GRU Driver: kernel services provide by driver
Jack Steiner [Wed, 30 Jul 2008 05:33:57 +0000 (22:33 -0700)]
This file contains functions for handling services provided to other
kernel modules that use the GRU.

Signed-off-by: Jack Steiner <steiner@sgi.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

drivers/misc/sgi-gru/grukservices.c [new file with mode: 0644]

diff --git a/drivers/misc/sgi-gru/grukservices.c b/drivers/misc/sgi-gru/grukservices.c
new file mode 100644 (file)
index 0000000..234d165
--- /dev/null
@@ -0,0 +1,679 @@
+/*
+ * SN Platform GRU Driver
+ *
+ *              KERNEL SERVICES THAT USE THE GRU
+ *
+ *  Copyright (c) 2008 Silicon Graphics, Inc.  All Rights Reserved.
+ *
+ *  This program is free software; you can redistribute it and/or modify
+ *  it under the terms of the GNU General Public License as published by
+ *  the Free Software Foundation; either version 2 of the License, or
+ *  (at your option) any later version.
+ *
+ *  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.
+ *
+ *  You should have received a copy of the GNU General Public License
+ *  along with this program; if not, write to the Free Software
+ *  Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307 USA
+ */
+
+#include <linux/kernel.h>
+#include <linux/errno.h>
+#include <linux/slab.h>
+#include <linux/mm.h>
+#include <linux/smp_lock.h>
+#include <linux/spinlock.h>
+#include <linux/device.h>
+#include <linux/miscdevice.h>
+#include <linux/proc_fs.h>
+#include <linux/interrupt.h>
+#include <linux/uaccess.h>
+#include "gru.h"
+#include "grulib.h"
+#include "grutables.h"
+#include "grukservices.h"
+#include "gru_instructions.h"
+#include <asm/uv/uv_hub.h>
+
+/*
+ * Kernel GRU Usage
+ *
+ * The following is an interim algorithm for management of kernel GRU
+ * resources. This will likely be replaced when we better understand the
+ * kernel/user requirements.
+ *
+ * At boot time, the kernel permanently reserves a fixed number of
+ * CBRs/DSRs for each cpu to use. The resources are all taken from
+ * the GRU chiplet 1 on the blade. This leaves the full set of resources
+ * of chiplet 0 available to be allocated to a single user.
+ */
+
+/* Blade percpu resources PERMANENTLY reserved for kernel use */
+#define GRU_NUM_KERNEL_CBR      1
+#define GRU_NUM_KERNEL_DSR_BYTES 256
+#define KERNEL_CTXNUM           15
+
+/* GRU instruction attributes for all instructions */
+#define IMA                    IMA_CB_DELAY
+
+/* GRU cacheline size is always 64 bytes - even on arches with 128 byte lines */
+#define __gru_cacheline_aligned__                               \
+       __attribute__((__aligned__(GRU_CACHE_LINE_BYTES)))
+
+#define MAGIC  0x1234567887654321UL
+
+/* Default retry count for GRU errors on kernel instructions */
+#define EXCEPTION_RETRY_LIMIT  3
+
+/* Status of message queue sections */
+#define MQS_EMPTY              0
+#define MQS_FULL               1
+#define MQS_NOOP               2
+
+/*----------------- RESOURCE MANAGEMENT -------------------------------------*/
+/* optimized for x86_64 */
+struct message_queue {
+       union gru_mesqhead      head __gru_cacheline_aligned__; /* CL 0 */
+       int                     qlines;                         /* DW 1 */
+       long                    hstatus[2];
+       void                    *next __gru_cacheline_aligned__;/* CL 1 */
+       void                    *limit;
+       void                    *start;
+       void                    *start2;
+       char                    data ____cacheline_aligned;     /* CL 2 */
+};
+
+/* First word in every message - used by mesq interface */
+struct message_header {
+       char    present;
+       char    present2;
+       char    lines;
+       char    fill;
+};
+
+#define QLINES(mq)     ((mq) + offsetof(struct message_queue, qlines))
+#define HSTATUS(mq, h) ((mq) + offsetof(struct message_queue, hstatus[h]))
+
+static int gru_get_cpu_resources(int dsr_bytes, void **cb, void **dsr)
+{
+       struct gru_blade_state *bs;
+       int lcpu;
+
+       BUG_ON(dsr_bytes > GRU_NUM_KERNEL_DSR_BYTES);
+       preempt_disable();
+       bs = gru_base[uv_numa_blade_id()];
+       lcpu = uv_blade_processor_id();
+       *cb = bs->kernel_cb + lcpu * GRU_HANDLE_STRIDE;
+       *dsr = bs->kernel_dsr + lcpu * GRU_NUM_KERNEL_DSR_BYTES;
+       return 0;
+}
+
+static void gru_free_cpu_resources(void *cb, void *dsr)
+{
+       preempt_enable();
+}
+
+int gru_get_cb_exception_detail(void *cb,
+               struct control_block_extended_exc_detail *excdet)
+{
+       struct gru_control_block_extended *cbe;
+
+       cbe = get_cbe(GRUBASE(cb), get_cb_number(cb));
+       excdet->opc = cbe->opccpy;
+       excdet->exopc = cbe->exopccpy;
+       excdet->ecause = cbe->ecause;
+       excdet->exceptdet0 = cbe->idef1upd;
+       excdet->exceptdet1 = cbe->idef3upd;
+       return 0;
+}
+
+char *gru_get_cb_exception_detail_str(int ret, void *cb,
+                                     char *buf, int size)
+{
+       struct gru_control_block_status *gen = (void *)cb;
+       struct control_block_extended_exc_detail excdet;
+
+       if (ret > 0 && gen->istatus == CBS_EXCEPTION) {
+               gru_get_cb_exception_detail(cb, &excdet);
+               snprintf(buf, size,
+                       "GRU exception: cb %p, opc %d, exopc %d, ecause 0x%x,"
+                       "excdet0 0x%lx, excdet1 0x%x",
+                       gen, excdet.opc, excdet.exopc, excdet.ecause,
+                       excdet.exceptdet0, excdet.exceptdet1);
+       } else {
+               snprintf(buf, size, "No exception");
+       }
+       return buf;
+}
+
+static int gru_wait_idle_or_exception(struct gru_control_block_status *gen)
+{
+       while (gen->istatus >= CBS_ACTIVE) {
+               cpu_relax();
+               barrier();
+       }
+       return gen->istatus;
+}
+
+static int gru_retry_exception(void *cb)
+{
+       struct gru_control_block_status *gen = (void *)cb;
+       struct control_block_extended_exc_detail excdet;
+       int retry = EXCEPTION_RETRY_LIMIT;
+
+       while (1)  {
+               if (gru_get_cb_message_queue_substatus(cb))
+                       break;
+               if (gru_wait_idle_or_exception(gen) == CBS_IDLE)
+                       return CBS_IDLE;
+
+               gru_get_cb_exception_detail(cb, &excdet);
+               if (excdet.ecause & ~EXCEPTION_RETRY_BITS)
+                       break;
+               if (retry-- == 0)
+                       break;
+               gen->icmd = 1;
+               gru_flush_cache(gen);
+       }
+       return CBS_EXCEPTION;
+}
+
+int gru_check_status_proc(void *cb)
+{
+       struct gru_control_block_status *gen = (void *)cb;
+       int ret;
+
+       ret = gen->istatus;
+       if (ret != CBS_EXCEPTION)
+               return ret;
+       return gru_retry_exception(cb);
+
+}
+
+int gru_wait_proc(void *cb)
+{
+       struct gru_control_block_status *gen = (void *)cb;
+       int ret;
+
+       ret = gru_wait_idle_or_exception(gen);
+       if (ret == CBS_EXCEPTION)
+               ret = gru_retry_exception(cb);
+
+       return ret;
+}
+
+void gru_abort(int ret, void *cb, char *str)
+{
+       char buf[GRU_EXC_STR_SIZE];
+
+       panic("GRU FATAL ERROR: %s - %s\n", str,
+             gru_get_cb_exception_detail_str(ret, cb, buf, sizeof(buf)));
+}
+
+void gru_wait_abort_proc(void *cb)
+{
+       int ret;
+
+       ret = gru_wait_proc(cb);
+       if (ret)
+               gru_abort(ret, cb, "gru_wait_abort");
+}
+
+
+/*------------------------------ MESSAGE QUEUES -----------------------------*/
+
+/* Internal status . These are NOT returned to the user. */
+#define MQIE_AGAIN             -1      /* try again */
+
+
+/*
+ * Save/restore the "present" flag that is in the second line of 2-line
+ * messages
+ */
+static inline int get_present2(void *p)
+{
+       struct message_header *mhdr = p + GRU_CACHE_LINE_BYTES;
+       return mhdr->present;
+}
+
+static inline void restore_present2(void *p, int val)
+{
+       struct message_header *mhdr = p + GRU_CACHE_LINE_BYTES;
+       mhdr->present = val;
+}
+
+/*
+ * Create a message queue.
+ *     qlines - message queue size in cache lines. Includes 2-line header.
+ */
+int gru_create_message_queue(void *p, unsigned int bytes)
+{
+       struct message_queue *mq = p;
+       unsigned int qlines;
+
+       qlines = bytes / GRU_CACHE_LINE_BYTES - 2;
+       memset(mq, 0, bytes);
+       mq->start = &mq->data;
+       mq->start2 = &mq->data + (qlines / 2 - 1) * GRU_CACHE_LINE_BYTES;
+       mq->next = &mq->data;
+       mq->limit = &mq->data + (qlines - 2) * GRU_CACHE_LINE_BYTES;
+       mq->qlines = qlines;
+       mq->hstatus[0] = 0;
+       mq->hstatus[1] = 1;
+       mq->head = gru_mesq_head(2, qlines / 2 + 1);
+       return 0;
+}
+EXPORT_SYMBOL_GPL(gru_create_message_queue);
+
+/*
+ * Send a NOOP message to a message queue
+ *     Returns:
+ *              0 - if queue is full after the send. This is the normal case
+ *                  but various races can change this.
+ *             -1 - if mesq sent successfully but queue not full
+ *             >0 - unexpected error. MQE_xxx returned
+ */
+static int send_noop_message(void *cb,
+                               unsigned long mq, void *mesg)
+{
+       const struct message_header noop_header = {
+                                       .present = MQS_NOOP, .lines = 1};
+       unsigned long m;
+       int substatus, ret;
+       struct message_header save_mhdr, *mhdr = mesg;
+
+       STAT(mesq_noop);
+       save_mhdr = *mhdr;
+       *mhdr = noop_header;
+       gru_mesq(cb, mq, gru_get_tri(mhdr), 1, IMA);
+       ret = gru_wait(cb);
+
+       if (ret) {
+               substatus = gru_get_cb_message_queue_substatus(cb);
+               switch (substatus) {
+               case CBSS_NO_ERROR:
+                       STAT(mesq_noop_unexpected_error);
+                       ret = MQE_UNEXPECTED_CB_ERR;
+                       break;
+               case CBSS_LB_OVERFLOWED:
+                       STAT(mesq_noop_lb_overflow);
+                       ret = MQE_CONGESTION;
+                       break;
+               case CBSS_QLIMIT_REACHED:
+                       STAT(mesq_noop_qlimit_reached);
+                       ret = 0;
+                       break;
+               case CBSS_AMO_NACKED:
+                       STAT(mesq_noop_amo_nacked);
+                       ret = MQE_CONGESTION;
+                       break;
+               case CBSS_PUT_NACKED:
+                       STAT(mesq_noop_put_nacked);
+                       m = mq + (gru_get_amo_value_head(cb) << 6);
+                       gru_vstore(cb, m, gru_get_tri(mesg), XTYPE_CL, 1, 1,
+                                               IMA);
+                       if (gru_wait(cb) == CBS_IDLE)
+                               ret = MQIE_AGAIN;
+                       else
+                               ret = MQE_UNEXPECTED_CB_ERR;
+                       break;
+               case CBSS_PAGE_OVERFLOW:
+               default:
+                       BUG();
+               }
+       }
+       *mhdr = save_mhdr;
+       return ret;
+}
+
+/*
+ * Handle a gru_mesq full.
+ */
+static int send_message_queue_full(void *cb,
+                          unsigned long mq, void *mesg, int lines)
+{
+       union gru_mesqhead mqh;
+       unsigned int limit, head;
+       unsigned long avalue;
+       int half, qlines, save;
+
+       /* Determine if switching to first/second half of q */
+       avalue = gru_get_amo_value(cb);
+       head = gru_get_amo_value_head(cb);
+       limit = gru_get_amo_value_limit(cb);
+
+       /*
+        * Fetch "qlines" from the queue header. Since the queue may be
+        * in memory that can't be accessed using socket addresses, use
+        * the GRU to access the data. Use DSR space from the message.
+        */
+       save = *(int *)mesg;
+       gru_vload(cb, QLINES(mq), gru_get_tri(mesg), XTYPE_W, 1, 1, IMA);
+       if (gru_wait(cb) != CBS_IDLE)
+               goto cberr;
+       qlines = *(int *)mesg;
+       *(int *)mesg = save;
+       half = (limit != qlines);
+
+       if (half)
+               mqh = gru_mesq_head(qlines / 2 + 1, qlines);
+       else
+               mqh = gru_mesq_head(2, qlines / 2 + 1);
+
+       /* Try to get lock for switching head pointer */
+       gru_gamir(cb, EOP_IR_CLR, HSTATUS(mq, half), XTYPE_DW, IMA);
+       if (gru_wait(cb) != CBS_IDLE)
+               goto cberr;
+       if (!gru_get_amo_value(cb)) {
+               STAT(mesq_qf_locked);
+               return MQE_QUEUE_FULL;
+       }
+
+       /* Got the lock. Send optional NOP if queue not full, */
+       if (head != limit) {
+               if (send_noop_message(cb, mq, mesg)) {
+                       gru_gamir(cb, EOP_IR_INC, HSTATUS(mq, half),
+                                       XTYPE_DW, IMA);
+                       if (gru_wait(cb) != CBS_IDLE)
+                               goto cberr;
+                       STAT(mesq_qf_noop_not_full);
+                       return MQIE_AGAIN;
+               }
+               avalue++;
+       }
+
+       /* Then flip queuehead to other half of queue. */
+       gru_gamer(cb, EOP_ERR_CSWAP, mq, XTYPE_DW, mqh.val, avalue, IMA);
+       if (gru_wait(cb) != CBS_IDLE)
+               goto cberr;
+
+       /* If not successfully in swapping queue head, clear the hstatus lock */
+       if (gru_get_amo_value(cb) != avalue) {
+               STAT(mesq_qf_switch_head_failed);
+               gru_gamir(cb, EOP_IR_INC, HSTATUS(mq, half), XTYPE_DW, IMA);
+               if (gru_wait(cb) != CBS_IDLE)
+                       goto cberr;
+       }
+       return MQIE_AGAIN;
+cberr:
+       STAT(mesq_qf_unexpected_error);
+       return MQE_UNEXPECTED_CB_ERR;
+}
+
+
+/*
+ * Handle a gru_mesq failure. Some of these failures are software recoverable
+ * or retryable.
+ */
+static int send_message_failure(void *cb,
+                               unsigned long mq,
+                               void *mesg,
+                               int lines)
+{
+       int substatus, ret = 0;
+       unsigned long m;
+
+       substatus = gru_get_cb_message_queue_substatus(cb);
+       switch (substatus) {
+       case CBSS_NO_ERROR:
+               STAT(mesq_send_unexpected_error);
+               ret = MQE_UNEXPECTED_CB_ERR;
+               break;
+       case CBSS_LB_OVERFLOWED:
+               STAT(mesq_send_lb_overflow);
+               ret = MQE_CONGESTION;
+               break;
+       case CBSS_QLIMIT_REACHED:
+               STAT(mesq_send_qlimit_reached);
+               ret = send_message_queue_full(cb, mq, mesg, lines);
+               break;
+       case CBSS_AMO_NACKED:
+               STAT(mesq_send_amo_nacked);
+               ret = MQE_CONGESTION;
+               break;
+       case CBSS_PUT_NACKED:
+               STAT(mesq_send_put_nacked);
+               m =mq + (gru_get_amo_value_head(cb) << 6);
+               gru_vstore(cb, m, gru_get_tri(mesg), XTYPE_CL, lines, 1, IMA);
+               if (gru_wait(cb) == CBS_IDLE)
+                       ret = MQE_OK;
+               else
+                       ret = MQE_UNEXPECTED_CB_ERR;
+               break;
+       default:
+               BUG();
+       }
+       return ret;
+}
+
+/*
+ * Send a message to a message queue
+ *     cb      GRU control block to use to send message
+ *     mq      message queue
+ *     mesg    message. ust be vaddr within a GSEG
+ *     bytes   message size (<= 2 CL)
+ */
+int gru_send_message_gpa(unsigned long mq, void *mesg, unsigned int bytes)
+{
+       struct message_header *mhdr;
+       void *cb;
+       void *dsr;
+       int istatus, clines, ret;
+
+       STAT(mesq_send);
+       BUG_ON(bytes < sizeof(int) || bytes > 2 * GRU_CACHE_LINE_BYTES);
+
+       clines = (bytes + GRU_CACHE_LINE_BYTES - 1) / GRU_CACHE_LINE_BYTES;
+       if (gru_get_cpu_resources(bytes, &cb, &dsr))
+               return MQE_BUG_NO_RESOURCES;
+       memcpy(dsr, mesg, bytes);
+       mhdr = dsr;
+       mhdr->present = MQS_FULL;
+       mhdr->lines = clines;
+       if (clines == 2) {
+               mhdr->present2 = get_present2(mhdr);
+               restore_present2(mhdr, MQS_FULL);
+       }
+
+       do {
+               ret = MQE_OK;
+               gru_mesq(cb, mq, gru_get_tri(mhdr), clines, IMA);
+               istatus = gru_wait(cb);
+               if (istatus != CBS_IDLE)
+                       ret = send_message_failure(cb, mq, dsr, clines);
+       } while (ret == MQIE_AGAIN);
+       gru_free_cpu_resources(cb, dsr);
+
+       if (ret)
+               STAT(mesq_send_failed);
+       return ret;
+}
+EXPORT_SYMBOL_GPL(gru_send_message_gpa);
+
+/*
+ * Advance the receive pointer for the queue to the next message.
+ */
+void gru_free_message(void *rmq, void *mesg)
+{
+       struct message_queue *mq = rmq;
+       struct message_header *mhdr = mq->next;
+       void *next, *pnext;
+       int half = -1;
+       int lines = mhdr->lines;
+
+       if (lines == 2)
+               restore_present2(mhdr, MQS_EMPTY);
+       mhdr->present = MQS_EMPTY;
+
+       pnext = mq->next;
+       next = pnext + GRU_CACHE_LINE_BYTES * lines;
+       if (next == mq->limit) {
+               next = mq->start;
+               half = 1;
+       } else if (pnext < mq->start2 && next >= mq->start2) {
+               half = 0;
+       }
+
+       if (half >= 0)
+               mq->hstatus[half] = 1;
+       mq->next = next;
+}
+EXPORT_SYMBOL_GPL(gru_free_message);
+
+/*
+ * Get next message from message queue. Return NULL if no message
+ * present. User must call next_message() to move to next message.
+ *     rmq     message queue
+ */
+void *gru_get_next_message(void *rmq)
+{
+       struct message_queue *mq = rmq;
+       struct message_header *mhdr = mq->next;
+       int present = mhdr->present;
+
+       /* skip NOOP messages */
+       STAT(mesq_receive);
+       while (present == MQS_NOOP) {
+               gru_free_message(rmq, mhdr);
+               mhdr = mq->next;
+               present = mhdr->present;
+       }
+
+       /* Wait for both halves of 2 line messages */
+       if (present == MQS_FULL && mhdr->lines == 2 &&
+                               get_present2(mhdr) == MQS_EMPTY)
+               present = MQS_EMPTY;
+
+       if (!present) {
+               STAT(mesq_receive_none);
+               return NULL;
+       }
+
+       if (mhdr->lines == 2)
+               restore_present2(mhdr, mhdr->present2);
+
+       return mhdr;
+}
+EXPORT_SYMBOL_GPL(gru_get_next_message);
+
+/* ---------------------- GRU DATA COPY FUNCTIONS ---------------------------*/
+
+/*
+ * Copy a block of data using the GRU resources
+ */
+int gru_copy_gpa(unsigned long dest_gpa, unsigned long src_gpa,
+                               unsigned int bytes)
+{
+       void *cb;
+       void *dsr;
+       int ret;
+
+       STAT(copy_gpa);
+       if (gru_get_cpu_resources(GRU_NUM_KERNEL_DSR_BYTES, &cb, &dsr))
+               return MQE_BUG_NO_RESOURCES;
+       gru_bcopy(cb, src_gpa, dest_gpa, gru_get_tri(dsr),
+                 XTYPE_B, bytes, GRU_NUM_KERNEL_DSR_BYTES, IMA);
+       ret = gru_wait(cb);
+       gru_free_cpu_resources(cb, dsr);
+       return ret;
+}
+EXPORT_SYMBOL_GPL(gru_copy_gpa);
+
+/* ------------------- KERNEL QUICKTESTS RUN AT STARTUP ----------------*/
+/*     Temp - will delete after we gain confidence in the GRU          */
+static __cacheline_aligned unsigned long word0;
+static __cacheline_aligned unsigned long word1;
+
+static int quicktest(struct gru_state *gru)
+{
+       void *cb;
+       void *ds;
+       unsigned long *p;
+
+       cb = get_gseg_base_address_cb(gru->gs_gru_base_vaddr, KERNEL_CTXNUM, 0);
+       ds = get_gseg_base_address_ds(gru->gs_gru_base_vaddr, KERNEL_CTXNUM, 0);
+       p = ds;
+       word0 = MAGIC;
+
+       gru_vload(cb, uv_gpa(&word0), 0, XTYPE_DW, 1, 1, IMA);
+       if (gru_wait(cb) != CBS_IDLE)
+               BUG();
+
+       if (*(unsigned long *)ds != MAGIC)
+               BUG();
+       gru_vstore(cb, uv_gpa(&word1), 0, XTYPE_DW, 1, 1, IMA);
+       if (gru_wait(cb) != CBS_IDLE)
+               BUG();
+
+       if (word0 != word1 || word0 != MAGIC) {
+               printk
+                   ("GRU quicktest err: gru %d, found 0x%lx, expected 0x%lx\n",
+                    gru->gs_gid, word1, MAGIC);
+               BUG();          /* ZZZ should not be fatal */
+       }
+
+       return 0;
+}
+
+
+int gru_kservices_init(struct gru_state *gru)
+{
+       struct gru_blade_state *bs;
+       struct gru_context_configuration_handle *cch;
+       unsigned long cbr_map, dsr_map;
+       int err, num, cpus_possible;
+
+       /*
+        * Currently, resources are reserved ONLY on the second chiplet
+        * on each blade. This leaves ALL resources on chiplet 0 available
+        * for user code.
+        */
+       bs = gru->gs_blade;
+       if (gru != &bs->bs_grus[1])
+               return 0;
+
+       cpus_possible = uv_blade_nr_possible_cpus(gru->gs_blade_id);
+
+       num = GRU_NUM_KERNEL_CBR * cpus_possible;
+       cbr_map = reserve_gru_cb_resources(gru, GRU_CB_COUNT_TO_AU(num), NULL);
+       gru->gs_reserved_cbrs += num;
+
+       num = GRU_NUM_KERNEL_DSR_BYTES * cpus_possible;
+       dsr_map = reserve_gru_ds_resources(gru, GRU_DS_BYTES_TO_AU(num), NULL);
+       gru->gs_reserved_dsr_bytes += num;
+
+       gru->gs_active_contexts++;
+       __set_bit(KERNEL_CTXNUM, &gru->gs_context_map);
+       cch = get_cch(gru->gs_gru_base_vaddr, KERNEL_CTXNUM);
+
+       bs->kernel_cb = get_gseg_base_address_cb(gru->gs_gru_base_vaddr,
+                                       KERNEL_CTXNUM, 0);
+       bs->kernel_dsr = get_gseg_base_address_ds(gru->gs_gru_base_vaddr,
+                                       KERNEL_CTXNUM, 0);
+
+       lock_cch_handle(cch);
+       cch->tfm_fault_bit_enable = 0;
+       cch->tlb_int_enable = 0;
+       cch->tfm_done_bit_enable = 0;
+       cch->unmap_enable = 1;
+       err = cch_allocate(cch, 0, cbr_map, dsr_map);
+       if (err) {
+               gru_dbg(grudev,
+                       "Unable to allocate kernel CCH: gru %d, err %d\n",
+                       gru->gs_gid, err);
+               BUG();
+       }
+       if (cch_start(cch)) {
+               gru_dbg(grudev, "Unable to start kernel CCH: gru %d, err %d\n",
+                       gru->gs_gid, err);
+               BUG();
+       }
+       unlock_cch_handle(cch);
+
+       if (options & GRU_QUICKLOOK)
+               quicktest(gru);
+       return 0;
+}