5 #include "microwatt_soc.h"
14 static inline void mtspr(int sprnum
, unsigned long val
)
16 __asm__
volatile("mtspr %0,%1" : : "i" (sprnum
), "r" (val
));
19 static inline uint32_t read32(const void *addr
)
21 return *(volatile uint32_t *)addr
;
24 static inline void write32(void *addr
, uint32_t value
)
26 *(volatile uint32_t *)addr
= value
;
36 uint32_t zero0
; // reserved
37 uint32_t zero1
; // reserved
43 void uart_writeuint32(uint32_t val
) {
44 const char lut
[] = { '0', '1', '2', '3', '4', '5', '6', '7',
45 '8', '9', 'A', 'B', 'C', 'D', 'E', 'F' };
46 uint8_t *val_arr
= (uint8_t*)(&val
);
49 for (i
= 0; i
< 4; i
++) {
50 putchar(lut
[(val_arr
[3-i
] >> 4) & 0xF]);
51 putchar(lut
[val_arr
[3-i
] & 0xF]);
55 void memcpy(void *dest
, void *src
, size_t n
) {
57 //cast src and dest to char*
58 char *src_char
= (char *)src
;
59 char *dest_char
= (char *)dest
;
62 if ((i
% 4096) == 0) {
68 dest_char
[i
] = src_char
[i
]; //copy contents byte by byte
78 #define TERCEL_SPI_REG_SYS_PHY_CFG1 0x10
79 #define TERCEL_SPI_REG_SYS_FLASH_CFG5 0x24
80 #define TERCEL_SPI_PHY_CLOCK_DIVISOR_MASK 0xff
81 #define TERCEL_SPI_PHY_CLOCK_DIVISOR_SHIFT 0
82 #define TERCEL_SPI_FLASH_EN_MULTCYC_READ_MASK 0x1
83 #define TERCEL_SPI_FLASH_EN_MULTCYC_READ_SHIFT 0
84 static inline uint32_t read_tercel_register(uint8_t reg
)
86 return readl((unsigned long)(SPI_FCTRL_BASE
+reg
));
89 static inline void write_tercel_register(uint8_t reg
, uint32_t value
)
91 writel(value
, (unsigned long)(SPI_FCTRL_BASE
+reg
));
94 // TODO: need to use this
95 // https://gitlab.raptorengineering.com/kestrel-collaboration/kestrel-firmware/bare-metal-firmware/-/blob/master/main.c#L2328
97 /* this is a "level 1" speed-up, which gets an initial improvement of 10-50x
98 * over the default speed (which is a scant 100 bytes per second).
100 static void crank_up_qspi_level1(void)
102 // WARNING: KESTREL SPECIFIC
103 // Set SPI clock cycle divider to 1
105 dword
= read_tercel_register(TERCEL_SPI_REG_SYS_PHY_CFG1
);
106 dword
&= ~(TERCEL_SPI_PHY_CLOCK_DIVISOR_MASK
<<
107 TERCEL_SPI_PHY_CLOCK_DIVISOR_SHIFT
);
108 dword
|= ((1 & TERCEL_SPI_PHY_CLOCK_DIVISOR_MASK
) <<
109 TERCEL_SPI_PHY_CLOCK_DIVISOR_SHIFT
);
110 write_tercel_register(TERCEL_SPI_REG_SYS_PHY_CFG1
, dword
);
111 // Enable read merging
112 dword
= read_tercel_register(TERCEL_SPI_REG_SYS_FLASH_CFG5
);
113 dword
|= (TERCEL_SPI_FLASH_EN_MULTCYC_READ_MASK
<<
114 TERCEL_SPI_FLASH_EN_MULTCYC_READ_SHIFT
);
115 write_tercel_register(TERCEL_SPI_REG_SYS_FLASH_CFG5
, dword
);
118 static bool fl_read(void *dst
, uint32_t offset
, uint32_t size
)
121 memcpy(d
, (void *)(unsigned long)(SPI_FLASH_BASE
+ offset
), size
);
125 static unsigned long copy_flash(unsigned int offset
, unsigned int dst_offs
)
129 unsigned int i
, poff
, size
, off
;
132 puts("Trying flash...\r\n");
133 if (!fl_read(&ehdr
, offset
, sizeof(ehdr
)))
135 if (!IS_ELF(ehdr
) || ehdr
.e_ident
[EI_CLASS
] != ELFCLASS64
) {
136 puts("Doesn't look like an elf64\r\n");
139 if (ehdr
.e_ident
[EI_DATA
] != ELFDATA2LSB
||
140 ehdr
.e_machine
!= EM_PPC64
) {
141 puts("Not a ppc64le binary\r\n");
145 poff
= offset
+ ehdr
.e_phoff
;
146 for (i
= 0; i
< ehdr
.e_phnum
; i
++) {
147 if (!fl_read(&ph
, poff
, sizeof(ph
)))
149 if (ph
.p_type
!= PT_LOAD
)
152 /* XXX Add bound checking ! */
154 addr
= (void *)ph
.p_vaddr
;
155 off
= offset
+ ph
.p_offset
;
156 //printf("Copy segment %d (0x%x bytes) to %p\n", i, size, addr);
157 puts("Copy segment ");
160 uart_writeuint32(size
);
162 uart_writeuint32((uint32_t)(unsigned long)addr
);
164 fl_read(addr
+dst_offs
, off
, size
);
165 poff
+= ehdr
.e_phentsize
;
168 puts("Booting from DRAM at");
169 uart_writeuint32((unsigned int)(dst_offs
+ehdr
.e_entry
));
172 puts("Dump DRAM\r\n");
173 for (i
= 0; i
< 64; i
++) {
174 uart_writeuint32(readl(dst_offs
+ehdr
.e_entry
+(i
*4)));
176 if ((i
& 7) == 7) puts("\r\n");
180 //flush_cpu_icache();
181 return dst_offs
+ehdr
.e_entry
;
184 for (i
= 0; i
< 8; i
++) {
185 uart_writeuint32(ehdr
.e_ident
[i
]);
194 // Defining gram_[read|write] allows a trace of all register
195 // accesses to be dumped to console for debugging purposes.
196 // To use, define GRAM_RW_FUNC in gram.h
197 uint32_t gram_read(const struct gramCtx
*ctx
, void *addr
) {
201 uart_writeuint32((unsigned long)addr
);
202 dword
= readl((unsigned long)addr
);
204 uart_writeuint32((unsigned long)dword
);
210 int gram_write(const struct gramCtx
*ctx
, void *addr
, uint32_t value
) {
211 puts("gram_write: ");
212 uart_writeuint32((unsigned long)addr
);
214 uart_writeuint32((unsigned long)value
);
215 writel(value
, (unsigned long)addr
);
222 const int kNumIterations
= 14;
223 int res
, failcnt
= 0;
225 unsigned long ftr
, spi_offs
=0x0;
226 volatile uint32_t *ram
= (uint32_t*)MEMORY_BASE
;
229 //puts("Firmware launched...\n");
232 puts(" Soc signature: ");
233 tmp
= readl(SYSCON_BASE
+ SYS_REG_SIGNATURE
);
234 uart_writeuint32(tmp
);
235 tmp
= readl(SYSCON_BASE
+ SYS_REG_SIGNATURE
+4);
236 uart_writeuint32(tmp
);
237 puts(" Soc features: ");
238 ftr
= readl(SYSCON_BASE
+ SYS_REG_INFO
);
239 if (ftr
& SYS_REG_INFO_HAS_UART
)
241 if (ftr
& SYS_REG_INFO_HAS_DRAM
)
243 if (ftr
& SYS_REG_INFO_HAS_BRAM
)
245 if (ftr
& SYS_REG_INFO_HAS_SPI_FLASH
)
247 if (ftr
& SYS_REG_INFO_HAS_LITEETH
)
251 if (ftr
& SYS_REG_INFO_HAS_SPI_FLASH
) {
252 // speed up the QSPI to at least a sane level
253 crank_up_qspi_level1();
255 puts("SPI Offset: ");
256 spi_offs
= readl(SYSCON_BASE
+ SYS_REG_SPI_INFO
);
257 uart_writeuint32(spi_offs
);
265 if (ftr
& SYS_REG_INFO_HAS_SPI_FLASH
) {
266 // print out configuration parameters for QSPI
267 volatile uint32_t *qspi_cfg
= (uint32_t*)SPI_FCTRL_BASE
;
268 for (int k
=0; k
< 2; k
++) {
269 tmp
= readl((unsigned long)&(qspi_cfg
[k
]));
273 uart_writeuint32(tmp
);
278 if (ftr
& SYS_REG_INFO_HAS_SPI_FLASH
) {
279 volatile uint32_t *qspi
= (uint32_t*)SPI_FLASH_BASE
+0x600000;
280 //volatile uint8_t *qspi_bytes = (uint8_t*)spi_offs;
281 // let's not, eh? writel(0xDEAF0123, (unsigned long)&(qspi[0]));
282 // tmp = readl((unsigned long)&(qspi[0]));
283 for (int i
=0;i
<2;i
++) {
284 tmp
= readl((unsigned long)&(qspi
[i
]));
285 uart_writeuint32(tmp
);
287 if ((i
& 0x7) == 0x7) puts("\r\n");
291 for (i=0;i<256;i++) {
292 tmp = readb((unsigned long)&(qspi_bytes[i]));
293 uart_writeuint32(tmp);
300 tmp
= readl((unsigned long)&(qspi
[0x1000/4]));
302 uart_writeuint32(tmp
);
306 unsigned char c
= getchar();
308 if (c
== 13) { // if CR send LF
311 tmp
= readl((unsigned long)&(qspi
[1<<i
]));
313 uart_writeuint32(1<<i
);
315 uart_writeuint32(tmp
);
326 volatile uint32_t *hyperram
= (uint32_t*)0x00000000; // at 0x0 for arty
327 writel(0xDEAF0123, (unsigned long)&(hyperram
[0]));
328 tmp
= readl((unsigned long)&(hyperram
[0]));
331 unsigned char c
= getchar();
333 if (c
== 13) { // if CR send LF
336 writel(0xDEAF0123+i
, (unsigned long)&(hyperram
[1<<i
]));
337 tmp
= readl((unsigned long)&(hyperram
[1<<i
]));
339 uart_writeuint32(1<<i
);
341 uart_writeuint32(tmp
);
350 // init DRAM only if SYSCON says it exists (duh)
351 if (ftr
& SYS_REG_INFO_HAS_DRAM
)
353 puts("DRAM init... ");
357 struct gramProfile profile
= {
359 0xb20, 0x806, 0x200, 0x0
366 struct gramProfile profile
= {
368 0x0320, 0x0006, 0x0200, 0x0000
374 struct gramProfile profile2
;
375 gram_init(&ctx
, &profile
, (void*)MEMORY_BASE
,
376 (void*)DRAM_CTRL_BASE
,
377 (void*)DRAM_INIT_BASE
);
380 puts("MR profile: ");
381 uart_writeuint32(profile
.mode_registers
[0]);
383 uart_writeuint32(profile
.mode_registers
[1]);
385 uart_writeuint32(profile
.mode_registers
[2]);
387 uart_writeuint32(profile
.mode_registers
[3]);
391 // Early read test for WB access sim
392 //uart_writeuint32(*ram);
396 for (size_t i
= 0; i
< 8; i
++) {
397 profile2
.rdly_p0
= i
;
398 gram_load_calibration(&ctx
, &profile2
);
399 gram_reset_burstdet(&ctx
);
401 for (size_t j
= 0; j
< 128; j
++) {
402 tmp
= readl((unsigned long)&(ram
[i
]));
404 if (gram_read_burstdet(&ctx
, 0)) {
413 for (size_t i
= 0; i
< 8; i
++) {
414 profile2
.rdly_p1
= i
;
415 gram_load_calibration(&ctx
, &profile2
);
416 gram_reset_burstdet(&ctx
);
417 for (size_t j
= 0; j
< 128; j
++) {
418 tmp
= readl((unsigned long)&(ram
[i
]));
420 if (gram_read_burstdet(&ctx
, 1)) {
428 puts("Auto calibrating... ");
429 res
= gram_generate_calibration(&ctx
, &profile2
);
430 if (res
!= GRAM_ERR_NONE
) {
432 gram_load_calibration(&ctx
, &profile
);
434 gram_load_calibration(&ctx
, &profile2
);
438 puts("Auto calibration profile:");
440 uart_writeuint32(profile2
.rdly_p0
);
442 uart_writeuint32(profile2
.rdly_p1
);
446 puts("Reloading built-in calibration profile...");
447 gram_load_calibration(&ctx
, &profile
);
449 puts("DRAM test... \n");
450 for (size_t i
= 0; i
< kNumIterations
; i
++) {
451 writel(0xDEAF0000 | i
*4, (unsigned long)&(ram
[i
]));
455 for (int dly
= 0; dly
< 8; dly
++) {
457 profile2
.rdly_p0
= dly
;
458 profile2
.rdly_p1
= dly
;
460 uart_writeuint32(profile2
.rdly_p0
);
462 uart_writeuint32(profile2
.rdly_p1
);
463 gram_load_calibration(&ctx
, &profile2
);
464 for (size_t i
= 0; i
< kNumIterations
; i
++) {
465 if (readl((unsigned long)&(ram
[i
])) != (0xDEAF0000 | i
*4)) {
467 uart_writeuint32((unsigned long)(&ram
[i
]));
469 uart_writeuint32(readl((unsigned long)&(ram
[i
])));
474 puts("Test canceled (more than 10 errors)\n");
482 for (size_t i
= 0; i
< kNumIterations
; i
++) {
483 if (readl((unsigned long)&(ram
[i
])) != (0xDEAF0000 | i
*4)) {
485 uart_writeuint32((unsigned long)(&ram
[i
]));
487 uart_writeuint32(readl((unsigned long)&(ram
[i
])));
492 puts("Test canceled (more than 10 errors)\n");
501 #if 0 // ooo, annoying: won't work. no idea why
502 // temporary hard-hack: boot directly from QSPI. really
503 // should do something like detect at least... something
504 if ((ftr
& SYS_REG_INFO_HAS_SPI_FLASH
))
506 // jump to absolute address
507 mtspr(8, SPI_FLASH_BASE
); // move address to LR
508 __asm__
volatile("blr");
513 // memcpy from SPI Flash then boot
514 if ((ftr
& SYS_REG_INFO_HAS_SPI_FLASH
) &&
518 puts("ELF @ QSPI\n");
519 // identify ELF, copy if present, and get the start address
520 unsigned long faddr = copy_flash(spi_offs,
523 // jump to absolute address
524 mtspr(8, faddr); // move address to LR
525 __asm__ volatile("blr");
527 // works with head.S which copies r3 into ctr then does bctr
532 // another terrible hack: copy from flash at offset 0x600000
533 // a block of size 0x600000 into mem address 0x600000, then
534 // jump to it. this allows a dtb image to be executed
536 volatile uint32_t *mem
= (uint32_t*)0x600000;
537 fl_read(mem
, 0x600000, 0x600000); // shorter (testing) 0x8000);
539 for (int i
=0;i
<256;i
++) {
540 tmp
= readl((unsigned long)&(mem
[i
]));
541 uart_writeuint32(tmp
);
543 if ((i
& 0x7) == 0x7) puts("\r\n");
546 mtspr(8, 0x600000); // move address to LR
547 __asm__
volatile("blr");