int
fork1(struct thread *td, struct fork_req *fr)
{
struct proc *p1, *newproc;
struct thread *td2;
struct vmspace *vm2;
struct file *fp_procdesc;
vm_ooffset_t mem_charged;
int error, nprocs_new, ok;
static int curfail;
static struct timeval lastfail;
int flags, pages;
flags = fr->fr_flags;
pages = fr->fr_pages;
if ((flags & RFSTOPPED) != 0)
MPASS(fr->fr_procp != NULL && fr->fr_pidp == NULL);
else
MPASS(fr->fr_procp == NULL);
/* Check for the undefined or unimplemented flags. */
if ((flags & ~(RFFLAGS | RFTSIGFLAGS(RFTSIGMASK))) != 0)
return (EINVAL);
/* Signal value requires RFTSIGZMB. */
if ((flags & RFTSIGFLAGS(RFTSIGMASK)) != 0 && (flags & RFTSIGZMB) == 0)
return (EINVAL);
/* Can't copy and clear. */
if ((flags & (RFFDG|RFCFDG)) == (RFFDG|RFCFDG))
return (EINVAL);
/* Check the validity of the signal number. */
if ((flags & RFTSIGZMB) != 0 && (u_int)RFTSIGNUM(flags) > _SIG_MAXSIG)
return (EINVAL);
if ((flags & RFPROCDESC) != 0) {
/* Can't not create a process yet get a process descriptor. */
if ((flags & RFPROC) == 0)
return (EINVAL);
/* Must provide a place to put a procdesc if creating one. */
if (fr->fr_pd_fd == NULL)
return (EINVAL);
/* Check if we are using supported flags. */
if ((fr->fr_pd_flags & ~PD_ALLOWED_AT_FORK) != 0)
return (EINVAL);
}
p1 = td->td_proc;
/*
* Here we don't create a new process, but we divorce
* certain parts of a process from itself.
*/
if ((flags & RFPROC) == 0) {
if (fr->fr_procp != NULL)
*fr->fr_procp = NULL;
else if (fr->fr_pidp != NULL)
*fr->fr_pidp = 0;
return (fork_norfproc(td, flags));
}
fp_procdesc = NULL;
newproc = NULL;
vm2 = NULL;
/*
* Increment the nprocs resource before allocations occur.
* Although process entries are dynamically created, we still
* keep a global limit on the maximum number we will
* create. There are hard-limits as to the number of processes
* that can run, established by the KVA and memory usage for
* the process data.
*
* Don't allow a nonprivileged user to use the last ten
* processes; don't let root exceed the limit.
*/
nprocs_new = atomic_fetchadd_int(&nprocs, 1) + 1;
if ((nprocs_new >= maxproc - 10 && priv_check_cred(td->td_ucred,
PRIV_MAXPROC, 0) != 0) || nprocs_new >= maxproc) {
error = EAGAIN;
sx_xlock(&allproc_lock);
if (ppsratecheck(&lastfail, &curfail, 1)) {
printf("maxproc limit exceeded by uid %u (pid %d); "
"see tuning(7) and login.conf(5)\n",
td->td_ucred->cr_ruid, p1->p_pid);
}
sx_xunlock(&allproc_lock);
goto fail2;
}
/*
* If required, create a process descriptor in the parent first; we
* will abandon it if something goes wrong. We don't finit() until
* later.
*/
if (flags & RFPROCDESC) {
error = procdesc_falloc(td, &fp_procdesc, fr->fr_pd_fd,
fr->fr_pd_flags, fr->fr_pd_fcaps);
if (error != 0)
goto fail2;
}
mem_charged = 0;
if (pages == 0)
pages = kstack_pages;
/* Allocate new proc. */
newproc = uma_zalloc(proc_zone, M_WAITOK);
td2 = FIRST_THREAD_IN_PROC(newproc);
if (td2 == NULL) {
td2 = thread_alloc(pages);
if (td2 == NULL) {
error = ENOMEM;
goto fail2;
}
proc_linkup(newproc, td2);
} else {
if (td2->td_kstack == 0 || td2->td_kstack_pages != pages) {
if (td2->td_kstack != 0)
vm_thread_dispose(td2);
if (!thread_alloc_stack(td2, pages)) {
error = ENOMEM;
goto fail2;
}
}
}
if ((flags & RFMEM) == 0) {
vm2 = vmspace_fork(p1->p_vmspace, &mem_charged);
if (vm2 == NULL) {
error = ENOMEM;
goto fail2;
}
if (!swap_reserve(mem_charged)) {
/*
* The swap reservation failed. The accounting
* from the entries of the copied vm2 will be
* subtracted in vmspace_free(), so force the
* reservation there.
*/
swap_reserve_force(mem_charged);
error = ENOMEM;
goto fail2;
}
} else
vm2 = NULL;
/*
* XXX: This is ugly; when we copy resource usage, we need to bump
* per-cred resource counters.
*/
proc_set_cred_init(newproc, crhold(td->td_ucred));
/*
* Initialize resource accounting for the child process.
*/
error = racct_proc_fork(p1, newproc);
if (error != 0) {
error = EAGAIN;
goto fail1;
}
#ifdef MAC
mac_proc_init(newproc);
#endif
newproc->p_klist = knlist_alloc(&newproc->p_mtx);
STAILQ_INIT(&newproc->p_ktr);
/* We have to lock the process tree while we look for a pid. */
sx_slock(&proctree_lock);
sx_xlock(&allproc_lock);
/*
* Increment the count of procs running with this uid. Don't allow
* a nonprivileged user to exceed their current limit.
*
* XXXRW: Can we avoid privilege here if it's not needed?
*/
error = priv_check_cred(td->td_ucred, PRIV_PROC_LIMIT, 0);
if (error == 0)
ok = chgproccnt(td->td_ucred->cr_ruidinfo, 1, 0);
else {
ok = chgproccnt(td->td_ucred->cr_ruidinfo, 1,
lim_cur(td, RLIMIT_NPROC));
}
if (ok) {
do_fork(td, fr, newproc, td2, vm2, fp_procdesc);
return (0);
}
error = EAGAIN;
sx_sunlock(&proctree_lock);
sx_xunlock(&allproc_lock);
#ifdef MAC
mac_proc_destroy(newproc);
#endif
racct_proc_exit(newproc);
fail1:
crfree(newproc->p_ucred);
newproc->p_ucred = NULL;
fail2:
if (vm2 != NULL)
vmspace_free(vm2);
uma_zfree(proc_zone, newproc);
if ((flags & RFPROCDESC) != 0 && fp_procdesc != NULL) {
fdclose(td, fp_procdesc, *fr->fr_pd_fd);
fdrop(fp_procdesc, td);
}
atomic_add_int(&nprocs, -1);
pause("fork", hz / 2);
return (error);
}
void *
initarm(void *arg, void *arg2)
{
struct pv_addr kernel_l1pt;
int loop;
u_int l1pagetable;
vm_offset_t freemempos;
vm_offset_t afterkern;
vm_offset_t lastaddr;
int i;
uint32_t memsize;
i = 0;
boothowto = 0;
set_cpufuncs();
cpufuncs.cf_sleep = s3c24x0_sleep;
lastaddr = fake_preload_metadata();
pcpu_init(pcpup, 0, sizeof(struct pcpu));
PCPU_SET(curthread, &thread0);
/* Do basic tuning, hz etc */
init_param1();
#define KERNEL_TEXT_BASE (KERNBASE)
freemempos = (lastaddr + PAGE_MASK) & ~PAGE_MASK;
/* Define a macro to simplify memory allocation */
#define valloc_pages(var, np) \
alloc_pages((var).pv_va, (np)); \
(var).pv_pa = (var).pv_va + (KERNPHYSADDR - KERNVIRTADDR);
#define alloc_pages(var, np) \
(var) = freemempos; \
freemempos += (np * PAGE_SIZE); \
memset((char *)(var), 0, ((np) * PAGE_SIZE));
while (((freemempos - L1_TABLE_SIZE) & (L1_TABLE_SIZE - 1)) != 0)
freemempos += PAGE_SIZE;
valloc_pages(kernel_l1pt, L1_TABLE_SIZE / PAGE_SIZE);
for (loop = 0; loop < NUM_KERNEL_PTS; ++loop) {
if (!(loop % (PAGE_SIZE / L2_TABLE_SIZE_REAL))) {
valloc_pages(kernel_pt_table[loop],
L2_TABLE_SIZE / PAGE_SIZE);
} else {
kernel_pt_table[loop].pv_va = freemempos -
(loop % (PAGE_SIZE / L2_TABLE_SIZE_REAL)) *
L2_TABLE_SIZE_REAL;
kernel_pt_table[loop].pv_pa =
kernel_pt_table[loop].pv_va - KERNVIRTADDR +
KERNPHYSADDR;
}
}
/*
* Allocate a page for the system page mapped to V0x00000000
* This page will just contain the system vectors and can be
* shared by all processes.
*/
valloc_pages(systempage, 1);
/* Allocate stacks for all modes */
valloc_pages(irqstack, IRQ_STACK_SIZE);
valloc_pages(abtstack, ABT_STACK_SIZE);
valloc_pages(undstack, UND_STACK_SIZE);
valloc_pages(kernelstack, KSTACK_PAGES);
valloc_pages(msgbufpv, round_page(msgbufsize) / PAGE_SIZE);
/*
* Now we start construction of the L1 page table
* We start by mapping the L2 page tables into the L1.
* This means that we can replace L1 mappings later on if necessary
*/
l1pagetable = kernel_l1pt.pv_va;
/* Map the L2 pages tables in the L1 page table */
pmap_link_l2pt(l1pagetable, ARM_VECTORS_HIGH,
&kernel_pt_table[KERNEL_PT_SYS]);
for (i = 0; i < KERNEL_PT_KERN_NUM; i++)
pmap_link_l2pt(l1pagetable, KERNBASE + i * L1_S_SIZE,
&kernel_pt_table[KERNEL_PT_KERN + i]);
pmap_map_chunk(l1pagetable, KERNBASE, PHYSADDR,
(((uint32_t)(lastaddr) - KERNBASE) + PAGE_SIZE) & ~(PAGE_SIZE - 1),
VM_PROT_READ|VM_PROT_WRITE, PTE_CACHE);
afterkern = round_page((lastaddr + L1_S_SIZE) & ~(L1_S_SIZE
- 1));
for (i = 0; i < KERNEL_PT_AFKERNEL_NUM; i++) {
pmap_link_l2pt(l1pagetable, afterkern + i * L1_S_SIZE,
&kernel_pt_table[KERNEL_PT_AFKERNEL + i]);
}
/* Map the vector page. */
pmap_map_entry(l1pagetable, ARM_VECTORS_HIGH, systempage.pv_pa,
VM_PROT_READ|VM_PROT_WRITE, PTE_CACHE);
/* Map the stack pages */
pmap_map_chunk(l1pagetable, irqstack.pv_va, irqstack.pv_pa,
IRQ_STACK_SIZE * PAGE_SIZE, VM_PROT_READ|VM_PROT_WRITE, PTE_CACHE);
pmap_map_chunk(l1pagetable, abtstack.pv_va, abtstack.pv_pa,
ABT_STACK_SIZE * PAGE_SIZE, VM_PROT_READ|VM_PROT_WRITE, PTE_CACHE);
pmap_map_chunk(l1pagetable, undstack.pv_va, undstack.pv_pa,
UND_STACK_SIZE * PAGE_SIZE, VM_PROT_READ|VM_PROT_WRITE, PTE_CACHE);
pmap_map_chunk(l1pagetable, kernelstack.pv_va, kernelstack.pv_pa,
KSTACK_PAGES * PAGE_SIZE, VM_PROT_READ|VM_PROT_WRITE, PTE_CACHE);
pmap_map_chunk(l1pagetable, kernel_l1pt.pv_va, kernel_l1pt.pv_pa,
L1_TABLE_SIZE, VM_PROT_READ|VM_PROT_WRITE, PTE_PAGETABLE);
pmap_map_chunk(l1pagetable, msgbufpv.pv_va, msgbufpv.pv_pa,
msgbufsize, VM_PROT_READ|VM_PROT_WRITE, PTE_CACHE);
for (loop = 0; loop < NUM_KERNEL_PTS; ++loop) {
pmap_map_chunk(l1pagetable, kernel_pt_table[loop].pv_va,
kernel_pt_table[loop].pv_pa, L2_TABLE_SIZE,
VM_PROT_READ|VM_PROT_WRITE, PTE_PAGETABLE);
}
pmap_devmap_bootstrap(l1pagetable, s3c24x0_devmap);
cpu_domains((DOMAIN_CLIENT << (PMAP_DOMAIN_KERNEL*2)) | DOMAIN_CLIENT);
setttb(kernel_l1pt.pv_pa);
cpu_tlb_flushID();
cpu_domains(DOMAIN_CLIENT << (PMAP_DOMAIN_KERNEL*2));
/*
* Pages were allocated during the secondary bootstrap for the
* stacks for different CPU modes.
* We must now set the r13 registers in the different CPU modes to
* point to these stacks.
* Since the ARM stacks use STMFD etc. we must set r13 to the top end
* of the stack memory.
*/
cpu_control(CPU_CONTROL_MMU_ENABLE, CPU_CONTROL_MMU_ENABLE);
set_stackptr(PSR_IRQ32_MODE,
irqstack.pv_va + IRQ_STACK_SIZE * PAGE_SIZE);
set_stackptr(PSR_ABT32_MODE,
abtstack.pv_va + ABT_STACK_SIZE * PAGE_SIZE);
set_stackptr(PSR_UND32_MODE,
undstack.pv_va + UND_STACK_SIZE * PAGE_SIZE);
/*
* We must now clean the cache again....
* Cleaning may be done by reading new data to displace any
* dirty data in the cache. This will have happened in setttb()
* but since we are boot strapping the addresses used for the read
* may have just been remapped and thus the cache could be out
* of sync. A re-clean after the switch will cure this.
* After booting there are no gross reloations of the kernel thus
* this problem will not occur after initarm().
*/
cpu_idcache_wbinv_all();
/* Disable all peripheral interrupts */
ioreg_write32(S3C24X0_INTCTL_BASE + INTCTL_INTMSK, ~0);
memsize = board_init();
/* Find pclk for uart */
switch(ioreg_read32(S3C24X0_GPIO_BASE + GPIO_GSTATUS1) >> 16) {
case 0x3241:
s3c2410_clock_freq2(S3C24X0_CLKMAN_BASE, NULL, NULL,
&s3c2410_pclk);
break;
case 0x3244:
s3c2440_clock_freq2(S3C24X0_CLKMAN_BASE, NULL, NULL,
&s3c2410_pclk);
break;
}
cninit();
/* Set stack for exception handlers */
data_abort_handler_address = (u_int)data_abort_handler;
prefetch_abort_handler_address = (u_int)prefetch_abort_handler;
undefined_handler_address = (u_int)undefinedinstruction_bounce;
undefined_init();
proc_linkup(&proc0, &thread0);
thread0.td_kstack = kernelstack.pv_va;
thread0.td_pcb = (struct pcb *)
(thread0.td_kstack + KSTACK_PAGES * PAGE_SIZE) - 1;
thread0.td_pcb->pcb_flags = 0;
thread0.td_frame = &proc0_tf;
pcpup->pc_curpcb = thread0.td_pcb;
arm_vector_init(ARM_VECTORS_HIGH, ARM_VEC_ALL);
pmap_curmaxkvaddr = afterkern + 0x100000 * (KERNEL_PT_KERN_NUM - 1);
/*
* ARM_USE_SMALL_ALLOC uses dump_avail, so it must be filled before
* calling pmap_bootstrap.
*/
dump_avail[0] = PHYSADDR;
dump_avail[1] = PHYSADDR + memsize;
dump_avail[2] = 0;
dump_avail[3] = 0;
pmap_bootstrap(freemempos,
KERNVIRTADDR + 3 * memsize,
&kernel_l1pt);
msgbufp = (void*)msgbufpv.pv_va;
msgbufinit(msgbufp, msgbufsize);
mutex_init();
physmem = memsize / PAGE_SIZE;
phys_avail[0] = virtual_avail - KERNVIRTADDR + KERNPHYSADDR;
phys_avail[1] = PHYSADDR + memsize;
phys_avail[2] = 0;
phys_avail[3] = 0;
init_param2(physmem);
kdb_init();
return ((void *)(kernelstack.pv_va + USPACE_SVC_STACK_TOP -
sizeof(struct pcb)));
}
