/*
* Vnode op for VM getpages.
* Wish wish .... get rid from multiple IO routines
*
* smbfs_getpages(struct vnode *a_vp, vm_page_t *a_m, int a_count,
* int a_reqpage, vm_ooffset_t a_offset)
*/
int
smbfs_getpages(struct vop_getpages_args *ap)
{
#ifdef SMBFS_RWGENERIC
return vop_stdgetpages(ap);
#else
int i, error, npages;
int doclose;
size_t size, toff, nextoff, count;
struct uio uio;
struct iovec iov;
vm_offset_t kva;
struct buf *bp;
struct vnode *vp;
struct thread *td = curthread; /* XXX */
struct ucred *cred;
struct smbmount *smp;
struct smbnode *np;
struct smb_cred scred;
vm_page_t *pages;
KKASSERT(td->td_proc);
vp = ap->a_vp;
cred = td->td_proc->p_ucred;
np = VTOSMB(vp);
smp = VFSTOSMBFS(vp->v_mount);
pages = ap->a_m;
count = (size_t)ap->a_count;
if (vp->v_object == NULL) {
kprintf("smbfs_getpages: called with non-merged cache vnode??\n");
return VM_PAGER_ERROR;
}
smb_makescred(&scred, td, cred);
bp = getpbuf_kva(&smbfs_pbuf_freecnt);
npages = btoc(count);
kva = (vm_offset_t) bp->b_data;
pmap_qenter(kva, pages, npages);
iov.iov_base = (caddr_t) kva;
iov.iov_len = count;
uio.uio_iov = &iov;
uio.uio_iovcnt = 1;
uio.uio_offset = IDX_TO_OFF(pages[0]->pindex);
uio.uio_resid = count;
uio.uio_segflg = UIO_SYSSPACE;
uio.uio_rw = UIO_READ;
uio.uio_td = td;
/*
* This is kinda nasty. Since smbfs is physically closing the
* fid on close(), we have to reopen it if necessary. There are
* other races here too, such as if another process opens the same
* file while we are blocked in read. XXX
*/
error = 0;
doclose = 0;
if (np->n_opencount == 0) {
error = smbfs_smb_open(np, SMB_AM_OPENREAD, &scred);
if (error == 0)
doclose = 1;
}
if (error == 0)
error = smb_read(smp->sm_share, np->n_fid, &uio, &scred);
if (doclose)
smbfs_smb_close(smp->sm_share, np->n_fid, NULL, &scred);
pmap_qremove(kva, npages);
relpbuf(bp, &smbfs_pbuf_freecnt);
if (error && (uio.uio_resid == count)) {
kprintf("smbfs_getpages: error %d\n",error);
for (i = 0; i < npages; i++) {
if (ap->a_reqpage != i)
vnode_pager_freepage(pages[i]);
}
return VM_PAGER_ERROR;
}
size = count - uio.uio_resid;
for (i = 0, toff = 0; i < npages; i++, toff = nextoff) {
vm_page_t m;
nextoff = toff + PAGE_SIZE;
m = pages[i];
m->flags &= ~PG_ZERO;
/*
* NOTE: pmap dirty bit should have already been cleared.
* We do not clear it here.
*/
if (nextoff <= size) {
m->valid = VM_PAGE_BITS_ALL;
m->dirty = 0;
} else {
int nvalid = ((size + DEV_BSIZE - 1) - toff) &
~(DEV_BSIZE - 1);
vm_page_set_validclean(m, 0, nvalid);
}
if (i != ap->a_reqpage) {
/*
* Whether or not to leave the page activated is up in
* the air, but we should put the page on a page queue
* somewhere (it already is in the object). Result:
* It appears that emperical results show that
* deactivating pages is best.
*/
/*
* Just in case someone was asking for this page we
* now tell them that it is ok to use.
*/
if (!error) {
if (m->flags & PG_REFERENCED)
vm_page_activate(m);
else
vm_page_deactivate(m);
vm_page_wakeup(m);
} else {
vnode_pager_freepage(m);
}
}
}
return 0;
#endif /* SMBFS_RWGENERIC */
}
static void
octeon_memory_init(void)
{
vm_paddr_t phys_end;
int64_t addr;
unsigned i, j;
phys_end = round_page(MIPS_KSEG0_TO_PHYS((vm_offset_t)&end));
if (cvmx_sysinfo_get()->board_type == CVMX_BOARD_TYPE_SIM) {
/* Simulator we limit to 96 meg */
phys_avail[0] = phys_end;
phys_avail[1] = 96 << 20;
dump_avail[0] = phys_avail[0];
dump_avail[1] = phys_avail[1];
realmem = physmem = btoc(phys_avail[1] - phys_avail[0]);
return;
}
/*
* Allocate memory from bootmem 1MB at a time and merge
* adjacent entries.
*/
i = 0;
while (i < PHYS_AVAIL_ENTRIES) {
/*
* If there is less than 2MB of memory available in 128-byte
* blocks, do not steal any more memory. We need to leave some
* memory for the command queues to be allocated out of.
*/
if (cvmx_bootmem_available_mem(128) < 2 << 20)
break;
addr = cvmx_bootmem_phy_alloc(1 << 20, phys_end,
~(vm_paddr_t)0, PAGE_SIZE, 0);
if (addr == -1)
break;
/*
* The SDK needs to be able to easily map any memory that might
* come to it e.g. in the form of an mbuf. Because on !n64 we
* can't direct-map some addresses and we don't want to manage
* temporary mappings within the SDK, don't feed memory that
* can't be direct-mapped to the kernel.
*/
#if !defined(__mips_n64)
if (!MIPS_DIRECT_MAPPABLE(addr + (1 << 20) - 1))
continue;
#endif
physmem += btoc(1 << 20);
if (i > 0 && phys_avail[i - 1] == addr) {
phys_avail[i - 1] += 1 << 20;
continue;
}
phys_avail[i + 0] = addr;
phys_avail[i + 1] = addr + (1 << 20);
i += 2;
}
for (j = 0; j < i; j++)
dump_avail[j] = phys_avail[j];
realmem = physmem;
}
/*
* Do all the stuff that locore normally does before calling main().
*/
void
mach_init(long fwhandle, long magic, long bootdata, long reserved)
{
void *kernend, *p0;
u_long first, last;
extern char edata[], end[];
int i;
uint32_t config;
/* XXX this code must run on the target CPU */
config = mips3_cp0_config_read();
config &= ~MIPS3_CONFIG_K0_MASK;
config |= 0x05; /* XXX. cacheable coherent */
mips3_cp0_config_write(config);
/* Zero BSS. XXXCGD: uh, is this really necessary still? */
memset(edata, 0, end - edata);
/*
* Copy the bootinfo structure from the boot loader.
* this has to be done before mips_vector_init is
* called because we may need CFE's TLB handler
*/
if (magic == BOOTINFO_MAGIC)
memcpy(&bootinfo, (struct bootinfo_v1 *)bootdata,
sizeof bootinfo);
else if (reserved == CFE_EPTSEAL) {
magic = BOOTINFO_MAGIC;
bzero(&bootinfo, sizeof bootinfo);
bootinfo.version = BOOTINFO_VERSION;
bootinfo.fwhandle = fwhandle;
bootinfo.fwentry = bootdata;
bootinfo.ssym = (vaddr_t)end;
bootinfo.esym = (vaddr_t)end;
}
kernend = (void *)mips_round_page(end);
#if NKSYMS || defined(DDB) || defined(LKM)
if (magic == BOOTINFO_MAGIC) {
ksym_start = (void *)bootinfo.ssym;
ksym_end = (void *)bootinfo.esym;
kernend = (void *)mips_round_page((vaddr_t)ksym_end);
}
#endif
consinit();
uvm_setpagesize();
/*
* Copy exception-dispatch code down to exception vector.
* Initialize locore-function vector.
* Clear out the I and D caches.
*/
mips_vector_init();
#ifdef DEBUG
printf("fwhandle=%08X magic=%08X bootdata=%08X reserved=%08X\n",
(u_int)fwhandle, (u_int)magic, (u_int)bootdata, (u_int)reserved);
#endif
strcpy(cpu_model, "sb1250");
if (magic == BOOTINFO_MAGIC) {
int idx;
int added;
uint64_t start, len, type;
cfe_init(bootinfo.fwhandle, bootinfo.fwentry);
cfe_present = 1;
idx = 0;
physmem = 0;
mem_cluster_cnt = 0;
while (cfe_enummem(idx, 0, &start, &len, &type) == 0) {
added = 0;
printf("Memory Block #%d start %08"PRIx64"X len %08"PRIx64"X: %s: ",
idx, start, len, (type == CFE_MI_AVAILABLE) ?
"Available" : "Reserved");
if ((type == CFE_MI_AVAILABLE) &&
(mem_cluster_cnt < VM_PHYSSEG_MAX)) {
/*
* XXX Ignore memory above 256MB for now, it
* XXX needs special handling.
*/
if (start < (256*1024*1024)) {
physmem += btoc(((int) len));
mem_clusters[mem_cluster_cnt].start =
(long) start;
mem_clusters[mem_cluster_cnt].size =
(long) len;
mem_cluster_cnt++;
added = 1;
}
}
if (added)
printf("added to map\n");
else
printf("not added to map\n");
idx++;
}
} else {
/*
* Handle the case of not being called from the firmware.
*/
/* XXX hardwire to 32MB; should be kernel config option */
physmem = 32 * 1024 * 1024 / 4096;
mem_clusters[0].start = 0;
mem_clusters[0].size = ctob(physmem);
mem_cluster_cnt = 1;
}
for (i = 0; i < sizeof(bootinfo.boot_flags); i++) {
switch (bootinfo.boot_flags[i]) {
case '\0':
break;
case ' ':
continue;
case '-':
while (bootinfo.boot_flags[i] != ' ' &&
bootinfo.boot_flags[i] != '\0') {
switch (bootinfo.boot_flags[i]) {
case 'a':
boothowto |= RB_ASKNAME;
break;
case 'd':
boothowto |= RB_KDB;
break;
case 's':
boothowto |= RB_SINGLE;
break;
}
i++;
}
}
}
/*
* Load the rest of the available pages into the VM system.
* The first chunk is tricky because we have to avoid the
* kernel, but the rest are easy.
*/
first = round_page(MIPS_KSEG0_TO_PHYS(kernend));
last = mem_clusters[0].start + mem_clusters[0].size;
uvm_page_physload(atop(first), atop(last), atop(first), atop(last),
VM_FREELIST_DEFAULT);
for (i = 1; i < mem_cluster_cnt; i++) {
first = round_page(mem_clusters[i].start);
last = mem_clusters[i].start + mem_clusters[i].size;
uvm_page_physload(atop(first), atop(last), atop(first),
atop(last), VM_FREELIST_DEFAULT);
}
/*
* Initialize error message buffer (at end of core).
*/
mips_init_msgbuf();
/*
* Allocate space for proc0's USPACE
*/
p0 = (void *)pmap_steal_memory(USPACE, NULL, NULL);
lwp0.l_addr = proc0paddr = (struct user *)p0;
lwp0.l_md.md_regs = (struct frame *)((char *)p0 + USPACE) - 1;
proc0paddr->u_pcb.pcb_context[11] =
MIPS_INT_MASK | MIPS_SR_INT_IE; /* SR */
pmap_bootstrap();
/*
* Initialize debuggers, and break into them, if appropriate.
*/
#if NKSYMS || defined(DDB) || defined(LKM)
ksyms_init(((uintptr_t)ksym_end - (uintptr_t)ksym_start),
ksym_start, ksym_end);
#endif
if (boothowto & RB_KDB) {
#if defined(DDB)
Debugger();
#endif
}
}
/*
* This is now called from local media FS's to operate against their
* own vnodes if they fail to implement VOP_PUTPAGES.
*
* This is typically called indirectly via the pageout daemon and
* clustering has already typically occurred, so in general we ask the
* underlying filesystem to write the data out asynchronously rather
* then delayed.
*/
int
vnode_pager_generic_putpages(struct vnode *vp, vm_page_t *ma, int bytecount,
int flags, int *rtvals)
{
int i;
vm_object_t object;
vm_page_t m;
int count;
int maxsize, ncount;
vm_ooffset_t poffset;
struct uio auio;
struct iovec aiov;
int error;
int ioflags;
int ppscheck = 0;
static struct timeval lastfail;
static int curfail;
object = vp->v_object;
count = bytecount / PAGE_SIZE;
for (i = 0; i < count; i++)
rtvals[i] = VM_PAGER_ERROR;
if ((int64_t)ma[0]->pindex < 0) {
printf("vnode_pager_putpages: attempt to write meta-data!!! -- 0x%lx(%lx)\n",
(long)ma[0]->pindex, (u_long)ma[0]->dirty);
rtvals[0] = VM_PAGER_BAD;
return VM_PAGER_BAD;
}
maxsize = count * PAGE_SIZE;
ncount = count;
poffset = IDX_TO_OFF(ma[0]->pindex);
/*
* If the page-aligned write is larger then the actual file we
* have to invalidate pages occurring beyond the file EOF. However,
* there is an edge case where a file may not be page-aligned where
* the last page is partially invalid. In this case the filesystem
* may not properly clear the dirty bits for the entire page (which
* could be VM_PAGE_BITS_ALL due to the page having been mmap()d).
* With the page locked we are free to fix-up the dirty bits here.
*
* We do not under any circumstances truncate the valid bits, as
* this will screw up bogus page replacement.
*/
VM_OBJECT_WLOCK(object);
if (maxsize + poffset > object->un_pager.vnp.vnp_size) {
if (object->un_pager.vnp.vnp_size > poffset) {
int pgoff;
maxsize = object->un_pager.vnp.vnp_size - poffset;
ncount = btoc(maxsize);
if ((pgoff = (int)maxsize & PAGE_MASK) != 0) {
/*
* If the object is locked and the following
* conditions hold, then the page's dirty
* field cannot be concurrently changed by a
* pmap operation.
*/
m = ma[ncount - 1];
vm_page_assert_sbusied(m);
KASSERT(!pmap_page_is_write_mapped(m),
("vnode_pager_generic_putpages: page %p is not read-only", m));
vm_page_clear_dirty(m, pgoff, PAGE_SIZE -
pgoff);
}
} else {
maxsize = 0;
ncount = 0;
}
if (ncount < count) {
for (i = ncount; i < count; i++) {
rtvals[i] = VM_PAGER_BAD;
}
}
}
VM_OBJECT_WUNLOCK(object);
/*
* pageouts are already clustered, use IO_ASYNC to force a bawrite()
* rather then a bdwrite() to prevent paging I/O from saturating
* the buffer cache. Dummy-up the sequential heuristic to cause
* large ranges to cluster. If neither IO_SYNC or IO_ASYNC is set,
* the system decides how to cluster.
*/
ioflags = IO_VMIO;
if (flags & (VM_PAGER_PUT_SYNC | VM_PAGER_PUT_INVAL))
ioflags |= IO_SYNC;
else if ((flags & VM_PAGER_CLUSTER_OK) == 0)
ioflags |= IO_ASYNC;
ioflags |= (flags & VM_PAGER_PUT_INVAL) ? IO_INVAL: 0;
ioflags |= IO_SEQMAX << IO_SEQSHIFT;
aiov.iov_base = (caddr_t) 0;
aiov.iov_len = maxsize;
auio.uio_iov = &aiov;
auio.uio_iovcnt = 1;
auio.uio_offset = poffset;
auio.uio_segflg = UIO_NOCOPY;
auio.uio_rw = UIO_WRITE;
auio.uio_resid = maxsize;
auio.uio_td = (struct thread *) 0;
error = VOP_WRITE(vp, &auio, ioflags, curthread->td_ucred);
PCPU_INC(cnt.v_vnodeout);
PCPU_ADD(cnt.v_vnodepgsout, ncount);
if (error) {
if ((ppscheck = ppsratecheck(&lastfail, &curfail, 1)))
printf("vnode_pager_putpages: I/O error %d\n", error);
}
if (auio.uio_resid) {
if (ppscheck || ppsratecheck(&lastfail, &curfail, 1))
printf("vnode_pager_putpages: residual I/O %zd at %lu\n",
auio.uio_resid, (u_long)ma[0]->pindex);
}
for (i = 0; i < ncount; i++) {
rtvals[i] = VM_PAGER_OK;
}
return rtvals[0];
}
static void
mips_init(void)
{
int i, j;
printf("entry: mips_init()\n");
#ifdef CFE
/*
* Query DRAM memory map from CFE.
*/
physmem = 0;
for (i = 0; i < 10; i += 2) {
int result;
uint64_t addr, len, type;
result = cfe_enummem(i / 2, 0, &addr, &len, &type);
if (result < 0) {
BCM_TRACE("There is no phys memory for: %d\n", i);
phys_avail[i] = phys_avail[i + 1] = 0;
break;
}
if (type != CFE_MI_AVAILABLE) {
BCM_TRACE("phys memory is not available: %d\n", i);
continue;
}
phys_avail[i] = addr;
if (i == 0 && addr == 0) {
/*
* If this is the first physical memory segment probed
* from CFE, omit the region at the start of physical
* memory where the kernel has been loaded.
*/
phys_avail[i] += MIPS_KSEG0_TO_PHYS(kernel_kseg0_end);
}
BCM_TRACE("phys memory is available for: %d\n", i);
BCM_TRACE(" => addr = %jx\n", addr);
BCM_TRACE(" => len = %jd\n", len);
phys_avail[i + 1] = addr + len;
physmem += len;
}
BCM_TRACE("Total phys memory is : %ld\n", physmem);
realmem = btoc(physmem);
#endif
for (j = 0; j < i; j++)
dump_avail[j] = phys_avail[j];
physmem = realmem;
init_param1();
init_param2(physmem);
mips_cpu_init();
pmap_bootstrap();
mips_proc0_init();
mutex_init();
kdb_init();
#ifdef KDB
if (boothowto & RB_KDB)
kdb_enter(KDB_WHY_BOOTFLAGS, "Boot flags requested debugger");
#endif
}
void
platform_start(__register_t a0, __register_t a1, __register_t a2,
__register_t a3)
{
struct bootinfo *bootinfop;
vm_offset_t kernend;
uint64_t platform_counter_freq;
int argc = a0;
char **argv = (char **)a1;
char **envp = (char **)a2;
long memsize;
#ifdef FDT
char buf[2048]; /* early stack supposedly big enough */
vm_offset_t dtbp = 0;
phandle_t chosen;
void *kmdp;
int dtb_needs_swap = 0; /* error */
size_t dtb_size = 0;
#ifndef FDT_DTB_STATIC_ONLY
struct fdt_header *dtb_rom, *dtb;
uint32_t *swapptr;
#endif
int fdt_source = FDT_SOURCE_NONE;
#endif
int i;
/* clear the BSS and SBSS segments */
kernend = (vm_offset_t)&end;
memset(&edata, 0, kernend - (vm_offset_t)(&edata));
mips_postboot_fixup();
mips_pcpu0_init();
/*
* Over time, we've changed out boot-time binary interface for the
* kernel. Miniboot simply provides a 'memsize' in a3, whereas the
* FreeBSD boot loader provides a 'bootinfo *' in a3. While slightly
* grody, we support both here by detecting 'pointer-like' values in
* a3 and assuming physical memory can never be that big.
*
* XXXRW: Pull more values than memsize out of bootinfop -- e.g.,
* module information.
*/
if (a3 >= 0x9800000000000000ULL) {
bootinfop = (void *)a3;
memsize = bootinfop->bi_memsize;
preload_metadata = (caddr_t)bootinfop->bi_modulep;
} else {
bootinfop = NULL;
memsize = a3;
}
kmdp = preload_search_by_type("elf kernel");
/*
* Configure more boot-time parameters passed in by loader.
*/
boothowto = MD_FETCH(kmdp, MODINFOMD_HOWTO, int);
init_static_kenv(MD_FETCH(kmdp, MODINFOMD_ENVP, char *), 0);
#ifdef FDT
#ifndef FDT_DTB_STATIC_ONLY
/*
* Find the dtb passed in by the boot loader (currently fictional).
*
* Prefer a dtb provided as a module to one from bootinfo as we may
* have loaded an alternative one or created a modified version.
*/
dtbp = MD_FETCH(kmdp, MODINFOMD_DTBP, vm_offset_t);
if (dtbp == (vm_offset_t)NULL &&
bootinfop != NULL && bootinfop->bi_dtb != (bi_ptr_t)NULL) {
dtbp = bootinfop->bi_dtb;
fdt_source = FDT_SOURCE_LOADER;
}
/* Try to find an FDT directly in the hardware */
if (dtbp == (vm_offset_t)NULL) {
dtb_rom = (void*)(intptr_t)0x900000007f010000;
if (dtb_rom->magic == FDT_MAGIC) {
dtb_needs_swap = 0;
dtb_size = dtb_rom->totalsize;
} else if (dtb_rom->magic == bswap32(FDT_MAGIC)) {
dtb_needs_swap = 1;
dtb_size = bswap32(dtb_rom->totalsize);
}
if (dtb_size != 0) {
/* Steal a bit of memory... */
dtb = (void *)kernel_kseg0_end;
/* Round alignment from linker script. */
kernel_kseg0_end += roundup2(dtb_size, 64 / 8);
memcpy(dtb, dtb_rom, dtb_size);
if (dtb_needs_swap)
for (swapptr = (uint32_t *)dtb;
swapptr < (uint32_t *)dtb + (dtb_size/sizeof(*dtb));
swapptr++)
*swapptr = bswap32(*swapptr);
dtbp = (vm_offset_t)dtb;
fdt_source = FDT_SOURCE_ROM;
}
}
#endif /* !FDT_DTB_STATIC_ONLY */
#if defined(FDT_DTB_STATIC)
/*
* In case the device tree blob was not retrieved (from metadata) try
* to use the statically embedded one.
*/
if (dtbp == (vm_offset_t)NULL) {
dtbp = (vm_offset_t)&fdt_static_dtb;
fdt_source = FDT_SOURCE_STATIC;
}
#endif
if (OF_install(OFW_FDT, 0) == FALSE)
while (1);
if (OF_init((void *)dtbp) != 0)
while (1);
/*
* Get bootargs from FDT if specified.
*/
chosen = OF_finddevice("/chosen");
if (OF_getprop(chosen, "bootargs", buf, sizeof(buf)) != -1)
_parse_bootargs(buf);
#endif
/*
* XXXRW: We have no way to compare wallclock time to cycle rate on
* BERI, so for now assume we run at the MALTA default (100MHz).
*/
platform_counter_freq = MIPS_DEFAULT_HZ;
mips_timer_early_init(platform_counter_freq);
cninit();
printf("entry: platform_start()\n");
#ifdef FDT
if (dtbp != (vm_offset_t)NULL) {
printf("Using FDT at %p from ", (void *)dtbp);
switch (fdt_source) {
case FDT_SOURCE_LOADER:
printf("loader");
break;
case FDT_SOURCE_ROM:
printf("ROM");
break;
case FDT_SOURCE_STATIC:
printf("kernel");
break;
default:
printf("unknown source %d", fdt_source);
break;
}
printf("\n");
}
if (dtb_size != 0 && dtb_needs_swap)
printf("FDT was byteswapped\n");
#endif
bootverbose = 1;
if (bootverbose) {
printf("cmd line: ");
for (i = 0; i < argc; i++)
printf("%s ", argv[i]);
printf("\n");
printf("envp:\n");
for (i = 0; envp[i]; i += 2)
printf("\t%s = %s\n", envp[i], envp[i+1]);
if (bootinfop != NULL)
printf("bootinfo found at %p\n", bootinfop);
printf("memsize = %p\n", (void *)memsize);
}
realmem = btoc(memsize);
mips_init();
mips_timer_init_params(platform_counter_freq, 0);
}
/*
* cpu_startup: allocate memory for variable-sized tables,
* initialize cpu, and do autoconfiguration.
*/
cpu_startup()
{
register unsigned i;
register caddr_t v, firstaddr;
int base, residual;
vm_offset_t minaddr, maxaddr;
vm_size_t size;
#ifdef BUFFERS_UNMANAGED
vm_offset_t bufmemp;
caddr_t buffermem;
int ix;
#endif
#ifdef DEBUG
extern int pmapdebug;
int opmapdebug = pmapdebug;
pmapdebug = 0;
#endif
/*
* Initialize error message buffer (at end of core).
* avail_end was pre-decremented in pmap_bootstrap to compensate.
*/
for (i = 0; i < btoc(sizeof (struct msgbuf)); i++)
pmap_enter(kernel_pmap, (vm_offset_t)msgbufp,
avail_end + i * NBPG, VM_PROT_ALL, TRUE);
msgbufmapped = 1;
/*
* Good {morning,afternoon,evening,night}.
*/
printf(version);
identifycpu();
printf("real mem = %d\n", ctob(physmem));
/*
* Allocate space for system data structures.
* The first available real memory address is in "firstaddr".
* The first available kernel virtual address is in "v".
* As pages of kernel virtual memory are allocated, "v" is incremented.
* As pages of memory are allocated and cleared,
* "firstaddr" is incremented.
* An index into the kernel page table corresponding to the
* virtual memory address maintained in "v" is kept in "mapaddr".
*/
/*
* Make two passes. The first pass calculates how much memory is
* needed and allocates it. The second pass assigns virtual
* addresses to the various data structures.
*/
firstaddr = 0;
again:
v = (caddr_t)firstaddr;
#define valloc(name, type, num) \
(name) = (type *)v; v = (caddr_t)((name)+(num))
#define valloclim(name, type, num, lim) \
(name) = (type *)v; v = (caddr_t)((lim) = ((name)+(num)))
valloc(cfree, struct cblock, nclist);
valloc(callout, struct callout, ncallout);
valloc(swapmap, struct map, nswapmap = maxproc * 2);
#ifdef SYSVSHM
valloc(shmsegs, struct shmid_ds, shminfo.shmmni);
#endif
/*
* Determine how many buffers to allocate.
* Since HPs tend to be long on memory and short on disk speed,
* we allocate more buffer space than the BSD standard of
* use 10% of memory for the first 2 Meg, 5% of remaining.
* We just allocate a flat 10%. Insure a minimum of 16 buffers.
* We allocate 1/2 as many swap buffer headers as file i/o buffers.
*/
if (bufpages == 0)
bufpages = physmem / 10 / CLSIZE;
if (nbuf == 0) {
nbuf = bufpages;
if (nbuf < 16)
nbuf = 16;
}
if (nswbuf == 0) {
nswbuf = (nbuf / 2) &~ 1; /* force even */
if (nswbuf > 256)
nswbuf = 256; /* sanity */
}
valloc(swbuf, struct buf, nswbuf);
valloc(buf, struct buf, nbuf);
/*
* End of first pass, size has been calculated so allocate memory
*/
if (firstaddr == 0) {
size = (vm_size_t)(v - firstaddr);
firstaddr = (caddr_t) kmem_alloc(kernel_map, round_page(size));
if (firstaddr == 0)
panic("startup: no room for tables");
#ifdef BUFFERS_UNMANAGED
buffermem = (caddr_t) kmem_alloc(kernel_map, bufpages*CLBYTES);
if (buffermem == 0)
panic("startup: no room for buffers");
#endif
goto again;
}
/*
* End of second pass, addresses have been assigned
*/
if ((vm_size_t)(v - firstaddr) != size)
panic("startup: table size inconsistency");
/*
* Now allocate buffers proper. They are different than the above
* in that they usually occupy more virtual memory than physical.
*/
size = MAXBSIZE * nbuf;
buffer_map = kmem_suballoc(kernel_map, (vm_offset_t *)&buffers,
&maxaddr, size, TRUE);
minaddr = (vm_offset_t)buffers;
if (vm_map_find(buffer_map, vm_object_allocate(size), (vm_offset_t)0,
&minaddr, size, FALSE) != KERN_SUCCESS)
panic("startup: cannot allocate buffers");
base = bufpages / nbuf;
residual = bufpages % nbuf;
#ifdef BUFFERS_UNMANAGED
bufmemp = (vm_offset_t) buffermem;
#endif
for (i = 0; i < nbuf; i++) {
vm_size_t curbufsize;
vm_offset_t curbuf;
/*
* First <residual> buffers get (base+1) physical pages
* allocated for them. The rest get (base) physical pages.
*
* The rest of each buffer occupies virtual space,
* but has no physical memory allocated for it.
*/
curbuf = (vm_offset_t)buffers + i * MAXBSIZE;
curbufsize = CLBYTES * (i < residual ? base+1 : base);
#ifdef BUFFERS_UNMANAGED
/*
* Move the physical pages over from buffermem.
*/
for (ix = 0; ix < curbufsize/CLBYTES; ix++) {
vm_offset_t pa;
pa = pmap_extract(kernel_pmap, bufmemp);
if (pa == 0)
panic("startup: unmapped buffer");
pmap_remove(kernel_pmap, bufmemp, bufmemp+CLBYTES);
pmap_enter(kernel_pmap,
(vm_offset_t)(curbuf + ix * CLBYTES),
pa, VM_PROT_READ|VM_PROT_WRITE, TRUE);
bufmemp += CLBYTES;
}
#else
vm_map_pageable(buffer_map, curbuf, curbuf+curbufsize, FALSE);
vm_map_simplify(buffer_map, curbuf);
#endif
}
#ifdef BUFFERS_UNMANAGED
#if 0
/*
* We would like to free the (now empty) original address range
* but too many bad things will happen if we try.
*/
kmem_free(kernel_map, (vm_offset_t)buffermem, bufpages*CLBYTES);
#endif
#endif
/*
* Allocate a submap for exec arguments. This map effectively
* limits the number of processes exec'ing at any time.
*/
exec_map = kmem_suballoc(kernel_map, &minaddr, &maxaddr,
16*NCARGS, TRUE);
/*
* Allocate a submap for physio
*/
phys_map = kmem_suballoc(kernel_map, &minaddr, &maxaddr,
VM_PHYS_SIZE, TRUE);
/*
* Finally, allocate mbuf pool. Since mclrefcnt is an off-size
* we use the more space efficient malloc in place of kmem_alloc.
*/
mclrefcnt = (char *)malloc(NMBCLUSTERS+CLBYTES/MCLBYTES,
M_MBUF, M_NOWAIT);
bzero(mclrefcnt, NMBCLUSTERS+CLBYTES/MCLBYTES);
mb_map = kmem_suballoc(kernel_map, (vm_offset_t *)&mbutl, &maxaddr,
VM_MBUF_SIZE, FALSE);
/*
* Initialize callouts
*/
callfree = callout;
for (i = 1; i < ncallout; i++)
callout[i-1].c_next = &callout[i];
callout[i-1].c_next = NULL;
#ifdef DEBUG
pmapdebug = opmapdebug;
#endif
printf("avail mem = %d\n", ptoa(cnt.v_free_count));
printf("using %d buffers containing %d bytes of memory\n",
nbuf, bufpages * CLBYTES);
/*
* Set up CPU-specific registers, cache, etc.
*/
initcpu();
/*
* Set up buffers, so they can be used to read disk labels.
*/
bufinit();
/*
* Configure the system.
*/
configure();
}
void
mach_init(void)
{
void *kernend;
uint32_t memsize;
extern char edata[], end[]; /* XXX */
/* clear the BSS segment */
kernend = (void *)mips_round_page(end);
memset(edata, 0, (char *)kernend - edata);
/* setup early console */
ingenic_putchar_init();
/* set CPU model info for sysctl_hw */
cpu_setmodel("Ingenic XBurst");
mips_vector_init(NULL, false);
cal_timer();
uvm_setpagesize();
/*
* Look at arguments passed to us and compute boothowto.
*/
boothowto = RB_AUTOBOOT;
#ifdef KADB
boothowto |= RB_KDB;
#endif
/*
* Determine the memory size.
*
* Note: Reserve the first page! That's where the trap
* vectors are located.
*/
memsize = 0x40000000;
printf("Memory size: 0x%08x\n", memsize);
physmem = btoc(memsize);
/*
* memory is at 0x20000000 with first 256MB mirrored to 0x00000000 so
* we can see them through KSEG*
* assume 1GB for now, the SoC can theoretically support up to 3GB
*/
mem_clusters[0].start = PAGE_SIZE;
mem_clusters[0].size = 0x10000000 - PAGE_SIZE;
mem_clusters[1].start = 0x30000000;
mem_clusters[1].size = 0x30000000;
mem_cluster_cnt = 2;
/*
* Load the available pages into the VM system.
*/
mips_page_physload(MIPS_KSEG0_START, (vaddr_t)kernend,
mem_clusters, mem_cluster_cnt, NULL, 0);
/*
* Initialize message buffer (at end of core).
*/
mips_init_msgbuf();
/*
* Initialize the virtual memory system.
*/
pmap_bootstrap();
/*
* Allocate uarea page for lwp0 and set it.
*/
mips_init_lwp0_uarea();
#ifdef MULTIPROCESSOR
mutex_init(&ingenic_ipi_lock, MUTEX_DEFAULT, IPL_HIGH);
mips_locoresw.lsw_send_ipi = ingenic_send_ipi;
mips_locoresw.lsw_cpu_init = ingenic_cpu_init;
#endif
apbus_init();
/*
* Initialize debuggers, and break into them, if appropriate.
*/
#ifdef DDB
if (boothowto & RB_KDB)
Debugger();
#endif
}
void
platform_start(__register_t a0 __unused, __register_t a1 __unused,
__register_t a2 __unused, __register_t a3 __unused)
{
uint64_t platform_counter_freq;
int argc, i;
char **argv, **envp;
vm_offset_t kernend;
/*
* clear the BSS and SBSS segments, this should be first call in
* the function
*/
kernend = (vm_offset_t)&end;
memset(&edata, 0, kernend - (vm_offset_t)(&edata));
mips_postboot_fixup();
/* Initialize pcpu stuff */
mips_pcpu0_init();
argc = a0;
argv = (char**)a1;
envp = (char**)a2;
/*
* Protect ourselves from garbage in registers
*/
if (MIPS_IS_VALID_PTR(envp)) {
for (i = 0; envp[i]; i += 2) {
if (strcmp(envp[i], "memsize") == 0)
realmem = btoc(strtoul(envp[i+1], NULL, 16));
}
}
/*
* Just wild guess. RedBoot let us down and didn't reported
* memory size
*/
if (realmem == 0)
realmem = btoc(32*1024*1024);
/*
* Allow build-time override in case Redboot lies
* or in other situations (eg where there's u-boot)
* where there isn't (yet) a convienent method of
* being told how much RAM is available.
*
* This happens on at least the Ubiquiti LS-SR71A
* board, where redboot says there's 16mb of RAM
* but in fact there's 32mb.
*/
#if defined(AR71XX_REALMEM)
realmem = btoc(AR71XX_REALMEM);
#endif
/* phys_avail regions are in bytes */
phys_avail[0] = MIPS_KSEG0_TO_PHYS(kernel_kseg0_end);
phys_avail[1] = ctob(realmem);
dump_avail[0] = phys_avail[0];
dump_avail[1] = phys_avail[1] - phys_avail[0];
physmem = realmem;
/*
* ns8250 uart code uses DELAY so ticker should be inititalized
* before cninit. And tick_init_params refers to hz, so * init_param1
* should be called first.
*/
init_param1();
/* Detect the system type - this is needed for subsequent chipset-specific calls */
ar71xx_detect_sys_type();
ar71xx_detect_sys_frequency();
platform_counter_freq = ar71xx_cpu_freq();
mips_timer_init_params(platform_counter_freq, 1);
cninit();
init_static_kenv(boot1_env, sizeof(boot1_env));
printf("CPU platform: %s\n", ar71xx_get_system_type());
printf("CPU Frequency=%d MHz\n", u_ar71xx_cpu_freq / 1000000);
printf("CPU DDR Frequency=%d MHz\n", u_ar71xx_ddr_freq / 1000000);
printf("CPU AHB Frequency=%d MHz\n", u_ar71xx_ahb_freq / 1000000);
printf("platform frequency: %lld\n", platform_counter_freq);
printf("arguments: \n");
printf(" a0 = %08x\n", a0);
printf(" a1 = %08x\n", a1);
printf(" a2 = %08x\n", a2);
printf(" a3 = %08x\n", a3);
printf("Cmd line:");
if (MIPS_IS_VALID_PTR(argv)) {
for (i = 0; i < argc; i++) {
printf(" %s", argv[i]);
parse_argv(argv[i]);
}
}
else
printf ("argv is invalid");
printf("\n");
printf("Environment:\n");
if (MIPS_IS_VALID_PTR(envp)) {
for (i = 0; envp[i]; i+=2) {
printf(" %s = %s\n", envp[i], envp[i+1]);
setenv(envp[i], envp[i+1]);
}
}
else
printf ("envp is invalid\n");
/* Redboot if_arge MAC address is in the environment */
ar71xx_redboot_get_macaddr();
init_param2(physmem);
mips_cpu_init();
pmap_bootstrap();
mips_proc0_init();
mutex_init();
/*
* Reset USB devices
*/
ar71xx_init_usb_peripheral();
kdb_init();
#ifdef KDB
if (boothowto & RB_KDB)
kdb_enter(KDB_WHY_BOOTFLAGS, "Boot flags requested debugger");
#endif
}
void
cpu_startup()
{
caddr_t v;
int sz;
vaddr_t minaddr, maxaddr;
extern unsigned int avail_end;
extern char cpu_model[];
/*
* Initialize error message buffer.
*/
initmsgbuf((caddr_t)msgbufp, round_page(MSGBUFSIZE));
/*
* Good {morning,afternoon,evening,night}.
* Also call CPU init on systems that need that.
*/
printf("%s%s [%08X %08X]\n", version, cpu_model, vax_cpudata, vax_siedata);
if (dep_call->cpu_conf)
(*dep_call->cpu_conf)();
printf("real mem = %u (%uMB)\n", avail_end,
avail_end/1024/1024);
physmem = btoc(avail_end);
mtpr(AST_NO, PR_ASTLVL);
spl0();
/*
* Find out how much space we need, allocate it, and then give
* everything true virtual addresses.
*/
sz = (int) allocsys((caddr_t)0);
if ((v = (caddr_t)uvm_km_zalloc(kernel_map, round_page(sz))) == 0)
panic("startup: no room for tables");
if (((unsigned long)allocsys(v) - (unsigned long)v) != sz)
panic("startup: table size inconsistency");
/*
* Determine how many buffers to allocate.
* We allocate bufcachepercent% of memory for buffer space.
*/
if (bufpages == 0)
bufpages = physmem * bufcachepercent / 100;
/* Restrict to at most 25% filled kvm */
if (bufpages >
(VM_MAX_KERNEL_ADDRESS-VM_MIN_KERNEL_ADDRESS) / PAGE_SIZE / 4)
bufpages = (VM_MAX_KERNEL_ADDRESS-VM_MIN_KERNEL_ADDRESS) /
PAGE_SIZE / 4;
/*
* Allocate a submap for exec arguments. This map effectively limits
* the number of processes exec'ing at any time.
*/
minaddr = vm_map_min(kernel_map);
exec_map = uvm_km_suballoc(kernel_map, &minaddr, &maxaddr,
16 * NCARGS, VM_MAP_PAGEABLE, FALSE, NULL);
#if VAX46 || VAX48 || VAX49 || VAX53
/*
* Allocate a submap for physio. This map effectively limits the
* number of processes doing physio at any one time.
*
* Note that machines on which all mass storage I/O controllers
* can perform address translation, do not need this.
*/
if (vax_boardtype == VAX_BTYP_46 || vax_boardtype == VAX_BTYP_48 ||
vax_boardtype == VAX_BTYP_49 || vax_boardtype == VAX_BTYP_1303)
phys_map = uvm_km_suballoc(kernel_map, &minaddr, &maxaddr,
VM_PHYS_SIZE, 0, FALSE, NULL);
#endif
printf("avail mem = %lu (%luMB)\n", ptoa(uvmexp.free),
ptoa(uvmexp.free)/1024/1024);
/*
* Set up buffers, so they can be used to read disk labels.
*/
bufinit();
#ifdef DDB
if (boothowto & RB_KDB)
Debugger();
#endif
/*
* Configure the system.
*/
if (boothowto & RB_CONFIG) {
#ifdef BOOT_CONFIG
user_config();
#else
printf("kernel does not support -c; continuing..\n");
#endif
}
}
static void
mips_init(void)
{
int i, j, cfe_mem_idx, tmp;
uint64_t maxmem;
#ifdef CFE_ENV
cfe_env_init();
#endif
TUNABLE_INT_FETCH("boothowto", &boothowto);
if (boothowto & RB_VERBOSE)
bootverbose++;
#ifdef MAXMEM
tmp = MAXMEM;
#else
tmp = 0;
#endif
TUNABLE_INT_FETCH("hw.physmem", &tmp);
maxmem = (uint64_t)tmp * 1024;
/*
* XXX
* If we used vm_paddr_t consistently in pmap, etc., we could
* use 64-bit page numbers on !n64 systems, too, like i386
* does with PAE.
*/
#if !defined(__mips_n64)
if (maxmem == 0 || maxmem > 0xffffffff)
maxmem = 0xffffffff;
#endif
#ifdef CFE
/*
* Query DRAM memory map from CFE.
*/
physmem = 0;
cfe_mem_idx = 0;
for (i = 0; i < 10; i += 2) {
int result;
uint64_t addr, len, type;
result = cfe_enummem(cfe_mem_idx++, 0, &addr, &len, &type);
if (result < 0) {
phys_avail[i] = phys_avail[i + 1] = 0;
break;
}
KASSERT(type == CFE_MI_AVAILABLE,
("CFE DRAM region is not available?"));
if (bootverbose)
printf("cfe_enummem: 0x%016jx/%ju.\n", addr, len);
if (maxmem != 0) {
if (addr >= maxmem) {
printf("Ignoring %ju bytes of memory at 0x%jx "
"that is above maxmem %dMB\n",
len, addr,
(int)(maxmem / (1024 * 1024)));
continue;
}
if (addr + len > maxmem) {
printf("Ignoring %ju bytes of memory "
"that is above maxmem %dMB\n",
(addr + len) - maxmem,
(int)(maxmem / (1024 * 1024)));
len = maxmem - addr;
}
}
phys_avail[i] = addr;
if (i == 0 && addr == 0) {
/*
* If this is the first physical memory segment probed
* from CFE, omit the region at the start of physical
* memory where the kernel has been loaded.
*/
phys_avail[i] += MIPS_KSEG0_TO_PHYS(kernel_kseg0_end);
}
phys_avail[i + 1] = addr + len;
physmem += len;
}
realmem = btoc(physmem);
#endif
for (j = 0; j < i; j++)
dump_avail[j] = phys_avail[j];
physmem = realmem;
init_param1();
init_param2(physmem);
mips_cpu_init();
/*
* Sibyte has a L1 data cache coherent with DMA. This includes
* on-chip network interfaces as well as PCI/HyperTransport bus
* masters.
*/
cpuinfo.cache_coherent_dma = TRUE;
/*
* XXX
* The kernel is running in 32-bit mode but the CFE is running in
* 64-bit mode. So the SR_KX bit in the status register is turned
* on by the CFE every time we call into it - for e.g. CFE_CONSOLE.
*
* This means that if get a TLB miss for any address above 0xc0000000
* and the SR_KX bit is set then we will end up in the XTLB exception
* vector.
*
* For now work around this by copying the TLB exception handling
* code to the XTLB exception vector.
*/
{
bcopy(MipsTLBMiss, (void *)MIPS_XTLB_MISS_EXC_VEC,
MipsTLBMissEnd - MipsTLBMiss);
mips_icache_sync_all();
mips_dcache_wbinv_all();
}
pmap_bootstrap();
mips_proc0_init();
mutex_init();
kdb_init();
#ifdef KDB
if (boothowto & RB_KDB)
kdb_enter(KDB_WHY_BOOTFLAGS, "Boot flags requested debugger");
#endif
}
void
platform_start(__register_t a0, __register_t a1, __register_t a2,
__register_t a3)
{
struct bootinfo *bootinfop;
vm_offset_t kernend;
uint64_t platform_counter_freq;
int argc = a0;
char **argv = (char **)a1;
char **envp = (char **)a2;
long memsize;
#ifdef FDT
char buf[2048]; /* early stack supposedly big enough */
vm_offset_t dtbp;
phandle_t chosen;
void *kmdp;
#endif
int i;
/* clear the BSS and SBSS segments */
kernend = (vm_offset_t)&end;
memset(&edata, 0, kernend - (vm_offset_t)(&edata));
mips_postboot_fixup();
mips_pcpu0_init();
/*
* Over time, we've changed out boot-time binary interface for the
* kernel. Miniboot simply provides a 'memsize' in a3, whereas the
* FreeBSD boot loader provides a 'bootinfo *' in a3. While slightly
* grody, we support both here by detecting 'pointer-like' values in
* a3 and assuming physical memory can never be that back.
*
* XXXRW: Pull more values than memsize out of bootinfop -- e.g.,
* module information.
*/
if (a3 >= 0x9800000000000000ULL) {
bootinfop = (void *)a3;
memsize = bootinfop->bi_memsize;
preload_metadata = (caddr_t)bootinfop->bi_modulep;
} else {
bootinfop = NULL;
memsize = a3;
}
#ifdef FDT
/*
* Find the dtb passed in by the boot loader (currently fictional).
*/
kmdp = preload_search_by_type("elf kernel");
dtbp = MD_FETCH(kmdp, MODINFOMD_DTBP, vm_offset_t);
#if defined(FDT_DTB_STATIC)
/*
* In case the device tree blob was not retrieved (from metadata) try
* to use the statically embedded one.
*/
if (dtbp == (vm_offset_t)NULL)
dtbp = (vm_offset_t)&fdt_static_dtb;
#else
#error "Non-static FDT not yet supported on BERI"
#endif
if (OF_install(OFW_FDT, 0) == FALSE)
while (1);
if (OF_init((void *)dtbp) != 0)
while (1);
/*
* Configure more boot-time parameters passed in by loader.
*/
boothowto = MD_FETCH(kmdp, MODINFOMD_HOWTO, int);
init_static_kenv(MD_FETCH(kmdp, MODINFOMD_ENVP, char *), 0);
/*
* Get bootargs from FDT if specified.
*/
chosen = OF_finddevice("/chosen");
if (OF_getprop(chosen, "bootargs", buf, sizeof(buf)) != -1)
_parse_bootargs(buf);
#endif
/*
* XXXRW: We have no way to compare wallclock time to cycle rate on
* BERI, so for now assume we run at the MALTA default (100MHz).
*/
platform_counter_freq = MIPS_DEFAULT_HZ;
mips_timer_early_init(platform_counter_freq);
cninit();
printf("entry: platform_start()\n");
bootverbose = 1;
if (bootverbose) {
printf("cmd line: ");
for (i = 0; i < argc; i++)
printf("%s ", argv[i]);
printf("\n");
printf("envp:\n");
for (i = 0; envp[i]; i += 2)
printf("\t%s = %s\n", envp[i], envp[i+1]);
if (bootinfop != NULL)
printf("bootinfo found at %p\n", bootinfop);
printf("memsize = %p\n", (void *)memsize);
}
realmem = btoc(memsize);
mips_init();
mips_timer_init_params(platform_counter_freq, 0);
}
void
platform_start(__register_t a0, __register_t a1, __register_t a2,
__register_t a3)
{
vm_offset_t kernend;
uint64_t platform_counter_freq;
int argc = a0;
char **argv = (char **)a1;
char **envp = (char **)a2;
unsigned int memsize = a3;
int i;
/* clear the BSS and SBSS segments */
kernend = round_page((vm_offset_t)&end);
memset(&edata, 0, kernend - (vm_offset_t)(&edata));
cninit();
printf("entry: platform_start()\n");
bootverbose = 1;
if (bootverbose) {
printf("cmd line: ");
for (i = 0; i < argc; i++)
printf("%s ", argv[i]);
printf("\n");
printf("envp:\n");
for (i = 0; envp[i]; i += 2)
printf("\t%s = %s\n", envp[i], envp[i+1]);
printf("memsize = %08x\n", memsize);
}
realmem = btoc(memsize);
mips_init();
do {
#if defined(TICK_USE_YAMON_FREQ)
/*
* If we are running on a board which uses YAMON firmware,
* then query CPU pipeline clock from the syscon object.
* If unsuccessful, use hard-coded default.
*/
platform_counter_freq = yamon_getcpufreq();
if (platform_counter_freq == 0)
platform_counter_freq = MIPS_DEFAULT_HZ;
#elif defined(TICK_USE_MALTA_RTC)
/*
* If we are running on a board with the MC146818 RTC,
* use it to determine CPU pipeline clock frequency.
*/
u_int64_t counterval[2];
/* Set RTC to binary mode. */
writertc(RTC_STATUSB, (rtcin(RTC_STATUSB) | RTCSB_BCD));
/* Busy-wait for falling edge of RTC update. */
while (((rtcin(RTC_STATUSA) & RTCSA_TUP) == 0))
;
while (((rtcin(RTC_STATUSA)& RTCSA_TUP) != 0))
;
counterval[0] = mips_rd_count();
/* Busy-wait for falling edge of RTC update. */
while (((rtcin(RTC_STATUSA) & RTCSA_TUP) == 0))
;
while (((rtcin(RTC_STATUSA)& RTCSA_TUP) != 0))
;
counterval[1] = mips_rd_count();
platform_counter_freq = counterval[1] - counterval[0];
#endif
} while(0);
mips_timer_init_params(platform_counter_freq, 0);
}
/*
* Vnode op for VM putpages.
* possible bug: all IO done in sync mode
* Note that vop_close always invalidate pages before close, so it's
* not necessary to open vnode.
*
* smbfs_putpages(struct vnode *a_vp, vm_page_t *a_m, int a_count, int a_sync,
* int *a_rtvals, vm_ooffset_t a_offset)
*/
int
smbfs_putpages(struct vop_putpages_args *ap)
{
int error;
struct vnode *vp = ap->a_vp;
struct thread *td = curthread; /* XXX */
struct ucred *cred;
#ifdef SMBFS_RWGENERIC
KKASSERT(td->td_proc);
cred = td->td_proc->p_ucred;
VOP_OPEN(vp, FWRITE, cred, NULL);
error = vop_stdputpages(ap);
VOP_CLOSE(vp, FWRITE, cred);
return error;
#else
struct uio uio;
struct iovec iov;
vm_offset_t kva;
struct buf *bp;
int i, npages, count;
int doclose;
int *rtvals;
struct smbmount *smp;
struct smbnode *np;
struct smb_cred scred;
vm_page_t *pages;
KKASSERT(td->td_proc);
cred = td->td_proc->p_ucred;
/* VOP_OPEN(vp, FWRITE, cred, NULL);*/
np = VTOSMB(vp);
smp = VFSTOSMBFS(vp->v_mount);
pages = ap->a_m;
count = ap->a_count;
rtvals = ap->a_rtvals;
npages = btoc(count);
for (i = 0; i < npages; i++) {
rtvals[i] = VM_PAGER_AGAIN;
}
bp = getpbuf_kva(&smbfs_pbuf_freecnt);
kva = (vm_offset_t) bp->b_data;
pmap_qenter(kva, pages, npages);
iov.iov_base = (caddr_t) kva;
iov.iov_len = count;
uio.uio_iov = &iov;
uio.uio_iovcnt = 1;
uio.uio_offset = IDX_TO_OFF(pages[0]->pindex);
uio.uio_resid = count;
uio.uio_segflg = UIO_SYSSPACE;
uio.uio_rw = UIO_WRITE;
uio.uio_td = td;
SMBVDEBUG("ofs=%d,resid=%d\n",(int)uio.uio_offset, uio.uio_resid);
smb_makescred(&scred, td, cred);
/*
* This is kinda nasty. Since smbfs is physically closing the
* fid on close(), we have to reopen it if necessary. There are
* other races here too, such as if another process opens the same
* file while we are blocked in read, or the file is open read-only
* XXX
*/
error = 0;
doclose = 0;
if (np->n_opencount == 0) {
error = smbfs_smb_open(np, SMB_AM_OPENRW, &scred);
if (error == 0)
doclose = 1;
}
if (error == 0)
error = smb_write(smp->sm_share, np->n_fid, &uio, &scred);
if (doclose)
smbfs_smb_close(smp->sm_share, np->n_fid, NULL, &scred);
/* VOP_CLOSE(vp, FWRITE, cred);*/
SMBVDEBUG("paged write done: %d\n", error);
pmap_qremove(kva, npages);
relpbuf(bp, &smbfs_pbuf_freecnt);
if (!error) {
int nwritten = round_page(count - uio.uio_resid) / PAGE_SIZE;
for (i = 0; i < nwritten; i++) {
rtvals[i] = VM_PAGER_OK;
vm_page_undirty(pages[i]);
}
}
return rtvals[0];
#endif /* SMBFS_RWGENERIC */
}
void
m88100_trap(unsigned type, struct trapframe *frame)
{
struct proc *p;
u_quad_t sticks = 0;
struct vm_map *map;
vaddr_t va, pcb_onfault;
vm_prot_t ftype;
int fault_type, pbus_type;
u_long fault_code;
unsigned nss, fault_addr;
struct vmspace *vm;
union sigval sv;
int result;
#ifdef DDB
int s;
#endif
int sig = 0;
extern struct vm_map *kernel_map;
extern caddr_t guarded_access_start;
extern caddr_t guarded_access_end;
extern caddr_t guarded_access_bad;
uvmexp.traps++;
if ((p = curproc) == NULL)
p = &proc0;
if (USERMODE(frame->tf_epsr)) {
sticks = p->p_sticks;
type += T_USER;
p->p_md.md_tf = frame; /* for ptrace/signals */
}
fault_type = 0;
fault_code = 0;
fault_addr = frame->tf_sxip & XIP_ADDR;
switch (type) {
default:
panictrap(frame->tf_vector, frame);
break;
/*NOTREACHED*/
#if defined(DDB)
case T_KDB_BREAK:
s = splhigh();
db_enable_interrupt();
ddb_break_trap(T_KDB_BREAK, (db_regs_t*)frame);
db_disable_interrupt();
splx(s);
return;
case T_KDB_ENTRY:
s = splhigh();
db_enable_interrupt();
ddb_entry_trap(T_KDB_ENTRY, (db_regs_t*)frame);
db_disable_interrupt();
splx(s);
return;
#endif /* DDB */
case T_ILLFLT:
printf("Unimplemented opcode!\n");
panictrap(frame->tf_vector, frame);
break;
case T_INT:
case T_INT+T_USER:
/* This function pointer is set in machdep.c
It calls m188_ext_int or sbc_ext_int depending
on the value of brdtyp - smurph */
(*md.interrupt_func)(T_INT, frame);
return;
case T_MISALGNFLT:
printf("kernel misaligned access exception @ 0x%08x\n",
frame->tf_sxip);
panictrap(frame->tf_vector, frame);
break;
case T_INSTFLT:
/* kernel mode instruction access fault.
* Should never, never happen for a non-paged kernel.
*/
#ifdef TRAPDEBUG
pbus_type = CMMU_PFSR_FAULT(frame->tf_ipfsr);
printf("Kernel Instruction fault #%d (%s) v = 0x%x, frame 0x%x cpu %d\n",
pbus_type, pbus_exception_type[pbus_type],
fault_addr, frame, frame->tf_cpu);
#endif
panictrap(frame->tf_vector, frame);
break;
case T_DATAFLT:
/* kernel mode data fault */
/* data fault on the user address? */
if ((frame->tf_dmt0 & DMT_DAS) == 0) {
type = T_DATAFLT + T_USER;
goto user_fault;
}
fault_addr = frame->tf_dma0;
if (frame->tf_dmt0 & (DMT_WRITE|DMT_LOCKBAR)) {
ftype = VM_PROT_READ|VM_PROT_WRITE;
fault_code = VM_PROT_WRITE;
} else {
ftype = VM_PROT_READ;
fault_code = VM_PROT_READ;
}
va = trunc_page((vaddr_t)fault_addr);
if (va == 0) {
panic("trap: bad kernel access at %x", fault_addr);
}
vm = p->p_vmspace;
map = kernel_map;
pbus_type = CMMU_PFSR_FAULT(frame->tf_dpfsr);
#ifdef TRAPDEBUG
printf("Kernel Data access fault #%d (%s) v = 0x%x, frame 0x%x cpu %d\n",
pbus_type, pbus_exception_type[pbus_type],
fault_addr, frame, frame->tf_cpu);
#endif
switch (pbus_type) {
case CMMU_PFSR_BERROR:
/*
* If it is a guarded access, bus error is OK.
*/
if ((frame->tf_sxip & XIP_ADDR) >=
(unsigned)&guarded_access_start &&
(frame->tf_sxip & XIP_ADDR) <=
(unsigned)&guarded_access_end) {
frame->tf_snip =
((unsigned)&guarded_access_bad ) | NIP_V;
frame->tf_sfip =
((unsigned)&guarded_access_bad + 4) | FIP_V;
frame->tf_sxip = 0;
/* We sort of resolved the fault ourselves
* because we know where it came from
* [guarded_access()]. But we must still think
* about the other possible transactions in
* dmt1 & dmt2. Mark dmt0 so that
* data_access_emulation skips it. XXX smurph
*/
frame->tf_dmt0 |= DMT_SKIP;
data_access_emulation((unsigned *)frame);
frame->tf_dpfsr = 0;
frame->tf_dmt0 = 0;
return;
}
break;
case CMMU_PFSR_SUCCESS:
/*
* The fault was resolved. Call data_access_emulation
* to drain the data unit pipe line and reset dmt0
* so that trap won't get called again.
*/
data_access_emulation((unsigned *)frame);
frame->tf_dpfsr = 0;
frame->tf_dmt0 = 0;
return;
case CMMU_PFSR_SFAULT:
case CMMU_PFSR_PFAULT:
if ((pcb_onfault = p->p_addr->u_pcb.pcb_onfault) != 0)
p->p_addr->u_pcb.pcb_onfault = 0;
result = uvm_fault(map, va, VM_FAULT_INVALID, ftype);
p->p_addr->u_pcb.pcb_onfault = pcb_onfault;
if (result == 0) {
/*
* We could resolve the fault. Call
* data_access_emulation to drain the data
* unit pipe line and reset dmt0 so that trap
* won't get called again.
*/
data_access_emulation((unsigned *)frame);
frame->tf_dpfsr = 0;
frame->tf_dmt0 = 0;
return;
}
break;
}
#ifdef TRAPDEBUG
printf("PBUS Fault %d (%s) va = 0x%x\n", pbus_type,
pbus_exception_type[pbus_type], va);
#endif
panictrap(frame->tf_vector, frame);
/* NOTREACHED */
case T_INSTFLT+T_USER:
/* User mode instruction access fault */
/* FALLTHROUGH */
case T_DATAFLT+T_USER:
user_fault:
if (type == T_INSTFLT + T_USER) {
pbus_type = CMMU_PFSR_FAULT(frame->tf_ipfsr);
#ifdef TRAPDEBUG
printf("User Instruction fault #%d (%s) v = 0x%x, frame 0x%x cpu %d\n",
pbus_type, pbus_exception_type[pbus_type],
fault_addr, frame, frame->tf_cpu);
#endif
} else {
fault_addr = frame->tf_dma0;
pbus_type = CMMU_PFSR_FAULT(frame->tf_dpfsr);
#ifdef TRAPDEBUG
printf("User Data access fault #%d (%s) v = 0x%x, frame 0x%x cpu %d\n",
pbus_type, pbus_exception_type[pbus_type],
fault_addr, frame, frame->tf_cpu);
#endif
}
if (frame->tf_dmt0 & (DMT_WRITE | DMT_LOCKBAR)) {
ftype = VM_PROT_READ | VM_PROT_WRITE;
fault_code = VM_PROT_WRITE;
} else {
ftype = VM_PROT_READ;
fault_code = VM_PROT_READ;
}
va = trunc_page((vaddr_t)fault_addr);
vm = p->p_vmspace;
map = &vm->vm_map;
if ((pcb_onfault = p->p_addr->u_pcb.pcb_onfault) != 0)
p->p_addr->u_pcb.pcb_onfault = 0;
/* Call uvm_fault() to resolve non-bus error faults */
switch (pbus_type) {
case CMMU_PFSR_SUCCESS:
result = 0;
break;
case CMMU_PFSR_BERROR:
result = EACCES;
break;
default:
result = uvm_fault(map, va, VM_FAULT_INVALID, ftype);
if (result == EACCES)
result = EFAULT;
break;
}
p->p_addr->u_pcb.pcb_onfault = pcb_onfault;
if ((caddr_t)va >= vm->vm_maxsaddr) {
if (result == 0) {
nss = btoc(USRSTACK - va);/* XXX check this */
if (nss > vm->vm_ssize)
vm->vm_ssize = nss;
}
}
/*
* This could be a fault caused in copyin*()
* while accessing user space.
*/
if (result != 0 && pcb_onfault != 0) {
frame->tf_snip = pcb_onfault | NIP_V;
frame->tf_sfip = (pcb_onfault + 4) | FIP_V;
frame->tf_sxip = 0;
/*
* Continue as if the fault had been resolved, but
* do not try to complete the faulting access.
*/
frame->tf_dmt0 |= DMT_SKIP;
result = 0;
}
if (result == 0) {
if (type == T_DATAFLT+T_USER) {
/*
* We could resolve the fault. Call
* data_access_emulation to drain the data unit
* pipe line and reset dmt0 so that trap won't
* get called again.
*/
data_access_emulation((unsigned *)frame);
frame->tf_dpfsr = 0;
frame->tf_dmt0 = 0;
} else {
/*
* back up SXIP, SNIP,
* clearing the Error bit
*/
frame->tf_sfip = frame->tf_snip & ~FIP_E;
frame->tf_snip = frame->tf_sxip & ~NIP_E;
frame->tf_ipfsr = 0;
}
} else {
sig = result == EACCES ? SIGBUS : SIGSEGV;
fault_type = result == EACCES ?
BUS_ADRERR : SEGV_MAPERR;
}
break;
case T_MISALGNFLT+T_USER:
/* Fix any misaligned ld.d or st.d instructions */
sig = double_reg_fixup(frame);
fault_type = BUS_ADRALN;
break;
case T_PRIVINFLT+T_USER:
case T_ILLFLT+T_USER:
#ifndef DDB
case T_KDB_BREAK:
case T_KDB_ENTRY:
#endif
case T_KDB_BREAK+T_USER:
case T_KDB_ENTRY+T_USER:
case T_KDB_TRACE:
case T_KDB_TRACE+T_USER:
sig = SIGILL;
break;
case T_BNDFLT+T_USER:
sig = SIGFPE;
break;
case T_ZERODIV+T_USER:
sig = SIGFPE;
fault_type = FPE_INTDIV;
break;
case T_OVFFLT+T_USER:
sig = SIGFPE;
fault_type = FPE_INTOVF;
break;
case T_FPEPFLT+T_USER:
case T_FPEIFLT+T_USER:
sig = SIGFPE;
break;
case T_SIGSYS+T_USER:
sig = SIGSYS;
break;
case T_SIGTRAP+T_USER:
sig = SIGTRAP;
fault_type = TRAP_TRACE;
break;
case T_STEPBPT+T_USER:
#ifdef PTRACE
/*
* This trap is used by the kernel to support single-step
* debugging (although any user could generate this trap
* which should probably be handled differently). When a
* process is continued by a debugger with the PT_STEP
* function of ptrace (single step), the kernel inserts
* one or two breakpoints in the user process so that only
* one instruction (or two in the case of a delayed branch)
* is executed. When this breakpoint is hit, we get the
* T_STEPBPT trap.
*/
{
unsigned va;
unsigned instr;
unsigned pc = PC_REGS(&frame->tf_regs);
/* read break instruction */
copyin((caddr_t)pc, &instr, sizeof(unsigned));
#if 0
printf("trap: %s (%d) breakpoint %x at %x: (adr %x ins %x)\n",
p->p_comm, p->p_pid, instr, pc,
p->p_md.md_ss_addr, p->p_md.md_ss_instr); /* XXX */
#endif
/* check and see if we got here by accident */
if ((p->p_md.md_ss_addr != pc &&
p->p_md.md_ss_taken_addr != pc) ||
instr != SSBREAKPOINT) {
sig = SIGTRAP;
fault_type = TRAP_TRACE;
break;
}
/* restore original instruction and clear BP */
va = p->p_md.md_ss_addr;
if (va != 0) {
instr = p->p_md.md_ss_instr;
ss_put_value(p, va, instr, sizeof(instr));
}
/* branch taken instruction */
instr = p->p_md.md_ss_taken_instr;
if (instr != 0) {
va = p->p_md.md_ss_taken_addr;
ss_put_value(p, va, instr, sizeof(instr));
}
#if 1
frame->tf_sfip = frame->tf_snip;
frame->tf_snip = pc | NIP_V;
#endif
p->p_md.md_ss_addr = 0;
p->p_md.md_ss_instr = 0;
p->p_md.md_ss_taken_addr = 0;
p->p_md.md_ss_taken_instr = 0;
sig = SIGTRAP;
fault_type = TRAP_BRKPT;
}
#else
sig = SIGTRAP;
fault_type = TRAP_TRACE;
#endif
break;
case T_USERBPT+T_USER:
/*
* This trap is meant to be used by debuggers to implement
* breakpoint debugging. When we get this trap, we just
* return a signal which gets caught by the debugger.
*/
frame->tf_sfip = frame->tf_snip;
frame->tf_snip = frame->tf_sxip;
sig = SIGTRAP;
fault_type = TRAP_BRKPT;
break;
case T_ASTFLT+T_USER:
uvmexp.softs++;
want_ast = 0;
if (p->p_flag & P_OWEUPC) {
p->p_flag &= ~P_OWEUPC;
ADDUPROF(p);
}
break;
}
/*
* If trap from supervisor mode, just return
*/
if (type < T_USER)
return;
if (sig) {
sv.sival_int = fault_addr;
trapsignal(p, sig, fault_code, fault_type, sv);
/*
* don't want multiple faults - we are going to
* deliver signal.
*/
frame->tf_dmt0 = 0;
frame->tf_ipfsr = frame->tf_dpfsr = 0;
}
userret(p, frame, sticks);
}
void
platform_start(__register_t a0, __register_t a1,
__register_t a2 __unused, __register_t a3 __unused)
{
uint64_t platform_counter_freq;
vm_offset_t kernend;
int argc = a0;
char **argv = (char **)a1;
int i, mem;
/* clear the BSS and SBSS segments */
kernend = (vm_offset_t)&end;
memset(&edata, 0, kernend - (vm_offset_t)(&edata));
mips_postboot_fixup();
/* Initialize pcpu stuff */
mips_pcpu0_init();
/*
* Looking for mem=XXM argument
*/
mem = 0; /* Just something to start with */
for (i=0; i < argc; i++) {
if (strncmp(argv[i], "mem=", 4) == 0) {
mem = strtol(argv[i] + 4, NULL, 0);
break;
}
}
bootverbose = 1;
if (mem > 0)
realmem = btoc(mem << 20);
else
realmem = btoc(32 << 20);
for (i = 0; i < 10; i++) {
phys_avail[i] = 0;
}
/* phys_avail regions are in bytes */
phys_avail[0] = MIPS_KSEG0_TO_PHYS(kernel_kseg0_end);
phys_avail[1] = ctob(realmem);
dump_avail[0] = phys_avail[0];
dump_avail[1] = phys_avail[1];
physmem = realmem;
/*
* ns8250 uart code uses DELAY so ticker should be inititalized
* before cninit. And tick_init_params refers to hz, so * init_param1
* should be called first.
*/
init_param1();
/* TODO: parse argc,argv */
platform_counter_freq = 330000000UL;
mips_timer_init_params(platform_counter_freq, 1);
cninit();
/* Panic here, after cninit */
if (mem == 0)
panic("No mem=XX parameter in arguments");
printf("cmd line: ");
for (i=0; i < argc; i++)
printf("%s ", argv[i]);
printf("\n");
init_param2(physmem);
mips_cpu_init();
pmap_bootstrap();
mips_proc0_init();
mutex_init();
kdb_init();
#ifdef KDB
if (boothowto & RB_KDB)
kdb_enter(KDB_WHY_BOOTFLAGS, "Boot flags requested debugger");
#endif
}
void
m88110_trap(unsigned type, struct trapframe *frame)
{
struct proc *p;
u_quad_t sticks = 0;
struct vm_map *map;
vaddr_t va, pcb_onfault;
vm_prot_t ftype;
int fault_type;
u_long fault_code;
unsigned nss, fault_addr;
struct vmspace *vm;
union sigval sv;
int result;
#ifdef DDB
int s;
#endif
int sig = 0;
pt_entry_t *pte;
extern struct vm_map *kernel_map;
extern unsigned guarded_access_start;
extern unsigned guarded_access_end;
extern unsigned guarded_access_bad;
extern pt_entry_t *pmap_pte(pmap_t, vaddr_t);
uvmexp.traps++;
if ((p = curproc) == NULL)
p = &proc0;
if (USERMODE(frame->tf_epsr)) {
sticks = p->p_sticks;
type += T_USER;
p->p_md.md_tf = frame; /* for ptrace/signals */
}
fault_type = 0;
fault_code = 0;
fault_addr = frame->tf_exip & XIP_ADDR;
switch (type) {
default:
panictrap(frame->tf_vector, frame);
break;
/*NOTREACHED*/
case T_197_READ+T_USER:
case T_197_READ:
printf("DMMU read miss: Hardware Table Searches should be enabled!\n");
panictrap(frame->tf_vector, frame);
break;
/*NOTREACHED*/
case T_197_WRITE+T_USER:
case T_197_WRITE:
printf("DMMU write miss: Hardware Table Searches should be enabled!\n");
panictrap(frame->tf_vector, frame);
break;
/*NOTREACHED*/
case T_197_INST+T_USER:
case T_197_INST:
printf("IMMU miss: Hardware Table Searches should be enabled!\n");
panictrap(frame->tf_vector, frame);
break;
/*NOTREACHED*/
#ifdef DDB
case T_KDB_TRACE:
s = splhigh();
db_enable_interrupt();
ddb_break_trap(T_KDB_TRACE, (db_regs_t*)frame);
db_disable_interrupt();
splx(s);
return;
case T_KDB_BREAK:
s = splhigh();
db_enable_interrupt();
ddb_break_trap(T_KDB_BREAK, (db_regs_t*)frame);
db_disable_interrupt();
splx(s);
return;
case T_KDB_ENTRY:
s = splhigh();
db_enable_interrupt();
ddb_entry_trap(T_KDB_ENTRY, (db_regs_t*)frame);
db_disable_interrupt();
/* skip one instruction */
if (frame->tf_exip & 1)
frame->tf_exip = frame->tf_enip;
else
frame->tf_exip += 4;
splx(s);
return;
#if 0
case T_ILLFLT:
s = splhigh();
db_enable_interrupt();
ddb_error_trap(type == T_ILLFLT ? "unimplemented opcode" :
"error fault", (db_regs_t*)frame);
db_disable_interrupt();
splx(s);
return;
#endif /* 0 */
#endif /* DDB */
case T_ILLFLT:
printf("Unimplemented opcode!\n");
panictrap(frame->tf_vector, frame);
break;
case T_NON_MASK:
case T_NON_MASK+T_USER:
(*md.interrupt_func)(T_NON_MASK, frame);
return;
case T_INT:
case T_INT+T_USER:
(*md.interrupt_func)(T_INT, frame);
return;
case T_MISALGNFLT:
printf("kernel mode misaligned access exception @ 0x%08x\n",
frame->tf_exip);
panictrap(frame->tf_vector, frame);
break;
/*NOTREACHED*/
case T_INSTFLT:
/* kernel mode instruction access fault.
* Should never, never happen for a non-paged kernel.
*/
#ifdef TRAPDEBUG
printf("Kernel Instruction fault exip %x isr %x ilar %x\n",
frame->tf_exip, frame->tf_isr, frame->tf_ilar);
#endif
panictrap(frame->tf_vector, frame);
break;
/*NOTREACHED*/
case T_DATAFLT:
/* kernel mode data fault */
/* data fault on the user address? */
if ((frame->tf_dsr & CMMU_DSR_SU) == 0) {
type = T_DATAFLT + T_USER;
goto m88110_user_fault;
}
#ifdef TRAPDEBUG
printf("Kernel Data access fault exip %x dsr %x dlar %x\n",
frame->tf_exip, frame->tf_dsr, frame->tf_dlar);
#endif
fault_addr = frame->tf_dlar;
if (frame->tf_dsr & CMMU_DSR_RW) {
ftype = VM_PROT_READ;
fault_code = VM_PROT_READ;
} else {
ftype = VM_PROT_READ|VM_PROT_WRITE;
fault_code = VM_PROT_WRITE;
}
va = trunc_page((vaddr_t)fault_addr);
if (va == 0) {
panic("trap: bad kernel access at %x", fault_addr);
}
vm = p->p_vmspace;
map = kernel_map;
if (frame->tf_dsr & CMMU_DSR_BE) {
/*
* If it is a guarded access, bus error is OK.
*/
if ((frame->tf_exip & XIP_ADDR) >=
(unsigned)&guarded_access_start &&
(frame->tf_exip & XIP_ADDR) <=
(unsigned)&guarded_access_end) {
frame->tf_exip = (unsigned)&guarded_access_bad;
return;
}
}
if (frame->tf_dsr & (CMMU_DSR_SI | CMMU_DSR_PI)) {
frame->tf_dsr &= ~CMMU_DSR_WE; /* undefined */
/*
* On a segment or a page fault, call uvm_fault() to
* resolve the fault.
*/
if ((pcb_onfault = p->p_addr->u_pcb.pcb_onfault) != 0)
p->p_addr->u_pcb.pcb_onfault = 0;
result = uvm_fault(map, va, VM_FAULT_INVALID, ftype);
p->p_addr->u_pcb.pcb_onfault = pcb_onfault;
if (result == 0)
return;
}
if (frame->tf_dsr & CMMU_DSR_WE) { /* write fault */
/*
* This could be a write protection fault or an
* exception to set the used and modified bits
* in the pte. Basically, if we got a write error,
* then we already have a pte entry that faulted
* in from a previous seg fault or page fault.
* Get the pte and check the status of the
* modified and valid bits to determine if this
* indeed a real write fault. XXX smurph
*/
pte = pmap_pte(map->pmap, va);
#ifdef DEBUG
if (pte == PT_ENTRY_NULL)
panic("NULL pte on write fault??");
#endif
if (!(*pte & PG_M) && !(*pte & PG_RO)) {
/* Set modified bit and try the write again. */
#ifdef TRAPDEBUG
printf("Corrected kernel write fault, map %x pte %x\n",
map->pmap, *pte);
#endif
*pte |= PG_M;
return;
#if 1 /* shouldn't happen */
} else {
/* must be a real wp fault */
#ifdef TRAPDEBUG
printf("Uncorrected kernel write fault, map %x pte %x\n",
map->pmap, *pte);
#endif
if ((pcb_onfault = p->p_addr->u_pcb.pcb_onfault) != 0)
p->p_addr->u_pcb.pcb_onfault = 0;
result = uvm_fault(map, va, VM_FAULT_INVALID, ftype);
p->p_addr->u_pcb.pcb_onfault = pcb_onfault;
if (result == 0)
return;
#endif
}
}
panictrap(frame->tf_vector, frame);
/* NOTREACHED */
case T_INSTFLT+T_USER:
/* User mode instruction access fault */
/* FALLTHROUGH */
case T_DATAFLT+T_USER:
m88110_user_fault:
if (type == T_INSTFLT+T_USER) {
ftype = VM_PROT_READ;
fault_code = VM_PROT_READ;
#ifdef TRAPDEBUG
printf("User Instruction fault exip %x isr %x ilar %x\n",
frame->tf_exip, frame->tf_isr, frame->tf_ilar);
#endif
} else {
fault_addr = frame->tf_dlar;
if (frame->tf_dsr & CMMU_DSR_RW) {
ftype = VM_PROT_READ;
fault_code = VM_PROT_READ;
} else {
ftype = VM_PROT_READ|VM_PROT_WRITE;
fault_code = VM_PROT_WRITE;
}
#ifdef TRAPDEBUG
printf("User Data access fault exip %x dsr %x dlar %x\n",
frame->tf_exip, frame->tf_dsr, frame->tf_dlar);
#endif
}
va = trunc_page((vaddr_t)fault_addr);
vm = p->p_vmspace;
map = &vm->vm_map;
if ((pcb_onfault = p->p_addr->u_pcb.pcb_onfault) != 0)
p->p_addr->u_pcb.pcb_onfault = 0;
/*
* Call uvm_fault() to resolve non-bus error faults
* whenever possible.
*/
if (type == T_DATAFLT+T_USER) {
/* data faults */
if (frame->tf_dsr & CMMU_DSR_BE) {
/* bus error */
result = EACCES;
} else
if (frame->tf_dsr & (CMMU_DSR_SI | CMMU_DSR_PI)) {
/* segment or page fault */
result = uvm_fault(map, va, VM_FAULT_INVALID, ftype);
p->p_addr->u_pcb.pcb_onfault = pcb_onfault;
if (result == EACCES)
result = EFAULT;
} else
if (frame->tf_dsr & (CMMU_DSR_CP | CMMU_DSR_WA)) {
/* copyback or write allocate error */
result = EACCES;
} else
if (frame->tf_dsr & CMMU_DSR_WE) {
/* write fault */
/* This could be a write protection fault or an
* exception to set the used and modified bits
* in the pte. Basically, if we got a write
* error, then we already have a pte entry that
* faulted in from a previous seg fault or page
* fault.
* Get the pte and check the status of the
* modified and valid bits to determine if this
* indeed a real write fault. XXX smurph
*/
pte = pmap_pte(vm_map_pmap(map), va);
#ifdef DEBUG
if (pte == PT_ENTRY_NULL)
panic("NULL pte on write fault??");
#endif
if (!(*pte & PG_M) && !(*pte & PG_RO)) {
/*
* Set modified bit and try the
* write again.
*/
#ifdef TRAPDEBUG
printf("Corrected userland write fault, map %x pte %x\n",
map->pmap, *pte);
#endif
*pte |= PG_M;
/*
* invalidate ATCs to force
* table search
*/
set_dcmd(CMMU_DCMD_INV_UATC);
return;
} else {
/* must be a real wp fault */
#ifdef TRAPDEBUG
printf("Uncorrected userland write fault, map %x pte %x\n",
map->pmap, *pte);
#endif
result = uvm_fault(map, va, VM_FAULT_INVALID, ftype);
p->p_addr->u_pcb.pcb_onfault = pcb_onfault;
if (result == EACCES)
result = EFAULT;
}
} else {
#ifdef TRAPDEBUG
printf("Unexpected Data access fault dsr %x\n",
frame->tf_dsr);
#endif
panictrap(frame->tf_vector, frame);
}
} else {
/* instruction faults */
if (frame->tf_isr &
(CMMU_ISR_BE | CMMU_ISR_SP | CMMU_ISR_TBE)) {
/* bus error, supervisor protection */
result = EACCES;
} else
if (frame->tf_isr & (CMMU_ISR_SI | CMMU_ISR_PI)) {
/* segment or page fault */
result = uvm_fault(map, va, VM_FAULT_INVALID, ftype);
p->p_addr->u_pcb.pcb_onfault = pcb_onfault;
if (result == EACCES)
result = EFAULT;
} else {
#ifdef TRAPDEBUG
printf("Unexpected Instruction fault isr %x\n",
frame->tf_isr);
#endif
panictrap(frame->tf_vector, frame);
}
}
if ((caddr_t)va >= vm->vm_maxsaddr) {
if (result == 0) {
nss = btoc(USRSTACK - va);/* XXX check this */
if (nss > vm->vm_ssize)
vm->vm_ssize = nss;
}
}
/*
* This could be a fault caused in copyin*()
* while accessing user space.
*/
if (result != 0 && pcb_onfault != 0) {
frame->tf_exip = pcb_onfault;
/*
* Continue as if the fault had been resolved.
*/
result = 0;
}
if (result != 0) {
sig = result == EACCES ? SIGBUS : SIGSEGV;
fault_type = result == EACCES ?
BUS_ADRERR : SEGV_MAPERR;
}
break;
case T_MISALGNFLT+T_USER:
/* Fix any misaligned ld.d or st.d instructions */
sig = double_reg_fixup(frame);
fault_type = BUS_ADRALN;
break;
case T_PRIVINFLT+T_USER:
case T_ILLFLT+T_USER:
#ifndef DDB
case T_KDB_BREAK:
case T_KDB_ENTRY:
case T_KDB_TRACE:
#endif
case T_KDB_BREAK+T_USER:
case T_KDB_ENTRY+T_USER:
case T_KDB_TRACE+T_USER:
sig = SIGILL;
break;
case T_BNDFLT+T_USER:
sig = SIGFPE;
break;
case T_ZERODIV+T_USER:
sig = SIGFPE;
fault_type = FPE_INTDIV;
break;
case T_OVFFLT+T_USER:
sig = SIGFPE;
fault_type = FPE_INTOVF;
break;
case T_FPEPFLT+T_USER:
case T_FPEIFLT+T_USER:
sig = SIGFPE;
break;
case T_SIGSYS+T_USER:
sig = SIGSYS;
break;
case T_SIGTRAP+T_USER:
sig = SIGTRAP;
fault_type = TRAP_TRACE;
break;
case T_STEPBPT+T_USER:
#ifdef PTRACE
/*
* This trap is used by the kernel to support single-step
* debugging (although any user could generate this trap
* which should probably be handled differently). When a
* process is continued by a debugger with the PT_STEP
* function of ptrace (single step), the kernel inserts
* one or two breakpoints in the user process so that only
* one instruction (or two in the case of a delayed branch)
* is executed. When this breakpoint is hit, we get the
* T_STEPBPT trap.
*/
{
unsigned instr;
unsigned pc = PC_REGS(&frame->tf_regs);
/* read break instruction */
copyin((caddr_t)pc, &instr, sizeof(unsigned));
#if 0
printf("trap: %s (%d) breakpoint %x at %x: (adr %x ins %x)\n",
p->p_comm, p->p_pid, instr, pc,
p->p_md.md_ss_addr, p->p_md.md_ss_instr); /* XXX */
#endif
/* check and see if we got here by accident */
#ifdef notyet
if (p->p_md.md_ss_addr != pc || instr != SSBREAKPOINT) {
sig = SIGTRAP;
fault_type = TRAP_TRACE;
break;
}
#endif
/* restore original instruction and clear BP */
instr = p->p_md.md_ss_instr;
if (instr != 0)
ss_put_value(p, pc, instr, sizeof(instr));
p->p_md.md_ss_addr = 0;
p->p_md.md_ss_instr = 0;
sig = SIGTRAP;
fault_type = TRAP_BRKPT;
}
#else
sig = SIGTRAP;
fault_type = TRAP_TRACE;
#endif
break;
case T_USERBPT+T_USER:
/*
* This trap is meant to be used by debuggers to implement
* breakpoint debugging. When we get this trap, we just
* return a signal which gets caught by the debugger.
*/
sig = SIGTRAP;
fault_type = TRAP_BRKPT;
break;
case T_ASTFLT+T_USER:
uvmexp.softs++;
want_ast = 0;
if (p->p_flag & P_OWEUPC) {
p->p_flag &= ~P_OWEUPC;
ADDUPROF(p);
}
break;
}
/*
* If trap from supervisor mode, just return
*/
if (type < T_USER)
return;
if (sig) {
sv.sival_int = fault_addr;
trapsignal(p, sig, fault_code, fault_type, sv);
}
userret(p, frame, sticks);
}
/*
* Vnode op for VM getpages.
* Wish wish .... get rid from multiple IO routines
*
* nwfs_getpages(struct vnode *a_vp, vm_page_t *a_m, int a_count,
* int a_reqpage, vm_ooffset_t a_offset)
*/
int
nwfs_getpages(struct vop_getpages_args *ap)
{
#ifndef NWFS_RWCACHE
return vnode_pager_generic_getpages(ap->a_vp, ap->a_m, ap->a_count,
ap->a_reqpage, ap->a_seqaccess);
#else
int i, error, npages;
size_t nextoff, toff;
size_t count;
size_t size;
struct uio uio;
struct iovec iov;
vm_offset_t kva;
struct buf *bp;
struct vnode *vp;
struct thread *td = curthread; /* XXX */
struct ucred *cred;
struct nwmount *nmp;
struct nwnode *np;
vm_page_t *pages;
KKASSERT(td->td_proc);
cred = td->td_proc->p_ucred;
vp = ap->a_vp;
np = VTONW(vp);
nmp = VFSTONWFS(vp->v_mount);
pages = ap->a_m;
count = (size_t)ap->a_count;
if (vp->v_object == NULL) {
kprintf("nwfs_getpages: called with non-merged cache vnode??\n");
return VM_PAGER_ERROR;
}
bp = getpbuf_kva(&nwfs_pbuf_freecnt);
npages = btoc(count);
kva = (vm_offset_t) bp->b_data;
pmap_qenter(kva, pages, npages);
iov.iov_base = (caddr_t) kva;
iov.iov_len = count;
uio.uio_iov = &iov;
uio.uio_iovcnt = 1;
uio.uio_offset = IDX_TO_OFF(pages[0]->pindex);
uio.uio_resid = count;
uio.uio_segflg = UIO_SYSSPACE;
uio.uio_rw = UIO_READ;
uio.uio_td = td;
error = ncp_read(NWFSTOCONN(nmp), &np->n_fh, &uio,cred);
pmap_qremove(kva, npages);
relpbuf(bp, &nwfs_pbuf_freecnt);
if (error && (uio.uio_resid == count)) {
kprintf("nwfs_getpages: error %d\n",error);
for (i = 0; i < npages; i++) {
if (ap->a_reqpage != i)
vnode_pager_freepage(pages[i]);
}
return VM_PAGER_ERROR;
}
size = count - uio.uio_resid;
for (i = 0, toff = 0; i < npages; i++, toff = nextoff) {
vm_page_t m;
nextoff = toff + PAGE_SIZE;
m = pages[i];
m->flags &= ~PG_ZERO;
/*
* NOTE: pmap dirty bit should have already been cleared.
* We do not clear it here.
*/
if (nextoff <= size) {
m->valid = VM_PAGE_BITS_ALL;
m->dirty = 0;
} else {
int nvalid = ((size + DEV_BSIZE - 1) - toff) &
~(DEV_BSIZE - 1);
vm_page_set_validclean(m, 0, nvalid);
}
if (i != ap->a_reqpage) {
/*
* Whether or not to leave the page activated is up in
* the air, but we should put the page on a page queue
* somewhere (it already is in the object). Result:
* It appears that emperical results show that
* deactivating pages is best.
*/
/*
* Just in case someone was asking for this page we
* now tell them that it is ok to use.
*/
if (!error) {
if (m->flags & PG_REFERENCED)
vm_page_activate(m);
else
vm_page_deactivate(m);
vm_page_wakeup(m);
} else {
vnode_pager_freepage(m);
}
}
}
return 0;
#endif /* NWFS_RWCACHE */
}
void
mach_init(void)
{
void *kernend;
uint32_t memsize;
extern char edata[], end[]; /* XXX */
/* clear the BSS segment */
kernend = (void *)mips_round_page(end);
memset(edata, 0, (char *)kernend - edata);
/* setup early console */
atheros_set_platformsw();
/* set CPU model info for sysctl_hw */
snprintf(cpu_model, 64, "Atheros %s", atheros_get_cpuname());
/*
* Set up the exception vectors and CPU-specific function
* vectors early on. We need the wbflush() vector set up
* before comcnattach() is called (or at least before the
* first printf() after that is called).
* Sets up mips_cpu_flags that may be queried by other
* functions called during startup.
* Also clears the I+D caches.
*/
mips_vector_init(NULL, false);
/*
* Calibrate timers.
*/
cal_timer();
/*
* Set the VM page size.
*/
uvm_setpagesize();
/*
* Look at arguments passed to us and compute boothowto.
*/
boothowto = RB_AUTOBOOT;
#ifdef KADB
boothowto |= RB_KDB;
#endif
/*
* This would be a good place to parse a boot command line, if
* we got one from the bootloader. Right now we have no way to
* get one from e.g. vxworks.
*/
/*
* Determine the memory size.
*
* Note: Reserve the first page! That's where the trap
* vectors are located.
*/
memsize = atheros_get_memsize();
printf("Memory size: 0x%08x\n", memsize);
physmem = btoc(memsize);
mem_clusters[mem_cluster_cnt].start = PAGE_SIZE;
mem_clusters[mem_cluster_cnt].size =
memsize - mem_clusters[mem_cluster_cnt].start;
mem_cluster_cnt++;
/*
* Load the available pages into the VM system.
*/
mips_page_physload(MIPS_KSEG0_START, (vaddr_t)kernend,
mem_clusters, mem_cluster_cnt, NULL, 0);
/*
* Initialize message buffer (at end of core).
*/
mips_init_msgbuf();
/*
* Initialize the virtual memory system.
*/
pmap_bootstrap();
/*
* Allocate uarea page for lwp0 and set it.
*/
mips_init_lwp0_uarea();
/*
* Initialize busses.
*/
atheros_bus_init();
/*
* Turn off (ignore) the hardware watchdog. If we got this
* far, then we shouldn't need it anymore.
*/
atheros_wdog_reload(0);
/*
* Turn off watchpoint that may have been enabled by the
* PROM. VxWorks bootloader seems to leave one set.
*/
__asm volatile (
"mtc0 $0, $%0\n\t"
"nop\n\t"
"nop\n\t" :: "n"(MIPS_COP_0_WATCH_LO));
/*
* Initialize debuggers, and break into them, if appropriate.
*/
#ifdef DDB
if (boothowto & RB_KDB)
Debugger();
#endif
}
/*
* Vnode op for VM putpages.
* possible bug: all IO done in sync mode
* Note that vop_close always invalidate pages before close, so it's
* not necessary to open vnode.
*
* nwfs_putpages(struct vnode *a_vp, vm_page_t *a_m, int a_count,
* int a_sync, int *a_rtvals, vm_ooffset_t a_offset)
*/
int
nwfs_putpages(struct vop_putpages_args *ap)
{
int error;
struct thread *td = curthread; /* XXX */
struct vnode *vp = ap->a_vp;
struct ucred *cred;
#ifndef NWFS_RWCACHE
KKASSERT(td->td_proc);
cred = td->td_proc->p_ucred; /* XXX */
VOP_OPEN(vp, FWRITE, cred, NULL);
error = vnode_pager_generic_putpages(ap->a_vp, ap->a_m, ap->a_count,
ap->a_sync, ap->a_rtvals);
VOP_CLOSE(vp, FWRITE, cred);
return error;
#else
struct uio uio;
struct iovec iov;
vm_offset_t kva;
struct buf *bp;
int i, npages, count;
int *rtvals;
struct nwmount *nmp;
struct nwnode *np;
vm_page_t *pages;
KKASSERT(td->td_proc);
cred = td->td_proc->p_ucred; /* XXX */
/* VOP_OPEN(vp, FWRITE, cred, NULL);*/
np = VTONW(vp);
nmp = VFSTONWFS(vp->v_mount);
pages = ap->a_m;
count = ap->a_count;
rtvals = ap->a_rtvals;
npages = btoc(count);
for (i = 0; i < npages; i++) {
rtvals[i] = VM_PAGER_AGAIN;
}
bp = getpbuf_kva(&nwfs_pbuf_freecnt);
kva = (vm_offset_t) bp->b_data;
pmap_qenter(kva, pages, npages);
iov.iov_base = (caddr_t) kva;
iov.iov_len = count;
uio.uio_iov = &iov;
uio.uio_iovcnt = 1;
uio.uio_offset = IDX_TO_OFF(pages[0]->pindex);
uio.uio_resid = count;
uio.uio_segflg = UIO_SYSSPACE;
uio.uio_rw = UIO_WRITE;
uio.uio_td = td;
NCPVNDEBUG("ofs=%d,resid=%d\n",(int)uio.uio_offset, uio.uio_resid);
error = ncp_write(NWFSTOCONN(nmp), &np->n_fh, &uio, cred);
/* VOP_CLOSE(vp, FWRITE, cred);*/
NCPVNDEBUG("paged write done: %d\n", error);
pmap_qremove(kva, npages);
relpbuf(bp, &nwfs_pbuf_freecnt);
if (!error) {
int nwritten = round_page(count - uio.uio_resid) / PAGE_SIZE;
for (i = 0; i < nwritten; i++) {
rtvals[i] = VM_PAGER_OK;
vm_page_undirty(pages[i]);
}
}
return rtvals[0];
#endif /* NWFS_RWCACHE */
}
int
physio(struct cdev *dev, struct uio *uio, int ioflag)
{
struct cdevsw *csw;
struct buf *pbuf;
struct bio *bp;
struct vm_page **pages;
caddr_t sa;
u_int iolen, poff;
int error, i, npages, maxpages;
vm_prot_t prot;
csw = dev->si_devsw;
npages = 0;
sa = NULL;
/* check if character device is being destroyed */
if (csw == NULL)
return (ENXIO);
/* XXX: sanity check */
if(dev->si_iosize_max < PAGE_SIZE) {
printf("WARNING: %s si_iosize_max=%d, using DFLTPHYS.\n",
devtoname(dev), dev->si_iosize_max);
dev->si_iosize_max = DFLTPHYS;
}
/*
* If the driver does not want I/O to be split, that means that we
* need to reject any requests that will not fit into one buffer.
*/
if (dev->si_flags & SI_NOSPLIT &&
(uio->uio_resid > dev->si_iosize_max || uio->uio_resid > MAXPHYS ||
uio->uio_iovcnt > 1)) {
/*
* Tell the user why his I/O was rejected.
*/
if (uio->uio_resid > dev->si_iosize_max)
uprintf("%s: request size=%zd > si_iosize_max=%d; "
"cannot split request\n", devtoname(dev),
uio->uio_resid, dev->si_iosize_max);
if (uio->uio_resid > MAXPHYS)
uprintf("%s: request size=%zd > MAXPHYS=%d; "
"cannot split request\n", devtoname(dev),
uio->uio_resid, MAXPHYS);
if (uio->uio_iovcnt > 1)
uprintf("%s: request vectors=%d > 1; "
"cannot split request\n", devtoname(dev),
uio->uio_iovcnt);
return (EFBIG);
}
/*
* Keep the process UPAGES from being swapped. Processes swapped
* out while holding pbufs, used by swapper, may lead to deadlock.
*/
PHOLD(curproc);
bp = g_alloc_bio();
if (uio->uio_segflg != UIO_USERSPACE) {
pbuf = NULL;
pages = NULL;
} else if ((dev->si_flags & SI_UNMAPPED) && unmapped_buf_allowed) {
pbuf = NULL;
maxpages = btoc(MIN(uio->uio_resid, MAXPHYS)) + 1;
pages = malloc(sizeof(*pages) * maxpages, M_DEVBUF, M_WAITOK);
} else {
pbuf = uma_zalloc(pbuf_zone, M_WAITOK);
sa = pbuf->b_data;
maxpages = btoc(MAXPHYS);
pages = pbuf->b_pages;
}
prot = VM_PROT_READ;
if (uio->uio_rw == UIO_READ)
prot |= VM_PROT_WRITE; /* Less backwards than it looks */
error = 0;
for (i = 0; i < uio->uio_iovcnt; i++) {
#ifdef RACCT
if (racct_enable) {
PROC_LOCK(curproc);
if (uio->uio_rw == UIO_READ) {
racct_add_force(curproc, RACCT_READBPS,
uio->uio_iov[i].iov_len);
racct_add_force(curproc, RACCT_READIOPS, 1);
} else {
racct_add_force(curproc, RACCT_WRITEBPS,
uio->uio_iov[i].iov_len);
racct_add_force(curproc, RACCT_WRITEIOPS, 1);
}
PROC_UNLOCK(curproc);
}
#endif /* RACCT */
while (uio->uio_iov[i].iov_len) {
g_reset_bio(bp);
if (uio->uio_rw == UIO_READ) {
bp->bio_cmd = BIO_READ;
curthread->td_ru.ru_inblock++;
} else {
bp->bio_cmd = BIO_WRITE;
curthread->td_ru.ru_oublock++;
}
bp->bio_offset = uio->uio_offset;
bp->bio_data = uio->uio_iov[i].iov_base;
bp->bio_length = uio->uio_iov[i].iov_len;
if (bp->bio_length > dev->si_iosize_max)
bp->bio_length = dev->si_iosize_max;
if (bp->bio_length > MAXPHYS)
bp->bio_length = MAXPHYS;
/*
* Make sure the pbuf can map the request.
* The pbuf has kvasize = MAXPHYS, so a request
* larger than MAXPHYS - PAGE_SIZE must be
* page aligned or it will be fragmented.
*/
poff = (vm_offset_t)bp->bio_data & PAGE_MASK;
if (pbuf && bp->bio_length + poff > pbuf->b_kvasize) {
if (dev->si_flags & SI_NOSPLIT) {
uprintf("%s: request ptr %p is not "
"on a page boundary; cannot split "
"request\n", devtoname(dev),
bp->bio_data);
error = EFBIG;
goto doerror;
}
bp->bio_length = pbuf->b_kvasize;
if (poff != 0)
bp->bio_length -= PAGE_SIZE;
}
bp->bio_bcount = bp->bio_length;
bp->bio_dev = dev;
if (pages) {
if ((npages = vm_fault_quick_hold_pages(
&curproc->p_vmspace->vm_map,
(vm_offset_t)bp->bio_data, bp->bio_length,
prot, pages, maxpages)) < 0) {
error = EFAULT;
goto doerror;
}
if (pbuf && sa) {
pmap_qenter((vm_offset_t)sa,
pages, npages);
bp->bio_data = sa + poff;
} else {
bp->bio_ma = pages;
bp->bio_ma_n = npages;
bp->bio_ma_offset = poff;
bp->bio_data = unmapped_buf;
bp->bio_flags |= BIO_UNMAPPED;
}
}
csw->d_strategy(bp);
if (uio->uio_rw == UIO_READ)
biowait(bp, "physrd");
else
biowait(bp, "physwr");
if (pages) {
if (pbuf)
pmap_qremove((vm_offset_t)sa, npages);
vm_page_unhold_pages(pages, npages);
}
iolen = bp->bio_length - bp->bio_resid;
if (iolen == 0 && !(bp->bio_flags & BIO_ERROR))
goto doerror; /* EOF */
uio->uio_iov[i].iov_len -= iolen;
uio->uio_iov[i].iov_base =
(char *)uio->uio_iov[i].iov_base + iolen;
uio->uio_resid -= iolen;
uio->uio_offset += iolen;
if (bp->bio_flags & BIO_ERROR) {
error = bp->bio_error;
goto doerror;
}
}
}
doerror:
if (pbuf)
uma_zfree(pbuf_zone, pbuf);
else if (pages)
free(pages, M_DEVBUF);
g_destroy_bio(bp);
PRELE(curproc);
return (error);
}
/*
* Machine-dependent startup code
*/
void
cpu_startup(void)
{
vaddr_t v;
vsize_t sz;
vaddr_t minaddr, maxaddr;
msgbuf_vaddr = PMAP_DIRECT_MAP(msgbuf_paddr);
initmsgbuf((caddr_t)msgbuf_vaddr, round_page(MSGBUFSIZE));
printf("%s", version);
printf("real mem = %u (%uK)\n", ctob(physmem), ctob(physmem)/1024);
if (physmem >= btoc(1ULL << 32)) {
extern int amdgart_enable;
amdgart_enable = 1;
}
/*
* Find out how much space we need, allocate it,
* and then give everything true virtual addresses.
*/
sz = allocsys(0);
if ((v = uvm_km_zalloc(kernel_map, round_page(sz))) == 0)
panic("startup: no room for tables");
if (allocsys(v) - v != sz)
panic("startup: table size inconsistency");
/*
* Now allocate buffers proper. They are different than the above
* in that they usually occupy more virtual memory than physical.
*/
setup_buffers(&maxaddr);
/*
* Allocate a submap for exec arguments. This map effectively
* limits the number of processes exec'ing at any time.
*/
minaddr = vm_map_min(kernel_map);
exec_map = uvm_km_suballoc(kernel_map, &minaddr, &maxaddr,
16*NCARGS, VM_MAP_PAGEABLE, FALSE, NULL);
/*
* Allocate a submap for physio
*/
minaddr = vm_map_min(kernel_map);
phys_map = uvm_km_suballoc(kernel_map, &minaddr, &maxaddr,
VM_PHYS_SIZE, 0, FALSE, NULL);
printf("avail mem = %lu (%luK)\n", ptoa(uvmexp.free),
ptoa(uvmexp.free)/1024);
printf("using %u buffers containing %u bytes (%uK) of memory\n",
nbuf, bufpages * PAGE_SIZE, bufpages * PAGE_SIZE / 1024);
bufinit();
if (boothowto & RB_CONFIG) {
#ifdef BOOT_CONFIG
user_config();
#else
printf("kernel does not support - c; continuing..\n");
#endif
}
/* Safe for i/o port / memory space allocation to use malloc now. */
x86_bus_space_mallocok();
}
/*
* Early initialization, before main() is called.
*/
void
luna68k_init(void)
{
volatile uint8_t *pio0 = (void *)0x49000000;
int sw1, i;
char *cp;
extern char bootarg[64];
extern paddr_t avail_start, avail_end;
/* initialize cn_tab for early console */
#if 1
cn_tab = &syscons;
#else
cn_tab = &romcons;
#endif
/*
* Tell the VM system about available physical memory. The
* luna68k only has one segment.
*/
uvm_page_physload(atop(avail_start), atop(avail_end),
atop(avail_start), atop(avail_end), VM_FREELIST_DEFAULT);
/*
* Initialize error message buffer (at end of core).
* avail_end was pre-decremented in pmap_bootstrap to compensate.
*/
for (i = 0; i < btoc(MSGBUFSIZE); i++)
pmap_kenter_pa((vaddr_t)msgbufaddr + i * PAGE_SIZE,
avail_end + i * PAGE_SIZE, VM_PROT_READ|VM_PROT_WRITE, 0);
pmap_update(pmap_kernel());
initmsgbuf(msgbufaddr, m68k_round_page(MSGBUFSIZE));
pio0[3] = 0xb6;
pio0[2] = 1 << 6; /* enable parity check */
pio0[3] = 0xb6;
sw1 = pio0[0]; /* dip sw1 value */
sw1 ^= 0xff;
sysconsole = !(sw1 & 0x2); /* console selection */
/*
* Check if boothowto and bootdev values are passed by our bootloader.
*/
if ((bootdev & B_MAGICMASK) == B_DEVMAGIC) {
/* Valid value is set; no need to parse bootarg. */
return;
}
/*
* No valid bootdev value is set.
* Assume we are booted by ROM monitor directly using a.out kernel
* and we have to parse bootarg passed from the monitor to set
* proper boothowto and check netboot.
*/
/* set default to "sd0a" with no howto flags */
bootdev = MAKEBOOTDEV(0, LUNA68K_BOOTADPT_SPC, 0, 0, 0);
boothowto = 0;
/*
* 'bootarg' on LUNA has:
* "<args of x command> ENADDR=<addr> HOST=<host> SERVER=<name>"
* where <addr> is MAC address of which network loader used (not
* necessarily same as one at 0x4101.FFE0), <host> and <name>
* are the values of HOST and SERVER environment variables.
*
* 'bootarg' on LUNA-II has "<args of x command>" only.
*
* NetBSD/luna68k cares only the first argment; any of "sda".
*/
bootarg[63] = '\0';
for (cp = bootarg; *cp != '\0'; cp++) {
if (*cp == '-') {
char c;
while ((c = *cp) != '\0' && c != ' ') {
BOOT_FLAG(c, boothowto);
cp++;
}
} else if (*cp == 'E' && memcmp("ENADDR=", cp, 7) == 0) {
bootdev =
MAKEBOOTDEV(0, LUNA68K_BOOTADPT_LANCE, 0, 0, 0);
}
}
}
void
platform_start(__register_t a0 __unused, __register_t a1 __unused,
__register_t a2 __unused, __register_t a3 __unused)
{
uint64_t platform_counter_freq;
int argc = 0, i;
char **argv = NULL, **envp = NULL;
vm_offset_t kernend;
/*
* clear the BSS and SBSS segments, this should be first call in
* the function
*/
kernend = (vm_offset_t)&end;
memset(&edata, 0, kernend - (vm_offset_t)(&edata));
mips_postboot_fixup();
/* Initialize pcpu stuff */
mips_pcpu0_init();
/*
* Until some more sensible abstractions for uboot/redboot
* environment handling, we have to make this a compile-time
* hack. The existing code handles the uboot environment
* very incorrectly so we should just ignore initialising
* the relevant pointers.
*/
#ifndef AR71XX_ENV_UBOOT
argc = a0;
argv = (char**)a1;
envp = (char**)a2;
#endif
/*
* Protect ourselves from garbage in registers
*/
if (MIPS_IS_VALID_PTR(envp)) {
for (i = 0; envp[i]; i += 2) {
if (strcmp(envp[i], "memsize") == 0)
realmem = btoc(strtoul(envp[i+1], NULL, 16));
else if (strcmp(envp[i], "bootverbose") == 0)
bootverbose = btoc(strtoul(envp[i+1], NULL, 10));
}
}
bootverbose = 1;
#ifdef AR71XX_ENV_ROUTERBOOT
/*
* RouterBoot informs the board memory as a command line argument.
*/
if (realmem == 0)
realmem = ar71xx_routerboot_get_mem(argc, argv);
#endif
/*
* Just wild guess. RedBoot let us down and didn't reported
* memory size
*/
if (realmem == 0)
realmem = btoc(32*1024*1024);
/*
* Allow build-time override in case Redboot lies
* or in other situations (eg where there's u-boot)
* where there isn't (yet) a convienent method of
* being told how much RAM is available.
*
* This happens on at least the Ubiquiti LS-SR71A
* board, where redboot says there's 16mb of RAM
* but in fact there's 32mb.
*/
#if defined(AR71XX_REALMEM)
realmem = btoc(AR71XX_REALMEM);
#endif
/* phys_avail regions are in bytes */
phys_avail[0] = MIPS_KSEG0_TO_PHYS(kernel_kseg0_end);
phys_avail[1] = ctob(realmem);
dump_avail[0] = phys_avail[0];
dump_avail[1] = phys_avail[1] - phys_avail[0];
physmem = realmem;
/*
* ns8250 uart code uses DELAY so ticker should be inititalized
* before cninit. And tick_init_params refers to hz, so * init_param1
* should be called first.
*/
init_param1();
/* Detect the system type - this is needed for subsequent chipset-specific calls */
ar71xx_detect_sys_type();
ar71xx_detect_sys_frequency();
platform_counter_freq = ar71xx_cpu_freq();
mips_timer_init_params(platform_counter_freq, 1);
cninit();
init_static_kenv(boot1_env, sizeof(boot1_env));
printf("CPU platform: %s\n", ar71xx_get_system_type());
printf("CPU Frequency=%d MHz\n", u_ar71xx_cpu_freq / 1000000);
printf("CPU DDR Frequency=%d MHz\n", u_ar71xx_ddr_freq / 1000000);
printf("CPU AHB Frequency=%d MHz\n", u_ar71xx_ahb_freq / 1000000);
printf("platform frequency: %lld MHz\n", platform_counter_freq / 1000000);
printf("CPU reference clock: %d MHz\n", u_ar71xx_refclk / 1000000);
printf("CPU MDIO clock: %d MHz\n", u_ar71xx_mdio_freq / 1000000);
printf("arguments: \n");
printf(" a0 = %08x\n", a0);
printf(" a1 = %08x\n", a1);
printf(" a2 = %08x\n", a2);
printf(" a3 = %08x\n", a3);
strcpy(cpu_model, ar71xx_get_system_type());
/*
* XXX this code is very redboot specific.
*/
printf("Cmd line:");
if (MIPS_IS_VALID_PTR(argv)) {
for (i = 0; i < argc; i++) {
printf(" %s", argv[i]);
parse_argv(argv[i]);
}
}
else
printf ("argv is invalid");
printf("\n");
printf("Environment:\n");
if (MIPS_IS_VALID_PTR(envp)) {
for (i = 0; envp[i]; i+=2) {
printf(" %s = %s\n", envp[i], envp[i+1]);
kern_setenv(envp[i], envp[i+1]);
}
}
else
printf ("envp is invalid\n");
/* Platform setup */
init_param2(physmem);
mips_cpu_init();
pmap_bootstrap();
mips_proc0_init();
mutex_init();
/*
* Reset USB devices
*/
ar71xx_init_usb_peripheral();
/*
* Reset internal ethernet switch, if one exists
*/
ar71xx_reset_ethernet_switch();
/*
* Initialise the gmac driver.
*/
ar71xx_init_gmac();
/* Redboot if_arge MAC address is in the environment */
(void) ar71xx_redboot_get_macaddr();
/* Various other boards need things to come out of EEPROM */
(void) ar71xx_platform_read_eeprom_mac();
/* Initialise the MAC address hint map */
ar71xx_platform_check_mac_hints();
kdb_init();
#ifdef KDB
if (boothowto & RB_KDB)
kdb_enter(KDB_WHY_BOOTFLAGS, "Boot flags requested debugger");
#endif
}
/*
* Release a buffer on to the free lists.
* Described in Bach (p. 46).
*/
void
brelse(struct buf *bp)
{
struct bqueues *bufq;
int s;
/* Block disk interrupts. */
s = splbio();
/*
* Determine which queue the buffer should be on, then put it there.
*/
/* If it's locked, don't report an error; try again later. */
if (ISSET(bp->b_flags, (B_LOCKED|B_ERROR)) == (B_LOCKED|B_ERROR))
CLR(bp->b_flags, B_ERROR);
/* If it's not cacheable, or an error, mark it invalid. */
if (ISSET(bp->b_flags, (B_NOCACHE|B_ERROR)))
SET(bp->b_flags, B_INVAL);
if ((bp->b_bufsize <= 0) || ISSET(bp->b_flags, B_INVAL)) {
/*
* If it's invalid or empty, dissociate it from its vnode
* and put on the head of the appropriate queue.
*/
if (LIST_FIRST(&bp->b_dep) != NULL)
buf_deallocate(bp);
if (ISSET(bp->b_flags, B_DELWRI)) {
CLR(bp->b_flags, B_DELWRI);
}
if (bp->b_vp) {
reassignbuf(bp);
brelvp(bp);
}
if (bp->b_bufsize <= 0) {
/* no data */
bufq = &bufqueues[BQ_EMPTY];
numemptybufs++;
} else {
/* invalid data */
bufq = &bufqueues[BQ_CLEAN];
numfreepages += btoc(bp->b_bufsize);
numcleanpages += btoc(bp->b_bufsize);
}
binsheadfree(bp, bufq);
} else {
/*
* It has valid data. Put it on the end of the appropriate
* queue, so that it'll stick around for as long as possible.
*/
if (ISSET(bp->b_flags, B_LOCKED))
/* locked in core */
bufq = &bufqueues[BQ_LOCKED];
else {
numfreepages += btoc(bp->b_bufsize);
if (!ISSET(bp->b_flags, B_DELWRI)) {
numcleanpages += btoc(bp->b_bufsize);
bufq = &bufqueues[BQ_CLEAN];
} else {
numdirtypages += btoc(bp->b_bufsize);
bufq = &bufqueues[BQ_DIRTY];
}
}
if (ISSET(bp->b_flags, B_AGE))
binsheadfree(bp, bufq);
else
binstailfree(bp, bufq);
}
/* Unlock the buffer. */
CLR(bp->b_flags, (B_AGE | B_ASYNC | B_BUSY | B_NOCACHE | B_DEFERRED));
/* Wake up syncer and cleaner processes waiting for buffers */
if (nobuffers) {
wakeup(&nobuffers);
nobuffers = 0;
}
/* Wake up any processes waiting for any buffer to become free. */
if (needbuffer && (numcleanpages > locleanpages)) {
needbuffer--;
wakeup_one(&needbuffer);
}
splx(s);
/* Wake up any processes waiting for _this_ buffer to become free. */
if (ISSET(bp->b_flags, B_WANTED)) {
CLR(bp->b_flags, B_WANTED);
wakeup(bp);
}
}
/*
struct vnop_getpages_args {
struct vnode *a_vp;
vm_page_t *a_m;
int a_count;
int a_reqpage;
vm_ooffset_t a_offset;
};
*/
static int
fuse_vnop_getpages(struct vop_getpages_args *ap)
{
int i, error, nextoff, size, toff, count, npages;
struct uio uio;
struct iovec iov;
vm_offset_t kva;
struct buf *bp;
struct vnode *vp;
struct thread *td;
struct ucred *cred;
vm_page_t *pages;
FS_DEBUG2G("heh\n");
vp = ap->a_vp;
KASSERT(vp->v_object, ("objectless vp passed to getpages"));
td = curthread; /* XXX */
cred = curthread->td_ucred; /* XXX */
pages = ap->a_m;
count = ap->a_count;
if (!fsess_opt_mmap(vnode_mount(vp))) {
FS_DEBUG("called on non-cacheable vnode??\n");
return (VM_PAGER_ERROR);
}
npages = btoc(count);
/*
* If the requested page is partially valid, just return it and
* allow the pager to zero-out the blanks. Partially valid pages
* can only occur at the file EOF.
*/
VM_OBJECT_WLOCK(vp->v_object);
fuse_vm_page_lock_queues();
if (pages[ap->a_reqpage]->valid != 0) {
for (i = 0; i < npages; ++i) {
if (i != ap->a_reqpage) {
fuse_vm_page_lock(pages[i]);
vm_page_free(pages[i]);
fuse_vm_page_unlock(pages[i]);
}
}
fuse_vm_page_unlock_queues();
VM_OBJECT_WUNLOCK(vp->v_object);
return 0;
}
fuse_vm_page_unlock_queues();
VM_OBJECT_WUNLOCK(vp->v_object);
/*
* We use only the kva address for the buffer, but this is extremely
* convienient and fast.
*/
bp = getpbuf(&fuse_pbuf_freecnt);
kva = (vm_offset_t)bp->b_data;
pmap_qenter(kva, pages, npages);
PCPU_INC(cnt.v_vnodein);
PCPU_ADD(cnt.v_vnodepgsin, npages);
iov.iov_base = (caddr_t)kva;
iov.iov_len = count;
uio.uio_iov = &iov;
uio.uio_iovcnt = 1;
uio.uio_offset = IDX_TO_OFF(pages[0]->pindex);
uio.uio_resid = count;
uio.uio_segflg = UIO_SYSSPACE;
uio.uio_rw = UIO_READ;
uio.uio_td = td;
error = fuse_io_dispatch(vp, &uio, IO_DIRECT, cred);
pmap_qremove(kva, npages);
relpbuf(bp, &fuse_pbuf_freecnt);
if (error && (uio.uio_resid == count)) {
FS_DEBUG("error %d\n", error);
VM_OBJECT_WLOCK(vp->v_object);
fuse_vm_page_lock_queues();
for (i = 0; i < npages; ++i) {
if (i != ap->a_reqpage) {
fuse_vm_page_lock(pages[i]);
vm_page_free(pages[i]);
fuse_vm_page_unlock(pages[i]);
}
}
fuse_vm_page_unlock_queues();
VM_OBJECT_WUNLOCK(vp->v_object);
return VM_PAGER_ERROR;
}
/*
* Calculate the number of bytes read and validate only that number
* of bytes. Note that due to pending writes, size may be 0. This
* does not mean that the remaining data is invalid!
*/
size = count - uio.uio_resid;
VM_OBJECT_WLOCK(vp->v_object);
fuse_vm_page_lock_queues();
for (i = 0, toff = 0; i < npages; i++, toff = nextoff) {
vm_page_t m;
nextoff = toff + PAGE_SIZE;
m = pages[i];
if (nextoff <= size) {
/*
* Read operation filled an entire page
*/
m->valid = VM_PAGE_BITS_ALL;
KASSERT(m->dirty == 0,
("fuse_getpages: page %p is dirty", m));
} else if (size > toff) {
/*
* Read operation filled a partial page.
*/
m->valid = 0;
vm_page_set_valid_range(m, 0, size - toff);
KASSERT(m->dirty == 0,
("fuse_getpages: page %p is dirty", m));
} else {
/*
* Read operation was short. If no error occured
* we may have hit a zero-fill section. We simply
* leave valid set to 0.
*/
;
}
if (i != ap->a_reqpage)
vm_page_readahead_finish(m);
}
fuse_vm_page_unlock_queues();
VM_OBJECT_WUNLOCK(vp->v_object);
return 0;
}
/*
* Do all the stuff that locore normally does before calling main().
* Process arguments passed to us by the prom monitor.
* Return the first page address following the system.
*/
void
mach_init(int x_boothowto, int x_bootdev, int x_bootname, int x_maxmem)
{
u_long first, last;
char *kernend;
struct btinfo_magic *bi_magic;
struct btinfo_bootarg *bi_arg;
struct btinfo_systype *bi_systype;
#if NKSYMS || defined(DDB) || defined(MODULAR)
struct btinfo_symtab *bi_sym;
int nsym = 0;
char *ssym, *esym;
ssym = esym = NULL; /* XXX: gcc */
#endif
bi_arg = NULL;
bootinfo = (void *)BOOTINFO_ADDR; /* XXX */
bi_magic = lookup_bootinfo(BTINFO_MAGIC);
if (bi_magic && bi_magic->magic == BOOTINFO_MAGIC) {
bi_arg = lookup_bootinfo(BTINFO_BOOTARG);
if (bi_arg) {
x_boothowto = bi_arg->howto;
x_bootdev = bi_arg->bootdev;
x_maxmem = bi_arg->maxmem;
}
#if NKSYMS || defined(DDB) || defined(MODULAR)
bi_sym = lookup_bootinfo(BTINFO_SYMTAB);
if (bi_sym) {
nsym = bi_sym->nsym;
ssym = (void *)bi_sym->ssym;
esym = (void *)bi_sym->esym;
}
#endif
bi_systype = lookup_bootinfo(BTINFO_SYSTYPE);
if (bi_systype)
systype = bi_systype->type;
} else {
/*
* Running kernel is loaded by non-native loader;
* clear the BSS segment here.
*/
memset(edata, 0, end - edata);
}
if (systype == 0)
systype = NEWS3400; /* XXX compatibility for old boot */
#ifdef news5000
if (systype == NEWS5000) {
int i;
char *bootspec = (char *)x_bootdev;
if (bi_arg == NULL)
panic("news5000 requires BTINFO_BOOTARG to boot");
_sip = (void *)bi_arg->sip;
x_maxmem = _sip->apbsi_memsize;
x_maxmem -= 0x00100000; /* reserve 1MB for ROM monitor */
if (strncmp(bootspec, "scsi", 4) == 0) {
x_bootdev = (5 << 28) | 0; /* magic, sd */
bootspec += 4;
if (*bootspec != '(' /*)*/)
goto bootspec_end;
i = strtoul(bootspec + 1, &bootspec, 10);
x_bootdev |= (i << 24); /* bus */
if (*bootspec != ',')
goto bootspec_end;
i = strtoul(bootspec + 1, &bootspec, 10);
x_bootdev |= (i / 10) << 20; /* controller */
x_bootdev |= (i % 10) << 16; /* unit */
if (*bootspec != ',')
goto bootspec_end;
i = strtoul(bootspec + 1, &bootspec, 10);
x_bootdev |= (i << 8); /* partition */
}
bootspec_end:
consinit();
}
#endif
/*
* Save parameters into kernel work area.
*/
*(int *)(MIPS_PHYS_TO_KSEG1(MACH_MAXMEMSIZE_ADDR)) = x_maxmem;
*(int *)(MIPS_PHYS_TO_KSEG1(MACH_BOOTDEV_ADDR)) = x_bootdev;
*(int *)(MIPS_PHYS_TO_KSEG1(MACH_BOOTSW_ADDR)) = x_boothowto;
kernend = (char *)mips_round_page(end);
#if NKSYMS || defined(DDB) || defined(MODULAR)
if (nsym)
kernend = (char *)mips_round_page(esym);
#endif
/*
* Set the VM page size.
*/
uvm_setpagesize();
boothowto = x_boothowto;
bootdev = x_bootdev;
physmem = btoc(x_maxmem);
/*
* Now that we know how much memory we have, initialize the
* mem cluster array.
*/
mem_clusters[0].start = 0; /* XXX is this correct? */
mem_clusters[0].size = ctob(physmem);
mem_cluster_cnt = 1;
/*
* Copy exception-dispatch code down to exception vector.
* Initialize locore-function vector.
* Clear out the I and D caches.
*/
mips_vector_init(NULL, false);
/*
* We know the CPU type now. Initialize our DMA tags (might
* need this early).
*/
newsmips_bus_dma_init();
#if NKSYMS || defined(DDB) || defined(MODULAR)
if (nsym)
ksyms_addsyms_elf(esym - ssym, ssym, esym);
#endif
#ifdef KADB
boothowto |= RB_KDB;
#endif
/*
* Check to see if a mini-root was loaded into memory. It resides
* at the start of the next page just after the end of BSS.
*/
if (boothowto & RB_MINIROOT)
kernend += round_page(mfs_initminiroot(kernend));
/*
* Load the rest of the available pages into the VM system.
*/
first = round_page(MIPS_KSEG0_TO_PHYS(kernend));
last = mem_clusters[0].start + mem_clusters[0].size;
uvm_page_physload(atop(first), atop(last), atop(first), atop(last),
VM_FREELIST_DEFAULT);
/*
* Initialize error message buffer (at end of core).
*/
mips_init_msgbuf();
/*
* Initialize the virtual memory system.
*/
pmap_bootstrap();
/*
* Allocate uarea page for lwp0 and set it.
*/
mips_init_lwp0_uarea();
/*
* Determine what model of computer we are running on.
*/
switch (systype) {
#ifdef news3400
case NEWS3400:
news3400_init();
strcpy(cpu_model, idrom.id_machine);
if (strcmp(cpu_model, "news3400") == 0 ||
strcmp(cpu_model, "news3200") == 0 ||
strcmp(cpu_model, "news3700") == 0) {
/*
* Set up interrupt handling and I/O addresses.
*/
hardware_intr = news3400_intr;
cpuspeed = 10;
} else {
printf("kernel not configured for machine %s\n",
cpu_model);
}
break;
#endif
#ifdef news5000
case NEWS5000:
news5000_init();
strcpy(cpu_model, idrom.id_machine);
if (strcmp(cpu_model, "news5000") == 0 ||
strcmp(cpu_model, "news5900") == 0) {
/*
* Set up interrupt handling and I/O addresses.
*/
hardware_intr = news5000_intr;
cpuspeed = 50; /* ??? XXX */
} else {
printf("kernel not configured for machine %s\n",
cpu_model);
}
break;
#endif
default:
printf("kernel not configured for systype %d\n", systype);
break;
}
}
/*
struct vnop_putpages_args {
struct vnode *a_vp;
vm_page_t *a_m;
int a_count;
int a_sync;
int *a_rtvals;
vm_ooffset_t a_offset;
};
*/
static int
fuse_vnop_putpages(struct vop_putpages_args *ap)
{
struct uio uio;
struct iovec iov;
vm_offset_t kva;
struct buf *bp;
int i, error, npages, count;
off_t offset;
int *rtvals;
struct vnode *vp;
struct thread *td;
struct ucred *cred;
vm_page_t *pages;
vm_ooffset_t fsize;
FS_DEBUG2G("heh\n");
vp = ap->a_vp;
KASSERT(vp->v_object, ("objectless vp passed to putpages"));
fsize = vp->v_object->un_pager.vnp.vnp_size;
td = curthread; /* XXX */
cred = curthread->td_ucred; /* XXX */
pages = ap->a_m;
count = ap->a_count;
rtvals = ap->a_rtvals;
npages = btoc(count);
offset = IDX_TO_OFF(pages[0]->pindex);
if (!fsess_opt_mmap(vnode_mount(vp))) {
FS_DEBUG("called on non-cacheable vnode??\n");
}
for (i = 0; i < npages; i++)
rtvals[i] = VM_PAGER_AGAIN;
/*
* When putting pages, do not extend file past EOF.
*/
if (offset + count > fsize) {
count = fsize - offset;
if (count < 0)
count = 0;
}
/*
* We use only the kva address for the buffer, but this is extremely
* convienient and fast.
*/
bp = getpbuf(&fuse_pbuf_freecnt);
kva = (vm_offset_t)bp->b_data;
pmap_qenter(kva, pages, npages);
PCPU_INC(cnt.v_vnodeout);
PCPU_ADD(cnt.v_vnodepgsout, count);
iov.iov_base = (caddr_t)kva;
iov.iov_len = count;
uio.uio_iov = &iov;
uio.uio_iovcnt = 1;
uio.uio_offset = offset;
uio.uio_resid = count;
uio.uio_segflg = UIO_SYSSPACE;
uio.uio_rw = UIO_WRITE;
uio.uio_td = td;
error = fuse_io_dispatch(vp, &uio, IO_DIRECT, cred);
pmap_qremove(kva, npages);
relpbuf(bp, &fuse_pbuf_freecnt);
if (!error) {
int nwritten = round_page(count - uio.uio_resid) / PAGE_SIZE;
for (i = 0; i < nwritten; i++) {
rtvals[i] = VM_PAGER_OK;
VM_OBJECT_WLOCK(pages[i]->object);
vm_page_undirty(pages[i]);
VM_OBJECT_WUNLOCK(pages[i]->object);
}
}
return rtvals[0];
}
/*
* Do all the stuff that locore normally does before calling main().
*/
void
mach_init(int32_t memsize32, u_int bim, int32_t bip32)
{
intptr_t memsize = (int32_t)memsize32;
char *kernend;
char *bip = (char *)(intptr_t)(int32_t)bip32;
u_long first, last;
extern char edata[], end[];
const char *bi_msg;
#if NKSYMS || defined(DDB) || defined(MODULAR)
char *ssym = 0;
struct btinfo_symtab *bi_syms;
#endif
struct btinfo_howto *bi_howto;
/*
* Clear the BSS segment (if needed).
*/
if (memcmp(((Elf_Ehdr *)end)->e_ident, ELFMAG, SELFMAG) == 0 &&
((Elf_Ehdr *)end)->e_ident[EI_CLASS] == ELFCLASS) {
esym = end;
#if NKSYMS || defined(DDB) || defined(MODULAR)
esym += ((Elf_Ehdr *)end)->e_entry;
#endif
kernend = (char *)mips_round_page(esym);
/*
* We don't have to clear BSS here
* since our bootloader already does it.
*/
#if 0
memset(edata, 0, end - edata);
#endif
} else {
kernend = (void *)mips_round_page(end);
/*
* No symbol table, so assume we are loaded by
* the firmware directly with "bfd" command.
* The firmware loader doesn't clear BSS of
* a loaded kernel, so do it here.
*/
memset(edata, 0, kernend - edata);
}
/*
* Copy exception-dispatch code down to exception vector.
* Initialize locore-function vector.
* Clear out the I and D caches.
*/
mips_vector_init(NULL, false);
/* Check for valid bootinfo passed from bootstrap */
if (bim == BOOTINFO_MAGIC) {
struct btinfo_magic *bi_magic;
bootinfo = bip;
bi_magic = lookup_bootinfo(BTINFO_MAGIC);
if (bi_magic == NULL) {
bi_msg = "missing bootinfo structure";
bim = (uintptr_t)bip;
} else if (bi_magic->magic != BOOTINFO_MAGIC) {
bi_msg = "invalid bootinfo structure";
bim = bi_magic->magic;
} else
bi_msg = NULL;
} else {
bi_msg = "invalid bootinfo (standalone boot?)";
}
#if NKSYMS || defined(DDB) || defined(MODULAR)
bi_syms = lookup_bootinfo(BTINFO_SYMTAB);
/* Load symbol table if present */
if (bi_syms != NULL) {
ssym = (void *)(intptr_t)bi_syms->ssym;
esym = (void *)(intptr_t)bi_syms->esym;
kernend = (void *)mips_round_page(esym);
}
#endif
bi_howto = lookup_bootinfo(BTINFO_HOWTO);
if (bi_howto != NULL)
boothowto = bi_howto->bi_howto;
cobalt_id = read_board_id();
if (cobalt_id >= COBALT_MODELS || cobalt_model[cobalt_id] == NULL)
cpu_setmodel("Cobalt unknown model (board ID %u)",
cobalt_id);
else
cpu_setmodel("%s", cobalt_model[cobalt_id]);
switch (cobalt_id) {
case COBALT_ID_QUBE2700:
case COBALT_ID_RAQ:
cpuspeed = 150; /* MHz */
break;
case COBALT_ID_QUBE2:
case COBALT_ID_RAQ2:
cpuspeed = 250; /* MHz */
break;
default:
/* assume the fastest, so that delay(9) works */
cpuspeed = 250;
break;
}
curcpu()->ci_cpu_freq = cpuspeed * 1000 * 1000;
curcpu()->ci_cycles_per_hz = (curcpu()->ci_cpu_freq + hz / 2) / hz;
curcpu()->ci_divisor_delay =
((curcpu()->ci_cpu_freq + (1000000 / 2)) / 1000000);
/* all models have Rm5200, which is CPU_MIPS_DOUBLE_COUNT */
curcpu()->ci_cycles_per_hz /= 2;
curcpu()->ci_divisor_delay /= 2;
physmem = btoc(memsize - MIPS_KSEG0_START);
consinit();
KASSERT(&lwp0 == curlwp);
if (bi_msg != NULL)
printf("%s: magic=%#x bip=%p\n", bi_msg, bim, bip);
uvm_setpagesize();
/*
* The boot command is passed in the top 512 bytes,
* so don't clobber that.
*/
mem_clusters[0].start = 0;
mem_clusters[0].size = ctob(physmem) - 512;
mem_cluster_cnt = 1;
memcpy(bootstring, (char *)(memsize - 512), 512);
memset((char *)(memsize - 512), 0, 512);
bootstring[511] = '\0';
decode_bootstring();
#if NKSYMS || defined(DDB) || defined(MODULAR)
/* init symbols if present */
if ((bi_syms != NULL) && (esym != NULL))
ksyms_addsyms_elf(esym - ssym, ssym, esym);
#endif
KASSERT(&lwp0 == curlwp);
#ifdef DDB
if (boothowto & RB_KDB)
Debugger();
#endif
#ifdef KGDB
if (boothowto & RB_KDB)
kgdb_connect(0);
#endif
/*
* Load the rest of the available pages into the VM system.
*/
first = round_page(MIPS_KSEG0_TO_PHYS(kernend));
last = mem_clusters[0].start + mem_clusters[0].size;
uvm_page_physload(atop(first), atop(last), atop(first), atop(last),
VM_FREELIST_DEFAULT);
/*
* Initialize error message buffer (at end of core).
*/
mips_init_msgbuf();
pmap_bootstrap();
/*
* Allocate space for proc0's USPACE.
*/
mips_init_lwp0_uarea();
}
void
mach_init(int argc, char **argv, yamon_env_var *envp, u_long memsize)
{
bus_space_handle_t sh;
void *kernend;
const char *cp;
u_long first, last;
void *v;
int freqok, howto, i;
const struct alchemy_board *board;
extern char edata[], end[]; /* XXX */
board = board_info();
KASSERT(board != NULL);
/* clear the BSS segment */
kernend = (void *)mips_round_page(end);
memset(edata, 0, (char *)kernend - edata);
/* set CPU model info for sysctl_hw */
strcpy(cpu_model, board->ab_name);
/* save the yamon environment pointer */
yamon_envp = envp;
/* Use YAMON callbacks for early console I/O */
cn_tab = &yamon_promcd;
/*
* Set up the exception vectors and CPU-specific function
* vectors early on. We need the wbflush() vector set up
* before comcnattach() is called (or at least before the
* first printf() after that is called).
* Sets up mips_cpu_flags that may be queried by other
* functions called during startup.
* Also clears the I+D caches.
*/
mips_vector_init();
/*
* Set the VM page size.
*/
uvm_setpagesize();
/*
* Use YAMON's CPU frequency if available.
*/
freqok = yamon_setcpufreq(1);
/*
* Initialize bus space tags.
*/
au_cpureg_bus_mem_init(&alchemy_cpuregt, &alchemy_cpuregt);
aubus_st = &alchemy_cpuregt;
/*
* Calibrate the timer if YAMON failed to tell us.
*/
if (!freqok) {
bus_space_map(aubus_st, PC_BASE, PC_SIZE, 0, &sh);
au_cal_timers(aubus_st, sh);
bus_space_unmap(aubus_st, sh, PC_SIZE);
}
/*
* Perform board-specific initialization.
*/
board->ab_init();
/*
* Bring up the console.
*/
#if NCOM > 0
#ifdef CONSPEED
if (aucomcnrate == 0)
aucomcnrate = CONSPEED;
#else /* !CONSPEED */
/*
* Learn default console speed. We use the YAMON environment,
* though we could probably also figure it out by checking the
* aucom registers directly.
*/
if ((aucomcnrate == 0) && ((cp = yamon_getenv("modetty0")) != NULL))
aucomcnrate = strtoul(cp, NULL, 0);
if (aucomcnrate == 0) {
printf("FATAL: `modetty0' YAMON variable not set. Set it\n");
printf(" to the speed of the console and try again.\n");
printf(" Or, build a kernel with the `CONSPEED' "
"option.\n");
panic("mach_init");
}
#endif /* CONSPEED */
/*
* Delay to allow firmware putchars to complete.
* FIFO depth * character time.
* character time = (1000000 / (defaultrate / 10))
*/
delay(160000000 / aucomcnrate);
if (com_aubus_cnattach(UART0_BASE, aucomcnrate) != 0)
panic("mach_init: unable to initialize serial console");
#else /* NCOM > 0 */
panic("mach_init: not configured to use serial console");
#endif /* NAUCOM > 0 */
/*
* Look at arguments passed to us and compute boothowto.
*/
boothowto = RB_AUTOBOOT;
#ifdef KADB
boothowto |= RB_KDB;
#endif
for (i = 1; i < argc; i++) {
for (cp = argv[i]; *cp; cp++) {
/* Ignore superfluous '-', if there is one */
if (*cp == '-')
continue;
howto = 0;
BOOT_FLAG(*cp, howto);
if (! howto)
printf("bootflag '%c' not recognised\n", *cp);
else
boothowto |= howto;
}
}
/*
* Determine the memory size. Use the `memsize' PMON
* variable. If that's not available, panic.
*
* Note: Reserve the first page! That's where the trap
* vectors are located.
*/
#if defined(MEMSIZE)
memsize = MEMSIZE;
#else
if (memsize == 0) {
if ((cp = yamon_getenv("memsize")) != NULL)
memsize = strtoul(cp, NULL, 0);
else {
printf("FATAL: `memsize' YAMON variable not set. Set it to\n");
printf(" the amount of memory (in MB) and try again.\n");
printf(" Or, build a kernel with the `MEMSIZE' "
"option.\n");
panic("mach_init");
}
}
#endif /* MEMSIZE */
printf("Memory size: 0x%08lx\n", memsize);
physmem = btoc(memsize);
mem_clusters[mem_cluster_cnt].start = PAGE_SIZE;
mem_clusters[mem_cluster_cnt].size =
memsize - mem_clusters[mem_cluster_cnt].start;
mem_cluster_cnt++;
/*
* Load the rest of the available pages into the VM system.
*/
first = round_page(MIPS_KSEG0_TO_PHYS(kernend));
last = mem_clusters[0].start + mem_clusters[0].size;
uvm_page_physload(atop(first), atop(last), atop(first), atop(last),
VM_FREELIST_DEFAULT);
/*
* Initialize message buffer (at end of core).
*/
mips_init_msgbuf();
/*
* Initialize the virtual memory system.
*/
pmap_bootstrap();
/*
* Init mapping for u page(s) for proc0.
*/
v = (void *) uvm_pageboot_alloc(USPACE);
lwp0.l_addr = proc0paddr = (struct user *)v;
lwp0.l_md.md_regs = (struct frame *)((char *)v + USPACE) - 1;
proc0paddr->u_pcb.pcb_context[11] =
MIPS_INT_MASK | MIPS_SR_INT_IE; /* SR */
/*
* Initialize debuggers, and break into them, if appropriate.
*/
#if NKSYMS || defined(DDB) || defined(LKM)
ksyms_init(0, 0, 0);
#endif
#ifdef DDB
if (boothowto & RB_KDB)
Debugger();
#endif
}
