1 | /* |
---|
2 | * \file : reset.S |
---|
3 | * \date : 01/12/2012 |
---|
4 | * \author: Cesar FUGUET & Manuel BOUYER & Alain Greiner |
---|
5 | * |
---|
6 | * This is a generic reset code for a generic multi-clusters / multi-processors |
---|
7 | * TSAR architecture (up to 256 clusters / up to 4 processors per cluster). |
---|
8 | * |
---|
9 | * There is one XICU, one TTY, one DMA, and one memory bank per cluster. |
---|
10 | * |
---|
11 | * This preloader uses a stack segment allocated in cluster 0 for processor 0. |
---|
12 | * The stack allocation is not performed for other processors as they do not |
---|
13 | * need it during the preloader execution. Therefore, this allocation should be |
---|
14 | * done by the loaded Operating System. |
---|
15 | * |
---|
16 | * The replicated XICU is used to awake the sleeping processors: |
---|
17 | * xicu_paddr_base = ICU_PADDR_BASE + (cluster_xy << 32) |
---|
18 | * |
---|
19 | * It is intended to be used with various operating systems or nano kernels, |
---|
20 | * including NetBSD, Linux, ALMOS, and GIET_VM. |
---|
21 | * |
---|
22 | * - Each processor initializes its Status Register (SR) to disable interrupts. |
---|
23 | * - Each processor initializes its Count Register. |
---|
24 | * - Each processor initialises its private XICU WTI mask register. |
---|
25 | * - Only processor 0 executes the reset_load_elf function to load into memory |
---|
26 | * the system specific boot-loader stored on disk at RESET_LOADER_LBA |
---|
27 | * - All other processors wait in a low power consumption mode that the |
---|
28 | * processor 0 wakes them using an IPI (Inter Processor Interruption) |
---|
29 | */ |
---|
30 | |
---|
31 | #include <defs.h> |
---|
32 | #include <mips32_registers.h> |
---|
33 | |
---|
34 | /* These define should be consistent with values defined in map.xml file */ |
---|
35 | |
---|
36 | .section .reset,"ax",@progbits |
---|
37 | |
---|
38 | .extern reset_putc |
---|
39 | .extern reset_getc |
---|
40 | .extern reset_ioc_read |
---|
41 | .extern reset_elf_loader |
---|
42 | .extern memcpy |
---|
43 | .extern reset_puts |
---|
44 | .extern reset_putx |
---|
45 | .extern reset_putd |
---|
46 | .extern reset_ioc_init |
---|
47 | .extern versionstr |
---|
48 | .extern dtb_start |
---|
49 | .extern dtb_addr |
---|
50 | |
---|
51 | .globl reset /* Makes reset an external symbol */ |
---|
52 | .ent reset |
---|
53 | |
---|
54 | .align 2 |
---|
55 | .set noreorder |
---|
56 | |
---|
57 | reset: |
---|
58 | b _reset /* 0xbfc0000 */ |
---|
59 | nop /* 0xbfc0004 */ |
---|
60 | |
---|
61 | /* Addresses of the functions provided by this reset code */ |
---|
62 | |
---|
63 | preloader_vector: |
---|
64 | .word RESET_VERSION /* 0xbfc0008 */ |
---|
65 | .word dtb_start /* 0xbfc000c */ |
---|
66 | .word reset_putc /* 0xbfc0010 */ |
---|
67 | .word reset_getc /* 0xbfc0014 */ |
---|
68 | .word reset_ioc_read /* 0xbfc0018 */ |
---|
69 | .word reset_elf_loader /* 0xbfc001C */ |
---|
70 | .word memcpy /* 0xbfc0020 */ |
---|
71 | .word reset_puts /* 0xbfc0024 */ |
---|
72 | .word reset_putx /* 0xbfc0028 */ |
---|
73 | .word reset_putd /* 0xbfc002C */ |
---|
74 | |
---|
75 | _reset: |
---|
76 | |
---|
77 | /* All processors Disable interruptions, keep STATUSbev enabled */ |
---|
78 | |
---|
79 | li k0, (1 << 22) |
---|
80 | mtc0 k0, CP0_STATUS |
---|
81 | |
---|
82 | /* |
---|
83 | * All processors compute gpid, lpid, cluster_xy |
---|
84 | * gpid = ebase[11:0] = X_WIDTH : Y_WIDTH : P_WIDTH |
---|
85 | * x y lpid |
---|
86 | * X, Y and LPID fields are left-aligned |
---|
87 | */ |
---|
88 | |
---|
89 | mfc0 k0, CP0_EBASE |
---|
90 | andi t0, k0, 0xFFF /* t0 <= gpid (<= 4096 procs) */ |
---|
91 | andi t1, t0, ((1<<P_WIDTH)-1) /* t1 <= lpid */ |
---|
92 | srl t2, t0, P_WIDTH /* t2 <= cluster_xy */ |
---|
93 | |
---|
94 | /* All processors initialize the count register in CP0 */ |
---|
95 | |
---|
96 | mtc0 zero, CP0_COUNT |
---|
97 | |
---|
98 | #if USE_32BIT |
---|
99 | /*** VERSION 1 : 32 bits ***/ |
---|
100 | |
---|
101 | /* |
---|
102 | * If the addresses are 32-bit wide, we need to compute the address |
---|
103 | * if the XICU for each cluster |
---|
104 | * All processors enable the WTI for XICU |
---|
105 | * Each processor may have IRQ_PER_PROC irq outputs from the XICU |
---|
106 | * In each cluster, the XICU base address depends on the cluster_xy |
---|
107 | */ |
---|
108 | la t3, SEG_XCU_BASE /* t3 <= ICU base address */ |
---|
109 | li t4, 1 /* t4 <= 1 */ |
---|
110 | sll t4, t4, X_WIDTH /* t4 <= 1 << X_WIDTH */ |
---|
111 | li t5, 1 /* t5 <= 1 */ |
---|
112 | sll t5, t5, Y_WIDTH /* t5 <= 1 << Y_WIDTH */ |
---|
113 | multu t4, t5 /* X_WIDTH * Y_WIDTH */ |
---|
114 | mflo t4 /* t4 <= X_WIDTH * Y_WIDTH */ |
---|
115 | lui t5, 0x8000 /* t5 <= 0x80000000 */ |
---|
116 | divu t5, t4 /* (Address increment per cluster) / 2*/ |
---|
117 | mflo t4 /* t4 <= Increment / 2 */ |
---|
118 | sll t4, t4, 1 /* t4 <= Address increment per clus. */ |
---|
119 | mult t4, t2 /* Cluster increment * Cluster num. */ |
---|
120 | mflo t4 /* Cluster base address */ |
---|
121 | addu t3, t3, t4 /* t3 <= XICU base address in clus. */ |
---|
122 | |
---|
123 | move t4, t1 /* t4 <= local_id */ |
---|
124 | li t5, IRQ_PER_PROCESSOR /* t5 <= IRQ_PER_PROCESSOR */ |
---|
125 | multu t4, t5 |
---|
126 | mflo t6 /* t6 <= IRQ_PER_PROC * local_id */ |
---|
127 | sll t4, t6, 2 /* t4 <= OUT_INDEX = t6 * 4 */ |
---|
128 | |
---|
129 | li t5, (0xC << 7) /* t5 <= FUNC = XICU_MSK_WTI */ |
---|
130 | or t4, t4, t5 /* t4 <= FUNC | INDEX | 00 */ |
---|
131 | or t5, t3, t4 /* t5 <= &XICU[MSK_WTI][OUT_INDEX] */ |
---|
132 | |
---|
133 | /* All processors set WTI mask */ |
---|
134 | |
---|
135 | li t4, 1 |
---|
136 | sllv t4, t4, t1 /* Set XICU[MSK_WTI][INDEX][local_id] */ |
---|
137 | sw t4, 0(t5) /* XICU[MSK_WTI][INDEX] <= t4 */ |
---|
138 | #else |
---|
139 | /*** VERSION 2 : 40 bits ***/ |
---|
140 | |
---|
141 | /* |
---|
142 | * All processors enable the WTI for XICU |
---|
143 | * Each processor may have IRQ_PER_PROC irq outputs from the XICU |
---|
144 | * In each cluster, the XICU base address depends on the cluster_xy |
---|
145 | */ |
---|
146 | la t3, SEG_XCU_BASE /* t3 <= ICU base address */ |
---|
147 | move t4, t1 /* t4 <= local_id */ |
---|
148 | li t5, IRQ_PER_PROCESSOR /* t5 <= IRQ_PER_PROCESSOR */ |
---|
149 | multu t4, t5 |
---|
150 | mflo t6 /* t6 <= IRQ_PER_PROC * local_id */ |
---|
151 | sll t4, t6, 2 /* t4 <= OUT_INDEX = t6 * 4 */ |
---|
152 | |
---|
153 | li t5, (0xC << 7) /* t5 <= FUNC = XICU_MSK_WTI */ |
---|
154 | or t4, t4, t5 /* t4 <= FUNC | INDEX | 00 */ |
---|
155 | or t5, t3, t4 /* t5 <= &XICU[MSK_WTI][OUT_INDEX] */ |
---|
156 | |
---|
157 | /* All processors set WTI mask using the physical address extension */ |
---|
158 | |
---|
159 | li t4, 1 |
---|
160 | sllv t4, t4, t1 /* Set XICU[MSK_WTI][INDEX][local_id] */ |
---|
161 | |
---|
162 | mtc2 t2, CP2_PADDR_EXT /* set PADDR extension */ |
---|
163 | sw t4, 0(t5) /* XICU[MSK_WTI][INDEX] <= t4 */ |
---|
164 | mtc2 zero, CP2_PADDR_EXT /* reset PADDR extension */ |
---|
165 | #endif |
---|
166 | |
---|
167 | /* |
---|
168 | * Only the bootstrap processor loads and executes the boot-loader |
---|
169 | * We have: |
---|
170 | * t0: global pid |
---|
171 | * t1: local pid |
---|
172 | * t2: cluster_xy |
---|
173 | * t3: xicu physical base address in bootstrap cluster |
---|
174 | */ |
---|
175 | |
---|
176 | li t4, BS_PROC |
---|
177 | bne t4, t0, _reset_wait |
---|
178 | nop |
---|
179 | |
---|
180 | /* Bootstrap Processor initializes stack pointer */ |
---|
181 | |
---|
182 | la k0, _stack |
---|
183 | li k1, RESET_STACK_SIZE /* k1 <= P0 stack size */ |
---|
184 | addu sp, k0, k1 /* P0 stack from base to (base + size) */ |
---|
185 | |
---|
186 | /* Bootstrap Processor displays version for this reset code */ |
---|
187 | |
---|
188 | la a0, versionstr |
---|
189 | jal reset_puts |
---|
190 | nop |
---|
191 | |
---|
192 | /* Bootstrap Processor initializes the block device */ |
---|
193 | |
---|
194 | jal reset_ioc_init |
---|
195 | nop |
---|
196 | |
---|
197 | /* |
---|
198 | * Bootstrap Processor jumps to the reset_elf_loader routine passing as argument |
---|
199 | * the block number in which is loaded the .elf file |
---|
200 | */ |
---|
201 | |
---|
202 | li a0, RESET_LOADER_LBA |
---|
203 | jal reset_elf_loader |
---|
204 | nop |
---|
205 | |
---|
206 | /* |
---|
207 | * Bootstrap Processor jumps to the entry address defined in the .elf file, and |
---|
208 | * returned by reset_elf_loader function. |
---|
209 | * First argument is pointer to the preloader function vectors other |
---|
210 | * function arguments are 0 |
---|
211 | */ |
---|
212 | |
---|
213 | la a0, preloader_vector |
---|
214 | lw a1, dtb_addr |
---|
215 | move a2, zero |
---|
216 | move a3, zero |
---|
217 | jr v0 |
---|
218 | nop |
---|
219 | |
---|
220 | /* |
---|
221 | * All processor (but processor 0) wait in low power mode until processor 0 |
---|
222 | * wakes them using an IPI. |
---|
223 | * We have: |
---|
224 | * t0: global id |
---|
225 | * t1: local id |
---|
226 | * t2: cluster id |
---|
227 | * t3: xicu physical base address in cluster 0 |
---|
228 | */ |
---|
229 | |
---|
230 | _reset_wait: |
---|
231 | |
---|
232 | sll t4, t1, 2 /* t4 <= local_id * 4 */ |
---|
233 | addu t5, t4, t3 /* t5 <= &XICU[WTI_REG][local_id] */ |
---|
234 | |
---|
235 | wait |
---|
236 | |
---|
237 | /* |
---|
238 | * All other processors, when exiting wait mode, read from XICU the address |
---|
239 | * to jump. |
---|
240 | * This address is the boot-loader entry address that has been written in |
---|
241 | * the mailbox by the IPI sent by processor 0 |
---|
242 | */ |
---|
243 | |
---|
244 | mtc2 t2, CP2_PADDR_EXT /* set PADDR extension */ |
---|
245 | lw k0, 0(t5) /* k0 <= XICU[WTI_REG][local_id] */ |
---|
246 | mtc2 zero, CP2_PADDR_EXT /* reset PADDR extension */ |
---|
247 | |
---|
248 | jr k0 |
---|
249 | nop |
---|
250 | |
---|
251 | /* Exception entry point */ |
---|
252 | |
---|
253 | .org 0x0380 |
---|
254 | _excep: |
---|
255 | mfc0 a0, CP0_STATUS /* first arg is status */ |
---|
256 | mfc0 a1, CP0_CAUSE /* second arg is cause */ |
---|
257 | mfc0 a2, CP0_EPC /* third argc is epc */ |
---|
258 | mfc2 a3, CP2_DBVAR /* fourth argc is dbvar */ |
---|
259 | nop |
---|
260 | j handle_except |
---|
261 | nop |
---|
262 | |
---|
263 | .end reset |
---|
264 | |
---|
265 | .set reorder |
---|
266 | |
---|
267 | .section .data |
---|
268 | |
---|
269 | _stack: |
---|
270 | |
---|
271 | .space RESET_STACK_SIZE |
---|
272 | |
---|
273 | /* |
---|
274 | * vim: tabstop=4 : shiftwidth=4 : expandtab |
---|
275 | */ |
---|