// Get the bootinfo and map it in.
static errval_t map_bootinfo(struct bootinfo **bootinfo)
{
errval_t err, msgerr;
struct monitor_blocking_rpc_client *cl = get_monitor_blocking_rpc_client();
assert(cl != NULL);
struct capref bootinfo_frame;
size_t bootinfo_size;
msgerr = cl->vtbl.get_bootinfo(cl, &err, &bootinfo_frame, &bootinfo_size);
if (err_is_fail(msgerr)) {
err = msgerr;
}
if (err_is_fail(err)) {
USER_PANIC_ERR(err, "failed in get_bootinfo");
return err;
}
err = vspace_map_one_frame((void**)bootinfo, bootinfo_size, bootinfo_frame,
NULL, NULL);
assert(err_is_ok(err));
return err;
}
/* allocate inrq */
static errval_t arm_allocirq(struct capref ep, uint32_t irq)
{
errval_t err, msgerr;
struct monitor_blocking_rpc_client *r = get_monitor_blocking_rpc_client();
err = r->vtbl.arm_irq_handle(r, ep, irq, &msgerr);
if (err_is_fail(err)){
return err;
} else {
return msgerr;
}
}
/* Allocate vector from local monitor */
static errval_t allocirq(struct capref ep, uint32_t *retvector)
{
errval_t err, msgerr;
uint32_t vector;
struct monitor_blocking_rpc_client *r = get_monitor_blocking_rpc_client();
err = r->vtbl.irq_handle(r, ep, &msgerr, &vector);
if (err_is_fail(err)){
return err;
} else if (err_is_fail(msgerr)) {
return msgerr;
} else {
*retvector = vector;
return msgerr;
}
}
int start_aps_x86_32_start(uint8_t core_id, genvaddr_t entry)
{
DEBUG("%s:%d: start_aps_x86_32_start\n", __FILE__, __LINE__);
// Copy the startup code to the real-mode address
uint8_t *real_src = (uint8_t *) &x86_32_start_ap;
uint8_t *real_end = (uint8_t *) &x86_32_start_ap_end;
struct capref bootcap;
struct acpi_rpc_client* acl = get_acpi_rpc_client();
errval_t error_code;
errval_t err = acl->vtbl.mm_realloc_range_proxy(acl, 16, 0x0,
&bootcap, &error_code);
if (err_is_fail(err)) {
USER_PANIC_ERR(err, "mm_alloc_range_proxy failed.");
}
if (err_is_fail(error_code)) {
USER_PANIC_ERR(error_code, "mm_alloc_range_proxy return failed.");
}
void* real_base;
err = vspace_map_one_frame(&real_base, 1<<16, bootcap, NULL, NULL);
uint8_t* real_dest = (uint8_t*)real_base + X86_32_REAL_MODE_LINEAR_OFFSET;
memcpy(real_dest, real_src, real_end - real_src);
/* Pointer to the entry point called from init_ap.S */
volatile uint64_t *absolute_entry_ptr = (volatile uint64_t *)
((
(lpaddr_t) &x86_32_init_ap_absolute_entry -
(lpaddr_t) &x86_32_start_ap
)
+
real_dest);
//copy the address of the function start (in boot.S) to the long-mode
//assembler code to be able to perform an absolute jump
*absolute_entry_ptr = entry;
// pointer to the shared global variable amongst all kernels
volatile uint64_t *ap_global = (volatile uint64_t *)
((
(lpaddr_t) &x86_32_init_ap_global -
(lpaddr_t) &x86_32_start_ap
)
+ real_dest);
genpaddr_t global;
struct monitor_blocking_rpc_client *mc =
get_monitor_blocking_rpc_client();
err = mc->vtbl.get_global_paddr(mc, &global);
if (err_is_fail(err)) {
DEBUG_ERR(err, "invoke spawn core");
return err_push(err, MON_ERR_SPAWN_CORE);
}
*ap_global = (uint64_t)(genpaddr_t)global;
// pointer to the pseudo-lock used to detect boot up of new core
volatile uint32_t *ap_wait = (volatile uint32_t *)
((lpaddr_t) &x86_32_init_ap_wait -
((lpaddr_t) &x86_32_start_ap) +
real_dest);
// Pointer to the lock variable in the realmode code
volatile uint8_t *ap_lock = (volatile uint8_t *)
((lpaddr_t) &x86_32_init_ap_lock -
((lpaddr_t) &x86_32_start_ap) +
real_dest);
*ap_wait = AP_STARTING_UP;
end = bench_tsc();
err = invoke_send_init_ipi(ipi_cap, core_id);
if (err_is_fail(err)) {
DEBUG_ERR(err, "invoke send init ipi");
return err;
}
err = invoke_send_start_ipi(ipi_cap, core_id, entry);
if (err_is_fail(err)) {
DEBUG_ERR(err, "invoke sipi");
return err;
}
//give the new core a bit time to start-up and set the lock
for (uint64_t i = 0; i < STARTUP_TIMEOUT; i++) {
if (*ap_lock != 0) {
break;
}
}
// If the lock is set, the core has been started, otherwise assume, that
// a core with this APIC ID doesn't exist.
if (*ap_lock != 0) {
while (*ap_wait != AP_STARTED);
trace_event(TRACE_SUBSYS_KERNEL, TRACE_EVENT_KERNEL_CORE_START_REQUEST_ACK, core_id);
*ap_lock = 0;
return 0;
}
assert(!"badness");
return -1;
}
static errval_t spawn(char *path, char *const argv[], char *argbuf,
size_t argbytes, char *const envp[],
struct capref inheritcn_cap, struct capref argcn_cap,
domainid_t *domainid)
{
errval_t err, msgerr;
/* read file into memory */
vfs_handle_t fh;
err = vfs_open(path, &fh);
if (err_is_fail(err)) {
return err_push(err, SPAWN_ERR_LOAD);
}
struct vfs_fileinfo info;
err = vfs_stat(fh, &info);
if (err_is_fail(err)) {
vfs_close(fh);
return err_push(err, SPAWN_ERR_LOAD);
}
assert(info.type == VFS_FILE);
uint8_t *image = malloc(info.size);
if (image == NULL) {
vfs_close(fh);
return err_push(err, SPAWN_ERR_LOAD);
}
size_t pos = 0, readlen;
do {
err = vfs_read(fh, &image[pos], info.size - pos, &readlen);
if (err_is_fail(err)) {
vfs_close(fh);
free(image);
return err_push(err, SPAWN_ERR_LOAD);
} else if (readlen == 0) {
vfs_close(fh);
free(image);
return SPAWN_ERR_LOAD; // XXX
} else {
pos += readlen;
}
} while (err_is_ok(err) && readlen > 0 && pos < info.size);
err = vfs_close(fh);
if (err_is_fail(err)) {
DEBUG_ERR(err, "failed to close file %s", path);
}
// find short name (last part of path)
char *name = strrchr(path, VFS_PATH_SEP);
if (name == NULL) {
name = path;
} else {
name++;
}
/* spawn the image */
struct spawninfo si;
err = spawn_load_image(&si, (lvaddr_t)image, info.size, CURRENT_CPU_TYPE,
name, my_core_id, argv, envp, inheritcn_cap,
argcn_cap);
if (err_is_fail(err)) {
free(image);
return err;
}
free(image);
/* request connection from monitor */
struct monitor_blocking_rpc_client *mrpc = get_monitor_blocking_rpc_client();
struct capref monep;
err = mrpc->vtbl.alloc_monitor_ep(mrpc, &msgerr, &monep);
if (err_is_fail(err)) {
return err_push(err, SPAWN_ERR_MONITOR_CLIENT);
} else if (err_is_fail(msgerr)) {
return msgerr;
}
/* copy connection into the new domain */
struct capref destep = {
.cnode = si.rootcn,
.slot = ROOTCN_SLOT_MONITOREP,
};
err = cap_copy(destep, monep);
if (err_is_fail(err)) {
spawn_free(&si);
cap_destroy(monep);
return err_push(err, SPAWN_ERR_MONITOR_CLIENT);
}
err = cap_destroy(monep);
if (err_is_fail(err)) {
return err_push(err, SPAWN_ERR_MONITOR_CLIENT);
}
debug_printf("spawning %s on core %u\n", path, my_core_id);
/* give the perfmon capability */
struct capref dest, src;
dest.cnode = si.taskcn;
dest.slot = TASKCN_SLOT_PERF_MON;
src.cnode = cnode_task;
src.slot = TASKCN_SLOT_PERF_MON;
err = cap_copy(dest, src);
if (err_is_fail(err)) {
return err_push(err, INIT_ERR_COPY_PERF_MON);
}
/* run the domain */
err = spawn_run(&si);
if (err_is_fail(err)) {
spawn_free(&si);
return err_push(err, SPAWN_ERR_RUN);
}
// Allocate domain id
struct ps_entry *pe = malloc(sizeof(struct ps_entry));
assert(pe != NULL);
memset(pe, 0, sizeof(struct ps_entry));
memcpy(pe->argv, argv, MAX_CMDLINE_ARGS*sizeof(*argv));
pe->argbuf = argbuf;
pe->argbytes = argbytes;
/*
* NB: It's important to keep a copy of the DCB *and* the root
* CNode around. We need to revoke both (in the right order, see
* kill_domain() below), so that we ensure no one else is
* referring to the domain's CSpace anymore. Especially the loop
* created by placing rootcn into its own address space becomes a
* problem here.
*/
err = slot_alloc(&pe->rootcn_cap);
assert(err_is_ok(err));
err = cap_copy(pe->rootcn_cap, si.rootcn_cap);
pe->rootcn = si.rootcn;
assert(err_is_ok(err));
err = slot_alloc(&pe->dcb);
assert(err_is_ok(err));
err = cap_copy(pe->dcb, si.dcb);
assert(err_is_ok(err));
pe->status = PS_STATUS_RUNNING;
err = ps_allocate(pe, domainid);
if(err_is_fail(err)) {
free(pe);
}
// Store in target dispatcher frame
struct dispatcher_generic *dg = get_dispatcher_generic(si.handle);
dg->domain_id = *domainid;
/* cleanup */
err = spawn_free(&si);
if (err_is_fail(err)) {
return err_push(err, SPAWN_ERR_FREE);
}
return SYS_ERR_OK;
}
static void retry_use_local_memserv_response(void *a)
{
errval_t err;
struct spawn_binding *b = (struct spawn_binding*)a;
err = b->tx_vtbl.use_local_memserv_response(b, NOP_CONT);
if (err_no(err) == FLOUNDER_ERR_TX_BUSY) {
// try again
err = b->register_send(b, get_default_waitset(),
MKCONT(retry_use_local_memserv_response,a));
}
if (err_is_fail(err)) {
DEBUG_ERR(err, "error sending use_local_memserv reply\n");
}
}
static errval_t init_allocators(void)
{
errval_t err, msgerr;
struct monitor_blocking_rpc_client *cl = get_monitor_blocking_rpc_client();
assert(cl != NULL);
// Get the bootinfo and map it in.
struct capref bootinfo_frame;
size_t bootinfo_size;
struct bootinfo *bootinfo;
msgerr = cl->vtbl.get_bootinfo(cl, &err, &bootinfo_frame, &bootinfo_size);
if (err_is_fail(msgerr) || err_is_fail(err)) {
USER_PANIC_ERR(err_is_fail(msgerr) ? msgerr : err, "failed in get_bootinfo");
}
err = vspace_map_one_frame((void**)&bootinfo, bootinfo_size, bootinfo_frame,
NULL, NULL);
assert(err_is_ok(err));
/* Initialize the memory allocator to handle PhysAddr caps */
static struct range_slot_allocator devframes_allocator;
err = range_slot_alloc_init(&devframes_allocator, PCI_CNODE_SLOTS, NULL);
if (err_is_fail(err)) {
return err_push(err, LIB_ERR_SLOT_ALLOC_INIT);
}
err = mm_init(&pci_mm_physaddr, ObjType_DevFrame, 0, 48,
/* This next parameter is important. It specifies the maximum
* amount that a cap may be "chunked" (i.e. broken up) at each
* level in the allocator. Setting it higher than 1 reduces the
* memory overhead of keeping all the intermediate caps around,
* but leads to problems if you chunk up a cap too small to be
* able to allocate a large subregion. This caused problems
* for me with a large framebuffer... -AB 20110810 */
1, /*was DEFAULT_CNODE_BITS,*/
slab_default_refill, slot_alloc_dynamic, &devframes_allocator, false);
if (err_is_fail(err)) {
return err_push(err, MM_ERR_MM_INIT);
}
// Request I/O Cap
struct capref requested_caps;
errval_t error_code;
err = cl->vtbl.get_io_cap(cl, &requested_caps, &error_code);
assert(err_is_ok(err) && err_is_ok(error_code));
// Copy into correct slot
struct capref caps_io = {
.cnode = cnode_task,
.slot = TASKCN_SLOT_IO
};
err = cap_copy(caps_io, requested_caps);
// XXX: The code below is confused about gen/l/paddrs.
// Caps should be managed in genpaddr, while the bus mgmt must be in lpaddr.
err = cl->vtbl.get_phyaddr_cap(cl, &requested_caps, &error_code);
assert(err_is_ok(err) && err_is_ok(error_code));
physical_caps = requested_caps;
// Build the capref for the first physical address capability
struct capref phys_cap;
phys_cap.cnode = build_cnoderef(requested_caps, PHYSADDRCN_BITS);
phys_cap.slot = 0;
struct cnoderef devcnode;
err = slot_alloc(&my_devframes_cnode);
assert(err_is_ok(err));
cslot_t slots;
err = cnode_create(&my_devframes_cnode, &devcnode, 255, &slots);
if (err_is_fail(err)) { USER_PANIC_ERR(err, "cnode create"); }
struct capref devframe;
devframe.cnode = devcnode;
devframe.slot = 0;
for (int i = 0; i < bootinfo->regions_length; i++) {
struct mem_region *mrp = &bootinfo->regions[i];
if (mrp->mr_type == RegionType_Module) {
skb_add_fact("memory_region(16'%" PRIxGENPADDR ",%u,%zu,%u,%tu).",
mrp->mr_base,
0,
mrp->mrmod_size,
mrp->mr_type,
mrp->mrmod_data);
}
else {
skb_add_fact("memory_region(16'%" PRIxGENPADDR ",%u,%zu,%u,%tu).",
mrp->mr_base,
mrp->mr_bits,
((size_t)1) << mrp->mr_bits,
mrp->mr_type,
mrp->mrmod_data);
}
if (mrp->mr_type == RegionType_PhyAddr ||
mrp->mr_type == RegionType_PlatformData) {
ACPI_DEBUG("Region %d: %"PRIxGENPADDR" - %"PRIxGENPADDR" %s\n",
i, mrp->mr_base,
mrp->mr_base + (((size_t)1)<<mrp->mr_bits),
mrp->mr_type == RegionType_PhyAddr ?
"physical address" : "platform data");
err = cap_retype(devframe, phys_cap, ObjType_DevFrame, mrp->mr_bits);
if (err_no(err) == SYS_ERR_REVOKE_FIRST) {
printf("cannot retype region %d: need to revoke first; ignoring it\n", i);
} else {
assert(err_is_ok(err));
err = mm_add(&pci_mm_physaddr, devframe,
mrp->mr_bits, mrp->mr_base);
if (err_is_fail(err)) {
USER_PANIC_ERR(err, "adding region %d FAILED\n", i);
}
}
phys_cap.slot++;
devframe.slot++;
}
}
return SYS_ERR_OK;
}
