1 Booting AArch64 Linux 2 ===================== 3 4Author: Will Deacon <will.deacon@arm.com> 5Date : 07 September 2012 6 7This document is based on the ARM booting document by Russell King and 8is relevant to all public releases of the AArch64 Linux kernel. 9 10The 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 12counterpart. EL2 is the hypervisor level and exists only in non-secure 13mode. EL3 is the highest priority level and exists only in secure mode. 14 15For the purposes of this document, we will use the term `boot loader' 16simply to define all software that executes on the CPU(s) before control 17is passed to the Linux kernel. This may include secure monitor and 18hypervisor code, or it may just be a handful of instructions for 19preparing a minimal boot environment. 20 21Essentially, the boot loader should provide (as a minimum) the 22following: 23 241. Setup and initialise the RAM 252. Setup the device tree 263. Decompress the kernel image 274. Call the kernel image 28 29 301. Setup and initialise RAM 31--------------------------- 32 33Requirement: MANDATORY 34 35The boot loader is expected to find and initialise all RAM that the 36kernel will use for volatile data storage in the system. It performs 37this in a machine dependent manner. (It may use internal algorithms 38to automatically locate and size all RAM, or it may use knowledge of 39the RAM in the machine, or any other method the boot loader designer 40sees fit.) 41 42 432. Setup the device tree 44------------------------- 45 46Requirement: MANDATORY 47 48The device tree blob (dtb) must be placed on an 8-byte boundary within 49the first 512 megabytes from the start of the kernel image and must not 50cross a 2-megabyte boundary. This is to allow the kernel to map the 51blob using a single section mapping in the initial page tables. 52 53 543. Decompress the kernel image 55------------------------------ 56 57Requirement: OPTIONAL 58 59The AArch64 kernel does not currently provide a decompressor and 60therefore requires decompression (gzip etc.) to be performed by the boot 61loader if a compressed Image target (e.g. Image.gz) is used. For 62bootloaders that do not implement this requirement, the uncompressed 63Image target is available instead. 64 65 664. Call the kernel image 67------------------------ 68 69Requirement: MANDATORY 70 71The decompressed kernel image contains a 64-byte header as follows: 72 73 u32 code0; /* Executable code */ 74 u32 code1; /* Executable code */ 75 u64 text_offset; /* Image load offset, little endian */ 76 u64 image_size; /* Effective Image size, little endian */ 77 u64 flags; /* kernel flags, little endian */ 78 u64 res2 = 0; /* reserved */ 79 u64 res3 = 0; /* reserved */ 80 u64 res4 = 0; /* reserved */ 81 u32 magic = 0x644d5241; /* Magic number, little endian, "ARM\x64" */ 82 u32 res5; /* reserved (used for PE COFF offset) */ 83 84 85Header notes: 86 87- As of v3.17, all fields are little endian unless stated otherwise. 88 89- code0/code1 are responsible for branching to stext. 90 91- when booting through EFI, code0/code1 are initially skipped. 92 res5 is an offset to the PE header and the PE header has the EFI 93 entry point (efi_stub_entry). When the stub has done its work, it 94 jumps to code0 to resume the normal boot process. 95 96- Prior to v3.17, the endianness of text_offset was not specified. In 97 these cases image_size is zero and text_offset is 0x80000 in the 98 endianness of the kernel. Where image_size is non-zero image_size is 99 little-endian and must be respected. Where image_size is zero, 100 text_offset can be assumed to be 0x80000. 101 102- The flags field (introduced in v3.17) is a little-endian 64-bit field 103 composed as follows: 104 Bit 0: Kernel endianness. 1 if BE, 0 if LE. 105 Bits 1-63: Reserved. 106 107- When image_size is zero, a bootloader should attempt to keep as much 108 memory as possible free for use by the kernel immediately after the 109 end of the kernel image. The amount of space required will vary 110 depending on selected features, and is effectively unbound. 111 112The Image must be placed text_offset bytes from a 2MB aligned base 113address near the start of usable system RAM and called there. Memory 114below that base address is currently unusable by Linux, and therefore it 115is strongly recommended that this location is the start of system RAM. 116At least image_size bytes from the start of the image must be free for 117use by the kernel. 118 119Any memory described to the kernel (even that below the 2MB aligned base 120address) which is not marked as reserved from the kernel e.g. with a 121memreserve region in the device tree) will be considered as available to 122the kernel. 123 124Before jumping into the kernel, the following conditions must be met: 125 126- Quiesce all DMA capable devices so that memory does not get 127 corrupted by bogus network packets or disk data. This will save 128 you many hours of debug. 129 130- Primary CPU general-purpose register settings 131 x0 = physical address of device tree blob (dtb) in system RAM. 132 x1 = 0 (reserved for future use) 133 x2 = 0 (reserved for future use) 134 x3 = 0 (reserved for future use) 135 136- CPU mode 137 All forms of interrupts must be masked in PSTATE.DAIF (Debug, SError, 138 IRQ and FIQ). 139 The CPU must be in either EL2 (RECOMMENDED in order to have access to 140 the virtualisation extensions) or non-secure EL1. 141 142- Caches, MMUs 143 The MMU must be off. 144 Instruction cache may be on or off. 145 The address range corresponding to the loaded kernel image must be 146 cleaned to the PoC. In the presence of a system cache or other 147 coherent masters with caches enabled, this will typically require 148 cache maintenance by VA rather than set/way operations. 149 System caches which respect the architected cache maintenance by VA 150 operations must be configured and may be enabled. 151 System caches which do not respect architected cache maintenance by VA 152 operations (not recommended) must be configured and disabled. 153 154- Architected timers 155 CNTFRQ must be programmed with the timer frequency and CNTVOFF must 156 be programmed with a consistent value on all CPUs. If entering the 157 kernel at EL1, CNTHCTL_EL2 must have EL1PCTEN (bit 0) set where 158 available. 159 160- Coherency 161 All CPUs to be booted by the kernel must be part of the same coherency 162 domain on entry to the kernel. This may require IMPLEMENTATION DEFINED 163 initialisation to enable the receiving of maintenance operations on 164 each CPU. 165 166- System registers 167 All writable architected system registers at the exception level where 168 the kernel image will be entered must be initialised by software at a 169 higher exception level to prevent execution in an UNKNOWN state. 170 171 For systems with a GICv3 interrupt controller: 172 - If EL3 is present: 173 ICC_SRE_EL3.Enable (bit 3) must be initialiased to 0b1. 174 ICC_SRE_EL3.SRE (bit 0) must be initialised to 0b1. 175 - If the kernel is entered at EL1: 176 ICC.SRE_EL2.Enable (bit 3) must be initialised to 0b1 177 ICC_SRE_EL2.SRE (bit 0) must be initialised to 0b1. 178 179The requirements described above for CPU mode, caches, MMUs, architected 180timers, coherency and system registers apply to all CPUs. All CPUs must 181enter the kernel in the same exception level. 182 183The boot loader is expected to enter the kernel on each CPU in the 184following manner: 185 186- The primary CPU must jump directly to the first instruction of the 187 kernel image. The device tree blob passed by this CPU must contain 188 an 'enable-method' property for each cpu node. The supported 189 enable-methods are described below. 190 191 It is expected that the bootloader will generate these device tree 192 properties and insert them into the blob prior to kernel entry. 193 194- CPUs with a "spin-table" enable-method must have a 'cpu-release-addr' 195 property in their cpu node. This property identifies a 196 naturally-aligned 64-bit zero-initalised memory location. 197 198 These CPUs should spin outside of the kernel in a reserved area of 199 memory (communicated to the kernel by a /memreserve/ region in the 200 device tree) polling their cpu-release-addr location, which must be 201 contained in the reserved region. A wfe instruction may be inserted 202 to reduce the overhead of the busy-loop and a sev will be issued by 203 the primary CPU. When a read of the location pointed to by the 204 cpu-release-addr returns a non-zero value, the CPU must jump to this 205 value. The value will be written as a single 64-bit little-endian 206 value, so CPUs must convert the read value to their native endianness 207 before jumping to it. 208 209- CPUs with a "psci" enable method should remain outside of 210 the kernel (i.e. outside of the regions of memory described to the 211 kernel in the memory node, or in a reserved area of memory described 212 to the kernel by a /memreserve/ region in the device tree). The 213 kernel will issue CPU_ON calls as described in ARM document number ARM 214 DEN 0022A ("Power State Coordination Interface System Software on ARM 215 processors") to bring CPUs into the kernel. 216 217 The device tree should contain a 'psci' node, as described in 218 Documentation/devicetree/bindings/arm/psci.txt. 219 220- Secondary CPU general-purpose register settings 221 x0 = 0 (reserved for future use) 222 x1 = 0 (reserved for future use) 223 x2 = 0 (reserved for future use) 224 x3 = 0 (reserved for future use) 225