IB/hfi1: Remove snoop/diag interface
authorDennis Dalessandro <dennis.dalessandro@intel.com>
Thu, 19 May 2016 12:26:10 +0000 (05:26 -0700)
committerDoug Ledford <dledford@redhat.com>
Thu, 26 May 2016 15:23:18 +0000 (11:23 -0400)
The snoop/diag interface is better served by an implementation which is
more general and usable by other drivers perhaps. Go ahead and remove
the code now and get rid of the char dev. We can put the feature back
when we have a more agreeable solution.

Reviewed-by: Dean Luick <dean.luick@intel.com>
Reviewed-by: Mike Marciniszyn <mike.marciniszyn@intel.com>
Signed-off-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
drivers/staging/rdma/hfi1/Makefile
drivers/staging/rdma/hfi1/diag.c [deleted file]
drivers/staging/rdma/hfi1/file_ops.c

index 8dc59382ee96cd781f643259eb334eef4adef458..9b5382c94b0c86d1a2fa96f8c60e9069bd832b0c 100644 (file)
@@ -7,7 +7,7 @@
 #
 obj-$(CONFIG_INFINIBAND_HFI1) += hfi1.o
 
-hfi1-y := affinity.o chip.o device.o diag.o driver.o efivar.o \
+hfi1-y := affinity.o chip.o device.o driver.o efivar.o \
        eprom.o file_ops.o firmware.o \
        init.o intr.o mad.o mmu_rb.o pcie.o pio.o pio_copy.o platform.o \
        qp.o qsfp.o rc.o ruc.o sdma.o sysfs.o trace.o twsi.o \
diff --git a/drivers/staging/rdma/hfi1/diag.c b/drivers/staging/rdma/hfi1/diag.c
deleted file mode 100644 (file)
index 3a679e4..0000000
+++ /dev/null
@@ -1,1925 +0,0 @@
-/*
- * Copyright(c) 2015, 2016 Intel Corporation.
- *
- * This file is provided under a dual BSD/GPLv2 license.  When using or
- * redistributing this file, you may do so under either license.
- *
- * GPL LICENSE SUMMARY
- *
- * This program is free software; you can redistribute it and/or modify
- * it under the terms of version 2 of the GNU General Public License as
- * published by the Free Software Foundation.
- *
- * This program is distributed in the hope that it will be useful, but
- * WITHOUT ANY WARRANTY; without even the implied warranty of
- * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
- * General Public License for more details.
- *
- * BSD LICENSE
- *
- * Redistribution and use in source and binary forms, with or without
- * modification, are permitted provided that the following conditions
- * are met:
- *
- *  - Redistributions of source code must retain the above copyright
- *    notice, this list of conditions and the following disclaimer.
- *  - Redistributions in binary form must reproduce the above copyright
- *    notice, this list of conditions and the following disclaimer in
- *    the documentation and/or other materials provided with the
- *    distribution.
- *  - Neither the name of Intel Corporation nor the names of its
- *    contributors may be used to endorse or promote products derived
- *    from this software without specific prior written permission.
- *
- * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
- * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
- * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
- * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
- * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
- * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
- * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
- * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
- * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
- * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
- * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
- *
- */
-
-/*
- * This file contains support for diagnostic functions.  It is accessed by
- * opening the hfi1_diag device, normally minor number 129.  Diagnostic use
- * of the chip may render the chip or board unusable until the driver
- * is unloaded, or in some cases, until the system is rebooted.
- *
- * Accesses to the chip through this interface are not similar to going
- * through the /sys/bus/pci resource mmap interface.
- */
-
-#include <linux/io.h>
-#include <linux/pci.h>
-#include <linux/poll.h>
-#include <linux/vmalloc.h>
-#include <linux/export.h>
-#include <linux/fs.h>
-#include <linux/uaccess.h>
-#include <linux/module.h>
-#include <rdma/ib_smi.h>
-#include "hfi.h"
-#include "device.h"
-#include "common.h"
-#include "verbs_txreq.h"
-#include "trace.h"
-
-#undef pr_fmt
-#define pr_fmt(fmt) DRIVER_NAME ": " fmt
-#define snoop_dbg(fmt, ...) \
-       hfi1_cdbg(SNOOP, fmt, ##__VA_ARGS__)
-
-/* Snoop option mask */
-#define SNOOP_DROP_SEND                BIT(0)
-#define SNOOP_USE_METADATA     BIT(1)
-#define SNOOP_SET_VL0TOVL15     BIT(2)
-
-static u8 snoop_flags;
-
-/*
- * Extract packet length from LRH header.
- * This is in Dwords so multiply by 4 to get size in bytes
- */
-#define HFI1_GET_PKT_LEN(x)      (((be16_to_cpu((x)->lrh[2]) & 0xFFF)) << 2)
-
-enum hfi1_filter_status {
-       HFI1_FILTER_HIT,
-       HFI1_FILTER_ERR,
-       HFI1_FILTER_MISS
-};
-
-/* snoop processing functions */
-rhf_rcv_function_ptr snoop_rhf_rcv_functions[8] = {
-       [RHF_RCV_TYPE_EXPECTED] = snoop_recv_handler,
-       [RHF_RCV_TYPE_EAGER]    = snoop_recv_handler,
-       [RHF_RCV_TYPE_IB]       = snoop_recv_handler,
-       [RHF_RCV_TYPE_ERROR]    = snoop_recv_handler,
-       [RHF_RCV_TYPE_BYPASS]   = snoop_recv_handler,
-       [RHF_RCV_TYPE_INVALID5] = process_receive_invalid,
-       [RHF_RCV_TYPE_INVALID6] = process_receive_invalid,
-       [RHF_RCV_TYPE_INVALID7] = process_receive_invalid
-};
-
-/* Snoop packet structure */
-struct snoop_packet {
-       struct list_head list;
-       u32 total_len;
-       u8 data[];
-};
-
-/* Do not make these an enum or it will blow up the capture_md */
-#define PKT_DIR_EGRESS 0x0
-#define PKT_DIR_INGRESS 0x1
-
-/* Packet capture metadata returned to the user with the packet. */
-struct capture_md {
-       u8 port;
-       u8 dir;
-       u8 reserved[6];
-       union {
-               u64 pbc;
-               u64 rhf;
-       } u;
-};
-
-static atomic_t diagpkt_count = ATOMIC_INIT(0);
-static struct cdev diagpkt_cdev;
-static struct device *diagpkt_device;
-
-static ssize_t diagpkt_write(struct file *fp, const char __user *data,
-                            size_t count, loff_t *off);
-
-static const struct file_operations diagpkt_file_ops = {
-       .owner = THIS_MODULE,
-       .write = diagpkt_write,
-       .llseek = noop_llseek,
-};
-
-/*
- * This is used for communication with user space for snoop extended IOCTLs
- */
-struct hfi1_link_info {
-       __be64 node_guid;
-       u8 port_mode;
-       u8 port_state;
-       u16 link_speed_active;
-       u16 link_width_active;
-       u16 vl15_init;
-       u8 port_number;
-       /*
-        * Add padding to make this a full IB SMP payload. Note: changing the
-        * size of this structure will make the IOCTLs created with _IOWR
-        * change.
-        * Be sure to run tests on all IOCTLs when making changes to this
-        * structure.
-        */
-       u8 res[47];
-};
-
-/*
- * This starts our ioctl sequence numbers *way* off from the ones
- * defined in ib_core.
- */
-#define SNOOP_CAPTURE_VERSION 0x1
-
-#define IB_IOCTL_MAGIC          0x1b /* See Documentation/ioctl-number.txt */
-#define HFI1_SNOOP_IOC_MAGIC IB_IOCTL_MAGIC
-#define HFI1_SNOOP_IOC_BASE_SEQ 0x80
-
-#define HFI1_SNOOP_IOCGETLINKSTATE \
-       _IO(HFI1_SNOOP_IOC_MAGIC, HFI1_SNOOP_IOC_BASE_SEQ)
-#define HFI1_SNOOP_IOCSETLINKSTATE \
-       _IO(HFI1_SNOOP_IOC_MAGIC, HFI1_SNOOP_IOC_BASE_SEQ + 1)
-#define HFI1_SNOOP_IOCCLEARQUEUE \
-       _IO(HFI1_SNOOP_IOC_MAGIC, HFI1_SNOOP_IOC_BASE_SEQ + 2)
-#define HFI1_SNOOP_IOCCLEARFILTER \
-       _IO(HFI1_SNOOP_IOC_MAGIC, HFI1_SNOOP_IOC_BASE_SEQ + 3)
-#define HFI1_SNOOP_IOCSETFILTER \
-       _IO(HFI1_SNOOP_IOC_MAGIC, HFI1_SNOOP_IOC_BASE_SEQ + 4)
-#define HFI1_SNOOP_IOCGETVERSION \
-       _IO(HFI1_SNOOP_IOC_MAGIC, HFI1_SNOOP_IOC_BASE_SEQ + 5)
-#define HFI1_SNOOP_IOCSET_OPTS \
-       _IO(HFI1_SNOOP_IOC_MAGIC, HFI1_SNOOP_IOC_BASE_SEQ + 6)
-
-/*
- * These offsets +6/+7 could change, but these are already known and used
- * IOCTL numbers so don't change them without a good reason.
- */
-#define HFI1_SNOOP_IOCGETLINKSTATE_EXTRA \
-       _IOWR(HFI1_SNOOP_IOC_MAGIC, HFI1_SNOOP_IOC_BASE_SEQ + 6, \
-               struct hfi1_link_info)
-#define HFI1_SNOOP_IOCSETLINKSTATE_EXTRA \
-       _IOWR(HFI1_SNOOP_IOC_MAGIC, HFI1_SNOOP_IOC_BASE_SEQ + 7, \
-               struct hfi1_link_info)
-
-static int hfi1_snoop_open(struct inode *in, struct file *fp);
-static ssize_t hfi1_snoop_read(struct file *fp, char __user *data,
-                              size_t pkt_len, loff_t *off);
-static ssize_t hfi1_snoop_write(struct file *fp, const char __user *data,
-                               size_t count, loff_t *off);
-static long hfi1_ioctl(struct file *fp, unsigned int cmd, unsigned long arg);
-static unsigned int hfi1_snoop_poll(struct file *fp,
-                                   struct poll_table_struct *wait);
-static int hfi1_snoop_release(struct inode *in, struct file *fp);
-
-struct hfi1_packet_filter_command {
-       int opcode;
-       int length;
-       void *value_ptr;
-};
-
-/* Can't re-use PKT_DIR_*GRESS here because 0 means no packets for this */
-#define HFI1_SNOOP_INGRESS 0x1
-#define HFI1_SNOOP_EGRESS  0x2
-
-enum hfi1_packet_filter_opcodes {
-       FILTER_BY_LID,
-       FILTER_BY_DLID,
-       FILTER_BY_MAD_MGMT_CLASS,
-       FILTER_BY_QP_NUMBER,
-       FILTER_BY_PKT_TYPE,
-       FILTER_BY_SERVICE_LEVEL,
-       FILTER_BY_PKEY,
-       FILTER_BY_DIRECTION,
-};
-
-static const struct file_operations snoop_file_ops = {
-       .owner = THIS_MODULE,
-       .open = hfi1_snoop_open,
-       .read = hfi1_snoop_read,
-       .unlocked_ioctl = hfi1_ioctl,
-       .poll = hfi1_snoop_poll,
-       .write = hfi1_snoop_write,
-       .release = hfi1_snoop_release
-};
-
-struct hfi1_filter_array {
-       int (*filter)(void *, void *, void *);
-};
-
-static int hfi1_filter_lid(void *ibhdr, void *packet_data, void *value);
-static int hfi1_filter_dlid(void *ibhdr, void *packet_data, void *value);
-static int hfi1_filter_mad_mgmt_class(void *ibhdr, void *packet_data,
-                                     void *value);
-static int hfi1_filter_qp_number(void *ibhdr, void *packet_data, void *value);
-static int hfi1_filter_ibpacket_type(void *ibhdr, void *packet_data,
-                                    void *value);
-static int hfi1_filter_ib_service_level(void *ibhdr, void *packet_data,
-                                       void *value);
-static int hfi1_filter_ib_pkey(void *ibhdr, void *packet_data, void *value);
-static int hfi1_filter_direction(void *ibhdr, void *packet_data, void *value);
-
-static const struct hfi1_filter_array hfi1_filters[] = {
-       { hfi1_filter_lid },
-       { hfi1_filter_dlid },
-       { hfi1_filter_mad_mgmt_class },
-       { hfi1_filter_qp_number },
-       { hfi1_filter_ibpacket_type },
-       { hfi1_filter_ib_service_level },
-       { hfi1_filter_ib_pkey },
-       { hfi1_filter_direction },
-};
-
-#define HFI1_MAX_FILTERS       ARRAY_SIZE(hfi1_filters)
-#define HFI1_DIAG_MINOR_BASE   129
-
-static int hfi1_snoop_add(struct hfi1_devdata *dd, const char *name);
-
-int hfi1_diag_add(struct hfi1_devdata *dd)
-{
-       char name[16];
-       int ret = 0;
-
-       snprintf(name, sizeof(name), "%s_diagpkt%d", class_name(),
-                dd->unit);
-       /*
-        * Do this for each device as opposed to the normal diagpkt
-        * interface which is one per host
-        */
-       ret = hfi1_snoop_add(dd, name);
-       if (ret)
-               dd_dev_err(dd, "Unable to init snoop/capture device");
-
-       snprintf(name, sizeof(name), "%s_diagpkt", class_name());
-       if (atomic_inc_return(&diagpkt_count) == 1) {
-               ret = hfi1_cdev_init(HFI1_DIAGPKT_MINOR, name,
-                                    &diagpkt_file_ops, &diagpkt_cdev,
-                                    &diagpkt_device, false);
-       }
-
-       return ret;
-}
-
-/* this must be called w/ dd->snoop_in_lock held */
-static void drain_snoop_list(struct list_head *queue)
-{
-       struct list_head *pos, *q;
-       struct snoop_packet *packet;
-
-       list_for_each_safe(pos, q, queue) {
-               packet = list_entry(pos, struct snoop_packet, list);
-               list_del(pos);
-               kfree(packet);
-       }
-}
-
-static void hfi1_snoop_remove(struct hfi1_devdata *dd)
-{
-       unsigned long flags = 0;
-
-       spin_lock_irqsave(&dd->hfi1_snoop.snoop_lock, flags);
-       drain_snoop_list(&dd->hfi1_snoop.queue);
-       hfi1_cdev_cleanup(&dd->hfi1_snoop.cdev, &dd->hfi1_snoop.class_dev);
-       spin_unlock_irqrestore(&dd->hfi1_snoop.snoop_lock, flags);
-}
-
-void hfi1_diag_remove(struct hfi1_devdata *dd)
-{
-       hfi1_snoop_remove(dd);
-       if (atomic_dec_and_test(&diagpkt_count))
-               hfi1_cdev_cleanup(&diagpkt_cdev, &diagpkt_device);
-       hfi1_cdev_cleanup(&dd->diag_cdev, &dd->diag_device);
-}
-
-/*
- * Allocated structure shared between the credit return mechanism and
- * diagpkt_send().
- */
-struct diagpkt_wait {
-       struct completion credits_returned;
-       int code;
-       atomic_t count;
-};
-
-/*
- * When each side is finished with the structure, they call this.
- * The last user frees the structure.
- */
-static void put_diagpkt_wait(struct diagpkt_wait *wait)
-{
-       if (atomic_dec_and_test(&wait->count))
-               kfree(wait);
-}
-
-/*
- * Callback from the credit return code.  Set the complete, which
- * will let diapkt_send() continue.
- */
-static void diagpkt_complete(void *arg, int code)
-{
-       struct diagpkt_wait *wait = (struct diagpkt_wait *)arg;
-
-       wait->code = code;
-       complete(&wait->credits_returned);
-       put_diagpkt_wait(wait); /* finished with the structure */
-}
-
-/**
- * diagpkt_send - send a packet
- * @dp: diag packet descriptor
- */
-static ssize_t diagpkt_send(struct diag_pkt *dp)
-{
-       struct hfi1_devdata *dd;
-       struct send_context *sc;
-       struct pio_buf *pbuf;
-       u32 *tmpbuf = NULL;
-       ssize_t ret = 0;
-       u32 pkt_len, total_len;
-       pio_release_cb credit_cb = NULL;
-       void *credit_arg = NULL;
-       struct diagpkt_wait *wait = NULL;
-       int trycount = 0;
-
-       dd = hfi1_lookup(dp->unit);
-       if (!dd || !(dd->flags & HFI1_PRESENT) || !dd->kregbase) {
-               ret = -ENODEV;
-               goto bail;
-       }
-       if (!(dd->flags & HFI1_INITTED)) {
-               /* no hardware, freeze, etc. */
-               ret = -ENODEV;
-               goto bail;
-       }
-
-       if (dp->version != _DIAG_PKT_VERS) {
-               dd_dev_err(dd, "Invalid version %u for diagpkt_write\n",
-                          dp->version);
-               ret = -EINVAL;
-               goto bail;
-       }
-
-       /* send count must be an exact number of dwords */
-       if (dp->len & 3) {
-               ret = -EINVAL;
-               goto bail;
-       }
-
-       /* there is only port 1 */
-       if (dp->port != 1) {
-               ret = -EINVAL;
-               goto bail;
-       }
-
-       /* need a valid context */
-       if (dp->sw_index >= dd->num_send_contexts) {
-               ret = -EINVAL;
-               goto bail;
-       }
-       /* can only use kernel contexts */
-       if (dd->send_contexts[dp->sw_index].type != SC_KERNEL &&
-           dd->send_contexts[dp->sw_index].type != SC_VL15) {
-               ret = -EINVAL;
-               goto bail;
-       }
-       /* must be allocated */
-       sc = dd->send_contexts[dp->sw_index].sc;
-       if (!sc) {
-               ret = -EINVAL;
-               goto bail;
-       }
-       /* must be enabled */
-       if (!(sc->flags & SCF_ENABLED)) {
-               ret = -EINVAL;
-               goto bail;
-       }
-
-       /* allocate a buffer and copy the data in */
-       tmpbuf = vmalloc(dp->len);
-       if (!tmpbuf) {
-               ret = -ENOMEM;
-               goto bail;
-       }
-
-       if (copy_from_user(tmpbuf,
-                          (const void __user *)(unsigned long)dp->data,
-                          dp->len)) {
-               ret = -EFAULT;
-               goto bail;
-       }
-
-       /*
-        * pkt_len is how much data we have to write, includes header and data.
-        * total_len is length of the packet in Dwords plus the PBC should not
-        * include the CRC.
-        */
-       pkt_len = dp->len >> 2;
-       total_len = pkt_len + 2; /* PBC + packet */
-
-       /* if 0, fill in a default */
-       if (dp->pbc == 0) {
-               struct hfi1_pportdata *ppd = dd->pport;
-
-               hfi1_cdbg(PKT, "Generating PBC");
-               dp->pbc = create_pbc(ppd, 0, 0, 0, total_len);
-       } else {
-               hfi1_cdbg(PKT, "Using passed in PBC");
-       }
-
-       hfi1_cdbg(PKT, "Egress PBC content is 0x%llx", dp->pbc);
-
-       /*
-        * The caller wants to wait until the packet is sent and to
-        * check for errors.  The best we can do is wait until
-        * the buffer credits are returned and check if any packet
-        * error has occurred.  If there are any late errors, this
-        * could miss it.  If there are other senders who generate
-        * an error, this may find it.  However, in general, it
-        * should catch most.
-        */
-       if (dp->flags & F_DIAGPKT_WAIT) {
-               /* always force a credit return */
-               dp->pbc |= PBC_CREDIT_RETURN;
-               /* turn on credit return interrupts */
-               sc_add_credit_return_intr(sc);
-               wait = kmalloc(sizeof(*wait), GFP_KERNEL);
-               if (!wait) {
-                       ret = -ENOMEM;
-                       goto bail;
-               }
-               init_completion(&wait->credits_returned);
-               atomic_set(&wait->count, 2);
-               wait->code = PRC_OK;
-
-               credit_cb = diagpkt_complete;
-               credit_arg = wait;
-       }
-
-retry:
-       pbuf = sc_buffer_alloc(sc, total_len, credit_cb, credit_arg);
-       if (!pbuf) {
-               if (trycount == 0) {
-                       /* force a credit return and try again */
-                       sc_return_credits(sc);
-                       trycount = 1;
-                       goto retry;
-               }
-               /*
-                * No send buffer means no credit callback.  Undo
-                * the wait set-up that was done above.  We free wait
-                * because the callback will never be called.
-                */
-               if (dp->flags & F_DIAGPKT_WAIT) {
-                       sc_del_credit_return_intr(sc);
-                       kfree(wait);
-                       wait = NULL;
-               }
-               ret = -ENOSPC;
-               goto bail;
-       }
-
-       pio_copy(dd, pbuf, dp->pbc, tmpbuf, pkt_len);
-       /* no flush needed as the HW knows the packet size */
-
-       ret = sizeof(*dp);
-
-       if (dp->flags & F_DIAGPKT_WAIT) {
-               /* wait for credit return */
-               ret = wait_for_completion_interruptible(
-                                               &wait->credits_returned);
-               /*
-                * If the wait returns an error, the wait was interrupted,
-                * e.g. with a ^C in the user program.  The callback is
-                * still pending.  This is OK as the wait structure is
-                * kmalloc'ed and the structure will free itself when
-                * all users are done with it.
-                *
-                * A context disable occurs on a send context restart, so
-                * include that in the list of errors below to check for.
-                * NOTE: PRC_FILL_ERR is at best informational and cannot
-                * be depended on.
-                */
-               if (!ret && (((wait->code & PRC_STATUS_ERR) ||
-                             (wait->code & PRC_FILL_ERR) ||
-                             (wait->code & PRC_SC_DISABLE))))
-                       ret = -EIO;
-
-               put_diagpkt_wait(wait); /* finished with the structure */
-               sc_del_credit_return_intr(sc);
-       }
-
-bail:
-       vfree(tmpbuf);
-       return ret;
-}
-
-static ssize_t diagpkt_write(struct file *fp, const char __user *data,
-                            size_t count, loff_t *off)
-{
-       struct hfi1_devdata *dd;
-       struct send_context *sc;
-       u8 vl;
-
-       struct diag_pkt dp;
-
-       if (count != sizeof(dp))
-               return -EINVAL;
-
-       if (copy_from_user(&dp, data, sizeof(dp)))
-               return -EFAULT;
-
-       /*
-       * The Send Context is derived from the PbcVL value
-       * if PBC is populated
-       */
-       if (dp.pbc) {
-               dd = hfi1_lookup(dp.unit);
-               if (!dd)
-                       return -ENODEV;
-               vl = (dp.pbc >> PBC_VL_SHIFT) & PBC_VL_MASK;
-               sc = dd->vld[vl].sc;
-               if (sc) {
-                       dp.sw_index = sc->sw_index;
-                       hfi1_cdbg(
-                              PKT,
-                              "Packet sent over VL %d via Send Context %u(%u)",
-                              vl, sc->sw_index, sc->hw_context);
-               }
-       }
-
-       return diagpkt_send(&dp);
-}
-
-static int hfi1_snoop_add(struct hfi1_devdata *dd, const char *name)
-{
-       int ret = 0;
-
-       dd->hfi1_snoop.mode_flag = 0;
-       spin_lock_init(&dd->hfi1_snoop.snoop_lock);
-       INIT_LIST_HEAD(&dd->hfi1_snoop.queue);
-       init_waitqueue_head(&dd->hfi1_snoop.waitq);
-
-       ret = hfi1_cdev_init(HFI1_SNOOP_CAPTURE_BASE + dd->unit, name,
-                            &snoop_file_ops,
-                            &dd->hfi1_snoop.cdev, &dd->hfi1_snoop.class_dev,
-                            false);
-
-       if (ret) {
-               dd_dev_err(dd, "Couldn't create %s device: %d", name, ret);
-               hfi1_cdev_cleanup(&dd->hfi1_snoop.cdev,
-                                 &dd->hfi1_snoop.class_dev);
-       }
-
-       return ret;
-}
-
-static struct hfi1_devdata *hfi1_dd_from_sc_inode(struct inode *in)
-{
-       int unit = iminor(in) - HFI1_SNOOP_CAPTURE_BASE;
-       struct hfi1_devdata *dd;
-
-       dd = hfi1_lookup(unit);
-       return dd;
-}
-
-/* clear or restore send context integrity checks */
-static void adjust_integrity_checks(struct hfi1_devdata *dd)
-{
-       struct send_context *sc;
-       unsigned long sc_flags;
-       int i;
-
-       spin_lock_irqsave(&dd->sc_lock, sc_flags);
-       for (i = 0; i < dd->num_send_contexts; i++) {
-               int enable;
-
-               sc = dd->send_contexts[i].sc;
-
-               if (!sc)
-                       continue;       /* not allocated */
-
-               enable = likely(!HFI1_CAP_IS_KSET(NO_INTEGRITY)) &&
-                        dd->hfi1_snoop.mode_flag != HFI1_PORT_SNOOP_MODE;
-
-               set_pio_integrity(sc);
-
-               if (enable) /* take HFI_CAP_* flags into account */
-                       hfi1_init_ctxt(sc);
-       }
-       spin_unlock_irqrestore(&dd->sc_lock, sc_flags);
-}
-
-static int hfi1_snoop_open(struct inode *in, struct file *fp)
-{
-       int ret;
-       int mode_flag = 0;
-       unsigned long flags = 0;
-       struct hfi1_devdata *dd;
-       struct list_head *queue;
-
-       mutex_lock(&hfi1_mutex);
-
-       dd = hfi1_dd_from_sc_inode(in);
-       if (!dd) {
-               ret = -ENODEV;
-               goto bail;
-       }
-
-       /*
-        * File mode determines snoop or capture. Some existing user
-        * applications expect the capture device to be able to be opened RDWR
-        * because they expect a dedicated capture device. For this reason we
-        * support a module param to force capture mode even if the file open
-        * mode matches snoop.
-        */
-       if ((fp->f_flags & O_ACCMODE) == O_RDONLY) {
-               snoop_dbg("Capture Enabled");
-               mode_flag = HFI1_PORT_CAPTURE_MODE;
-       } else if ((fp->f_flags & O_ACCMODE) == O_RDWR) {
-               snoop_dbg("Snoop Enabled");
-               mode_flag = HFI1_PORT_SNOOP_MODE;
-       } else {
-               snoop_dbg("Invalid");
-               ret =  -EINVAL;
-               goto bail;
-       }
-       queue = &dd->hfi1_snoop.queue;
-
-       /*
-        * We are not supporting snoop and capture at the same time.
-        */
-       spin_lock_irqsave(&dd->hfi1_snoop.snoop_lock, flags);
-       if (dd->hfi1_snoop.mode_flag) {
-               ret = -EBUSY;
-               spin_unlock_irqrestore(&dd->hfi1_snoop.snoop_lock, flags);
-               goto bail;
-       }
-
-       dd->hfi1_snoop.mode_flag = mode_flag;
-       drain_snoop_list(queue);
-
-       dd->hfi1_snoop.filter_callback = NULL;
-       dd->hfi1_snoop.filter_value = NULL;
-
-       /*
-        * Send side packet integrity checks are not helpful when snooping so
-        * disable and re-enable when we stop snooping.
-        */
-       if (mode_flag == HFI1_PORT_SNOOP_MODE) {
-               /* clear after snoop mode is on */
-               adjust_integrity_checks(dd); /* clear */
-
-               /*
-                * We also do not want to be doing the DLID LMC check for
-                * ingressed packets.
-                */
-               dd->hfi1_snoop.dcc_cfg = read_csr(dd, DCC_CFG_PORT_CONFIG1);
-               write_csr(dd, DCC_CFG_PORT_CONFIG1,
-                         (dd->hfi1_snoop.dcc_cfg >> 32) << 32);
-       }
-
-       /*
-        * As soon as we set these function pointers the recv and send handlers
-        * are active. This is a race condition so we must make sure to drain
-        * the queue and init filter values above. Technically we should add
-        * locking here but all that will happen is on recv a packet will get
-        * allocated and get stuck on the snoop_lock before getting added to the
-        * queue. Same goes for send.
-        */
-       dd->rhf_rcv_function_map = snoop_rhf_rcv_functions;
-       dd->process_pio_send = snoop_send_pio_handler;
-       dd->process_dma_send = snoop_send_pio_handler;
-       dd->pio_inline_send = snoop_inline_pio_send;
-
-       spin_unlock_irqrestore(&dd->hfi1_snoop.snoop_lock, flags);
-       ret = 0;
-
-bail:
-       mutex_unlock(&hfi1_mutex);
-
-       return ret;
-}
-
-static int hfi1_snoop_release(struct inode *in, struct file *fp)
-{
-       unsigned long flags = 0;
-       struct hfi1_devdata *dd;
-       int mode_flag;
-
-       dd = hfi1_dd_from_sc_inode(in);
-       if (!dd)
-               return -ENODEV;
-
-       spin_lock_irqsave(&dd->hfi1_snoop.snoop_lock, flags);
-
-       /* clear the snoop mode before re-adjusting send context CSRs */
-       mode_flag = dd->hfi1_snoop.mode_flag;
-       dd->hfi1_snoop.mode_flag = 0;
-
-       /*
-        * Drain the queue and clear the filters we are done with it. Don't
-        * forget to restore the packet integrity checks
-        */
-       drain_snoop_list(&dd->hfi1_snoop.queue);
-       if (mode_flag == HFI1_PORT_SNOOP_MODE) {
-               /* restore after snoop mode is clear */
-               adjust_integrity_checks(dd); /* restore */
-
-               /*
-                * Also should probably reset the DCC_CONFIG1 register for DLID
-                * checking on incoming packets again. Use the value saved when
-                * opening the snoop device.
-                */
-               write_csr(dd, DCC_CFG_PORT_CONFIG1, dd->hfi1_snoop.dcc_cfg);
-       }
-
-       dd->hfi1_snoop.filter_callback = NULL;
-       kfree(dd->hfi1_snoop.filter_value);
-       dd->hfi1_snoop.filter_value = NULL;
-
-       /*
-        * User is done snooping and capturing, return control to the normal
-        * handler. Re-enable SDMA handling.
-        */
-       dd->rhf_rcv_function_map = dd->normal_rhf_rcv_functions;
-       dd->process_pio_send = hfi1_verbs_send_pio;
-       dd->process_dma_send = hfi1_verbs_send_dma;
-       dd->pio_inline_send = pio_copy;
-
-       spin_unlock_irqrestore(&dd->hfi1_snoop.snoop_lock, flags);
-
-       snoop_dbg("snoop/capture device released");
-
-       return 0;
-}
-
-static unsigned int hfi1_snoop_poll(struct file *fp,
-                                   struct poll_table_struct *wait)
-{
-       int ret = 0;
-       unsigned long flags = 0;
-
-       struct hfi1_devdata *dd;
-
-       dd = hfi1_dd_from_sc_inode(fp->f_inode);
-       if (!dd)
-               return -ENODEV;
-
-       spin_lock_irqsave(&dd->hfi1_snoop.snoop_lock, flags);
-
-       poll_wait(fp, &dd->hfi1_snoop.waitq, wait);
-       if (!list_empty(&dd->hfi1_snoop.queue))
-               ret |= POLLIN | POLLRDNORM;
-
-       spin_unlock_irqrestore(&dd->hfi1_snoop.snoop_lock, flags);
-       return ret;
-}
-
-static ssize_t hfi1_snoop_write(struct file *fp, const char __user *data,
-                               size_t count, loff_t *off)
-{
-       struct diag_pkt dpkt;
-       struct hfi1_devdata *dd;
-       size_t ret;
-       u8 byte_two, sl, sc5, sc4, vl, byte_one;
-       struct send_context *sc;
-       u32 len;
-       u64 pbc;
-       struct hfi1_ibport *ibp;
-       struct hfi1_pportdata *ppd;
-
-       dd = hfi1_dd_from_sc_inode(fp->f_inode);
-       if (!dd)
-               return -ENODEV;
-
-       ppd = dd->pport;
-       snoop_dbg("received %lu bytes from user", count);
-
-       memset(&dpkt, 0, sizeof(struct diag_pkt));
-       dpkt.version = _DIAG_PKT_VERS;
-       dpkt.unit = dd->unit;
-       dpkt.port = 1;
-
-       if (likely(!(snoop_flags & SNOOP_USE_METADATA))) {
-               /*
-               * We need to generate the PBC and not let diagpkt_send do it,
-               * to do this we need the VL and the length in dwords.
-               * The VL can be determined by using the SL and looking up the
-               * SC. Then the SC can be converted into VL. The exception to
-               * this is those packets which are from an SMI queue pair.
-               * Since we can't detect anything about the QP here we have to
-               * rely on the SC. If its 0xF then we assume its SMI and
-               * do not look at the SL.
-               */
-               if (copy_from_user(&byte_one, data, 1))
-                       return -EINVAL;
-
-               if (copy_from_user(&byte_two, data + 1, 1))
-                       return -EINVAL;
-
-               sc4 = (byte_one >> 4) & 0xf;
-               if (sc4 == 0xF) {
-                       snoop_dbg("Detected VL15 packet ignoring SL in packet");
-                       vl = sc4;
-               } else {
-                       sl = (byte_two >> 4) & 0xf;
-                       ibp = to_iport(&dd->verbs_dev.rdi.ibdev, 1);
-                       sc5 = ibp->sl_to_sc[sl];
-                       vl = sc_to_vlt(dd, sc5);
-                       if (vl != sc4) {
-                               snoop_dbg("VL %d does not match SC %d of packet",
-                                         vl, sc4);
-                               return -EINVAL;
-                       }
-               }
-
-               sc = dd->vld[vl].sc; /* Look up the context based on VL */
-               if (sc) {
-                       dpkt.sw_index = sc->sw_index;
-                       snoop_dbg("Sending on context %u(%u)", sc->sw_index,
-                                 sc->hw_context);
-               } else {
-                       snoop_dbg("Could not find context for vl %d", vl);
-                       return -EINVAL;
-               }
-
-               len = (count >> 2) + 2; /* Add in PBC */
-               pbc = create_pbc(ppd, 0, 0, vl, len);
-       } else {
-               if (copy_from_user(&pbc, data, sizeof(pbc)))
-                       return -EINVAL;
-               vl = (pbc >> PBC_VL_SHIFT) & PBC_VL_MASK;
-               sc = dd->vld[vl].sc; /* Look up the context based on VL */
-               if (sc) {
-                       dpkt.sw_index = sc->sw_index;
-               } else {
-                       snoop_dbg("Could not find context for vl %d", vl);
-                       return -EINVAL;
-               }
-               data += sizeof(pbc);
-               count -= sizeof(pbc);
-       }
-       dpkt.len = count;
-       dpkt.data = (unsigned long)data;
-
-       snoop_dbg("PBC: vl=0x%llx Length=0x%llx",
-                 (pbc >> 12) & 0xf,
-                 (pbc & 0xfff));
-
-       dpkt.pbc = pbc;
-       ret = diagpkt_send(&dpkt);
-       /*
-        * diagpkt_send only returns number of bytes in the diagpkt so patch
-        * that up here before returning.
-        */
-       if (ret == sizeof(dpkt))
-               return count;
-
-       return ret;
-}
-
-static ssize_t hfi1_snoop_read(struct file *fp, char __user *data,
-                              size_t pkt_len, loff_t *off)
-{
-       ssize_t ret = 0;
-       unsigned long flags = 0;
-       struct snoop_packet *packet = NULL;
-       struct hfi1_devdata *dd;
-
-       dd = hfi1_dd_from_sc_inode(fp->f_inode);
-       if (!dd)
-               return -ENODEV;
-
-       spin_lock_irqsave(&dd->hfi1_snoop.snoop_lock, flags);
-
-       while (list_empty(&dd->hfi1_snoop.queue)) {
-               spin_unlock_irqrestore(&dd->hfi1_snoop.snoop_lock, flags);
-
-               if (fp->f_flags & O_NONBLOCK)
-                       return -EAGAIN;
-
-               if (wait_event_interruptible(
-                               dd->hfi1_snoop.waitq,
-                               !list_empty(&dd->hfi1_snoop.queue)))
-                       return -EINTR;
-
-               spin_lock_irqsave(&dd->hfi1_snoop.snoop_lock, flags);
-       }
-
-       if (!list_empty(&dd->hfi1_snoop.queue)) {
-               packet = list_entry(dd->hfi1_snoop.queue.next,
-                                   struct snoop_packet, list);
-               list_del(&packet->list);
-               spin_unlock_irqrestore(&dd->hfi1_snoop.snoop_lock, flags);
-               if (pkt_len >= packet->total_len) {
-                       if (copy_to_user(data, packet->data,
-                                        packet->total_len))
-                               ret = -EFAULT;
-                       else
-                               ret = packet->total_len;
-               } else {
-                       ret = -EINVAL;
-               }
-
-               kfree(packet);
-       } else {
-               spin_unlock_irqrestore(&dd->hfi1_snoop.snoop_lock, flags);
-       }
-
-       return ret;
-}
-
-/**
- * hfi1_assign_snoop_link_credits -- Set up credits for VL15 and others
- * @ppd : ptr to hfi1 port data
- * @value : options from user space
- *
- * Assumes the rest of the CM credit registers are zero from a
- * previous global or credit reset.
- * Leave shared count at zero for both global and all vls.
- * In snoop mode ideally we don't use shared credits
- * Reserve 8.5k for VL15
- * If total credits less than 8.5kbytes return error.
- * Divide the rest of the credits across VL0 to VL7 and if
- * each of these levels has less than 34 credits (at least 2048 + 128 bytes)
- * return with an error.
- * The credit registers will be reset to zero on link negotiation or link up
- * so this function should be activated from user space only if the port has
- * gone past link negotiation and link up.
- *
- * Return -- 0 if successful else error condition
- *
- */
-static long hfi1_assign_snoop_link_credits(struct hfi1_pportdata *ppd,
-                                          int value)
-{
-#define  OPA_MIN_PER_VL_CREDITS  34  /* 2048 + 128 bytes */
-       struct buffer_control t;
-       int i;
-       struct hfi1_devdata *dd = ppd->dd;
-       u16  total_credits = (value >> 16) & 0xffff;
-       u16  vl15_credits = dd->vl15_init / 2;
-       u16  per_vl_credits;
-       __be16 be_per_vl_credits;
-
-       if (ppd->host_link_state & HLS_DOWN)
-               goto err_exit;
-       if (total_credits  <  vl15_credits)
-               goto err_exit;
-
-       per_vl_credits = (total_credits - vl15_credits) / TXE_NUM_DATA_VL;
-
-       if (per_vl_credits < OPA_MIN_PER_VL_CREDITS)
-               goto err_exit;
-
-       memset(&t, 0, sizeof(t));
-       be_per_vl_credits = cpu_to_be16(per_vl_credits);
-
-       for (i = 0; i < TXE_NUM_DATA_VL; i++)
-               t.vl[i].dedicated = be_per_vl_credits;
-
-       t.vl[15].dedicated  = cpu_to_be16(vl15_credits);
-       return set_buffer_control(ppd, &t);
-
-err_exit:
-       snoop_dbg("port_state = 0x%x, total_credits = %d, vl15_credits = %d",
-                 ppd->host_link_state, total_credits, vl15_credits);
-
-       return -EINVAL;
-}
-
-static long hfi1_ioctl(struct file *fp, unsigned int cmd, unsigned long arg)
-{
-       struct hfi1_devdata *dd;
-       void *filter_value = NULL;
-       long ret = 0;
-       int value = 0;
-       u8 phys_state = 0;
-       u8 link_state = 0;
-       u16 dev_state = 0;
-       unsigned long flags = 0;
-       unsigned long *argp = NULL;
-       struct hfi1_packet_filter_command filter_cmd = {0};
-       int mode_flag = 0;
-       struct hfi1_pportdata *ppd = NULL;
-       unsigned int index;
-       struct hfi1_link_info link_info;
-       int read_cmd, write_cmd, read_ok, write_ok;
-
-       dd = hfi1_dd_from_sc_inode(fp->f_inode);
-       if (!dd)
-               return -ENODEV;
-
-       mode_flag = dd->hfi1_snoop.mode_flag;
-       read_cmd = _IOC_DIR(cmd) & _IOC_READ;
-       write_cmd = _IOC_DIR(cmd) & _IOC_WRITE;
-       write_ok = access_ok(VERIFY_WRITE, (void __user *)arg, _IOC_SIZE(cmd));
-       read_ok = access_ok(VERIFY_READ, (void __user *)arg, _IOC_SIZE(cmd));
-
-       if ((read_cmd && !write_ok) || (write_cmd && !read_ok))
-               return -EFAULT;
-
-       if (!capable(CAP_SYS_ADMIN))
-               return -EPERM;
-
-       if ((mode_flag & HFI1_PORT_CAPTURE_MODE) &&
-           (cmd != HFI1_SNOOP_IOCCLEARQUEUE) &&
-           (cmd != HFI1_SNOOP_IOCCLEARFILTER) &&
-           (cmd != HFI1_SNOOP_IOCSETFILTER))
-               /* Capture devices are allowed only 3 operations
-                * 1.Clear capture queue
-                * 2.Clear capture filter
-                * 3.Set capture filter
-                * Other are invalid.
-                */
-               return -EINVAL;
-
-       switch (cmd) {
-       case HFI1_SNOOP_IOCSETLINKSTATE_EXTRA:
-               memset(&link_info, 0, sizeof(link_info));
-
-               if (copy_from_user(&link_info,
-                                  (struct hfi1_link_info __user *)arg,
-                                  sizeof(link_info)))
-                       return -EFAULT;
-
-               value = link_info.port_state;
-               index = link_info.port_number;
-               if (index > dd->num_pports - 1)
-                       return -EINVAL;
-
-               ppd = &dd->pport[index];
-               if (!ppd)
-                       return -EINVAL;
-
-               /* What we want to transition to */
-               phys_state = (value >> 4) & 0xF;
-               link_state = value & 0xF;
-               snoop_dbg("Setting link state 0x%x", value);
-
-               switch (link_state) {
-               case IB_PORT_NOP:
-                       if (phys_state == 0)
-                               break;
-                               /* fall through */
-               case IB_PORT_DOWN:
-                       switch (phys_state) {
-                       case 0:
-                               dev_state = HLS_DN_DOWNDEF;
-                               break;
-                       case 2:
-                               dev_state = HLS_DN_POLL;
-                               break;
-                       case 3:
-                               dev_state = HLS_DN_DISABLE;
-                               break;
-                       default:
-                               return -EINVAL;
-                       }
-                       ret = set_link_state(ppd, dev_state);
-                       break;
-               case IB_PORT_ARMED:
-                       ret = set_link_state(ppd, HLS_UP_ARMED);
-                       if (!ret)
-                               send_idle_sma(dd, SMA_IDLE_ARM);
-                       break;
-               case IB_PORT_ACTIVE:
-                       ret = set_link_state(ppd, HLS_UP_ACTIVE);
-                       if (!ret)
-                               send_idle_sma(dd, SMA_IDLE_ACTIVE);
-                       break;
-               default:
-                       return -EINVAL;
-               }
-
-               if (ret)
-                       break;
-               /* fall through */
-       case HFI1_SNOOP_IOCGETLINKSTATE:
-       case HFI1_SNOOP_IOCGETLINKSTATE_EXTRA:
-               if (cmd == HFI1_SNOOP_IOCGETLINKSTATE_EXTRA) {
-                       memset(&link_info, 0, sizeof(link_info));
-                       if (copy_from_user(&link_info,
-                                          (struct hfi1_link_info __user *)arg,
-                                          sizeof(link_info)))
-                               return -EFAULT;
-                       index = link_info.port_number;
-               } else {
-                       ret = __get_user(index, (int __user *)arg);
-                       if (ret !=  0)
-                               break;
-               }
-
-               if (index > dd->num_pports - 1)
-                       return -EINVAL;
-
-               ppd = &dd->pport[index];
-               if (!ppd)
-                       return -EINVAL;
-
-               value = hfi1_ibphys_portstate(ppd);
-               value <<= 4;
-               value |= driver_lstate(ppd);
-
-               snoop_dbg("Link port | Link State: %d", value);
-
-               if ((cmd == HFI1_SNOOP_IOCGETLINKSTATE_EXTRA) ||
-                   (cmd == HFI1_SNOOP_IOCSETLINKSTATE_EXTRA)) {
-                       link_info.port_state = value;
-                       link_info.node_guid = cpu_to_be64(ppd->guid);
-                       link_info.link_speed_active =
-                                               ppd->link_speed_active;
-                       link_info.link_width_active =
-                                               ppd->link_width_active;
-                       if (copy_to_user((struct hfi1_link_info __user *)arg,
-                                        &link_info, sizeof(link_info)))
-                               return -EFAULT;
-               } else {
-                       ret = __put_user(value, (int __user *)arg);
-               }
-               break;
-
-       case HFI1_SNOOP_IOCCLEARQUEUE:
-               snoop_dbg("Clearing snoop queue");
-               spin_lock_irqsave(&dd->hfi1_snoop.snoop_lock, flags);
-               drain_snoop_list(&dd->hfi1_snoop.queue);
-               spin_unlock_irqrestore(&dd->hfi1_snoop.snoop_lock, flags);
-               break;
-
-       case HFI1_SNOOP_IOCCLEARFILTER:
-               snoop_dbg("Clearing filter");
-               spin_lock_irqsave(&dd->hfi1_snoop.snoop_lock, flags);
-               if (dd->hfi1_snoop.filter_callback) {
-                       /* Drain packets first */
-                       drain_snoop_list(&dd->hfi1_snoop.queue);
-                       dd->hfi1_snoop.filter_callback = NULL;
-               }
-               kfree(dd->hfi1_snoop.filter_value);
-               dd->hfi1_snoop.filter_value = NULL;
-               spin_unlock_irqrestore(&dd->hfi1_snoop.snoop_lock, flags);
-               break;
-
-       case HFI1_SNOOP_IOCSETFILTER:
-               snoop_dbg("Setting filter");
-               /* just copy command structure */
-               argp = (unsigned long *)arg;
-               if (copy_from_user(&filter_cmd, (void __user *)argp,
-                                  sizeof(filter_cmd)))
-                       return -EFAULT;
-
-               if (filter_cmd.opcode >= HFI1_MAX_FILTERS) {
-                       pr_alert("Invalid opcode in request\n");
-                       return -EINVAL;
-               }
-
-               snoop_dbg("Opcode %d Len %d Ptr %p",
-                         filter_cmd.opcode, filter_cmd.length,
-                         filter_cmd.value_ptr);
-
-               filter_value = kcalloc(filter_cmd.length, sizeof(u8),
-                                      GFP_KERNEL);
-               if (!filter_value)
-                       return -ENOMEM;
-
-               /* copy remaining data from userspace */
-               if (copy_from_user((u8 *)filter_value,
-                                  (void __user *)filter_cmd.value_ptr,
-                                  filter_cmd.length)) {
-                       kfree(filter_value);
-                       return -EFAULT;
-               }
-               /* Drain packets first */
-               spin_lock_irqsave(&dd->hfi1_snoop.snoop_lock, flags);
-               drain_snoop_list(&dd->hfi1_snoop.queue);
-               dd->hfi1_snoop.filter_callback =
-                       hfi1_filters[filter_cmd.opcode].filter;
-               /* just in case we see back to back sets */
-               kfree(dd->hfi1_snoop.filter_value);
-               dd->hfi1_snoop.filter_value = filter_value;
-               spin_unlock_irqrestore(&dd->hfi1_snoop.snoop_lock, flags);
-               break;
-       case HFI1_SNOOP_IOCGETVERSION:
-               value = SNOOP_CAPTURE_VERSION;
-               snoop_dbg("Getting version: %d", value);
-               ret = __put_user(value, (int __user *)arg);
-               break;
-       case HFI1_SNOOP_IOCSET_OPTS:
-               snoop_flags = 0;
-               ret = __get_user(value, (int __user *)arg);
-               if (ret != 0)
-                       break;
-
-               snoop_dbg("Setting snoop option %d", value);
-               if (value & SNOOP_DROP_SEND)
-                       snoop_flags |= SNOOP_DROP_SEND;
-               if (value & SNOOP_USE_METADATA)
-                       snoop_flags |= SNOOP_USE_METADATA;
-               if (value & (SNOOP_SET_VL0TOVL15)) {
-                       ppd = &dd->pport[0];  /* first port will do */
-                       ret = hfi1_assign_snoop_link_credits(ppd, value);
-               }
-               break;
-       default:
-               return -ENOTTY;
-       }
-
-       return ret;
-}
-
-static void snoop_list_add_tail(struct snoop_packet *packet,
-                               struct hfi1_devdata *dd)
-{
-       unsigned long flags = 0;
-
-       spin_lock_irqsave(&dd->hfi1_snoop.snoop_lock, flags);
-       if (likely((dd->hfi1_snoop.mode_flag & HFI1_PORT_SNOOP_MODE) ||
-                  (dd->hfi1_snoop.mode_flag & HFI1_PORT_CAPTURE_MODE))) {
-               list_add_tail(&packet->list, &dd->hfi1_snoop.queue);
-               snoop_dbg("Added packet to list");
-       }
-
-       /*
-        * Technically we can could have closed the snoop device while waiting
-        * on the above lock and it is gone now. The snoop mode_flag will
-        * prevent us from adding the packet to the queue though.
-        */
-
-       spin_unlock_irqrestore(&dd->hfi1_snoop.snoop_lock, flags);
-       wake_up_interruptible(&dd->hfi1_snoop.waitq);
-}
-
-static inline int hfi1_filter_check(void *val, const char *msg)
-{
-       if (!val) {
-               snoop_dbg("Error invalid %s value for filter", msg);
-               return HFI1_FILTER_ERR;
-       }
-       return 0;
-}
-
-static int hfi1_filter_lid(void *ibhdr, void *packet_data, void *value)
-{
-       struct hfi1_ib_header *hdr;
-       int ret;
-
-       ret = hfi1_filter_check(ibhdr, "header");
-       if (ret)
-               return ret;
-       ret = hfi1_filter_check(value, "user");
-       if (ret)
-               return ret;
-       hdr = (struct hfi1_ib_header *)ibhdr;
-
-       if (*((u16 *)value) == be16_to_cpu(hdr->lrh[3])) /* matches slid */
-               return HFI1_FILTER_HIT; /* matched */
-
-       return HFI1_FILTER_MISS; /* Not matched */
-}
-
-static int hfi1_filter_dlid(void *ibhdr, void *packet_data, void *value)
-{
-       struct hfi1_ib_header *hdr;
-       int ret;
-
-       ret = hfi1_filter_check(ibhdr, "header");
-       if (ret)
-               return ret;
-       ret = hfi1_filter_check(value, "user");
-       if (ret)
-               return ret;
-
-       hdr = (struct hfi1_ib_header *)ibhdr;
-
-       if (*((u16 *)value) == be16_to_cpu(hdr->lrh[1]))
-               return HFI1_FILTER_HIT;
-
-       return HFI1_FILTER_MISS;
-}
-
-/* Not valid for outgoing packets, send handler passes null for data*/
-static int hfi1_filter_mad_mgmt_class(void *ibhdr, void *packet_data,
-                                     void *value)
-{
-       struct hfi1_ib_header *hdr;
-       struct hfi1_other_headers *ohdr = NULL;
-       struct ib_smp *smp = NULL;
-       u32 qpn = 0;
-       int ret;
-
-       ret = hfi1_filter_check(ibhdr, "header");
-       if (ret)
-               return ret;
-       ret = hfi1_filter_check(packet_data, "packet_data");
-       if (ret)
-               return ret;
-       ret = hfi1_filter_check(value, "user");
-       if (ret)
-               return ret;
-
-       hdr = (struct hfi1_ib_header *)ibhdr;
-
-       /* Check for GRH */
-       if ((be16_to_cpu(hdr->lrh[0]) & 3) == HFI1_LRH_BTH)
-               ohdr = &hdr->u.oth; /* LRH + BTH + DETH */
-       else
-               ohdr = &hdr->u.l.oth; /* LRH + GRH + BTH + DETH */
-
-       qpn = be32_to_cpu(ohdr->bth[1]) & 0x00FFFFFF;
-       if (qpn <= 1) {
-               smp = (struct ib_smp *)packet_data;
-               if (*((u8 *)value) == smp->mgmt_class)
-                       return HFI1_FILTER_HIT;
-               else
-                       return HFI1_FILTER_MISS;
-       }
-       return HFI1_FILTER_ERR;
-}
-
-static int hfi1_filter_qp_number(void *ibhdr, void *packet_data, void *value)
-{
-       struct hfi1_ib_header *hdr;
-       struct hfi1_other_headers *ohdr = NULL;
-       int ret;
-
-       ret = hfi1_filter_check(ibhdr, "header");
-       if (ret)
-               return ret;
-       ret = hfi1_filter_check(value, "user");
-       if (ret)
-               return ret;
-
-       hdr = (struct hfi1_ib_header *)ibhdr;
-
-       /* Check for GRH */
-       if ((be16_to_cpu(hdr->lrh[0]) & 3) == HFI1_LRH_BTH)
-               ohdr = &hdr->u.oth; /* LRH + BTH + DETH */
-       else
-               ohdr = &hdr->u.l.oth; /* LRH + GRH + BTH + DETH */
-       if (*((u32 *)value) == (be32_to_cpu(ohdr->bth[1]) & 0x00FFFFFF))
-               return HFI1_FILTER_HIT;
-
-       return HFI1_FILTER_MISS;
-}
-
-static int hfi1_filter_ibpacket_type(void *ibhdr, void *packet_data,
-                                    void *value)
-{
-       u32 lnh = 0;
-       u8 opcode = 0;
-       struct hfi1_ib_header *hdr;
-       struct hfi1_other_headers *ohdr = NULL;
-       int ret;
-
-       ret = hfi1_filter_check(ibhdr, "header");
-       if (ret)
-               return ret;
-       ret = hfi1_filter_check(value, "user");
-       if (ret)
-               return ret;
-
-       hdr = (struct hfi1_ib_header *)ibhdr;
-
-       lnh = (be16_to_cpu(hdr->lrh[0]) & 3);
-
-       if (lnh == HFI1_LRH_BTH)
-               ohdr = &hdr->u.oth;
-       else if (lnh == HFI1_LRH_GRH)
-               ohdr = &hdr->u.l.oth;
-       else
-               return HFI1_FILTER_ERR;
-
-       opcode = be32_to_cpu(ohdr->bth[0]) >> 24;
-
-       if (*((u8 *)value) == ((opcode >> 5) & 0x7))
-               return HFI1_FILTER_HIT;
-
-       return HFI1_FILTER_MISS;
-}
-
-static int hfi1_filter_ib_service_level(void *ibhdr, void *packet_data,
-                                       void *value)
-{
-       struct hfi1_ib_header *hdr;
-       int ret;
-
-       ret = hfi1_filter_check(ibhdr, "header");
-       if (ret)
-               return ret;
-       ret = hfi1_filter_check(value, "user");
-       if (ret)
-               return ret;
-
-       hdr = (struct hfi1_ib_header *)ibhdr;
-
-       if ((*((u8 *)value)) == ((be16_to_cpu(hdr->lrh[0]) >> 4) & 0xF))
-               return HFI1_FILTER_HIT;
-
-       return HFI1_FILTER_MISS;
-}
-
-static int hfi1_filter_ib_pkey(void *ibhdr, void *packet_data, void *value)
-{
-       u32 lnh = 0;
-       struct hfi1_ib_header *hdr;
-       struct hfi1_other_headers *ohdr = NULL;
-       int ret;
-
-       ret = hfi1_filter_check(ibhdr, "header");
-       if (ret)
-               return ret;
-       ret = hfi1_filter_check(value, "user");
-       if (ret)
-               return ret;
-
-       hdr = (struct hfi1_ib_header *)ibhdr;
-
-       lnh = (be16_to_cpu(hdr->lrh[0]) & 3);
-       if (lnh == HFI1_LRH_BTH)
-               ohdr = &hdr->u.oth;
-       else if (lnh == HFI1_LRH_GRH)
-               ohdr = &hdr->u.l.oth;
-       else
-               return HFI1_FILTER_ERR;
-
-       /* P_key is 16-bit entity, however top most bit indicates
-        * type of membership. 0 for limited and 1 for Full.
-        * Limited members cannot accept information from other
-        * Limited members, but communication is allowed between
-        * every other combination of membership.
-        * Hence we'll omit comparing top-most bit while filtering
-        */
-
-       if ((*(u16 *)value & 0x7FFF) ==
-               ((be32_to_cpu(ohdr->bth[0])) & 0x7FFF))
-               return HFI1_FILTER_HIT;
-
-       return HFI1_FILTER_MISS;
-}
-
-/*
- * If packet_data is NULL then this is coming from one of the send functions.
- * Thus we know if its an ingressed or egressed packet.
- */
-static int hfi1_filter_direction(void *ibhdr, void *packet_data, void *value)
-{
-       u8 user_dir = *(u8 *)value;
-       int ret;
-
-       ret = hfi1_filter_check(value, "user");
-       if (ret)
-               return ret;
-
-       if (packet_data) {
-               /* Incoming packet */
-               if (user_dir & HFI1_SNOOP_INGRESS)
-                       return HFI1_FILTER_HIT;
-       } else {
-               /* Outgoing packet */
-               if (user_dir & HFI1_SNOOP_EGRESS)
-                       return HFI1_FILTER_HIT;
-       }
-
-       return HFI1_FILTER_MISS;
-}
-
-/*
- * Allocate a snoop packet. The structure that is stored in the ring buffer, not
- * to be confused with an hfi packet type.
- */
-static struct snoop_packet *allocate_snoop_packet(u32 hdr_len,
-                                                 u32 data_len,
-                                                 u32 md_len)
-{
-       struct snoop_packet *packet;
-
-       packet = kzalloc(sizeof(*packet) + hdr_len + data_len
-                        + md_len,
-                        GFP_ATOMIC | __GFP_NOWARN);
-       if (likely(packet))
-               INIT_LIST_HEAD(&packet->list);
-
-       return packet;
-}
-
-/*
- * Instead of having snoop and capture code intermixed with the recv functions,
- * both the interrupt handler and hfi1_ib_rcv() we are going to hijack the call
- * and land in here for snoop/capture but if not enabled the call will go
- * through as before. This gives us a single point to constrain all of the snoop
- * snoop recv logic. There is nothing special that needs to happen for bypass
- * packets. This routine should not try to look into the packet. It just copied
- * it. There is no guarantee for filters when it comes to bypass packets as
- * there is no specific support. Bottom line is this routine does now even know
- * what a bypass packet is.
- */
-int snoop_recv_handler(struct hfi1_packet *packet)
-{
-       struct hfi1_pportdata *ppd = packet->rcd->ppd;
-       struct hfi1_ib_header *hdr = packet->hdr;
-       int header_size = packet->hlen;
-       void *data = packet->ebuf;
-       u32 tlen = packet->tlen;
-       struct snoop_packet *s_packet = NULL;
-       int ret;
-       int snoop_mode = 0;
-       u32 md_len = 0;
-       struct capture_md md;
-
-       snoop_dbg("PACKET IN: hdr size %d tlen %d data %p", header_size, tlen,
-                 data);
-
-       trace_snoop_capture(ppd->dd, header_size, hdr, tlen - header_size,
-                           data);
-
-       if (!ppd->dd->hfi1_snoop.filter_callback) {
-               snoop_dbg("filter not set");
-               ret = HFI1_FILTER_HIT;
-       } else {
-               ret = ppd->dd->hfi1_snoop.filter_callback(hdr, data,
-                                       ppd->dd->hfi1_snoop.filter_value);
-       }
-
-       switch (ret) {
-       case HFI1_FILTER_ERR:
-               snoop_dbg("Error in filter call");
-               break;
-       case HFI1_FILTER_MISS:
-               snoop_dbg("Filter Miss");
-               break;
-       case HFI1_FILTER_HIT:
-
-               if (ppd->dd->hfi1_snoop.mode_flag & HFI1_PORT_SNOOP_MODE)
-                       snoop_mode = 1;
-               if ((snoop_mode == 0) ||
-                   unlikely(snoop_flags & SNOOP_USE_METADATA))
-                       md_len = sizeof(struct capture_md);
-
-               s_packet = allocate_snoop_packet(header_size,
-                                                tlen - header_size,
-                                                md_len);
-
-               if (unlikely(!s_packet)) {
-                       dd_dev_warn_ratelimited(ppd->dd, "Unable to allocate snoop/capture packet\n");
-                       break;
-               }
-
-               if (md_len > 0) {
-                       memset(&md, 0, sizeof(struct capture_md));
-                       md.port = 1;
-                       md.dir = PKT_DIR_INGRESS;
-                       md.u.rhf = packet->rhf;
-                       memcpy(s_packet->data, &md, md_len);
-               }
-
-               /* We should always have a header */
-               if (hdr) {
-                       memcpy(s_packet->data + md_len, hdr, header_size);
-               } else {
-                       dd_dev_err(ppd->dd, "Unable to copy header to snoop/capture packet\n");
-                       kfree(s_packet);
-                       break;
-               }
-
-               /*
-                * Packets with no data are possible. If there is no data needed
-                * to take care of the last 4 bytes which are normally included
-                * with data buffers and are included in tlen.  Since we kzalloc
-                * the buffer we do not need to set any values but if we decide
-                * not to use kzalloc we should zero them.
-                */
-               if (data)
-                       memcpy(s_packet->data + header_size + md_len, data,
-                              tlen - header_size);
-
-               s_packet->total_len = tlen + md_len;
-               snoop_list_add_tail(s_packet, ppd->dd);
-
-               /*
-                * If we are snooping the packet not capturing then throw away
-                * after adding to the list.
-                */
-               snoop_dbg("Capturing packet");
-               if (ppd->dd->hfi1_snoop.mode_flag & HFI1_PORT_SNOOP_MODE) {
-                       snoop_dbg("Throwing packet away");
-                       /*
-                        * If we are dropping the packet we still may need to
-                        * handle the case where error flags are set, this is
-                        * normally done by the type specific handler but that
-                        * won't be called in this case.
-                        */
-                       if (unlikely(rhf_err_flags(packet->rhf)))
-                               handle_eflags(packet);
-
-                       /* throw the packet on the floor */
-                       return RHF_RCV_CONTINUE;
-               }
-               break;
-       default:
-               break;
-       }
-
-       /*
-        * We do not care what type of packet came in here - just pass it off
-        * to the normal handler.
-        */
-       return ppd->dd->normal_rhf_rcv_functions[rhf_rcv_type(packet->rhf)]
-                       (packet);
-}
-
-/*
- * Handle snooping and capturing packets when sdma is being used.
- */
-int snoop_send_dma_handler(struct rvt_qp *qp, struct hfi1_pkt_state *ps,
-                          u64 pbc)
-{
-       pr_alert("Snooping/Capture of Send DMA Packets Is Not Supported!\n");
-       snoop_dbg("Unsupported Operation");
-       return hfi1_verbs_send_dma(qp, ps, 0);
-}
-
-/*
- * Handle snooping and capturing packets when pio is being used. Does not handle
- * bypass packets. The only way to send a bypass packet currently is to use the
- * diagpkt interface. When that interface is enable snoop/capture is not.
- */
-int snoop_send_pio_handler(struct rvt_qp *qp, struct hfi1_pkt_state *ps,
-                          u64 pbc)
-{
-       u32 hdrwords = qp->s_hdrwords;
-       struct rvt_sge_state *ss = qp->s_cur_sge;
-       u32 len = qp->s_cur_size;
-       u32 dwords = (len + 3) >> 2;
-       u32 plen = hdrwords + dwords + 2; /* includes pbc */
-       struct hfi1_pportdata *ppd = ps->ppd;
-       struct snoop_packet *s_packet = NULL;
-       u32 *hdr = (u32 *)&ps->s_txreq->phdr.hdr;
-       u32 length = 0;
-       struct rvt_sge_state temp_ss;
-       void *data = NULL;
-       void *data_start = NULL;
-       int ret;
-       int snoop_mode = 0;
-       int md_len = 0;
-       struct capture_md md;
-       u32 vl;
-       u32 hdr_len = hdrwords << 2;
-       u32 tlen = HFI1_GET_PKT_LEN(&ps->s_txreq->phdr.hdr);
-
-       md.u.pbc = 0;
-
-       snoop_dbg("PACKET OUT: hdrword %u len %u plen %u dwords %u tlen %u",
-                 hdrwords, len, plen, dwords, tlen);
-       if (ppd->dd->hfi1_snoop.mode_flag & HFI1_PORT_SNOOP_MODE)
-               snoop_mode = 1;
-       if ((snoop_mode == 0) ||
-           unlikely(snoop_flags & SNOOP_USE_METADATA))
-               md_len = sizeof(struct capture_md);
-
-       /* not using ss->total_len as arg 2 b/c that does not count CRC */
-       s_packet = allocate_snoop_packet(hdr_len, tlen - hdr_len, md_len);
-
-       if (unlikely(!s_packet)) {
-               dd_dev_warn_ratelimited(ppd->dd, "Unable to allocate snoop/capture packet\n");
-               goto out;
-       }
-
-       s_packet->total_len = tlen + md_len;
-
-       if (md_len > 0) {
-               memset(&md, 0, sizeof(struct capture_md));
-               md.port = 1;
-               md.dir = PKT_DIR_EGRESS;
-               if (likely(pbc == 0)) {
-                       vl = be16_to_cpu(ps->s_txreq->phdr.hdr.lrh[0]) >> 12;
-                       md.u.pbc = create_pbc(ppd, 0, qp->s_srate, vl, plen);
-               } else {
-                       md.u.pbc = 0;
-               }
-               memcpy(s_packet->data, &md, md_len);
-       } else {
-               md.u.pbc = pbc;
-       }
-
-       /* Copy header */
-       if (likely(hdr)) {
-               memcpy(s_packet->data + md_len, hdr, hdr_len);
-       } else {
-               dd_dev_err(ppd->dd,
-                          "Unable to copy header to snoop/capture packet\n");
-               kfree(s_packet);
-               goto out;
-       }
-
-       if (ss) {
-               data = s_packet->data + hdr_len + md_len;
-               data_start = data;
-
-               /*
-                * Copy SGE State
-                * The update_sge() function below will not modify the
-                * individual SGEs in the array. It will make a copy each time
-                * and operate on that. So we only need to copy this instance
-                * and it won't impact PIO.
-                */
-               temp_ss = *ss;
-               length = len;
-
-               snoop_dbg("Need to copy %d bytes", length);
-               while (length) {
-                       void *addr = temp_ss.sge.vaddr;
-                       u32 slen = temp_ss.sge.length;
-
-                       if (slen > length) {
-                               slen = length;
-                               snoop_dbg("slen %d > len %d", slen, length);
-                       }
-                       snoop_dbg("copy %d to %p", slen, addr);
-                       memcpy(data, addr, slen);
-                       update_sge(&temp_ss, slen);
-                       length -= slen;
-                       data += slen;
-                       snoop_dbg("data is now %p bytes left %d", data, length);
-               }
-               snoop_dbg("Completed SGE copy");
-       }
-
-       /*
-        * Why do the filter check down here? Because the event tracing has its
-        * own filtering and we need to have the walked the SGE list.
-        */
-       if (!ppd->dd->hfi1_snoop.filter_callback) {
-               snoop_dbg("filter not set\n");
-               ret = HFI1_FILTER_HIT;
-       } else {
-               ret = ppd->dd->hfi1_snoop.filter_callback(
-                                       &ps->s_txreq->phdr.hdr,
-                                       NULL,
-                                       ppd->dd->hfi1_snoop.filter_value);
-       }
-
-       switch (ret) {
-       case HFI1_FILTER_ERR:
-               snoop_dbg("Error in filter call");
-               /* fall through */
-       case HFI1_FILTER_MISS:
-               snoop_dbg("Filter Miss");
-               kfree(s_packet);
-               break;
-       case HFI1_FILTER_HIT:
-               snoop_dbg("Capturing packet");
-               snoop_list_add_tail(s_packet, ppd->dd);
-
-               if (unlikely((snoop_flags & SNOOP_DROP_SEND) &&
-                            (ppd->dd->hfi1_snoop.mode_flag &
-                             HFI1_PORT_SNOOP_MODE))) {
-                       unsigned long flags;
-
-                       snoop_dbg("Dropping packet");
-                       if (qp->s_wqe) {
-                               spin_lock_irqsave(&qp->s_lock, flags);
-                               hfi1_send_complete(
-                                       qp,
-                                       qp->s_wqe,
-                                       IB_WC_SUCCESS);
-                               spin_unlock_irqrestore(&qp->s_lock, flags);
-                       } else if (qp->ibqp.qp_type == IB_QPT_RC) {
-                               spin_lock_irqsave(&qp->s_lock, flags);
-                               hfi1_rc_send_complete(qp,
-                                                     &ps->s_txreq->phdr.hdr);
-                               spin_unlock_irqrestore(&qp->s_lock, flags);
-                       }
-
-                       /*
-                        * If snoop is dropping the packet we need to put the
-                        * txreq back because no one else will.
-                        */
-                       hfi1_put_txreq(ps->s_txreq);
-                       return 0;
-               }
-               break;
-       default:
-               kfree(s_packet);
-               break;
-       }
-out:
-       return hfi1_verbs_send_pio(qp, ps, md.u.pbc);
-}
-
-/*
- * Callers of this must pass a hfi1_ib_header type for the from ptr. Currently
- * this can be used anywhere, but the intention is for inline ACKs for RC and
- * CCA packets. We don't restrict this usage though.
- */
-void snoop_inline_pio_send(struct hfi1_devdata *dd, struct pio_buf *pbuf,
-                          u64 pbc, const void *from, size_t count)
-{
-       int snoop_mode = 0;
-       int md_len = 0;
-       struct capture_md md;
-       struct snoop_packet *s_packet = NULL;
-
-       /*
-        * count is in dwords so we need to convert to bytes.
-        * We also need to account for CRC which would be tacked on by hardware.
-        */
-       int packet_len = (count << 2) + 4;
-       int ret;
-
-       snoop_dbg("ACK OUT: len %d", packet_len);
-
-       if (!dd->hfi1_snoop.filter_callback) {
-               snoop_dbg("filter not set");
-               ret = HFI1_FILTER_HIT;
-       } else {
-               ret = dd->hfi1_snoop.filter_callback(
-                               (struct hfi1_ib_header *)from,
-                               NULL,
-                               dd->hfi1_snoop.filter_value);
-       }
-
-       switch (ret) {
-       case HFI1_FILTER_ERR:
-               snoop_dbg("Error in filter call");
-               /* fall through */
-       case HFI1_FILTER_MISS:
-               snoop_dbg("Filter Miss");
-               break;
-       case HFI1_FILTER_HIT:
-               snoop_dbg("Capturing packet");
-               if (dd->hfi1_snoop.mode_flag & HFI1_PORT_SNOOP_MODE)
-                       snoop_mode = 1;
-               if ((snoop_mode == 0) ||
-                   unlikely(snoop_flags & SNOOP_USE_METADATA))
-                       md_len = sizeof(struct capture_md);
-
-               s_packet = allocate_snoop_packet(packet_len, 0, md_len);
-
-               if (unlikely(!s_packet)) {
-                       dd_dev_warn_ratelimited(dd, "Unable to allocate snoop/capture packet\n");
-                       goto inline_pio_out;
-               }
-
-               s_packet->total_len = packet_len + md_len;
-
-               /* Fill in the metadata for the packet */
-               if (md_len > 0) {
-                       memset(&md, 0, sizeof(struct capture_md));
-                       md.port = 1;
-                       md.dir = PKT_DIR_EGRESS;
-                       md.u.pbc = pbc;
-                       memcpy(s_packet->data, &md, md_len);
-               }
-
-               /* Add the packet data which is a single buffer */
-               memcpy(s_packet->data + md_len, from, packet_len);
-
-               snoop_list_add_tail(s_packet, dd);
-
-               if (unlikely((snoop_flags & SNOOP_DROP_SEND) && snoop_mode)) {
-                       snoop_dbg("Dropping packet");
-                       return;
-               }
-               break;
-       default:
-               break;
-       }
-
-inline_pio_out:
-       pio_copy(dd, pbuf, pbc, from, count);
-}
index 2eddd3305c6a66db4e579c7a7ed995247752dfec..4287c806642e1057cbe91531416cc2d22b3788d7 100644 (file)
@@ -1500,13 +1500,7 @@ static int user_add(struct hfi1_devdata *dd)
  */
 int hfi1_device_create(struct hfi1_devdata *dd)
 {
-       int r, ret;
-
-       r = user_add(dd);
-       ret = hfi1_diag_add(dd);
-       if (r && !ret)
-               ret = r;
-       return ret;
+       return user_add(dd);
 }
 
 /*
@@ -1516,5 +1510,4 @@ int hfi1_device_create(struct hfi1_devdata *dd)
 void hfi1_device_remove(struct hfi1_devdata *dd)
 {
        user_remove(dd);
-       hfi1_diag_remove(dd);
 }