Add the rt linux 4.1.3-rt3 as base
[kvmfornfv.git] / kernel / drivers / infiniband / hw / cxgb4 / resource.c
diff --git a/kernel/drivers/infiniband/hw/cxgb4/resource.c b/kernel/drivers/infiniband/hw/cxgb4/resource.c
new file mode 100644 (file)
index 0000000..67df71a
--- /dev/null
@@ -0,0 +1,453 @@
+/*
+ * Copyright (c) 2009-2010 Chelsio, Inc. All rights reserved.
+ *
+ * This software is available to you under a choice of one of two
+ * licenses.  You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the
+ * OpenIB.org BSD license below:
+ *
+ *     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.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+/* Crude resource management */
+#include <linux/spinlock.h>
+#include <linux/genalloc.h>
+#include <linux/ratelimit.h>
+#include "iw_cxgb4.h"
+
+static int c4iw_init_qid_table(struct c4iw_rdev *rdev)
+{
+       u32 i;
+
+       if (c4iw_id_table_alloc(&rdev->resource.qid_table,
+                               rdev->lldi.vr->qp.start,
+                               rdev->lldi.vr->qp.size,
+                               rdev->lldi.vr->qp.size, 0))
+               return -ENOMEM;
+
+       for (i = rdev->lldi.vr->qp.start;
+               i < rdev->lldi.vr->qp.start + rdev->lldi.vr->qp.size; i++)
+               if (!(i & rdev->qpmask))
+                       c4iw_id_free(&rdev->resource.qid_table, i);
+       return 0;
+}
+
+/* nr_* must be power of 2 */
+int c4iw_init_resource(struct c4iw_rdev *rdev, u32 nr_tpt, u32 nr_pdid)
+{
+       int err = 0;
+       err = c4iw_id_table_alloc(&rdev->resource.tpt_table, 0, nr_tpt, 1,
+                                       C4IW_ID_TABLE_F_RANDOM);
+       if (err)
+               goto tpt_err;
+       err = c4iw_init_qid_table(rdev);
+       if (err)
+               goto qid_err;
+       err = c4iw_id_table_alloc(&rdev->resource.pdid_table, 0,
+                                       nr_pdid, 1, 0);
+       if (err)
+               goto pdid_err;
+       return 0;
+ pdid_err:
+       c4iw_id_table_free(&rdev->resource.qid_table);
+ qid_err:
+       c4iw_id_table_free(&rdev->resource.tpt_table);
+ tpt_err:
+       return -ENOMEM;
+}
+
+/*
+ * returns 0 if no resource available
+ */
+u32 c4iw_get_resource(struct c4iw_id_table *id_table)
+{
+       u32 entry;
+       entry = c4iw_id_alloc(id_table);
+       if (entry == (u32)(-1))
+               return 0;
+       return entry;
+}
+
+void c4iw_put_resource(struct c4iw_id_table *id_table, u32 entry)
+{
+       PDBG("%s entry 0x%x\n", __func__, entry);
+       c4iw_id_free(id_table, entry);
+}
+
+u32 c4iw_get_cqid(struct c4iw_rdev *rdev, struct c4iw_dev_ucontext *uctx)
+{
+       struct c4iw_qid_list *entry;
+       u32 qid;
+       int i;
+
+       mutex_lock(&uctx->lock);
+       if (!list_empty(&uctx->cqids)) {
+               entry = list_entry(uctx->cqids.next, struct c4iw_qid_list,
+                                  entry);
+               list_del(&entry->entry);
+               qid = entry->qid;
+               kfree(entry);
+       } else {
+               qid = c4iw_get_resource(&rdev->resource.qid_table);
+               if (!qid)
+                       goto out;
+               mutex_lock(&rdev->stats.lock);
+               rdev->stats.qid.cur += rdev->qpmask + 1;
+               mutex_unlock(&rdev->stats.lock);
+               for (i = qid+1; i & rdev->qpmask; i++) {
+                       entry = kmalloc(sizeof *entry, GFP_KERNEL);
+                       if (!entry)
+                               goto out;
+                       entry->qid = i;
+                       list_add_tail(&entry->entry, &uctx->cqids);
+               }
+
+               /*
+                * now put the same ids on the qp list since they all
+                * map to the same db/gts page.
+                */
+               entry = kmalloc(sizeof *entry, GFP_KERNEL);
+               if (!entry)
+                       goto out;
+               entry->qid = qid;
+               list_add_tail(&entry->entry, &uctx->qpids);
+               for (i = qid+1; i & rdev->qpmask; i++) {
+                       entry = kmalloc(sizeof *entry, GFP_KERNEL);
+                       if (!entry)
+                               goto out;
+                       entry->qid = i;
+                       list_add_tail(&entry->entry, &uctx->qpids);
+               }
+       }
+out:
+       mutex_unlock(&uctx->lock);
+       PDBG("%s qid 0x%x\n", __func__, qid);
+       mutex_lock(&rdev->stats.lock);
+       if (rdev->stats.qid.cur > rdev->stats.qid.max)
+               rdev->stats.qid.max = rdev->stats.qid.cur;
+       mutex_unlock(&rdev->stats.lock);
+       return qid;
+}
+
+void c4iw_put_cqid(struct c4iw_rdev *rdev, u32 qid,
+                  struct c4iw_dev_ucontext *uctx)
+{
+       struct c4iw_qid_list *entry;
+
+       entry = kmalloc(sizeof *entry, GFP_KERNEL);
+       if (!entry)
+               return;
+       PDBG("%s qid 0x%x\n", __func__, qid);
+       entry->qid = qid;
+       mutex_lock(&uctx->lock);
+       list_add_tail(&entry->entry, &uctx->cqids);
+       mutex_unlock(&uctx->lock);
+}
+
+u32 c4iw_get_qpid(struct c4iw_rdev *rdev, struct c4iw_dev_ucontext *uctx)
+{
+       struct c4iw_qid_list *entry;
+       u32 qid;
+       int i;
+
+       mutex_lock(&uctx->lock);
+       if (!list_empty(&uctx->qpids)) {
+               entry = list_entry(uctx->qpids.next, struct c4iw_qid_list,
+                                  entry);
+               list_del(&entry->entry);
+               qid = entry->qid;
+               kfree(entry);
+       } else {
+               qid = c4iw_get_resource(&rdev->resource.qid_table);
+               if (!qid) {
+                       mutex_lock(&rdev->stats.lock);
+                       rdev->stats.qid.fail++;
+                       mutex_unlock(&rdev->stats.lock);
+                       goto out;
+               }
+               mutex_lock(&rdev->stats.lock);
+               rdev->stats.qid.cur += rdev->qpmask + 1;
+               mutex_unlock(&rdev->stats.lock);
+               for (i = qid+1; i & rdev->qpmask; i++) {
+                       entry = kmalloc(sizeof *entry, GFP_KERNEL);
+                       if (!entry)
+                               goto out;
+                       entry->qid = i;
+                       list_add_tail(&entry->entry, &uctx->qpids);
+               }
+
+               /*
+                * now put the same ids on the cq list since they all
+                * map to the same db/gts page.
+                */
+               entry = kmalloc(sizeof *entry, GFP_KERNEL);
+               if (!entry)
+                       goto out;
+               entry->qid = qid;
+               list_add_tail(&entry->entry, &uctx->cqids);
+               for (i = qid; i & rdev->qpmask; i++) {
+                       entry = kmalloc(sizeof *entry, GFP_KERNEL);
+                       if (!entry)
+                               goto out;
+                       entry->qid = i;
+                       list_add_tail(&entry->entry, &uctx->cqids);
+               }
+       }
+out:
+       mutex_unlock(&uctx->lock);
+       PDBG("%s qid 0x%x\n", __func__, qid);
+       mutex_lock(&rdev->stats.lock);
+       if (rdev->stats.qid.cur > rdev->stats.qid.max)
+               rdev->stats.qid.max = rdev->stats.qid.cur;
+       mutex_unlock(&rdev->stats.lock);
+       return qid;
+}
+
+void c4iw_put_qpid(struct c4iw_rdev *rdev, u32 qid,
+                  struct c4iw_dev_ucontext *uctx)
+{
+       struct c4iw_qid_list *entry;
+
+       entry = kmalloc(sizeof *entry, GFP_KERNEL);
+       if (!entry)
+               return;
+       PDBG("%s qid 0x%x\n", __func__, qid);
+       entry->qid = qid;
+       mutex_lock(&uctx->lock);
+       list_add_tail(&entry->entry, &uctx->qpids);
+       mutex_unlock(&uctx->lock);
+}
+
+void c4iw_destroy_resource(struct c4iw_resource *rscp)
+{
+       c4iw_id_table_free(&rscp->tpt_table);
+       c4iw_id_table_free(&rscp->qid_table);
+       c4iw_id_table_free(&rscp->pdid_table);
+}
+
+/*
+ * PBL Memory Manager.  Uses Linux generic allocator.
+ */
+
+#define MIN_PBL_SHIFT 8                        /* 256B == min PBL size (32 entries) */
+
+u32 c4iw_pblpool_alloc(struct c4iw_rdev *rdev, int size)
+{
+       unsigned long addr = gen_pool_alloc(rdev->pbl_pool, size);
+       PDBG("%s addr 0x%x size %d\n", __func__, (u32)addr, size);
+       mutex_lock(&rdev->stats.lock);
+       if (addr) {
+               rdev->stats.pbl.cur += roundup(size, 1 << MIN_PBL_SHIFT);
+               if (rdev->stats.pbl.cur > rdev->stats.pbl.max)
+                       rdev->stats.pbl.max = rdev->stats.pbl.cur;
+       } else
+               rdev->stats.pbl.fail++;
+       mutex_unlock(&rdev->stats.lock);
+       return (u32)addr;
+}
+
+void c4iw_pblpool_free(struct c4iw_rdev *rdev, u32 addr, int size)
+{
+       PDBG("%s addr 0x%x size %d\n", __func__, addr, size);
+       mutex_lock(&rdev->stats.lock);
+       rdev->stats.pbl.cur -= roundup(size, 1 << MIN_PBL_SHIFT);
+       mutex_unlock(&rdev->stats.lock);
+       gen_pool_free(rdev->pbl_pool, (unsigned long)addr, size);
+}
+
+int c4iw_pblpool_create(struct c4iw_rdev *rdev)
+{
+       unsigned pbl_start, pbl_chunk, pbl_top;
+
+       rdev->pbl_pool = gen_pool_create(MIN_PBL_SHIFT, -1);
+       if (!rdev->pbl_pool)
+               return -ENOMEM;
+
+       pbl_start = rdev->lldi.vr->pbl.start;
+       pbl_chunk = rdev->lldi.vr->pbl.size;
+       pbl_top = pbl_start + pbl_chunk;
+
+       while (pbl_start < pbl_top) {
+               pbl_chunk = min(pbl_top - pbl_start + 1, pbl_chunk);
+               if (gen_pool_add(rdev->pbl_pool, pbl_start, pbl_chunk, -1)) {
+                       PDBG("%s failed to add PBL chunk (%x/%x)\n",
+                            __func__, pbl_start, pbl_chunk);
+                       if (pbl_chunk <= 1024 << MIN_PBL_SHIFT) {
+                               printk(KERN_WARNING MOD
+                                      "Failed to add all PBL chunks (%x/%x)\n",
+                                      pbl_start,
+                                      pbl_top - pbl_start);
+                               return 0;
+                       }
+                       pbl_chunk >>= 1;
+               } else {
+                       PDBG("%s added PBL chunk (%x/%x)\n",
+                            __func__, pbl_start, pbl_chunk);
+                       pbl_start += pbl_chunk;
+               }
+       }
+
+       return 0;
+}
+
+void c4iw_pblpool_destroy(struct c4iw_rdev *rdev)
+{
+       gen_pool_destroy(rdev->pbl_pool);
+}
+
+/*
+ * RQT Memory Manager.  Uses Linux generic allocator.
+ */
+
+#define MIN_RQT_SHIFT 10       /* 1KB == min RQT size (16 entries) */
+
+u32 c4iw_rqtpool_alloc(struct c4iw_rdev *rdev, int size)
+{
+       unsigned long addr = gen_pool_alloc(rdev->rqt_pool, size << 6);
+       PDBG("%s addr 0x%x size %d\n", __func__, (u32)addr, size << 6);
+       if (!addr)
+               pr_warn_ratelimited(MOD "%s: Out of RQT memory\n",
+                                   pci_name(rdev->lldi.pdev));
+       mutex_lock(&rdev->stats.lock);
+       if (addr) {
+               rdev->stats.rqt.cur += roundup(size << 6, 1 << MIN_RQT_SHIFT);
+               if (rdev->stats.rqt.cur > rdev->stats.rqt.max)
+                       rdev->stats.rqt.max = rdev->stats.rqt.cur;
+       } else
+               rdev->stats.rqt.fail++;
+       mutex_unlock(&rdev->stats.lock);
+       return (u32)addr;
+}
+
+void c4iw_rqtpool_free(struct c4iw_rdev *rdev, u32 addr, int size)
+{
+       PDBG("%s addr 0x%x size %d\n", __func__, addr, size << 6);
+       mutex_lock(&rdev->stats.lock);
+       rdev->stats.rqt.cur -= roundup(size << 6, 1 << MIN_RQT_SHIFT);
+       mutex_unlock(&rdev->stats.lock);
+       gen_pool_free(rdev->rqt_pool, (unsigned long)addr, size << 6);
+}
+
+int c4iw_rqtpool_create(struct c4iw_rdev *rdev)
+{
+       unsigned rqt_start, rqt_chunk, rqt_top;
+
+       rdev->rqt_pool = gen_pool_create(MIN_RQT_SHIFT, -1);
+       if (!rdev->rqt_pool)
+               return -ENOMEM;
+
+       rqt_start = rdev->lldi.vr->rq.start;
+       rqt_chunk = rdev->lldi.vr->rq.size;
+       rqt_top = rqt_start + rqt_chunk;
+
+       while (rqt_start < rqt_top) {
+               rqt_chunk = min(rqt_top - rqt_start + 1, rqt_chunk);
+               if (gen_pool_add(rdev->rqt_pool, rqt_start, rqt_chunk, -1)) {
+                       PDBG("%s failed to add RQT chunk (%x/%x)\n",
+                            __func__, rqt_start, rqt_chunk);
+                       if (rqt_chunk <= 1024 << MIN_RQT_SHIFT) {
+                               printk(KERN_WARNING MOD
+                                      "Failed to add all RQT chunks (%x/%x)\n",
+                                      rqt_start, rqt_top - rqt_start);
+                               return 0;
+                       }
+                       rqt_chunk >>= 1;
+               } else {
+                       PDBG("%s added RQT chunk (%x/%x)\n",
+                            __func__, rqt_start, rqt_chunk);
+                       rqt_start += rqt_chunk;
+               }
+       }
+       return 0;
+}
+
+void c4iw_rqtpool_destroy(struct c4iw_rdev *rdev)
+{
+       gen_pool_destroy(rdev->rqt_pool);
+}
+
+/*
+ * On-Chip QP Memory.
+ */
+#define MIN_OCQP_SHIFT 12      /* 4KB == min ocqp size */
+
+u32 c4iw_ocqp_pool_alloc(struct c4iw_rdev *rdev, int size)
+{
+       unsigned long addr = gen_pool_alloc(rdev->ocqp_pool, size);
+       PDBG("%s addr 0x%x size %d\n", __func__, (u32)addr, size);
+       if (addr) {
+               mutex_lock(&rdev->stats.lock);
+               rdev->stats.ocqp.cur += roundup(size, 1 << MIN_OCQP_SHIFT);
+               if (rdev->stats.ocqp.cur > rdev->stats.ocqp.max)
+                       rdev->stats.ocqp.max = rdev->stats.ocqp.cur;
+               mutex_unlock(&rdev->stats.lock);
+       }
+       return (u32)addr;
+}
+
+void c4iw_ocqp_pool_free(struct c4iw_rdev *rdev, u32 addr, int size)
+{
+       PDBG("%s addr 0x%x size %d\n", __func__, addr, size);
+       mutex_lock(&rdev->stats.lock);
+       rdev->stats.ocqp.cur -= roundup(size, 1 << MIN_OCQP_SHIFT);
+       mutex_unlock(&rdev->stats.lock);
+       gen_pool_free(rdev->ocqp_pool, (unsigned long)addr, size);
+}
+
+int c4iw_ocqp_pool_create(struct c4iw_rdev *rdev)
+{
+       unsigned start, chunk, top;
+
+       rdev->ocqp_pool = gen_pool_create(MIN_OCQP_SHIFT, -1);
+       if (!rdev->ocqp_pool)
+               return -ENOMEM;
+
+       start = rdev->lldi.vr->ocq.start;
+       chunk = rdev->lldi.vr->ocq.size;
+       top = start + chunk;
+
+       while (start < top) {
+               chunk = min(top - start + 1, chunk);
+               if (gen_pool_add(rdev->ocqp_pool, start, chunk, -1)) {
+                       PDBG("%s failed to add OCQP chunk (%x/%x)\n",
+                            __func__, start, chunk);
+                       if (chunk <= 1024 << MIN_OCQP_SHIFT) {
+                               printk(KERN_WARNING MOD
+                                      "Failed to add all OCQP chunks (%x/%x)\n",
+                                      start, top - start);
+                               return 0;
+                       }
+                       chunk >>= 1;
+               } else {
+                       PDBG("%s added OCQP chunk (%x/%x)\n",
+                            __func__, start, chunk);
+                       start += chunk;
+               }
+       }
+       return 0;
+}
+
+void c4iw_ocqp_pool_destroy(struct c4iw_rdev *rdev)
+{
+       gen_pool_destroy(rdev->ocqp_pool);
+}