[linux-2.6-block.git] / net / rds / ib_fmr.c

/*
 * Copyright (c) 2016 Oracle.  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.
 */

#include "ib_mr.h"

struct rds_ib_mr *rds_ib_alloc_fmr(struct rds_ib_device *rds_ibdev, int npages)
{
	struct rds_ib_mr_pool *pool;
	struct rds_ib_mr *ibmr = NULL;
	struct rds_ib_fmr *fmr;
	int err = 0;

	if (npages <= RDS_MR_8K_MSG_SIZE)
		pool = rds_ibdev->mr_8k_pool;
	else
		pool = rds_ibdev->mr_1m_pool;

	ibmr = rds_ib_try_reuse_ibmr(pool);
	if (ibmr)
		return ibmr;

	ibmr = kzalloc_node(sizeof(*ibmr), GFP_KERNEL,
			    rdsibdev_to_node(rds_ibdev));
	if (!ibmr) {
		err = -ENOMEM;
		goto out_no_cigar;
	}

	fmr = &ibmr->u.fmr;
	fmr->fmr = ib_alloc_fmr(rds_ibdev->pd,
			(IB_ACCESS_LOCAL_WRITE |
			 IB_ACCESS_REMOTE_READ |
			 IB_ACCESS_REMOTE_WRITE |
			 IB_ACCESS_REMOTE_ATOMIC),
			&pool->fmr_attr);
	if (IS_ERR(fmr->fmr)) {
		err = PTR_ERR(fmr->fmr);
		fmr->fmr = NULL;
		pr_warn("RDS/IB: %s failed (err=%d)\n", __func__, err);
		goto out_no_cigar;
	}

	ibmr->pool = pool;
	if (pool->pool_type == RDS_IB_MR_8K_POOL)
		rds_ib_stats_inc(s_ib_rdma_mr_8k_alloc);
	else
		rds_ib_stats_inc(s_ib_rdma_mr_1m_alloc);

	return ibmr;

out_no_cigar:
	kfree(ibmr);
	atomic_dec(&pool->item_count);

	return ERR_PTR(err);
}

int rds_ib_map_fmr(struct rds_ib_device *rds_ibdev, struct rds_ib_mr *ibmr,
		   struct scatterlist *sg, unsigned int nents)
{
	struct ib_device *dev = rds_ibdev->dev;
	struct rds_ib_fmr *fmr = &ibmr->u.fmr;
	struct scatterlist *scat = sg;
	u64 io_addr = 0;
	u64 *dma_pages;
	u32 len;
	int page_cnt, sg_dma_len;
	int i, j;
	int ret;

	sg_dma_len = ib_dma_map_sg(dev, sg, nents, DMA_BIDIRECTIONAL);
	if (unlikely(!sg_dma_len)) {
		pr_warn("RDS/IB: %s failed!\n", __func__);
		return -EBUSY;
	}

	len = 0;
	page_cnt = 0;

	for (i = 0; i < sg_dma_len; ++i) {
		unsigned int dma_len = ib_sg_dma_len(dev, &scat[i]);
		u64 dma_addr = ib_sg_dma_address(dev, &scat[i]);

		if (dma_addr & ~PAGE_MASK) {
			if (i > 0)
				return -EINVAL;
			else
				++page_cnt;
		}
		if ((dma_addr + dma_len) & ~PAGE_MASK) {
			if (i < sg_dma_len - 1)
				return -EINVAL;
			else
				++page_cnt;
		}

		len += dma_len;
	}

	page_cnt += len >> PAGE_SHIFT;
	if (page_cnt > ibmr->pool->fmr_attr.max_pages)
		return -EINVAL;

	dma_pages = kmalloc_node(sizeof(u64) * page_cnt, GFP_ATOMIC,
				 rdsibdev_to_node(rds_ibdev));
	if (!dma_pages)
		return -ENOMEM;

	page_cnt = 0;
	for (i = 0; i < sg_dma_len; ++i) {
		unsigned int dma_len = ib_sg_dma_len(dev, &scat[i]);
		u64 dma_addr = ib_sg_dma_address(dev, &scat[i]);

		for (j = 0; j < dma_len; j += PAGE_SIZE)
			dma_pages[page_cnt++] =
				(dma_addr & PAGE_MASK) + j;
	}

	ret = ib_map_phys_fmr(fmr->fmr, dma_pages, page_cnt, io_addr);
	if (ret)
		goto out;

	/* Success - we successfully remapped the MR, so we can
	 * safely tear down the old mapping.
	 */
	rds_ib_teardown_mr(ibmr);

	ibmr->sg = scat;
	ibmr->sg_len = nents;
	ibmr->sg_dma_len = sg_dma_len;
	ibmr->remap_count++;

	if (ibmr->pool->pool_type == RDS_IB_MR_8K_POOL)
		rds_ib_stats_inc(s_ib_rdma_mr_8k_used);
	else
		rds_ib_stats_inc(s_ib_rdma_mr_1m_used);
	ret = 0;

out:
	kfree(dma_pages);

	return ret;
}

struct rds_ib_mr *rds_ib_reg_fmr(struct rds_ib_device *rds_ibdev,
				 struct scatterlist *sg,
				 unsigned long nents,
				 u32 *key)
{
	struct rds_ib_mr *ibmr = NULL;
	struct rds_ib_fmr *fmr;
	int ret;

	ibmr = rds_ib_alloc_fmr(rds_ibdev, nents);
	if (IS_ERR(ibmr))
		return ibmr;

	ibmr->device = rds_ibdev;
	fmr = &ibmr->u.fmr;
	ret = rds_ib_map_fmr(rds_ibdev, ibmr, sg, nents);
	if (ret == 0)
		*key = fmr->fmr->rkey;
	else
		rds_ib_free_mr(ibmr, 0);

	return ibmr;
}

void rds_ib_unreg_fmr(struct list_head *list, unsigned int *nfreed,
		      unsigned long *unpinned, unsigned int goal)
{
	struct rds_ib_mr *ibmr, *next;
	struct rds_ib_fmr *fmr;
	LIST_HEAD(fmr_list);
	int ret = 0;
	unsigned int freed = *nfreed;

	/* String all ib_mr's onto one list and hand them to  ib_unmap_fmr */
	list_for_each_entry(ibmr, list, unmap_list) {
		fmr = &ibmr->u.fmr;
		list_add(&fmr->fmr->list, &fmr_list);
	}

	ret = ib_unmap_fmr(&fmr_list);
	if (ret)
		pr_warn("RDS/IB: FMR invalidation failed (err=%d)\n", ret);

	/* Now we can destroy the DMA mapping and unpin any pages */
	list_for_each_entry_safe(ibmr, next, list, unmap_list) {
		fmr = &ibmr->u.fmr;
		*unpinned += ibmr->sg_len;
		__rds_ib_teardown_mr(ibmr);
		if (freed < goal ||
		    ibmr->remap_count >= ibmr->pool->fmr_attr.max_maps) {
			if (ibmr->pool->pool_type == RDS_IB_MR_8K_POOL)
				rds_ib_stats_inc(s_ib_rdma_mr_8k_free);
			else
				rds_ib_stats_inc(s_ib_rdma_mr_1m_free);
			list_del(&ibmr->unmap_list);
			ib_dealloc_fmr(fmr->fmr);
			kfree(ibmr);
			freed++;
		}
	}
	*nfreed = freed;
}

void rds_ib_free_fmr_list(struct rds_ib_mr *ibmr)
{
	struct rds_ib_mr_pool *pool = ibmr->pool;

	if (ibmr->remap_count >= pool->fmr_attr.max_maps)
		llist_add(&ibmr->llnode, &pool->drop_list);
	else
		llist_add(&ibmr->llnode, &pool->free_list);
}
Commit	Line	Data
f6df683f	1	/*
	2	* Copyright (c) 2016 Oracle. All rights reserved.
	3	*
	4	* This software is available to you under a choice of one of two
	5	* licenses. You may choose to be licensed under the terms of the GNU
	6	* General Public License (GPL) Version 2, available from the file
	7	* COPYING in the main directory of this source tree, or the
	8	* OpenIB.org BSD license below:
	9	*
	10	* Redistribution and use in source and binary forms, with or
	11	* without modification, are permitted provided that the following
	12	* conditions are met:
	13	*
	14	* - Redistributions of source code must retain the above
	15	* copyright notice, this list of conditions and the following
	16	* disclaimer.
	17	*
	18	* - Redistributions in binary form must reproduce the above
	19	* copyright notice, this list of conditions and the following
	20	* disclaimer in the documentation and/or other materials
	21	* provided with the distribution.
	22	*
	23	* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
	24	* EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
	25	* MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
	26	* NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
	27	* BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
	28	* ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
	29	* CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
	30	* SOFTWARE.
	31	*/
	32
	33	#include "ib_mr.h"
	34
	35	struct rds_ib_mr rds_ib_alloc_fmr(struct rds_ib_device rds_ibdev, int npages)
	36	{
	37	struct rds_ib_mr_pool *pool;
	38	struct rds_ib_mr *ibmr = NULL;
a69365a3	39	struct rds_ib_fmr *fmr;
490ea596	40	int err = 0;
f6df683f	41
	42	if (npages <= RDS_MR_8K_MSG_SIZE)
	43	pool = rds_ibdev->mr_8k_pool;
	44	else
	45	pool = rds_ibdev->mr_1m_pool;
	46
490ea596	47	ibmr = rds_ib_try_reuse_ibmr(pool);
	48	if (ibmr)
	49	return ibmr;
f6df683f	50
	51	ibmr = kzalloc_node(sizeof(*ibmr), GFP_KERNEL,
	52	rdsibdev_to_node(rds_ibdev));
	53	if (!ibmr) {
	54	err = -ENOMEM;
	55	goto out_no_cigar;
	56	}
	57
a69365a3	58	fmr = &ibmr->u.fmr;
a69365a3	59	fmr->fmr = ib_alloc_fmr(rds_ibdev->pd,
f6df683f	60	(IB_ACCESS_LOCAL_WRITE \|
	61	IB_ACCESS_REMOTE_READ \|
	62	IB_ACCESS_REMOTE_WRITE \|
	63	IB_ACCESS_REMOTE_ATOMIC),
	64	&pool->fmr_attr);
a69365a3	65	if (IS_ERR(fmr->fmr)) {
	66	err = PTR_ERR(fmr->fmr);
	67	fmr->fmr = NULL;
f6df683f	68	pr_warn("RDS/IB: %s failed (err=%d)\n", __func__, err);
	69	goto out_no_cigar;
	70	}
	71
	72	ibmr->pool = pool;
	73	if (pool->pool_type == RDS_IB_MR_8K_POOL)
	74	rds_ib_stats_inc(s_ib_rdma_mr_8k_alloc);
	75	else
	76	rds_ib_stats_inc(s_ib_rdma_mr_1m_alloc);
	77
	78	return ibmr;
	79
	80	out_no_cigar:
ea69c883	81	kfree(ibmr);
f6df683f	82	atomic_dec(&pool->item_count);
ea69c883	83
f6df683f	84	return ERR_PTR(err);
	85	}
	86
	87	int rds_ib_map_fmr(struct rds_ib_device rds_ibdev, struct rds_ib_mr ibmr,
	88	struct scatterlist *sg, unsigned int nents)
	89	{
	90	struct ib_device *dev = rds_ibdev->dev;
a69365a3	91	struct rds_ib_fmr *fmr = &ibmr->u.fmr;
f6df683f	92	struct scatterlist *scat = sg;
	93	u64 io_addr = 0;
	94	u64 *dma_pages;
	95	u32 len;
	96	int page_cnt, sg_dma_len;
	97	int i, j;
	98	int ret;
	99
	100	sg_dma_len = ib_dma_map_sg(dev, sg, nents, DMA_BIDIRECTIONAL);
	101	if (unlikely(!sg_dma_len)) {
	102	pr_warn("RDS/IB: %s failed!\n", __func__);
	103	return -EBUSY;
	104	}
	105
	106	len = 0;
	107	page_cnt = 0;
	108
	109	for (i = 0; i < sg_dma_len; ++i) {
	110	unsigned int dma_len = ib_sg_dma_len(dev, &scat[i]);
	111	u64 dma_addr = ib_sg_dma_address(dev, &scat[i]);
	112
	113	if (dma_addr & ~PAGE_MASK) {
	114	if (i > 0)
	115	return -EINVAL;
	116	else
	117	++page_cnt;
	118	}
	119	if ((dma_addr + dma_len) & ~PAGE_MASK) {
	120	if (i < sg_dma_len - 1)
	121	return -EINVAL;
	122	else
	123	++page_cnt;
	124	}
	125
	126	len += dma_len;
	127	}
	128
	129	page_cnt += len >> PAGE_SHIFT;
	130	if (page_cnt > ibmr->pool->fmr_attr.max_pages)
	131	return -EINVAL;
	132
	133	dma_pages = kmalloc_node(sizeof(u64) * page_cnt, GFP_ATOMIC,
	134	rdsibdev_to_node(rds_ibdev));
	135	if (!dma_pages)
	136	return -ENOMEM;
	137
	138	page_cnt = 0;
	139	for (i = 0; i < sg_dma_len; ++i) {
	140	unsigned int dma_len = ib_sg_dma_len(dev, &scat[i]);
	141	u64 dma_addr = ib_sg_dma_address(dev, &scat[i]);
	142
	143	for (j = 0; j < dma_len; j += PAGE_SIZE)
	144	dma_pages[page_cnt++] =
	145	(dma_addr & PAGE_MASK) + j;
	146	}
	147
a69365a3	148	ret = ib_map_phys_fmr(fmr->fmr, dma_pages, page_cnt, io_addr);
f6df683f	149	if (ret)
	150	goto out;
	151
	152	/* Success - we successfully remapped the MR, so we can
	153	* safely tear down the old mapping.
	154	*/
	155	rds_ib_teardown_mr(ibmr);
	156
	157	ibmr->sg = scat;
	158	ibmr->sg_len = nents;
	159	ibmr->sg_dma_len = sg_dma_len;
	160	ibmr->remap_count++;
	161
	162	if (ibmr->pool->pool_type == RDS_IB_MR_8K_POOL)
	163	rds_ib_stats_inc(s_ib_rdma_mr_8k_used);
	164	else
	165	rds_ib_stats_inc(s_ib_rdma_mr_1m_used);
	166	ret = 0;
	167
	168	out:
	169	kfree(dma_pages);
	170
	171	return ret;
	172	}
490ea596	173
	174	struct rds_ib_mr rds_ib_reg_fmr(struct rds_ib_device rds_ibdev,
	175	struct scatterlist *sg,
	176	unsigned long nents,
	177	u32 *key)
	178	{
	179	struct rds_ib_mr *ibmr = NULL;
	180	struct rds_ib_fmr *fmr;
	181	int ret;
	182
	183	ibmr = rds_ib_alloc_fmr(rds_ibdev, nents);
	184	if (IS_ERR(ibmr))
	185	return ibmr;
	186
	187	ibmr->device = rds_ibdev;
	188	fmr = &ibmr->u.fmr;
	189	ret = rds_ib_map_fmr(rds_ibdev, ibmr, sg, nents);
	190	if (ret == 0)
	191	*key = fmr->fmr->rkey;
	192	else
	193	rds_ib_free_mr(ibmr, 0);
	194
	195	return ibmr;
	196	}
	197
	198	void rds_ib_unreg_fmr(struct list_head list, unsigned int nfreed,
	199	unsigned long *unpinned, unsigned int goal)
	200	{
	201	struct rds_ib_mr ibmr, next;
	202	struct rds_ib_fmr *fmr;
	203	LIST_HEAD(fmr_list);
	204	int ret = 0;
	205	unsigned int freed = *nfreed;
	206
	207	/* String all ib_mr's onto one list and hand them to ib_unmap_fmr */
	208	list_for_each_entry(ibmr, list, unmap_list) {
	209	fmr = &ibmr->u.fmr;
	210	list_add(&fmr->fmr->list, &fmr_list);
	211	}
	212
	213	ret = ib_unmap_fmr(&fmr_list);
	214	if (ret)
	215	pr_warn("RDS/IB: FMR invalidation failed (err=%d)\n", ret);
	216
	217	/* Now we can destroy the DMA mapping and unpin any pages */
	218	list_for_each_entry_safe(ibmr, next, list, unmap_list) {
	219	fmr = &ibmr->u.fmr;
	220	*unpinned += ibmr->sg_len;
	221	__rds_ib_teardown_mr(ibmr);
	222	if (freed < goal \|\|
	223	ibmr->remap_count >= ibmr->pool->fmr_attr.max_maps) {
	224	if (ibmr->pool->pool_type == RDS_IB_MR_8K_POOL)
	225	rds_ib_stats_inc(s_ib_rdma_mr_8k_free);
	226	else
	227	rds_ib_stats_inc(s_ib_rdma_mr_1m_free);
	228	list_del(&ibmr->unmap_list);
	229	ib_dealloc_fmr(fmr->fmr);
	230	kfree(ibmr);
	231	freed++;
	232	}
	233	}
	234	*nfreed = freed;
	235	}
	236
237	void rds_ib_free_fmr_list(struct rds_ib_mr *ibmr)
238	{
239	struct rds_ib_mr_pool *pool = ibmr->pool;
240
241	if (ibmr->remap_count >= pool->fmr_attr.max_maps)
242	llist_add(&ibmr->llnode, &pool->drop_list);
243	else
244	llist_add(&ibmr->llnode, &pool->free_list);
245	}