xref: /openbmc/qemu/hw/riscv/boot.c (revision 5b5968c4)
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     return harts->harts[0].env.misa_mxl_max == MXL_RV32;
40 }
41 
42 /*
43  * Return the per-socket PLIC hart topology configuration string
44  * (caller must free with g_free())
45  */
46 char *riscv_plic_hart_config_string(int hart_count)
47 {
48     g_autofree const char **vals = g_new(const char *, hart_count + 1);
49     int i;
50 
51     for (i = 0; i < hart_count; i++) {
52         CPUState *cs = qemu_get_cpu(i);
53         CPURISCVState *env = &RISCV_CPU(cs)->env;
54 
55         if (kvm_enabled()) {
56             vals[i] = "S";
57         } else if (riscv_has_ext(env, RVS)) {
58             vals[i] = "MS";
59         } else {
60             vals[i] = "M";
61         }
62     }
63     vals[i] = NULL;
64 
65     /* g_strjoinv() obliges us to cast away const here */
66     return g_strjoinv(",", (char **)vals);
67 }
68 
69 target_ulong riscv_calc_kernel_start_addr(RISCVHartArrayState *harts,
70                                           target_ulong firmware_end_addr) {
71     if (riscv_is_32bit(harts)) {
72         return QEMU_ALIGN_UP(firmware_end_addr, 4 * MiB);
73     } else {
74         return QEMU_ALIGN_UP(firmware_end_addr, 2 * MiB);
75     }
76 }
77 
78 const char *riscv_default_firmware_name(RISCVHartArrayState *harts)
79 {
80     if (riscv_is_32bit(harts)) {
81         return RISCV32_BIOS_BIN;
82     }
83 
84     return RISCV64_BIOS_BIN;
85 }
86 
87 static char *riscv_find_bios(const char *bios_filename)
88 {
89     char *filename;
90 
91     filename = qemu_find_file(QEMU_FILE_TYPE_BIOS, bios_filename);
92     if (filename == NULL) {
93         if (!qtest_enabled()) {
94             /*
95              * We only ship OpenSBI binary bios images in the QEMU source.
96              * For machines that use images other than the default bios,
97              * running QEMU test will complain hence let's suppress the error
98              * report for QEMU testing.
99              */
100             error_report("Unable to find the RISC-V BIOS \"%s\"",
101                          bios_filename);
102             exit(1);
103         }
104     }
105 
106     return filename;
107 }
108 
109 char *riscv_find_firmware(const char *firmware_filename,
110                           const char *default_machine_firmware)
111 {
112     char *filename = NULL;
113 
114     if ((!firmware_filename) || (!strcmp(firmware_filename, "default"))) {
115         /*
116          * The user didn't specify -bios, or has specified "-bios default".
117          * That means we are going to load the OpenSBI binary included in
118          * the QEMU source.
119          */
120         filename = riscv_find_bios(default_machine_firmware);
121     } else if (strcmp(firmware_filename, "none")) {
122         filename = riscv_find_bios(firmware_filename);
123     }
124 
125     return filename;
126 }
127 
128 target_ulong riscv_find_and_load_firmware(MachineState *machine,
129                                           const char *default_machine_firmware,
130                                           hwaddr firmware_load_addr,
131                                           symbol_fn_t sym_cb)
132 {
133     char *firmware_filename;
134     target_ulong firmware_end_addr = firmware_load_addr;
135 
136     firmware_filename = riscv_find_firmware(machine->firmware,
137                                             default_machine_firmware);
138 
139     if (firmware_filename) {
140         /* If not "none" load the firmware */
141         firmware_end_addr = riscv_load_firmware(firmware_filename,
142                                                 firmware_load_addr, sym_cb);
143         g_free(firmware_filename);
144     }
145 
146     return firmware_end_addr;
147 }
148 
149 target_ulong riscv_load_firmware(const char *firmware_filename,
150                                  hwaddr firmware_load_addr,
151                                  symbol_fn_t sym_cb)
152 {
153     uint64_t firmware_entry, firmware_end;
154     ssize_t firmware_size;
155 
156     g_assert(firmware_filename != NULL);
157 
158     if (load_elf_ram_sym(firmware_filename, NULL, NULL, NULL,
159                          &firmware_entry, NULL, &firmware_end, NULL,
160                          0, EM_RISCV, 1, 0, NULL, true, sym_cb) > 0) {
161         return firmware_end;
162     }
163 
164     firmware_size = load_image_targphys_as(firmware_filename,
165                                            firmware_load_addr,
166                                            current_machine->ram_size, NULL);
167 
168     if (firmware_size > 0) {
169         return firmware_load_addr + firmware_size;
170     }
171 
172     error_report("could not load firmware '%s'", firmware_filename);
173     exit(1);
174 }
175 
176 static void riscv_load_initrd(MachineState *machine, uint64_t kernel_entry)
177 {
178     const char *filename = machine->initrd_filename;
179     uint64_t mem_size = machine->ram_size;
180     void *fdt = machine->fdt;
181     hwaddr start, end;
182     ssize_t size;
183 
184     g_assert(filename != NULL);
185 
186     /*
187      * We want to put the initrd far enough into RAM that when the
188      * kernel is uncompressed it will not clobber the initrd. However
189      * on boards without much RAM we must ensure that we still leave
190      * enough room for a decent sized initrd, and on boards with large
191      * amounts of RAM we must avoid the initrd being so far up in RAM
192      * that it is outside lowmem and inaccessible to the kernel.
193      * So for boards with less  than 256MB of RAM we put the initrd
194      * halfway into RAM, and for boards with 256MB of RAM or more we put
195      * the initrd at 128MB.
196      */
197     start = kernel_entry + MIN(mem_size / 2, 128 * MiB);
198 
199     size = load_ramdisk(filename, start, mem_size - start);
200     if (size == -1) {
201         size = load_image_targphys(filename, start, mem_size - start);
202         if (size == -1) {
203             error_report("could not load ramdisk '%s'", filename);
204             exit(1);
205         }
206     }
207 
208     /* Some RISC-V machines (e.g. opentitan) don't have a fdt. */
209     if (fdt) {
210         end = start + size;
211         qemu_fdt_setprop_cell(fdt, "/chosen", "linux,initrd-start", start);
212         qemu_fdt_setprop_cell(fdt, "/chosen", "linux,initrd-end", end);
213     }
214 }
215 
216 target_ulong riscv_load_kernel(MachineState *machine,
217                                RISCVHartArrayState *harts,
218                                target_ulong kernel_start_addr,
219                                bool load_initrd,
220                                symbol_fn_t sym_cb)
221 {
222     const char *kernel_filename = machine->kernel_filename;
223     uint64_t kernel_load_base, kernel_entry;
224     void *fdt = machine->fdt;
225 
226     g_assert(kernel_filename != NULL);
227 
228     /*
229      * NB: Use low address not ELF entry point to ensure that the fw_dynamic
230      * behaviour when loading an ELF matches the fw_payload, fw_jump and BBL
231      * behaviour, as well as fw_dynamic with a raw binary, all of which jump to
232      * the (expected) load address load address. This allows kernels to have
233      * separate SBI and ELF entry points (used by FreeBSD, for example).
234      */
235     if (load_elf_ram_sym(kernel_filename, NULL, NULL, NULL,
236                          NULL, &kernel_load_base, NULL, NULL, 0,
237                          EM_RISCV, 1, 0, NULL, true, sym_cb) > 0) {
238         kernel_entry = kernel_load_base;
239         goto out;
240     }
241 
242     if (load_uimage_as(kernel_filename, &kernel_entry, NULL, NULL,
243                        NULL, NULL, NULL) > 0) {
244         goto out;
245     }
246 
247     if (load_image_targphys_as(kernel_filename, kernel_start_addr,
248                                current_machine->ram_size, NULL) > 0) {
249         kernel_entry = kernel_start_addr;
250         goto out;
251     }
252 
253     error_report("could not load kernel '%s'", kernel_filename);
254     exit(1);
255 
256 out:
257     /*
258      * For 32 bit CPUs 'kernel_entry' can be sign-extended by
259      * load_elf_ram_sym().
260      */
261     if (riscv_is_32bit(harts)) {
262         kernel_entry = extract64(kernel_entry, 0, 32);
263     }
264 
265     if (load_initrd && machine->initrd_filename) {
266         riscv_load_initrd(machine, kernel_entry);
267     }
268 
269     if (fdt && machine->kernel_cmdline && *machine->kernel_cmdline) {
270         qemu_fdt_setprop_string(fdt, "/chosen", "bootargs",
271                                 machine->kernel_cmdline);
272     }
273 
274     return kernel_entry;
275 }
276 
277 /*
278  * This function makes an assumption that the DRAM interval
279  * 'dram_base' + 'dram_size' is contiguous.
280  *
281  * Considering that 'dram_end' is the lowest value between
282  * the end of the DRAM block and MachineState->ram_size, the
283  * FDT location will vary according to 'dram_base':
284  *
285  * - if 'dram_base' is less that 3072 MiB, the FDT will be
286  * put at the lowest value between 3072 MiB and 'dram_end';
287  *
288  * - if 'dram_base' is higher than 3072 MiB, the FDT will be
289  * put at 'dram_end'.
290  *
291  * The FDT is fdt_packed() during the calculation.
292  */
293 uint64_t riscv_compute_fdt_addr(hwaddr dram_base, hwaddr dram_size,
294                                 MachineState *ms)
295 {
296     int ret = fdt_pack(ms->fdt);
297     hwaddr dram_end, temp;
298     int fdtsize;
299 
300     /* Should only fail if we've built a corrupted tree */
301     g_assert(ret == 0);
302 
303     fdtsize = fdt_totalsize(ms->fdt);
304     if (fdtsize <= 0) {
305         error_report("invalid device-tree");
306         exit(1);
307     }
308 
309     /*
310      * A dram_size == 0, usually from a MemMapEntry[].size element,
311      * means that the DRAM block goes all the way to ms->ram_size.
312      */
313     dram_end = dram_base;
314     dram_end += dram_size ? MIN(ms->ram_size, dram_size) : ms->ram_size;
315 
316     /*
317      * We should put fdt as far as possible to avoid kernel/initrd overwriting
318      * its content. But it should be addressable by 32 bit system as well.
319      * Thus, put it at an 2MB aligned address that less than fdt size from the
320      * end of dram or 3GB whichever is lesser.
321      */
322     temp = (dram_base < 3072 * MiB) ? MIN(dram_end, 3072 * MiB) : dram_end;
323 
324     return QEMU_ALIGN_DOWN(temp - fdtsize, 2 * MiB);
325 }
326 
327 /*
328  * 'fdt_addr' is received as hwaddr because boards might put
329  * the FDT beyond 32-bit addressing boundary.
330  */
331 void riscv_load_fdt(hwaddr fdt_addr, void *fdt)
332 {
333     uint32_t fdtsize = fdt_totalsize(fdt);
334 
335     /* copy in the device tree */
336     qemu_fdt_dumpdtb(fdt, fdtsize);
337 
338     rom_add_blob_fixed_as("fdt", fdt, fdtsize, fdt_addr,
339                           &address_space_memory);
340     qemu_register_reset_nosnapshotload(qemu_fdt_randomize_seeds,
341                         rom_ptr_for_as(&address_space_memory, fdt_addr, fdtsize));
342 }
343 
344 void riscv_rom_copy_firmware_info(MachineState *machine, hwaddr rom_base,
345                                   hwaddr rom_size, uint32_t reset_vec_size,
346                                   uint64_t kernel_entry)
347 {
348     struct fw_dynamic_info dinfo;
349     size_t dinfo_len;
350 
351     if (sizeof(dinfo.magic) == 4) {
352         dinfo.magic = cpu_to_le32(FW_DYNAMIC_INFO_MAGIC_VALUE);
353         dinfo.version = cpu_to_le32(FW_DYNAMIC_INFO_VERSION);
354         dinfo.next_mode = cpu_to_le32(FW_DYNAMIC_INFO_NEXT_MODE_S);
355         dinfo.next_addr = cpu_to_le32(kernel_entry);
356     } else {
357         dinfo.magic = cpu_to_le64(FW_DYNAMIC_INFO_MAGIC_VALUE);
358         dinfo.version = cpu_to_le64(FW_DYNAMIC_INFO_VERSION);
359         dinfo.next_mode = cpu_to_le64(FW_DYNAMIC_INFO_NEXT_MODE_S);
360         dinfo.next_addr = cpu_to_le64(kernel_entry);
361     }
362     dinfo.options = 0;
363     dinfo.boot_hart = 0;
364     dinfo_len = sizeof(dinfo);
365 
366     /**
367      * copy the dynamic firmware info. This information is specific to
368      * OpenSBI but doesn't break any other firmware as long as they don't
369      * expect any certain value in "a2" register.
370      */
371     if (dinfo_len > (rom_size - reset_vec_size)) {
372         error_report("not enough space to store dynamic firmware info");
373         exit(1);
374     }
375 
376     rom_add_blob_fixed_as("mrom.finfo", &dinfo, dinfo_len,
377                            rom_base + reset_vec_size,
378                            &address_space_memory);
379 }
380 
381 void riscv_setup_rom_reset_vec(MachineState *machine, RISCVHartArrayState *harts,
382                                hwaddr start_addr,
383                                hwaddr rom_base, hwaddr rom_size,
384                                uint64_t kernel_entry,
385                                uint64_t fdt_load_addr)
386 {
387     int i;
388     uint32_t start_addr_hi32 = 0x00000000;
389     uint32_t fdt_load_addr_hi32 = 0x00000000;
390 
391     if (!riscv_is_32bit(harts)) {
392         start_addr_hi32 = start_addr >> 32;
393         fdt_load_addr_hi32 = fdt_load_addr >> 32;
394     }
395     /* reset vector */
396     uint32_t reset_vec[10] = {
397         0x00000297,                  /* 1:  auipc  t0, %pcrel_hi(fw_dyn) */
398         0x02828613,                  /*     addi   a2, t0, %pcrel_lo(1b) */
399         0xf1402573,                  /*     csrr   a0, mhartid  */
400         0,
401         0,
402         0x00028067,                  /*     jr     t0 */
403         start_addr,                  /* start: .dword */
404         start_addr_hi32,
405         fdt_load_addr,               /* fdt_laddr: .dword */
406         fdt_load_addr_hi32,
407                                      /* fw_dyn: */
408     };
409     if (riscv_is_32bit(harts)) {
410         reset_vec[3] = 0x0202a583;   /*     lw     a1, 32(t0) */
411         reset_vec[4] = 0x0182a283;   /*     lw     t0, 24(t0) */
412     } else {
413         reset_vec[3] = 0x0202b583;   /*     ld     a1, 32(t0) */
414         reset_vec[4] = 0x0182b283;   /*     ld     t0, 24(t0) */
415     }
416 
417     if (!harts->harts[0].cfg.ext_icsr) {
418         /*
419          * The Zicsr extension has been disabled, so let's ensure we don't
420          * run the CSR instruction. Let's fill the address with a non
421          * compressed nop.
422          */
423         reset_vec[2] = 0x00000013;   /*     addi   x0, x0, 0 */
424     }
425 
426     /* copy in the reset vector in little_endian byte order */
427     for (i = 0; i < ARRAY_SIZE(reset_vec); i++) {
428         reset_vec[i] = cpu_to_le32(reset_vec[i]);
429     }
430     rom_add_blob_fixed_as("mrom.reset", reset_vec, sizeof(reset_vec),
431                           rom_base, &address_space_memory);
432     riscv_rom_copy_firmware_info(machine, rom_base, rom_size, sizeof(reset_vec),
433                                  kernel_entry);
434 }
435 
436 void riscv_setup_direct_kernel(hwaddr kernel_addr, hwaddr fdt_addr)
437 {
438     CPUState *cs;
439 
440     for (cs = first_cpu; cs; cs = CPU_NEXT(cs)) {
441         RISCVCPU *riscv_cpu = RISCV_CPU(cs);
442         riscv_cpu->env.kernel_addr = kernel_addr;
443         riscv_cpu->env.fdt_addr = fdt_addr;
444     }
445 }
446 
447 void riscv_setup_firmware_boot(MachineState *machine)
448 {
449     if (machine->kernel_filename) {
450         FWCfgState *fw_cfg;
451         fw_cfg = fw_cfg_find();
452 
453         assert(fw_cfg);
454         /*
455          * Expose the kernel, the command line, and the initrd in fw_cfg.
456          * We don't process them here at all, it's all left to the
457          * firmware.
458          */
459         load_image_to_fw_cfg(fw_cfg,
460                              FW_CFG_KERNEL_SIZE, FW_CFG_KERNEL_DATA,
461                              machine->kernel_filename,
462                              true);
463         load_image_to_fw_cfg(fw_cfg,
464                              FW_CFG_INITRD_SIZE, FW_CFG_INITRD_DATA,
465                              machine->initrd_filename, false);
466 
467         if (machine->kernel_cmdline) {
468             fw_cfg_add_i32(fw_cfg, FW_CFG_CMDLINE_SIZE,
469                            strlen(machine->kernel_cmdline) + 1);
470             fw_cfg_add_string(fw_cfg, FW_CFG_CMDLINE_DATA,
471                               machine->kernel_cmdline);
472         }
473     }
474 }
475