source: soft/giet_vm/giet_kernel/kernel_init.c @ 364

Last change on this file since 364 was 345, checked in by cfuguet, 10 years ago

giet_vm optimizations:

  • Several modifications in GIET_VM in order to support compilation with GCC optimizations (-O2) activated.
  • Adding missing volatile in some global variables.
  • Using ioread and iowrite utility functions in peripheral drivers which prevent GCC to remove writes or reads in hardware memory mapped registers.
  • Code refactoring of stdio printf functions. Now, shr_printf and tty_printf function reuse the same function body. The only difference is that shr_printf wraps printf function call with TTY get lock and release lock.
  • Property svn:executable set to *
File size: 13.3 KB
Line 
1///////////////////////////////////////////////////////////////////////////////////
2// File     : kernel_init.c
3// Date     : 26/05/2012
4// Authors  : alain greiner & mohamed karaoui
5// Copyright (c) UPMC-LIP6
6////////////////////////////////////////////////////////////////////////////////////
7// The kernel_init.c file is part of the GIET-VM nano-kernel.
8//
9// This nano-kernel has been written for the MIPS32 processor.
10// The virtual adresses are on 32 bits and use the (unsigned int) type, but the
11// physicals addresses can have up to 40 bits, and use the (unsigned long long) type.
12// It natively supports clusterised shared mmemory multi-processors architectures,
13// where each processor is identified by a composite index [x,y,lpid],
14// and where there is one physical memory bank per cluster.
15//
16// The kernel_init() function is executed in parallel by all procesors,
17// and completes the system initialisation that has been started by processor[0,0,0]
18// in the boot_init() function. It makes the following assuptions, regarding the work
19// bone by the boot code:
20//
21// 1) The page tables associated to the various vspaces have been build
22//    in physical memory, and can be used by the kernel code.
23//
24// 2) All schedulers (this include all task contexts) have been initialised,
25//    Both the virtual and the physical base addresses of the page tables
26//    are available in the CTX_PTAB and CTX_PTPR slots.
27//
28// 3) The CP0_SCHED register of each processor contains a pointer on its
29//    private scheduler (virtual address).
30//
31// 4) The CP2_PTPR register of each processor contains a pointer on
32//    the vspace_0 page table (physical address>>13).
33//
34// 5) For all processors, the MMU is activated (CP2_MODE contains 0xF).
35//
36// This code must be loaded in .kinit section, in order to control seg_kinit_base,
37// as this address is used by the boot code to jump into kernel code.
38//
39// Each processor performs the following actions:
40// 1/ contribute to _schedulers_paddr[] array initialisation.
41// 2/ contribute to _ptabs_paddr[] and _ptabs_vaddr arrays initialisation
42// 3/ completes task context initialisation for ech allocated task
43// 4/ compute and set the ICU mask for its private ICU channel
44// 5/ initialise its private TICK timer (if tasks > 0)
45// 6/ initialise the "idle" task context in its private scheduler
46// 7/ initialise SP, SR, PTPR, EPC registers and jump to user code with an eret.
47////////////////////////////////////////////////////////////////////////////////////
48
49#include <giet_config.h>
50
51// kernel libraries
52#include <utils.h>
53#include <fat32.h>
54
55//for peripheral initialisation
56#include <dma_driver.h>
57#include <fbf_driver.h>
58#include <tty_driver.h>
59#include <icu_driver.h>
60#include <xcu_driver.h>
61#include <ioc_driver.h>
62#include <mmc_driver.h>
63#include <mwr_driver.h>
64#include <nic_driver.h>
65#include <tim_driver.h>
66
67#include <ctx_handler.h>
68#include <irq_handler.h>
69
70#include <mapping_info.h>
71#include <mips32_registers.h>
72
73#if !defined(X_SIZE)
74# error: You must define X_SIZE in the hard_config.h file
75#endif
76
77#if !defined(Y_SIZE)
78# error: You must define Y_SIZE in the hard_config.h file
79#endif
80
81#if !defined(Y_WIDTH)
82# error: You must define Y_WIDTH in the hard_config.h file
83#endif
84
85#if !defined(Y_WIDTH)
86# error: You must define Y_WIDTH in the hard_config.h file
87#endif
88
89#if !defined(NB_PROCS_MAX)
90# error: You must define NB_PROCS_MAX in the hard_config.h file
91#endif
92
93#if !defined(NB_TOTAL_PROCS)
94# error: You must define NB_TOTAL_PROCS in the hard_config.h file
95#endif
96
97#if !defined(USE_XCU)
98# error: You must define USE_XCU in the hard_config.h file
99#endif
100
101#if !defined(IDLE_TASK_INDEX)
102# error: You must define IDLE_TASK_INDEX in the giet_config.h file
103#endif
104
105#if !defined(GIET_TICK_VALUE)
106# error: You must define GIET_TICK_VALUE in the giet_config.h file
107#endif
108
109#if !defined(GIET_NB_VSPACE_MAX)
110# error: You must define GIET_NB_VSPACE_MAX in the giet_config.h file
111#endif
112
113///////////////////////////////////////////////////////////////////////////////////
114// array of pointers on the page tables (virtual addresses)
115///////////////////////////////////////////////////////////////////////////////////
116
117__attribute__((section (".kdata"))) 
118volatile unsigned int _ptabs_vaddr[GIET_NB_VSPACE_MAX];    // virtual addresses
119
120__attribute__((section (".kdata")))       
121volatile unsigned int _ptabs_ptprs[GIET_NB_VSPACE_MAX];    // physical addresses >> 13
122
123///////////////////////////////////////////////////////////////////////////////////
124// array of pointers on the schedulers (physical addresses)
125///////////////////////////////////////////////////////////////////////////////////
126
127__attribute__((section (".kdata"))) 
128volatile static_scheduler_t* _schedulers[NB_PROCS_MAX<<(X_WIDTH+Y_WIDTH)]; // virtual addresses
129
130////////////////////////////////////////////////////////////////////////////////////
131// staks for the "idle" tasks (512 bytes for each processor)
132////////////////////////////////////////////////////////////////////////////////////
133
134__attribute__((section (".kdata"))) 
135volatile unsigned int _idle_stack[X_SIZE * Y_SIZE * NB_PROCS_MAX * 128 ]; 
136
137////////////////////////////////////////////////////////////////////////////////////
138// Synchonisation Barrier before jumping to user code
139////////////////////////////////////////////////////////////////////////////////////
140
141__attribute__((section (".kdata"))) 
142volatile unsigned int _init_barrier = 0;
143
144///////////////////////////////////////////////////////////////////////////////////
145__attribute__((section (".kinit"))) void kernel_init() 
146{
147    unsigned int global_pid = _get_procid();
148    unsigned int cluster_xy = global_pid / NB_PROCS_MAX;
149    unsigned int x          = cluster_xy >> Y_WIDTH;
150    unsigned int y          = cluster_xy & ((1<<Y_WIDTH)-1);
151    unsigned int lpid       = global_pid % NB_PROCS_MAX;
152    unsigned int pid        = ((( x * Y_SIZE) + y) * NB_PROCS_MAX) + lpid;
153
154    // This last initialisation phase is done sequencially:
155    while( pid != _init_barrier ) asm volatile ( "nop" );
156
157    // Step 1 : each processor get its scheduler virtual address
158    //          and contribute to initialise the _schedulers[] array
159
160    static_scheduler_t* psched     = (static_scheduler_t*)_get_sched();
161    unsigned int        tasks      = psched->tasks;
162
163    _schedulers[global_pid] = psched;
164
165#if GIET_DEBUG_INIT
166_printf("\n[GIET DEBUG INIT] Processor[%d,%d,%d] starts kernel init\n"
167        " - scheduler vbase = %x\n"
168        " - tasks           = %d\n",
169        x, y, lpid, (unsigned int)psched, tasks );
170#endif
171
172    // step 2 : each processor that is allocated at least one task loops
173    //          on all allocated tasks:
174    //          - contributes to _ptabs_vaddr[] & _ptabs_ptprs[] initialisation.
175    //          - set CTX_RA slot  with the kernel _ctx_eret() virtual address.
176    //          - set CTX_EPC slot that must contain the task entry point,
177    //            and contain only at this point the virtual address of the memory
178    //            location containing this entry point. We must switch the PTPR
179    //            to use the page table corresponding to the task.
180
181    unsigned int ltid;
182
183    for (ltid = 0; ltid < tasks; ltid++) 
184    {
185        unsigned int vsid = _get_task_slot( global_pid, ltid , CTX_VSID_ID ); 
186        unsigned int ptab = _get_task_slot( global_pid, ltid , CTX_PTAB_ID ); 
187        unsigned int ptpr = _get_task_slot( global_pid, ltid , CTX_PTPR_ID ); 
188
189        // initialize PTABS arrays
190        _ptabs_vaddr[vsid] = ptab;
191        _ptabs_ptprs[vsid] = ptpr;
192
193#if GIET_DEBUG_INIT
194_printf("\n[GIET DEBUG INIT] Processor[%d,%d,%d] contributes to PTABS arrays\n"
195        " - ptabs_vaddr[%d] = %x / ptpr_paddr[%d] = %l\n",
196        x, y, lpid, 
197        vsid, ptab, vsid, ((unsigned long long)ptpr)<<13 );
198#endif
199
200        // set the ptpr to use the task page table
201        asm volatile( "mtc2    %0,   $0   \n"
202                      : : "r" (ptpr) );
203
204        // compute ctx_ra
205        unsigned int ctx_ra = (unsigned int)(&_ctx_eret);
206        _set_task_slot( global_pid, ltid, CTX_RA_ID, ctx_ra );
207
208        // compute ctx_epc
209        unsigned int* ptr = (unsigned int*)_get_task_slot( global_pid, ltid, CTX_EPC_ID );
210        _set_task_slot( global_pid, ltid, CTX_EPC_ID, *ptr );
211
212#if GIET_DEBUG_INIT
213_printf("\n[GIET DEBUG INIT] Processor[%d,%d,%d] updates context for task %d\n"
214        " - ctx_epc   = %x\n"
215        " - ctx_ra    = %x\n",
216        x, y, lpid, ltid,
217        _get_task_slot( global_pid, ltid, CTX_EPC_ID ),
218        _get_task_slot( global_pid, ltid, CTX_RA_ID ) );
219#endif
220
221    }  // end for tasks
222
223    // step 4 : compute and set ICU or XCU masks
224
225    unsigned int isr_switch_index = 0xFFFFFFFF;
226    unsigned int hwi_mask = 0;
227    unsigned int pti_mask = 0;
228    unsigned int wti_mask = 0;
229    unsigned int irq_id;            // IN_IRQ index
230    unsigned int entry;             // interrupt vector entry
231
232    for (irq_id = 0; irq_id < 32; irq_id++) 
233    {
234        entry = psched->hwi_vector[irq_id];
235        if ( entry & 0x80000000 ) hwi_mask = hwi_mask | (1<<irq_id);
236        if ( (entry & 0x0000FFFF) == ISR_TICK ) isr_switch_index = irq_id;
237
238        entry = psched->pti_vector[irq_id];
239        if ( entry & 0x80000000 ) pti_mask = pti_mask | (1<<irq_id);
240        if ( (entry & 0x0000FFFF) == ISR_TICK ) isr_switch_index = irq_id;
241
242        entry = psched->wti_vector[irq_id];
243        if ( entry & 0x80000000 ) wti_mask = wti_mask | (1<<irq_id);
244        if ( (entry & 0x0000FFFF) == ISR_TICK ) isr_switch_index = irq_id;
245    }
246
247#if GIET_DEBUG_INIT
248_printf("\n[GIET DEBUG INIT] Processor[%d,%d,%d] sets XCU masks\n"
249        " - ICU HWI_MASK = %x\n"
250        " - ICU WTI_MASK = %x\n"
251        " - ICU PTI_MASK = %x\n",
252        x, y, lpid, hwi_mask, wti_mask, pti_mask );
253#endif
254
255    unsigned int channel = lpid * IRQ_PER_PROCESSOR; 
256
257#if USE_XCU
258    _xcu_set_mask( cluster_xy, channel, hwi_mask, IRQ_TYPE_HWI ); 
259    _xcu_set_mask( cluster_xy, channel, wti_mask, IRQ_TYPE_WTI );
260    _xcu_set_mask( cluster_xy, channel, pti_mask, IRQ_TYPE_PTI );
261#else
262    _icu_set_mask( cluster_xy, channel, hwi_mask );   
263#endif
264
265    // step 5 : start TICK timer if at least one task
266    if (tasks > 0) 
267    {
268        // one ISR_TICK must be defined for each proc
269        if (isr_switch_index == 0xFFFFFFFF) 
270        {
271            _printf("\n[GIET ERROR] ISR_TICK not found for processor[%d,%d,%d]\n",
272                    x, y, lpid );
273            _exit();
274        }
275
276        // start system timer
277
278#if USE_XCU
279        _xcu_timer_start( cluster_xy, isr_switch_index, GIET_TICK_VALUE ); 
280#else
281        _timer_start( cluster_xy, isr_switch_index, GIET_TICK_VALUE ); 
282#endif
283
284    }
285
286#if GIET_DEBUG_INIT
287_printf("\n[GIET DEBUG INIT] Processor[%d,%d,%d] starts TICK timer\n",
288        x, y, lpid );
289#endif
290
291    // step 6 : each processor updates the idle_task context:
292    //          (only CTX_SP, CTX_RA, CTX_EPC).
293    //          The stack size is 512 bytes, reserved in seg_kdata.
294    //          The PTPR register, the CTX_PTPR and CTX_PTAB slots
295    //          have been initialised in boot code.
296
297    unsigned int p = ((x * Y_SIZE) + y) * NB_PROCS_MAX + lpid; 
298
299    unsigned int stack = (unsigned int)_idle_stack + ((p + 1)<<9);
300
301    _set_task_slot( global_pid, IDLE_TASK_INDEX, CTX_SP_ID,  stack);
302    _set_task_slot( global_pid, IDLE_TASK_INDEX, CTX_RA_ID,  (unsigned int) &_ctx_eret);
303    _set_task_slot( global_pid, IDLE_TASK_INDEX, CTX_EPC_ID, (unsigned int) &_idle_task);
304
305#if GIET_DEBUG_INIT
306_printf("\n[GIET DEBUG INIT] Processor[%d,%d,%d] initializes IDLE task\n",
307        x, y, lpid );
308#endif
309
310    // step 7 : when all processors reach the synchronisation barrier,
311    //          each processor set registers SP, SR, PTPR, EPC,
312    //          with the values corresponding to the first allocated task,
313    //          or to the idle_task if there is no task allocated,
314    //          and jump to user code
315
316    if (tasks == 0) 
317    {
318        ltid = IDLE_TASK_INDEX;
319
320        _printf("\n[GIET WARNING] No task allocated to processor[%d,%d,%d]\n",
321                x, y, lpid );
322    }
323    else
324    {
325        ltid = 0;
326    }
327
328    unsigned int sp_value   = _get_task_slot(global_pid, ltid, CTX_SP_ID);
329    unsigned int sr_value   = _get_task_slot(global_pid, ltid, CTX_SR_ID);
330    unsigned int ptpr_value = _get_task_slot(global_pid, ltid, CTX_PTPR_ID);
331    unsigned int epc_value  = _get_task_slot(global_pid, ltid, CTX_EPC_ID);
332
333#if GIET_DEBUG_INIT
334_printf("\n[GIET DEBUG INIT] Processor[%d,%d,%d] reach barrier at cycle %d\n",
335        x, y, lpid, _get_proctime() );
336#endif
337
338    // increment barrier counter
339    _init_barrier++;
340
341    // busy waiting until all processors synchronized
342    while ( _init_barrier != NB_TOTAL_PROCS );
343
344    // set registers and jump to user code
345    asm volatile ( "move  $29,  %0                  \n"   /* SP <= ctx[CTX_SP_ID] */
346                   "mtc0  %1,   $12                 \n"   /* SR <= ctx[CTX_SR_ID] */
347                   "mtc2  %2,   $0                  \n"   /* PTPR <= ctx[CTX_PTPR] */
348                   "mtc0  %3,   $14                 \n"   /* EPC <= ctx[CTX_EPC]  */
349                   "eret                            \n"   /* jump to user code  */
350                   "nop                             \n"
351                   : 
352                   : "r"(sp_value), "r"(sr_value), "r"(ptpr_value), "r"(epc_value)
353                   : "$29", "memory" );
354
355} // end kernel_init()
356
357
358// Local Variables:
359// tab-width: 4
360// c-basic-offset: 4
361// c-file-offsets:((innamespace . 0)(inline-open . 0))
362// indent-tabs-mode: nil
363// End:
364// vim: filetype=c:expandtab:shiftwidth=4:tabstop=4:softtabstop=4
365
Note: See TracBrowser for help on using the repository browser.