1===============================================
2Memory Tagging Extension (MTE) in AArch64 Linux
3===============================================
4
5Authors: Vincenzo Frascino <vincenzo.frascino@arm.com>
6         Catalin Marinas <catalin.marinas@arm.com>
7
8Date: 2020-02-25
9
10This document describes the provision of the Memory Tagging Extension
11functionality in AArch64 Linux.
12
13Introduction
14============
15
16ARMv8.5 based processors introduce the Memory Tagging Extension (MTE)
17feature. MTE is built on top of the ARMv8.0 virtual address tagging TBI
18(Top Byte Ignore) feature and allows software to access a 4-bit
19allocation tag for each 16-byte granule in the physical address space.
20Such memory range must be mapped with the Normal-Tagged memory
21attribute. A logical tag is derived from bits 59-56 of the virtual
22address used for the memory access. A CPU with MTE enabled will compare
23the logical tag against the allocation tag and potentially raise an
24exception on mismatch, subject to system registers configuration.
25
26Userspace Support
27=================
28
29When ``CONFIG_ARM64_MTE`` is selected and Memory Tagging Extension is
30supported by the hardware, the kernel advertises the feature to
31userspace via ``HWCAP2_MTE``.
32
33PROT_MTE
34--------
35
36To access the allocation tags, a user process must enable the Tagged
37memory attribute on an address range using a new ``prot`` flag for
38``mmap()`` and ``mprotect()``:
39
40``PROT_MTE`` - Pages allow access to the MTE allocation tags.
41
42The allocation tag is set to 0 when such pages are first mapped in the
43user address space and preserved on copy-on-write. ``MAP_SHARED`` is
44supported and the allocation tags can be shared between processes.
45
46**Note**: ``PROT_MTE`` is only supported on ``MAP_ANONYMOUS`` and
47RAM-based file mappings (``tmpfs``, ``memfd``). Passing it to other
48types of mapping will result in ``-EINVAL`` returned by these system
49calls.
50
51**Note**: The ``PROT_MTE`` flag (and corresponding memory type) cannot
52be cleared by ``mprotect()``.
53
54**Note**: ``madvise()`` memory ranges with ``MADV_DONTNEED`` and
55``MADV_FREE`` may have the allocation tags cleared (set to 0) at any
56point after the system call.
57
58Tag Check Faults
59----------------
60
61When ``PROT_MTE`` is enabled on an address range and a mismatch between
62the logical and allocation tags occurs on access, there are three
63configurable behaviours:
64
65- *Ignore* - This is the default mode. The CPU (and kernel) ignores the
66  tag check fault.
67
68- *Synchronous* - The kernel raises a ``SIGSEGV`` synchronously, with
69  ``.si_code = SEGV_MTESERR`` and ``.si_addr = <fault-address>``. The
70  memory access is not performed. If ``SIGSEGV`` is ignored or blocked
71  by the offending thread, the containing process is terminated with a
72  ``coredump``.
73
74- *Asynchronous* - The kernel raises a ``SIGSEGV``, in the offending
75  thread, asynchronously following one or multiple tag check faults,
76  with ``.si_code = SEGV_MTEAERR`` and ``.si_addr = 0`` (the faulting
77  address is unknown).
78
79- *Asymmetric* - Reads are handled as for synchronous mode while writes
80  are handled as for asynchronous mode.
81
82The user can select the above modes, per thread, using the
83``prctl(PR_SET_TAGGED_ADDR_CTRL, flags, 0, 0, 0)`` system call where ``flags``
84contains any number of the following values in the ``PR_MTE_TCF_MASK``
85bit-field:
86
87- ``PR_MTE_TCF_NONE``  - *Ignore* tag check faults
88                         (ignored if combined with other options)
89- ``PR_MTE_TCF_SYNC``  - *Synchronous* tag check fault mode
90- ``PR_MTE_TCF_ASYNC`` - *Asynchronous* tag check fault mode
91
92If no modes are specified, tag check faults are ignored. If a single
93mode is specified, the program will run in that mode. If multiple
94modes are specified, the mode is selected as described in the "Per-CPU
95preferred tag checking modes" section below.
96
97The current tag check fault configuration can be read using the
98``prctl(PR_GET_TAGGED_ADDR_CTRL, 0, 0, 0, 0)`` system call. If
99multiple modes were requested then all will be reported.
100
101Tag checking can also be disabled for a user thread by setting the
102``PSTATE.TCO`` bit with ``MSR TCO, #1``.
103
104**Note**: Signal handlers are always invoked with ``PSTATE.TCO = 0``,
105irrespective of the interrupted context. ``PSTATE.TCO`` is restored on
106``sigreturn()``.
107
108**Note**: There are no *match-all* logical tags available for user
109applications.
110
111**Note**: Kernel accesses to the user address space (e.g. ``read()``
112system call) are not checked if the user thread tag checking mode is
113``PR_MTE_TCF_NONE`` or ``PR_MTE_TCF_ASYNC``. If the tag checking mode is
114``PR_MTE_TCF_SYNC``, the kernel makes a best effort to check its user
115address accesses, however it cannot always guarantee it. Kernel accesses
116to user addresses are always performed with an effective ``PSTATE.TCO``
117value of zero, regardless of the user configuration.
118
119Excluding Tags in the ``IRG``, ``ADDG`` and ``SUBG`` instructions
120-----------------------------------------------------------------
121
122The architecture allows excluding certain tags to be randomly generated
123via the ``GCR_EL1.Exclude`` register bit-field. By default, Linux
124excludes all tags other than 0. A user thread can enable specific tags
125in the randomly generated set using the ``prctl(PR_SET_TAGGED_ADDR_CTRL,
126flags, 0, 0, 0)`` system call where ``flags`` contains the tags bitmap
127in the ``PR_MTE_TAG_MASK`` bit-field.
128
129**Note**: The hardware uses an exclude mask but the ``prctl()``
130interface provides an include mask. An include mask of ``0`` (exclusion
131mask ``0xffff``) results in the CPU always generating tag ``0``.
132
133Per-CPU preferred tag checking mode
134-----------------------------------
135
136On some CPUs the performance of MTE in stricter tag checking modes
137is similar to that of less strict tag checking modes. This makes it
138worthwhile to enable stricter checks on those CPUs when a less strict
139checking mode is requested, in order to gain the error detection
140benefits of the stricter checks without the performance downsides. To
141support this scenario, a privileged user may configure a stricter
142tag checking mode as the CPU's preferred tag checking mode.
143
144The preferred tag checking mode for each CPU is controlled by
145``/sys/devices/system/cpu/cpu<N>/mte_tcf_preferred``, to which a
146privileged user may write the value ``async``, ``sync`` or ``asymm``.  The
147default preferred mode for each CPU is ``async``.
148
149To allow a program to potentially run in the CPU's preferred tag
150checking mode, the user program may set multiple tag check fault mode
151bits in the ``flags`` argument to the ``prctl(PR_SET_TAGGED_ADDR_CTRL,
152flags, 0, 0, 0)`` system call. If both synchronous and asynchronous
153modes are requested then asymmetric mode may also be selected by the
154kernel. If the CPU's preferred tag checking mode is in the task's set
155of provided tag checking modes, that mode will be selected. Otherwise,
156one of the modes in the task's mode will be selected by the kernel
157from the task's mode set using the preference order:
158
159	1. Asynchronous
160	2. Asymmetric
161	3. Synchronous
162
163Note that there is no way for userspace to request multiple modes and
164also disable asymmetric mode.
165
166Initial process state
167---------------------
168
169On ``execve()``, the new process has the following configuration:
170
171- ``PR_TAGGED_ADDR_ENABLE`` set to 0 (disabled)
172- No tag checking modes are selected (tag check faults ignored)
173- ``PR_MTE_TAG_MASK`` set to 0 (all tags excluded)
174- ``PSTATE.TCO`` set to 0
175- ``PROT_MTE`` not set on any of the initial memory maps
176
177On ``fork()``, the new process inherits the parent's configuration and
178memory map attributes with the exception of the ``madvise()`` ranges
179with ``MADV_WIPEONFORK`` which will have the data and tags cleared (set
180to 0).
181
182The ``ptrace()`` interface
183--------------------------
184
185``PTRACE_PEEKMTETAGS`` and ``PTRACE_POKEMTETAGS`` allow a tracer to read
186the tags from or set the tags to a tracee's address space. The
187``ptrace()`` system call is invoked as ``ptrace(request, pid, addr,
188data)`` where:
189
190- ``request`` - one of ``PTRACE_PEEKMTETAGS`` or ``PTRACE_POKEMTETAGS``.
191- ``pid`` - the tracee's PID.
192- ``addr`` - address in the tracee's address space.
193- ``data`` - pointer to a ``struct iovec`` where ``iov_base`` points to
194  a buffer of ``iov_len`` length in the tracer's address space.
195
196The tags in the tracer's ``iov_base`` buffer are represented as one
1974-bit tag per byte and correspond to a 16-byte MTE tag granule in the
198tracee's address space.
199
200**Note**: If ``addr`` is not aligned to a 16-byte granule, the kernel
201will use the corresponding aligned address.
202
203``ptrace()`` return value:
204
205- 0 - tags were copied, the tracer's ``iov_len`` was updated to the
206  number of tags transferred. This may be smaller than the requested
207  ``iov_len`` if the requested address range in the tracee's or the
208  tracer's space cannot be accessed or does not have valid tags.
209- ``-EPERM`` - the specified process cannot be traced.
210- ``-EIO`` - the tracee's address range cannot be accessed (e.g. invalid
211  address) and no tags copied. ``iov_len`` not updated.
212- ``-EFAULT`` - fault on accessing the tracer's memory (``struct iovec``
213  or ``iov_base`` buffer) and no tags copied. ``iov_len`` not updated.
214- ``-EOPNOTSUPP`` - the tracee's address does not have valid tags (never
215  mapped with the ``PROT_MTE`` flag). ``iov_len`` not updated.
216
217**Note**: There are no transient errors for the requests above, so user
218programs should not retry in case of a non-zero system call return.
219
220``PTRACE_GETREGSET`` and ``PTRACE_SETREGSET`` with ``addr ==
221``NT_ARM_TAGGED_ADDR_CTRL`` allow ``ptrace()`` access to the tagged
222address ABI control and MTE configuration of a process as per the
223``prctl()`` options described in
224Documentation/arch/arm64/tagged-address-abi.rst and above. The corresponding
225``regset`` is 1 element of 8 bytes (``sizeof(long))``).
226
227Core dump support
228-----------------
229
230The allocation tags for user memory mapped with ``PROT_MTE`` are dumped
231in the core file as additional ``PT_AARCH64_MEMTAG_MTE`` segments. The
232program header for such segment is defined as:
233
234:``p_type``: ``PT_AARCH64_MEMTAG_MTE``
235:``p_flags``: 0
236:``p_offset``: segment file offset
237:``p_vaddr``: segment virtual address, same as the corresponding
238  ``PT_LOAD`` segment
239:``p_paddr``: 0
240:``p_filesz``: segment size in file, calculated as ``p_mem_sz / 32``
241  (two 4-bit tags cover 32 bytes of memory)
242:``p_memsz``: segment size in memory, same as the corresponding
243  ``PT_LOAD`` segment
244:``p_align``: 0
245
246The tags are stored in the core file at ``p_offset`` as two 4-bit tags
247in a byte. With the tag granule of 16 bytes, a 4K page requires 128
248bytes in the core file.
249
250Example of correct usage
251========================
252
253*MTE Example code*
254
255.. code-block:: c
256
257    /*
258     * To be compiled with -march=armv8.5-a+memtag
259     */
260    #include <errno.h>
261    #include <stdint.h>
262    #include <stdio.h>
263    #include <stdlib.h>
264    #include <unistd.h>
265    #include <sys/auxv.h>
266    #include <sys/mman.h>
267    #include <sys/prctl.h>
268
269    /*
270     * From arch/arm64/include/uapi/asm/hwcap.h
271     */
272    #define HWCAP2_MTE              (1 << 18)
273
274    /*
275     * From arch/arm64/include/uapi/asm/mman.h
276     */
277    #define PROT_MTE                 0x20
278
279    /*
280     * From include/uapi/linux/prctl.h
281     */
282    #define PR_SET_TAGGED_ADDR_CTRL 55
283    #define PR_GET_TAGGED_ADDR_CTRL 56
284    # define PR_TAGGED_ADDR_ENABLE  (1UL << 0)
285    # define PR_MTE_TCF_SHIFT       1
286    # define PR_MTE_TCF_NONE        (0UL << PR_MTE_TCF_SHIFT)
287    # define PR_MTE_TCF_SYNC        (1UL << PR_MTE_TCF_SHIFT)
288    # define PR_MTE_TCF_ASYNC       (2UL << PR_MTE_TCF_SHIFT)
289    # define PR_MTE_TCF_MASK        (3UL << PR_MTE_TCF_SHIFT)
290    # define PR_MTE_TAG_SHIFT       3
291    # define PR_MTE_TAG_MASK        (0xffffUL << PR_MTE_TAG_SHIFT)
292
293    /*
294     * Insert a random logical tag into the given pointer.
295     */
296    #define insert_random_tag(ptr) ({                       \
297            uint64_t __val;                                 \
298            asm("irg %0, %1" : "=r" (__val) : "r" (ptr));   \
299            __val;                                          \
300    })
301
302    /*
303     * Set the allocation tag on the destination address.
304     */
305    #define set_tag(tagged_addr) do {                                      \
306            asm volatile("stg %0, [%0]" : : "r" (tagged_addr) : "memory"); \
307    } while (0)
308
309    int main()
310    {
311            unsigned char *a;
312            unsigned long page_sz = sysconf(_SC_PAGESIZE);
313            unsigned long hwcap2 = getauxval(AT_HWCAP2);
314
315            /* check if MTE is present */
316            if (!(hwcap2 & HWCAP2_MTE))
317                    return EXIT_FAILURE;
318
319            /*
320             * Enable the tagged address ABI, synchronous or asynchronous MTE
321             * tag check faults (based on per-CPU preference) and allow all
322             * non-zero tags in the randomly generated set.
323             */
324            if (prctl(PR_SET_TAGGED_ADDR_CTRL,
325                      PR_TAGGED_ADDR_ENABLE | PR_MTE_TCF_SYNC | PR_MTE_TCF_ASYNC |
326                      (0xfffe << PR_MTE_TAG_SHIFT),
327                      0, 0, 0)) {
328                    perror("prctl() failed");
329                    return EXIT_FAILURE;
330            }
331
332            a = mmap(0, page_sz, PROT_READ | PROT_WRITE,
333                     MAP_PRIVATE | MAP_ANONYMOUS, -1, 0);
334            if (a == MAP_FAILED) {
335                    perror("mmap() failed");
336                    return EXIT_FAILURE;
337            }
338
339            /*
340             * Enable MTE on the above anonymous mmap. The flag could be passed
341             * directly to mmap() and skip this step.
342             */
343            if (mprotect(a, page_sz, PROT_READ | PROT_WRITE | PROT_MTE)) {
344                    perror("mprotect() failed");
345                    return EXIT_FAILURE;
346            }
347
348            /* access with the default tag (0) */
349            a[0] = 1;
350            a[1] = 2;
351
352            printf("a[0] = %hhu a[1] = %hhu\n", a[0], a[1]);
353
354            /* set the logical and allocation tags */
355            a = (unsigned char *)insert_random_tag(a);
356            set_tag(a);
357
358            printf("%p\n", a);
359
360            /* non-zero tag access */
361            a[0] = 3;
362            printf("a[0] = %hhu a[1] = %hhu\n", a[0], a[1]);
363
364            /*
365             * If MTE is enabled correctly the next instruction will generate an
366             * exception.
367             */
368            printf("Expecting SIGSEGV...\n");
369            a[16] = 0xdd;
370
371            /* this should not be printed in the PR_MTE_TCF_SYNC mode */
372            printf("...haven't got one\n");
373
374            return EXIT_FAILURE;
375    }
376