1 // SPDX-License-Identifier: GPL-2.0-only
3 * Copyright (C) 2020 ARM Ltd.
6 #include <linux/bitops.h>
7 #include <linux/kernel.h>
9 #include <linux/prctl.h>
10 #include <linux/sched.h>
11 #include <linux/sched/mm.h>
12 #include <linux/string.h>
13 #include <linux/swap.h>
14 #include <linux/swapops.h>
15 #include <linux/thread_info.h>
16 #include <linux/types.h>
17 #include <linux/uio.h>
19 #include <asm/barrier.h>
20 #include <asm/cpufeature.h>
22 #include <asm/mte-kasan.h>
23 #include <asm/ptrace.h>
24 #include <asm/sysreg.h>
26 static void mte_sync_page_tags(struct page *page, pte_t *ptep, bool check_swap)
28 pte_t old_pte = READ_ONCE(*ptep);
30 if (check_swap && is_swap_pte(old_pte)) {
31 swp_entry_t entry = pte_to_swp_entry(old_pte);
33 if (!non_swap_entry(entry) && mte_restore_tags(entry, page))
37 page_kasan_tag_reset(page);
39 * We need smp_wmb() in between setting the flags and clearing the
40 * tags because if another thread reads page->flags and builds a
41 * tagged address out of it, there is an actual dependency to the
42 * memory access, but on the current thread we do not guarantee that
43 * the new page->flags are visible before the tags were updated.
46 mte_clear_page_tags(page_address(page));
49 void mte_sync_tags(pte_t *ptep, pte_t pte)
51 struct page *page = pte_page(pte);
52 long i, nr_pages = compound_nr(page);
53 bool check_swap = nr_pages == 1;
55 /* if PG_mte_tagged is set, tags have already been initialised */
56 for (i = 0; i < nr_pages; i++, page++) {
57 if (!test_and_set_bit(PG_mte_tagged, &page->flags))
58 mte_sync_page_tags(page, ptep, check_swap);
62 int memcmp_pages(struct page *page1, struct page *page2)
67 addr1 = page_address(page1);
68 addr2 = page_address(page2);
69 ret = memcmp(addr1, addr2, PAGE_SIZE);
71 if (!system_supports_mte() || ret)
75 * If the page content is identical but at least one of the pages is
76 * tagged, return non-zero to avoid KSM merging. If only one of the
77 * pages is tagged, set_pte_at() may zero or change the tags of the
78 * other page via mte_sync_tags().
80 if (test_bit(PG_mte_tagged, &page1->flags) ||
81 test_bit(PG_mte_tagged, &page2->flags))
82 return addr1 != addr2;
87 u8 mte_get_mem_tag(void *addr)
89 if (!system_supports_mte())
92 asm(__MTE_PREAMBLE "ldg %0, [%0]"
95 return mte_get_ptr_tag(addr);
98 u8 mte_get_random_tag(void)
102 if (!system_supports_mte())
105 asm(__MTE_PREAMBLE "irg %0, %0"
108 return mte_get_ptr_tag(addr);
111 void *mte_set_mem_tag_range(void *addr, size_t size, u8 tag)
115 if ((!system_supports_mte()) || (size == 0))
118 /* Make sure that size is MTE granule aligned. */
119 WARN_ON(size & (MTE_GRANULE_SIZE - 1));
121 /* Make sure that the address is MTE granule aligned. */
122 WARN_ON((u64)addr & (MTE_GRANULE_SIZE - 1));
125 ptr = (void *)__tag_set(ptr, tag);
127 mte_assign_mem_tag_range(ptr, size);
132 static void update_sctlr_el1_tcf0(u64 tcf0)
134 /* ISB required for the kernel uaccess routines */
135 sysreg_clear_set(sctlr_el1, SCTLR_EL1_TCF0_MASK, tcf0);
139 static void set_sctlr_el1_tcf0(u64 tcf0)
142 * mte_thread_switch() checks current->thread.sctlr_tcf0 as an
143 * optimisation. Disable preemption so that it does not see
144 * the variable update before the SCTLR_EL1.TCF0 one.
147 current->thread.sctlr_tcf0 = tcf0;
148 update_sctlr_el1_tcf0(tcf0);
152 static void update_gcr_el1_excl(u64 incl)
154 u64 excl = ~incl & SYS_GCR_EL1_EXCL_MASK;
157 * Note that 'incl' is an include mask (controlled by the user via
158 * prctl()) while GCR_EL1 accepts an exclude mask.
159 * No need for ISB since this only affects EL0 currently, implicit
162 sysreg_clear_set_s(SYS_GCR_EL1, SYS_GCR_EL1_EXCL_MASK, excl);
165 static void set_gcr_el1_excl(u64 incl)
167 current->thread.gcr_user_incl = incl;
168 update_gcr_el1_excl(incl);
171 void flush_mte_state(void)
173 if (!system_supports_mte())
176 /* clear any pending asynchronous tag fault */
178 write_sysreg_s(0, SYS_TFSRE0_EL1);
179 clear_thread_flag(TIF_MTE_ASYNC_FAULT);
180 /* disable tag checking */
181 set_sctlr_el1_tcf0(SCTLR_EL1_TCF0_NONE);
182 /* reset tag generation mask */
186 void mte_thread_switch(struct task_struct *next)
188 if (!system_supports_mte())
191 /* avoid expensive SCTLR_EL1 accesses if no change */
192 if (current->thread.sctlr_tcf0 != next->thread.sctlr_tcf0)
193 update_sctlr_el1_tcf0(next->thread.sctlr_tcf0);
194 update_gcr_el1_excl(next->thread.gcr_user_incl);
197 void mte_suspend_exit(void)
199 if (!system_supports_mte())
202 update_gcr_el1_excl(current->thread.gcr_user_incl);
205 long set_mte_ctrl(struct task_struct *task, unsigned long arg)
208 u64 gcr_incl = (arg & PR_MTE_TAG_MASK) >> PR_MTE_TAG_SHIFT;
210 if (!system_supports_mte())
213 switch (arg & PR_MTE_TCF_MASK) {
214 case PR_MTE_TCF_NONE:
215 tcf0 = SCTLR_EL1_TCF0_NONE;
217 case PR_MTE_TCF_SYNC:
218 tcf0 = SCTLR_EL1_TCF0_SYNC;
220 case PR_MTE_TCF_ASYNC:
221 tcf0 = SCTLR_EL1_TCF0_ASYNC;
227 if (task != current) {
228 task->thread.sctlr_tcf0 = tcf0;
229 task->thread.gcr_user_incl = gcr_incl;
231 set_sctlr_el1_tcf0(tcf0);
232 set_gcr_el1_excl(gcr_incl);
238 long get_mte_ctrl(struct task_struct *task)
242 if (!system_supports_mte())
245 ret = task->thread.gcr_user_incl << PR_MTE_TAG_SHIFT;
247 switch (task->thread.sctlr_tcf0) {
248 case SCTLR_EL1_TCF0_NONE:
249 ret |= PR_MTE_TCF_NONE;
251 case SCTLR_EL1_TCF0_SYNC:
252 ret |= PR_MTE_TCF_SYNC;
254 case SCTLR_EL1_TCF0_ASYNC:
255 ret |= PR_MTE_TCF_ASYNC;
263 * Access MTE tags in another process' address space as given in mm. Update
264 * the number of tags copied. Return 0 if any tags copied, error otherwise.
265 * Inspired by __access_remote_vm().
267 static int __access_remote_tags(struct mm_struct *mm, unsigned long addr,
268 struct iovec *kiov, unsigned int gup_flags)
270 struct vm_area_struct *vma;
271 void __user *buf = kiov->iov_base;
272 size_t len = kiov->iov_len;
274 int write = gup_flags & FOLL_WRITE;
276 if (!access_ok(buf, len))
279 if (mmap_read_lock_killable(mm))
283 unsigned long tags, offset;
285 struct page *page = NULL;
287 ret = get_user_pages_remote(mm, addr, 1, gup_flags, &page,
293 * Only copy tags if the page has been mapped as PROT_MTE
294 * (PG_mte_tagged set). Otherwise the tags are not valid and
295 * not accessible to user. Moreover, an mprotect(PROT_MTE)
296 * would cause the existing tags to be cleared if the page
297 * was never mapped with PROT_MTE.
299 if (!test_bit(PG_mte_tagged, &page->flags)) {
305 /* limit access to the end of the page */
306 offset = offset_in_page(addr);
307 tags = min(len, (PAGE_SIZE - offset) / MTE_GRANULE_SIZE);
309 maddr = page_address(page);
311 tags = mte_copy_tags_from_user(maddr + offset, buf, tags);
312 set_page_dirty_lock(page);
314 tags = mte_copy_tags_to_user(buf, maddr + offset, tags);
318 /* error accessing the tracer's buffer */
324 addr += tags * MTE_GRANULE_SIZE;
326 mmap_read_unlock(mm);
328 /* return an error if no tags copied */
329 kiov->iov_len = buf - kiov->iov_base;
330 if (!kiov->iov_len) {
331 /* check for error accessing the tracee's address space */
342 * Copy MTE tags in another process' address space at 'addr' to/from tracer's
343 * iovec buffer. Return 0 on success. Inspired by ptrace_access_vm().
345 static int access_remote_tags(struct task_struct *tsk, unsigned long addr,
346 struct iovec *kiov, unsigned int gup_flags)
348 struct mm_struct *mm;
351 mm = get_task_mm(tsk);
355 if (!tsk->ptrace || (current != tsk->parent) ||
356 ((get_dumpable(mm) != SUID_DUMP_USER) &&
357 !ptracer_capable(tsk, mm->user_ns))) {
362 ret = __access_remote_tags(mm, addr, kiov, gup_flags);
368 int mte_ptrace_copy_tags(struct task_struct *child, long request,
369 unsigned long addr, unsigned long data)
373 struct iovec __user *uiov = (void __user *)data;
374 unsigned int gup_flags = FOLL_FORCE;
376 if (!system_supports_mte())
379 if (get_user(kiov.iov_base, &uiov->iov_base) ||
380 get_user(kiov.iov_len, &uiov->iov_len))
383 if (request == PTRACE_POKEMTETAGS)
384 gup_flags |= FOLL_WRITE;
386 /* align addr to the MTE tag granule */
387 addr &= MTE_GRANULE_MASK;
389 ret = access_remote_tags(child, addr, &kiov, gup_flags);
391 ret = put_user(kiov.iov_len, &uiov->iov_len);