Linux kernel mirror (for testing)
git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git
kernel
os
linux
1/* SPDX-License-Identifier: GPL-2.0-only */
2/*
3 * Copyright (c) 2024-2025, NVIDIA CORPORATION & AFFILIATES
4 *
5 * This header is included before the format. It contains definitions
6 * that are required to compile the format. The header order is:
7 * pt_defs.h
8 * fmt_XX.h
9 * pt_common.h
10 */
11#ifndef __GENERIC_PT_DEFS_H
12#define __GENERIC_PT_DEFS_H
13
14#include <linux/generic_pt/common.h>
15
16#include <linux/types.h>
17#include <linux/atomic.h>
18#include <linux/bits.h>
19#include <linux/limits.h>
20#include <linux/bug.h>
21#include <linux/kconfig.h>
22#include "pt_log2.h"
23
24/* Header self-compile default defines */
25#ifndef pt_write_attrs
26typedef u64 pt_vaddr_t;
27typedef u64 pt_oaddr_t;
28#endif
29
30struct pt_table_p;
31
32enum {
33 PT_VADDR_MAX = sizeof(pt_vaddr_t) == 8 ? U64_MAX : U32_MAX,
34 PT_VADDR_MAX_LG2 = sizeof(pt_vaddr_t) == 8 ? 64 : 32,
35 PT_OADDR_MAX = sizeof(pt_oaddr_t) == 8 ? U64_MAX : U32_MAX,
36 PT_OADDR_MAX_LG2 = sizeof(pt_oaddr_t) == 8 ? 64 : 32,
37};
38
39/*
40 * The format instantiation can have features wired off or on to optimize the
41 * code gen. Supported features are just a reflection of what the current set of
42 * kernel users want to use.
43 */
44#ifndef PT_SUPPORTED_FEATURES
45#define PT_SUPPORTED_FEATURES 0
46#endif
47
48/*
49 * When in debug mode we compile all formats with all features. This allows the
50 * kunit to test the full matrix. SIGN_EXTEND can't co-exist with DYNAMIC_TOP or
51 * FULL_VA. DMA_INCOHERENT requires a SW bit that not all formats have
52 */
53#if IS_ENABLED(CONFIG_DEBUG_GENERIC_PT)
54enum {
55 PT_ORIG_SUPPORTED_FEATURES = PT_SUPPORTED_FEATURES,
56 PT_DEBUG_SUPPORTED_FEATURES =
57 UINT_MAX &
58 ~((PT_ORIG_SUPPORTED_FEATURES & BIT(PT_FEAT_DMA_INCOHERENT) ?
59 0 :
60 BIT(PT_FEAT_DMA_INCOHERENT))) &
61 ~((PT_ORIG_SUPPORTED_FEATURES & BIT(PT_FEAT_SIGN_EXTEND)) ?
62 BIT(PT_FEAT_DYNAMIC_TOP) | BIT(PT_FEAT_FULL_VA) :
63 BIT(PT_FEAT_SIGN_EXTEND)),
64};
65#undef PT_SUPPORTED_FEATURES
66#define PT_SUPPORTED_FEATURES PT_DEBUG_SUPPORTED_FEATURES
67#endif
68
69#ifndef PT_FORCE_ENABLED_FEATURES
70#define PT_FORCE_ENABLED_FEATURES 0
71#endif
72
73/**
74 * DOC: Generic Page Table Language
75 *
76 * Language used in Generic Page Table
77 * VA
78 * The input address to the page table, often the virtual address.
79 * OA
80 * The output address from the page table, often the physical address.
81 * leaf
82 * An entry that results in an output address.
83 * start/end
84 * An half-open range, e.g. [0,0) refers to no VA.
85 * start/last
86 * An inclusive closed range, e.g. [0,0] refers to the VA 0
87 * common
88 * The generic page table container struct pt_common
89 * level
90 * Level 0 is always a table of only leaves with no futher table pointers.
91 * Increasing levels increase the size of the table items. The least
92 * significant VA bits used to index page tables are used to index the Level
93 * 0 table. The various labels for table levels used by HW descriptions are
94 * not used.
95 * top_level
96 * The inclusive highest level of the table. A two-level table
97 * has a top level of 1.
98 * table
99 * A linear array of translation items for that level.
100 * index
101 * The position in a table of an element: item = table[index]
102 * item
103 * A single index in a table
104 * entry
105 * A single logical element in a table. If contiguous pages are not
106 * supported then item and entry are the same thing, otherwise entry refers
107 * to all the items that comprise a single contiguous translation.
108 * item/entry_size
109 * The number of bytes of VA the table index translates for.
110 * If the item is a table entry then the next table covers
111 * this size. If the entry translates to an output address then the
112 * full OA is: OA | (VA % entry_size)
113 * contig_count
114 * The number of consecutive items fused into a single entry.
115 * item_size * contig_count is the size of that entry's translation.
116 * lg2
117 * Indicates the value is encoded as log2, i.e. 1<<x is the actual value.
118 * Normally the compiler is fine to optimize divide and mod with log2 values
119 * automatically when inlining, however if the values are not constant
120 * expressions it can't. So we do it by hand; we want to avoid 64-bit
121 * divmod.
122 */
123
124/* Returned by pt_load_entry() and for_each_pt_level_entry() */
125enum pt_entry_type {
126 PT_ENTRY_EMPTY,
127 /* Entry is valid and points to a lower table level */
128 PT_ENTRY_TABLE,
129 /* Entry is valid and returns an output address */
130 PT_ENTRY_OA,
131};
132
133struct pt_range {
134 struct pt_common *common;
135 struct pt_table_p *top_table;
136 pt_vaddr_t va;
137 pt_vaddr_t last_va;
138 u8 top_level;
139 u8 max_vasz_lg2;
140};
141
142/*
143 * Similar to xa_state, this records information about an in-progress parse at a
144 * single level.
145 */
146struct pt_state {
147 struct pt_range *range;
148 struct pt_table_p *table;
149 struct pt_table_p *table_lower;
150 u64 entry;
151 enum pt_entry_type type;
152 unsigned short index;
153 unsigned short end_index;
154 u8 level;
155};
156
157#define pt_cur_table(pts, type) ((type *)((pts)->table))
158
159/*
160 * Try to install a new table pointer. The locking methodology requires this to
161 * be atomic (multiple threads can race to install a pointer). The losing
162 * threads will fail the atomic and return false. They should free any memory
163 * and reparse the table level again.
164 */
165#if !IS_ENABLED(CONFIG_GENERIC_ATOMIC64)
166static inline bool pt_table_install64(struct pt_state *pts, u64 table_entry)
167{
168 u64 *entryp = pt_cur_table(pts, u64) + pts->index;
169 u64 old_entry = pts->entry;
170 bool ret;
171
172 /*
173 * Ensure the zero'd table content itself is visible before its PTE can
174 * be. release is a NOP on !SMP, but the HW is still doing an acquire.
175 */
176 if (!IS_ENABLED(CONFIG_SMP))
177 dma_wmb();
178 ret = try_cmpxchg64_release(entryp, &old_entry, table_entry);
179 if (ret)
180 pts->entry = table_entry;
181 return ret;
182}
183#endif
184
185static inline bool pt_table_install32(struct pt_state *pts, u32 table_entry)
186{
187 u32 *entryp = pt_cur_table(pts, u32) + pts->index;
188 u32 old_entry = pts->entry;
189 bool ret;
190
191 /*
192 * Ensure the zero'd table content itself is visible before its PTE can
193 * be. release is a NOP on !SMP, but the HW is still doing an acquire.
194 */
195 if (!IS_ENABLED(CONFIG_SMP))
196 dma_wmb();
197 ret = try_cmpxchg_release(entryp, &old_entry, table_entry);
198 if (ret)
199 pts->entry = table_entry;
200 return ret;
201}
202
203#define PT_SUPPORTED_FEATURE(feature_nr) (PT_SUPPORTED_FEATURES & BIT(feature_nr))
204
205static __always_inline bool pt_feature(const struct pt_common *common,
206 unsigned int feature_nr)
207{
208 if (PT_FORCE_ENABLED_FEATURES & BIT(feature_nr))
209 return true;
210 if (!PT_SUPPORTED_FEATURE(feature_nr))
211 return false;
212 return common->features & BIT(feature_nr);
213}
214
215static __always_inline bool pts_feature(const struct pt_state *pts,
216 unsigned int feature_nr)
217{
218 return pt_feature(pts->range->common, feature_nr);
219}
220
221/*
222 * PT_WARN_ON is used for invariants that the kunit should be checking can't
223 * happen.
224 */
225#if IS_ENABLED(CONFIG_DEBUG_GENERIC_PT)
226#define PT_WARN_ON WARN_ON
227#else
228static inline bool PT_WARN_ON(bool condition)
229{
230 return false;
231}
232#endif
233
234/* These all work on the VA type */
235#define log2_to_int(a_lg2) log2_to_int_t(pt_vaddr_t, a_lg2)
236#define log2_to_max_int(a_lg2) log2_to_max_int_t(pt_vaddr_t, a_lg2)
237#define log2_div(a, b_lg2) log2_div_t(pt_vaddr_t, a, b_lg2)
238#define log2_div_eq(a, b, c_lg2) log2_div_eq_t(pt_vaddr_t, a, b, c_lg2)
239#define log2_mod(a, b_lg2) log2_mod_t(pt_vaddr_t, a, b_lg2)
240#define log2_mod_eq_max(a, b_lg2) log2_mod_eq_max_t(pt_vaddr_t, a, b_lg2)
241#define log2_set_mod(a, val, b_lg2) log2_set_mod_t(pt_vaddr_t, a, val, b_lg2)
242#define log2_set_mod_max(a, b_lg2) log2_set_mod_max_t(pt_vaddr_t, a, b_lg2)
243#define log2_mul(a, b_lg2) log2_mul_t(pt_vaddr_t, a, b_lg2)
244#define vaffs(a) ffs_t(pt_vaddr_t, a)
245#define vafls(a) fls_t(pt_vaddr_t, a)
246#define vaffz(a) ffz_t(pt_vaddr_t, a)
247
248/*
249 * The full VA (fva) versions permit the lg2 value to be == PT_VADDR_MAX_LG2 and
250 * generate a useful defined result. The non-fva versions will malfunction at
251 * this extreme.
252 */
253static inline pt_vaddr_t fvalog2_div(pt_vaddr_t a, unsigned int b_lg2)
254{
255 if (PT_SUPPORTED_FEATURE(PT_FEAT_FULL_VA) && b_lg2 == PT_VADDR_MAX_LG2)
256 return 0;
257 return log2_div_t(pt_vaddr_t, a, b_lg2);
258}
259
260static inline pt_vaddr_t fvalog2_mod(pt_vaddr_t a, unsigned int b_lg2)
261{
262 if (PT_SUPPORTED_FEATURE(PT_FEAT_FULL_VA) && b_lg2 == PT_VADDR_MAX_LG2)
263 return a;
264 return log2_mod_t(pt_vaddr_t, a, b_lg2);
265}
266
267static inline bool fvalog2_div_eq(pt_vaddr_t a, pt_vaddr_t b,
268 unsigned int c_lg2)
269{
270 if (PT_SUPPORTED_FEATURE(PT_FEAT_FULL_VA) && c_lg2 == PT_VADDR_MAX_LG2)
271 return true;
272 return log2_div_eq_t(pt_vaddr_t, a, b, c_lg2);
273}
274
275static inline pt_vaddr_t fvalog2_set_mod(pt_vaddr_t a, pt_vaddr_t val,
276 unsigned int b_lg2)
277{
278 if (PT_SUPPORTED_FEATURE(PT_FEAT_FULL_VA) && b_lg2 == PT_VADDR_MAX_LG2)
279 return val;
280 return log2_set_mod_t(pt_vaddr_t, a, val, b_lg2);
281}
282
283static inline pt_vaddr_t fvalog2_set_mod_max(pt_vaddr_t a, unsigned int b_lg2)
284{
285 if (PT_SUPPORTED_FEATURE(PT_FEAT_FULL_VA) && b_lg2 == PT_VADDR_MAX_LG2)
286 return PT_VADDR_MAX;
287 return log2_set_mod_max_t(pt_vaddr_t, a, b_lg2);
288}
289
290/* These all work on the OA type */
291#define oalog2_to_int(a_lg2) log2_to_int_t(pt_oaddr_t, a_lg2)
292#define oalog2_to_max_int(a_lg2) log2_to_max_int_t(pt_oaddr_t, a_lg2)
293#define oalog2_div(a, b_lg2) log2_div_t(pt_oaddr_t, a, b_lg2)
294#define oalog2_div_eq(a, b, c_lg2) log2_div_eq_t(pt_oaddr_t, a, b, c_lg2)
295#define oalog2_mod(a, b_lg2) log2_mod_t(pt_oaddr_t, a, b_lg2)
296#define oalog2_mod_eq_max(a, b_lg2) log2_mod_eq_max_t(pt_oaddr_t, a, b_lg2)
297#define oalog2_set_mod(a, val, b_lg2) log2_set_mod_t(pt_oaddr_t, a, val, b_lg2)
298#define oalog2_set_mod_max(a, b_lg2) log2_set_mod_max_t(pt_oaddr_t, a, b_lg2)
299#define oalog2_mul(a, b_lg2) log2_mul_t(pt_oaddr_t, a, b_lg2)
300#define oaffs(a) ffs_t(pt_oaddr_t, a)
301#define oafls(a) fls_t(pt_oaddr_t, a)
302#define oaffz(a) ffz_t(pt_oaddr_t, a)
303
304static inline uintptr_t _pt_top_set(struct pt_table_p *table_mem,
305 unsigned int top_level)
306{
307 return top_level | (uintptr_t)table_mem;
308}
309
310static inline void pt_top_set(struct pt_common *common,
311 struct pt_table_p *table_mem,
312 unsigned int top_level)
313{
314 WRITE_ONCE(common->top_of_table, _pt_top_set(table_mem, top_level));
315}
316
317static inline void pt_top_set_level(struct pt_common *common,
318 unsigned int top_level)
319{
320 pt_top_set(common, NULL, top_level);
321}
322
323static inline unsigned int pt_top_get_level(const struct pt_common *common)
324{
325 return READ_ONCE(common->top_of_table) % (1 << PT_TOP_LEVEL_BITS);
326}
327
328static inline bool pt_check_install_leaf_args(struct pt_state *pts,
329 pt_oaddr_t oa,
330 unsigned int oasz_lg2);
331
332#endif