Context Navigation

source: soft/giet_vm/giet_boot/boot.c @ 566

Last change on this file since 566 was 564, checked in by alain, 10 years ago
Cosmetic.
File size: 78.1 KB

Line
1	///////////////////////////////////////////////////////////////////////////////////
2	// File : boot.c
3	// Date : 01/11/2013
4	// Author : alain greiner
5	// Copyright (c) UPMC-LIP6
6	///////////////////////////////////////////////////////////////////////////////////
7	// The boot.c file contains the bootloader for the GIET-VM static OS.
8	//
9	// This code has been written for the MIPS32 processor.
10	// The virtual adresses are on 32 bits and use the (unsigned int) type. The
11	// physicals addresses can have up to 40 bits, and use type (unsigned long long).
12	// It natively supports clusterised shared memory multi-processors architectures,
13	// where each processor is identified by a composite index [x,y,p],
14	// and where there is one physical memory bank per cluster.
15	//
16	// The boot.elf file is stored on disk and is loaded into memory by proc[0,0,0],
17	// executing the generic preloader (stored in ROM). The boot-loader code itself
18	// is executed in parallel by all proc[x,y,0], and performs the following tasks:
19	// - load into memory various binary files, from a FAT32 file system.
20	// - build the various page tables (one page table per vspace).
21	// - initialize the shedulers (one scheduler per processor).
22	// - initialize the external peripherals.
23	//
24	// 1) The binary files to be loaded are:
25	// - the "map.bin" file contains the hardware architecture description and the
26	// mapping directives. It must be stored in the the seg_boot_mapping segment
27	// (at address SEG_BOOT_MAPPING_BASE defined in hard_config.h file).
28	// - the "kernel.elf" file contains the kernel binary code and data.
29	// - the various "application.elf" files.
30	//
31	// 2) The "map.bin" file contains the C binary structure defining:
32	// - the hardware architecture: number of clusters, number or processors,
33	// size of the memory segments, and peripherals in each cluster.
34	// - The structure of the various multi-threaded software applications:
35	// number of tasks, communication channels.
36	// - The mapping: placement of virtual segments (vseg) in the physical
37	// segments (pseg), placement of software tasks on the processors,
38	//
39	// 3) The GIET-VM uses the paged virtual memory to provides two services:
40	// - classical memory protection, when several independant applications compiled
41	// in different virtual spaces are executing on the same hardware platform.
42	// - data placement in NUMA architectures, to control the placement
43	// of the software objects (vsegs) on the physical memory banks (psegs).
44	//
45	// The max number of vspaces (GIET_NB_VSPACE_MAX) is a configuration parameter.
46	// For each application, the tasks are statically allocateded on processors.
47	// The page table are statically build in the boot phase, and they do not
48	// change during execution.
49	// The GIET_VM uses both small pages (4 Kbytes), and big pages (2 Mbytes).
50	//
51	// Each page table (one page table per virtual space) is monolithic, and
52	// contains one PT1 (8 Kbytes) and a variable number of PT2s (4 Kbytes each).
53	// For each vspace, the number of PT2s is defined by the size of the PTAB vseg
54	// in the mapping.
55	// The PT1 is indexed by the ix1 field (11 bits) of the VPN. An entry is 32 bits.
56	// A PT2 is indexed the ix2 field (9 bits) of the VPN. An entry is 64 bits.
57	// The first word contains the flags, the second word contains the PPN.
58	// The page tables are distributed/replicated in all clusters.
59	///////////////////////////////////////////////////////////////////////////////////
60	// Implementation Notes:
61	//
62	// 1) The cluster_id variable is a linear index in the mapping_info array.
63	// The cluster_xy variable is the tological index = x << Y_WIDTH + y
64	//
65	// 2) We set the _tty0_boot_mode variable to force the _printf() function to use
66	// the tty0_spin_lock for exclusive access to TTY0.
67	///////////////////////////////////////////////////////////////////////////////////
68
69	#include <giet_config.h>
70	#include <hard_config.h>
71	#include <mapping_info.h>
72	#include <kernel_malloc.h>
73	#include <memspace.h>
74	#include <tty_driver.h>
75	#include <xcu_driver.h>
76	#include <bdv_driver.h>
77	#include <hba_driver.h>
78	#include <sdc_driver.h>
79	#include <cma_driver.h>
80	#include <nic_driver.h>
81	#include <iob_driver.h>
82	#include <pic_driver.h>
83	#include <mwr_driver.h>
84	#include <dma_driver.h>
85	#include <ctx_handler.h>
86	#include <irq_handler.h>
87	#include <vmem.h>
88	#include <pmem.h>
89	#include <utils.h>
90	#include <tty0.h>
91	#include <kernel_locks.h>
92	#include <kernel_barriers.h>
93	#include <elf-types.h>
94	#include <fat32.h>
95	#include <mips32_registers.h>
96	#include <stdarg.h>
97
98	#if !defined(X_SIZE)
99	# error: The X_SIZE value must be defined in the 'hard_config.h' file !
100	#endif
101
102	#if !defined(Y_SIZE)
103	# error: The Y_SIZE value must be defined in the 'hard_config.h' file !
104	#endif
105
106	#if !defined(X_WIDTH)
107	# error: The X_WIDTH value must be defined in the 'hard_config.h' file !
108	#endif
109
110	#if !defined(Y_WIDTH)
111	# error: The Y_WIDTH value must be defined in the 'hard_config.h' file !
112	#endif
113
114	#if !defined(SEG_BOOT_MAPPING_BASE)
115	# error: The SEG_BOOT_MAPPING_BASE value must be defined in the hard_config.h file !
116	#endif
117
118	#if !defined(NB_PROCS_MAX)
119	# error: The NB_PROCS_MAX value must be defined in the 'hard_config.h' file !
120	#endif
121
122	#if !defined(GIET_NB_VSPACE_MAX)
123	# error: The GIET_NB_VSPACE_MAX value must be defined in the 'giet_config.h' file !
124	#endif
125
126	#if !defined(GIET_ELF_BUFFER_SIZE)
127	# error: The GIET_ELF_BUFFER_SIZE value must be defined in the giet_config.h file !
128	#endif
129
130	////////////////////////////////////////////////////////////////////////////
131	// Global variables for boot code
132	////////////////////////////////////////////////////////////////////////////
133
134	// FAT internal representation for boot code
135	__attribute__((section(".kdata")))
136	fat32_fs_t _fat __attribute__((aligned(512)));
137
138	// Temporaty buffer used to load one complete .elf file
139	__attribute__((section(".kdata")))
140	char _boot_elf_buffer[GIET_ELF_BUFFER_SIZE] __attribute__((aligned(512)));
141
142	// Physical memory allocators array (one per cluster)
143	__attribute__((section(".kdata")))
144	pmem_alloc_t boot_pmem_alloc[X_SIZE][Y_SIZE];
145
146	// Distributed kernel heap (one per cluster)
147	// __attribute__((section(".kdata")))
148	// kernel_heap_t kernel_heap[X_SIZE][Y_SIZE];
149
150	// Schedulers virtual base addresses array (one per processor)
151	__attribute__((section(".kdata")))
152	static_scheduler_t* _schedulers[X_SIZE][Y_SIZE][NB_PROCS_MAX];
153
154	// Page tables virtual base addresses (one per vspace and per cluster)
155	__attribute__((section(".kdata")))
156	unsigned int _ptabs_vaddr[GIET_NB_VSPACE_MAX][X_SIZE][Y_SIZE];
157
158	// Page tables physical base addresses (one per vspace and per cluster)
159	__attribute__((section(".kdata")))
160	paddr_t _ptabs_paddr[GIET_NB_VSPACE_MAX][X_SIZE][Y_SIZE];
161
162	// Page tables pt2 allocators (one per vspace and per cluster)
163	__attribute__((section(".kdata")))
164	unsigned int _ptabs_next_pt2[GIET_NB_VSPACE_MAX][X_SIZE][Y_SIZE];
165
166	// Page tables max_pt2 (same value for all page tables)
167	__attribute__((section(".kdata")))
168	unsigned int _ptabs_max_pt2;
169
170	// boot code uses a spin lock to protect TTY0
171	__attribute__((section(".kdata")))
172	unsigned int _tty0_boot_mode = 1;
173
174	__attribute__((section(".kdata")))
175	spin_lock_t _ptabs_spin_lock[GIET_NB_VSPACE_MAX][X_SIZE][Y_SIZE];
176
177	// barrier used by boot code for parallel execution
178	__attribute__((section(".kdata")))
179	simple_barrier_t _barrier_all_clusters;
180
181	//////////////////////////////////////////////////////////////////////////////
182	// Extern variables
183	//////////////////////////////////////////////////////////////////////////////
184
185	// this variable is defined in the tty0.c file
186	extern spin_lock_t _tty0_spin_lock;
187
188	extern void boot_entry();
189
190	//////////////////////////////////////////////////////////////////////////////
191	// This function registers a new PTE1 in the page table defined
192	// by the vspace_id argument, and the (x,y) coordinates.
193	// It updates only the first level PT1.
194	// As each vseg is mapped by a different processor, the PT1 entry cannot
195	// be concurrently accessed, and we don't need to take any lock.
196	//////////////////////////////////////////////////////////////////////////////
197	void boot_add_pte1( unsigned int vspace_id,
198	unsigned int x,
199	unsigned int y,
200	unsigned int vpn, // 20 bits right-justified
201	unsigned int flags, // 10 bits left-justified
202	unsigned int ppn ) // 28 bits right-justified
203	{
204	// compute index in PT1
205	unsigned int ix1 = vpn >> 9; // 11 bits for ix1
206
207	// get page table physical base address
208	paddr_t pt1_pbase = _ptabs_paddr[vspace_id][x][y];
209
210	if ( pt1_pbase == 0 )
211	{
212	_printf("\n[BOOT ERROR] in boot_add_pte1() : no PTAB in cluster[%d,%d]"
213	" containing processors\n", x , y );
214	_exit();
215	}
216
217	// compute pte1 : 2 bits V T / 8 bits flags / 3 bits RSVD / 19 bits bppi
218	unsigned int pte1 = PTE_V \|
219	(flags & 0x3FC00000) \|
220	((ppn>>9) & 0x0007FFFF);
221
222	// write pte1 in PT1
223	_physical_write( pt1_pbase + 4*ix1, pte1 );
224
225	asm volatile ("sync");
226
227	} // end boot_add_pte1()
228
229	//////////////////////////////////////////////////////////////////////////////
230	// This function registers a new PTE2 in the page table defined
231	// by the vspace_id argument, and the (x,y) coordinates.
232	// It updates both the first level PT1 and the second level PT2.
233	// As the set of PT2s is implemented as a fixed size array (no dynamic
234	// allocation), this function checks a possible overflow of the PT2 array.
235	// As a given entry in PT1 can be shared by several vsegs, mapped by
236	// different processors, we need to take the lock protecting PTAB[v][x]y].
237	//////////////////////////////////////////////////////////////////////////////
238	void boot_add_pte2( unsigned int vspace_id,
239	unsigned int x,
240	unsigned int y,
241	unsigned int vpn, // 20 bits right-justified
242	unsigned int flags, // 10 bits left-justified
243	unsigned int ppn ) // 28 bits right-justified
244	{
245	unsigned int ix1;
246	unsigned int ix2;
247	paddr_t pt2_pbase; // PT2 physical base address
248	paddr_t pte2_paddr; // PTE2 physical address
249	unsigned int pt2_id; // PT2 index
250	unsigned int ptd; // PTD : entry in PT1
251
252	ix1 = vpn >> 9; // 11 bits for ix1
253	ix2 = vpn & 0x1FF; // 9 bits for ix2
254
255	// get page table physical base address
256	paddr_t pt1_pbase = _ptabs_paddr[vspace_id][x][y];
257
258	if ( pt1_pbase == 0 )
259	{
260	_printf("\n[BOOT ERROR] in boot_add_pte2() : no PTAB for vspace %d "
261	"in cluster[%d,%d]\n", vspace_id , x , y );
262	_exit();
263	}
264
265	// get lock protecting PTAB[vspace_id][x][y]
266	_spin_lock_acquire( &_ptabs_spin_lock[vspace_id][x][y] );
267
268	// get ptd in PT1
269	ptd = _physical_read( pt1_pbase + 4 * ix1 );
270
271	if ((ptd & PTE_V) == 0) // undefined PTD: compute PT2 base address,
272	// and set a new PTD in PT1
273	{
274	// get a new pt2_id
275	pt2_id = _ptabs_next_pt2[vspace_id][x][y];
276	_ptabs_next_pt2[vspace_id][x][y] = pt2_id + 1;
277
278	// check overflow
279	if (pt2_id == _ptabs_max_pt2)
280	{
281	_printf("\n[BOOT ERROR] in boot_add_pte2() : PTAB[%d,%d,%d]"
282	" contains not enough PT2s\n", vspace_id, x, y );
283	_exit();
284	}
285
286	pt2_pbase = pt1_pbase + PT1_SIZE + PT2_SIZE * pt2_id;
287	ptd = PTE_V \| PTE_T \| (unsigned int) (pt2_pbase >> 12);
288
289	// set PTD into PT1
290	_physical_write( pt1_pbase + 4*ix1, ptd);
291	}
292	else // valid PTD: compute PT2 base address
293	{
294	pt2_pbase = ((paddr_t)(ptd & 0x0FFFFFFF)) << 12;
295	}
296
297	// set PTE in PT2 : flags & PPN in two 32 bits words
298	pte2_paddr = pt2_pbase + 8 * ix2;
299	_physical_write(pte2_paddr , (PTE_V \| flags) );
300	_physical_write(pte2_paddr + 4 , ppn );
301
302	// release lock protecting PTAB[vspace_id][x][y]
303	_spin_lock_release( &_ptabs_spin_lock[vspace_id][x][y] );
304
305	asm volatile ("sync");
306
307	} // end boot_add_pte2()
308
309	////////////////////////////////////////////////////////////////////////////////////
310	// Align the value of paddr or vaddr to the required alignement,
311	// defined by alignPow2 == L2(alignement).
312	////////////////////////////////////////////////////////////////////////////////////
313	paddr_t paddr_align_to( paddr_t paddr, unsigned int alignPow2 )
314	{
315	paddr_t mask = (1 << alignPow2) - 1;
316	return ((paddr + mask) & ~mask);
317	}
318
319	unsigned int vaddr_align_to( unsigned int vaddr, unsigned int alignPow2 )
320	{
321	unsigned int mask = (1 << alignPow2) - 1;
322	return ((vaddr + mask) & ~mask);
323	}
324
325	/////////////////////////////////////////////////////////////////////////////////////
326	// This function map a vseg identified by the vseg pointer.
327	//
328	// A given vseg can be mapped in a Big Physical Pages (BPP: 2 Mbytes) or in a
329	// Small Physical Pages (SPP: 4 Kbytes), depending on the "big" attribute of vseg,
330	// with the following rules:
331	// - SPP : There is only one vseg in a small physical page, but a single vseg
332	// can cover several contiguous small physical pages.
333	// - BPP : It can exist several vsegs in a single big physical page, and a single
334	// vseg can cover several contiguous big physical pages.
335	//
336	// 1) First step: it computes various vseg attributes and checks
337	// alignment constraints.
338	//
339	// 2) Second step: it allocates the required number of contiguous physical pages,
340	// computes the physical base address (if the vseg is not identity mapping),
341	// and register it in the vseg pbase field.
342	// Only the vsegs used by the boot code and the peripheral vsegs
343	// can be identity mapping. The first big physical page in cluster[0,0]
344	// is reserved for the boot vsegs.
345	//
346	// 3) Third step (only for vseg that have the VSEG_TYPE_PTAB): the M page tables
347	// associated to the M vspaces must be packed in the same vseg.
348	// We divide this vseg in M sub-segments, and compute the vbase and pbase
349	// addresses for M page tables, and register these addresses in the _ptabs_paddr
350	// and _ptabs_vaddr arrays.
351	//
352	/////////////////////////////////////////////////////////////////////////////////////
353	void boot_vseg_map( mapping_vseg_t* vseg,
354	unsigned int vspace_id )
355	{
356	mapping_header_t* header = (mapping_header_t *)SEG_BOOT_MAPPING_BASE;
357	mapping_cluster_t* cluster = _get_cluster_base(header);
358	mapping_pseg_t* pseg = _get_pseg_base(header);
359
360	//////////// First step : compute vseg attributes
361
362	// compute destination cluster pointer & coordinates
363	pseg = pseg + vseg->psegid;
364	cluster = cluster + pseg->clusterid;
365	unsigned int x_dest = cluster->x;
366	unsigned int y_dest = cluster->y;
367
368	// compute the "big" vseg attribute
369	unsigned int big = vseg->big;
370
371	// all vsegs must be aligned on 4Kbytes
372	if ( vseg->vbase & 0x00000FFF )
373	{
374	_printf("\n[BOOT ERROR] vseg %s not aligned : vbase = %x\n",
375	vseg->name, vseg->vbase );
376	_exit();
377	}
378
379	// compute the "is_ram" vseg attribute
380	unsigned int is_ram;
381	if ( pseg->type == PSEG_TYPE_RAM ) is_ram = 1;
382	else is_ram = 0;
383
384	// compute the "is_ptab" attribute
385	unsigned int is_ptab;
386	if ( vseg->type == VSEG_TYPE_PTAB ) is_ptab = 1;
387	else is_ptab = 0;
388
389	// compute actual vspace index
390	unsigned int vsid;
391	if ( vspace_id == 0xFFFFFFFF ) vsid = 0;
392	else vsid = vspace_id;
393
394	//////////// Second step : compute ppn and npages
395	//////////// - if identity mapping : ppn <= vpn
396	//////////// - if vseg is periph : ppn <= pseg.base >> 12
397	//////////// - if vseg is ram : ppn <= physical memory allocator
398
399	unsigned int ppn; // first physical page index (28 bits = \|x\|y\|bppi\|sppi\|)
400	unsigned int vpn; // first virtual page index (20 bits = \|ix1\|ix2\|)
401	unsigned int vpn_max; // last virtual page index (20 bits = \|ix1\|ix2\|)
402
403	vpn = vseg->vbase >> 12;
404	vpn_max = (vseg->vbase + vseg->length - 1) >> 12;
405
406	// compute npages
407	unsigned int npages; // number of required (big or small) pages
408	if ( big == 0 ) npages = vpn_max - vpn + 1; // number of small pages
409	else npages = (vpn_max>>9) - (vpn>>9) + 1; // number of big pages
410
411	// compute ppn
412	if ( vseg->ident ) // identity mapping
413	{
414	ppn = vpn;
415	}
416	else // not identity mapping
417	{
418	if ( is_ram ) // RAM : physical memory allocation required
419	{
420	// compute pointer on physical memory allocator in dest cluster
421	pmem_alloc_t* palloc = &boot_pmem_alloc[x_dest][y_dest];
422
423	if ( big == 0 ) // SPP : small physical pages
424	{
425	// allocate contiguous small physical pages
426	ppn = _get_small_ppn( palloc, npages );
427	}
428	else // BPP : big physical pages
429	{
430
431	// one big page can be shared by several vsegs
432	// we must chek if BPP already allocated
433	if ( is_ptab ) // It cannot be mapped
434	{
435	ppn = _get_big_ppn( palloc, npages );
436	}
437	else // It can be mapped
438	{
439	unsigned int ix1 = vpn >> 9; // 11 bits
440	paddr_t paddr = _ptabs_paddr[vsid][x_dest][y_dest] + (ix1<<2);
441	unsigned int pte1 = _physical_read( paddr );
442
443	if ( (pte1 & PTE_V) == 0 ) // BPP not allocated yet
444	{
445	// allocate contiguous big physical pages
446	ppn = _get_big_ppn( palloc, npages );
447	}
448	else // BPP already allocated
449	{
450	// test if new vseg has the same mode bits than
451	// the other vsegs in the same big page
452	unsigned int pte1_mode = 0;
453	if (pte1 & PTE_C) pte1_mode \|= C_MODE_MASK;
454	if (pte1 & PTE_X) pte1_mode \|= X_MODE_MASK;
455	if (pte1 & PTE_W) pte1_mode \|= W_MODE_MASK;
456	if (pte1 & PTE_U) pte1_mode \|= U_MODE_MASK;
457	if (vseg->mode != pte1_mode)
458	{
459	_printf("\n[BOOT ERROR] in boot_vseg_map() : "
460	"vseg %s has different flags than another vseg "
461	"in the same BPP\n", vseg->name );
462	_exit();
463	}
464	ppn = ((pte1 << 9) & 0x0FFFFE00);
465	}
466	}
467	ppn = ppn \| (vpn & 0x1FF);
468	}
469	}
470	else // PERI : no memory allocation required
471	{
472	ppn = pseg->base >> 12;
473	}
474	}
475
476	// update vseg.pbase field and update vsegs chaining
477	vseg->pbase = ((paddr_t)ppn) << 12;
478	vseg->mapped = 1;
479
480
481	//////////// Third step : (only if the vseg is a page table)
482	//////////// - compute the physical & virtual base address for each vspace
483	//////////// by dividing the vseg in several sub-segments.
484	//////////// - register it in _ptabs_vaddr & _ptabs_paddr arrays,
485	//////////// and initialize next_pt2 allocators.
486	//////////// - reset all entries in first level page tables
487
488	if ( is_ptab )
489	{
490	unsigned int vs; // vspace index
491	unsigned int nspaces; // number of vspaces
492	unsigned int nsp; // number of small pages for one PTAB
493	unsigned int offset; // address offset for current PTAB
494
495	nspaces = header->vspaces;
496	offset = 0;
497
498	// each PTAB must be aligned on a 8 Kbytes boundary
499	nsp = ( vseg->length >> 12 ) / nspaces;
500	if ( (nsp & 0x1) == 0x1 ) nsp = nsp - 1;
501
502	// compute max_pt2
503	_ptabs_max_pt2 = ((nsp<<12) - PT1_SIZE) / PT2_SIZE;
504
505	for ( vs = 0 ; vs < nspaces ; vs++ )
506	{
507	_ptabs_vaddr [vs][x_dest][y_dest] = (vpn + offset) << 12;
508	_ptabs_paddr [vs][x_dest][y_dest] = ((paddr_t)(ppn + offset)) << 12;
509	_ptabs_next_pt2[vs][x_dest][y_dest] = 0;
510	offset += nsp;
511
512	// reset all entries in PT1 (8 Kbytes)
513	_physical_memset( _ptabs_paddr[vs][x_dest][y_dest], PT1_SIZE, 0 );
514	}
515	}
516
517	asm volatile ("sync");
518
519	#if BOOT_DEBUG_PT
520	if ( big )
521	_printf("\n[BOOT] vseg %s : cluster[%d,%d] / "
522	"vbase = %x / length = %x / BIG / npages = %d / pbase = %l\n",
523	vseg->name, x_dest, y_dest, vseg->vbase, vseg->length, npages, vseg-> pbase );
524	else
525	_printf("\n[BOOT] vseg %s : cluster[%d,%d] / "
526	"vbase = %x / length = %x / SMALL / npages = %d / pbase = %l\n",
527	vseg->name, x_dest, y_dest, vseg->vbase, vseg->length, npages, vseg-> pbase );
528	#endif
529
530	} // end boot_vseg_map()
531
532	/////////////////////////////////////////////////////////////////////////////////////
533	// For the vseg defined by the vseg pointer, this function register PTEs
534	// in one or several page tables.
535	// It is a global vseg (kernel vseg) if (vspace_id == 0xFFFFFFFF).
536	// The number of involved PTABs depends on the "local" and "global" attributes:
537	// - PTEs are replicated in all vspaces for a global vseg.
538	// - PTEs are replicated in all clusters containing procs for a non local vseg.
539	/////////////////////////////////////////////////////////////////////////////////////
540	void boot_vseg_pte( mapping_vseg_t* vseg,
541	unsigned int vspace_id )
542	{
543	// compute the "global" vseg attribute and actual vspace index
544	unsigned int global;
545	unsigned int vsid;
546	if ( vspace_id == 0xFFFFFFFF )
547	{
548	global = 1;
549	vsid = 0;
550	}
551	else
552	{
553	global = 0;
554	vsid = vspace_id;
555	}
556
557	// compute the "local" and "big" attributes
558	unsigned int local = vseg->local;
559	unsigned int big = vseg->big;
560
561	// compute vseg flags
562	// The three flags (Local, Remote and Dirty) are set to 1
563	// to avoid hardware update for these flags, because GIET_VM
564	// does use these flags.
565	unsigned int flags = 0;
566	if (vseg->mode & C_MODE_MASK) flags \|= PTE_C;
567	if (vseg->mode & X_MODE_MASK) flags \|= PTE_X;
568	if (vseg->mode & W_MODE_MASK) flags \|= PTE_W;
569	if (vseg->mode & U_MODE_MASK) flags \|= PTE_U;
570	if ( global ) flags \|= PTE_G;
571	flags \|= PTE_L;
572	flags \|= PTE_R;
573	flags \|= PTE_D;
574
575	// compute VPN, PPN and number of pages (big or small)
576	unsigned int vpn = vseg->vbase >> 12;
577	unsigned int vpn_max = (vseg->vbase + vseg->length - 1) >> 12;
578	unsigned int ppn = (unsigned int)(vseg->pbase >> 12);
579	unsigned int npages;
580	if ( big == 0 ) npages = vpn_max - vpn + 1;
581	else npages = (vpn_max>>9) - (vpn>>9) + 1;
582
583	// compute destination cluster coordinates, for local vsegs
584	mapping_header_t* header = (mapping_header_t *)SEG_BOOT_MAPPING_BASE;
585	mapping_cluster_t* cluster = _get_cluster_base(header);
586	mapping_pseg_t* pseg = _get_pseg_base(header);
587	mapping_pseg_t* pseg_dest = &pseg[vseg->psegid];
588	mapping_cluster_t* cluster_dest = &cluster[pseg_dest->clusterid];
589	unsigned int x_dest = cluster_dest->x;
590	unsigned int y_dest = cluster_dest->y;
591
592	unsigned int p; // iterator for physical page index
593	unsigned int x; // iterator for cluster x coordinate
594	unsigned int y; // iterator for cluster y coordinate
595	unsigned int v; // iterator for vspace index
596
597	// loop on PTEs
598	for ( p = 0 ; p < npages ; p++ )
599	{
600	if ( (local != 0) && (global == 0) ) // one cluster / one vspace
601	{
602	if ( big ) // big pages => PTE1s
603	{
604	boot_add_pte1( vsid,
605	x_dest,
606	y_dest,
607	vpn + (p<<9),
608	flags,
609	ppn + (p<<9) );
610	}
611	else // small pages => PTE2s
612	{
613	boot_add_pte2( vsid,
614	x_dest,
615	y_dest,
616	vpn + p,
617	flags,
618	ppn + p );
619	}
620	}
621	else if ( (local == 0) && (global == 0) ) // all clusters / one vspace
622	{
623	for ( x = 0 ; x < X_SIZE ; x++ )
624	{
625	for ( y = 0 ; y < Y_SIZE ; y++ )
626	{
627	if ( cluster[(x * Y_SIZE) + y].procs )
628	{
629	if ( big ) // big pages => PTE1s
630	{
631	boot_add_pte1( vsid,
632	x,
633	y,
634	vpn + (p<<9),
635	flags,
636	ppn + (p<<9) );
637	}
638	else // small pages => PTE2s
639	{
640	boot_add_pte2( vsid,
641	x,
642	y,
643	vpn + p,
644	flags,
645	ppn + p );
646	}
647	}
648	}
649	}
650	}
651	else if ( (local != 0) && (global != 0) ) // one cluster / all vspaces
652	{
653	for ( v = 0 ; v < header->vspaces ; v++ )
654	{
655	if ( big ) // big pages => PTE1s
656	{
657	boot_add_pte1( v,
658	x_dest,
659	y_dest,
660	vpn + (p<<9),
661	flags,
662	ppn + (p<<9) );
663	}
664	else // small pages = PTE2s
665	{
666	boot_add_pte2( v,
667	x_dest,
668	y_dest,
669	vpn + p,
670	flags,
671	ppn + p );
672	}
673	}
674	}
675	else if ( (local == 0) && (global != 0) ) // all clusters / all vspaces
676	{
677	for ( x = 0 ; x < X_SIZE ; x++ )
678	{
679	for ( y = 0 ; y < Y_SIZE ; y++ )
680	{
681	if ( cluster[(x * Y_SIZE) + y].procs )
682	{
683	for ( v = 0 ; v < header->vspaces ; v++ )
684	{
685	if ( big ) // big pages => PTE1s
686	{
687	boot_add_pte1( v,
688	x,
689	y,
690	vpn + (p<<9),
691	flags,
692	ppn + (p<<9) );
693	}
694	else // small pages -> PTE2s
695	{
696	boot_add_pte2( v,
697	x,
698	y,
699	vpn + p,
700	flags,
701	ppn + p );
702	}
703	}
704	}
705	}
706	}
707	}
708	} // end for pages
709
710	asm volatile ("sync");
711
712	} // end boot_vseg_pte()
713
714
715	///////////////////////////////////////////////////////////////////////////////
716	// This function is executed by processor[x][y][0] in each cluster
717	// containing at least one processor.
718	// It initialises all page table for all global or private vsegs
719	// mapped in cluster[x][y], as specified in the mapping.
720	// In each cluster all page tables for the different vspaces must be
721	// packed in one vseg occupying one single BPP (Big Physical Page).
722	//
723	// For each vseg, the mapping is done in two steps:
724	// 1) mapping : the boot_vseg_map() function allocates contiguous BPPs
725	// or SPPs (if the vseg is not associated to a peripheral), and register
726	// the physical base address in the vseg pbase field. It initialises the
727	// _ptabs_vaddr[] and _ptabs_paddr[] arrays if the vseg is a PTAB.
728	//
729	// 2) page table initialisation : the boot_vseg_pte() function initialise
730	// the PTEs (both PTE1 and PTE2) in one or several page tables:
731	// - PTEs are replicated in all vspaces for a global vseg.
732	// - PTEs are replicated in all clusters for a non local vseg.
733	//
734	// We must handle vsegs in the following order
735	// 1) global vseg containing PTAB mapped in cluster[x][y],
736	// 2) global vsegs occupying more than one BPP mapped in cluster[x][y],
737	// 3) others global vsegs mapped in cluster[x][y],
738	// 4) all private vsegs in all user spaces mapped in cluster[x][y].
739	///////////////////////////////////////////////////////////////////////////////
740	void boot_ptab_init( unsigned int cx,
741	unsigned int cy )
742	{
743	mapping_header_t* header = (mapping_header_t *)SEG_BOOT_MAPPING_BASE;
744	mapping_vspace_t* vspace = _get_vspace_base(header);
745	mapping_vseg_t* vseg = _get_vseg_base(header);
746	mapping_cluster_t* cluster ;
747	mapping_pseg_t* pseg ;
748
749	unsigned int vspace_id;
750	unsigned int vseg_id;
751
752	unsigned int procid = _get_procid();
753	unsigned int lpid = procid & ((1<<P_WIDTH)-1);
754
755	if( lpid )
756	{
757	_printf("\n[BOOT ERROR] in boot_ptab_init() : "
758	"P[%d][%d][%d] should not execute it\n", cx, cy, lpid );
759	_exit();
760	}
761
762	if ( header->vspaces == 0 )
763	{
764	_printf("\n[BOOT ERROR] in boot_ptab_init() : "
765	"mapping %s contains no vspace\n", header->name );
766	_exit();
767	}
768
769	///////// Phase 1 : global vseg containing the PTAB (two barriers required)
770
771	// get PTAB global vseg in cluster(cx,cy)
772	unsigned int found = 0;
773	for (vseg_id = 0; vseg_id < header->globals; vseg_id++)
774	{
775	pseg = _get_pseg_base(header) + vseg[vseg_id].psegid;
776	cluster = _get_cluster_base(header) + pseg->clusterid;
777	if ( (vseg[vseg_id].type == VSEG_TYPE_PTAB) &&
778	(cluster->x == cx) && (cluster->y == cy) )
779	{
780	found = 1;
781	break;
782	}
783	}
784	if ( found == 0 )
785	{
786	_printf("\n[BOOT ERROR] in boot_ptab_init() : "
787	"cluster[%d][%d] contains no PTAB vseg\n", cx , cy );
788	_exit();
789	}
790
791	boot_vseg_map( &vseg[vseg_id], 0xFFFFFFFF );
792
793	//////////////////////////////////////////////
794	_simple_barrier_wait( &_barrier_all_clusters );
795	//////////////////////////////////////////////
796
797	boot_vseg_pte( &vseg[vseg_id], 0xFFFFFFFF );
798
799	//////////////////////////////////////////////
800	_simple_barrier_wait( &_barrier_all_clusters );
801	//////////////////////////////////////////////
802
803	///////// Phase 2 : global vsegs occupying more than one BPP
804
805	for (vseg_id = 0; vseg_id < header->globals; vseg_id++)
806	{
807	pseg = _get_pseg_base(header) + vseg[vseg_id].psegid;
808	cluster = _get_cluster_base(header) + pseg->clusterid;
809	if ( (vseg[vseg_id].length > 0x200000) &&
810	(vseg[vseg_id].mapped == 0) &&
811	(cluster->x == cx) && (cluster->y == cy) )
812	{
813	boot_vseg_map( &vseg[vseg_id], 0xFFFFFFFF );
814	boot_vseg_pte( &vseg[vseg_id], 0xFFFFFFFF );
815	}
816	}
817
818	///////// Phase 3 : all others global vsegs
819
820	for (vseg_id = 0; vseg_id < header->globals; vseg_id++)
821	{
822	pseg = _get_pseg_base(header) + vseg[vseg_id].psegid;
823	cluster = _get_cluster_base(header) + pseg->clusterid;
824	if ( (vseg[vseg_id].mapped == 0) &&
825	(cluster->x == cx) && (cluster->y == cy) )
826	{
827	boot_vseg_map( &vseg[vseg_id], 0xFFFFFFFF );
828	boot_vseg_pte( &vseg[vseg_id], 0xFFFFFFFF );
829	}
830	}
831
832	///////// Phase 4 : all private vsegs
833
834	for (vspace_id = 0; vspace_id < header->vspaces; vspace_id++)
835	{
836	for (vseg_id = vspace[vspace_id].vseg_offset;
837	vseg_id < (vspace[vspace_id].vseg_offset + vspace[vspace_id].vsegs);
838	vseg_id++)
839	{
840	pseg = _get_pseg_base(header) + vseg[vseg_id].psegid;
841	cluster = _get_cluster_base(header) + pseg->clusterid;
842	if ( (cluster->x == cx) && (cluster->y == cy) )
843	{
844	boot_vseg_map( &vseg[vseg_id], vspace_id );
845	boot_vseg_pte( &vseg[vseg_id], vspace_id );
846	}
847	}
848	}
849
850	//////////////////////////////////////////////
851	_simple_barrier_wait( &_barrier_all_clusters );
852	//////////////////////////////////////////////
853
854	} // end boot_ptab_init()
855
856	////////////////////////////////////////////////////////////////////////////////
857	// This function should be executed by P[0][0][0] only. It complete the
858	// page table initialisation, taking care of all global vsegs that are
859	// not mapped in a cluster containing a processor, and have not been
860	// handled by the boot_ptab_init(x,y) function.
861	// An example of such vsegs are the external peripherals in TSAR_LETI platform.
862	////////////////////////////////////////////////////////////////////////////////
863	void boot_ptab_extend()
864	{
865
866	mapping_header_t* header = (mapping_header_t *)SEG_BOOT_MAPPING_BASE;
867	mapping_vseg_t* vseg = _get_vseg_base(header);
868
869	unsigned int vseg_id;
870
871	for (vseg_id = 0; vseg_id < header->globals; vseg_id++)
872	{
873	if ( vseg[vseg_id].mapped == 0 )
874	{
875	boot_vseg_map( &vseg[vseg_id], 0xFFFFFFFF );
876	boot_vseg_pte( &vseg[vseg_id], 0xFFFFFFFF );
877	}
878	}
879	} // end boot_ptab_extend()
880
881	///////////////////////////////////////////////////////////////////////////////
882	// This function returns in the vbase and length buffers the virtual base
883	// address and the length of the segment allocated to the schedulers array
884	// in the cluster defined by the clusterid argument.
885	///////////////////////////////////////////////////////////////////////////////
886	void boot_get_sched_vaddr( unsigned int cluster_id,
887	unsigned int* vbase,
888	unsigned int* length )
889	{
890	mapping_header_t* header = (mapping_header_t *)SEG_BOOT_MAPPING_BASE;
891	mapping_vseg_t* vseg = _get_vseg_base(header);
892	mapping_pseg_t* pseg = _get_pseg_base(header);
893
894	unsigned int vseg_id;
895	unsigned int found = 0;
896
897	for ( vseg_id = 0 ; (vseg_id < header->vsegs) && (found == 0) ; vseg_id++ )
898	{
899	if ( (vseg[vseg_id].type == VSEG_TYPE_SCHED) &&
900	(pseg[vseg[vseg_id].psegid].clusterid == cluster_id ) )
901	{
902	*vbase = vseg[vseg_id].vbase;
903	*length = vseg[vseg_id].length;
904	found = 1;
905	}
906	}
907	if ( found == 0 )
908	{
909	mapping_cluster_t* cluster = _get_cluster_base(header);
910	_printf("\n[BOOT ERROR] No vseg of type SCHED in cluster [%d,%d]\n",
911	cluster[cluster_id].x, cluster[cluster_id].y );
912	_exit();
913	}
914	} // end boot_get_sched_vaddr()
915
916	#if BOOT_DEBUG_SCHED
917	/////////////////////////////////////////////////////////////////////////////
918	// This debug function should be executed by only one procesor.
919	// It loops on all processors in all clusters to display
920	// the HWI / PTI / WTI interrupt vectors for each processor.
921	/////////////////////////////////////////////////////////////////////////////
922	void boot_sched_irq_display()
923	{
924	unsigned int cx;
925	unsigned int cy;
926	unsigned int lpid;
927	unsigned int slot;
928	unsigned int entry;
929	unsigned int type;
930	unsigned int channel;
931
932	mapping_header_t* header = (mapping_header_t *)SEG_BOOT_MAPPING_BASE;
933	mapping_cluster_t* cluster = _get_cluster_base(header);
934
935	static_scheduler_t* psched;
936
937	for ( cx = 0 ; cx < X_SIZE ; cx++ )
938	{
939	for ( cy = 0 ; cy < Y_SIZE ; cy++ )
940	{
941	unsigned int cluster_id = (cx * Y_SIZE) + cy;
942	unsigned int nprocs = cluster[cluster_id].procs;
943
944	for ( lpid = 0 ; lpid < nprocs ; lpid++ )
945	{
946	psched = _schedulers[cx][cy][lpid];
947
948	_printf("\n[BOOT] interrupt vectors for proc[%d,%d,%d]\n",
949	cx , cy , lpid );
950
951	for ( slot = 0 ; slot < 32 ; slot++ )
952	{
953	entry = psched->hwi_vector[slot];
954	type = entry & 0xFFFF;
955	channel = entry >> 16;
956	if ( type != ISR_DEFAULT )
957	_printf(" - HWI : index = %d / type = %s / channel = %d\n",
958	slot , _isr_type_str[type] , channel );
959	}
960	for ( slot = 0 ; slot < 32 ; slot++ )
961	{
962	entry = psched->wti_vector[slot];
963	type = entry & 0xFFFF;
964	channel = entry >> 16;
965	if ( type != ISR_DEFAULT )
966	_printf(" - WTI : index = %d / type = %s / channel = %d\n",
967	slot , _isr_type_str[type] , channel );
968	}
969	for ( slot = 0 ; slot < 32 ; slot++ )
970	{
971	entry = psched->pti_vector[slot];
972	type = entry & 0xFFFF;
973	channel = entry >> 16;
974	if ( type != ISR_DEFAULT )
975	_printf(" - PTI : index = %d / type = %s / channel = %d\n",
976	slot , _isr_type_str[type] , channel );
977	}
978	}
979	}
980	}
981	} // end boot_sched_irq_display()
982	#endif
983
984
985	////////////////////////////////////////////////////////////////////////////////////
986	// This function is executed in parallel by all processors P[x][y][0].
987	// It initialises all schedulers in cluster [x][y]. The MMU must be activated.
988	// It is split in two phases separated by a synchronisation barrier.
989	// - In Step 1, it initialises the _schedulers[x][y][l] pointers array, the
990	// idle_task context, the HWI / PTI / WTI interrupt vectors,
991	// and the CU HWI / PTI / WTI masks.
992	// - In Step 2, it scan all tasks in all vspaces to complete the tasks contexts,
993	// initialisation as specified in the mapping_info data structure,
994	// and set the CP0_SCHED register.
995	////////////////////////////////////////////////////////////////////////////////////
996	void boot_scheduler_init( unsigned int x,
997	unsigned int y )
998	{
999	mapping_header_t* header = (mapping_header_t *)SEG_BOOT_MAPPING_BASE;
1000	mapping_cluster_t* cluster = _get_cluster_base(header);
1001	mapping_vspace_t* vspace = _get_vspace_base(header);
1002	mapping_vseg_t* vseg = _get_vseg_base(header);
1003	mapping_task_t* task = _get_task_base(header);
1004	mapping_periph_t* periph = _get_periph_base(header);
1005	mapping_irq_t* irq = _get_irq_base(header);
1006
1007	unsigned int periph_id;
1008	unsigned int irq_id;
1009	unsigned int vspace_id;
1010	unsigned int vseg_id;
1011	unsigned int task_id;
1012
1013	unsigned int sched_vbase; // schedulers array vbase address
1014	unsigned int sched_length; // schedulers array length
1015	static_scheduler_t* psched; // pointer on processor scheduler
1016
1017	unsigned int cluster_id = (x * Y_SIZE) + y;
1018	unsigned int cluster_xy = (x << Y_WIDTH) + y;
1019	unsigned int nprocs = cluster[cluster_id].procs;
1020	unsigned int lpid;
1021
1022	if ( nprocs > 8 )
1023	{
1024	_printf("\n[BOOT ERROR] cluster[%d,%d] contains more than 8 procs\n", x, y );
1025	_exit();
1026	}
1027
1028	////////////////////////////////////////////////////////////////////////////////
1029	// Step 1 : - initialize the schedulers[] array of pointers,
1030	// - initialize the "tasks" and "current variables.
1031	// - initialise the idle task context.
1032	// - initialize the HWI, PTI and WTI interrupt vectors.
1033	// - initialize the XCU masks for HWI / WTI / PTI interrupts.
1034	//
1035	// The general policy for interrupts routing is the following:
1036	// - the local HWI are statically allocatedted to local processors.
1037	// - the nprocs first PTI are allocated for TICK (one per processor).
1038	// - we allocate 4 WTI per processor: the first one is for WAKUP,
1039	// the 3 others WTI are used for external interrupts (from PIC),
1040	// and are dynamically allocated by kernel on demand.
1041	///////////////////////////////////////////////////////////////////////////////
1042
1043	// get scheduler array virtual base address in cluster[x,y]
1044	boot_get_sched_vaddr( cluster_id, &sched_vbase, &sched_length );
1045
1046	if ( sched_length < (nprocs<<13) ) // 8 Kbytes per scheduler
1047	{
1048	_printf("\n[BOOT ERROR] Sched segment too small in cluster[%d,%d]\n",
1049	x, y );
1050	_exit();
1051	}
1052
1053	// loop on local processors
1054	for ( lpid = 0 ; lpid < nprocs ; lpid++ )
1055	{
1056	// get scheduler pointer and initialise the schedulers pointers array
1057	psched = (static_scheduler_t*)(sched_vbase + (lpid<<13));
1058	_schedulers[x][y][lpid] = psched;
1059
1060	// initialise the "tasks" and "current" variables default values
1061	psched->tasks = 0;
1062	psched->current = IDLE_TASK_INDEX;
1063
1064	// set default values for HWI / PTI / SWI vectors (valid bit = 0)
1065	unsigned int slot;
1066	for (slot = 0; slot < 32; slot++)
1067	{
1068	psched->hwi_vector[slot] = 0;
1069	psched->pti_vector[slot] = 0;
1070	psched->wti_vector[slot] = 0;
1071	}
1072
1073	// initializes the idle_task context:
1074	// - the SR slot is 0xFF03 because this task run in kernel mode.
1075	// - it uses the page table of vspace[0]
1076	// - it uses the kernel TTY terminal
1077	// - slots containing addresses (SP,RA,EPC) are initialised by kernel_init()
1078
1079	psched->context[IDLE_TASK_INDEX][CTX_CR_ID] = 0;
1080	psched->context[IDLE_TASK_INDEX][CTX_SR_ID] = 0xFF03;
1081	psched->context[IDLE_TASK_INDEX][CTX_PTPR_ID] = _ptabs_paddr[0][x][y]>>13;
1082	psched->context[IDLE_TASK_INDEX][CTX_PTAB_ID] = _ptabs_vaddr[0][x][y];
1083	psched->context[IDLE_TASK_INDEX][CTX_TTY_ID] = 0;
1084	psched->context[IDLE_TASK_INDEX][CTX_LTID_ID] = IDLE_TASK_INDEX;
1085	psched->context[IDLE_TASK_INDEX][CTX_VSID_ID] = 0;
1086	psched->context[IDLE_TASK_INDEX][CTX_RUN_ID] = 1;
1087	}
1088
1089	// HWI / PTI / WTI masks (up to 8 local processors)
1090	unsigned int hwi_mask[8] = {0,0,0,0,0,0,0,0};
1091	unsigned int pti_mask[8] = {0,0,0,0,0,0,0,0};
1092	unsigned int wti_mask[8] = {0,0,0,0,0,0,0,0};
1093
1094	// scan local peripherals to get and check local XCU
1095	mapping_periph_t* xcu = NULL;
1096	unsigned int min = cluster[cluster_id].periph_offset ;
1097	unsigned int max = min + cluster[cluster_id].periphs ;
1098
1099	for ( periph_id = min ; periph_id < max ; periph_id++ )
1100	{
1101	if( periph[periph_id].type == PERIPH_TYPE_XCU )
1102	{
1103	xcu = &periph[periph_id];
1104
1105	// check nb_hwi_in
1106	if ( xcu->arg0 < xcu->irqs )
1107	{
1108	_printf("\n[BOOT ERROR] Not enough HWI inputs for XCU[%d,%d]"
1109	" : nb_hwi = %d / nb_irqs = %d\n",
1110	x , y , xcu->arg0 , xcu->irqs );
1111	_exit();
1112	}
1113	// check nb_pti_in
1114	if ( xcu->arg2 < nprocs )
1115	{
1116	_printf("\n[BOOT ERROR] Not enough PTI inputs for XCU[%d,%d]\n",
1117	x, y );
1118	_exit();
1119	}
1120	// check nb_wti_in
1121	if ( xcu->arg1 < (4 * nprocs) )
1122	{
1123	_printf("\n[BOOT ERROR] Not enough WTI inputs for XCU[%d,%d]\n",
1124	x, y );
1125	_exit();
1126	}
1127	// check nb_irq_out
1128	if ( xcu->channels < (nprocs * header->irq_per_proc) )
1129	{
1130	_printf("\n[BOOT ERROR] Not enough outputs for XCU[%d,%d]\n",
1131	x, y );
1132	_exit();
1133	}
1134	}
1135	}
1136
1137	if ( xcu == NULL )
1138	{
1139	_printf("\n[BOOT ERROR] missing XCU in cluster[%d,%d]\n", x , y );
1140	_exit();
1141	}
1142
1143	// HWI interrupt vector definition
1144	// scan HWI connected to local XCU
1145	// for round-robin allocation to local processors
1146	lpid = 0;
1147	for ( irq_id = xcu->irq_offset ;
1148	irq_id < xcu->irq_offset + xcu->irqs ;
1149	irq_id++ )
1150	{
1151	unsigned int type = irq[irq_id].srctype;
1152	unsigned int srcid = irq[irq_id].srcid;
1153	unsigned int isr = irq[irq_id].isr & 0xFFFF;
1154	unsigned int channel = irq[irq_id].channel << 16;
1155
1156	if ( (type != IRQ_TYPE_HWI) \|\| (srcid > 31) )
1157	{
1158	_printf("\n[BOOT ERROR] Bad IRQ in cluster[%d,%d]\n", x, y );
1159	_exit();
1160	}
1161
1162	// register entry in HWI interrupt vector
1163	_schedulers[x][y][lpid]->hwi_vector[srcid] = isr \| channel;
1164
1165	// update XCU HWI mask for P[x,y,lpid]
1166	hwi_mask[lpid] = hwi_mask[lpid] \| (1<<srcid);
1167
1168	lpid = (lpid + 1) % nprocs;
1169	} // end for irqs
1170
1171	// PTI interrupt vector definition
1172	// one PTI for TICK per processor
1173	for ( lpid = 0 ; lpid < nprocs ; lpid++ )
1174	{
1175	// register entry in PTI interrupt vector
1176	_schedulers[x][y][lpid]->pti_vector[lpid] = ISR_TICK;
1177
1178	// update XCU PTI mask for P[x,y,lpid]
1179	pti_mask[lpid] = pti_mask[lpid] \| (1<<lpid);
1180	}
1181
1182	// WTI interrupt vector definition
1183	// 4 WTI per processor, first for WAKUP
1184	for ( lpid = 0 ; lpid < nprocs ; lpid++ )
1185	{
1186	// register WAKUP ISR in WTI interrupt vector
1187	_schedulers[x][y][lpid]->wti_vector[lpid] = ISR_WAKUP;
1188
1189	// update XCU WTI mask for P[x,y,lpid] (4 entries per proc)
1190	wti_mask[lpid] = wti_mask[lpid] \| (0x1<<(lpid ));
1191	wti_mask[lpid] = wti_mask[lpid] \| (0x1<<(lpid + NB_PROCS_MAX ));
1192	wti_mask[lpid] = wti_mask[lpid] \| (0x1<<(lpid + 2*NB_PROCS_MAX));
1193	wti_mask[lpid] = wti_mask[lpid] \| (0x1<<(lpid + 3*NB_PROCS_MAX));
1194	}
1195
1196	// set the XCU masks for HWI / WTI / PTI interrupts
1197	for ( lpid = 0 ; lpid < nprocs ; lpid++ )
1198	{
1199	unsigned int channel = lpid * IRQ_PER_PROCESSOR;
1200
1201	_xcu_set_mask( cluster_xy, channel, hwi_mask[lpid], IRQ_TYPE_HWI );
1202	_xcu_set_mask( cluster_xy, channel, wti_mask[lpid], IRQ_TYPE_WTI );
1203	_xcu_set_mask( cluster_xy, channel, pti_mask[lpid], IRQ_TYPE_PTI );
1204	}
1205
1206	//////////////////////////////////////////////
1207	_simple_barrier_wait( &_barrier_all_clusters );
1208	//////////////////////////////////////////////
1209
1210	#if BOOT_DEBUG_SCHED
1211	if ( cluster_xy == 0 ) boot_sched_irq_display();
1212	_simple_barrier_wait( &_barrier_all_clusters );
1213	#endif
1214
1215	///////////////////////////////////////////////////////////////////////////////
1216	// Step 2 : Initialise the tasks context. The context of task placed
1217	// on processor P must be stored in the scheduler of P.
1218	// This require two nested loops: loop on the tasks, and loop
1219	// on the local processors. We complete the scheduler when the
1220	// required placement fit one local processor.
1221	///////////////////////////////////////////////////////////////////////////////
1222
1223	for (vspace_id = 0; vspace_id < header->vspaces; vspace_id++)
1224	{
1225	// We must set the PTPR depending on the vspace, because the start_vector
1226	// and the stack address are defined in virtual space.
1227	_set_mmu_ptpr( (unsigned int)(_ptabs_paddr[vspace_id][x][y] >> 13) );
1228
1229	// loop on the tasks in vspace (task_id is the global index in mapping)
1230	for (task_id = vspace[vspace_id].task_offset;
1231	task_id < (vspace[vspace_id].task_offset + vspace[vspace_id].tasks);
1232	task_id++)
1233	{
1234	// get the required task placement coordinates [x,y,p]
1235	unsigned int req_x = cluster[task[task_id].clusterid].x;
1236	unsigned int req_y = cluster[task[task_id].clusterid].y;
1237	unsigned int req_p = task[task_id].proclocid;
1238
1239	// ctx_sr : value required before an eret instruction
1240	unsigned int ctx_sr = 0x2000FF13;
1241
1242	// ctx_ptpr : page table physical base address (shifted by 13 bit)
1243	unsigned int ctx_ptpr = (_ptabs_paddr[vspace_id][req_x][req_y] >> 13);
1244
1245	// ctx_ptab : page_table virtual base address
1246	unsigned int ctx_ptab = _ptabs_vaddr[vspace_id][req_x][req_y];
1247
1248	// ctx_epc : Get the virtual address of the memory location containing
1249	// the task entry point : the start_vector is stored by GCC in the
1250	// seg_data segment, and we must wait the .elf loading to get
1251	// the entry point value...
1252	vseg_id = vspace[vspace_id].start_vseg_id;
1253	unsigned int ctx_epc = vseg[vseg_id].vbase + (task[task_id].startid)*4;
1254
1255	// ctx_sp : Get the vseg containing the stack
1256	vseg_id = task[task_id].stack_vseg_id;
1257	unsigned int ctx_sp = vseg[vseg_id].vbase + vseg[vseg_id].length;
1258
1259	// get vspace thread index
1260	unsigned int thread_id = task[task_id].trdid;
1261
1262	// loop on the local processors
1263	for ( lpid = 0 ; lpid < nprocs ; lpid++ )
1264	{
1265	if ( (x == req_x) && (y == req_y) && (req_p == lpid) ) // fit
1266	{
1267	// pointer on selected scheduler
1268	psched = _schedulers[x][y][lpid];
1269
1270	// get local task index in scheduler
1271	unsigned int ltid = psched->tasks;
1272
1273	// update the "tasks" and "current" fields in scheduler:
1274	psched->tasks = ltid + 1;
1275	psched->current = 0;
1276
1277	// initializes the task context
1278	psched->context[ltid][CTX_CR_ID] = 0;
1279	psched->context[ltid][CTX_SR_ID] = ctx_sr;
1280	psched->context[ltid][CTX_SP_ID] = ctx_sp;
1281	psched->context[ltid][CTX_EPC_ID] = ctx_epc;
1282	psched->context[ltid][CTX_PTPR_ID] = ctx_ptpr;
1283	psched->context[ltid][CTX_PTAB_ID] = ctx_ptab;
1284	psched->context[ltid][CTX_LTID_ID] = ltid;
1285	psched->context[ltid][CTX_GTID_ID] = task_id;
1286	psched->context[ltid][CTX_TRDID_ID] = thread_id;
1287	psched->context[ltid][CTX_VSID_ID] = vspace_id;
1288	psched->context[ltid][CTX_RUN_ID] = 1;
1289
1290	psched->context[ltid][CTX_TTY_ID] = 0xFFFFFFFF;
1291	psched->context[ltid][CTX_CMA_FB_ID] = 0xFFFFFFFF;
1292	psched->context[ltid][CTX_CMA_RX_ID] = 0xFFFFFFFF;
1293	psched->context[ltid][CTX_CMA_TX_ID] = 0xFFFFFFFF;
1294	psched->context[ltid][CTX_NIC_RX_ID] = 0xFFFFFFFF;
1295	psched->context[ltid][CTX_NIC_TX_ID] = 0xFFFFFFFF;
1296	psched->context[ltid][CTX_TIM_ID] = 0xFFFFFFFF;
1297	psched->context[ltid][CTX_HBA_ID] = 0xFFFFFFFF;
1298
1299	#if BOOT_DEBUG_SCHED
1300	_printf("\nTask %s in vspace %s allocated to P[%d,%d,%d]\n"
1301	" - ctx[LTID] = %d\n"
1302	" - ctx[SR] = %x\n"
1303	" - ctx[SP] = %x\n"
1304	" - ctx[EPC] = %x\n"
1305	" - ctx[PTPR] = %x\n"
1306	" - ctx[PTAB] = %x\n"
1307	" - ctx[VSID] = %d\n"
1308	" - ctx[TRDID] = %d\n",
1309	task[task_id].name,
1310	vspace[vspace_id].name,
1311	x, y, lpid,
1312	psched->context[ltid][CTX_LTID_ID],
1313	psched->context[ltid][CTX_SR_ID],
1314	psched->context[ltid][CTX_SP_ID],
1315	psched->context[ltid][CTX_EPC_ID],
1316	psched->context[ltid][CTX_PTPR_ID],
1317	psched->context[ltid][CTX_PTAB_ID],
1318	psched->context[ltid][CTX_VSID_ID],
1319	psched->context[ltid][CTX_TRDID_ID] );
1320	#endif
1321	} // end if FIT
1322	} // end for loop on local procs
1323	} // end loop on tasks
1324	} // end loop on vspaces
1325	} // end boot_scheduler_init()
1326
1327
1328
1329	//////////////////////////////////////////////////////////////////////////////////
1330	// This function loads the map.bin file from block device.
1331	//////////////////////////////////////////////////////////////////////////////////
1332	void boot_mapping_init()
1333	{
1334	// open file "map.bin"
1335	int fd_id = _fat_open( 0, "map.bin", 0 ); // no IRQ / no creation
1336
1337	if ( fd_id == -1 )
1338	{
1339	_printf("\n[BOOT ERROR] : map.bin file not found \n");
1340	_exit();
1341	}
1342
1343	#if BOOT_DEBUG_MAPPING
1344	_printf("\n[BOOT] map.bin file successfully open at cycle %d\n",
1345	_get_proctime() );
1346	#endif
1347
1348	// get "map.bin" file size (from fat) and check it
1349	unsigned int size = _fat.fd[fd_id].file_size;
1350
1351	if ( size > SEG_BOOT_MAPPING_SIZE )
1352	{
1353	_printf("\n[BOOT ERROR] : segment too small for map.bin file\n");
1354	_exit();
1355	}
1356
1357	#if BOOT_DEBUG_MAPPING
1358	_printf("\n[BOOT] map.bin buf_pbase = %x / buf_size = %x / file_size = %x\n",
1359	SEG_BOOT_MAPPING_BASE , SEG_BOOT_MAPPING_SIZE , size );
1360	#endif
1361
1362	// load "map.bin" file into buffer
1363	unsigned int nblocks = size >> 9;
1364	unsigned int offset = size & 0x1FF;
1365	if ( offset ) nblocks++;
1366
1367	unsigned int ok = _fat_read( 0, // No IRQ
1368	fd_id,
1369	(unsigned int*)SEG_BOOT_MAPPING_BASE,
1370	nblocks,
1371	0 ); // offset
1372	if ( ok == -1 )
1373	{
1374	_printf("\n[BOOT ERROR] : unable to load map.bin file \n");
1375	_exit();
1376	}
1377
1378	#if BOOT_DEBUG_MAPPING
1379	_printf("\n[BOOT] map.bin file successfully loaded at cycle %d\n", _get_proctime() );
1380	#endif
1381
1382	// check mapping signature, number of clusters, number of vspaces
1383	mapping_header_t * header = (mapping_header_t *)SEG_BOOT_MAPPING_BASE;
1384	if ( (header->signature != IN_MAPPING_SIGNATURE) \|\|
1385	(header->x_size != X_SIZE) \|\|
1386	(header->y_size != Y_SIZE) \|\|
1387	(header->vspaces > GIET_NB_VSPACE_MAX) )
1388	{
1389	_printf("\n[BOOT ERROR] Illegal mapping signature: %x\n", header->signature );
1390	_exit();
1391	}
1392
1393	#if BOOT_DEBUG_MAPPING
1394	unsigned int line;
1395	unsigned int* pointer = (unsigned int*)SEG_BOOT_MAPPING_BASE;
1396	_printf("\n[BOOT] First block of mapping\n");
1397	for ( line = 0 ; line < 8 ; line++ )
1398	{
1399	_printf(" \| %x \| %x \| %x \| %x \| %x \| %x \| %x \| %x \|\n",
1400	*(pointer + 0),
1401	*(pointer + 1),
1402	*(pointer + 2),
1403	*(pointer + 3),
1404	*(pointer + 4),
1405	*(pointer + 5),
1406	*(pointer + 6),
1407	*(pointer + 7) );
1408
1409	pointer = pointer + 8;
1410	}
1411	#endif
1412
1413	#if BOOT_DEBUG_MAPPING
1414	_printf("\n[BOOT] map.bin file checked at cycle %d\n", _get_proctime() );
1415	#endif
1416
1417	// close file "map.bin"
1418	_fat_close( fd_id );
1419
1420	} // end boot_mapping_init()
1421
1422
1423	///////////////////////////////////////////////////
1424	void boot_dma_copy( unsigned int cluster_xy,
1425	unsigned long long dst_paddr,
1426	unsigned long long src_paddr,
1427	unsigned int size )
1428	{
1429	// size must be multiple of 64 bytes
1430	if ( size & 0x3F ) size = (size & (~0x3F)) + 0x40;
1431
1432	unsigned int mode = MODE_DMA_NO_IRQ;
1433
1434	unsigned int src = 0;
1435	unsigned int src_lsb = (unsigned int)src_paddr;
1436	unsigned int src_msb = (unsigned int)(src_paddr>>32);
1437
1438	unsigned int dst = 1;
1439	unsigned int dst_lsb = (unsigned int)dst_paddr;
1440	unsigned int dst_msb = (unsigned int)(dst_paddr>>32);
1441
1442	// initializes src channel
1443	_mwr_set_channel_register( cluster_xy , src , MWR_CHANNEL_MODE , mode );
1444	_mwr_set_channel_register( cluster_xy , src , MWR_CHANNEL_SIZE , size );
1445	_mwr_set_channel_register( cluster_xy , src , MWR_CHANNEL_BUFFER_LSB , src_lsb );
1446	_mwr_set_channel_register( cluster_xy , src , MWR_CHANNEL_BUFFER_MSB , src_msb );
1447	_mwr_set_channel_register( cluster_xy , src , MWR_CHANNEL_RUNNING , 1 );
1448
1449	// initializes dst channel
1450	_mwr_set_channel_register( cluster_xy , dst , MWR_CHANNEL_MODE , mode );
1451	_mwr_set_channel_register( cluster_xy , dst , MWR_CHANNEL_SIZE , size );
1452	_mwr_set_channel_register( cluster_xy , dst , MWR_CHANNEL_BUFFER_LSB , dst_lsb );
1453	_mwr_set_channel_register( cluster_xy , dst , MWR_CHANNEL_BUFFER_MSB , dst_msb );
1454	_mwr_set_channel_register( cluster_xy , dst , MWR_CHANNEL_RUNNING , 1 );
1455
1456	// start CPY coprocessor (write non-zero value into config register)
1457	_mwr_set_coproc_register( cluster_xy, 0 , 1 );
1458
1459	// poll dst channel status register to detect completion
1460	unsigned int status;
1461	do
1462	{
1463	status = _mwr_get_channel_register( cluster_xy , dst , MWR_CHANNEL_STATUS );
1464	} while ( status == MWR_CHANNEL_BUSY );
1465
1466	if ( status )
1467	{
1468	_printf("\n[BOOT ERROR] in boot_dma_copy()\n");
1469	_exit();
1470	}
1471
1472	// stop CPY coprocessor and DMA channels
1473	_mwr_set_channel_register( cluster_xy , src , MWR_CHANNEL_RUNNING , 0 );
1474	_mwr_set_channel_register( cluster_xy , dst , MWR_CHANNEL_RUNNING , 0 );
1475	_mwr_set_coproc_register ( cluster_xy , 0 , 0 );
1476
1477	} // end boot_dma_copy()
1478
1479	//////////////////////////////////////////////////////////////////////////////////
1480	// This function load all loadable segments contained in the .elf file identified
1481	// by the "pathname" argument. Some loadable segments can be copied in several
1482	// clusters: same virtual address but different physical addresses.
1483	// - It open the file.
1484	// - It loads the complete file in the dedicated _boot_elf_buffer.
1485	// - It copies each loadable segments at the virtual address defined in
1486	// the .elf file, making several copies if the target vseg is not local.
1487	// - It closes the file.
1488	// This function is supposed to be executed by all processors[x,y,0].
1489	//
1490	// Note: We must use physical addresses to reach the destination buffers that
1491	// can be located in remote clusters. We use either a _physical_memcpy(),
1492	// or a _dma_physical_copy() if DMA is available.
1493	//////////////////////////////////////////////////////////////////////////////////
1494	void load_one_elf_file( unsigned int is_kernel, // kernel file if non zero
1495	char* pathname,
1496	unsigned int vspace_id ) // to scan the proper vspace
1497	{
1498	mapping_header_t * header = (mapping_header_t *)SEG_BOOT_MAPPING_BASE;
1499	mapping_vspace_t * vspace = _get_vspace_base(header);
1500	mapping_vseg_t * vseg = _get_vseg_base(header);
1501
1502	unsigned int procid = _get_procid();
1503	unsigned int cxy = procid >> P_WIDTH;
1504	unsigned int x = cxy >> Y_WIDTH;
1505	unsigned int y = cxy & ((1<<Y_WIDTH)-1);
1506	unsigned int p = procid & ((1<<P_WIDTH)-1);
1507
1508	#if BOOT_DEBUG_ELF
1509	_printf("\n[DEBUG BOOT_ELF] load_one_elf_file() : P[%d,%d,%d] enters for %s\n",
1510	x , y , p , pathname );
1511	#endif
1512
1513	Elf32_Ehdr* elf_header_ptr = NULL; // avoid a warning
1514
1515	int fd_id = 0; // avoid a warning
1516
1517	// only P[0,0,0] load file from FAT
1518	if ( (cxy == 0) && (p == 0) )
1519	{
1520	// open .elf file
1521	fd_id = _fat_open( 0 , pathname , 0 ); // no IRQ / no creation
1522
1523	if ( fd_id < 0 )
1524	{
1525	_printf("\n[BOOT ERROR] load_one_elf_file() : %s not found\n",
1526	pathname );
1527	_exit();
1528	}
1529
1530	// check buffer size versus file size
1531	if ( _fat.fd[fd_id].file_size > GIET_ELF_BUFFER_SIZE )
1532	{
1533	_printf("\n[BOOT ERROR] in load_one_elf_file() : %s / size = %x "
1534	"larger than GIET_ELF_BUFFER_SIZE = %x\n",
1535	pathname , _fat.fd[fd_id].file_size , GIET_ELF_BUFFER_SIZE );
1536	_exit();
1537	}
1538
1539	// compute number of sectors
1540	unsigned int nbytes = _fat.fd[fd_id].file_size;
1541	unsigned int nsectors = nbytes>>9;
1542	if( nbytes & 0x1FF) nsectors++;
1543
1544	// load file to elf buffer
1545	if( _fat_read( 0, // no IRQ
1546	fd_id,
1547	_boot_elf_buffer,
1548	nsectors,
1549	0 ) != nsectors ) // offset
1550	{
1551	_printf("\n[BOOT ERROR] load_one_elf_file() : unexpected EOF for %s\n",
1552	pathname );
1553	_exit();
1554	}
1555
1556	// Check ELF Magic Number in ELF header
1557	Elf32_Ehdr* ptr = (Elf32_Ehdr*)_boot_elf_buffer;
1558
1559	if ( (ptr->e_ident[EI_MAG0] != ELFMAG0) \|\|
1560	(ptr->e_ident[EI_MAG1] != ELFMAG1) \|\|
1561	(ptr->e_ident[EI_MAG2] != ELFMAG2) \|\|
1562	(ptr->e_ident[EI_MAG3] != ELFMAG3) )
1563	{
1564	_printf("\n[BOOT ERROR] load_one_elf_file() : %s not ELF format\n",
1565	pathname );
1566	_exit();
1567	}
1568
1569	#if BOOT_DEBUG_ELF
1570	_printf("\n[DEBUG BOOT_ELF] load_one_elf_file() : P[%d,%d,%d] load %s at cycle %d\n",
1571	x , y , p , pathname , _get_proctime() );
1572	#endif
1573
1574	} // end if P[0,0,0]
1575
1576	//////////////////////////////////////////////
1577	_simple_barrier_wait( &_barrier_all_clusters );
1578	//////////////////////////////////////////////
1579
1580	// Each processor P[x,y,0] copy replicated segments in cluster[x,y]
1581	elf_header_ptr = (Elf32_Ehdr*)_boot_elf_buffer;
1582
1583	// get program header table pointer
1584	unsigned int offset = elf_header_ptr->e_phoff;
1585	if( offset == 0 )
1586	{
1587	_printf("\n[BOOT ERROR] load_one_elf_file() : file %s "
1588	"does not contain loadable segment\n", pathname );
1589	_exit();
1590	}
1591
1592	Elf32_Phdr* elf_pht_ptr = (Elf32_Phdr*)(_boot_elf_buffer + offset);
1593
1594	// get number of segments
1595	unsigned int nsegments = elf_header_ptr->e_phnum;
1596
1597	// First loop on loadable segments in the .elf file
1598	unsigned int seg_id;
1599	for (seg_id = 0 ; seg_id < nsegments ; seg_id++)
1600	{
1601	if(elf_pht_ptr[seg_id].p_type == PT_LOAD)
1602	{
1603	// Get segment attributes
1604	unsigned int seg_vaddr = elf_pht_ptr[seg_id].p_vaddr;
1605	unsigned int seg_offset = elf_pht_ptr[seg_id].p_offset;
1606	unsigned int seg_filesz = elf_pht_ptr[seg_id].p_filesz;
1607	unsigned int seg_memsz = elf_pht_ptr[seg_id].p_memsz;
1608
1609	if( seg_memsz != seg_filesz )
1610	{
1611	_printf("\n[BOOT ERROR] load_one_elf_file() : segment at vaddr = %x\n"
1612	" in file %s has memsize = %x / filesize = %x \n"
1613	" check that all global variables are in data segment\n",
1614	seg_vaddr, pathname , seg_memsz , seg_filesz );
1615	_exit();
1616	}
1617
1618	unsigned int src_vaddr = (unsigned int)_boot_elf_buffer + seg_offset;
1619
1620	// search all vsegs matching the virtual address
1621	unsigned int vseg_first;
1622	unsigned int vseg_last;
1623	unsigned int vseg_id;
1624	unsigned int found = 0;
1625	if ( is_kernel )
1626	{
1627	vseg_first = 0;
1628	vseg_last = header->globals;
1629	}
1630	else
1631	{
1632	vseg_first = vspace[vspace_id].vseg_offset;
1633	vseg_last = vseg_first + vspace[vspace_id].vsegs;
1634	}
1635
1636	// Second loop on vsegs in the mapping
1637	for ( vseg_id = vseg_first ; vseg_id < vseg_last ; vseg_id++ )
1638	{
1639	if ( seg_vaddr == vseg[vseg_id].vbase ) // matching
1640	{
1641	found = 1;
1642
1643	// get destination buffer physical address, size, coordinates
1644	paddr_t seg_paddr = vseg[vseg_id].pbase;
1645	unsigned int seg_size = vseg[vseg_id].length;
1646	unsigned int cluster_xy = (unsigned int)(seg_paddr>>32);
1647	unsigned int cx = cluster_xy >> Y_WIDTH;
1648	unsigned int cy = cluster_xy & ((1<<Y_WIDTH)-1);
1649
1650	// check vseg size
1651	if ( seg_size < seg_filesz )
1652	{
1653	_printf("\n[BOOT ERROR] in load_one_elf_file() : vseg %s "
1654	"is to small for loadable segment %x in file %s\n",
1655	vseg[vseg_id].name , seg_vaddr , pathname );
1656	_exit();
1657	}
1658
1659	// P[x,y,0] copy the segment from boot buffer in cluster[0,0]
1660	// to destination buffer in cluster[x,y], using DMA if available
1661	if ( (cx == x) && (cy == y) )
1662	{
1663	if( USE_MWR_CPY )
1664	{
1665	boot_dma_copy( cluster_xy, // DMA in cluster[x,y]
1666	seg_paddr,
1667	(paddr_t)src_vaddr,
1668	seg_filesz );
1669	#if BOOT_DEBUG_ELF
1670	_printf("\n[DEBUG BOOT_ELF] load_one_elf_file() : DMA[%d,%d] copy segment %d :\n"
1671	" vaddr = %x / size = %x / paddr = %l\n",
1672	x , y , seg_id , seg_vaddr , seg_memsz , seg_paddr );
1673	#endif
1674	}
1675	else
1676	{
1677	_physical_memcpy( seg_paddr, // dest paddr
1678	(paddr_t)src_vaddr, // source paddr
1679	seg_filesz ); // size
1680	#if BOOT_DEBUG_ELF
1681	_printf("\n[DEBUG BOOT_ELF] load_one_elf_file() : P[%d,%d,%d] copy segment %d :\n"
1682	" vaddr = %x / size = %x / paddr = %l\n",
1683	x , y , p , seg_id , seg_vaddr , seg_memsz , seg_paddr );
1684	#endif
1685	}
1686	}
1687	}
1688	} // end for vsegs
1689
1690	// check at least one matching vseg
1691	if ( found == 0 )
1692	{
1693	_printf("\n[BOOT ERROR] in load_one_elf_file() : vseg for loadable "
1694	"segment %x in file %s not found "
1695	"check consistency between the .py and .ld files\n",
1696	seg_vaddr, pathname );
1697	_exit();
1698	}
1699	}
1700	} // end for loadable segments
1701
1702	//////////////////////////////////////////////
1703	_simple_barrier_wait( &_barrier_all_clusters );
1704	//////////////////////////////////////////////
1705
1706	// only P[0,0,0] close the file
1707	if ( (cxy == 0) && (p == 0) )
1708	{
1709	// close .elf file
1710	_fat_close( fd_id );
1711
1712	_printf("\n[BOOT] File %s loaded at cycle %d\n",
1713	pathname , _get_proctime() );
1714	}
1715
1716	} // end load_one_elf_file()
1717
1718
1719	/////i////////////////////////////////////////////////////////////////////////////////
1720	// This function uses the map.bin data structure to load the "kernel.elf" file
1721	// as well as the various "application.elf" files into memory.
1722	// - The "preloader.elf" file is not loaded, because it has been burned in the ROM.
1723	// - The "boot.elf" file is not loaded, because it has been loaded by the preloader.
1724	// This function scans all vsegs defined in the map.bin data structure to collect
1725	// all .elf files pathnames, and calls the load_one_elf_file() for each .elf file.
1726	// As the code can be replicated in several vsegs, the same code can be copied
1727	// in one or several clusters by the load_one_elf_file() function.
1728	//////////////////////////////////////////////////////////////////////////////////////
1729	void boot_elf_load()
1730	{
1731	mapping_header_t* header = (mapping_header_t *)SEG_BOOT_MAPPING_BASE;
1732	mapping_vspace_t* vspace = _get_vspace_base( header );
1733	mapping_vseg_t* vseg = _get_vseg_base( header );
1734
1735	unsigned int vspace_id;
1736	unsigned int vseg_id;
1737	unsigned int found;
1738
1739	// Scan all global vsegs to find the pathname to the kernel.elf file
1740	found = 0;
1741	for( vseg_id = 0 ; vseg_id < header->globals ; vseg_id++ )
1742	{
1743	if(vseg[vseg_id].type == VSEG_TYPE_ELF)
1744	{
1745	found = 1;
1746	break;
1747	}
1748	}
1749
1750	// We need one kernel.elf file
1751	if (found == 0)
1752	{
1753	_printf("\n[BOOT ERROR] boot_elf_load() : kernel.elf file not found\n");
1754	_exit();
1755	}
1756
1757	// Load the kernel
1758	load_one_elf_file( 1, // kernel file
1759	vseg[vseg_id].binpath, // file pathname
1760	0 ); // vspace 0
1761
1762	// loop on the vspaces, scanning all vsegs in the vspace,
1763	// to find the pathname of the .elf file associated to the vspace.
1764	for( vspace_id = 0 ; vspace_id < header->vspaces ; vspace_id++ )
1765	{
1766	// loop on the private vsegs
1767	unsigned int found = 0;
1768	for (vseg_id = vspace[vspace_id].vseg_offset;
1769	vseg_id < (vspace[vspace_id].vseg_offset + vspace[vspace_id].vsegs);
1770	vseg_id++)
1771	{
1772	if(vseg[vseg_id].type == VSEG_TYPE_ELF)
1773	{
1774	found = 1;
1775	break;
1776	}
1777	}
1778
1779	// We want one .elf file per vspace
1780	if (found == 0)
1781	{
1782	_printf("\n[BOOT ERROR] boot_elf_load() : "
1783	".elf file not found for vspace %s\n", vspace[vspace_id].name );
1784	_exit();
1785	}
1786
1787	load_one_elf_file( 0, // not a kernel file
1788	vseg[vseg_id].binpath, // file pathname
1789	vspace_id ); // vspace index
1790
1791	} // end for vspaces
1792
1793	} // end boot_elf_load()
1794
1795
1796	/////////////////////////////////////////////////////////////////////////////////
1797	// This function is executed in parallel by all processors[x][y][0].
1798	// It initialises the physical memory allocator in each cluster containing
1799	// a RAM pseg.
1800	/////////////////////////////////////////////////////////////////////////////////
1801	void boot_pmem_init( unsigned int cx,
1802	unsigned int cy )
1803	{
1804	mapping_header_t* header = (mapping_header_t *)SEG_BOOT_MAPPING_BASE;
1805	mapping_cluster_t* cluster = _get_cluster_base(header);
1806	mapping_pseg_t* pseg = _get_pseg_base(header);
1807
1808	unsigned int pseg_id;
1809	unsigned int procid = _get_procid();
1810	unsigned int lpid = procid & ((1<<P_WIDTH)-1);
1811
1812	if( lpid )
1813	{
1814	_printf("\n[BOOT ERROR] boot_pmem_init() : "
1815	"P[%d][%d][%d] should not execute it\n", cx, cy, lpid );
1816	_exit();
1817	}
1818
1819	// scan the psegs in local cluster to find pseg of type RAM
1820	unsigned int found = 0;
1821	unsigned int cluster_id = cx * Y_SIZE + cy;
1822	unsigned int pseg_min = cluster[cluster_id].pseg_offset;
1823	unsigned int pseg_max = pseg_min + cluster[cluster_id].psegs;
1824	for ( pseg_id = pseg_min ; pseg_id < pseg_max ; pseg_id++ )
1825	{
1826	if ( pseg[pseg_id].type == PSEG_TYPE_RAM )
1827	{
1828	unsigned int base = (unsigned int)pseg[pseg_id].base;
1829	unsigned int size = (unsigned int)pseg[pseg_id].length;
1830	_pmem_alloc_init( cx, cy, base, size );
1831	found = 1;
1832
1833	#if BOOT_DEBUG_PT
1834	_printf("\n[BOOT] pmem allocator initialised in cluster[%d][%d]"
1835	" : base = %x / size = %x\n", cx , cy , base , size );
1836	#endif
1837	break;
1838	}
1839	}
1840
1841	if ( found == 0 )
1842	{
1843	_printf("\n[BOOT ERROR] boot_pmem_init() : no RAM in cluster[%d][%d]\n",
1844	cx , cy );
1845	_exit();
1846	}
1847	} // end boot_pmem_init()
1848
1849	/////////////////////////////////////////////////////////////////////////
1850	// This function is the entry point of the boot code for all processors.
1851	/////////////////////////////////////////////////////////////////////////
1852	void boot_init()
1853	{
1854
1855	unsigned int gpid = _get_procid();
1856	unsigned int cx = gpid >> (Y_WIDTH + P_WIDTH);
1857	unsigned int cy = (gpid >> P_WIDTH) & ((1<<Y_WIDTH)-1);
1858	unsigned int lpid = gpid & ((1 << P_WIDTH) -1);
1859
1860	//////////////////////////////////////////////////////////
1861	// Phase ONE : only P[0][0][0] execute it
1862	//////////////////////////////////////////////////////////
1863	if ( gpid == 0 )
1864	{
1865	unsigned int cid; // index for loop on clusters
1866
1867	// initialises the TTY0 spin lock
1868	_spin_lock_init( &_tty0_spin_lock );
1869
1870	_printf("\n[BOOT] P[0,0,0] starts at cycle %d\n", _get_proctime() );
1871
1872	// initialises the IOC peripheral
1873	if ( USE_IOC_BDV != 0 ) _bdv_init();
1874	else if ( USE_IOC_HBA != 0 ) _hba_init();
1875	else if ( USE_IOC_SDC != 0 ) _sdc_init();
1876	else if ( USE_IOC_RDK == 0 )
1877	{
1878	_printf("\n[BOOT ERROR] boot_init() : no IOC peripheral\n");
1879	_exit();
1880	}
1881
1882	// initialises the FAT
1883	_fat_init( 0 ); // no IRQ
1884
1885	_printf("\n[BOOT] FAT initialised at cycle %d\n", _get_proctime() );
1886
1887	// Load the map.bin file into memory
1888	boot_mapping_init();
1889
1890	mapping_header_t* header = (mapping_header_t *)SEG_BOOT_MAPPING_BASE;
1891	mapping_cluster_t* cluster = _get_cluster_base(header);
1892
1893	_printf("\n[BOOT] Mapping %s loaded at cycle %d\n",
1894	header->name , _get_proctime() );
1895
1896	// initialises the barrier for all clusters containing processors
1897	unsigned int nclusters = 0;
1898	for ( cid = 0 ; cid < X_SIZE*Y_SIZE ; cid++ )
1899	{
1900	if ( cluster[cid].procs ) nclusters++ ;
1901	}
1902
1903	_simple_barrier_init( &_barrier_all_clusters , nclusters );
1904
1905	// wake up all processors P[x][y][0]
1906	for ( cid = 1 ; cid < X_SIZE*Y_SIZE ; cid++ )
1907	{
1908	unsigned int x = cluster[cid].x;
1909	unsigned int y = cluster[cid].y;
1910	unsigned int cluster_xy = (x << Y_WIDTH) + y;
1911
1912	if ( cluster[cid].procs )
1913	{
1914	unsigned long long paddr = (((unsigned long long)cluster_xy)<<32) +
1915	SEG_XCU_BASE+XCU_REG( XCU_WTI_REG , 0 );
1916
1917	_physical_write( paddr , (unsigned int)boot_entry );
1918	}
1919	}
1920
1921	_printf("\n[BOOT] Processors P[x,y,0] start at cycle %d\n",
1922	_get_proctime() );
1923	}
1924
1925	/////////////////////////////////////////////////////////////////
1926	// Phase TWO : All processors P[x][y][0] execute it in parallel
1927	/////////////////////////////////////////////////////////////////
1928	if( lpid == 0 )
1929	{
1930	// Initializes physical memory allocator in cluster[cx][cy]
1931	boot_pmem_init( cx , cy );
1932
1933	// Build page table in cluster[cx][cy]
1934	boot_ptab_init( cx , cy );
1935
1936	//////////////////////////////////////////////
1937	_simple_barrier_wait( &_barrier_all_clusters );
1938	//////////////////////////////////////////////
1939
1940	// P[0][0][0] complete page tables with vsegs
1941	// mapped in clusters without processors
1942	if ( gpid == 0 )
1943	{
1944	// complete page tables initialisation
1945	boot_ptab_extend();
1946
1947	_printf("\n[BOOT] Physical memory allocators and page tables"
1948	" initialized at cycle %d\n", _get_proctime() );
1949	}
1950
1951	//////////////////////////////////////////////
1952	_simple_barrier_wait( &_barrier_all_clusters );
1953	//////////////////////////////////////////////
1954
1955	// All processors P[x,y,0] activate MMU (using local PTAB)
1956	_set_mmu_ptpr( (unsigned int)(_ptabs_paddr[0][cx][cy]>>13) );
1957	_set_mmu_mode( 0xF );
1958
1959	// Each processor P[x,y,0] initialises all schedulers in cluster[x,y]
1960	boot_scheduler_init( cx , cy );
1961
1962	// Each processor P[x][y][0] initialises its CP0_SCHED register
1963	_set_sched( (unsigned int)_schedulers[cx][cy][0] );
1964
1965	//////////////////////////////////////////////
1966	_simple_barrier_wait( &_barrier_all_clusters );
1967	//////////////////////////////////////////////
1968
1969	if ( gpid == 0 )
1970	{
1971	_printf("\n[BOOT] Schedulers initialised at cycle %d\n",
1972	_get_proctime() );
1973	}
1974
1975	// All processor P[x,y,0] contributes to load .elf files into clusters.
1976	boot_elf_load();
1977
1978	//////////////////////////////////////////////
1979	_simple_barrier_wait( &_barrier_all_clusters );
1980	//////////////////////////////////////////////
1981
1982	// Each processor P[x][y][0] wake up other processors in same cluster
1983	mapping_header_t* header = (mapping_header_t *)SEG_BOOT_MAPPING_BASE;
1984	mapping_cluster_t* cluster = _get_cluster_base(header);
1985	unsigned int cluster_xy = (cx << Y_WIDTH) + cy;
1986	unsigned int cluster_id = (cx * Y_SIZE) + cy;
1987	unsigned int p;
1988	for ( p = 1 ; p < cluster[cluster_id].procs ; p++ )
1989	{
1990	_xcu_send_wti( cluster_xy , p , (unsigned int)boot_entry );
1991	}
1992
1993	// only P[0][0][0] makes display
1994	if ( gpid == 0 )
1995	{
1996	_printf("\n[BOOT] All processors start at cycle %d\n",
1997	_get_proctime() );
1998	}
1999	}
2000	// All other processors activate MMU (using local PTAB)
2001	if ( lpid != 0 )
2002	{
2003	_set_mmu_ptpr( (unsigned int)(_ptabs_paddr[0][cx][cy]>>13) );
2004	_set_mmu_mode( 0xF );
2005	}
2006
2007	// All processors set CP0_SCHED register
2008	_set_sched( (unsigned int)_schedulers[cx][cy][lpid] );
2009
2010	// All processors reset BEV bit in SR to use GIET_VM exception handler
2011	_set_sr( 0 );
2012
2013	// Each proocessor get kernel entry virtual address
2014	unsigned int kernel_entry = (unsigned int)&kernel_init_vbase;
2015
2016	#if BOOT_DEBUG_ELF
2017	_printf("\n[DEBUG BOOT_ELF] P[%d,%d,%d] exit boot & jump to %x at cycle %d\n",
2018	cx, cy, lpid, kernel_entry , _get_proctime() );
2019	#endif
2020
2021	// All processors jump to kernel_init
2022	asm volatile( "jr %0" ::"r"(kernel_entry) );
2023
2024	} // end boot_init()
2025
2026
2027	// Local Variables:
2028	// tab-width: 4
2029	// c-basic-offset: 4
2030	// c-file-offsets:((innamespace . 0)(inline-open . 0))
2031	// indent-tabs-mode: nil
2032	// End:
2033	// vim: filetype=c:expandtab:shiftwidth=4:tabstop=4:softtabstop=4
2034

Note: See TracBrowser for help on using the repository browser.

Download in other formats: