BOOT_CODE bool_t
init_sys_state(
cpu_id_t cpu_id,
mem_p_regs_t mem_p_regs,
dev_p_regs_t* dev_p_regs,
ui_info_t ui_info,
p_region_t boot_mem_reuse_p_reg,
/* parameters below not modeled in abstract specification */
uint32_t num_drhu,
paddr_t* drhu_list,
acpi_rmrr_list_t *rmrr_list
)
{
cap_t root_cnode_cap;
vptr_t bi_frame_vptr;
vptr_t ipcbuf_vptr;
cap_t it_vspace_cap;
cap_t it_ap_cap;
cap_t ipcbuf_cap;
pptr_t bi_frame_pptr;
create_frames_of_region_ret_t create_frames_ret;
#ifdef CONFIG_ENABLE_BENCHMARKS
vm_attributes_t buffer_attr = {{ 0 }};
word_t paddr;
pde_t pde;
#endif /* CONFIG_ENABLE_BENCHMARKS */
/* convert from physical addresses to kernel pptrs */
region_t ui_reg = paddr_to_pptr_reg(ui_info.p_reg);
region_t boot_mem_reuse_reg = paddr_to_pptr_reg(boot_mem_reuse_p_reg);
/* convert from physical addresses to userland vptrs */
v_region_t ui_v_reg;
v_region_t it_v_reg;
ui_v_reg.start = ui_info.p_reg.start - ui_info.pv_offset;
ui_v_reg.end = ui_info.p_reg.end - ui_info.pv_offset;
ipcbuf_vptr = ui_v_reg.end;
bi_frame_vptr = ipcbuf_vptr + BIT(PAGE_BITS);
/* The region of the initial thread is the user image + ipcbuf and boot info */
it_v_reg.start = ui_v_reg.start;
it_v_reg.end = bi_frame_vptr + BIT(PAGE_BITS);
init_freemem(ui_info.p_reg, mem_p_regs);
/* initialise virtual-memory-related data structures (not in abstract spec) */
if (!init_vm_state()) {
return false;
}
#ifdef CONFIG_ENABLE_BENCHMARKS
/* allocate and create the log buffer */
buffer_attr.words[0] = IA32_PAT_MT_WRITE_THROUGH;
paddr = pptr_to_paddr((void *) alloc_region(pageBitsForSize(X86_LargePage)));
/* allocate a large frame for logging */
pde = x86_make_pde_mapping(paddr, buffer_attr);
ia32KSGlobalPD[IA32_KSLOG_IDX] = pde;
/* flush the tlb */
invalidateTranslationAll();
/* if we crash here, the log isn't working */
#ifdef CONFIG_DEBUG_BUILD
#if CONFIG_MAX_NUM_TRACE_POINTS > 0
printf("Testing log\n");
ksLog[0].data = 0xdeadbeef;
printf("Wrote to ksLog %x\n", ksLog[0].data);
assert(ksLog[0].data == 0xdeadbeef);
#endif /* CONFIG_MAX_NUM_TRACE_POINTS */
#endif /* CONFIG_DEBUG_BUILD */
#endif /* CONFIG_ENABLE_BENCHMARKS */
/* create the root cnode */
root_cnode_cap = create_root_cnode();
/* create the IO port cap */
write_slot(
SLOT_PTR(pptr_of_cap(root_cnode_cap), seL4_CapIOPort),
cap_io_port_cap_new(
0, /* first port */
NUM_IO_PORTS - 1 /* last port */
)
);
/* create the cap for managing thread domains */
create_domain_cap(root_cnode_cap);
/* create the IRQ CNode */
if (!create_irq_cnode()) {
return false;
}
/* initialise the IRQ states and provide the IRQ control cap */
init_irqs(root_cnode_cap);
/* create the bootinfo frame */
bi_frame_pptr = allocate_bi_frame(0, 1, ipcbuf_vptr);
if (!bi_frame_pptr) {
return false;
}
/* Construct an initial address space with enough virtual addresses
* to cover the user image + ipc buffer and bootinfo frames */
it_vspace_cap = create_it_address_space(root_cnode_cap, it_v_reg);
if (cap_get_capType(it_vspace_cap) == cap_null_cap) {
return false;
}
/* Create and map bootinfo frame cap */
create_bi_frame_cap(
root_cnode_cap,
it_vspace_cap,
bi_frame_pptr,
bi_frame_vptr
);
/* create the initial thread's IPC buffer */
ipcbuf_cap = create_ipcbuf_frame(root_cnode_cap, it_vspace_cap, ipcbuf_vptr);
if (cap_get_capType(ipcbuf_cap) == cap_null_cap) {
return false;
}
/* create all userland image frames */
create_frames_ret =
create_frames_of_region(
root_cnode_cap,
it_vspace_cap,
ui_reg,
true,
ui_info.pv_offset
);
if (!create_frames_ret.success) {
return false;
}
ndks_boot.bi_frame->userImageFrames = create_frames_ret.region;
/* create the initial thread's ASID pool */
it_ap_cap = create_it_asid_pool(root_cnode_cap);
if (cap_get_capType(it_ap_cap) == cap_null_cap) {
return false;
}
write_it_asid_pool(it_ap_cap, it_vspace_cap);
/*
* Initialise the NULL FPU state. This is different from merely zero'ing it
* out (i.e., the NULL FPU state is non-zero), and must be performed before
* the first thread is created.
*/
resetFpu();
saveFpuState(&x86KSnullFpuState);
x86KSfpuOwner = NULL;
/* create the idle thread */
if (!create_idle_thread()) {
return false;
}
/* create the initial thread */
if (!create_initial_thread(
root_cnode_cap,
it_vspace_cap,
ui_info.v_entry,
bi_frame_vptr,
ipcbuf_vptr,
ipcbuf_cap
)) {
return false;
}
if (config_set(CONFIG_IOMMU)) {
/* initialise VTD-related data structures and the IOMMUs */
if (!vtd_init(cpu_id, num_drhu, rmrr_list)) {
return false;
}
/* write number of IOMMU PT levels into bootinfo */
ndks_boot.bi_frame->numIOPTLevels = x86KSnumIOPTLevels;
/* write IOSpace master cap */
write_slot(SLOT_PTR(pptr_of_cap(root_cnode_cap), seL4_CapIOSpace), master_iospace_cap());
} else {
ndks_boot.bi_frame->numIOPTLevels = -1;
}
/* convert the remaining free memory into UT objects and provide the caps */
if (!create_untypeds(root_cnode_cap, boot_mem_reuse_reg)) {
return false;
}
/* WARNING: alloc_region() must not be called anymore after here! */
/* create device frames */
if (!create_device_frames(root_cnode_cap, dev_p_regs)) {
return false;
}
/* finalise the bootinfo frame */
bi_finalise();
return true;
}
BOOT_CODE bool_t
init_sys_state(
cpu_id_t cpu_id,
mem_p_regs_t mem_p_regs,
ui_info_t ui_info,
p_region_t boot_mem_reuse_p_reg,
/* parameters below not modeled in abstract specification */
uint32_t num_drhu,
paddr_t* drhu_list,
acpi_rmrr_list_t *rmrr_list,
seL4_X86_BootInfo_VBE *vbe
)
{
cap_t root_cnode_cap;
vptr_t extra_bi_frame_vptr;
vptr_t bi_frame_vptr;
vptr_t ipcbuf_vptr;
cap_t it_vspace_cap;
cap_t it_ap_cap;
cap_t ipcbuf_cap;
pptr_t bi_frame_pptr;
word_t extra_bi_size = sizeof(seL4_BootInfoHeader);
region_t extra_bi_region;
pptr_t extra_bi_offset = 0;
create_frames_of_region_ret_t create_frames_ret;
create_frames_of_region_ret_t extra_bi_ret;
/* convert from physical addresses to kernel pptrs */
region_t ui_reg = paddr_to_pptr_reg(ui_info.p_reg);
region_t boot_mem_reuse_reg = paddr_to_pptr_reg(boot_mem_reuse_p_reg);
/* convert from physical addresses to userland vptrs */
v_region_t ui_v_reg;
v_region_t it_v_reg;
ui_v_reg.start = ui_info.p_reg.start - ui_info.pv_offset;
ui_v_reg.end = ui_info.p_reg.end - ui_info.pv_offset;
ipcbuf_vptr = ui_v_reg.end;
bi_frame_vptr = ipcbuf_vptr + BIT(PAGE_BITS);
extra_bi_frame_vptr = bi_frame_vptr + BIT(PAGE_BITS);
if (vbe->vbeMode != -1) {
extra_bi_size += sizeof(seL4_X86_BootInfo_VBE);
}
/* The region of the initial thread is the user image + ipcbuf and boot info */
it_v_reg.start = ui_v_reg.start;
it_v_reg.end = ROUND_UP(extra_bi_frame_vptr + extra_bi_size, PAGE_BITS);
init_freemem(ui_info.p_reg, mem_p_regs);
/* create the root cnode */
root_cnode_cap = create_root_cnode();
/* create the IO port cap */
write_slot(
SLOT_PTR(pptr_of_cap(root_cnode_cap), seL4_CapIOPort),
cap_io_port_cap_new(
0, /* first port */
NUM_IO_PORTS - 1, /* last port */
VPID_INVALID
)
);
/* create the cap for managing thread domains */
create_domain_cap(root_cnode_cap);
/* create the IRQ CNode */
if (!create_irq_cnode()) {
return false;
}
/* initialise the IRQ states and provide the IRQ control cap */
init_irqs(root_cnode_cap);
/* create the bootinfo frame */
bi_frame_pptr = allocate_bi_frame(0, ksNumCPUs, ipcbuf_vptr);
if (!bi_frame_pptr) {
return false;
}
extra_bi_region = allocate_extra_bi_region(extra_bi_size);
if (extra_bi_region.start == 0) {
return false;
}
/* populate vbe info block */
if (vbe->vbeMode != -1) {
vbe->header.id = SEL4_BOOTINFO_HEADER_X86_VBE;
vbe->header.len = sizeof(seL4_X86_BootInfo_VBE);
memcpy((void*)(extra_bi_region.start + extra_bi_offset), vbe, sizeof(seL4_X86_BootInfo_VBE));
extra_bi_offset += sizeof(seL4_X86_BootInfo_VBE);
}
/* provde a chunk for any leftover padding in the extended boot info */
seL4_BootInfoHeader padding_header;
padding_header.id = SEL4_BOOTINFO_HEADER_PADDING;
padding_header.len = (extra_bi_region.end - extra_bi_region.start) - extra_bi_offset;
memcpy((void*)(extra_bi_region.start + extra_bi_offset), &padding_header, sizeof(seL4_BootInfoHeader));
/* Construct an initial address space with enough virtual addresses
* to cover the user image + ipc buffer and bootinfo frames */
it_vspace_cap = create_it_address_space(root_cnode_cap, it_v_reg);
if (cap_get_capType(it_vspace_cap) == cap_null_cap) {
return false;
}
/* Create and map bootinfo frame cap */
create_bi_frame_cap(
root_cnode_cap,
it_vspace_cap,
bi_frame_pptr,
bi_frame_vptr
);
/* create and map extra bootinfo region */
extra_bi_ret =
create_frames_of_region(
root_cnode_cap,
it_vspace_cap,
extra_bi_region,
true,
pptr_to_paddr((void*)(extra_bi_region.start - extra_bi_frame_vptr))
);
if (!extra_bi_ret.success) {
return false;
}
ndks_boot.bi_frame->extraBIPages = extra_bi_ret.region;
/* create the initial thread's IPC buffer */
ipcbuf_cap = create_ipcbuf_frame(root_cnode_cap, it_vspace_cap, ipcbuf_vptr);
if (cap_get_capType(ipcbuf_cap) == cap_null_cap) {
return false;
}
/* create all userland image frames */
create_frames_ret =
create_frames_of_region(
root_cnode_cap,
it_vspace_cap,
ui_reg,
true,
ui_info.pv_offset
);
if (!create_frames_ret.success) {
return false;
}
ndks_boot.bi_frame->userImageFrames = create_frames_ret.region;
/* create the initial thread's ASID pool */
it_ap_cap = create_it_asid_pool(root_cnode_cap);
if (cap_get_capType(it_ap_cap) == cap_null_cap) {
return false;
}
write_it_asid_pool(it_ap_cap, it_vspace_cap);
ndks_boot.bi_frame->archInfo = tsc_init();
/* create the idle thread */
if (!create_idle_thread()) {
return false;
}
/* create the initial thread */
if (!create_initial_thread(
root_cnode_cap,
it_vspace_cap,
ui_info.v_entry,
bi_frame_vptr,
ipcbuf_vptr,
ipcbuf_cap
)) {
return false;
}
if (config_set(CONFIG_IOMMU)) {
/* initialise VTD-related data structures and the IOMMUs */
if (!vtd_init(cpu_id, num_drhu, rmrr_list)) {
return false;
}
/* write number of IOMMU PT levels into bootinfo */
ndks_boot.bi_frame->numIOPTLevels = x86KSnumIOPTLevels;
/* write IOSpace master cap */
write_slot(SLOT_PTR(pptr_of_cap(root_cnode_cap), seL4_CapIOSpace), master_iospace_cap());
} else {
ndks_boot.bi_frame->numIOPTLevels = -1;
}
/* create all of the untypeds. Both devices and kernel window memory */
if (!create_untypeds(root_cnode_cap, boot_mem_reuse_reg)) {
return false;
}
/* WARNING: alloc_region() must not be called anymore after here! */
/* finalise the bootinfo frame */
bi_finalise();
return true;
}
int
main(int argc, char *argv[], char *envp[])
{
#ifdef LISP_FEATURE_WIN32
/* Exception handling support structure. Evil Win32 hack. */
struct lisp_exception_frame exception_frame;
#endif
/* the name of the core file we're to execute. Note that this is
* a malloc'ed string which should be freed eventually. */
char *core = 0;
char **sbcl_argv = 0;
os_vm_offset_t embedded_core_offset = 0;
char *runtime_path = 0;
/* other command line options */
boolean noinform = 0;
boolean end_runtime_options = 0;
boolean disable_lossage_handler_p = 0;
lispobj initial_function;
const char *sbcl_home = getenv("SBCL_HOME");
interrupt_init();
block_blockable_signals(0, 0);
setlocale(LC_ALL, "");
runtime_options = NULL;
/* Check early to see if this executable has an embedded core,
* which also populates runtime_options if the core has runtime
* options */
runtime_path = os_get_runtime_executable_path();
if (runtime_path) {
os_vm_offset_t offset = search_for_embedded_core(runtime_path);
if (offset != -1) {
embedded_core_offset = offset;
core = runtime_path;
} else {
free(runtime_path);
}
}
/* Parse our part of the command line (aka "runtime options"),
* stripping out those options that we handle. */
if (runtime_options != NULL) {
dynamic_space_size = runtime_options->dynamic_space_size;
thread_control_stack_size = runtime_options->thread_control_stack_size;
sbcl_argv = argv;
} else {
int argi = 1;
runtime_options = successful_malloc(sizeof(struct runtime_options));
while (argi < argc) {
char *arg = argv[argi];
if (0 == strcmp(arg, "--script")) {
/* This is both a runtime and a toplevel option. As a
* runtime option, it is equivalent to --noinform.
* This exits, and does not increment argi, so that
* TOPLEVEL-INIT sees the option. */
noinform = 1;
end_runtime_options = 1;
disable_lossage_handler_p = 1;
lose_on_corruption_p = 1;
break;
} else if (0 == strcmp(arg, "--noinform")) {
noinform = 1;
++argi;
} else if (0 == strcmp(arg, "--core")) {
if (core) {
lose("more than one core file specified\n");
} else {
++argi;
if (argi >= argc) {
lose("missing filename for --core argument\n");
}
core = copied_string(argv[argi]);
++argi;
}
} else if (0 == strcmp(arg, "--help")) {
/* I think this is the (or a) usual convention: upon
* seeing "--help" we immediately print our help
* string and exit, ignoring everything else. */
print_help();
exit(0);
} else if (0 == strcmp(arg, "--version")) {
/* As in "--help" case, I think this is expected. */
print_version();
exit(0);
} else if (0 == strcmp(arg, "--dynamic-space-size")) {
++argi;
if (argi >= argc)
lose("missing argument for --dynamic-space-size");
errno = 0;
dynamic_space_size = strtol(argv[argi++], 0, 0) << 20;
if (errno)
lose("argument to --dynamic-space-size is not a number");
# ifdef MAX_DYNAMIC_SPACE_END
if (!((DYNAMIC_SPACE_START <
DYNAMIC_SPACE_START+dynamic_space_size) &&
(DYNAMIC_SPACE_START+dynamic_space_size <=
MAX_DYNAMIC_SPACE_END)))
lose("specified --dynamic-space-size too large");
# endif
} else if (0 == strcmp(arg, "--control-stack-size")) {
++argi;
if (argi >= argc)
lose("missing argument for --control-stack-size");
errno = 0;
thread_control_stack_size = strtol(argv[argi++], 0, 0) << 20;
if (errno)
lose("argument to --control-stack-size is not a number");
} else if (0 == strcmp(arg, "--debug-environment")) {
int n = 0;
printf("; Commandline arguments:\n");
while (n < argc) {
printf("; %2d: \"%s\"\n", n, argv[n]);
++n;
}
n = 0;
printf(";\n; Environment:\n");
while (ENVIRON[n]) {
printf("; %2d: \"%s\"\n", n, ENVIRON[n]);
++n;
}
++argi;
} else if (0 == strcmp(arg, "--disable-ldb")) {
disable_lossage_handler_p = 1;
++argi;
} else if (0 == strcmp(arg, "--lose-on-corruption")) {
lose_on_corruption_p = 1;
++argi;
} else if (0 == strcmp(arg, "--end-runtime-options")) {
end_runtime_options = 1;
++argi;
break;
} else {
/* This option was unrecognized as a runtime option,
* so it must be a toplevel option or a user option,
* so we must be past the end of the runtime option
* section. */
break;
}
}
/* This is where we strip out those options that we handle. We
* also take this opportunity to make sure that we don't find
* an out-of-place "--end-runtime-options" option. */
{
char *argi0 = argv[argi];
int argj = 1;
/* (argc - argi) for the arguments, one for the binary,
and one for the terminating NULL. */
sbcl_argv = successful_malloc((2 + argc - argi) * sizeof(char *));
sbcl_argv[0] = argv[0];
while (argi < argc) {
char *arg = argv[argi++];
/* If we encounter --end-runtime-options for the first
* time after the point where we had to give up on
* runtime options, then the point where we had to
* give up on runtime options must've been a user
* error. */
if (!end_runtime_options &&
0 == strcmp(arg, "--end-runtime-options")) {
lose("bad runtime option \"%s\"\n", argi0);
}
sbcl_argv[argj++] = arg;
}
sbcl_argv[argj] = 0;
}
}
/* Align down to multiple of page_table page size, and to the appropriate
* stack alignment. */
dynamic_space_size &= ~(PAGE_BYTES-1);
thread_control_stack_size &= ~(CONTROL_STACK_ALIGNMENT_BYTES-1);
/* Preserve the runtime options for possible future core saving */
runtime_options->dynamic_space_size = dynamic_space_size;
runtime_options->thread_control_stack_size = thread_control_stack_size;
/* KLUDGE: os_vm_page_size is set by os_init(), and on some
* systems (e.g. Alpha) arch_init() needs need os_vm_page_size, so
* it must follow os_init(). -- WHN 2000-01-26 */
os_init(argv, envp);
arch_init();
gc_init();
validate();
/* If no core file was specified, look for one. */
if (!core) {
core = search_for_core();
}
/* Make sure that SBCL_HOME is set and not the empty string,
unless loading an embedded core. */
if (!(sbcl_home && *sbcl_home) && embedded_core_offset == 0) {
char *envstring, *copied_core, *dir;
char *stem = "SBCL_HOME=";
copied_core = copied_string(core);
dir = dirname(copied_core);
envstring = (char *) calloc(strlen(stem) +
strlen(dir) +
1,
sizeof(char));
sprintf(envstring, "%s%s", stem, dir);
putenv(envstring);
free(copied_core);
}
if (!noinform && embedded_core_offset == 0) {
print_banner();
fflush(stdout);
}
#if defined(SVR4) || defined(__linux__)
tzset();
#endif
define_var("nil", NIL, 1);
define_var("t", T, 1);
if (!disable_lossage_handler_p)
enable_lossage_handler();
globals_init();
initial_function = load_core_file(core, embedded_core_offset);
if (initial_function == NIL) {
lose("couldn't find initial function\n");
}
#ifdef LISP_FEATURE_HPUX
/* -1 = CLOSURE_FUN_OFFSET, 23 = SIMPLE_FUN_CODE_OFFSET, we are
* not in LANGUAGE_ASSEMBLY so we cant reach them. */
return_from_lisp_stub = (void *) ((char *)*((unsigned long *)
((char *)initial_function + -1)) + 23);
#endif
gc_initialize_pointers();
arch_install_interrupt_handlers();
#ifndef LISP_FEATURE_WIN32
os_install_interrupt_handlers();
#else
/* wos_install_interrupt_handlers(handler); */
wos_install_interrupt_handlers(&exception_frame);
#endif
/* Pass core filename and the processed argv into Lisp. They'll
* need to be processed further there, to do locale conversion.
*/
core_string = core;
posix_argv = sbcl_argv;
FSHOW((stderr, "/funcalling initial_function=0x%lx\n",
(unsigned long)initial_function));
#ifdef LISP_FEATURE_WIN32
fprintf(stderr, "\n\
This is experimental prerelease support for the Windows platform: use\n\
at your own risk. \"Your Kitten of Death awaits!\"\n");
fflush(stdout);
fflush(stderr);
#endif
create_initial_thread(initial_function);
lose("CATS. CATS ARE NICE.\n");
return 0;
}
static BOOT_CODE bool_t try_init_kernel(
paddr_t ui_p_reg_start,
paddr_t ui_p_reg_end,
sword_t pv_offset,
vptr_t v_entry,
paddr_t dtb_addr_start,
paddr_t dtb_addr_end
)
{
cap_t root_cnode_cap;
cap_t it_ap_cap;
cap_t it_pd_cap;
cap_t ipcbuf_cap;
region_t ui_reg = paddr_to_pptr_reg((p_region_t) {
ui_p_reg_start, ui_p_reg_end
});
region_t dtb_reg;
word_t extra_bi_size = sizeof(seL4_BootInfoHeader) + (dtb_addr_end - dtb_addr_start);
region_t extra_bi_region;
pptr_t extra_bi_offset = 0;
vptr_t extra_bi_frame_vptr;
pptr_t bi_frame_pptr;
vptr_t bi_frame_vptr;
vptr_t ipcbuf_vptr;
create_frames_of_region_ret_t create_frames_ret;
create_frames_of_region_ret_t extra_bi_ret;
/* convert from physical addresses to userland vptrs */
v_region_t ui_v_reg;
v_region_t it_v_reg;
ui_v_reg.start = ui_p_reg_start - pv_offset;
ui_v_reg.end = ui_p_reg_end - pv_offset;
ipcbuf_vptr = ui_v_reg.end;
bi_frame_vptr = ipcbuf_vptr + BIT(PAGE_BITS);
extra_bi_frame_vptr = bi_frame_vptr + BIT(PAGE_BITS);
/* If no DTB was provided, skip allocating extra bootinfo */
if (dtb_addr_start == 0) {
extra_bi_size = 0;
dtb_reg = (region_t) {
0, 0
};
} else {
dtb_reg = paddr_to_pptr_reg((p_region_t) {
dtb_addr_start, ROUND_UP(dtb_addr_end, PAGE_BITS)
});
}
/* The region of the initial thread is the user image + ipcbuf and boot info */
it_v_reg.start = ui_v_reg.start;
it_v_reg.end = extra_bi_frame_vptr;
if (it_v_reg.end > kernelBase) {
printf("Userland image virtual end address too high\n");
return false;
}
/* setup virtual memory for the kernel */
map_kernel_window();
/* initialise the CPU */
if (!init_cpu()) {
return false;
}
/* debug output via serial port is only available from here */
printf("Bootstrapping kernel\n");
/* initialise the platform */
init_plat();
/* make the free memory available to alloc_region() */
arch_init_freemem(ui_reg, dtb_reg);
/* create the root cnode */
root_cnode_cap = create_root_cnode();
if (cap_get_capType(root_cnode_cap) == cap_null_cap) {
return false;
}
/* create the cap for managing thread domains */
create_domain_cap(root_cnode_cap);
/* initialise the IRQ states and provide the IRQ control cap */
init_irqs(root_cnode_cap);
/* create the bootinfo frame */
bi_frame_pptr = allocate_bi_frame(0, CONFIG_MAX_NUM_NODES, ipcbuf_vptr);
if (!bi_frame_pptr) {
return false;
}
/* create extra bootinfo region - will return an empty allocation if extra_bi_size = 0 */
extra_bi_region = allocate_extra_bi_region(extra_bi_size);
if (extra_bi_region.start == 0) {
return false;
}
/* update initial thread virtual address range for extra bootinfo */
it_v_reg.end += extra_bi_region.end - extra_bi_region.start;
if (it_v_reg.end > kernelBase) {
printf("Userland extra bootinfo end address too high\n");
return false;
}
/* put DTB in the bootinfo block, if present. */
seL4_BootInfoHeader header;
if (dtb_reg.start) {
header.id = SEL4_BOOTINFO_HEADER_FDT;
header.len = sizeof(header) + dtb_reg.end - dtb_reg.start;
*(seL4_BootInfoHeader *)(extra_bi_region.start + extra_bi_offset) = header;
extra_bi_offset += sizeof(header);
memcpy((void *)(extra_bi_region.start + extra_bi_offset), (void *)dtb_reg.start,
dtb_reg.end - dtb_reg.start);
extra_bi_offset += dtb_reg.end - dtb_reg.start;
}
if ((extra_bi_region.end - extra_bi_region.start) - extra_bi_offset > 0) {
/* provde a chunk for any leftover padding in the extended boot info */
header.id = SEL4_BOOTINFO_HEADER_PADDING;
header.len = (extra_bi_region.end - extra_bi_region.start) - extra_bi_offset;
*(seL4_BootInfoHeader *)(extra_bi_region.start + extra_bi_offset) = header;
}
if (config_set(CONFIG_ARM_SMMU)) {
ndks_boot.bi_frame->ioSpaceCaps = create_iospace_caps(root_cnode_cap);
if (ndks_boot.bi_frame->ioSpaceCaps.start == 0 &&
ndks_boot.bi_frame->ioSpaceCaps.end == 0) {
return false;
}
} else {
ndks_boot.bi_frame->ioSpaceCaps = S_REG_EMPTY;
}
/* Construct an initial address space with enough virtual addresses
* to cover the user image + ipc buffer and bootinfo frames */
it_pd_cap = create_it_address_space(root_cnode_cap, it_v_reg);
if (cap_get_capType(it_pd_cap) == cap_null_cap) {
return false;
}
/* Create and map bootinfo frame cap */
create_bi_frame_cap(
root_cnode_cap,
it_pd_cap,
bi_frame_pptr,
bi_frame_vptr
);
/* create and map extra bootinfo region */
if (extra_bi_size > 0) {
extra_bi_ret =
create_frames_of_region(
root_cnode_cap,
it_pd_cap,
extra_bi_region,
true,
pptr_to_paddr((void *)extra_bi_region.start) - extra_bi_frame_vptr
);
if (!extra_bi_ret.success) {
return false;
}
ndks_boot.bi_frame->extraBIPages = extra_bi_ret.region;
}
/* create the initial thread's IPC buffer */
ipcbuf_cap = create_ipcbuf_frame(root_cnode_cap, it_pd_cap, ipcbuf_vptr);
if (cap_get_capType(ipcbuf_cap) == cap_null_cap) {
return false;
}
/* create all userland image frames */
create_frames_ret =
create_frames_of_region(
root_cnode_cap,
it_pd_cap,
ui_reg,
true,
pv_offset
);
if (!create_frames_ret.success) {
return false;
}
ndks_boot.bi_frame->userImageFrames = create_frames_ret.region;
/* create/initialise the initial thread's ASID pool */
it_ap_cap = create_it_asid_pool(root_cnode_cap);
if (cap_get_capType(it_ap_cap) == cap_null_cap) {
return false;
}
write_it_asid_pool(it_ap_cap, it_pd_cap);
/* create the idle thread */
if (!create_idle_thread()) {
return false;
}
/* Before creating the initial thread (which also switches to it)
* we clean the cache so that any page table information written
* as a result of calling create_frames_of_region will be correctly
* read by the hardware page table walker */
cleanInvalidateL1Caches();
/* create the initial thread */
tcb_t *initial = create_initial_thread(
root_cnode_cap,
it_pd_cap,
v_entry,
bi_frame_vptr,
ipcbuf_vptr,
ipcbuf_cap
);
if (initial == NULL) {
return false;
}
init_core_state(initial);
/* create all of the untypeds. Both devices and kernel window memory */
if (!create_untypeds(
root_cnode_cap,
(region_t) {
kernelBase, (pptr_t)ki_boot_end
} /* reusable boot code/data */
)) {
return false;
}
/* no shared-frame caps (ARM has no multikernel support) */
ndks_boot.bi_frame->sharedFrames = S_REG_EMPTY;
/* finalise the bootinfo frame */
bi_finalise();
/* make everything written by the kernel visible to userland. Cleaning to PoC is not
* strictly neccessary, but performance is not critical here so clean and invalidate
* everything to PoC */
cleanInvalidateL1Caches();
invalidateLocalTLB();
if (config_set(CONFIG_ARM_HYPERVISOR_SUPPORT)) {
invalidateHypTLB();
}
ksNumCPUs = 1;
/* initialize BKL before booting up other cores */
SMP_COND_STATEMENT(clh_lock_init());
SMP_COND_STATEMENT(release_secondary_cpus());
/* grab BKL before leaving the kernel */
NODE_LOCK_SYS;
printf("Booting all finished, dropped to user space\n");
/* kernel successfully initialized */
return true;
}
BOOT_CODE bool_t
init_node_state(
p_region_t avail_p_reg,
p_region_t sh_p_reg,
dev_p_regs_t* dev_p_regs,
ui_info_t ui_info,
p_region_t boot_mem_reuse_p_reg,
node_id_t node_id,
uint32_t num_nodes,
/* parameters below not modeled in abstract specification */
pdpte_t* kernel_pdpt,
pde_t* kernel_pd,
pte_t* kernel_pt
#ifdef CONFIG_IOMMU
, cpu_id_t cpu_id,
uint32_t num_drhu,
paddr_t* drhu_list,
uint32_t num_passthrough_dev,
dev_id_t* passthrough_dev_list,
uint32_t* pci_bus_used_bitmap
#endif
)
{
cap_t root_cnode_cap;
vptr_t bi_frame_vptr;
vptr_t ipcbuf_vptr;
cap_t it_vspace_cap;
cap_t it_ap_cap;
cap_t ipcbuf_cap;
pptr_t bi_frame_pptr;
create_frames_of_region_ret_t create_frames_ret;
int i;
#ifdef CONFIG_BENCHMARK
vm_attributes_t buffer_attr = {{ 0 }};
uint32_t paddr;
pde_t pde;
#endif /* CONFIG_BENCHMARK */
/* convert from physical addresses to kernel pptrs */
region_t avail_reg = paddr_to_pptr_reg(avail_p_reg);
region_t ui_reg = paddr_to_pptr_reg(ui_info.p_reg);
region_t sh_reg = paddr_to_pptr_reg(sh_p_reg);
region_t boot_mem_reuse_reg = paddr_to_pptr_reg(boot_mem_reuse_p_reg);
/* convert from physical addresses to userland vptrs */
v_region_t ui_v_reg;
v_region_t it_v_reg;
ui_v_reg.start = ui_info.p_reg.start - ui_info.pv_offset;
ui_v_reg.end = ui_info.p_reg.end - ui_info.pv_offset;
ipcbuf_vptr = ui_v_reg.end;
bi_frame_vptr = ipcbuf_vptr + BIT(PAGE_BITS);
/* The region of the initial thread is the user image + ipcbuf and boot info */
it_v_reg.start = ui_v_reg.start;
it_v_reg.end = bi_frame_vptr + BIT(PAGE_BITS);
/* make the free memory available to alloc_region() */
ndks_boot.freemem[0] = avail_reg;
for (i = 1; i < MAX_NUM_FREEMEM_REG; i++) {
ndks_boot.freemem[i] = REG_EMPTY;
}
/* initialise virtual-memory-related data structures (not in abstract spec) */
if (!init_vm_state(kernel_pdpt, kernel_pd, kernel_pt)) {
return false;
}
#ifdef CONFIG_BENCHMARK
/* allocate and create the log buffer */
buffer_attr.words[0] = IA32_PAT_MT_WRITE_THROUGH;
paddr = pptr_to_paddr((void *) alloc_region(pageBitsForSize(IA32_LargePage)));
/* allocate a large frame for logging */
pde = pde_pde_large_new(
paddr, /* page_base_address */
vm_attributes_get_ia32PATBit(buffer_attr), /* pat */
0, /* avl_cte_depth */
1, /* global */
0, /* dirty */
0, /* accessed */
vm_attributes_get_ia32PCDBit(buffer_attr), /* cache_disabled */
vm_attributes_get_ia32PWTBit(buffer_attr), /* write_through */
0, /* super_user */
1, /* read_write */
1 /* present */
);
/* TODO this shouldn't be hardcoded */
ia32KSkernelPD[IA32_KSLOG_IDX] = pde;
/* flush the tlb */
invalidatePageStructureCache();
/* if we crash here, the log isn't working */
#ifdef CONFIG_DEBUG_BUILD
printf("Testing log\n");
ksLog[0] = 0xdeadbeef;
printf("Wrote to ksLog %x\n", ksLog[0]);
assert(ksLog[0] == 0xdeadbeef);
#endif /* CONFIG_DEBUG_BUILD */
#endif /* CONFIG_BENCHMARK */
/* create the root cnode */
root_cnode_cap = create_root_cnode();
/* create the IO port cap */
write_slot(
SLOT_PTR(pptr_of_cap(root_cnode_cap), BI_CAP_IO_PORT),
cap_io_port_cap_new(
0, /* first port */
NUM_IO_PORTS - 1 /* last port */
)
);
/* create the cap for managing thread domains */
create_domain_cap(root_cnode_cap);
/* create the IRQ CNode */
if (!create_irq_cnode()) {
return false;
}
/* initialise the IRQ states and provide the IRQ control cap */
init_irqs(root_cnode_cap, node_id != 0);
/* create the bootinfo frame */
bi_frame_pptr = allocate_bi_frame(node_id, num_nodes, ipcbuf_vptr);
if (!bi_frame_pptr) {
return false;
}
/* Construct an initial address space with enough virtual addresses
* to cover the user image + ipc buffer and bootinfo frames */
it_vspace_cap = create_it_address_space(root_cnode_cap, it_v_reg);
if (cap_get_capType(it_vspace_cap) == cap_null_cap) {
return false;
}
/* Create and map bootinfo frame cap */
create_bi_frame_cap(
root_cnode_cap,
it_vspace_cap,
bi_frame_pptr,
bi_frame_vptr
);
/* create the initial thread's IPC buffer */
ipcbuf_cap = create_ipcbuf_frame(root_cnode_cap, it_vspace_cap, ipcbuf_vptr);
if (cap_get_capType(ipcbuf_cap) == cap_null_cap) {
return false;
}
/* create all userland image frames */
create_frames_ret =
create_frames_of_region(
root_cnode_cap,
it_vspace_cap,
ui_reg,
true,
ui_info.pv_offset
);
if (!create_frames_ret.success) {
return false;
}
ndks_boot.bi_frame->ui_frame_caps = create_frames_ret.region;
/* create the initial thread's ASID pool */
it_ap_cap = create_it_asid_pool(root_cnode_cap);
if (cap_get_capType(it_ap_cap) == cap_null_cap) {
return false;
}
write_it_asid_pool(it_ap_cap, it_vspace_cap);
/*
* Initialise the NULL FPU state. This is different from merely zero'ing it
* out (i.e., the NULL FPU state is non-zero), and must be performed before
* the first thread is created.
*/
resetFpu();
saveFpuState(&ia32KSnullFpuState);
ia32KSfpuOwner = NULL;
/* create the idle thread */
if (!create_idle_thread()) {
return false;
}
/* create the initial thread */
if (!create_initial_thread(
root_cnode_cap,
it_vspace_cap,
ui_info.v_entry,
bi_frame_vptr,
ipcbuf_vptr,
ipcbuf_cap
)) {
return false;
}
#ifdef CONFIG_IOMMU
/* initialise VTD-related data structures and the IOMMUs */
if (!vtd_init(cpu_id, num_drhu, pci_bus_used_bitmap, num_passthrough_dev, passthrough_dev_list)) {
return false;
}
/* write number of IOMMU PT levels into bootinfo */
ndks_boot.bi_frame->num_iopt_levels = ia32KSnumIOPTLevels;
/* write IOSpace master cap */
write_slot(SLOT_PTR(pptr_of_cap(root_cnode_cap), BI_CAP_IO_SPACE), master_iospace_cap());
#endif
/* convert the remaining free memory into UT objects and provide the caps */
if (!create_untypeds(root_cnode_cap, boot_mem_reuse_reg)) {
return false;
}
/* WARNING: alloc_region() must not be called anymore after here! */
/* create device frames */
if (!create_device_frames(root_cnode_cap, dev_p_regs)) {
return false;
}
/* create all shared frames */
create_frames_ret =
create_frames_of_region(
root_cnode_cap,
it_vspace_cap,
sh_reg,
false,
0
);
if (!create_frames_ret.success) {
return false;
}
ndks_boot.bi_frame->sh_frame_caps = create_frames_ret.region;;
/* finalise the bootinfo frame */
bi_finalise();
#ifdef DEBUG
ia32KSconsolePort = console_port_of_node(node_id);
ia32KSdebugPort = debug_port_of_node(node_id);
#endif
return true;
}
