1==========================
2AArch64 TAGGED ADDRESS ABI
3==========================
4
5Authors: Vincenzo Frascino <vincenzo.frascino@arm.com>
6         Catalin Marinas <catalin.marinas@arm.com>
7
8Date: 21 August 2019
9
10This document describes the usage and semantics of the Tagged Address
11ABI on AArch64 Linux.
12
131. Introduction
14---------------
15
16On AArch64 the ``TCR_EL1.TBI0`` bit is set by default, allowing
17userspace (EL0) to perform memory accesses through 64-bit pointers with
18a non-zero top byte. This document describes the relaxation of the
19syscall ABI that allows userspace to pass certain tagged pointers to
20kernel syscalls.
21
222. AArch64 Tagged Address ABI
23-----------------------------
24
25From the kernel syscall interface perspective and for the purposes of
26this document, a "valid tagged pointer" is a pointer with a potentially
27non-zero top-byte that references an address in the user process address
28space obtained in one of the following ways:
29
30- ``mmap()`` syscall where either:
31
32  - flags have the ``MAP_ANONYMOUS`` bit set or
33  - the file descriptor refers to a regular file (including those
34    returned by ``memfd_create()``) or ``/dev/zero``
35
36- ``brk()`` syscall (i.e. the heap area between the initial location of
37  the program break at process creation and its current location).
38
39- any memory mapped by the kernel in the address space of the process
40  during creation and with the same restrictions as for ``mmap()`` above
41  (e.g. data, bss, stack).
42
43The AArch64 Tagged Address ABI has two stages of relaxation depending on
44how the user addresses are used by the kernel:
45
461. User addresses not accessed by the kernel but used for address space
47   management (e.g. ``mprotect()``, ``madvise()``). The use of valid
48   tagged pointers in this context is allowed with these exceptions:
49
50   - ``brk()``, ``mmap()`` and the ``new_address`` argument to
51     ``mremap()`` as these have the potential to alias with existing
52     user addresses.
53
54     NOTE: This behaviour changed in v5.6 and so some earlier kernels may
55     incorrectly accept valid tagged pointers for the ``brk()``,
56     ``mmap()`` and ``mremap()`` system calls.
57
58   - The ``range.start``, ``start`` and ``dst`` arguments to the
59     ``UFFDIO_*`` ``ioctl()``s used on a file descriptor obtained from
60     ``userfaultfd()``, as fault addresses subsequently obtained by reading
61     the file descriptor will be untagged, which may otherwise confuse
62     tag-unaware programs.
63
64     NOTE: This behaviour changed in v5.14 and so some earlier kernels may
65     incorrectly accept valid tagged pointers for this system call.
66
672. User addresses accessed by the kernel (e.g. ``write()``). This ABI
68   relaxation is disabled by default and the application thread needs to
69   explicitly enable it via ``prctl()`` as follows:
70
71   - ``PR_SET_TAGGED_ADDR_CTRL``: enable or disable the AArch64 Tagged
72     Address ABI for the calling thread.
73
74     The ``(unsigned int) arg2`` argument is a bit mask describing the
75     control mode used:
76
77     - ``PR_TAGGED_ADDR_ENABLE``: enable AArch64 Tagged Address ABI.
78       Default status is disabled.
79
80     Arguments ``arg3``, ``arg4``, and ``arg5`` must be 0.
81
82   - ``PR_GET_TAGGED_ADDR_CTRL``: get the status of the AArch64 Tagged
83     Address ABI for the calling thread.
84
85     Arguments ``arg2``, ``arg3``, ``arg4``, and ``arg5`` must be 0.
86
87   The ABI properties described above are thread-scoped, inherited on
88   clone() and fork() and cleared on exec().
89
90   Calling ``prctl(PR_SET_TAGGED_ADDR_CTRL, PR_TAGGED_ADDR_ENABLE, 0, 0, 0)``
91   returns ``-EINVAL`` if the AArch64 Tagged Address ABI is globally
92   disabled by ``sysctl abi.tagged_addr_disabled=1``. The default
93   ``sysctl abi.tagged_addr_disabled`` configuration is 0.
94
95When the AArch64 Tagged Address ABI is enabled for a thread, the
96following behaviours are guaranteed:
97
98- All syscalls except the cases mentioned in section 3 can accept any
99  valid tagged pointer.
100
101- The syscall behaviour is undefined for invalid tagged pointers: it may
102  result in an error code being returned, a (fatal) signal being raised,
103  or other modes of failure.
104
105- The syscall behaviour for a valid tagged pointer is the same as for
106  the corresponding untagged pointer.
107
108
109A definition of the meaning of tagged pointers on AArch64 can be found
110in Documentation/arm64/tagged-pointers.rst.
111
1123. AArch64 Tagged Address ABI Exceptions
113-----------------------------------------
114
115The following system call parameters must be untagged regardless of the
116ABI relaxation:
117
118- ``prctl()`` other than pointers to user data either passed directly or
119  indirectly as arguments to be accessed by the kernel.
120
121- ``ioctl()`` other than pointers to user data either passed directly or
122  indirectly as arguments to be accessed by the kernel.
123
124- ``shmat()`` and ``shmdt()``.
125
126- ``brk()`` (since kernel v5.6).
127
128- ``mmap()`` (since kernel v5.6).
129
130- ``mremap()``, the ``new_address`` argument (since kernel v5.6).
131
132Any attempt to use non-zero tagged pointers may result in an error code
133being returned, a (fatal) signal being raised, or other modes of
134failure.
135
1364. Example of correct usage
137---------------------------
138.. code-block:: c
139
140   #include <stdlib.h>
141   #include <string.h>
142   #include <unistd.h>
143   #include <sys/mman.h>
144   #include <sys/prctl.h>
145
146   #define PR_SET_TAGGED_ADDR_CTRL	55
147   #define PR_TAGGED_ADDR_ENABLE	(1UL << 0)
148
149   #define TAG_SHIFT		56
150
151   int main(void)
152   {
153   	int tbi_enabled = 0;
154   	unsigned long tag = 0;
155   	char *ptr;
156
157   	/* check/enable the tagged address ABI */
158   	if (!prctl(PR_SET_TAGGED_ADDR_CTRL, PR_TAGGED_ADDR_ENABLE, 0, 0, 0))
159   		tbi_enabled = 1;
160
161   	/* memory allocation */
162   	ptr = mmap(NULL, sysconf(_SC_PAGE_SIZE), PROT_READ | PROT_WRITE,
163   		   MAP_PRIVATE | MAP_ANONYMOUS, -1, 0);
164   	if (ptr == MAP_FAILED)
165   		return 1;
166
167   	/* set a non-zero tag if the ABI is available */
168   	if (tbi_enabled)
169   		tag = rand() & 0xff;
170   	ptr = (char *)((unsigned long)ptr | (tag << TAG_SHIFT));
171
172   	/* memory access to a tagged address */
173   	strcpy(ptr, "tagged pointer\n");
174
175   	/* syscall with a tagged pointer */
176   	write(1, ptr, strlen(ptr));
177
178   	return 0;
179   }
180