void
cpu_set_tss_gates(struct cpu_info *ci)
{
struct segment_descriptor sd;
ci->ci_doubleflt_stack = (char *)uvm_km_alloc(kernel_map, USPACE);
cpu_init_tss(&ci->ci_doubleflt_tss, ci->ci_doubleflt_stack,
IDTVEC(tss_trap08));
setsegment(&sd, &ci->ci_doubleflt_tss, sizeof(struct i386tss) - 1,
SDT_SYS386TSS, SEL_KPL, 0, 0);
ci->ci_gdt[GTRAPTSS_SEL].sd = sd;
setgate(&idt[8], NULL, 0, SDT_SYSTASKGT, SEL_KPL,
GSEL(GTRAPTSS_SEL, SEL_KPL));
#if defined(DDB) && defined(MULTIPROCESSOR)
/*
* Set up separate handler for the DDB IPI, so that it doesn't
* stomp on a possibly corrupted stack.
*
* XXX overwriting the gate set in db_machine_init.
* Should rearrange the code so that it's set only once.
*/
ci->ci_ddbipi_stack = (char *)uvm_km_alloc(kernel_map, USPACE);
cpu_init_tss(&ci->ci_ddbipi_tss, ci->ci_ddbipi_stack,
Xintrddbipi);
setsegment(&sd, &ci->ci_ddbipi_tss, sizeof(struct i386tss) - 1,
SDT_SYS386TSS, SEL_KPL, 0, 0);
ci->ci_gdt[GIPITSS_SEL].sd = sd;
setgate(&idt[ddb_vec], NULL, 0, SDT_SYSTASKGT, SEL_KPL,
GSEL(GIPITSS_SEL, SEL_KPL));
#endif
}
void trap_init()
{
int i;
extern uint32_t trap_handlers[];
for(i = 0;i < 256; i++) {
setgate(idt[i], 0, _KERNEL_CS_, trap_handlers[i], 0);
}
setgate(idt[T_SYSCALL], 0, _KERNEL_CS_, trap_handlers[T_SYSCALL], 3);
trap_init_percpu();
}
void
idt_vec_set(int vec, void (*function)(void))
{
KASSERT(mutex_owned(&cpu_lock) || !mp_online);
KASSERT(idt_allocmap[vec] == 1);
setgate(&idt[vec], function, 0, SDT_SYS386IGT, SEL_KPL,
GSEL(GCODE_SEL, SEL_KPL));
}
void
idt_vec_set(int vec, void (*function)(void))
{
/*
* Vector should be allocated, so no locking needed.
*/
KASSERT(idt_allocmap[vec] == 1);
setgate(&idt[vec], function, 0, SDT_SYS386IGT, SEL_KPL,
GSEL(GCODE_SEL, SEL_KPL));
}
/*
* Fill in default interrupt table (in case of spurious interrupt
* during configuration of kernel), setup interrupt control unit
*/
void
intr_default_setup(void)
{
int i;
/* icu vectors */
for (i = 0; i < NUM_LEGACY_IRQS; i++) {
idt_allocmap[ICU_OFFSET + i] = 1;
setgate(&idt[ICU_OFFSET + i],
i8259_stubs[i].ist_entry, 0, SDT_SYS386IGT,
SEL_KPL, GSEL(GCODE_SEL, SEL_KPL));
}
/*
* Eventually might want to check if it's actually there.
*/
i8259_default_setup();
}
/*
* Fill in default interrupt table (in case of spurious interrupt
* during configuration of kernel, setup interrupt control unit
*/
void
isa_defaultirq(void)
{
int i;
/* icu vectors */
for (i = 0; i < ICU_LEN; i++)
setgate(&idt[ICU_OFFSET + i], IDTVEC(intr)[i], 0,
SDT_SYS386IGT, SEL_KPL, GICODE_SEL);
/* initialize 8259's */
outb(IO_ICU1, 0x11); /* reset; program device, four bytes */
outb(IO_ICU1+1, ICU_OFFSET); /* starting at this vector index */
outb(IO_ICU1+1, 1 << IRQ_SLAVE); /* slave on line 2 */
#ifdef AUTO_EOI_1
outb(IO_ICU1+1, 2 | 1); /* auto EOI, 8086 mode */
#else
outb(IO_ICU1+1, 1); /* 8086 mode */
#endif
outb(IO_ICU1+1, 0xff); /* leave interrupts masked */
outb(IO_ICU1, 0x68); /* special mask mode (if available) */
outb(IO_ICU1, 0x0a); /* Read IRR by default. */
#ifdef REORDER_IRQ
outb(IO_ICU1, 0xc0 | (3 - 1)); /* pri order 3-7, 0-2 (com2 first) */
#endif
outb(IO_ICU2, 0x11); /* reset; program device, four bytes */
outb(IO_ICU2+1, ICU_OFFSET+8); /* staring at this vector index */
outb(IO_ICU2+1, IRQ_SLAVE);
#ifdef AUTO_EOI_2
outb(IO_ICU2+1, 2 | 1); /* auto EOI, 8086 mode */
#else
outb(IO_ICU2+1, 1); /* 8086 mode */
#endif
outb(IO_ICU2+1, 0xff); /* leave interrupts masked */
outb(IO_ICU2, 0x68); /* special mask mode (if available) */
outb(IO_ICU2, 0x0a); /* Read IRR by default. */
}
void
init_x86_64(paddr_t first_avail)
{
extern void consinit(void);
extern struct extent *iomem_ex;
struct region_descriptor region;
struct mem_segment_descriptor *ldt_segp;
int x, first16q, ist;
u_int64_t seg_start, seg_end;
u_int64_t seg_start1, seg_end1;
cpu_init_msrs(&cpu_info_primary);
proc0.p_addr = proc0paddr;
cpu_info_primary.ci_curpcb = &proc0.p_addr->u_pcb;
x86_bus_space_init();
consinit(); /* XXX SHOULD NOT BE DONE HERE */
/*
* Initailize PAGE_SIZE-dependent variables.
*/
uvm_setpagesize();
#if 0
uvmexp.ncolors = 2;
#endif
/*
* Boot arguments are in a single page specified by /boot.
*
* We require the "new" vector form, as well as memory ranges
* to be given in bytes rather than KB.
*
* locore copies the data into bootinfo[] for us.
*/
if ((bootapiver & (BAPIV_VECTOR | BAPIV_BMEMMAP)) ==
(BAPIV_VECTOR | BAPIV_BMEMMAP)) {
if (bootinfo_size >= sizeof(bootinfo))
panic("boot args too big");
getbootinfo(bootinfo, bootinfo_size);
} else
panic("invalid /boot");
avail_start = PAGE_SIZE; /* BIOS leaves data in low memory */
/* and VM system doesn't work with phys 0 */
#ifdef MULTIPROCESSOR
if (avail_start < MP_TRAMPOLINE + PAGE_SIZE)
avail_start = MP_TRAMPOLINE + PAGE_SIZE;
#endif
/*
* Call pmap initialization to make new kernel address space.
* We must do this before loading pages into the VM system.
*/
pmap_bootstrap(VM_MIN_KERNEL_ADDRESS,
IOM_END + trunc_page(KBTOB(biosextmem)));
if (avail_start != PAGE_SIZE)
pmap_prealloc_lowmem_ptps();
if (mem_cluster_cnt == 0) {
/*
* Allocate the physical addresses used by RAM from the iomem
* extent map. This is done before the addresses are
* page rounded just to make sure we get them all.
*/
if (extent_alloc_region(iomem_ex, 0, KBTOB(biosbasemem),
EX_NOWAIT)) {
/* XXX What should we do? */
printf("WARNING: CAN'T ALLOCATE BASE MEMORY FROM "
"IOMEM EXTENT MAP!\n");
}
mem_clusters[0].start = 0;
mem_clusters[0].size = trunc_page(KBTOB(biosbasemem));
physmem += atop(mem_clusters[0].size);
if (extent_alloc_region(iomem_ex, IOM_END, KBTOB(biosextmem),
EX_NOWAIT)) {
/* XXX What should we do? */
printf("WARNING: CAN'T ALLOCATE EXTENDED MEMORY FROM "
"IOMEM EXTENT MAP!\n");
}
#if 0
#if NISADMA > 0
/*
* Some motherboards/BIOSes remap the 384K of RAM that would
* normally be covered by the ISA hole to the end of memory
* so that it can be used. However, on a 16M system, this
* would cause bounce buffers to be allocated and used.
* This is not desirable behaviour, as more than 384K of
* bounce buffers might be allocated. As a work-around,
* we round memory down to the nearest 1M boundary if
* we're using any isadma devices and the remapped memory
* is what puts us over 16M.
*/
if (biosextmem > (15*1024) && biosextmem < (16*1024)) {
char pbuf[9];
format_bytes(pbuf, sizeof(pbuf),
biosextmem - (15*1024));
printf("Warning: ignoring %s of remapped memory\n",
pbuf);
biosextmem = (15*1024);
}
#endif
#endif
mem_clusters[1].start = IOM_END;
mem_clusters[1].size = trunc_page(KBTOB(biosextmem));
physmem += atop(mem_clusters[1].size);
mem_cluster_cnt = 2;
avail_end = IOM_END + trunc_page(KBTOB(biosextmem));
}
/*
* If we have 16M of RAM or less, just put it all on
* the default free list. Otherwise, put the first
* 16M of RAM on a lower priority free list (so that
* all of the ISA DMA'able memory won't be eaten up
* first-off).
*/
if (avail_end <= (16 * 1024 * 1024))
first16q = VM_FREELIST_DEFAULT;
else
first16q = VM_FREELIST_FIRST16;
/* Make sure the end of the space used by the kernel is rounded. */
first_avail = round_page(first_avail);
kern_end = KERNBASE + first_avail;
/*
* Now, load the memory clusters (which have already been
* rounded and truncated) into the VM system.
*
* NOTE: WE ASSUME THAT MEMORY STARTS AT 0 AND THAT THE KERNEL
* IS LOADED AT IOM_END (1M).
*/
for (x = 0; x < mem_cluster_cnt; x++) {
seg_start = mem_clusters[x].start;
seg_end = mem_clusters[x].start + mem_clusters[x].size;
seg_start1 = 0;
seg_end1 = 0;
if (seg_start > 0xffffffffULL) {
printf("skipping %lld bytes of memory above 4GB\n",
seg_end - seg_start);
continue;
}
if (seg_end > 0x100000000ULL) {
printf("skipping %lld bytes of memory above 4GB\n",
seg_end - 0x100000000ULL);
seg_end = 0x100000000ULL;
}
/*
* Skip memory before our available starting point.
*/
if (seg_end <= avail_start)
continue;
if (avail_start >= seg_start && avail_start < seg_end) {
if (seg_start != 0)
panic("init_x86_64: memory doesn't start at 0");
seg_start = avail_start;
if (seg_start == seg_end)
continue;
}
/*
* If this segment contains the kernel, split it
* in two, around the kernel.
*/
if (seg_start <= IOM_END && first_avail <= seg_end) {
seg_start1 = first_avail;
seg_end1 = seg_end;
seg_end = IOM_END;
}
/* First hunk */
if (seg_start != seg_end) {
if (seg_start <= (16 * 1024 * 1024) &&
first16q != VM_FREELIST_DEFAULT) {
u_int64_t tmp;
if (seg_end > (16 * 1024 * 1024))
tmp = (16 * 1024 * 1024);
else
tmp = seg_end;
#if DEBUG_MEMLOAD
printf("loading 0x%qx-0x%qx (0x%lx-0x%lx)\n",
(unsigned long long)seg_start,
(unsigned long long)tmp,
atop(seg_start), atop(tmp));
#endif
uvm_page_physload(atop(seg_start),
atop(tmp), atop(seg_start),
atop(tmp), first16q);
seg_start = tmp;
}
if (seg_start != seg_end) {
#if DEBUG_MEMLOAD
printf("loading 0x%qx-0x%qx (0x%lx-0x%lx)\n",
(unsigned long long)seg_start,
(unsigned long long)seg_end,
atop(seg_start), atop(seg_end));
#endif
uvm_page_physload(atop(seg_start),
atop(seg_end), atop(seg_start),
atop(seg_end), VM_FREELIST_DEFAULT);
}
}
/* Second hunk */
if (seg_start1 != seg_end1) {
if (seg_start1 <= (16 * 1024 * 1024) &&
first16q != VM_FREELIST_DEFAULT) {
u_int64_t tmp;
if (seg_end1 > (16 * 1024 * 1024))
tmp = (16 * 1024 * 1024);
else
tmp = seg_end1;
#if DEBUG_MEMLOAD
printf("loading 0x%qx-0x%qx (0x%lx-0x%lx)\n",
(unsigned long long)seg_start1,
(unsigned long long)tmp,
atop(seg_start1), atop(tmp));
#endif
uvm_page_physload(atop(seg_start1),
atop(tmp), atop(seg_start1),
atop(tmp), first16q);
seg_start1 = tmp;
}
if (seg_start1 != seg_end1) {
#if DEBUG_MEMLOAD
printf("loading 0x%qx-0x%qx (0x%lx-0x%lx)\n",
(unsigned long long)seg_start1,
(unsigned long long)seg_end1,
atop(seg_start1), atop(seg_end1));
#endif
uvm_page_physload(atop(seg_start1),
atop(seg_end1), atop(seg_start1),
atop(seg_end1), VM_FREELIST_DEFAULT);
}
}
}
/*
* Steal memory for the message buffer (at end of core).
*/
{
struct vm_physseg *vps = NULL;
psize_t sz = round_page(MSGBUFSIZE);
psize_t reqsz = sz;
for (x = 0; x < vm_nphysseg; x++) {
vps = &vm_physmem[x];
if (ptoa(vps->avail_end) == avail_end)
break;
}
if (x == vm_nphysseg)
panic("init_x86_64: can't find end of memory");
/* Shrink so it'll fit in the last segment. */
if ((vps->avail_end - vps->avail_start) < atop(sz))
sz = ptoa(vps->avail_end - vps->avail_start);
vps->avail_end -= atop(sz);
vps->end -= atop(sz);
msgbuf_paddr = ptoa(vps->avail_end);
/* Remove the last segment if it now has no pages. */
if (vps->start == vps->end) {
for (vm_nphysseg--; x < vm_nphysseg; x++)
vm_physmem[x] = vm_physmem[x + 1];
}
/* Now find where the new avail_end is. */
for (avail_end = 0, x = 0; x < vm_nphysseg; x++)
if (vm_physmem[x].avail_end > avail_end)
avail_end = vm_physmem[x].avail_end;
avail_end = ptoa(avail_end);
/* Warn if the message buffer had to be shrunk. */
if (sz != reqsz)
printf("WARNING: %ld bytes not available for msgbuf "
"in last cluster (%ld used)\n", reqsz, sz);
}
/*
* XXXfvdl todo: acpi wakeup code.
*/
pmap_growkernel(VM_MIN_KERNEL_ADDRESS + 32 * 1024 * 1024);
pmap_kenter_pa(idt_vaddr, idt_paddr, VM_PROT_READ|VM_PROT_WRITE);
pmap_kenter_pa(idt_vaddr + PAGE_SIZE, idt_paddr + PAGE_SIZE,
VM_PROT_READ|VM_PROT_WRITE);
pmap_kenter_pa(lo32_vaddr, lo32_paddr, VM_PROT_READ|VM_PROT_WRITE);
idt = (struct gate_descriptor *)idt_vaddr;
gdtstore = (char *)(idt + NIDT);
ldtstore = gdtstore + DYNSEL_START;
/* make gdt gates and memory segments */
set_mem_segment(GDT_ADDR_MEM(gdtstore, GCODE_SEL), 0, 0xfffff, SDT_MEMERA,
SEL_KPL, 1, 0, 1);
set_mem_segment(GDT_ADDR_MEM(gdtstore, GDATA_SEL), 0, 0xfffff, SDT_MEMRWA,
SEL_KPL, 1, 0, 1);
set_sys_segment(GDT_ADDR_SYS(gdtstore, GLDT_SEL), ldtstore, LDT_SIZE - 1,
SDT_SYSLDT, SEL_KPL, 0);
set_mem_segment(GDT_ADDR_MEM(gdtstore, GUCODE_SEL), 0,
atop(VM_MAXUSER_ADDRESS) - 1, SDT_MEMERA, SEL_UPL, 1, 0, 1);
set_mem_segment(GDT_ADDR_MEM(gdtstore, GUDATA_SEL), 0,
atop(VM_MAXUSER_ADDRESS) - 1, SDT_MEMRWA, SEL_UPL, 1, 0, 1);
/* make ldt gates and memory segments */
setgate((struct gate_descriptor *)(ldtstore + LSYS5CALLS_SEL),
&IDTVEC(oosyscall), 0, SDT_SYS386CGT, SEL_UPL,
GSEL(GCODE_SEL, SEL_KPL));
*(struct mem_segment_descriptor *)(ldtstore + LUCODE_SEL) =
*GDT_ADDR_MEM(gdtstore, GUCODE_SEL);
*(struct mem_segment_descriptor *)(ldtstore + LUDATA_SEL) =
*GDT_ADDR_MEM(gdtstore, GUDATA_SEL);
/*
* 32 bit GDT entries.
*/
set_mem_segment(GDT_ADDR_MEM(gdtstore, GUCODE32_SEL), 0,
atop(VM_MAXUSER_ADDRESS) - 1, SDT_MEMERA, SEL_UPL, 1, 1, 0);
set_mem_segment(GDT_ADDR_MEM(gdtstore, GUDATA32_SEL), 0,
atop(VM_MAXUSER_ADDRESS) - 1, SDT_MEMRWA, SEL_UPL, 1, 1, 0);
/*
* 32 bit LDT entries.
*/
ldt_segp = (struct mem_segment_descriptor *)(ldtstore + LUCODE32_SEL);
set_mem_segment(ldt_segp, 0, atop(VM_MAXUSER_ADDRESS32) - 1,
SDT_MEMERA, SEL_UPL, 1, 1, 0);
ldt_segp = (struct mem_segment_descriptor *)(ldtstore + LUDATA32_SEL);
set_mem_segment(ldt_segp, 0, atop(VM_MAXUSER_ADDRESS32) - 1,
SDT_MEMRWA, SEL_UPL, 1, 1, 0);
/*
* Other entries.
*/
memcpy((struct gate_descriptor *)(ldtstore + LSOL26CALLS_SEL),
(struct gate_descriptor *)(ldtstore + LSYS5CALLS_SEL),
sizeof (struct gate_descriptor));
memcpy((struct gate_descriptor *)(ldtstore + LBSDICALLS_SEL),
(struct gate_descriptor *)(ldtstore + LSYS5CALLS_SEL),
sizeof (struct gate_descriptor));
/* exceptions */
for (x = 0; x < 32; x++) {
ist = (x == 8) ? 1 : 0;
setgate(&idt[x], IDTVEC(exceptions)[x], ist, SDT_SYS386IGT,
(x == 3 || x == 4) ? SEL_UPL : SEL_KPL,
GSEL(GCODE_SEL, SEL_KPL));
idt_allocmap[x] = 1;
}
/* new-style interrupt gate for syscalls */
setgate(&idt[128], &IDTVEC(osyscall), 0, SDT_SYS386IGT, SEL_UPL,
GSEL(GCODE_SEL, SEL_KPL));
idt_allocmap[128] = 1;
setregion(®ion, gdtstore, DYNSEL_START - 1);
lgdt(®ion);
cpu_init_idt();
#ifdef DDB
db_machine_init();
ddb_init();
if (boothowto & RB_KDB)
Debugger();
#endif
#ifdef KGDB
kgdb_port_init();
if (boothowto & RB_KDB) {
kgdb_debug_init = 1;
kgdb_connect(1);
}
#endif
intr_default_setup();
softintr_init();
splraise(IPL_IPI);
enable_intr();
/* Make sure maxproc is sane */
if (maxproc > cpu_maxproc())
maxproc = cpu_maxproc();
}