void
ibd_init(void)
{
pnode_t chosen;
char *mtuprop = "ipib-frame-size";
char *bcastprop = "ipib-broadcast";
char *addrprop = "ipib-address";
char *cidprop = "client-id";
int cidlen;
uint8_t dhcpcid[DHCP_MAX_CID_LEN];
mac_state.mac_addr_len = IPOIB_ADDRL;
mac_state.mac_addr_buf = bkmem_alloc(mac_state.mac_addr_len);
if (mac_state.mac_addr_buf == NULL)
prom_panic("ibd_init: Cannot allocate memory.");
chosen = prom_finddevice("/chosen");
if (chosen == OBP_NONODE || chosen == OBP_BADNODE)
prom_panic("ibd_init: Cannot find /chosen.");
if (prom_getprop(chosen, addrprop, (caddr_t)mac_state.mac_addr_buf) !=
IPOIB_ADDRL)
prom_panic("ibd_init: Cannot find /chosen:ipib-address\n.");
if (prom_getprop(chosen, bcastprop, (caddr_t)&ibdbroadcastaddr) !=
IPOIB_ADDRL)
prom_panic("ibd_init: Cannot find /chosen:ipib-broadcast\n.");
if (((cidlen = prom_getproplen(chosen, cidprop)) <= 0) ||
(cidlen > DHCP_MAX_CID_LEN) || (prom_getprop(chosen, cidprop,
(caddr_t)&dhcpcid) != cidlen))
prom_panic("ibd_init: Invalid /chosen:client-id\n.");
dhcp_set_client_id(dhcpcid, cidlen);
/*
* Note that prom reports mtu including 20 bytes of
* addressing information.
*/
if (prom_getprop(chosen, mtuprop,
(caddr_t)&mac_state.mac_mtu) <= 0)
mac_state.mac_mtu = IBDSIZE + IPOIB_ADDRL;
/*
* Tell upper layers that we can support a little
* more. We will be taking off these 20 bytes at
* the start before we invoke prom_write() to send
* over the wire.
*/
mac_state.mac_arp_timeout = IBD_ARP_TIMEOUT;
mac_state.mac_in_timeout = IBD_IN_TIMEOUT;
mac_state.mac_arp = ibd_arp;
mac_state.mac_rarp = ibd_revarp;
mac_state.mac_header_len = ibd_header_len;
mac_state.mac_input = ibd_input;
mac_state.mac_output = ibd_output;
}
/*
* prom_walk_devs() implements a generic walker for the OBP tree; this
* implementation uses an explicitly managed stack in order to save the
* overhead of a recursive implementation.
*/
void
prom_walk_devs(pnode_t node, int (*cb)(pnode_t, void *, void *), void *arg,
void *result)
{
pnode_t stack[OBP_STACKDEPTH];
int stackidx = 0;
if (node == OBP_NONODE || node == OBP_BADNODE) {
prom_panic("Invalid node specified as root of prom tree walk");
}
stack[0] = node;
for (;;) {
pnode_t curnode = stack[stackidx];
pnode_t child;
/*
* We're out of stuff to do at this level, bump back up a level
* in the tree, and move to the next node; if the new level
* will be level -1, we're done.
*/
if (curnode == OBP_NONODE || curnode == OBP_BADNODE) {
stackidx--;
if (stackidx < 0)
return;
stack[stackidx] = prom_nextnode(stack[stackidx]);
continue;
}
switch ((*cb)(curnode, arg, result)) {
case PROM_WALK_TERMINATE:
return;
case PROM_WALK_CONTINUE:
/*
* If curnode has a child, traverse to it,
* otherwise move to curnode's sibling.
*/
child = prom_childnode(curnode);
if (child != OBP_NONODE && child != OBP_BADNODE) {
stackidx++;
stack[stackidx] = child;
} else {
stack[stackidx] =
prom_nextnode(stack[stackidx]);
}
break;
default:
prom_panic("unrecognized walk directive");
}
}
}
void
resalloc_init(void)
{
char iarch[128];
if (impl_name(iarch, sizeof (iarch)) < 0) {
dprintf("boot: resalloc_init: failed to read iarch\n");
return;
}
dprintf("boot: resalloc_init: got iarch %s\n", iarch);
/*
* Some versions of SG/LW8 firmware can actually handle the entire 10MB,
* but we don't have the ability to check for the firmware version here.
*/
if (strcmp(iarch, "SUNW,Sun-Fire") == 0 ||
strcmp(iarch, "SUNW,Netra-T12") == 0) {
is_sg = 1;
sg_addr = MAPPEDMEM_MINTOP;
sg_len = MAPPEDMEM_FULLTOP - MAPPEDMEM_MINTOP;
if (prom_alloc(sg_addr, sg_len, 1) != sg_addr)
prom_panic("can't extend sg bootmem");
}
top_bootmem = MAPPEDMEM_FULLTOP;
dprintf("boot: resalloc_init: boosted top_bootmem to %p\n",
(void *)top_bootmem);
}
/*
* ARP client side
* Broadcasts to determine MAC address given network order IP address.
* See RFC 826
*
* Returns TRUE if successful, FALSE otherwise.
*/
static int
ibd_arp(struct in_addr *ip, void *hap, uint32_t timeout)
{
ipoib_mac_t *ep = (ipoib_mac_t *)hap;
struct arp_packet out;
int result;
if (!initialized)
prom_panic("IPoIB device is not initialized.");
bzero((char *)&out, sizeof (struct arp_packet));
out.arp_eh.ipoib_rhdr.ipoib_type = htons(ETHERTYPE_ARP);
out.arp_ea.arp_op = htons(ARPOP_REQUEST);
bcopy((caddr_t)&ibdbroadcastaddr, (caddr_t)&out.arp_ea.arp_tha,
IPOIB_ADDRL);
bcopy((caddr_t)ip, (caddr_t)out.arp_ea.arp_tpa,
sizeof (struct in_addr));
result = ibd_comarp(&out, timeout);
if (result && (ep != NULL)) {
bcopy((caddr_t)&out.arp_ea.arp_sha, (caddr_t)ep, IPOIB_ADDRL);
}
return (result);
}
/*
* Given the boot path in the native firmware format use
* the redirection string to mutate the boot path to the new device.
* Fix up the 'v2path' so that it matches the new firmware path.
*/
void
redirect_boot_path(char *bpath, char *redirect)
{
char slicec = *redirect;
char *p = bpath + strlen(bpath);
/*
* If the redirection character doesn't fall in this
* range, something went horribly wrong.
*/
if (slicec < '0' || slicec > '7') {
printf("boot: bad redirection slice '%c'\n", slicec);
return;
}
/*
* Handle fully qualified OpenBoot pathname.
*/
while (--p >= bpath && *p != '@' && *p != '/')
if (*p == ':')
break;
if (*p++ == ':') {
/*
* Convert slice number to partition 'letter'.
*/
*p++ = 'a' + slicec - '0';
*p = '\0';
v2path = bpath;
return;
}
prom_panic("redirect_boot_path: mangled boot path!");
}
/*ARGSUSED*/
int
bootprog(char *bpath, char *bargs, boolean_t user_specified_filename)
{
boolean_t once = B_FALSE;
systype = set_fstype(v2path, bpath);
loop:
/*
* Beware: the following code may be executed twice, with different
* bpath's if we discover a redirection file.
*/
if (verbosemode) {
printf("device path '%s'\n", bpath);
if (strcmp(bpath, v2path) != 0)
printf("client path '%s'\n", v2path);
}
if (mountroot(bpath) != SUCCESS)
prom_panic("Could not mount filesystem.");
/*
* kernname (default-name) might have changed if mountroot() called
* boot_nfs_mountroot(), and it called set_default_filename().
*/
if (!user_specified_filename)
(void) strcpy(filename, kernname);
if (verbosemode)
printf("standalone = `%s', args = `%s'\n", filename, bargs);
set_client_bootargs(filename, bargs);
if (!once &&
(strcmp(systype, "ufs") == 0 || strcmp(systype, "hsfs") == 0)) {
char redirect[OBP_MAXPATHLEN];
post_mountroot(filename, redirect);
/*
* If we return at all, it's because we discovered
* a redirection file - the 'redirect' string now contains
* the name of the disk slice we should be looking at.
*
* Unmount the filesystem, tweak the boot path and retry
* the whole operation one more time.
*/
closeall(1);
once = B_TRUE;
redirect_boot_path(bpath, redirect);
if (verbosemode)
printf("%sboot: using '%s'\n", systype, bpath);
goto loop;
/*NOTREACHED*/
}
return (0);
}
int
set_bcache(fileid_t *fp)
{
/*
* Insert Disk Block Cache Entry:
*
* In this case, we actually read the requested block into a
* dynamically allocated buffer before inserting it into the
* cache. If the read fails, we return a non-zero value.
*
* The search keys for disk blocks are the block number and
* buffer size. The data associated with each entry is the
* corresponding data buffer.
*/
bc_t *bcp;
if (fp->fi_memp = bkmem_alloc(x_len = fp->fi_count)) {
/*
* We were able to succesffully allocate an input
* buffer, now read the data into it.
*/
if (diskread(fp) != 0) {
/*
* I/O error on read. Free the input buffer,
* print an error message, and bail out.
*/
bkmem_free(fp->fi_memp, x_len);
printf("disk read error\n");
return (-1);
}
x_blkno = fp->fi_blocknum;
x_dev = fp->fi_devp->di_dcookie;
bcp = (bc_t *)
set_cache(&bc_hash[BC_HASH(x_dev, x_blkno, x_len)],
&bc_head, 0);
bcp->bc_blk = x_blkno;
bcp->bc_hdr.dev = x_dev;
bcp->bc_hdr.size = x_len;
bcp->bc_hdr.data = (void *)fp->fi_memp;
} else {
/*
* We could be a bit more convervative here by
* calling "set_cache" before we try to allocate a
* buffer (thereby giving us a chance to re-use a
* previously allocated buffer) but the error recovery
* is a bit trickier, and if we're that short on memory
* we'll have trouble elsewhere anyway!
*/
prom_panic("can't read - no memory");
}
return (0);
}
/*
* Compare the version of boot that boot says it is against
* the version of boot the kernel expects.
*/
int
check_boot_version(int boots_version)
{
if (boots_version == BO_VERSION)
return (0);
prom_printf("Wrong boot interface - kernel needs v%d found v%d\n",
BO_VERSION, boots_version);
prom_panic("halting");
/*NOTREACHED*/
}
/*
* Initializes our "clock" to the creation date of /timestamp, which is
* made on the fly for us by the web server. Thereafter, time() will keep
* time sort of up to date.
*/
void
init_boot_time(void)
{
struct stat sb;
if (start_time == 0) {
if (stat("/timestamp", &sb) < 0)
prom_panic("init_boot_time: cannot stat /timestamp");
start_time = sb.st_ctim.tv_sec;
secs_since_boot = prom_gettime() / 1000;
}
}
void *
bkmem_alloc(size_t s)
{
static caddr_t next;
caddr_t ret;
if (next == NULL)
next = (caddr_t)roundup((uintptr_t)&_end, MINALLOC);
ret = next;
next += roundup(s, MINALLOC);
if (next >= TOPMEM)
prom_panic("out of memory");
return (ret);
}
/*
* This routine will find the next PAGESIZE chunk in the
* low MAPPEDMEM_MINTOP. It is analogous to valloc(). It is only for boot
* scratch memory, because child scratch memory goes up in
* the the high memory. We just need to verify that the
* pages are on the list. The calling routine will actually
* remove them.
*/
static caddr_t
get_low_vpage(size_t numpages, enum RESOURCES type)
{
size_t bytes;
caddr_t v;
if (!numpages)
return (0);
/* We know the page is mapped because the 1st MAPPEDMEM_MINTOP is 1:1 */
bytes = numpages * pagesize;
if (scratchmemp + bytes <= top_bootmem) {
v = scratchmemp;
scratchmemp += bytes;
return (v);
}
/*
* If we run out of scratch memory, look in the freelist
*/
if ((v = vpage_from_freelist(bytes)) != NULL)
return (v);
/*
* Try really hard for allocations that can't fail. Look in the area
* that we've reserved for them.
*/
if (type == RES_BOOTSCRATCH_NOFAIL) {
if (scratchresvp + bytes <= top_resvmem) {
v = scratchresvp;
scratchresvp += bytes;
dprintf("using %lu bytes of reserved mem (%lu left)\n",
bytes, top_resvmem - scratchresvp);
return (v);
} else {
printf("boot: failed to allocate %lu bytes from "
"reserved scratch memory\n", bytes);
prom_panic("boot: scratch memory overflow.\n");
}
}
return (NULL);
}
/*
* Get pages from boot and hash them into the kernel's vp.
* Used after page structs have been allocated, but before segkmem is ready.
*/
void *
boot_alloc(void *inaddr, size_t size, uint_t align)
{
caddr_t addr = inaddr;
if (bootops == NULL)
prom_panic("boot_alloc: attempt to allocate memory after "
"BOP_GONE");
size = ptob(btopr(size));
#ifdef __sparc
if (bop_alloc_chunk(addr, size, align) != (caddr_t)addr)
panic("boot_alloc: bop_alloc_chunk failed");
#else
if (BOP_ALLOC(bootops, addr, size, align) != addr)
panic("boot_alloc: BOP_ALLOC failed");
#endif
boot_mapin((caddr_t)addr, size);
return (addr);
}
int
set_rdcache(int dev, char *name, int pnum, int inum)
{
/*
* Reliably set the dcache
*
* This routine is the same as set_dcache except that it
* return 1 if the entry could not be entered into
* the cache without a purge.
*/
int len = strlen(name) + 1;
dc_t *dcp =
(dc_t *)set_cache(&dc_hash[DC_HASH(dev, name, len)],
&dc_head, 1);
if (dcp == NULL)
return (1);
if ((dcp->dc_hdr.data = (void *)bkmem_alloc(len)) == NULL) {
/*
* Not enough memory to make a copy of the name!
* There's probably not enough to do much else either!
*/
prom_panic("no memory for directory cache");
/* NOTREACHED */
}
/*
* Allocate a buffer for the pathname component, and
* make this the "data" portion of the generalize
* "cache_t" struct. Also fill in the cache-specific
* fields (pnum, inum).
*/
dcp->dc_pnum = pnum;
dcp->dc_inum = inum;
dcp->dc_hdr.dev = dev;
dcp->dc_hdr.size = len;
bcopy(name, (char *)dcp->dc_hdr.data, len);
return (0);
}
void
set_dcache(int dev, char *name, int pnum, int inum)
{
/*
* Build Directory Cache Entry:
*
* This routine creates directory cache entries to be retrieved later
* via "get_dcache". The cache key is composed of three parts: The
* device specifier, the file name ("name"), and the file number of
* the directory containing that name ("pnum"). The data portion of
* the entry consists of the file number ("inum").
*/
int len = strlen(name)+1;
dc_t *dcp =
(dc_t *)set_cache(&dc_hash[DC_HASH(dev, name, len)], &dc_head, 0);
if (dcp->dc_hdr.data = (void *)bkmem_alloc(len)) {
/*
* Allocate a buffer for the pathname component, and
* make this the "data" portion of the generalize
* "cache_t" struct. Also fill in the cache-specific
* fields (pnum, inum).
*/
dcp->dc_pnum = pnum;
dcp->dc_inum = inum;
dcp->dc_hdr.dev = dev;
dcp->dc_hdr.size = len;
bcopy(name, (char *)dcp->dc_hdr.data, len);
} else {
/*
* Not enough memory to make a copy of the name!
* There's probably not enough to do much else either!
*/
prom_panic("no memory for directory cache");
}
}
/*
* This routine remaps the kernel using large ttes
* All entries except locked ones will be removed from the tlb.
* It assumes that both the text and data segments reside in a separate
* 4mb virtual and physical contigous memory chunk. This routine
* is only executed by the first cpu. The remaining cpus execute
* sfmmu_mp_startup() instead.
* XXX It assumes that the start of the text segment is KERNELBASE. It should
* actually be based on start.
*/
void
sfmmu_remap_kernel(void)
{
pfn_t pfn;
uint_t attr;
int flags;
extern char end[];
extern struct as kas;
textva = (caddr_t)(KERNELBASE & MMU_PAGEMASK4M);
pfn = va_to_pfn(textva);
if (pfn == PFN_INVALID)
prom_panic("can't find kernel text pfn");
pfn &= TTE_PFNMASK(TTE4M);
attr = PROC_TEXT | HAT_NOSYNC;
flags = HAT_LOAD_LOCK | SFMMU_NO_TSBLOAD;
sfmmu_memtte(&ktext_tte, pfn, attr, TTE4M);
/*
* We set the lock bit in the tte to lock the translation in
* the tlb.
*/
TTE_SET_LOCKED(&ktext_tte);
sfmmu_tteload(kas.a_hat, &ktext_tte, textva, NULL, flags);
datava = (caddr_t)((uintptr_t)end & MMU_PAGEMASK4M);
pfn = va_to_pfn(datava);
if (pfn == PFN_INVALID)
prom_panic("can't find kernel data pfn");
pfn &= TTE_PFNMASK(TTE4M);
attr = PROC_DATA | HAT_NOSYNC;
sfmmu_memtte(&kdata_tte, pfn, attr, TTE4M);
/*
* We set the lock bit in the tte to lock the translation in
* the tlb. We also set the mod bit to avoid taking dirty bit
* traps on kernel data.
*/
TTE_SET_LOCKED(&kdata_tte);
TTE_SET_LOFLAGS(&kdata_tte, 0, TTE_HWWR_INT);
sfmmu_tteload(kas.a_hat, &kdata_tte, datava,
(struct page *)NULL, flags);
/*
* create bigktsb ttes if necessary.
*/
if (enable_bigktsb) {
int i = 0;
caddr_t va = ktsb_base;
size_t tsbsz = ktsb_sz;
tte_t tte;
ASSERT(va >= datava + MMU_PAGESIZE4M);
ASSERT(tsbsz >= MMU_PAGESIZE4M);
ASSERT(IS_P2ALIGNED(tsbsz, tsbsz));
ASSERT(IS_P2ALIGNED(va, tsbsz));
attr = PROC_DATA | HAT_NOSYNC;
while (tsbsz != 0) {
ASSERT(i < MAX_BIGKTSB_TTES);
pfn = va_to_pfn(va);
ASSERT(pfn != PFN_INVALID);
ASSERT((pfn & ~TTE_PFNMASK(TTE4M)) == 0);
sfmmu_memtte(&tte, pfn, attr, TTE4M);
ASSERT(TTE_IS_MOD(&tte));
/*
* No need to lock if we use physical addresses.
* Since we invalidate the kernel TSB using virtual
* addresses, it's an optimization to load them now
* so that we won't have to load them later.
*/
if (!ktsb_phys) {
TTE_SET_LOCKED(&tte);
}
sfmmu_tteload(kas.a_hat, &tte, va, NULL, flags);
bigktsb_ttes[i] = tte;
va += MMU_PAGESIZE4M;
tsbsz -= MMU_PAGESIZE4M;
i++;
}
bigktsb_nttes = i;
}
sfmmu_set_tlb();
}
/*
* Handle a IP datagram addressed to our MAC address or to the link
* layer broadcast address. Also respond to ARP requests. Generates
* inetgrams as long as there's data and the mac level IP timeout timer
* hasn't expired. As soon as there is no data, we try for
* IBD_INPUT_ATTEMPTS for more, then exit the loop, even if there is time
* left, since we expect to have data waiting for us when we're called, we just
* don't know how much.
*
* We workaround slow proms (some proms have hard sleeps for as much as 3msec)
* even though there are is data waiting.
*
* Returns the total number of MEDIA_LVL frames placed on the socket.
* Caller is expected to free up the inetgram resources.
*/
static int
ibd_input(int index)
{
struct inetgram *inp;
ipoib_ptxhdr_t *eh;
int frames = 0; /* successful frames */
int attempts = 0; /* failed attempts after success */
int16_t len = 0, data_len;
uint32_t timeout, reltime;
uint32_t pre_pr, post_pr; /* prom_read interval */
#ifdef DEBUG
int failures = 0; /* total failures */
int total_attempts = 0; /* total prom_read */
int no_data = 0; /* no data in prom */
int arps = 0; /* arp requests processed */
uint32_t tot_pr = 0; /* prom_read time */
uint32_t tot_pc = 0; /* inetgram creation time */
uint32_t pre_pc;
uint32_t now;
#endif /* DEBUG */
if (!initialized)
prom_panic("IPoIB device is not initialized.");
if ((reltime = sockets[index].in_timeout) == 0)
reltime = mac_state.mac_in_timeout;
timeout = prom_gettime() + reltime;
do {
if (frames > IBD_MAX_FRAMES) {
/* someone is trying a denial of service attack */
break;
}
/*
* The following is being paranoid about possible bugs
* where prom_read() returns a nonzero length, even when
* it's not read a packet; it zeroes out the header to
* compensate. Paranoia from calvin prom (V2) days.
*/
bzero(mac_state.mac_buf, sizeof (ipoib_ptxhdr_t));
/*
* Prom_read() will return 0 or -2 if no data is present. A
* return value of -1 means an error has occurred. We adjust
* the timeout by calling the time spent in prom_read() "free".
* prom_read() returns the number of bytes actually read, but
* will only copy "len" bytes into our buffer. Adjust in
* case the MTU is wrong.
*/
pre_pr = prom_gettime();
len = prom_read(mac_state.mac_dev, mac_state.mac_buf,
mac_state.mac_mtu, 0, NETWORK);
post_pr = prom_gettime();
timeout += (post_pr - pre_pr);
#ifdef DEBUG
tot_pr += (post_pr - pre_pr);
total_attempts++;
#endif /* DEBUG */
if (len > mac_state.mac_mtu) {
dprintf("ibd_input: adjusting MTU %d -> %d\n",
mac_state.mac_mtu, len);
bkmem_free(mac_state.mac_buf, mac_state.mac_mtu);
mac_state.mac_mtu = len;
mac_state.mac_buf = bkmem_alloc(mac_state.mac_mtu);
if (mac_state.mac_buf == NULL) {
prom_panic("ibd_input: Cannot reallocate "
"netbuf memory.");
}
len = 0; /* pretend there was no data */
}
if (len == -1) {
#ifdef DEBUG
failures++;
#endif /* DEBUG */
break;
}
if (len == 0 || len == -2) {
if (frames != 0)
attempts++;
#ifdef DEBUG
no_data++;
#endif /* DEBUG */
continue;
}
eh = (ipoib_ptxhdr_t *)mac_state.mac_buf;
if (eh->ipoib_rhdr.ipoib_type == ntohs(ETHERTYPE_IP) &&
len >= (sizeof (ipoib_ptxhdr_t) + sizeof (struct ip))) {
int offset;
#ifdef DEBUG
pre_pc = prom_gettime();
#endif /* DEBUG */
inp = (struct inetgram *)bkmem_zalloc(
sizeof (struct inetgram));
if (inp == NULL) {
errno = ENOMEM;
return (frames == 0 ? -1 : frames);
}
offset = sizeof (ipoib_ptxhdr_t);
data_len = len - offset;
inp->igm_mp = allocb(data_len, 0);
if (inp->igm_mp == NULL) {
errno = ENOMEM;
bkmem_free((caddr_t)inp,
sizeof (struct inetgram));
return (frames == 0 ? -1 : frames);
}
bcopy((caddr_t)(mac_state.mac_buf + offset),
inp->igm_mp->b_rptr, data_len);
inp->igm_mp->b_wptr += data_len;
inp->igm_level = NETWORK_LVL;
add_grams(&sockets[index].inq, inp);
frames++;
attempts = 0;
#ifdef DEBUG
tot_pc += prom_gettime() - pre_pc;
#endif /* DEBUG */
continue;
}
if (eh->ipoib_rhdr.ipoib_type == ntohs(ETHERTYPE_ARP) &&
len >= sizeof (struct arp_packet)) {
struct in_addr ip;
struct ibd_arp *ea;
#ifdef DEBUG
printf("ibd_input: ARP message received\n");
arps++;
#endif /* DEBUG */
ea = (struct ibd_arp *)(mac_state.mac_buf +
sizeof (ipoib_ptxhdr_t));
if (ea->arp_pro != ntohs(ETHERTYPE_IP))
continue;
ipv4_getipaddr(&ip);
ip.s_addr = ntohl(ip.s_addr);
if (ea->arp_op == ntohs(ARPOP_REQUEST) &&
ip.s_addr != INADDR_ANY &&
(bcmp((caddr_t)ea->arp_tpa, (caddr_t)&ip,
sizeof (struct in_addr)) == 0)) {
ea->arp_op = htons(ARPOP_REPLY);
bcopy((caddr_t)&ea->arp_sha,
(caddr_t)&eh->ipoib_dest, IPOIB_ADDRL);
bcopy((caddr_t)&ea->arp_sha,
(caddr_t)&ea->arp_tha, IPOIB_ADDRL);
bcopy((caddr_t)ea->arp_spa,
(caddr_t)ea->arp_tpa,
sizeof (struct in_addr));
bcopy(mac_state.mac_addr_buf,
(caddr_t)&ea->arp_sha,
mac_state.mac_addr_len);
bcopy((caddr_t)&ip, (caddr_t)ea->arp_spa,
sizeof (struct in_addr));
(void) prom_write(mac_state.mac_dev,
mac_state.mac_buf,
sizeof (struct arp_packet), 0, NETWORK);
/* don't charge for ARP replies */
timeout += reltime;
}
}
} while (attempts < IBD_INPUT_ATTEMPTS &&
#ifdef DEBUG
(now = prom_gettime()) < timeout);
#else
prom_gettime() < timeout);
#endif /* DEBUG */
#ifdef DEBUG
printf("ibd_input(%d): T/S/N/A/F/P/M: %d/%d/%d/%d/%d/%d/%d "
"T/O: %d < %d = %s\n", index, total_attempts, frames, no_data,
arps, failures, tot_pr, tot_pc, now, timeout,
(now < timeout) ? "TRUE" : "FALSE");
#endif /* DEBUG */
return (frames);
}
/*
* Reverse ARP client side
* Determine our Internet address given our MAC address
* See RFC 903
*/
static void
ibd_revarp(void)
{
prom_panic("IPoIB can not boot with RARP.");
}
static int
amd64_config_cpu(void)
{
struct amd64_cpuid_regs __vcr, *vcr = &__vcr;
uint32_t maxeax;
uint32_t max_maxeax = 0x100;
char vendor[13];
int isamd64 = 0;
uint32_t stdfeatures = 0, xtdfeatures = 0;
uint64_t efer;
/*
* This check may seem silly, but if the C preprocesor symbol __amd64
* is #defined during compilation, something that may outwardly seem
* like a good idea, uts/common/sys/isa_defs.h will #define _LP64,
* which will cause uts/common/sys/int_types.h to typedef uint64_t as
* an unsigned long - which is only 4 bytes in size when using a 32-bit
* compiler.
*
* If that happens, all the page table translation routines will fail
* horribly, so check the size of uint64_t just to insure some degree
* of sanity in future operations.
*/
/*LINTED [sizeof result is invarient]*/
if (sizeof (uint64_t) != 8)
prom_panic("grub compiled improperly, unable to boot "
"64-bit AMD64 executables");
/*
* If the CPU doesn't support the CPUID instruction, it's definitely
* not an AMD64.
*/
if (amd64_cpuid_supported() == 0)
return (0);
amd64_cpuid_insn(0, vcr);
maxeax = vcr->r_eax;
{
/*LINTED [vendor string from cpuid data]*/
uint32_t *iptr = (uint32_t *)vendor;
*iptr++ = vcr->r_ebx;
*iptr++ = vcr->r_edx;
*iptr++ = vcr->r_ecx;
vendor[12] = '\0';
}
if (maxeax > max_maxeax) {
grub_printf("cpu: warning, maxeax was 0x%x -> 0x%x\n",
maxeax, max_maxeax);
maxeax = max_maxeax;
}
if (maxeax < 1)
return (0); /* no additional functions, not an AMD64 */
else {
uint_t family, model, step;
amd64_cpuid_insn(1, vcr);
/*
* All AMD64/IA32e processors technically SHOULD report
* themselves as being in family 0xf, but for some reason
* Simics doesn't, and this may change in the future, so
* don't error out if it's not true.
*/
if ((family = BITX(vcr->r_eax, 11, 8)) == 0xf)
family += BITX(vcr->r_eax, 27, 20);
if ((model = BITX(vcr->r_eax, 7, 4)) == 0xf)
model += BITX(vcr->r_eax, 19, 16) << 4;
step = BITX(vcr->r_eax, 3, 0);
grub_printf("cpu: '%s' family %d model %d step %d\n",
vendor, family, model, step);
stdfeatures = vcr->r_edx;
}
amd64_cpuid_insn(0x80000000, vcr);
if (vcr->r_eax & 0x80000000) {
uint32_t xmaxeax = vcr->r_eax;
const uint32_t max_xmaxeax = 0x80000100;
if (xmaxeax > max_xmaxeax) {
grub_printf("amd64: warning, xmaxeax was "
"0x%x -> 0x%x\n", xmaxeax, max_xmaxeax);
xmaxeax = max_xmaxeax;
}
if (xmaxeax >= 0x80000001) {
amd64_cpuid_insn(0x80000001, vcr);
xtdfeatures = vcr->r_edx;
}
}
if (BITX(xtdfeatures, 29, 29)) /* long mode */
isamd64++;
else
grub_printf("amd64: CPU does NOT support long mode\n");
if (!BITX(stdfeatures, 0, 0)) {
grub_printf("amd64: CPU does NOT support FPU\n");
isamd64--;
}
if (!BITX(stdfeatures, 4, 4)) {
grub_printf("amd64: CPU does NOT support TSC\n");
isamd64--;
}
if (!BITX(stdfeatures, 5, 5)) {
grub_printf("amd64: CPU does NOT support MSRs\n");
isamd64--;
}
if (!BITX(stdfeatures, 6, 6)) {
grub_printf("amd64: CPU does NOT support PAE\n");
isamd64--;
}
if (!BITX(stdfeatures, 8, 8)) {
grub_printf("amd64: CPU does NOT support CX8\n");
isamd64--;
}
if (!BITX(stdfeatures, 13, 13)) {
grub_printf("amd64: CPU does NOT support PGE\n");
isamd64--;
}
if (!BITX(stdfeatures, 19, 19)) {
grub_printf("amd64: CPU does NOT support CLFSH\n");
isamd64--;
}
if (!BITX(stdfeatures, 23, 23)) {
grub_printf("amd64: CPU does NOT support MMX\n");
isamd64--;
}
if (!BITX(stdfeatures, 24, 24)) {
grub_printf("amd64: CPU does NOT support FXSR\n");
isamd64--;
}
if (!BITX(stdfeatures, 25, 25)) {
grub_printf("amd64: CPU does NOT support SSE\n");
isamd64--;
}
if (!BITX(stdfeatures, 26, 26)) {
grub_printf("amd64: CPU does NOT support SSE2\n");
isamd64--;
}
if (isamd64 < 1) {
grub_printf("amd64: CPU does not support amd64 executables.\n");
return (0);
}
amd64_rdmsr(MSR_AMD_EFER, &efer);
if (efer & AMD_EFER_SCE)
grub_printf("amd64: EFER_SCE (syscall/sysret) already "
"enabled\n");
if (efer & AMD_EFER_NXE)
grub_printf("amd64: EFER_NXE (no-exec prot) already enabled\n");
if (efer & AMD_EFER_LME)
grub_printf("amd64: EFER_LME (long mode) already enabled\n");
return (detect_target_operating_mode());
}
/*
* Standalone's approximation of abort().
*/
void
abort(void)
{
prom_panic("fatal error; aborting");
}
/* ARGSUSED */
static int
ibd_output(int index, struct inetgram *ogp)
{
int header_len, result;
ipoib_ptxhdr_t eh;
struct ip *ip;
struct in_addr tmpip, ipdst;
int broadcast = FALSE;
int size;
mblk_t *mp;
if (!initialized)
prom_panic("IPoIB device is not initialized.");
if (ogp->igm_level != MEDIA_LVL) {
dprintf("ibd_output: frame type wrong: socket: %d\n",
index * SOCKETTYPE);
errno = EINVAL;
return (-1);
}
header_len = IPOIB_HDRSIZE + IPOIB_ADDRL;
mp = ogp->igm_mp;
size = mp->b_wptr - mp->b_rptr;
if (size > (mac_state.mac_mtu - IPOIB_ADDRL)) {
dprintf("ibd_output: frame size too big: %d\n", size);
errno = E2BIG;
return (-1);
}
size += header_len;
ip = (struct ip *)(mp->b_rptr);
eh.ipoib_rhdr.ipoib_type = htons(ETHERTYPE_IP);
eh.ipoib_rhdr.ipoib_mbz = 0;
bcopy((caddr_t)&ip->ip_dst, (caddr_t)&ipdst, sizeof (ipdst));
if (ipdst.s_addr == htonl(INADDR_BROADCAST))
broadcast = TRUE; /* limited broadcast */
if (!broadcast) {
struct in_addr mask;
ipv4_getnetmask(&mask);
mask.s_addr = htonl(mask.s_addr);
if (mask.s_addr != htonl(INADDR_BROADCAST) &&
(ipdst.s_addr & ~mask.s_addr) == 0) {
broadcast = TRUE; /* directed broadcast */
} else {
if (ogp->igm_router.s_addr != htonl(INADDR_ANY))
tmpip.s_addr = ogp->igm_router.s_addr;
else
tmpip.s_addr = ipdst.s_addr;
result = mac_get_arp(&tmpip, (void *)&eh.ipoib_dest,
IPOIB_ADDRL, mac_state.mac_arp_timeout);
if (!result) {
errno = ETIMEDOUT;
dprintf("ibd_output: ARP request for %s "
"timed out.\n", inet_ntoa(tmpip));
return (-1);
}
}
}
if (broadcast)
bcopy((caddr_t)&ibdbroadcastaddr, (caddr_t)&eh.ipoib_dest,
IPOIB_ADDRL);
/* add the ibd header */
mp->b_rptr -= sizeof (eh);
bcopy((caddr_t)&eh, mp->b_rptr, sizeof (eh));
#ifdef DEBUG
printf("ibd_output(%d): level(%d) frame(0x%x) len(%d)\n",
index, ogp->igm_level, mp->b_rptr, size);
#endif /* DEBUG */
return (prom_write(mac_state.mac_dev, (char *)mp->b_rptr, size,
0, NETWORK));
}