2 * Copyright (c) 2003, 2004, 2005, 2006 PathScale, Inc. All rights reserved.
4 * This software is available to you under a choice of one of two
5 * licenses. You may choose to be licensed under the terms of the GNU
6 * General Public License (GPL) Version 2, available from the file
7 * COPYING in the main directory of this source tree, or the
8 * OpenIB.org BSD license below:
10 * Redistribution and use in source and binary forms, with or
11 * without modification, are permitted provided that the following
14 * - Redistributions of source code must retain the above
15 * copyright notice, this list of conditions and the following
18 * - Redistributions in binary form must reproduce the above
19 * copyright notice, this list of conditions and the following
20 * disclaimer in the documentation and/or other materials
21 * provided with the distribution.
23 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
24 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
25 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
26 * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
27 * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
28 * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
29 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
34 * These are the routines used by layered drivers, currently just the
35 * layered ethernet driver and verbs layer.
39 #include <linux/pci.h>
40 #include <asm/byteorder.h>
42 #include "ipath_kernel.h"
43 #include "ips_common.h"
44 #include "ipath_layer.h"
46 /* Acquire before ipath_devs_lock. */
47 static DEFINE_MUTEX(ipath_layer_mutex);
49 u16 ipath_layer_rcv_opcode;
50 static int (*layer_intr)(void *, u32);
51 static int (*layer_rcv)(void *, void *, struct sk_buff *);
52 static int (*layer_rcv_lid)(void *, void *);
53 static int (*verbs_piobufavail)(void *);
54 static void (*verbs_rcv)(void *, void *, void *, u32);
55 static int ipath_verbs_registered;
57 static void *(*layer_add_one)(int, struct ipath_devdata *);
58 static void (*layer_remove_one)(void *);
59 static void *(*verbs_add_one)(int, struct ipath_devdata *);
60 static void (*verbs_remove_one)(void *);
61 static void (*verbs_timer_cb)(void *);
63 int __ipath_layer_intr(struct ipath_devdata *dd, u32 arg)
67 if (dd->ipath_layer.l_arg && layer_intr)
68 ret = layer_intr(dd->ipath_layer.l_arg, arg);
73 int ipath_layer_intr(struct ipath_devdata *dd, u32 arg)
77 mutex_lock(&ipath_layer_mutex);
79 ret = __ipath_layer_intr(dd, arg);
81 mutex_unlock(&ipath_layer_mutex);
86 int __ipath_layer_rcv(struct ipath_devdata *dd, void *hdr,
91 if (dd->ipath_layer.l_arg && layer_rcv)
92 ret = layer_rcv(dd->ipath_layer.l_arg, hdr, skb);
97 int __ipath_layer_rcv_lid(struct ipath_devdata *dd, void *hdr)
101 if (dd->ipath_layer.l_arg && layer_rcv_lid)
102 ret = layer_rcv_lid(dd->ipath_layer.l_arg, hdr);
107 int __ipath_verbs_piobufavail(struct ipath_devdata *dd)
111 if (dd->verbs_layer.l_arg && verbs_piobufavail)
112 ret = verbs_piobufavail(dd->verbs_layer.l_arg);
117 int __ipath_verbs_rcv(struct ipath_devdata *dd, void *rc, void *ebuf,
122 if (dd->verbs_layer.l_arg && verbs_rcv) {
123 verbs_rcv(dd->verbs_layer.l_arg, rc, ebuf, tlen);
130 int ipath_layer_set_linkstate(struct ipath_devdata *dd, u8 newstate)
136 case IPATH_IB_LINKDOWN:
137 ipath_set_ib_lstate(dd, INFINIPATH_IBCC_LINKINITCMD_POLL <<
138 INFINIPATH_IBCC_LINKINITCMD_SHIFT);
143 case IPATH_IB_LINKDOWN_SLEEP:
144 ipath_set_ib_lstate(dd, INFINIPATH_IBCC_LINKINITCMD_SLEEP <<
145 INFINIPATH_IBCC_LINKINITCMD_SHIFT);
150 case IPATH_IB_LINKDOWN_DISABLE:
151 ipath_set_ib_lstate(dd,
152 INFINIPATH_IBCC_LINKINITCMD_DISABLE <<
153 INFINIPATH_IBCC_LINKINITCMD_SHIFT);
158 case IPATH_IB_LINKINIT:
159 if (dd->ipath_flags & IPATH_LINKINIT) {
163 ipath_set_ib_lstate(dd, INFINIPATH_IBCC_LINKCMD_INIT <<
164 INFINIPATH_IBCC_LINKCMD_SHIFT);
165 lstate = IPATH_LINKINIT;
168 case IPATH_IB_LINKARM:
169 if (dd->ipath_flags & IPATH_LINKARMED) {
173 if (!(dd->ipath_flags &
174 (IPATH_LINKINIT | IPATH_LINKACTIVE))) {
178 ipath_set_ib_lstate(dd, INFINIPATH_IBCC_LINKCMD_ARMED <<
179 INFINIPATH_IBCC_LINKCMD_SHIFT);
181 * Since the port can transition to ACTIVE by receiving
182 * a non VL 15 packet, wait for either state.
184 lstate = IPATH_LINKARMED | IPATH_LINKACTIVE;
187 case IPATH_IB_LINKACTIVE:
188 if (dd->ipath_flags & IPATH_LINKACTIVE) {
192 if (!(dd->ipath_flags & IPATH_LINKARMED)) {
196 ipath_set_ib_lstate(dd, INFINIPATH_IBCC_LINKCMD_ACTIVE <<
197 INFINIPATH_IBCC_LINKCMD_SHIFT);
198 lstate = IPATH_LINKACTIVE;
202 ipath_dbg("Invalid linkstate 0x%x requested\n", newstate);
206 ret = ipath_wait_linkstate(dd, lstate, 2000);
212 EXPORT_SYMBOL_GPL(ipath_layer_set_linkstate);
215 * ipath_layer_set_mtu - set the MTU
216 * @dd: the infinipath device
219 * we can handle "any" incoming size, the issue here is whether we
220 * need to restrict our outgoing size. For now, we don't do any
221 * sanity checking on this, and we don't deal with what happens to
222 * programs that are already running when the size changes.
223 * NOTE: changing the MTU will usually cause the IBC to go back to
224 * link initialize (IPATH_IBSTATE_INIT) state...
226 int ipath_layer_set_mtu(struct ipath_devdata *dd, u16 arg)
233 * mtu is IB data payload max. It's the largest power of 2 less
234 * than piosize (or even larger, since it only really controls the
235 * largest we can receive; we can send the max of the mtu and
236 * piosize). We check that it's one of the valid IB sizes.
238 if (arg != 256 && arg != 512 && arg != 1024 && arg != 2048 &&
240 ipath_dbg("Trying to set invalid mtu %u, failing\n", arg);
244 if (dd->ipath_ibmtu == arg) {
245 ret = 0; /* same as current */
249 piosize = dd->ipath_ibmaxlen;
250 dd->ipath_ibmtu = arg;
252 if (arg >= (piosize - IPATH_PIO_MAXIBHDR)) {
253 /* Only if it's not the initial value (or reset to it) */
254 if (piosize != dd->ipath_init_ibmaxlen) {
255 dd->ipath_ibmaxlen = piosize;
258 } else if ((arg + IPATH_PIO_MAXIBHDR) != dd->ipath_ibmaxlen) {
259 piosize = arg + IPATH_PIO_MAXIBHDR;
260 ipath_cdbg(VERBOSE, "ibmaxlen was 0x%x, setting to 0x%x "
261 "(mtu 0x%x)\n", dd->ipath_ibmaxlen, piosize,
263 dd->ipath_ibmaxlen = piosize;
269 * set the IBC maxpktlength to the size of our pio
272 u64 ibc = dd->ipath_ibcctrl;
273 ibc &= ~(INFINIPATH_IBCC_MAXPKTLEN_MASK <<
274 INFINIPATH_IBCC_MAXPKTLEN_SHIFT);
276 piosize = piosize - 2 * sizeof(u32); /* ignore pbc */
277 dd->ipath_ibmaxlen = piosize;
278 piosize /= sizeof(u32); /* in words */
280 * for ICRC, which we only send in diag test pkt mode, and
281 * we don't need to worry about that for mtu
285 ibc |= piosize << INFINIPATH_IBCC_MAXPKTLEN_SHIFT;
286 dd->ipath_ibcctrl = ibc;
287 ipath_write_kreg(dd, dd->ipath_kregs->kr_ibcctrl,
289 dd->ipath_f_tidtemplate(dd);
298 EXPORT_SYMBOL_GPL(ipath_layer_set_mtu);
300 int ipath_set_sps_lid(struct ipath_devdata *dd, u32 arg, u8 lmc)
302 ipath_stats.sps_lid[dd->ipath_unit] = arg;
306 mutex_lock(&ipath_layer_mutex);
308 if (dd->ipath_layer.l_arg && layer_intr)
309 layer_intr(dd->ipath_layer.l_arg, IPATH_LAYER_INT_LID);
311 mutex_unlock(&ipath_layer_mutex);
316 EXPORT_SYMBOL_GPL(ipath_set_sps_lid);
318 int ipath_layer_set_guid(struct ipath_devdata *dd, __be64 guid)
320 /* XXX - need to inform anyone who cares this just happened. */
321 dd->ipath_guid = guid;
325 EXPORT_SYMBOL_GPL(ipath_layer_set_guid);
327 __be64 ipath_layer_get_guid(struct ipath_devdata *dd)
329 return dd->ipath_guid;
332 EXPORT_SYMBOL_GPL(ipath_layer_get_guid);
334 u32 ipath_layer_get_nguid(struct ipath_devdata *dd)
336 return dd->ipath_nguid;
339 EXPORT_SYMBOL_GPL(ipath_layer_get_nguid);
341 int ipath_layer_query_device(struct ipath_devdata *dd, u32 * vendor,
342 u32 * boardrev, u32 * majrev, u32 * minrev)
344 *vendor = dd->ipath_vendorid;
345 *boardrev = dd->ipath_boardrev;
346 *majrev = dd->ipath_majrev;
347 *minrev = dd->ipath_minrev;
352 EXPORT_SYMBOL_GPL(ipath_layer_query_device);
354 u32 ipath_layer_get_flags(struct ipath_devdata *dd)
356 return dd->ipath_flags;
359 EXPORT_SYMBOL_GPL(ipath_layer_get_flags);
361 struct device *ipath_layer_get_device(struct ipath_devdata *dd)
363 return &dd->pcidev->dev;
366 EXPORT_SYMBOL_GPL(ipath_layer_get_device);
368 u16 ipath_layer_get_deviceid(struct ipath_devdata *dd)
370 return dd->ipath_deviceid;
373 EXPORT_SYMBOL_GPL(ipath_layer_get_deviceid);
375 u64 ipath_layer_get_lastibcstat(struct ipath_devdata *dd)
377 return dd->ipath_lastibcstat;
380 EXPORT_SYMBOL_GPL(ipath_layer_get_lastibcstat);
382 u32 ipath_layer_get_ibmtu(struct ipath_devdata *dd)
384 return dd->ipath_ibmtu;
387 EXPORT_SYMBOL_GPL(ipath_layer_get_ibmtu);
389 void ipath_layer_add(struct ipath_devdata *dd)
391 mutex_lock(&ipath_layer_mutex);
394 dd->ipath_layer.l_arg =
395 layer_add_one(dd->ipath_unit, dd);
398 dd->verbs_layer.l_arg =
399 verbs_add_one(dd->ipath_unit, dd);
401 mutex_unlock(&ipath_layer_mutex);
404 void ipath_layer_del(struct ipath_devdata *dd)
406 mutex_lock(&ipath_layer_mutex);
408 if (dd->ipath_layer.l_arg && layer_remove_one) {
409 layer_remove_one(dd->ipath_layer.l_arg);
410 dd->ipath_layer.l_arg = NULL;
413 if (dd->verbs_layer.l_arg && verbs_remove_one) {
414 verbs_remove_one(dd->verbs_layer.l_arg);
415 dd->verbs_layer.l_arg = NULL;
418 mutex_unlock(&ipath_layer_mutex);
421 int ipath_layer_register(void *(*l_add)(int, struct ipath_devdata *),
422 void (*l_remove)(void *),
423 int (*l_intr)(void *, u32),
424 int (*l_rcv)(void *, void *, struct sk_buff *),
426 int (*l_rcv_lid)(void *, void *))
428 struct ipath_devdata *dd, *tmp;
431 mutex_lock(&ipath_layer_mutex);
433 layer_add_one = l_add;
434 layer_remove_one = l_remove;
437 layer_rcv_lid = l_rcv_lid;
438 ipath_layer_rcv_opcode = l_rcv_opcode;
440 spin_lock_irqsave(&ipath_devs_lock, flags);
442 list_for_each_entry_safe(dd, tmp, &ipath_dev_list, ipath_list) {
443 if (!(dd->ipath_flags & IPATH_INITTED))
446 if (dd->ipath_layer.l_arg)
449 if (!(*dd->ipath_statusp & IPATH_STATUS_SMA))
450 *dd->ipath_statusp |= IPATH_STATUS_OIB_SMA;
452 spin_unlock_irqrestore(&ipath_devs_lock, flags);
453 dd->ipath_layer.l_arg = l_add(dd->ipath_unit, dd);
454 spin_lock_irqsave(&ipath_devs_lock, flags);
457 spin_unlock_irqrestore(&ipath_devs_lock, flags);
458 mutex_unlock(&ipath_layer_mutex);
463 EXPORT_SYMBOL_GPL(ipath_layer_register);
465 void ipath_layer_unregister(void)
467 struct ipath_devdata *dd, *tmp;
470 mutex_lock(&ipath_layer_mutex);
471 spin_lock_irqsave(&ipath_devs_lock, flags);
473 list_for_each_entry_safe(dd, tmp, &ipath_dev_list, ipath_list) {
474 if (dd->ipath_layer.l_arg && layer_remove_one) {
475 spin_unlock_irqrestore(&ipath_devs_lock, flags);
476 layer_remove_one(dd->ipath_layer.l_arg);
477 spin_lock_irqsave(&ipath_devs_lock, flags);
478 dd->ipath_layer.l_arg = NULL;
482 spin_unlock_irqrestore(&ipath_devs_lock, flags);
484 layer_add_one = NULL;
485 layer_remove_one = NULL;
488 layer_rcv_lid = NULL;
490 mutex_unlock(&ipath_layer_mutex);
493 EXPORT_SYMBOL_GPL(ipath_layer_unregister);
495 static void __ipath_verbs_timer(unsigned long arg)
497 struct ipath_devdata *dd = (struct ipath_devdata *) arg;
500 * If port 0 receive packet interrupts are not available, or
501 * can be missed, poll the receive queue
503 if (dd->ipath_flags & IPATH_POLL_RX_INTR)
506 /* Handle verbs layer timeouts. */
507 if (dd->verbs_layer.l_arg && verbs_timer_cb)
508 verbs_timer_cb(dd->verbs_layer.l_arg);
510 mod_timer(&dd->verbs_layer.l_timer, jiffies + 1);
514 * ipath_verbs_register - verbs layer registration
515 * @l_piobufavail: callback for when PIO buffers become available
516 * @l_rcv: callback for receiving a packet
517 * @l_timer_cb: timer callback
518 * @ipath_devdata: device data structure is put here
520 int ipath_verbs_register(void *(*l_add)(int, struct ipath_devdata *),
521 void (*l_remove)(void *arg),
522 int (*l_piobufavail) (void *arg),
523 void (*l_rcv) (void *arg, void *rhdr,
524 void *data, u32 tlen),
525 void (*l_timer_cb) (void *arg))
527 struct ipath_devdata *dd, *tmp;
530 mutex_lock(&ipath_layer_mutex);
532 verbs_add_one = l_add;
533 verbs_remove_one = l_remove;
534 verbs_piobufavail = l_piobufavail;
536 verbs_timer_cb = l_timer_cb;
538 spin_lock_irqsave(&ipath_devs_lock, flags);
540 list_for_each_entry_safe(dd, tmp, &ipath_dev_list, ipath_list) {
541 if (!(dd->ipath_flags & IPATH_INITTED))
544 if (dd->verbs_layer.l_arg)
547 spin_unlock_irqrestore(&ipath_devs_lock, flags);
548 dd->verbs_layer.l_arg = l_add(dd->ipath_unit, dd);
549 spin_lock_irqsave(&ipath_devs_lock, flags);
552 spin_unlock_irqrestore(&ipath_devs_lock, flags);
553 mutex_unlock(&ipath_layer_mutex);
555 ipath_verbs_registered = 1;
560 EXPORT_SYMBOL_GPL(ipath_verbs_register);
562 void ipath_verbs_unregister(void)
564 struct ipath_devdata *dd, *tmp;
567 mutex_lock(&ipath_layer_mutex);
568 spin_lock_irqsave(&ipath_devs_lock, flags);
570 list_for_each_entry_safe(dd, tmp, &ipath_dev_list, ipath_list) {
571 *dd->ipath_statusp &= ~IPATH_STATUS_OIB_SMA;
573 if (dd->verbs_layer.l_arg && verbs_remove_one) {
574 spin_unlock_irqrestore(&ipath_devs_lock, flags);
575 verbs_remove_one(dd->verbs_layer.l_arg);
576 spin_lock_irqsave(&ipath_devs_lock, flags);
577 dd->verbs_layer.l_arg = NULL;
581 spin_unlock_irqrestore(&ipath_devs_lock, flags);
583 verbs_add_one = NULL;
584 verbs_remove_one = NULL;
585 verbs_piobufavail = NULL;
587 verbs_timer_cb = NULL;
589 mutex_unlock(&ipath_layer_mutex);
592 EXPORT_SYMBOL_GPL(ipath_verbs_unregister);
594 int ipath_layer_open(struct ipath_devdata *dd, u32 * pktmax)
599 mutex_lock(&ipath_layer_mutex);
601 if (!dd->ipath_layer.l_arg) {
606 ret = ipath_setrcvhdrsize(dd, NUM_OF_EXTRA_WORDS_IN_HEADER_QUEUE);
611 *pktmax = dd->ipath_ibmaxlen;
613 if (*dd->ipath_statusp & IPATH_STATUS_IB_READY)
614 intval |= IPATH_LAYER_INT_IF_UP;
615 if (ipath_stats.sps_lid[dd->ipath_unit])
616 intval |= IPATH_LAYER_INT_LID;
617 if (ipath_stats.sps_mlid[dd->ipath_unit])
618 intval |= IPATH_LAYER_INT_BCAST;
620 * do this on open, in case low level is already up and
621 * just layered driver was reloaded, etc.
624 layer_intr(dd->ipath_layer.l_arg, intval);
628 mutex_unlock(&ipath_layer_mutex);
633 EXPORT_SYMBOL_GPL(ipath_layer_open);
635 u16 ipath_layer_get_lid(struct ipath_devdata *dd)
637 return dd->ipath_lid;
640 EXPORT_SYMBOL_GPL(ipath_layer_get_lid);
643 * ipath_layer_get_mac - get the MAC address
644 * @dd: the infinipath device
645 * @mac: the MAC is put here
647 * This is the EUID-64 OUI octets (top 3), then
648 * skip the next 2 (which should both be zero or 0xff).
649 * The returned MAC is in network order
650 * mac points to at least 6 bytes of buffer
651 * We assume that by the time the LID is set, that the GUID is as valid
652 * as it's ever going to be, rather than adding yet another status bit.
655 int ipath_layer_get_mac(struct ipath_devdata *dd, u8 * mac)
659 guid = (u8 *) &dd->ipath_guid;
667 if ((guid[3] || guid[4]) && !(guid[3] == 0xff && guid[4] == 0xff))
668 ipath_dbg("Warning, guid bytes 3 and 4 not 0 or 0xffff: "
669 "%x %x\n", guid[3], guid[4]);
673 EXPORT_SYMBOL_GPL(ipath_layer_get_mac);
675 u16 ipath_layer_get_bcast(struct ipath_devdata *dd)
677 return dd->ipath_mlid;
680 EXPORT_SYMBOL_GPL(ipath_layer_get_bcast);
682 u32 ipath_layer_get_cr_errpkey(struct ipath_devdata *dd)
684 return ipath_read_creg32(dd, dd->ipath_cregs->cr_errpkey);
687 EXPORT_SYMBOL_GPL(ipath_layer_get_cr_errpkey);
689 static void update_sge(struct ipath_sge_state *ss, u32 length)
691 struct ipath_sge *sge = &ss->sge;
693 sge->vaddr += length;
694 sge->length -= length;
695 sge->sge_length -= length;
696 if (sge->sge_length == 0) {
698 *sge = *ss->sg_list++;
699 } else if (sge->length == 0 && sge->mr != NULL) {
700 if (++sge->n >= IPATH_SEGSZ) {
701 if (++sge->m >= sge->mr->mapsz)
705 sge->vaddr = sge->mr->map[sge->m]->segs[sge->n].vaddr;
706 sge->length = sge->mr->map[sge->m]->segs[sge->n].length;
710 #ifdef __LITTLE_ENDIAN
711 static inline u32 get_upper_bits(u32 data, u32 shift)
713 return data >> shift;
716 static inline u32 set_upper_bits(u32 data, u32 shift)
718 return data << shift;
721 static inline u32 clear_upper_bytes(u32 data, u32 n, u32 off)
723 data <<= ((sizeof(u32) - n) * BITS_PER_BYTE);
724 data >>= ((sizeof(u32) - n - off) * BITS_PER_BYTE);
728 static inline u32 get_upper_bits(u32 data, u32 shift)
730 return data << shift;
733 static inline u32 set_upper_bits(u32 data, u32 shift)
735 return data >> shift;
738 static inline u32 clear_upper_bytes(u32 data, u32 n, u32 off)
740 data >>= ((sizeof(u32) - n) * BITS_PER_BYTE);
741 data <<= ((sizeof(u32) - n - off) * BITS_PER_BYTE);
746 static void copy_io(u32 __iomem *piobuf, struct ipath_sge_state *ss,
754 u32 len = ss->sge.length;
760 if (len > ss->sge.sge_length)
761 len = ss->sge.sge_length;
762 /* If the source address is not aligned, try to align it. */
763 off = (unsigned long)ss->sge.vaddr & (sizeof(u32) - 1);
765 u32 *addr = (u32 *)((unsigned long)ss->sge.vaddr &
767 u32 v = get_upper_bits(*addr, off * BITS_PER_BYTE);
770 y = sizeof(u32) - off;
773 if (len + extra >= sizeof(u32)) {
774 data |= set_upper_bits(v, extra *
776 len = sizeof(u32) - extra;
781 __raw_writel(data, piobuf);
786 /* Clear unused upper bytes */
787 data |= clear_upper_bytes(v, len, extra);
795 /* Source address is aligned. */
796 u32 *addr = (u32 *) ss->sge.vaddr;
797 int shift = extra * BITS_PER_BYTE;
798 int ushift = 32 - shift;
801 while (l >= sizeof(u32)) {
804 data |= set_upper_bits(v, shift);
805 __raw_writel(data, piobuf);
806 data = get_upper_bits(v, ushift);
812 * We still have 'extra' number of bytes leftover.
817 if (l + extra >= sizeof(u32)) {
818 data |= set_upper_bits(v, shift);
819 len -= l + extra - sizeof(u32);
824 __raw_writel(data, piobuf);
829 /* Clear unused upper bytes */
830 data |= clear_upper_bytes(v, l,
838 } else if (len == length) {
842 } else if (len == length) {
846 * Need to round up for the last dword in the
850 __iowrite32_copy(piobuf, ss->sge.vaddr, w - 1);
852 last = ((u32 *) ss->sge.vaddr)[w - 1];
857 __iowrite32_copy(piobuf, ss->sge.vaddr, w);
860 extra = len & (sizeof(u32) - 1);
862 u32 v = ((u32 *) ss->sge.vaddr)[w];
864 /* Clear unused upper bytes */
865 data = clear_upper_bytes(v, extra, 0);
871 /* must flush early everything before trigger word */
873 __raw_writel(last, piobuf);
874 /* be sure trigger word is written */
876 update_sge(ss, length);
880 * ipath_verbs_send - send a packet from the verbs layer
881 * @dd: the infinipath device
882 * @hdrwords: the number of works in the header
883 * @hdr: the packet header
884 * @len: the length of the packet in bytes
885 * @ss: the SGE to send
887 * This is like ipath_sma_send_pkt() in that we need to be able to send
888 * packets after the chip is initialized (MADs) but also like
889 * ipath_layer_send_hdr() since its used by the verbs layer.
891 int ipath_verbs_send(struct ipath_devdata *dd, u32 hdrwords,
892 u32 *hdr, u32 len, struct ipath_sge_state *ss)
898 /* +1 is for the qword padding of pbc */
899 plen = hdrwords + ((len + 3) >> 2) + 1;
900 if (unlikely((plen << 2) > dd->ipath_ibmaxlen)) {
901 ipath_dbg("packet len 0x%x too long, failing\n", plen);
906 /* Get a PIO buffer to use. */
907 piobuf = ipath_getpiobuf(dd, NULL);
908 if (unlikely(piobuf == NULL)) {
914 * Write len to control qword, no flags.
915 * We have to flush after the PBC for correctness on some cpus
916 * or WC buffer can be written out of order.
918 writeq(plen, piobuf);
923 * If there is just the header portion, must flush before
924 * writing last word of header for correctness, and after
925 * the last header word (trigger word).
927 __iowrite32_copy(piobuf, hdr, hdrwords - 1);
929 __raw_writel(hdr[hdrwords - 1], piobuf + hdrwords - 1);
935 __iowrite32_copy(piobuf, hdr, hdrwords);
938 /* The common case is aligned and contained in one segment. */
939 if (likely(ss->num_sge == 1 && len <= ss->sge.length &&
940 !((unsigned long)ss->sge.vaddr & (sizeof(u32) - 1)))) {
943 /* Need to round up for the last dword in the packet. */
945 __iowrite32_copy(piobuf, ss->sge.vaddr, w - 1);
946 /* must flush early everything before trigger word */
948 __raw_writel(((u32 *) ss->sge.vaddr)[w - 1],
950 /* be sure trigger word is written */
956 copy_io(piobuf, ss, len);
963 EXPORT_SYMBOL_GPL(ipath_verbs_send);
965 int ipath_layer_snapshot_counters(struct ipath_devdata *dd, u64 *swords,
966 u64 *rwords, u64 *spkts, u64 *rpkts,
971 if (!(dd->ipath_flags & IPATH_INITTED)) {
972 /* no hardware, freeze, etc. */
973 ipath_dbg("unit %u not usable\n", dd->ipath_unit);
977 *swords = ipath_snap_cntr(dd, dd->ipath_cregs->cr_wordsendcnt);
978 *rwords = ipath_snap_cntr(dd, dd->ipath_cregs->cr_wordrcvcnt);
979 *spkts = ipath_snap_cntr(dd, dd->ipath_cregs->cr_pktsendcnt);
980 *rpkts = ipath_snap_cntr(dd, dd->ipath_cregs->cr_pktrcvcnt);
981 *xmit_wait = ipath_snap_cntr(dd, dd->ipath_cregs->cr_sendstallcnt);
989 EXPORT_SYMBOL_GPL(ipath_layer_snapshot_counters);
992 * ipath_layer_get_counters - get various chip counters
993 * @dd: the infinipath device
994 * @cntrs: counters are placed here
996 * Return the counters needed by recv_pma_get_portcounters().
998 int ipath_layer_get_counters(struct ipath_devdata *dd,
999 struct ipath_layer_counters *cntrs)
1003 if (!(dd->ipath_flags & IPATH_INITTED)) {
1004 /* no hardware, freeze, etc. */
1005 ipath_dbg("unit %u not usable\n", dd->ipath_unit);
1009 cntrs->symbol_error_counter =
1010 ipath_snap_cntr(dd, dd->ipath_cregs->cr_ibsymbolerrcnt);
1011 cntrs->link_error_recovery_counter =
1012 ipath_snap_cntr(dd, dd->ipath_cregs->cr_iblinkerrrecovcnt);
1013 cntrs->link_downed_counter =
1014 ipath_snap_cntr(dd, dd->ipath_cregs->cr_iblinkdowncnt);
1015 cntrs->port_rcv_errors =
1016 ipath_snap_cntr(dd, dd->ipath_cregs->cr_rxdroppktcnt) +
1017 ipath_snap_cntr(dd, dd->ipath_cregs->cr_rcvovflcnt) +
1018 ipath_snap_cntr(dd, dd->ipath_cregs->cr_portovflcnt) +
1019 ipath_snap_cntr(dd, dd->ipath_cregs->cr_errrcvflowctrlcnt) +
1020 ipath_snap_cntr(dd, dd->ipath_cregs->cr_err_rlencnt) +
1021 ipath_snap_cntr(dd, dd->ipath_cregs->cr_invalidrlencnt) +
1022 ipath_snap_cntr(dd, dd->ipath_cregs->cr_erricrccnt) +
1023 ipath_snap_cntr(dd, dd->ipath_cregs->cr_errvcrccnt) +
1024 ipath_snap_cntr(dd, dd->ipath_cregs->cr_errlpcrccnt) +
1025 ipath_snap_cntr(dd, dd->ipath_cregs->cr_errlinkcnt) +
1026 ipath_snap_cntr(dd, dd->ipath_cregs->cr_badformatcnt);
1027 cntrs->port_rcv_remphys_errors =
1028 ipath_snap_cntr(dd, dd->ipath_cregs->cr_rcvebpcnt);
1029 cntrs->port_xmit_discards =
1030 ipath_snap_cntr(dd, dd->ipath_cregs->cr_unsupvlcnt);
1031 cntrs->port_xmit_data =
1032 ipath_snap_cntr(dd, dd->ipath_cregs->cr_wordsendcnt);
1033 cntrs->port_rcv_data =
1034 ipath_snap_cntr(dd, dd->ipath_cregs->cr_wordrcvcnt);
1035 cntrs->port_xmit_packets =
1036 ipath_snap_cntr(dd, dd->ipath_cregs->cr_pktsendcnt);
1037 cntrs->port_rcv_packets =
1038 ipath_snap_cntr(dd, dd->ipath_cregs->cr_pktrcvcnt);
1046 EXPORT_SYMBOL_GPL(ipath_layer_get_counters);
1048 int ipath_layer_want_buffer(struct ipath_devdata *dd)
1050 set_bit(IPATH_S_PIOINTBUFAVAIL, &dd->ipath_sendctrl);
1051 ipath_write_kreg(dd, dd->ipath_kregs->kr_sendctrl,
1052 dd->ipath_sendctrl);
1057 EXPORT_SYMBOL_GPL(ipath_layer_want_buffer);
1059 int ipath_layer_send_hdr(struct ipath_devdata *dd, struct ether_header *hdr)
1062 u32 __iomem *piobuf;
1067 if (!(dd->ipath_flags & IPATH_RCVHDRSZ_SET)) {
1068 ipath_dbg("send while not open\n");
1071 if ((dd->ipath_flags & (IPATH_LINKUNK | IPATH_LINKDOWN)) ||
1072 dd->ipath_lid == 0) {
1074 * lid check is for when sma hasn't yet configured
1077 ipath_cdbg(VERBOSE, "send while not ready, "
1078 "mylid=%u, flags=0x%x\n",
1079 dd->ipath_lid, dd->ipath_flags);
1082 vlsllnh = *((__be16 *) hdr);
1083 if (vlsllnh != htons(IPS_LRH_BTH)) {
1084 ipath_dbg("Warning: lrh[0] wrong (%x, not %x); "
1085 "not sending\n", be16_to_cpu(vlsllnh),
1092 /* Get a PIO buffer to use. */
1093 piobuf = ipath_getpiobuf(dd, NULL);
1094 if (piobuf == NULL) {
1099 plen = (sizeof(*hdr) >> 2); /* actual length */
1100 ipath_cdbg(EPKT, "0x%x+1w pio %p\n", plen, piobuf);
1102 writeq(plen+1, piobuf); /* len (+1 for pad) to pbc, no flags */
1106 count = plen-1; /* amount we can copy before trigger word */
1107 __iowrite32_copy(piobuf, uhdr, count);
1109 __raw_writel(uhdr[count], piobuf + count);
1110 ipath_flush_wc(); /* ensure it's sent, now */
1112 ipath_stats.sps_ether_spkts++; /* ether packet sent */
1118 EXPORT_SYMBOL_GPL(ipath_layer_send_hdr);
1120 int ipath_layer_set_piointbufavail_int(struct ipath_devdata *dd)
1122 set_bit(IPATH_S_PIOINTBUFAVAIL, &dd->ipath_sendctrl);
1124 ipath_write_kreg(dd, dd->ipath_kregs->kr_sendctrl,
1125 dd->ipath_sendctrl);
1129 EXPORT_SYMBOL_GPL(ipath_layer_set_piointbufavail_int);
1131 int ipath_layer_enable_timer(struct ipath_devdata *dd)
1134 * HT-400 has a design flaw where the chip and kernel idea
1135 * of the tail register don't always agree, and therefore we won't
1136 * get an interrupt on the next packet received.
1137 * If the board supports per packet receive interrupts, use it.
1138 * Otherwise, the timer function periodically checks for packets
1139 * to cover this case.
1140 * Either way, the timer is needed for verbs layer related
1143 if (dd->ipath_flags & IPATH_GPIO_INTR) {
1144 ipath_write_kreg(dd, dd->ipath_kregs->kr_debugportselect,
1145 0x2074076542310ULL);
1146 /* Enable GPIO bit 2 interrupt */
1147 ipath_write_kreg(dd, dd->ipath_kregs->kr_gpio_mask,
1151 init_timer(&dd->verbs_layer.l_timer);
1152 dd->verbs_layer.l_timer.function = __ipath_verbs_timer;
1153 dd->verbs_layer.l_timer.data = (unsigned long)dd;
1154 dd->verbs_layer.l_timer.expires = jiffies + 1;
1155 add_timer(&dd->verbs_layer.l_timer);
1160 EXPORT_SYMBOL_GPL(ipath_layer_enable_timer);
1162 int ipath_layer_disable_timer(struct ipath_devdata *dd)
1164 /* Disable GPIO bit 2 interrupt */
1165 if (dd->ipath_flags & IPATH_GPIO_INTR)
1166 ipath_write_kreg(dd, dd->ipath_kregs->kr_gpio_mask, 0);
1168 del_timer_sync(&dd->verbs_layer.l_timer);
1173 EXPORT_SYMBOL_GPL(ipath_layer_disable_timer);
1176 * ipath_layer_set_verbs_flags - set the verbs layer flags
1177 * @dd: the infinipath device
1178 * @flags: the flags to set
1180 int ipath_layer_set_verbs_flags(struct ipath_devdata *dd, unsigned flags)
1182 struct ipath_devdata *ss;
1183 unsigned long lflags;
1185 spin_lock_irqsave(&ipath_devs_lock, lflags);
1187 list_for_each_entry(ss, &ipath_dev_list, ipath_list) {
1188 if (!(ss->ipath_flags & IPATH_INITTED))
1190 if ((flags & IPATH_VERBS_KERNEL_SMA) &&
1191 !(*ss->ipath_statusp & IPATH_STATUS_SMA))
1192 *ss->ipath_statusp |= IPATH_STATUS_OIB_SMA;
1194 *ss->ipath_statusp &= ~IPATH_STATUS_OIB_SMA;
1197 spin_unlock_irqrestore(&ipath_devs_lock, lflags);
1202 EXPORT_SYMBOL_GPL(ipath_layer_set_verbs_flags);
1205 * ipath_layer_get_npkeys - return the size of the PKEY table for port 0
1206 * @dd: the infinipath device
1208 unsigned ipath_layer_get_npkeys(struct ipath_devdata *dd)
1210 return ARRAY_SIZE(dd->ipath_pd[0]->port_pkeys);
1213 EXPORT_SYMBOL_GPL(ipath_layer_get_npkeys);
1216 * ipath_layer_get_pkey - return the indexed PKEY from the port 0 PKEY table
1217 * @dd: the infinipath device
1218 * @index: the PKEY index
1220 unsigned ipath_layer_get_pkey(struct ipath_devdata *dd, unsigned index)
1224 if (index >= ARRAY_SIZE(dd->ipath_pd[0]->port_pkeys))
1227 ret = dd->ipath_pd[0]->port_pkeys[index];
1232 EXPORT_SYMBOL_GPL(ipath_layer_get_pkey);
1235 * ipath_layer_get_pkeys - return the PKEY table for port 0
1236 * @dd: the infinipath device
1237 * @pkeys: the pkey table is placed here
1239 int ipath_layer_get_pkeys(struct ipath_devdata *dd, u16 * pkeys)
1241 struct ipath_portdata *pd = dd->ipath_pd[0];
1243 memcpy(pkeys, pd->port_pkeys, sizeof(pd->port_pkeys));
1248 EXPORT_SYMBOL_GPL(ipath_layer_get_pkeys);
1251 * rm_pkey - decrecment the reference count for the given PKEY
1252 * @dd: the infinipath device
1253 * @key: the PKEY index
1255 * Return true if this was the last reference and the hardware table entry
1256 * needs to be changed.
1258 static int rm_pkey(struct ipath_devdata *dd, u16 key)
1263 for (i = 0; i < ARRAY_SIZE(dd->ipath_pkeys); i++) {
1264 if (dd->ipath_pkeys[i] != key)
1266 if (atomic_dec_and_test(&dd->ipath_pkeyrefs[i])) {
1267 dd->ipath_pkeys[i] = 0;
1281 * add_pkey - add the given PKEY to the hardware table
1282 * @dd: the infinipath device
1285 * Return an error code if unable to add the entry, zero if no change,
1286 * or 1 if the hardware PKEY register needs to be updated.
1288 static int add_pkey(struct ipath_devdata *dd, u16 key)
1291 u16 lkey = key & 0x7FFF;
1295 if (lkey == 0x7FFF) {
1300 /* Look for an empty slot or a matching PKEY. */
1301 for (i = 0; i < ARRAY_SIZE(dd->ipath_pkeys); i++) {
1302 if (!dd->ipath_pkeys[i]) {
1306 /* If it matches exactly, try to increment the ref count */
1307 if (dd->ipath_pkeys[i] == key) {
1308 if (atomic_inc_return(&dd->ipath_pkeyrefs[i]) > 1) {
1312 /* Lost the race. Look for an empty slot below. */
1313 atomic_dec(&dd->ipath_pkeyrefs[i]);
1317 * It makes no sense to have both the limited and unlimited
1318 * PKEY set at the same time since the unlimited one will
1319 * disable the limited one.
1321 if ((dd->ipath_pkeys[i] & 0x7FFF) == lkey) {
1330 for (i = 0; i < ARRAY_SIZE(dd->ipath_pkeys); i++) {
1331 if (!dd->ipath_pkeys[i] &&
1332 atomic_inc_return(&dd->ipath_pkeyrefs[i]) == 1) {
1333 /* for ipathstats, etc. */
1334 ipath_stats.sps_pkeys[i] = lkey;
1335 dd->ipath_pkeys[i] = key;
1347 * ipath_layer_set_pkeys - set the PKEY table for port 0
1348 * @dd: the infinipath device
1349 * @pkeys: the PKEY table
1351 int ipath_layer_set_pkeys(struct ipath_devdata *dd, u16 * pkeys)
1353 struct ipath_portdata *pd;
1357 pd = dd->ipath_pd[0];
1359 for (i = 0; i < ARRAY_SIZE(pd->port_pkeys); i++) {
1361 u16 okey = pd->port_pkeys[i];
1366 * The value of this PKEY table entry is changing.
1367 * Remove the old entry in the hardware's array of PKEYs.
1370 changed |= rm_pkey(dd, okey);
1372 int ret = add_pkey(dd, key);
1379 pd->port_pkeys[i] = key;
1384 pkey = (u64) dd->ipath_pkeys[0] |
1385 ((u64) dd->ipath_pkeys[1] << 16) |
1386 ((u64) dd->ipath_pkeys[2] << 32) |
1387 ((u64) dd->ipath_pkeys[3] << 48);
1388 ipath_cdbg(VERBOSE, "p0 new pkey reg %llx\n",
1389 (unsigned long long) pkey);
1390 ipath_write_kreg(dd, dd->ipath_kregs->kr_partitionkey,
1396 EXPORT_SYMBOL_GPL(ipath_layer_set_pkeys);
1399 * ipath_layer_get_linkdowndefaultstate - get the default linkdown state
1400 * @dd: the infinipath device
1402 * Returns zero if the default is POLL, 1 if the default is SLEEP.
1404 int ipath_layer_get_linkdowndefaultstate(struct ipath_devdata *dd)
1406 return !!(dd->ipath_ibcctrl & INFINIPATH_IBCC_LINKDOWNDEFAULTSTATE);
1409 EXPORT_SYMBOL_GPL(ipath_layer_get_linkdowndefaultstate);
1412 * ipath_layer_set_linkdowndefaultstate - set the default linkdown state
1413 * @dd: the infinipath device
1414 * @sleep: the new state
1416 * Note that this will only take effect when the link state changes.
1418 int ipath_layer_set_linkdowndefaultstate(struct ipath_devdata *dd,
1422 dd->ipath_ibcctrl |= INFINIPATH_IBCC_LINKDOWNDEFAULTSTATE;
1424 dd->ipath_ibcctrl &= ~INFINIPATH_IBCC_LINKDOWNDEFAULTSTATE;
1425 ipath_write_kreg(dd, dd->ipath_kregs->kr_ibcctrl,
1430 EXPORT_SYMBOL_GPL(ipath_layer_set_linkdowndefaultstate);
1432 int ipath_layer_get_phyerrthreshold(struct ipath_devdata *dd)
1434 return (dd->ipath_ibcctrl >>
1435 INFINIPATH_IBCC_PHYERRTHRESHOLD_SHIFT) &
1436 INFINIPATH_IBCC_PHYERRTHRESHOLD_MASK;
1439 EXPORT_SYMBOL_GPL(ipath_layer_get_phyerrthreshold);
1442 * ipath_layer_set_phyerrthreshold - set the physical error threshold
1443 * @dd: the infinipath device
1444 * @n: the new threshold
1446 * Note that this will only take effect when the link state changes.
1448 int ipath_layer_set_phyerrthreshold(struct ipath_devdata *dd, unsigned n)
1452 v = (dd->ipath_ibcctrl >> INFINIPATH_IBCC_PHYERRTHRESHOLD_SHIFT) &
1453 INFINIPATH_IBCC_PHYERRTHRESHOLD_MASK;
1455 dd->ipath_ibcctrl &=
1456 ~(INFINIPATH_IBCC_PHYERRTHRESHOLD_MASK <<
1457 INFINIPATH_IBCC_PHYERRTHRESHOLD_SHIFT);
1458 dd->ipath_ibcctrl |=
1459 (u64) n << INFINIPATH_IBCC_PHYERRTHRESHOLD_SHIFT;
1460 ipath_write_kreg(dd, dd->ipath_kregs->kr_ibcctrl,
1466 EXPORT_SYMBOL_GPL(ipath_layer_set_phyerrthreshold);
1468 int ipath_layer_get_overrunthreshold(struct ipath_devdata *dd)
1470 return (dd->ipath_ibcctrl >>
1471 INFINIPATH_IBCC_OVERRUNTHRESHOLD_SHIFT) &
1472 INFINIPATH_IBCC_OVERRUNTHRESHOLD_MASK;
1475 EXPORT_SYMBOL_GPL(ipath_layer_get_overrunthreshold);
1478 * ipath_layer_set_overrunthreshold - set the overrun threshold
1479 * @dd: the infinipath device
1480 * @n: the new threshold
1482 * Note that this will only take effect when the link state changes.
1484 int ipath_layer_set_overrunthreshold(struct ipath_devdata *dd, unsigned n)
1488 v = (dd->ipath_ibcctrl >> INFINIPATH_IBCC_OVERRUNTHRESHOLD_SHIFT) &
1489 INFINIPATH_IBCC_OVERRUNTHRESHOLD_MASK;
1491 dd->ipath_ibcctrl &=
1492 ~(INFINIPATH_IBCC_OVERRUNTHRESHOLD_MASK <<
1493 INFINIPATH_IBCC_OVERRUNTHRESHOLD_SHIFT);
1494 dd->ipath_ibcctrl |=
1495 (u64) n << INFINIPATH_IBCC_OVERRUNTHRESHOLD_SHIFT;
1496 ipath_write_kreg(dd, dd->ipath_kregs->kr_ibcctrl,
1502 EXPORT_SYMBOL_GPL(ipath_layer_set_overrunthreshold);
1504 int ipath_layer_get_boardname(struct ipath_devdata *dd, char *name,
1507 return dd->ipath_f_get_boardname(dd, name, namelen);
1509 EXPORT_SYMBOL_GPL(ipath_layer_get_boardname);
1511 u32 ipath_layer_get_rcvhdrentsize(struct ipath_devdata *dd)
1513 return dd->ipath_rcvhdrentsize;
1515 EXPORT_SYMBOL_GPL(ipath_layer_get_rcvhdrentsize);