iommu_amdv1 *iommu_table, } #define pt_iommu_fmt_init amdv1pt_iommu_fmt_init

+#ifndef PT_FMT_VARIANT static inline void amdv1pt_iommu_fmt_hw_info(struct pt_iommu_amdv1 *table, const struct pt_range *top_range, @@ -384,6 +397,7 @@ 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 +#endif

#if defined(GENERIC_PT_KUNIT) static const struct pt_iommu_amdv1_cfg amdv1_kunit_fmt_cfgs[] = { diff --git a/drivers/iommu/generic_pt/fmt/iommu_mock.c b/drivers/iommu/generic_pt/fmt/iommu_mock.c new file mode 100644 index 00000000000000..74e597cba9d9cd --- /dev/null +++ b/drivers/iommu/generic_pt/fmt/iommu_mock.c @@ -0,0 +1,10 @@ +// SPDX-License-Identifier: GPL-2.0-only +/* + * Copyright (c) 2024-2025, NVIDIA CORPORATION & AFFILIATES + */ +#define AMDV1_IOMMUFD_SELFTEST 1 +#define PT_FMT amdv1 +#define PT_FMT_VARIANT mock +#define PT_SUPPORTED_FEATURES 0 + +#include "iommu_template.h" diff --git a/include/linux/generic_pt/iommu.h b/include/linux/generic_pt/iommu.h index 03a906fbe12a83..848a5fb76272a9 100644 --- a/include/linux/generic_pt/iommu.h +++ b/include/linux/generic_pt/iommu.h @@ -237,6 +237,12 @@ struct pt_iommu_amdv1_hw_info {

IOMMU_FORMAT(amdv1, amdpt);

+/* amdv1_mock is used by the iommufd selftest */ +#define pt_iommu_amdv1_mock pt_iommu_amdv1 +#define pt_iommu_amdv1_mock_cfg pt_iommu_amdv1_cfg +struct pt_iommu_amdv1_mock_hw_info; +IOMMU_PROTOTYPES(amdv1_mock); + #undef IOMMU_PROTOTYPES #undef IOMMU_FORMAT #endif

This is used by x86 CPUs and can be used in AMD/VT-d x86 IOMMUs. When a x86 IOMMU is running SVA the MM will be using this 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%.

Reviewed-by: Kevin Tian kevin.tian@intel.com Reviewed-by: Vasant Hegde vasant.hegde@amd.com Tested-by: Alejandro Jimenez alejandro.j.jimenez@oracle.com Tested-by: Pasha Tatashin pasha.tatashin@soleen.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 | 255 ++++++++++++++++++++ include/linux/generic_pt/common.h | 13 + include/linux/generic_pt/iommu.h | 11 + 8 files changed, 325 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 81652cd9c69fe8..6dcb771b3c582a 100644 --- a/drivers/iommu/generic_pt/Kconfig +++ b/drivers/iommu/generic_pt/Kconfig @@ -42,10 +42,21 @@ 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..18d736d14b2d0d --- /dev/null +++ b/drivers/iommu/generic_pt/fmt/x86_64.h @@ -0,0 +1,255 @@ +/* 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), +}; + +static inline pt_oaddr_t x86_64_pt_table_pa(const struct pt_state *pts) +{ + u64 entry = pts->entry; + + if (pts_feature(pts, PT_FEAT_X86_64_AMD_ENCRYPT_TABLES)) + entry = __sme_clr(entry); + return oalog2_mul(FIELD_GET(X86_64_FMT_OA, 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) +{ + u64 entry = pts->entry; + + if (pts_feature(pts, PT_FEAT_X86_64_AMD_ENCRYPT_TABLES)) + entry = __sme_clr(entry); + return oalog2_mul(FIELD_GET(X86_64_FMT_OA, 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; + + if (!pt_check_install_leaf_args(pts, oa, oasz_lg2)) + return; + + 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 21e33489cbf20d..96f8a6a7d60e10 100644 --- a/include/linux/generic_pt/common.h +++ b/include/linux/generic_pt/common.h @@ -151,4 +151,17 @@ 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 f2a763aba08846..fde7ccf007c50c 100644 --- a/include/linux/generic_pt/iommu.h +++ b/include/linux/generic_pt/iommu.h @@ -255,6 +255,17 @@ 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

None of this is used anymore, delete it.

Reviewed-by: Alejandro Jimenez alejandro.j.jimenez@oracle.com Reviewed-by: Vasant Hegde vasant.hegde@amd.com Tested-by: Alejandro Jimenez alejandro.j.jimenez@oracle.com Tested-by: Pasha Tatashin pasha.tatashin@soleen.com Signed-off-by: Jason Gunthorpe jgg@nvidia.com --- drivers/iommu/amd/Makefile | 2 +- drivers/iommu/amd/amd_iommu_types.h | 98 ----- drivers/iommu/amd/io_pgtable.c | 575 ---------------------------- drivers/iommu/amd/io_pgtable_v2.c | 370 ------------------ drivers/iommu/io-pgtable.c | 4 - include/linux/io-pgtable.h | 2 - 6 files changed, 1 insertion(+), 1050 deletions(-) delete mode 100644 drivers/iommu/amd/io_pgtable.c delete mode 100644 drivers/iommu/amd/io_pgtable_v2.c

diff --git a/drivers/iommu/amd/Makefile b/drivers/iommu/amd/Makefile index 59c04a67f39825..5412a563c6979c 100644 --- a/drivers/iommu/amd/Makefile +++ b/drivers/iommu/amd/Makefile @@ -1,3 +1,3 @@ # SPDX-License-Identifier: GPL-2.0-only -obj-y += iommu.o init.o quirks.o io_pgtable.o io_pgtable_v2.o ppr.o pasid.o +obj-y += iommu.o init.o quirks.o ppr.o pasid.o obj-$(CONFIG_AMD_IOMMU_DEBUGFS) += debugfs.o diff --git a/drivers/iommu/amd/amd_iommu_types.h b/drivers/iommu/amd/amd_iommu_types.h index d90a285b44eb3a..4b4a37fad70ef2 100644 --- a/drivers/iommu/amd/amd_iommu_types.h +++ b/drivers/iommu/amd/amd_iommu_types.h @@ -18,7 +18,6 @@ #include <linux/spinlock.h> #include <linux/pci.h> #include <linux/irqreturn.h> -#include <linux/io-pgtable.h> #include <linux/generic_pt/iommu.h>

/* @@ -338,76 +337,7 @@ #define GUEST_PGTABLE_4_LEVEL 0x00 #define GUEST_PGTABLE_5_LEVEL 0x01

-#define PM_LEVEL_SHIFT(x) (12 + ((x) * 9)) -#define PM_LEVEL_SIZE(x) (((x) < 6) ? \ - ((1ULL << PM_LEVEL_SHIFT((x))) - 1): \ - (0xffffffffffffffffULL)) -#define PM_LEVEL_INDEX(x, a) (((a) >> PM_LEVEL_SHIFT((x))) & 0x1ffULL) -#define PM_LEVEL_ENC(x) (((x) << 9) & 0xe00ULL) -#define PM_LEVEL_PDE(x, a) ((a) | PM_LEVEL_ENC((x)) | \ - IOMMU_PTE_PR | IOMMU_PTE_IR | IOMMU_PTE_IW) -#define PM_PTE_LEVEL(pte) (((pte) >> 9) & 0x7ULL) - -#define PM_MAP_4k 0 #define PM_ADDR_MASK 0x000ffffffffff000ULL -#define PM_MAP_MASK(lvl) (PM_ADDR_MASK & \ - (~((1ULL << (12 + ((lvl) * 9))) - 1))) -#define PM_ALIGNED(lvl, addr) ((PM_MAP_MASK(lvl) & (addr)) == (addr)) - -/* - * Returns the page table level to use for a given page size - * Pagesize is expected to be a power-of-two - */ -#define PAGE_SIZE_LEVEL(pagesize) \ - ((__ffs(pagesize) - 12) / 9) -/* - * Returns the number of ptes to use for a given page size - * Pagesize is expected to be a power-of-two - */ -#define PAGE_SIZE_PTE_COUNT(pagesize) \ - (1ULL << ((__ffs(pagesize) - 12) % 9)) - -/* - * Aligns a given io-virtual address to a given page size - * Pagesize is expected to be a power-of-two - */ -#define PAGE_SIZE_ALIGN(address, pagesize) \ - ((address) & ~((pagesize) - 1)) -/* - * Creates an IOMMU PTE for an address and a given pagesize - * The PTE has no permission bits set - * Pagesize is expected to be a power-of-two larger than 4096 - */ -#define PAGE_SIZE_PTE(address, pagesize) \ - (((address) | ((pagesize) - 1)) & \ - (~(pagesize >> 1)) & PM_ADDR_MASK) - -/* - * Takes a PTE value with mode=0x07 and returns the page size it maps - */ -#define PTE_PAGE_SIZE(pte) \ - (1ULL << (1 + ffz(((pte) | 0xfffULL)))) - -/* - * Takes a page-table level and returns the default page-size for this level - */ -#define PTE_LEVEL_PAGE_SIZE(level) \ - (1ULL << (12 + (9 * (level)))) - -/* - * The IOPTE dirty bit - */ -#define IOMMU_PTE_HD_BIT (6) - -/* - * Bit value definition for I/O PTE fields - */ -#define IOMMU_PTE_PR BIT_ULL(0) -#define IOMMU_PTE_HD BIT_ULL(IOMMU_PTE_HD_BIT) -#define IOMMU_PTE_U BIT_ULL(59) -#define IOMMU_PTE_FC BIT_ULL(60) -#define IOMMU_PTE_IR BIT_ULL(61) -#define IOMMU_PTE_IW BIT_ULL(62)

/* * Bit value definition for DTE fields @@ -437,12 +367,6 @@ /* DTE[128:179] | DTE[184:191] */ #define DTE_DATA2_INTR_MASK ~GENMASK_ULL(55, 52)

-#define IOMMU_PAGE_MASK (((1ULL << 52) - 1) & ~0xfffULL) -#define IOMMU_PTE_PRESENT(pte) ((pte) & IOMMU_PTE_PR) -#define IOMMU_PTE_DIRTY(pte) ((pte) & IOMMU_PTE_HD) -#define IOMMU_PTE_PAGE(pte) (iommu_phys_to_virt((pte) & IOMMU_PAGE_MASK)) -#define IOMMU_PTE_MODE(pte) (((pte) >> 9) & 0x07) - #define IOMMU_PROT_MASK 0x03 #define IOMMU_PROT_IR 0x01 #define IOMMU_PROT_IW 0x02 @@ -535,19 +459,6 @@ struct amd_irte_ops;

#define AMD_IOMMU_FLAG_TRANS_PRE_ENABLED (1 << 0)

-#define io_pgtable_to_data(x) \ - container_of((x), struct amd_io_pgtable, pgtbl) - -#define io_pgtable_ops_to_data(x) \ - io_pgtable_to_data(io_pgtable_ops_to_pgtable(x)) - -#define io_pgtable_ops_to_domain(x) \ - container_of(io_pgtable_ops_to_data(x), \ - struct protection_domain, iop) - -#define io_pgtable_cfg_to_data(x) \ - container_of((x), struct amd_io_pgtable, pgtbl.cfg) - struct gcr3_tbl_info { u64 *gcr3_tbl; /* Guest CR3 table */ int glx; /* Number of levels for GCR3 table */ @@ -555,14 +466,6 @@ struct gcr3_tbl_info { u16 domid; /* Per device domain ID */ };

-struct amd_io_pgtable { - seqcount_t seqcount; /* Protects root/mode update */ - struct io_pgtable pgtbl; - int mode; - u64 *root; - u64 *pgd; /* v2 pgtable pgd pointer */ -}; - enum protection_domain_mode { PD_MODE_NONE, PD_MODE_V1, @@ -597,7 +500,6 @@ struct protection_domain { struct pt_iommu_x86_64 amdv2; }; struct list_head dev_list; /* List of all devices in this domain */ - 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 */ enum protection_domain_mode pd_mode; /* Track page table type */ diff --git a/drivers/iommu/amd/io_pgtable.c b/drivers/iommu/amd/io_pgtable.c deleted file mode 100644 index f64244938c9af7..00000000000000 --- a/drivers/iommu/amd/io_pgtable.c +++ /dev/null @@ -1,575 +0,0 @@ -// SPDX-License-Identifier: GPL-2.0-only -/* - * CPU-agnostic AMD IO page table allocator. - * - * Copyright (C) 2020 Advanced Micro Devices, Inc. - * Author: Suravee Suthikulpanit suravee.suthikulpanit@amd.com - */ - -#define pr_fmt(fmt) "AMD-Vi: " fmt -#define dev_fmt(fmt) pr_fmt(fmt) - -#include <linux/atomic.h> -#include <linux/bitops.h> -#include <linux/io-pgtable.h> -#include <linux/kernel.h> -#include <linux/sizes.h> -#include <linux/slab.h> -#include <linux/types.h> -#include <linux/dma-mapping.h> -#include <linux/seqlock.h> - -#include <asm/barrier.h> - -#include "amd_iommu_types.h" -#include "amd_iommu.h" -#include "../iommu-pages.h" - -/* - * Helper function to get the first pte of a large mapping - */ -static u64 *first_pte_l7(u64 *pte, unsigned long *page_size, - unsigned long *count) -{ - unsigned long pte_mask, pg_size, cnt; - u64 *fpte; - - pg_size = PTE_PAGE_SIZE(*pte); - cnt = PAGE_SIZE_PTE_COUNT(pg_size); - pte_mask = ~((cnt << 3) - 1); - fpte = (u64 *)(((unsigned long)pte) & pte_mask); - - if (page_size) - *page_size = pg_size; - - if (count) - *count = cnt; - - return fpte; -} - -static void free_pt_lvl(u64 *pt, struct iommu_pages_list *freelist, int lvl) -{ - u64 *p; - int i; - - for (i = 0; i < 512; ++i) { - /* PTE present? */ - if (!IOMMU_PTE_PRESENT(pt[i])) - continue; - - /* Large PTE? */ - if (PM_PTE_LEVEL(pt[i]) == 0 || - PM_PTE_LEVEL(pt[i]) == 7) - continue; - - /* - * Free the next level. No need to look at l1 tables here since - * they can only contain leaf PTEs; just free them directly. - */ - p = IOMMU_PTE_PAGE(pt[i]); - if (lvl > 2) - free_pt_lvl(p, freelist, lvl - 1); - else - iommu_pages_list_add(freelist, p); - } - - iommu_pages_list_add(freelist, pt); -} - -static void free_sub_pt(u64 *root, int mode, struct iommu_pages_list *freelist) -{ - switch (mode) { - case PAGE_MODE_NONE: - case PAGE_MODE_7_LEVEL: - break; - case PAGE_MODE_1_LEVEL: - iommu_pages_list_add(freelist, root); - break; - case PAGE_MODE_2_LEVEL: - case PAGE_MODE_3_LEVEL: - case PAGE_MODE_4_LEVEL: - case PAGE_MODE_5_LEVEL: - case PAGE_MODE_6_LEVEL: - free_pt_lvl(root, freelist, mode); - break; - default: - BUG(); - } -} - -/* - * This function is used to add another level to an IO page table. Adding - * another level increases the size of the address space by 9 bits to a size up - * to 64 bits. - */ -static bool increase_address_space(struct amd_io_pgtable *pgtable, - unsigned long address, - unsigned int page_size_level, - gfp_t gfp) -{ - struct io_pgtable_cfg *cfg = &pgtable->pgtbl.cfg; - struct protection_domain *domain = - container_of(pgtable, struct protection_domain, iop); - unsigned long flags; - bool ret = true; - u64 *pte; - - pte = iommu_alloc_pages_node_sz(cfg->amd.nid, gfp, SZ_4K); - if (!pte) - return false; - - spin_lock_irqsave(&domain->lock, flags); - - if (address <= PM_LEVEL_SIZE(pgtable->mode) && - pgtable->mode - 1 >= page_size_level) - goto out; - - ret = false; - if (WARN_ON_ONCE(pgtable->mode == amd_iommu_hpt_level)) - goto out; - - *pte = PM_LEVEL_PDE(pgtable->mode, iommu_virt_to_phys(pgtable->root)); - - write_seqcount_begin(&pgtable->seqcount); - pgtable->root = pte; - pgtable->mode += 1; - write_seqcount_end(&pgtable->seqcount); - - pte = NULL; - ret = true; - -out: - spin_unlock_irqrestore(&domain->lock, flags); - iommu_free_pages(pte); - - return ret; -} - -static u64 *alloc_pte(struct amd_io_pgtable *pgtable, - unsigned long address, - unsigned long page_size, - u64 **pte_page, - gfp_t gfp, - bool *updated) -{ - unsigned long last_addr = address + (page_size - 1); - struct io_pgtable_cfg *cfg = &pgtable->pgtbl.cfg; - unsigned int seqcount; - int level, end_lvl; - u64 *pte, *page; - - BUG_ON(!is_power_of_2(page_size)); - - while (last_addr > PM_LEVEL_SIZE(pgtable->mode) || - pgtable->mode - 1 < PAGE_SIZE_LEVEL(page_size)) { - /* - * Return an error if there is no memory to update the - * page-table. - */ - if (!increase_address_space(pgtable, last_addr, - PAGE_SIZE_LEVEL(page_size), gfp)) - return NULL; - } - - - do { - seqcount = read_seqcount_begin(&pgtable->seqcount); - - level = pgtable->mode - 1; - pte = &pgtable->root[PM_LEVEL_INDEX(level, address)]; - } while (read_seqcount_retry(&pgtable->seqcount, seqcount)); - - - address = PAGE_SIZE_ALIGN(address, page_size); - end_lvl = PAGE_SIZE_LEVEL(page_size); - - while (level > end_lvl) { - u64 __pte, __npte; - int pte_level; - - __pte = *pte; - pte_level = PM_PTE_LEVEL(__pte); - - /* - * If we replace a series of large PTEs, we need - * to tear down all of them. - */ - if (IOMMU_PTE_PRESENT(__pte) && - pte_level == PAGE_MODE_7_LEVEL) { - unsigned long count, i; - u64 *lpte; - - lpte = first_pte_l7(pte, NULL, &count); - - /* - * Unmap the replicated PTEs that still match the - * original large mapping - */ - for (i = 0; i < count; ++i) - cmpxchg64(&lpte[i], __pte, 0ULL); - - *updated = true; - continue; - } - - if (!IOMMU_PTE_PRESENT(__pte) || - pte_level == PAGE_MODE_NONE) { - page = iommu_alloc_pages_node_sz(cfg->amd.nid, gfp, - SZ_4K); - - if (!page) - return NULL; - - __npte = PM_LEVEL_PDE(level, iommu_virt_to_phys(page)); - - /* pte could have been changed somewhere. */ - if (!try_cmpxchg64(pte, &__pte, __npte)) - iommu_free_pages(page); - else if (IOMMU_PTE_PRESENT(__pte)) - *updated = true; - - continue; - } - - /* No level skipping support yet */ - if (pte_level != level) - return NULL; - - level -= 1; - - pte = IOMMU_PTE_PAGE(__pte); - - if (pte_page && level == end_lvl) - *pte_page = pte; - - pte = &pte[PM_LEVEL_INDEX(level, address)]; - } - - return pte; -} - -/* - * This function checks if there is a PTE for a given dma address. If - * there is one, it returns the pointer to it. - */ -static u64 *fetch_pte(struct amd_io_pgtable *pgtable, - unsigned long address, - unsigned long *page_size) -{ - int level; - unsigned int seqcount; - u64 *pte; - - *page_size = 0; - - if (address > PM_LEVEL_SIZE(pgtable->mode)) - return NULL; - - do { - seqcount = read_seqcount_begin(&pgtable->seqcount); - level = pgtable->mode - 1; - pte = &pgtable->root[PM_LEVEL_INDEX(level, address)]; - } while (read_seqcount_retry(&pgtable->seqcount, seqcount)); - - *page_size = PTE_LEVEL_PAGE_SIZE(level); - - while (level > 0) { - - /* Not Present */ - if (!IOMMU_PTE_PRESENT(*pte)) - return NULL; - - /* Large PTE */ - if (PM_PTE_LEVEL(*pte) == PAGE_MODE_7_LEVEL || - PM_PTE_LEVEL(*pte) == PAGE_MODE_NONE) - break; - - /* No level skipping support yet */ - if (PM_PTE_LEVEL(*pte) != level) - return NULL; - - level -= 1; - - /* Walk to the next level */ - pte = IOMMU_PTE_PAGE(*pte); - pte = &pte[PM_LEVEL_INDEX(level, address)]; - *page_size = PTE_LEVEL_PAGE_SIZE(level); - } - - /* - * If we have a series of large PTEs, make - * sure to return a pointer to the first one. - */ - if (PM_PTE_LEVEL(*pte) == PAGE_MODE_7_LEVEL) - pte = first_pte_l7(pte, page_size, NULL); - - return pte; -} - -static void free_clear_pte(u64 *pte, u64 pteval, - struct iommu_pages_list *freelist) -{ - u64 *pt; - int mode; - - while (!try_cmpxchg64(pte, &pteval, 0)) - pr_warn("AMD-Vi: IOMMU pte changed since we read it\n"); - - if (!IOMMU_PTE_PRESENT(pteval)) - return; - - pt = IOMMU_PTE_PAGE(pteval); - mode = IOMMU_PTE_MODE(pteval); - - free_sub_pt(pt, mode, freelist); -} - -/* - * Generic mapping functions. It maps a physical address into a DMA - * address space. It allocates the page table pages if necessary. - * In the future it can be extended to a generic mapping function - * supporting all features of AMD IOMMU page tables like level skipping - * and full 64 bit address spaces. - */ -static int iommu_v1_map_pages(struct io_pgtable_ops *ops, unsigned long iova, - phys_addr_t paddr, size_t pgsize, size_t pgcount, - int prot, gfp_t gfp, size_t *mapped) -{ - struct amd_io_pgtable *pgtable = io_pgtable_ops_to_data(ops); - struct iommu_pages_list freelist = IOMMU_PAGES_LIST_INIT(freelist); - bool updated = false; - u64 __pte, *pte; - int ret, i, count; - size_t size = pgcount << __ffs(pgsize); - unsigned long o_iova = iova; - - BUG_ON(!IS_ALIGNED(iova, pgsize)); - BUG_ON(!IS_ALIGNED(paddr, pgsize)); - - ret = -EINVAL; - if (!(prot & IOMMU_PROT_MASK)) - goto out; - - while (pgcount > 0) { - count = PAGE_SIZE_PTE_COUNT(pgsize); - pte = alloc_pte(pgtable, iova, pgsize, NULL, gfp, &updated); - - ret = -ENOMEM; - if (!pte) - goto out; - - for (i = 0; i < count; ++i) - free_clear_pte(&pte[i], pte[i], &freelist); - - if (!iommu_pages_list_empty(&freelist)) - updated = true; - - if (count > 1) { - __pte = PAGE_SIZE_PTE(__sme_set(paddr), pgsize); - __pte |= PM_LEVEL_ENC(7) | IOMMU_PTE_PR | IOMMU_PTE_FC; - } else - __pte = __sme_set(paddr) | IOMMU_PTE_PR | IOMMU_PTE_FC; - - if (prot & IOMMU_PROT_IR) - __pte |= IOMMU_PTE_IR; - if (prot & IOMMU_PROT_IW) - __pte |= IOMMU_PTE_IW; - - for (i = 0; i < count; ++i) - pte[i] = __pte; - - iova += pgsize; - paddr += pgsize; - pgcount--; - if (mapped) - *mapped += pgsize; - } - - ret = 0; - -out: - if (updated) { - struct protection_domain *dom = io_pgtable_ops_to_domain(ops); - unsigned long flags; - - spin_lock_irqsave(&dom->lock, flags); - /* - * Flush domain TLB(s) and wait for completion. Any Device-Table - * Updates and flushing already happened in - * increase_address_space(). - */ - amd_iommu_domain_flush_pages(dom, o_iova, size); - spin_unlock_irqrestore(&dom->lock, flags); - } - - /* Everything flushed out, free pages now */ - iommu_put_pages_list(&freelist); - - return ret; -} - -static unsigned long iommu_v1_unmap_pages(struct io_pgtable_ops *ops, - unsigned long iova, - size_t pgsize, size_t pgcount, - struct iommu_iotlb_gather *gather) -{ - struct amd_io_pgtable *pgtable = io_pgtable_ops_to_data(ops); - unsigned long long unmapped; - unsigned long unmap_size; - u64 *pte; - size_t size = pgcount << __ffs(pgsize); - - BUG_ON(!is_power_of_2(pgsize)); - - unmapped = 0; - - while (unmapped < size) { - pte = fetch_pte(pgtable, iova, &unmap_size); - if (pte) { - int i, count; - - count = PAGE_SIZE_PTE_COUNT(unmap_size); - for (i = 0; i < count; i++) - pte[i] = 0ULL; - } else { - return unmapped; - } - - iova = (iova & ~(unmap_size - 1)) + unmap_size; - unmapped += unmap_size; - } - - return unmapped; -} - -static phys_addr_t iommu_v1_iova_to_phys(struct io_pgtable_ops *ops, unsigned long iova) -{ - struct amd_io_pgtable *pgtable = io_pgtable_ops_to_data(ops); - unsigned long offset_mask, pte_pgsize; - u64 *pte, __pte; - - pte = fetch_pte(pgtable, iova, &pte_pgsize); - - if (!pte || !IOMMU_PTE_PRESENT(*pte)) - return 0; - - offset_mask = pte_pgsize - 1; - __pte = __sme_clr(*pte & PM_ADDR_MASK); - - return (__pte & ~offset_mask) | (iova & offset_mask); -} - -static bool pte_test_and_clear_dirty(u64 *ptep, unsigned long size, - unsigned long flags) -{ - bool test_only = flags & IOMMU_DIRTY_NO_CLEAR; - bool dirty = false; - int i, count; - - /* - * 2.2.3.2 Host Dirty Support - * When a non-default page size is used , software must OR the - * Dirty bits in all of the replicated host PTEs used to map - * the page. The IOMMU does not guarantee the Dirty bits are - * set in all of the replicated PTEs. Any portion of the page - * may have been written even if the Dirty bit is set in only - * one of the replicated PTEs. - */ - count = PAGE_SIZE_PTE_COUNT(size); - for (i = 0; i < count && test_only; i++) { - if (test_bit(IOMMU_PTE_HD_BIT, (unsigned long *)&ptep[i])) { - dirty = true; - break; - } - } - - for (i = 0; i < count && !test_only; i++) { - if (test_and_clear_bit(IOMMU_PTE_HD_BIT, - (unsigned long *)&ptep[i])) { - dirty = true; - } - } - - return dirty; -} - -static int iommu_v1_read_and_clear_dirty(struct io_pgtable_ops *ops, - unsigned long iova, size_t size, - unsigned long flags, - struct iommu_dirty_bitmap *dirty) -{ - struct amd_io_pgtable *pgtable = io_pgtable_ops_to_data(ops); - unsigned long end = iova + size - 1; - - do { - unsigned long pgsize = 0; - u64 *ptep, pte; - - ptep = fetch_pte(pgtable, iova, &pgsize); - if (ptep) - pte = READ_ONCE(*ptep); - if (!ptep || !IOMMU_PTE_PRESENT(pte)) { - pgsize = pgsize ?: PTE_LEVEL_PAGE_SIZE(0); - iova += pgsize; - continue; - } - - /* - * Mark the whole IOVA range as dirty even if only one of - * the replicated PTEs were marked dirty. - */ - if (pte_test_and_clear_dirty(ptep, pgsize, flags)) - iommu_dirty_bitmap_record(dirty, iova, pgsize); - iova += pgsize; - } while (iova < end); - - return 0; -} - -/* - * ---------------------------------------------------- - */ -static void v1_free_pgtable(struct io_pgtable *iop) -{ - struct amd_io_pgtable *pgtable = container_of(iop, struct amd_io_pgtable, pgtbl); - struct iommu_pages_list freelist = IOMMU_PAGES_LIST_INIT(freelist); - - if (pgtable->mode == PAGE_MODE_NONE) - return; - - /* Page-table is not visible to IOMMU anymore, so free it */ - BUG_ON(pgtable->mode < PAGE_MODE_NONE || - pgtable->mode > amd_iommu_hpt_level); - - free_sub_pt(pgtable->root, pgtable->mode, &freelist); - iommu_put_pages_list(&freelist); -} - -static struct io_pgtable *v1_alloc_pgtable(struct io_pgtable_cfg *cfg, void *cookie) -{ - struct amd_io_pgtable *pgtable = io_pgtable_cfg_to_data(cfg); - - pgtable->root = - iommu_alloc_pages_node_sz(cfg->amd.nid, GFP_KERNEL, SZ_4K); - if (!pgtable->root) - return NULL; - pgtable->mode = PAGE_MODE_3_LEVEL; - seqcount_init(&pgtable->seqcount); - - cfg->pgsize_bitmap = amd_iommu_pgsize_bitmap; - cfg->ias = IOMMU_IN_ADDR_BIT_SIZE; - cfg->oas = IOMMU_OUT_ADDR_BIT_SIZE; - - pgtable->pgtbl.ops.map_pages = iommu_v1_map_pages; - pgtable->pgtbl.ops.unmap_pages = iommu_v1_unmap_pages; - pgtable->pgtbl.ops.iova_to_phys = iommu_v1_iova_to_phys; - pgtable->pgtbl.ops.read_and_clear_dirty = iommu_v1_read_and_clear_dirty; - - return &pgtable->pgtbl; -} - -struct io_pgtable_init_fns io_pgtable_amd_iommu_v1_init_fns = { - .alloc = v1_alloc_pgtable, - .free = v1_free_pgtable, -}; diff --git a/drivers/iommu/amd/io_pgtable_v2.c b/drivers/iommu/amd/io_pgtable_v2.c deleted file mode 100644 index b47941353ccbb8..00000000000000 --- a/drivers/iommu/amd/io_pgtable_v2.c +++ /dev/null @@ -1,370 +0,0 @@ -// SPDX-License-Identifier: GPL-2.0-only -/* - * CPU-agnostic AMD IO page table v2 allocator. - * - * Copyright (C) 2022, 2023 Advanced Micro Devices, Inc. - * Author: Suravee Suthikulpanit suravee.suthikulpanit@amd.com - * Author: Vasant Hegde vasant.hegde@amd.com - */ - -#define pr_fmt(fmt) "AMD-Vi: " fmt -#define dev_fmt(fmt) pr_fmt(fmt) - -#include <linux/bitops.h> -#include <linux/io-pgtable.h> -#include <linux/kernel.h> - -#include <asm/barrier.h> - -#include "amd_iommu_types.h" -#include "amd_iommu.h" -#include "../iommu-pages.h" - -#define IOMMU_PAGE_PRESENT BIT_ULL(0) /* Is present */ -#define IOMMU_PAGE_RW BIT_ULL(1) /* Writeable */ -#define IOMMU_PAGE_USER BIT_ULL(2) /* Userspace addressable */ -#define IOMMU_PAGE_PWT BIT_ULL(3) /* Page write through */ -#define IOMMU_PAGE_PCD BIT_ULL(4) /* Page cache disabled */ -#define IOMMU_PAGE_ACCESS BIT_ULL(5) /* Was accessed (updated by IOMMU) */ -#define IOMMU_PAGE_DIRTY BIT_ULL(6) /* Was written to (updated by IOMMU) */ -#define IOMMU_PAGE_PSE BIT_ULL(7) /* Page Size Extensions */ -#define IOMMU_PAGE_NX BIT_ULL(63) /* No execute */ - -#define MAX_PTRS_PER_PAGE 512 - -#define IOMMU_PAGE_SIZE_2M BIT_ULL(21) -#define IOMMU_PAGE_SIZE_1G BIT_ULL(30) - - -static inline int get_pgtable_level(void) -{ - return amd_iommu_gpt_level; -} - -static inline bool is_large_pte(u64 pte) -{ - return (pte & IOMMU_PAGE_PSE); -} - -static inline u64 set_pgtable_attr(u64 *page) -{ - u64 prot; - - prot = IOMMU_PAGE_PRESENT | IOMMU_PAGE_RW | IOMMU_PAGE_USER; - prot |= IOMMU_PAGE_ACCESS; - - return (iommu_virt_to_phys(page) | prot); -} - -static inline void *get_pgtable_pte(u64 pte) -{ - return iommu_phys_to_virt(pte & PM_ADDR_MASK); -} - -static u64 set_pte_attr(u64 paddr, u64 pg_size, int prot) -{ - u64 pte; - - pte = __sme_set(paddr & PM_ADDR_MASK); - pte |= IOMMU_PAGE_PRESENT | IOMMU_PAGE_USER; - pte |= IOMMU_PAGE_ACCESS | IOMMU_PAGE_DIRTY; - - if (prot & IOMMU_PROT_IW) - pte |= IOMMU_PAGE_RW; - - /* Large page */ - if (pg_size == IOMMU_PAGE_SIZE_1G || pg_size == IOMMU_PAGE_SIZE_2M) - pte |= IOMMU_PAGE_PSE; - - return pte; -} - -static inline u64 get_alloc_page_size(u64 size) -{ - if (size >= IOMMU_PAGE_SIZE_1G) - return IOMMU_PAGE_SIZE_1G; - - if (size >= IOMMU_PAGE_SIZE_2M) - return IOMMU_PAGE_SIZE_2M; - - return PAGE_SIZE; -} - -static inline int page_size_to_level(u64 pg_size) -{ - if (pg_size == IOMMU_PAGE_SIZE_1G) - return PAGE_MODE_3_LEVEL; - if (pg_size == IOMMU_PAGE_SIZE_2M) - return PAGE_MODE_2_LEVEL; - - return PAGE_MODE_1_LEVEL; -} - -static void free_pgtable(u64 *pt, int level) -{ - u64 *p; - int i; - - for (i = 0; i < MAX_PTRS_PER_PAGE; i++) { - /* PTE present? */ - if (!IOMMU_PTE_PRESENT(pt[i])) - continue; - - if (is_large_pte(pt[i])) - continue; - - /* - * Free the next level. No need to look at l1 tables here since - * they can only contain leaf PTEs; just free them directly. - */ - p = get_pgtable_pte(pt[i]); - if (level > 2) - free_pgtable(p, level - 1); - else - iommu_free_pages(p); - } - - iommu_free_pages(pt); -} - -/* Allocate page table */ -static u64 *v2_alloc_pte(int nid, u64 *pgd, unsigned long iova, - unsigned long pg_size, gfp_t gfp, bool *updated) -{ - u64 *pte, *page; - int level, end_level; - - level = get_pgtable_level() - 1; - end_level = page_size_to_level(pg_size); - pte = &pgd[PM_LEVEL_INDEX(level, iova)]; - iova = PAGE_SIZE_ALIGN(iova, PAGE_SIZE); - - while (level >= end_level) { - u64 __pte, __npte; - - __pte = *pte; - - if (IOMMU_PTE_PRESENT(__pte) && is_large_pte(__pte)) { - /* Unmap large pte */ - cmpxchg64(pte, *pte, 0ULL); - *updated = true; - continue; - } - - if (!IOMMU_PTE_PRESENT(__pte)) { - page = iommu_alloc_pages_node_sz(nid, gfp, SZ_4K); - if (!page) - return NULL; - - __npte = set_pgtable_attr(page); - /* pte could have been changed somewhere. */ - if (!try_cmpxchg64(pte, &__pte, __npte)) - iommu_free_pages(page); - else if (IOMMU_PTE_PRESENT(__pte)) - *updated = true; - - continue; - } - - level -= 1; - pte = get_pgtable_pte(__pte); - pte = &pte[PM_LEVEL_INDEX(level, iova)]; - } - - /* Tear down existing pte entries */ - if (IOMMU_PTE_PRESENT(*pte)) { - u64 *__pte; - - *updated = true; - __pte = get_pgtable_pte(*pte); - cmpxchg64(pte, *pte, 0ULL); - if (pg_size == IOMMU_PAGE_SIZE_1G) - free_pgtable(__pte, end_level - 1); - else if (pg_size == IOMMU_PAGE_SIZE_2M) - iommu_free_pages(__pte); - } - - return pte; -} - -/* - * This function checks if there is a PTE for a given dma address. - * If there is one, it returns the pointer to it. - */ -static u64 *fetch_pte(struct amd_io_pgtable *pgtable, - unsigned long iova, unsigned long *page_size) -{ - u64 *pte; - int level; - - level = get_pgtable_level() - 1; - pte = &pgtable->pgd[PM_LEVEL_INDEX(level, iova)]; - /* Default page size is 4K */ - *page_size = PAGE_SIZE; - - while (level) { - /* Not present */ - if (!IOMMU_PTE_PRESENT(*pte)) - return NULL; - - /* Walk to the next level */ - pte = get_pgtable_pte(*pte); - pte = &pte[PM_LEVEL_INDEX(level - 1, iova)]; - - /* Large page */ - if (is_large_pte(*pte)) { - if (level == PAGE_MODE_3_LEVEL) - *page_size = IOMMU_PAGE_SIZE_1G; - else if (level == PAGE_MODE_2_LEVEL) - *page_size = IOMMU_PAGE_SIZE_2M; - else - return NULL; /* Wrongly set PSE bit in PTE */ - - break; - } - - level -= 1; - } - - return pte; -} - -static int iommu_v2_map_pages(struct io_pgtable_ops *ops, unsigned long iova, - phys_addr_t paddr, size_t pgsize, size_t pgcount, - int prot, gfp_t gfp, size_t *mapped) -{ - struct amd_io_pgtable *pgtable = io_pgtable_ops_to_data(ops); - struct io_pgtable_cfg *cfg = &pgtable->pgtbl.cfg; - u64 *pte; - unsigned long map_size; - unsigned long mapped_size = 0; - unsigned long o_iova = iova; - size_t size = pgcount << __ffs(pgsize); - int ret = 0; - bool updated = false; - - if (WARN_ON(!pgsize || (pgsize & cfg->pgsize_bitmap) != pgsize) || !pgcount) - return -EINVAL; - - if (!(prot & IOMMU_PROT_MASK)) - return -EINVAL; - - while (mapped_size < size) { - map_size = get_alloc_page_size(pgsize); - pte = v2_alloc_pte(cfg->amd.nid, pgtable->pgd, - iova, map_size, gfp, &updated); - if (!pte) { - ret = -ENOMEM; - goto out; - } - - *pte = set_pte_attr(paddr, map_size, prot); - - iova += map_size; - paddr += map_size; - mapped_size += map_size; - } - -out: - if (updated) { - struct protection_domain *pdom = io_pgtable_ops_to_domain(ops); - unsigned long flags; - - spin_lock_irqsave(&pdom->lock, flags); - amd_iommu_domain_flush_pages(pdom, o_iova, size); - spin_unlock_irqrestore(&pdom->lock, flags); - } - - if (mapped) - *mapped += mapped_size; - - return ret; -} - -static unsigned long iommu_v2_unmap_pages(struct io_pgtable_ops *ops, - unsigned long iova, - size_t pgsize, size_t pgcount, - struct iommu_iotlb_gather *gather) -{ - struct amd_io_pgtable *pgtable = io_pgtable_ops_to_data(ops); - struct io_pgtable_cfg *cfg = &pgtable->pgtbl.cfg; - unsigned long unmap_size; - unsigned long unmapped = 0; - size_t size = pgcount << __ffs(pgsize); - u64 *pte; - - if (WARN_ON(!pgsize || (pgsize & cfg->pgsize_bitmap) != pgsize || !pgcount)) - return 0; - - while (unmapped < size) { - pte = fetch_pte(pgtable, iova, &unmap_size); - if (!pte) - return unmapped; - - *pte = 0ULL; - - iova = (iova & ~(unmap_size - 1)) + unmap_size; - unmapped += unmap_size; - } - - return unmapped; -} - -static phys_addr_t iommu_v2_iova_to_phys(struct io_pgtable_ops *ops, unsigned long iova) -{ - struct amd_io_pgtable *pgtable = io_pgtable_ops_to_data(ops); - unsigned long offset_mask, pte_pgsize; - u64 *pte, __pte; - - pte = fetch_pte(pgtable, iova, &pte_pgsize); - if (!pte || !IOMMU_PTE_PRESENT(*pte)) - return 0; - - offset_mask = pte_pgsize - 1; - __pte = __sme_clr(*pte & PM_ADDR_MASK); - - return (__pte & ~offset_mask) | (iova & offset_mask); -} - -/* - * ---------------------------------------------------- - */ -static void v2_free_pgtable(struct io_pgtable *iop) -{ - struct amd_io_pgtable *pgtable = container_of(iop, struct amd_io_pgtable, pgtbl); - - if (!pgtable || !pgtable->pgd) - return; - - /* Free page table */ - free_pgtable(pgtable->pgd, get_pgtable_level()); - pgtable->pgd = NULL; -} - -static struct io_pgtable *v2_alloc_pgtable(struct io_pgtable_cfg *cfg, void *cookie) -{ - struct amd_io_pgtable *pgtable = io_pgtable_cfg_to_data(cfg); - int ias = IOMMU_IN_ADDR_BIT_SIZE; - - pgtable->pgd = iommu_alloc_pages_node_sz(cfg->amd.nid, GFP_KERNEL, SZ_4K); - if (!pgtable->pgd) - return NULL; - - if (get_pgtable_level() == PAGE_MODE_5_LEVEL) - ias = 57; - - pgtable->pgtbl.ops.map_pages = iommu_v2_map_pages; - pgtable->pgtbl.ops.unmap_pages = iommu_v2_unmap_pages; - pgtable->pgtbl.ops.iova_to_phys = iommu_v2_iova_to_phys; - - cfg->pgsize_bitmap = AMD_IOMMU_PGSIZES_V2; - cfg->ias = ias; - cfg->oas = IOMMU_OUT_ADDR_BIT_SIZE; - - return &pgtable->pgtbl; -} - -struct io_pgtable_init_fns io_pgtable_amd_iommu_v2_init_fns = { - .alloc = v2_alloc_pgtable, - .free = v2_free_pgtable, -}; diff --git a/drivers/iommu/io-pgtable.c b/drivers/iommu/io-pgtable.c index 8841c1487f0048..843fec8e8a511c 100644 --- a/drivers/iommu/io-pgtable.c +++ b/drivers/iommu/io-pgtable.c @@ -28,10 +28,6 @@ io_pgtable_init_table[IO_PGTABLE_NUM_FMTS] = { #ifdef CONFIG_IOMMU_IO_PGTABLE_ARMV7S [ARM_V7S] = &io_pgtable_arm_v7s_init_fns, #endif -#ifdef CONFIG_AMD_IOMMU - [AMD_IOMMU_V1] = &io_pgtable_amd_iommu_v1_init_fns, - [AMD_IOMMU_V2] = &io_pgtable_amd_iommu_v2_init_fns, -#endif };

static int check_custom_allocator(enum io_pgtable_fmt fmt, diff --git a/include/linux/io-pgtable.h b/include/linux/io-pgtable.h index 8a823c6f2b4a88..7a1516011ccf72 100644 --- a/include/linux/io-pgtable.h +++ b/include/linux/io-pgtable.h @@ -15,8 +15,6 @@ enum io_pgtable_fmt { ARM_64_LPAE_S2, ARM_V7S, ARM_MALI_LPAE, - AMD_IOMMU_V1, - AMD_IOMMU_V2, APPLE_DART, APPLE_DART2, IO_PGTABLE_NUM_FMTS,

On 12/5/25 2:46 PM, Jason Gunthorpe wrote:

On Fri, Dec 05, 2025 at 10:40:49AM +0800, Lai, Yi wrote:

...

Hi Alejandro Jimenez,

Greetings!

I used Syzkaller and found that there is WARNING in iommufd_fops_release in linux-next next-20251203.

After bisection and the first bad commit is: " 789a5913b29c iommu/amd: Use the generic iommu page table "

This bisect didn't work out, it is close, but the commit is

Thank you Jason for debugging this. I was wondering what could be the possible relation with the AMD changes, since the commit doesn't touch iommufd/main.c, and the syskaller logs show Intel HW so the AMD driver code is not involved.

Thank you, Alejandro

e93d5945ed5b ("iommufd: Change the selftest to use iommupt instead of xarray")

It is a refcount leak:

@@ -1215,8 +1215,10 @@ static int iommufd_test_md_check_pa(struct iommufd_ucmd *ucmd, page_size = 1 << __ffs(mock->domain.pgsize_bitmap); if (iova % page_size || length % page_size || (uintptr_t)uptr % page_size ||

•       check_add_overflow((uintptr_t)uptr, (uintptr_t)length, &end))
  

•           return -EINVAL;
  

•       check_add_overflow((uintptr_t)uptr, (uintptr_t)length, &end)) {
  

•           rc = -EINVAL;
  

•           goto out_put;
  

•   }
  
  

I'll send a patch, thanks

Jason