1=====================
2Booting AArch64 Linux
3=====================
4
5Author: Will Deacon <[email protected]>
6
7Date  : 07 September 2012
8
9This document is based on the ARM booting document by Russell King and
10is relevant to all public releases of the AArch64 Linux kernel.
11
12The AArch64 exception model is made up of a number of exception levels
13(EL0 - EL3), with EL0, EL1 and EL2 having a secure and a non-secure
14counterpart.  EL2 is the hypervisor level, EL3 is the highest priority
15level and exists only in secure mode. Both are architecturally optional.
16
17For the purposes of this document, we will use the term `boot loader`
18simply to define all software that executes on the CPU(s) before control
19is passed to the Linux kernel.  This may include secure monitor and
20hypervisor code, or it may just be a handful of instructions for
21preparing a minimal boot environment.
22
23Essentially, the boot loader should provide (as a minimum) the
24following:
25
261. Setup and initialise the RAM
272. Setup the device tree
283. Decompress the kernel image
294. Call the kernel image
30
31
321. Setup and initialise RAM
33---------------------------
34
35Requirement: MANDATORY
36
37The boot loader is expected to find and initialise all RAM that the
38kernel will use for volatile data storage in the system.  It performs
39this in a machine dependent manner.  (It may use internal algorithms
40to automatically locate and size all RAM, or it may use knowledge of
41the RAM in the machine, or any other method the boot loader designer
42sees fit.)
43
44For Arm Confidential Compute Realms this includes ensuring that all
45protected RAM has a Realm IPA state (RIPAS) of "RAM".
46
47
482. Setup the device tree
49-------------------------
50
51Requirement: MANDATORY
52
53The device tree blob (dtb) must be placed on an 8-byte boundary and must
54not exceed 2 megabytes in size. Since the dtb will be mapped cacheable
55using blocks of up to 2 megabytes in size, it must not be placed within
56any 2M region which must be mapped with any specific attributes.
57
58NOTE: versions prior to v4.2 also require that the DTB be placed within
59the 512 MB region starting at text_offset bytes below the kernel Image.
60
613. Decompress the kernel image
62------------------------------
63
64Requirement: OPTIONAL
65
66The AArch64 kernel does not currently provide a decompressor and
67therefore requires decompression (gzip etc.) to be performed by the boot
68loader if a compressed Image target (e.g. Image.gz) is used.  For
69bootloaders that do not implement this requirement, the uncompressed
70Image target is available instead.
71
72
734. Call the kernel image
74------------------------
75
76Requirement: MANDATORY
77
78The decompressed kernel image contains a 64-byte header as follows::
79
80  u32 code0;			/* Executable code */
81  u32 code1;			/* Executable code */
82  u64 text_offset;		/* Image load offset, little endian */
83  u64 image_size;		/* Effective Image size, little endian */
84  u64 flags;			/* kernel flags, little endian */
85  u64 res2	= 0;		/* reserved */
86  u64 res3	= 0;		/* reserved */
87  u64 res4	= 0;		/* reserved */
88  u32 magic	= 0x644d5241;	/* Magic number, little endian, "ARM\x64" */
89  u32 res5;			/* reserved (used for PE COFF offset) */
90
91
92Header notes:
93
94- As of v3.17, all fields are little endian unless stated otherwise.
95
96- code0/code1 are responsible for branching to stext.
97
98- when booting through EFI, code0/code1 are initially skipped.
99  res5 is an offset to the PE header and the PE header has the EFI
100  entry point (efi_stub_entry).  When the stub has done its work, it
101  jumps to code0 to resume the normal boot process.
102
103- Prior to v3.17, the endianness of text_offset was not specified.  In
104  these cases image_size is zero and text_offset is 0x80000 in the
105  endianness of the kernel.  Where image_size is non-zero image_size is
106  little-endian and must be respected.  Where image_size is zero,
107  text_offset can be assumed to be 0x80000.
108
109- The flags field (introduced in v3.17) is a little-endian 64-bit field
110  composed as follows:
111
112  ============= ===============================================================
113  Bit 0		Kernel endianness.  1 if BE, 0 if LE.
114  Bit 1-2	Kernel Page size.
115
116			* 0 - Unspecified.
117			* 1 - 4K
118			* 2 - 16K
119			* 3 - 64K
120  Bit 3		Kernel physical placement
121
122			0
123			  2MB aligned base should be as close as possible
124			  to the base of DRAM, since memory below it is not
125			  accessible via the linear mapping
126			1
127			  2MB aligned base such that all image_size bytes
128			  counted from the start of the image are within
129			  the 48-bit addressable range of physical memory
130  Bits 4-63	Reserved.
131  ============= ===============================================================
132
133- When image_size is zero, a bootloader should attempt to keep as much
134  memory as possible free for use by the kernel immediately after the
135  end of the kernel image. The amount of space required will vary
136  depending on selected features, and is effectively unbound.
137
138The Image must be placed text_offset bytes from a 2MB aligned base
139address anywhere in usable system RAM and called there. The region
140between the 2 MB aligned base address and the start of the image has no
141special significance to the kernel, and may be used for other purposes.
142At least image_size bytes from the start of the image must be free for
143use by the kernel.
144NOTE: versions prior to v4.6 cannot make use of memory below the
145physical offset of the Image so it is recommended that the Image be
146placed as close as possible to the start of system RAM.
147
148If an initrd/initramfs is passed to the kernel at boot, it must reside
149entirely within a 1 GB aligned physical memory window of up to 32 GB in
150size that fully covers the kernel Image as well.
151
152Any memory described to the kernel (even that below the start of the
153image) which is not marked as reserved from the kernel (e.g., with a
154memreserve region in the device tree) will be considered as available to
155the kernel.
156
157Before jumping into the kernel, the following conditions must be met:
158
159- Quiesce all DMA capable devices so that memory does not get
160  corrupted by bogus network packets or disk data.  This will save
161  you many hours of debug.
162
163- Primary CPU general-purpose register settings:
164
165    - x0 = physical address of device tree blob (dtb) in system RAM.
166    - x1 = 0 (reserved for future use)
167    - x2 = 0 (reserved for future use)
168    - x3 = 0 (reserved for future use)
169
170- CPU mode
171
172  All forms of interrupts must be masked in PSTATE.DAIF (Debug, SError,
173  IRQ and FIQ).
174  The CPU must be in non-secure state, either in EL2 (RECOMMENDED in order
175  to have access to the virtualisation extensions), or in EL1.
176
177- Caches, MMUs
178
179  The MMU must be off.
180
181  The instruction cache may be on or off, and must not hold any stale
182  entries corresponding to the loaded kernel image.
183
184  The address range corresponding to the loaded kernel image must be
185  cleaned to the PoC. In the presence of a system cache or other
186  coherent masters with caches enabled, this will typically require
187  cache maintenance by VA rather than set/way operations.
188  System caches which respect the architected cache maintenance by VA
189  operations must be configured and may be enabled.
190  System caches which do not respect architected cache maintenance by VA
191  operations (not recommended) must be configured and disabled.
192
193- Architected timers
194
195  CNTFRQ must be programmed with the timer frequency and CNTVOFF must
196  be programmed with a consistent value on all CPUs.  If entering the
197  kernel at EL1, CNTHCTL_EL2 must have EL1PCTEN (bit 0) set where
198  available.
199
200- Coherency
201
202  All CPUs to be booted by the kernel must be part of the same coherency
203  domain on entry to the kernel.  This may require IMPLEMENTATION DEFINED
204  initialisation to enable the receiving of maintenance operations on
205  each CPU.
206
207- System registers
208
209  All writable architected system registers at or below the exception
210  level where the kernel image will be entered must be initialised by
211  software at a higher exception level to prevent execution in an UNKNOWN
212  state.
213
214  For all systems:
215  - If EL3 is present:
216
217    - SCR_EL3.FIQ must have the same value across all CPUs the kernel is
218      executing on.
219    - The value of SCR_EL3.FIQ must be the same as the one present at boot
220      time whenever the kernel is executing.
221
222  - If EL3 is present and the kernel is entered at EL2:
223
224    - SCR_EL3.HCE (bit 8) must be initialised to 0b1.
225
226  For systems with a GICv3 interrupt controller to be used in v3 mode:
227  - If EL3 is present:
228
229      - ICC_SRE_EL3.Enable (bit 3) must be initialised to 0b1.
230      - ICC_SRE_EL3.SRE (bit 0) must be initialised to 0b1.
231      - ICC_CTLR_EL3.PMHE (bit 6) must be set to the same value across
232        all CPUs the kernel is executing on, and must stay constant
233        for the lifetime of the kernel.
234
235  - If the kernel is entered at EL1:
236
237      - ICC.SRE_EL2.Enable (bit 3) must be initialised to 0b1
238      - ICC_SRE_EL2.SRE (bit 0) must be initialised to 0b1.
239
240  - The DT or ACPI tables must describe a GICv3 interrupt controller.
241
242  For systems with a GICv3 interrupt controller to be used in
243  compatibility (v2) mode:
244
245  - If EL3 is present:
246
247      ICC_SRE_EL3.SRE (bit 0) must be initialised to 0b0.
248
249  - If the kernel is entered at EL1:
250
251      ICC_SRE_EL2.SRE (bit 0) must be initialised to 0b0.
252
253  - The DT or ACPI tables must describe a GICv2 interrupt controller.
254
255  For CPUs with pointer authentication functionality:
256
257  - If EL3 is present:
258
259    - SCR_EL3.APK (bit 16) must be initialised to 0b1
260    - SCR_EL3.API (bit 17) must be initialised to 0b1
261
262  - If the kernel is entered at EL1:
263
264    - HCR_EL2.APK (bit 40) must be initialised to 0b1
265    - HCR_EL2.API (bit 41) must be initialised to 0b1
266
267  For CPUs with Activity Monitors Unit v1 (AMUv1) extension present:
268
269  - If EL3 is present:
270
271    - CPTR_EL3.TAM (bit 30) must be initialised to 0b0
272    - CPTR_EL2.TAM (bit 30) must be initialised to 0b0
273    - AMCNTENSET0_EL0 must be initialised to 0b1111
274    - AMCNTENSET1_EL0 must be initialised to a platform specific value
275      having 0b1 set for the corresponding bit for each of the auxiliary
276      counters present.
277
278  - If the kernel is entered at EL1:
279
280    - AMCNTENSET0_EL0 must be initialised to 0b1111
281    - AMCNTENSET1_EL0 must be initialised to a platform specific value
282      having 0b1 set for the corresponding bit for each of the auxiliary
283      counters present.
284
285  For CPUs with the Fine Grained Traps (FEAT_FGT) extension present:
286
287  - If EL3 is present and the kernel is entered at EL2:
288
289    - SCR_EL3.FGTEn (bit 27) must be initialised to 0b1.
290
291  For CPUs with the Fine Grained Traps 2 (FEAT_FGT2) extension present:
292
293  - If EL3 is present and the kernel is entered at EL2:
294
295    - SCR_EL3.FGTEn2 (bit 59) must be initialised to 0b1.
296
297  For CPUs with support for HCRX_EL2 (FEAT_HCX) present:
298
299  - If EL3 is present and the kernel is entered at EL2:
300
301    - SCR_EL3.HXEn (bit 38) must be initialised to 0b1.
302
303  For CPUs with Advanced SIMD and floating point support:
304
305  - If EL3 is present:
306
307    - CPTR_EL3.TFP (bit 10) must be initialised to 0b0.
308
309  - If EL2 is present and the kernel is entered at EL1:
310
311    - CPTR_EL2.TFP (bit 10) must be initialised to 0b0.
312
313  For CPUs with the Scalable Vector Extension (FEAT_SVE) present:
314
315  - if EL3 is present:
316
317    - CPTR_EL3.EZ (bit 8) must be initialised to 0b1.
318
319    - ZCR_EL3.LEN must be initialised to the same value for all CPUs the
320      kernel is executed on.
321
322  - If the kernel is entered at EL1 and EL2 is present:
323
324    - CPTR_EL2.TZ (bit 8) must be initialised to 0b0.
325
326    - CPTR_EL2.ZEN (bits 17:16) must be initialised to 0b11.
327
328    - ZCR_EL2.LEN must be initialised to the same value for all CPUs the
329      kernel will execute on.
330
331  For CPUs with the Scalable Matrix Extension (FEAT_SME):
332
333  - If EL3 is present:
334
335    - CPTR_EL3.ESM (bit 12) must be initialised to 0b1.
336
337    - SCR_EL3.EnTP2 (bit 41) must be initialised to 0b1.
338
339    - SMCR_EL3.LEN must be initialised to the same value for all CPUs the
340      kernel will execute on.
341
342 - If the kernel is entered at EL1 and EL2 is present:
343
344    - CPTR_EL2.TSM (bit 12) must be initialised to 0b0.
345
346    - CPTR_EL2.SMEN (bits 25:24) must be initialised to 0b11.
347
348    - SCTLR_EL2.EnTP2 (bit 60) must be initialised to 0b1.
349
350    - SMCR_EL2.LEN must be initialised to the same value for all CPUs the
351      kernel will execute on.
352
353    - HWFGRTR_EL2.nTPIDR2_EL0 (bit 55) must be initialised to 0b01.
354
355    - HWFGWTR_EL2.nTPIDR2_EL0 (bit 55) must be initialised to 0b01.
356
357    - HWFGRTR_EL2.nSMPRI_EL1 (bit 54) must be initialised to 0b01.
358
359    - HWFGWTR_EL2.nSMPRI_EL1 (bit 54) must be initialised to 0b01.
360
361  For CPUs with the Scalable Matrix Extension FA64 feature (FEAT_SME_FA64):
362
363  - If EL3 is present:
364
365    - SMCR_EL3.FA64 (bit 31) must be initialised to 0b1.
366
367 - If the kernel is entered at EL1 and EL2 is present:
368
369    - SMCR_EL2.FA64 (bit 31) must be initialised to 0b1.
370
371  For CPUs with the Memory Tagging Extension feature (FEAT_MTE2):
372
373  - If EL3 is present:
374
375    - SCR_EL3.ATA (bit 26) must be initialised to 0b1.
376
377  - If the kernel is entered at EL1 and EL2 is present:
378
379    - HCR_EL2.ATA (bit 56) must be initialised to 0b1.
380
381  For CPUs with the Scalable Matrix Extension version 2 (FEAT_SME2):
382
383  - If EL3 is present:
384
385    - SMCR_EL3.EZT0 (bit 30) must be initialised to 0b1.
386
387 - If the kernel is entered at EL1 and EL2 is present:
388
389    - SMCR_EL2.EZT0 (bit 30) must be initialised to 0b1.
390
391  For CPUs with the Performance Monitors Extension (FEAT_PMUv3p9):
392
393 - If EL3 is present:
394
395    - MDCR_EL3.EnPM2 (bit 7) must be initialised to 0b1.
396
397 - If the kernel is entered at EL1 and EL2 is present:
398
399    - HDFGRTR2_EL2.nPMICNTR_EL0 (bit 2) must be initialised to 0b1.
400    - HDFGRTR2_EL2.nPMICFILTR_EL0 (bit 3) must be initialised to 0b1.
401    - HDFGRTR2_EL2.nPMUACR_EL1 (bit 4) must be initialised to 0b1.
402
403    - HDFGWTR2_EL2.nPMICNTR_EL0 (bit 2) must be initialised to 0b1.
404    - HDFGWTR2_EL2.nPMICFILTR_EL0 (bit 3) must be initialised to 0b1.
405    - HDFGWTR2_EL2.nPMUACR_EL1 (bit 4) must be initialised to 0b1.
406
407  For CPUs with Memory Copy and Memory Set instructions (FEAT_MOPS):
408
409  - If the kernel is entered at EL1 and EL2 is present:
410
411    - HCRX_EL2.MSCEn (bit 11) must be initialised to 0b1.
412
413    - HCRX_EL2.MCE2 (bit 10) must be initialised to 0b1 and the hypervisor
414      must handle MOPS exceptions as described in :ref:`arm64_mops_hyp`.
415
416  For CPUs with the Extended Translation Control Register feature (FEAT_TCR2):
417
418  - If EL3 is present:
419
420    - SCR_EL3.TCR2En (bit 43) must be initialised to 0b1.
421
422 - If the kernel is entered at EL1 and EL2 is present:
423
424    - HCRX_EL2.TCR2En (bit 14) must be initialised to 0b1.
425
426  For CPUs with the Stage 1 Permission Indirection Extension feature (FEAT_S1PIE):
427
428  - If EL3 is present:
429
430    - SCR_EL3.PIEn (bit 45) must be initialised to 0b1.
431
432  - If the kernel is entered at EL1 and EL2 is present:
433
434    - HFGRTR_EL2.nPIR_EL1 (bit 58) must be initialised to 0b1.
435
436    - HFGWTR_EL2.nPIR_EL1 (bit 58) must be initialised to 0b1.
437
438    - HFGRTR_EL2.nPIRE0_EL1 (bit 57) must be initialised to 0b1.
439
440    - HFGRWR_EL2.nPIRE0_EL1 (bit 57) must be initialised to 0b1.
441
442 - For CPUs with Guarded Control Stacks (FEAT_GCS):
443
444  - GCSCR_EL1 must be initialised to 0.
445
446  - GCSCRE0_EL1 must be initialised to 0.
447
448  - If EL3 is present:
449
450    - SCR_EL3.GCSEn (bit 39) must be initialised to 0b1.
451
452  - If EL2 is present:
453
454    - GCSCR_EL2 must be initialised to 0.
455
456 - If the kernel is entered at EL1 and EL2 is present:
457
458    - HCRX_EL2.GCSEn must be initialised to 0b1.
459
460    - HFGITR_EL2.nGCSEPP (bit 59) must be initialised to 0b1.
461
462    - HFGITR_EL2.nGCSSTR_EL1 (bit 58) must be initialised to 0b1.
463
464    - HFGITR_EL2.nGCSPUSHM_EL1 (bit 57) must be initialised to 0b1.
465
466    - HFGRTR_EL2.nGCS_EL1 (bit 53) must be initialised to 0b1.
467
468    - HFGRTR_EL2.nGCS_EL0 (bit 52) must be initialised to 0b1.
469
470    - HFGWTR_EL2.nGCS_EL1 (bit 53) must be initialised to 0b1.
471
472    - HFGWTR_EL2.nGCS_EL0 (bit 52) must be initialised to 0b1.
473
474 - For CPUs with debug architecture i.e FEAT_Debugv8pN (all versions):
475
476 - If EL3 is present:
477
478   - MDCR_EL3.TDA (bit 9) must be initialized to 0b0
479
480 - For CPUs with FEAT_PMUv3:
481
482 - If EL3 is present:
483
484   - MDCR_EL3.TPM (bit 6) must be initialized to 0b0
485
486The requirements described above for CPU mode, caches, MMUs, architected
487timers, coherency and system registers apply to all CPUs.  All CPUs must
488enter the kernel in the same exception level.  Where the values documented
489disable traps it is permissible for these traps to be enabled so long as
490those traps are handled transparently by higher exception levels as though
491the values documented were set.
492
493The boot loader is expected to enter the kernel on each CPU in the
494following manner:
495
496- The primary CPU must jump directly to the first instruction of the
497  kernel image.  The device tree blob passed by this CPU must contain
498  an 'enable-method' property for each cpu node.  The supported
499  enable-methods are described below.
500
501  It is expected that the bootloader will generate these device tree
502  properties and insert them into the blob prior to kernel entry.
503
504- CPUs with a "spin-table" enable-method must have a 'cpu-release-addr'
505  property in their cpu node.  This property identifies a
506  naturally-aligned 64-bit zero-initalised memory location.
507
508  These CPUs should spin outside of the kernel in a reserved area of
509  memory (communicated to the kernel by a /memreserve/ region in the
510  device tree) polling their cpu-release-addr location, which must be
511  contained in the reserved region.  A wfe instruction may be inserted
512  to reduce the overhead of the busy-loop and a sev will be issued by
513  the primary CPU.  When a read of the location pointed to by the
514  cpu-release-addr returns a non-zero value, the CPU must jump to this
515  value.  The value will be written as a single 64-bit little-endian
516  value, so CPUs must convert the read value to their native endianness
517  before jumping to it.
518
519- CPUs with a "psci" enable method should remain outside of
520  the kernel (i.e. outside of the regions of memory described to the
521  kernel in the memory node, or in a reserved area of memory described
522  to the kernel by a /memreserve/ region in the device tree).  The
523  kernel will issue CPU_ON calls as described in ARM document number ARM
524  DEN 0022A ("Power State Coordination Interface System Software on ARM
525  processors") to bring CPUs into the kernel.
526
527  The device tree should contain a 'psci' node, as described in
528  Documentation/devicetree/bindings/arm/psci.yaml.
529
530- Secondary CPU general-purpose register settings
531
532  - x0 = 0 (reserved for future use)
533  - x1 = 0 (reserved for future use)
534  - x2 = 0 (reserved for future use)
535  - x3 = 0 (reserved for future use)
536