xref: /openbmc/qemu/hw/riscv/boot.c (revision 7c08eefc)
1 /*
2  * QEMU RISC-V Boot Helper
3  *
4  * Copyright (c) 2017 SiFive, Inc.
5  * Copyright (c) 2019 Alistair Francis <alistair.francis@wdc.com>
6  *
7  * This program is free software; you can redistribute it and/or modify it
8  * under the terms and conditions of the GNU General Public License,
9  * version 2 or later, as published by the Free Software Foundation.
10  *
11  * This program is distributed in the hope it will be useful, but WITHOUT
12  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
13  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
14  * more details.
15  *
16  * You should have received a copy of the GNU General Public License along with
17  * this program.  If not, see <http://www.gnu.org/licenses/>.
18  */
19 
20 #include "qemu/osdep.h"
21 #include "qemu/datadir.h"
22 #include "qemu/units.h"
23 #include "qemu/error-report.h"
24 #include "exec/cpu-defs.h"
25 #include "hw/boards.h"
26 #include "hw/loader.h"
27 #include "hw/riscv/boot.h"
28 #include "hw/riscv/boot_opensbi.h"
29 #include "elf.h"
30 #include "sysemu/device_tree.h"
31 #include "sysemu/qtest.h"
32 #include "sysemu/kvm.h"
33 #include "sysemu/reset.h"
34 
35 #include <libfdt.h>
36 
37 bool riscv_is_32bit(RISCVHartArrayState *harts)
38 {
39     RISCVCPUClass *mcc = RISCV_CPU_GET_CLASS(&harts->harts[0]);
40     return mcc->misa_mxl_max == MXL_RV32;
41 }
42 
43 /*
44  * Return the per-socket PLIC hart topology configuration string
45  * (caller must free with g_free())
46  */
47 char *riscv_plic_hart_config_string(int hart_count)
48 {
49     g_autofree const char **vals = g_new(const char *, hart_count + 1);
50     int i;
51 
52     for (i = 0; i < hart_count; i++) {
53         CPUState *cs = qemu_get_cpu(i);
54         CPURISCVState *env = &RISCV_CPU(cs)->env;
55 
56         if (kvm_enabled()) {
57             vals[i] = "S";
58         } else if (riscv_has_ext(env, RVS)) {
59             vals[i] = "MS";
60         } else {
61             vals[i] = "M";
62         }
63     }
64     vals[i] = NULL;
65 
66     /* g_strjoinv() obliges us to cast away const here */
67     return g_strjoinv(",", (char **)vals);
68 }
69 
70 target_ulong riscv_calc_kernel_start_addr(RISCVHartArrayState *harts,
71                                           target_ulong firmware_end_addr) {
72     if (riscv_is_32bit(harts)) {
73         return QEMU_ALIGN_UP(firmware_end_addr, 4 * MiB);
74     } else {
75         return QEMU_ALIGN_UP(firmware_end_addr, 2 * MiB);
76     }
77 }
78 
79 const char *riscv_default_firmware_name(RISCVHartArrayState *harts)
80 {
81     if (riscv_is_32bit(harts)) {
82         return RISCV32_BIOS_BIN;
83     }
84 
85     return RISCV64_BIOS_BIN;
86 }
87 
88 static char *riscv_find_bios(const char *bios_filename)
89 {
90     char *filename;
91 
92     filename = qemu_find_file(QEMU_FILE_TYPE_BIOS, bios_filename);
93     if (filename == NULL) {
94         if (!qtest_enabled()) {
95             /*
96              * We only ship OpenSBI binary bios images in the QEMU source.
97              * For machines that use images other than the default bios,
98              * running QEMU test will complain hence let's suppress the error
99              * report for QEMU testing.
100              */
101             error_report("Unable to find the RISC-V BIOS \"%s\"",
102                          bios_filename);
103             exit(1);
104         }
105     }
106 
107     return filename;
108 }
109 
110 char *riscv_find_firmware(const char *firmware_filename,
111                           const char *default_machine_firmware)
112 {
113     char *filename = NULL;
114 
115     if ((!firmware_filename) || (!strcmp(firmware_filename, "default"))) {
116         /*
117          * The user didn't specify -bios, or has specified "-bios default".
118          * That means we are going to load the OpenSBI binary included in
119          * the QEMU source.
120          */
121         filename = riscv_find_bios(default_machine_firmware);
122     } else if (strcmp(firmware_filename, "none")) {
123         filename = riscv_find_bios(firmware_filename);
124     }
125 
126     return filename;
127 }
128 
129 target_ulong riscv_find_and_load_firmware(MachineState *machine,
130                                           const char *default_machine_firmware,
131                                           hwaddr firmware_load_addr,
132                                           symbol_fn_t sym_cb)
133 {
134     char *firmware_filename;
135     target_ulong firmware_end_addr = firmware_load_addr;
136 
137     firmware_filename = riscv_find_firmware(machine->firmware,
138                                             default_machine_firmware);
139 
140     if (firmware_filename) {
141         /* If not "none" load the firmware */
142         firmware_end_addr = riscv_load_firmware(firmware_filename,
143                                                 firmware_load_addr, sym_cb);
144         g_free(firmware_filename);
145     }
146 
147     return firmware_end_addr;
148 }
149 
150 target_ulong riscv_load_firmware(const char *firmware_filename,
151                                  hwaddr firmware_load_addr,
152                                  symbol_fn_t sym_cb)
153 {
154     uint64_t firmware_entry, firmware_end;
155     ssize_t firmware_size;
156 
157     g_assert(firmware_filename != NULL);
158 
159     if (load_elf_ram_sym(firmware_filename, NULL, NULL, NULL,
160                          &firmware_entry, NULL, &firmware_end, NULL,
161                          0, EM_RISCV, 1, 0, NULL, true, sym_cb) > 0) {
162         return firmware_end;
163     }
164 
165     firmware_size = load_image_targphys_as(firmware_filename,
166                                            firmware_load_addr,
167                                            current_machine->ram_size, NULL);
168 
169     if (firmware_size > 0) {
170         return firmware_load_addr + firmware_size;
171     }
172 
173     error_report("could not load firmware '%s'", firmware_filename);
174     exit(1);
175 }
176 
177 static void riscv_load_initrd(MachineState *machine, uint64_t kernel_entry)
178 {
179     const char *filename = machine->initrd_filename;
180     uint64_t mem_size = machine->ram_size;
181     void *fdt = machine->fdt;
182     hwaddr start, end;
183     ssize_t size;
184 
185     g_assert(filename != NULL);
186 
187     /*
188      * We want to put the initrd far enough into RAM that when the
189      * kernel is uncompressed it will not clobber the initrd. However
190      * on boards without much RAM we must ensure that we still leave
191      * enough room for a decent sized initrd, and on boards with large
192      * amounts of RAM, we put the initrd at 512MB to allow large kernels
193      * to boot.
194      * So for boards with less than 1GB of RAM we put the initrd
195      * halfway into RAM, and for boards with 1GB of RAM or more we put
196      * the initrd at 512MB.
197      */
198     start = kernel_entry + MIN(mem_size / 2, 512 * MiB);
199 
200     size = load_ramdisk(filename, start, mem_size - start);
201     if (size == -1) {
202         size = load_image_targphys(filename, start, mem_size - start);
203         if (size == -1) {
204             error_report("could not load ramdisk '%s'", filename);
205             exit(1);
206         }
207     }
208 
209     /* Some RISC-V machines (e.g. opentitan) don't have a fdt. */
210     if (fdt) {
211         end = start + size;
212         qemu_fdt_setprop_u64(fdt, "/chosen", "linux,initrd-start", start);
213         qemu_fdt_setprop_u64(fdt, "/chosen", "linux,initrd-end", end);
214     }
215 }
216 
217 target_ulong riscv_load_kernel(MachineState *machine,
218                                RISCVHartArrayState *harts,
219                                target_ulong kernel_start_addr,
220                                bool load_initrd,
221                                symbol_fn_t sym_cb)
222 {
223     const char *kernel_filename = machine->kernel_filename;
224     uint64_t kernel_load_base, kernel_entry;
225     void *fdt = machine->fdt;
226 
227     g_assert(kernel_filename != NULL);
228 
229     /*
230      * NB: Use low address not ELF entry point to ensure that the fw_dynamic
231      * behaviour when loading an ELF matches the fw_payload, fw_jump and BBL
232      * behaviour, as well as fw_dynamic with a raw binary, all of which jump to
233      * the (expected) load address load address. This allows kernels to have
234      * separate SBI and ELF entry points (used by FreeBSD, for example).
235      */
236     if (load_elf_ram_sym(kernel_filename, NULL, NULL, NULL,
237                          NULL, &kernel_load_base, NULL, NULL, 0,
238                          EM_RISCV, 1, 0, NULL, true, sym_cb) > 0) {
239         kernel_entry = kernel_load_base;
240         goto out;
241     }
242 
243     if (load_uimage_as(kernel_filename, &kernel_entry, NULL, NULL,
244                        NULL, NULL, NULL) > 0) {
245         goto out;
246     }
247 
248     if (load_image_targphys_as(kernel_filename, kernel_start_addr,
249                                current_machine->ram_size, NULL) > 0) {
250         kernel_entry = kernel_start_addr;
251         goto out;
252     }
253 
254     error_report("could not load kernel '%s'", kernel_filename);
255     exit(1);
256 
257 out:
258     /*
259      * For 32 bit CPUs 'kernel_entry' can be sign-extended by
260      * load_elf_ram_sym().
261      */
262     if (riscv_is_32bit(harts)) {
263         kernel_entry = extract64(kernel_entry, 0, 32);
264     }
265 
266     if (load_initrd && machine->initrd_filename) {
267         riscv_load_initrd(machine, kernel_entry);
268     }
269 
270     if (fdt && machine->kernel_cmdline && *machine->kernel_cmdline) {
271         qemu_fdt_setprop_string(fdt, "/chosen", "bootargs",
272                                 machine->kernel_cmdline);
273     }
274 
275     return kernel_entry;
276 }
277 
278 /*
279  * This function makes an assumption that the DRAM interval
280  * 'dram_base' + 'dram_size' is contiguous.
281  *
282  * Considering that 'dram_end' is the lowest value between
283  * the end of the DRAM block and MachineState->ram_size, the
284  * FDT location will vary according to 'dram_base':
285  *
286  * - if 'dram_base' is less that 3072 MiB, the FDT will be
287  * put at the lowest value between 3072 MiB and 'dram_end';
288  *
289  * - if 'dram_base' is higher than 3072 MiB, the FDT will be
290  * put at 'dram_end'.
291  *
292  * The FDT is fdt_packed() during the calculation.
293  */
294 uint64_t riscv_compute_fdt_addr(hwaddr dram_base, hwaddr dram_size,
295                                 MachineState *ms)
296 {
297     int ret = fdt_pack(ms->fdt);
298     hwaddr dram_end, temp;
299     int fdtsize;
300 
301     /* Should only fail if we've built a corrupted tree */
302     g_assert(ret == 0);
303 
304     fdtsize = fdt_totalsize(ms->fdt);
305     if (fdtsize <= 0) {
306         error_report("invalid device-tree");
307         exit(1);
308     }
309 
310     /*
311      * A dram_size == 0, usually from a MemMapEntry[].size element,
312      * means that the DRAM block goes all the way to ms->ram_size.
313      */
314     dram_end = dram_base;
315     dram_end += dram_size ? MIN(ms->ram_size, dram_size) : ms->ram_size;
316 
317     /*
318      * We should put fdt as far as possible to avoid kernel/initrd overwriting
319      * its content. But it should be addressable by 32 bit system as well.
320      * Thus, put it at an 2MB aligned address that less than fdt size from the
321      * end of dram or 3GB whichever is lesser.
322      */
323     temp = (dram_base < 3072 * MiB) ? MIN(dram_end, 3072 * MiB) : dram_end;
324 
325     return QEMU_ALIGN_DOWN(temp - fdtsize, 2 * MiB);
326 }
327 
328 /*
329  * 'fdt_addr' is received as hwaddr because boards might put
330  * the FDT beyond 32-bit addressing boundary.
331  */
332 void riscv_load_fdt(hwaddr fdt_addr, void *fdt)
333 {
334     uint32_t fdtsize = fdt_totalsize(fdt);
335 
336     /* copy in the device tree */
337     qemu_fdt_dumpdtb(fdt, fdtsize);
338 
339     rom_add_blob_fixed_as("fdt", fdt, fdtsize, fdt_addr,
340                           &address_space_memory);
341     qemu_register_reset_nosnapshotload(qemu_fdt_randomize_seeds,
342                         rom_ptr_for_as(&address_space_memory, fdt_addr, fdtsize));
343 }
344 
345 void riscv_rom_copy_firmware_info(MachineState *machine, hwaddr rom_base,
346                                   hwaddr rom_size, uint32_t reset_vec_size,
347                                   uint64_t kernel_entry)
348 {
349     struct fw_dynamic_info dinfo;
350     size_t dinfo_len;
351 
352     if (sizeof(dinfo.magic) == 4) {
353         dinfo.magic = cpu_to_le32(FW_DYNAMIC_INFO_MAGIC_VALUE);
354         dinfo.version = cpu_to_le32(FW_DYNAMIC_INFO_VERSION);
355         dinfo.next_mode = cpu_to_le32(FW_DYNAMIC_INFO_NEXT_MODE_S);
356         dinfo.next_addr = cpu_to_le32(kernel_entry);
357     } else {
358         dinfo.magic = cpu_to_le64(FW_DYNAMIC_INFO_MAGIC_VALUE);
359         dinfo.version = cpu_to_le64(FW_DYNAMIC_INFO_VERSION);
360         dinfo.next_mode = cpu_to_le64(FW_DYNAMIC_INFO_NEXT_MODE_S);
361         dinfo.next_addr = cpu_to_le64(kernel_entry);
362     }
363     dinfo.options = 0;
364     dinfo.boot_hart = 0;
365     dinfo_len = sizeof(dinfo);
366 
367     /**
368      * copy the dynamic firmware info. This information is specific to
369      * OpenSBI but doesn't break any other firmware as long as they don't
370      * expect any certain value in "a2" register.
371      */
372     if (dinfo_len > (rom_size - reset_vec_size)) {
373         error_report("not enough space to store dynamic firmware info");
374         exit(1);
375     }
376 
377     rom_add_blob_fixed_as("mrom.finfo", &dinfo, dinfo_len,
378                            rom_base + reset_vec_size,
379                            &address_space_memory);
380 }
381 
382 void riscv_setup_rom_reset_vec(MachineState *machine, RISCVHartArrayState *harts,
383                                hwaddr start_addr,
384                                hwaddr rom_base, hwaddr rom_size,
385                                uint64_t kernel_entry,
386                                uint64_t fdt_load_addr)
387 {
388     int i;
389     uint32_t start_addr_hi32 = 0x00000000;
390     uint32_t fdt_load_addr_hi32 = 0x00000000;
391 
392     if (!riscv_is_32bit(harts)) {
393         start_addr_hi32 = start_addr >> 32;
394         fdt_load_addr_hi32 = fdt_load_addr >> 32;
395     }
396     /* reset vector */
397     uint32_t reset_vec[10] = {
398         0x00000297,                  /* 1:  auipc  t0, %pcrel_hi(fw_dyn) */
399         0x02828613,                  /*     addi   a2, t0, %pcrel_lo(1b) */
400         0xf1402573,                  /*     csrr   a0, mhartid  */
401         0,
402         0,
403         0x00028067,                  /*     jr     t0 */
404         start_addr,                  /* start: .dword */
405         start_addr_hi32,
406         fdt_load_addr,               /* fdt_laddr: .dword */
407         fdt_load_addr_hi32,
408                                      /* fw_dyn: */
409     };
410     if (riscv_is_32bit(harts)) {
411         reset_vec[3] = 0x0202a583;   /*     lw     a1, 32(t0) */
412         reset_vec[4] = 0x0182a283;   /*     lw     t0, 24(t0) */
413     } else {
414         reset_vec[3] = 0x0202b583;   /*     ld     a1, 32(t0) */
415         reset_vec[4] = 0x0182b283;   /*     ld     t0, 24(t0) */
416     }
417 
418     if (!harts->harts[0].cfg.ext_zicsr) {
419         /*
420          * The Zicsr extension has been disabled, so let's ensure we don't
421          * run the CSR instruction. Let's fill the address with a non
422          * compressed nop.
423          */
424         reset_vec[2] = 0x00000013;   /*     addi   x0, x0, 0 */
425     }
426 
427     /* copy in the reset vector in little_endian byte order */
428     for (i = 0; i < ARRAY_SIZE(reset_vec); i++) {
429         reset_vec[i] = cpu_to_le32(reset_vec[i]);
430     }
431     rom_add_blob_fixed_as("mrom.reset", reset_vec, sizeof(reset_vec),
432                           rom_base, &address_space_memory);
433     riscv_rom_copy_firmware_info(machine, rom_base, rom_size, sizeof(reset_vec),
434                                  kernel_entry);
435 }
436 
437 void riscv_setup_direct_kernel(hwaddr kernel_addr, hwaddr fdt_addr)
438 {
439     CPUState *cs;
440 
441     for (cs = first_cpu; cs; cs = CPU_NEXT(cs)) {
442         RISCVCPU *riscv_cpu = RISCV_CPU(cs);
443         riscv_cpu->env.kernel_addr = kernel_addr;
444         riscv_cpu->env.fdt_addr = fdt_addr;
445     }
446 }
447 
448 void riscv_setup_firmware_boot(MachineState *machine)
449 {
450     if (machine->kernel_filename) {
451         FWCfgState *fw_cfg;
452         fw_cfg = fw_cfg_find();
453 
454         assert(fw_cfg);
455         /*
456          * Expose the kernel, the command line, and the initrd in fw_cfg.
457          * We don't process them here at all, it's all left to the
458          * firmware.
459          */
460         load_image_to_fw_cfg(fw_cfg,
461                              FW_CFG_KERNEL_SIZE, FW_CFG_KERNEL_DATA,
462                              machine->kernel_filename,
463                              true);
464         load_image_to_fw_cfg(fw_cfg,
465                              FW_CFG_INITRD_SIZE, FW_CFG_INITRD_DATA,
466                              machine->initrd_filename, false);
467 
468         if (machine->kernel_cmdline) {
469             fw_cfg_add_i32(fw_cfg, FW_CFG_CMDLINE_SIZE,
470                            strlen(machine->kernel_cmdline) + 1);
471             fw_cfg_add_string(fw_cfg, FW_CFG_CMDLINE_DATA,
472                               machine->kernel_cmdline);
473         }
474     }
475 }
476