4 Author: Will Deacon <will.deacon@arm.com>
5 Date : 07 September 2012
7 This document is based on the ARM booting document by Russell King and
8 is relevant to all public releases of the AArch64 Linux kernel.
10 The AArch64 exception model is made up of a number of exception levels
11 (EL0 - EL3), with EL0 and EL1 having a secure and a non-secure
12 counterpart. EL2 is the hypervisor level and exists only in non-secure
13 mode. EL3 is the highest priority level and exists only in secure mode.
15 For the purposes of this document, we will use the term `boot loader'
16 simply to define all software that executes on the CPU(s) before control
17 is passed to the Linux kernel. This may include secure monitor and
18 hypervisor code, or it may just be a handful of instructions for
19 preparing a minimal boot environment.
21 Essentially, the boot loader should provide (as a minimum) the
24 1. Setup and initialise the RAM
25 2. Setup the device tree
26 3. Decompress the kernel image
27 4. Call the kernel image
30 1. Setup and initialise RAM
31 ---------------------------
33 Requirement: MANDATORY
35 The boot loader is expected to find and initialise all RAM that the
36 kernel will use for volatile data storage in the system. It performs
37 this in a machine dependent manner. (It may use internal algorithms
38 to automatically locate and size all RAM, or it may use knowledge of
39 the RAM in the machine, or any other method the boot loader designer
43 2. Setup the device tree
44 -------------------------
46 Requirement: MANDATORY
48 The device tree blob (dtb) must be placed on an 8-byte boundary and must
49 not exceed 2 megabytes in size. Since the dtb will be mapped cacheable
50 using blocks of up to 2 megabytes in size, it must not be placed within
51 any 2M region which must be mapped with any specific attributes.
53 NOTE: versions prior to v4.2 also require that the DTB be placed within
54 the 512 MB region starting at text_offset bytes below the kernel Image.
56 3. Decompress the kernel image
57 ------------------------------
61 The AArch64 kernel does not currently provide a decompressor and
62 therefore requires decompression (gzip etc.) to be performed by the boot
63 loader if a compressed Image target (e.g. Image.gz) is used. For
64 bootloaders that do not implement this requirement, the uncompressed
65 Image target is available instead.
68 4. Call the kernel image
69 ------------------------
71 Requirement: MANDATORY
73 The decompressed kernel image contains a 64-byte header as follows:
75 u32 code0; /* Executable code */
76 u32 code1; /* Executable code */
77 u64 text_offset; /* Image load offset, little endian */
78 u64 image_size; /* Effective Image size, little endian */
79 u64 flags; /* kernel flags, little endian */
80 u64 res2 = 0; /* reserved */
81 u64 res3 = 0; /* reserved */
82 u64 res4 = 0; /* reserved */
83 u32 magic = 0x644d5241; /* Magic number, little endian, "ARM\x64" */
84 u32 res5; /* reserved (used for PE COFF offset) */
89 - As of v3.17, all fields are little endian unless stated otherwise.
91 - code0/code1 are responsible for branching to stext.
93 - when booting through EFI, code0/code1 are initially skipped.
94 res5 is an offset to the PE header and the PE header has the EFI
95 entry point (efi_stub_entry). When the stub has done its work, it
96 jumps to code0 to resume the normal boot process.
98 - Prior to v3.17, the endianness of text_offset was not specified. In
99 these cases image_size is zero and text_offset is 0x80000 in the
100 endianness of the kernel. Where image_size is non-zero image_size is
101 little-endian and must be respected. Where image_size is zero,
102 text_offset can be assumed to be 0x80000.
104 - The flags field (introduced in v3.17) is a little-endian 64-bit field
106 Bit 0: Kernel endianness. 1 if BE, 0 if LE.
107 Bit 1-2: Kernel Page size.
114 - When image_size is zero, a bootloader should attempt to keep as much
115 memory as possible free for use by the kernel immediately after the
116 end of the kernel image. The amount of space required will vary
117 depending on selected features, and is effectively unbound.
119 The Image must be placed text_offset bytes from a 2MB aligned base
120 address near the start of usable system RAM and called there. Memory
121 below that base address is currently unusable by Linux, and therefore it
122 is strongly recommended that this location is the start of system RAM.
123 The region between the 2 MB aligned base address and the start of the
124 image has no special significance to the kernel, and may be used for
126 At least image_size bytes from the start of the image must be free for
129 Any memory described to the kernel (even that below the start of the
130 image) which is not marked as reserved from the kernel (e.g., with a
131 memreserve region in the device tree) will be considered as available to
134 Before jumping into the kernel, the following conditions must be met:
136 - Quiesce all DMA capable devices so that memory does not get
137 corrupted by bogus network packets or disk data. This will save
138 you many hours of debug.
140 - Primary CPU general-purpose register settings
141 x0 = physical address of device tree blob (dtb) in system RAM.
142 x1 = 0 (reserved for future use)
143 x2 = 0 (reserved for future use)
144 x3 = 0 (reserved for future use)
147 All forms of interrupts must be masked in PSTATE.DAIF (Debug, SError,
149 The CPU must be in either EL2 (RECOMMENDED in order to have access to
150 the virtualisation extensions) or non-secure EL1.
154 Instruction cache may be on or off.
155 The address range corresponding to the loaded kernel image must be
156 cleaned to the PoC. In the presence of a system cache or other
157 coherent masters with caches enabled, this will typically require
158 cache maintenance by VA rather than set/way operations.
159 System caches which respect the architected cache maintenance by VA
160 operations must be configured and may be enabled.
161 System caches which do not respect architected cache maintenance by VA
162 operations (not recommended) must be configured and disabled.
165 CNTFRQ must be programmed with the timer frequency and CNTVOFF must
166 be programmed with a consistent value on all CPUs. If entering the
167 kernel at EL1, CNTHCTL_EL2 must have EL1PCTEN (bit 0) set where
171 All CPUs to be booted by the kernel must be part of the same coherency
172 domain on entry to the kernel. This may require IMPLEMENTATION DEFINED
173 initialisation to enable the receiving of maintenance operations on
177 All writable architected system registers at the exception level where
178 the kernel image will be entered must be initialised by software at a
179 higher exception level to prevent execution in an UNKNOWN state.
181 For systems with a GICv3 interrupt controller to be used in v3 mode:
183 ICC_SRE_EL3.Enable (bit 3) must be initialiased to 0b1.
184 ICC_SRE_EL3.SRE (bit 0) must be initialised to 0b1.
185 - If the kernel is entered at EL1:
186 ICC.SRE_EL2.Enable (bit 3) must be initialised to 0b1
187 ICC_SRE_EL2.SRE (bit 0) must be initialised to 0b1.
188 - The DT or ACPI tables must describe a GICv3 interrupt controller.
190 For systems with a GICv3 interrupt controller to be used in
191 compatibility (v2) mode:
193 ICC_SRE_EL3.SRE (bit 0) must be initialised to 0b0.
194 - If the kernel is entered at EL1:
195 ICC_SRE_EL2.SRE (bit 0) must be initialised to 0b0.
196 - The DT or ACPI tables must describe a GICv2 interrupt controller.
198 The requirements described above for CPU mode, caches, MMUs, architected
199 timers, coherency and system registers apply to all CPUs. All CPUs must
200 enter the kernel in the same exception level.
202 The boot loader is expected to enter the kernel on each CPU in the
205 - The primary CPU must jump directly to the first instruction of the
206 kernel image. The device tree blob passed by this CPU must contain
207 an 'enable-method' property for each cpu node. The supported
208 enable-methods are described below.
210 It is expected that the bootloader will generate these device tree
211 properties and insert them into the blob prior to kernel entry.
213 - CPUs with a "spin-table" enable-method must have a 'cpu-release-addr'
214 property in their cpu node. This property identifies a
215 naturally-aligned 64-bit zero-initalised memory location.
217 These CPUs should spin outside of the kernel in a reserved area of
218 memory (communicated to the kernel by a /memreserve/ region in the
219 device tree) polling their cpu-release-addr location, which must be
220 contained in the reserved region. A wfe instruction may be inserted
221 to reduce the overhead of the busy-loop and a sev will be issued by
222 the primary CPU. When a read of the location pointed to by the
223 cpu-release-addr returns a non-zero value, the CPU must jump to this
224 value. The value will be written as a single 64-bit little-endian
225 value, so CPUs must convert the read value to their native endianness
226 before jumping to it.
228 - CPUs with a "psci" enable method should remain outside of
229 the kernel (i.e. outside of the regions of memory described to the
230 kernel in the memory node, or in a reserved area of memory described
231 to the kernel by a /memreserve/ region in the device tree). The
232 kernel will issue CPU_ON calls as described in ARM document number ARM
233 DEN 0022A ("Power State Coordination Interface System Software on ARM
234 processors") to bring CPUs into the kernel.
236 The device tree should contain a 'psci' node, as described in
237 Documentation/devicetree/bindings/arm/psci.txt.
239 - Secondary CPU general-purpose register settings
240 x0 = 0 (reserved for future use)
241 x1 = 0 (reserved for future use)
242 x2 = 0 (reserved for future use)
243 x3 = 0 (reserved for future use)