ipath_driver.c

来自「LINUX 2.6.17.4的源码」· C语言 代码 · 共 1,993 行 · 第 1/4 页

		   "kregvirt %p\n", addr, dd->ipath_kregbase,
		   dd->ipath_kregvirt);

	/*
	 * clear ipath_flags here instead of in ipath_init_chip as it is set
	 * by ipath_setup_htconfig.
	 */
	dd->ipath_flags = 0;

	if (dd->ipath_f_bus(dd, pdev))
		ipath_dev_err(dd, "Failed to setup config space; "
			      "continuing anyway\n");

	/*
	 * set up our interrupt handler; SA_SHIRQ probably not needed,
	 * since MSI interrupts shouldn't be shared but won't  hurt for now.
	 * check 0 irq after we return from chip-specific bus setup, since
	 * that can affect this due to setup
	 */
	if (!pdev->irq)
		ipath_dev_err(dd, "irq is 0, BIOS error?  Interrupts won't "
			      "work\n");
	else {
		ret = request_irq(pdev->irq, ipath_intr, SA_SHIRQ,
				  IPATH_DRV_NAME, dd);
		if (ret) {
			ipath_dev_err(dd, "Couldn't setup irq handler, "
				      "irq=%u: %d\n", pdev->irq, ret);
			goto bail_iounmap;
		}
	}

	ret = ipath_init_chip(dd, 0);	/* do the chip-specific init */
	if (ret)
		goto bail_iounmap;

	ret = ipath_enable_wc(dd);

	if (ret) {
		ipath_dev_err(dd, "Write combining not enabled "
			      "(err %d): performance may be poor\n",
			      -ret);
		ret = 0;
	}

	ipath_device_create_group(&pdev->dev, dd);
	ipathfs_add_device(dd);
	ipath_user_add(dd);
	ipath_layer_add(dd);

	goto bail;

bail_iounmap:
	iounmap((volatile void __iomem *) dd->ipath_kregbase);

bail_regions:
	pci_release_regions(pdev);

bail_disable:
	pci_disable_device(pdev);

bail_devdata:
	ipath_free_devdata(pdev, dd);

bail_rcvhdrtail:
	cleanup_port0_rcvhdrtail(pdev);

bail:
	return ret;
}

static void __devexit ipath_remove_one(struct pci_dev *pdev)
{
	struct ipath_devdata *dd;

	ipath_cdbg(VERBOSE, "removing, pdev=%p\n", pdev);
	if (!pdev)
		return;

	dd = pci_get_drvdata(pdev);
	ipath_layer_del(dd);
	ipath_user_del(dd);
	ipathfs_remove_device(dd);
	ipath_device_remove_group(&pdev->dev, dd);
	ipath_cdbg(VERBOSE, "Releasing pci memory regions, dd %p, "
		   "unit %u\n", dd, (u32) dd->ipath_unit);
	if (dd->ipath_kregbase) {
		ipath_cdbg(VERBOSE, "Unmapping kregbase %p\n",
			   dd->ipath_kregbase);
		iounmap((volatile void __iomem *) dd->ipath_kregbase);
		dd->ipath_kregbase = NULL;
	}
	pci_release_regions(pdev);
	ipath_cdbg(VERBOSE, "calling pci_disable_device\n");
	pci_disable_device(pdev);

	ipath_free_devdata(pdev, dd);
	cleanup_port0_rcvhdrtail(pdev);
}

/* general driver use */
DEFINE_MUTEX(ipath_mutex);

static DEFINE_SPINLOCK(ipath_pioavail_lock);

/**
 * ipath_disarm_piobufs - cancel a range of PIO buffers
 * @dd: the infinipath device
 * @first: the first PIO buffer to cancel
 * @cnt: the number of PIO buffers to cancel
 *
 * cancel a range of PIO buffers, used when they might be armed, but
 * not triggered.  Used at init to ensure buffer state, and also user
 * process close, in case it died while writing to a PIO buffer
 * Also after errors.
 */
void ipath_disarm_piobufs(struct ipath_devdata *dd, unsigned first,
			  unsigned cnt)
{
	unsigned i, last = first + cnt;
	u64 sendctrl, sendorig;

	ipath_cdbg(PKT, "disarm %u PIObufs first=%u\n", cnt, first);
	sendorig = dd->ipath_sendctrl | INFINIPATH_S_DISARM;
	for (i = first; i < last; i++) {
		sendctrl = sendorig |
			(i << INFINIPATH_S_DISARMPIOBUF_SHIFT);
		ipath_write_kreg(dd, dd->ipath_kregs->kr_sendctrl,
				 sendctrl);
	}

	/*
	 * Write it again with current value, in case ipath_sendctrl changed
	 * while we were looping; no critical bits that would require
	 * locking.
	 *
	 * Write a 0, and then the original value, reading scratch in
	 * between.  This seems to avoid a chip timing race that causes
	 * pioavail updates to memory to stop.
	 */
	ipath_write_kreg(dd, dd->ipath_kregs->kr_sendctrl,
			 0);
	sendorig = ipath_read_kreg64(dd, dd->ipath_kregs->kr_scratch);
	ipath_write_kreg(dd, dd->ipath_kregs->kr_sendctrl,
			 dd->ipath_sendctrl);
}

/**
 * ipath_wait_linkstate - wait for an IB link state change to occur
 * @dd: the infinipath device
 * @state: the state to wait for
 * @msecs: the number of milliseconds to wait
 *
 * wait up to msecs milliseconds for IB link state change to occur for
 * now, take the easy polling route.  Currently used only by
 * ipath_layer_set_linkstate.  Returns 0 if state reached, otherwise
 * -ETIMEDOUT state can have multiple states set, for any of several
 * transitions.
 */
int ipath_wait_linkstate(struct ipath_devdata *dd, u32 state, int msecs)
{
	dd->ipath_sma_state_wanted = state;
	wait_event_interruptible_timeout(ipath_sma_state_wait,
					 (dd->ipath_flags & state),
					 msecs_to_jiffies(msecs));
	dd->ipath_sma_state_wanted = 0;

	if (!(dd->ipath_flags & state)) {
		u64 val;
		ipath_cdbg(SMA, "Didn't reach linkstate %s within %u ms\n",
			   /* test INIT ahead of DOWN, both can be set */
			   (state & IPATH_LINKINIT) ? "INIT" :
			   ((state & IPATH_LINKDOWN) ? "DOWN" :
			    ((state & IPATH_LINKARMED) ? "ARM" : "ACTIVE")),
			   msecs);
		val = ipath_read_kreg64(dd, dd->ipath_kregs->kr_ibcstatus);
		ipath_cdbg(VERBOSE, "ibcc=%llx ibcstatus=%llx (%s)\n",
			   (unsigned long long) ipath_read_kreg64(
				   dd, dd->ipath_kregs->kr_ibcctrl),
			   (unsigned long long) val,
			   ipath_ibcstatus_str[val & 0xf]);
	}
	return (dd->ipath_flags & state) ? 0 : -ETIMEDOUT;
}

void ipath_decode_err(char *buf, size_t blen, ipath_err_t err)
{
	*buf = '\0';
	if (err & INFINIPATH_E_RHDRLEN)
		strlcat(buf, "rhdrlen ", blen);
	if (err & INFINIPATH_E_RBADTID)
		strlcat(buf, "rbadtid ", blen);
	if (err & INFINIPATH_E_RBADVERSION)
		strlcat(buf, "rbadversion ", blen);
	if (err & INFINIPATH_E_RHDR)
		strlcat(buf, "rhdr ", blen);
	if (err & INFINIPATH_E_RLONGPKTLEN)
		strlcat(buf, "rlongpktlen ", blen);
	if (err & INFINIPATH_E_RSHORTPKTLEN)
		strlcat(buf, "rshortpktlen ", blen);
	if (err & INFINIPATH_E_RMAXPKTLEN)
		strlcat(buf, "rmaxpktlen ", blen);
	if (err & INFINIPATH_E_RMINPKTLEN)
		strlcat(buf, "rminpktlen ", blen);
	if (err & INFINIPATH_E_RFORMATERR)
		strlcat(buf, "rformaterr ", blen);
	if (err & INFINIPATH_E_RUNSUPVL)
		strlcat(buf, "runsupvl ", blen);
	if (err & INFINIPATH_E_RUNEXPCHAR)
		strlcat(buf, "runexpchar ", blen);
	if (err & INFINIPATH_E_RIBFLOW)
		strlcat(buf, "ribflow ", blen);
	if (err & INFINIPATH_E_REBP)
		strlcat(buf, "EBP ", blen);
	if (err & INFINIPATH_E_SUNDERRUN)
		strlcat(buf, "sunderrun ", blen);
	if (err & INFINIPATH_E_SPIOARMLAUNCH)
		strlcat(buf, "spioarmlaunch ", blen);
	if (err & INFINIPATH_E_SUNEXPERRPKTNUM)
		strlcat(buf, "sunexperrpktnum ", blen);
	if (err & INFINIPATH_E_SDROPPEDDATAPKT)
		strlcat(buf, "sdroppeddatapkt ", blen);
	if (err & INFINIPATH_E_SDROPPEDSMPPKT)
		strlcat(buf, "sdroppedsmppkt ", blen);
	if (err & INFINIPATH_E_SMAXPKTLEN)
		strlcat(buf, "smaxpktlen ", blen);
	if (err & INFINIPATH_E_SMINPKTLEN)
		strlcat(buf, "sminpktlen ", blen);
	if (err & INFINIPATH_E_SUNSUPVL)
		strlcat(buf, "sunsupVL ", blen);
	if (err & INFINIPATH_E_SPKTLEN)
		strlcat(buf, "spktlen ", blen);
	if (err & INFINIPATH_E_INVALIDADDR)
		strlcat(buf, "invalidaddr ", blen);
	if (err & INFINIPATH_E_RICRC)
		strlcat(buf, "CRC ", blen);
	if (err & INFINIPATH_E_RVCRC)
		strlcat(buf, "VCRC ", blen);
	if (err & INFINIPATH_E_RRCVEGRFULL)
		strlcat(buf, "rcvegrfull ", blen);
	if (err & INFINIPATH_E_RRCVHDRFULL)
		strlcat(buf, "rcvhdrfull ", blen);
	if (err & INFINIPATH_E_IBSTATUSCHANGED)
		strlcat(buf, "ibcstatuschg ", blen);
	if (err & INFINIPATH_E_RIBLOSTLINK)
		strlcat(buf, "riblostlink ", blen);
	if (err & INFINIPATH_E_HARDWARE)
		strlcat(buf, "hardware ", blen);
	if (err & INFINIPATH_E_RESET)
		strlcat(buf, "reset ", blen);
}

/**
 * get_rhf_errstring - decode RHF errors
 * @err: the err number
 * @msg: the output buffer
 * @len: the length of the output buffer
 *
 * only used one place now, may want more later
 */
static void get_rhf_errstring(u32 err, char *msg, size_t len)
{
	/* if no errors, and so don't need to check what's first */
	*msg = '\0';

	if (err & INFINIPATH_RHF_H_ICRCERR)
		strlcat(msg, "icrcerr ", len);
	if (err & INFINIPATH_RHF_H_VCRCERR)
		strlcat(msg, "vcrcerr ", len);
	if (err & INFINIPATH_RHF_H_PARITYERR)
		strlcat(msg, "parityerr ", len);
	if (err & INFINIPATH_RHF_H_LENERR)
		strlcat(msg, "lenerr ", len);
	if (err & INFINIPATH_RHF_H_MTUERR)
		strlcat(msg, "mtuerr ", len);
	if (err & INFINIPATH_RHF_H_IHDRERR)
		/* infinipath hdr checksum error */
		strlcat(msg, "ipathhdrerr ", len);
	if (err & INFINIPATH_RHF_H_TIDERR)
		strlcat(msg, "tiderr ", len);
	if (err & INFINIPATH_RHF_H_MKERR)
		/* bad port, offset, etc. */
		strlcat(msg, "invalid ipathhdr ", len);
	if (err & INFINIPATH_RHF_H_IBERR)
		strlcat(msg, "iberr ", len);
	if (err & INFINIPATH_RHF_L_SWA)
		strlcat(msg, "swA ", len);
	if (err & INFINIPATH_RHF_L_SWB)
		strlcat(msg, "swB ", len);
}

/**
 * ipath_get_egrbuf - get an eager buffer
 * @dd: the infinipath device
 * @bufnum: the eager buffer to get
 * @err: unused
 *
 * must only be called if ipath_pd[port] is known to be allocated
 */
static inline void *ipath_get_egrbuf(struct ipath_devdata *dd, u32 bufnum,
				     int err)
{
	return dd->ipath_port0_skbs ?
		(void *)dd->ipath_port0_skbs[bufnum]->data : NULL;
}

/**
 * ipath_alloc_skb - allocate an skb and buffer with possible constraints
 * @dd: the infinipath device
 * @gfp_mask: the sk_buff SFP mask
 */
struct sk_buff *ipath_alloc_skb(struct ipath_devdata *dd,
				gfp_t gfp_mask)
{
	struct sk_buff *skb;
	u32 len;

	/*
	 * Only fully supported way to handle this is to allocate lots
	 * extra, align as needed, and then do skb_reserve().  That wastes
	 * a lot of memory...  I'll have to hack this into infinipath_copy
	 * also.
	 */

	/*
	 * We need 4 extra bytes for unaligned transfer copying
	 */
	if (dd->ipath_flags & IPATH_4BYTE_TID) {
		/* we need a 4KB multiple alignment, and there is no way
		 * to do it except to allocate extra and then skb_reserve
		 * enough to bring it up to the right alignment.
		 */
		len = dd->ipath_ibmaxlen + 4 + (1 << 11) - 1;
	}
	else
		len = dd->ipath_ibmaxlen + 4;
	skb = __dev_alloc_skb(len, gfp_mask);
	if (!skb) {
		ipath_dev_err(dd, "Failed to allocate skbuff, length %u\n",
			      len);
		goto bail;
	}
	if (dd->ipath_flags & IPATH_4BYTE_TID) {
		u32 una = ((1 << 11) - 1) & (unsigned long)(skb->data + 4);
		if (una)
			skb_reserve(skb, 4 + (1 << 11) - una);
		else
			skb_reserve(skb, 4);
	} else
		skb_reserve(skb, 4);

bail:
	return skb;
}

/**
 * ipath_rcv_layer - receive a packet for the layered (ethernet) driver
 * @dd: the infinipath device
 * @etail: the sk_buff number
 * @tlen: the total packet length
 * @hdr: the ethernet header
 *
 * Separate routine for better overall optimization
 */
static void ipath_rcv_layer(struct ipath_devdata *dd, u32 etail,
			    u32 tlen, struct ether_header *hdr)
{
	u32 elen;
	u8 pad, *bthbytes;
	struct sk_buff *skb, *nskb;

	if (dd->ipath_port0_skbs && hdr->sub_opcode == OPCODE_ENCAP) {
		/*
		 * Allocate a new sk_buff to replace the one we give
		 * to the network stack.
		 */
		nskb = ipath_alloc_skb(dd, GFP_ATOMIC);
		if (!nskb) {
			/* count OK packets that we drop */
			ipath_stats.sps_krdrops++;
			return;
		}

		bthbytes = (u8 *) hdr->bth;
		pad = (bthbytes[1] >> 4) & 3;
		/* +CRC32 */
		elen = tlen - (sizeof(*hdr) + pad + sizeof(u32));

		skb = dd->ipath_port0_skbs[etail];
		dd->ipath_port0_skbs[etail] = nskb;
		skb_put(skb, elen);

		dd->ipath_f_put_tid(dd, etail + (u64 __iomem *)
				    ((char __iomem *) dd->ipath_kregbase
				     + dd->ipath_rcvegrbase), 0,
				    virt_to_phys(nskb->data));

		__ipath_layer_rcv(dd, hdr, skb);

		/* another ether packet received */
		ipath_stats.sps_ether_rpkts++;
	}
	else if (hdr->sub_opcode == OPCODE_LID_ARP)
		__ipath_layer_rcv_lid(dd, hdr);
}

/*
 * ipath_kreceive - receive a packet
 * @dd: the infinipath device
 *
 * called from interrupt handler for errors or receive interrupt
 */
void ipath_kreceive(struct ipath_devdata *dd)
{
	u64 *rc;
	void *ebuf;
	const u32 rsize = dd->ipath_rcvhdrentsize;	/* words */
	const u32 maxcnt = dd->ipath_rcvhdrcnt * rsize;	/* words */
	u32 etail = -1, l, hdrqtail;
	struct ips_message_header *hdr;
	u32 eflags, i, etype, tlen, pkttot = 0;
	static u64 totcalls;	/* stats, may eventually remove */
	char emsg[128];

	if (!dd->ipath_hdrqtailptr) {
		ipath_dev_err(dd,
			      "hdrqtailptr not set, can't do receives\n");
		goto bail;
	}

	/* There is already a thread processing this queue. */
	if (test_and_set_bit(0, &dd->ipath_rcv_pending))
		goto bail;

	if (dd->ipath_port0head ==
	    (u32)le64_to_cpu(*dd->ipath_hdrqtailptr))
		goto done;

gotmore:
	/*
	 * read only once at start.  If in flood situation, this helps
	 * performance slightly.  If more arrive while we are processing,
	 * we'll come back here and do them
	 */
	hdrqtail = (u32)le64_to_cpu(*dd->ipath_hdrqtailptr);

	for (i = 0, l = dd->ipath_port0head; l != hdrqtail; i++) {
		u32 qp;
		u8 *bthbytes;

		rc = (u64 *) (dd->ipath_pd[0]->port_rcvhdrq + (l << 2));
		hdr = (struct ips_message_header *)&rc[1];
		/*
		 * could make a network order version of IPATH_KD_QP, and
		 * do the obvious shift before masking to speed this up.
		 */
		qp = ntohl(hdr->bth[1]) & 0xffffff;
		bthbytes = (u8 *) hdr->bth;

		eflags = ips_get_hdr_err_flags((__le32 *) rc);
		etype = ips_get_rcv_type((__le32 *) rc);
		/* total length */
		tlen = ips_get_length_in_bytes((__le32 *) rc);
		ebuf = NULL;
		if (etype != RCVHQ_RCV_TYPE_EXPECTED) {
			/*
			 * it turns out that the chips uses an eager buffer
			 * for all non-expected packets, whether it "needs"
			 * one or not.  So always get the index, but don't
			 * set ebuf (so we try to copy data) unless the
			 * length requires it.
			 */
			etail = ips_get_index((__le32 *) rc);
			if (tlen > sizeof(*hdr) ||
			    etype == RCVHQ_RCV_TYPE_NON_KD)
				ebuf = ipath_get_egrbuf(dd, etail, 0);
		}

		/*
		 * both tiderr and ipathhdrerr are set for all plain IB
		 * packets; only ipathhdrerr should be set.
		 */

		if (etype != RCVHQ_RCV_TYPE_NON_KD && etype !=
		    RCVHQ_RCV_TYPE_ERROR && ips_get_ipath_ver(
			    hdr->iph.ver_port_tid_offset) !=
		    IPS_PROTO_VERSION) {
			ipath_cdbg(PKT, "Bad InfiniPath protocol version "
				   "%x\n", etype);
		}

		if (eflags & ~(INFINIPATH_RHF_H_TIDERR |
			       INFINIPATH_RHF_H_IHDRERR)) {
			get_rhf_errstring(eflags, emsg, sizeof emsg);
			ipath_cdbg(PKT, "RHFerrs %x hdrqtail=%x typ=%u "
				   "tlen=%x opcode=%x egridx=%x: %s\n",
				   eflags, l, etype, tlen, bthbytes[0],
				   ips_get_index((__le32 *) rc), emsg);
		} else if (etype == RCVHQ_RCV_TYPE_NON_KD) {