[linux-2.6-block.git] / drivers / infiniband / ulp / iser / iser_memory.c

/*
 * Copyright (c) 2004, 2005, 2006 Voltaire, Inc. All rights reserved.
 * Copyright (c) 2013 Mellanox Technologies. 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 <linux/module.h>
#include <linux/kernel.h>
#include <linux/slab.h>
#include <linux/mm.h>
#include <linux/highmem.h>
#include <linux/scatterlist.h>

#include "iscsi_iser.h"

#define ISER_KMALLOC_THRESHOLD 0x20000 /* 128K - kmalloc limit */

/**
 * iser_start_rdma_unaligned_sg
 */
static int iser_start_rdma_unaligned_sg(struct iscsi_iser_task *iser_task,
					enum iser_data_dir cmd_dir)
{
	int dma_nents;
	struct ib_device *dev;
	char *mem = NULL;
	struct iser_data_buf *data = &iser_task->data[cmd_dir];
	unsigned long  cmd_data_len = data->data_len;

	if (cmd_data_len > ISER_KMALLOC_THRESHOLD)
		mem = (void *)__get_free_pages(GFP_ATOMIC,
		      ilog2(roundup_pow_of_two(cmd_data_len)) - PAGE_SHIFT);
	else
		mem = kmalloc(cmd_data_len, GFP_ATOMIC);

	if (mem == NULL) {
		iser_err("Failed to allocate mem size %d %d for copying sglist\n",
			 data->size,(int)cmd_data_len);
		return -ENOMEM;
	}

	if (cmd_dir == ISER_DIR_OUT) {
		/* copy the unaligned sg the buffer which is used for RDMA */
		struct scatterlist *sgl = (struct scatterlist *)data->buf;
		struct scatterlist *sg;
		int i;
		char *p, *from;

		p = mem;
		for_each_sg(sgl, sg, data->size, i) {
			from = kmap_atomic(sg_page(sg));
			memcpy(p,
			       from + sg->offset,
			       sg->length);
			kunmap_atomic(from);
			p += sg->length;
		}
	}

	sg_init_one(&iser_task->data_copy[cmd_dir].sg_single, mem, cmd_data_len);
	iser_task->data_copy[cmd_dir].buf  =
		&iser_task->data_copy[cmd_dir].sg_single;
	iser_task->data_copy[cmd_dir].size = 1;

	iser_task->data_copy[cmd_dir].copy_buf  = mem;

	dev = iser_task->iser_conn->ib_conn->device->ib_device;
	dma_nents = ib_dma_map_sg(dev,
				  &iser_task->data_copy[cmd_dir].sg_single,
				  1,
				  (cmd_dir == ISER_DIR_OUT) ?
				  DMA_TO_DEVICE : DMA_FROM_DEVICE);
	BUG_ON(dma_nents == 0);

	iser_task->data_copy[cmd_dir].dma_nents = dma_nents;
	return 0;
}

/**
 * iser_finalize_rdma_unaligned_sg
 */
void iser_finalize_rdma_unaligned_sg(struct iscsi_iser_task *iser_task,
				     enum iser_data_dir         cmd_dir)
{
	struct ib_device *dev;
	struct iser_data_buf *mem_copy;
	unsigned long  cmd_data_len;

	dev = iser_task->iser_conn->ib_conn->device->ib_device;
	mem_copy = &iser_task->data_copy[cmd_dir];

	ib_dma_unmap_sg(dev, &mem_copy->sg_single, 1,
			(cmd_dir == ISER_DIR_OUT) ?
			DMA_TO_DEVICE : DMA_FROM_DEVICE);

	if (cmd_dir == ISER_DIR_IN) {
		char *mem;
		struct scatterlist *sgl, *sg;
		unsigned char *p, *to;
		unsigned int sg_size;
		int i;

		/* copy back read RDMA to unaligned sg */
		mem	= mem_copy->copy_buf;

		sgl	= (struct scatterlist *)iser_task->data[ISER_DIR_IN].buf;
		sg_size = iser_task->data[ISER_DIR_IN].size;

		p = mem;
		for_each_sg(sgl, sg, sg_size, i) {
			to = kmap_atomic(sg_page(sg));
			memcpy(to + sg->offset,
			       p,
			       sg->length);
			kunmap_atomic(to);
			p += sg->length;
		}
	}

	cmd_data_len = iser_task->data[cmd_dir].data_len;

	if (cmd_data_len > ISER_KMALLOC_THRESHOLD)
		free_pages((unsigned long)mem_copy->copy_buf,
			   ilog2(roundup_pow_of_two(cmd_data_len)) - PAGE_SHIFT);
	else
		kfree(mem_copy->copy_buf);

	mem_copy->copy_buf = NULL;
}

#define IS_4K_ALIGNED(addr)	((((unsigned long)addr) & ~MASK_4K) == 0)

/**
 * iser_sg_to_page_vec - Translates scatterlist entries to physical addresses
 * and returns the length of resulting physical address array (may be less than
 * the original due to possible compaction).
 *
 * we build a "page vec" under the assumption that the SG meets the RDMA
 * alignment requirements. Other then the first and last SG elements, all
 * the "internal" elements can be compacted into a list whose elements are
 * dma addresses of physical pages. The code supports also the weird case
 * where --few fragments of the same page-- are present in the SG as
 * consecutive elements. Also, it handles one entry SG.
 */

static int iser_sg_to_page_vec(struct iser_data_buf *data,
			       struct ib_device *ibdev, u64 *pages,
			       int *offset, int *data_size)
{
	struct scatterlist *sg, *sgl = (struct scatterlist *)data->buf;
	u64 start_addr, end_addr, page, chunk_start = 0;
	unsigned long total_sz = 0;
	unsigned int dma_len;
	int i, new_chunk, cur_page, last_ent = data->dma_nents - 1;

	/* compute the offset of first element */
	*offset = (u64) sgl[0].offset & ~MASK_4K;

	new_chunk = 1;
	cur_page  = 0;
	for_each_sg(sgl, sg, data->dma_nents, i) {
		start_addr = ib_sg_dma_address(ibdev, sg);
		if (new_chunk)
			chunk_start = start_addr;
		dma_len = ib_sg_dma_len(ibdev, sg);
		end_addr = start_addr + dma_len;
		total_sz += dma_len;

		/* collect page fragments until aligned or end of SG list */
		if (!IS_4K_ALIGNED(end_addr) && i < last_ent) {
			new_chunk = 0;
			continue;
		}
		new_chunk = 1;

		/* address of the first page in the contiguous chunk;
		   masking relevant for the very first SG entry,
		   which might be unaligned */
		page = chunk_start & MASK_4K;
		do {
			pages[cur_page++] = page;
			page += SIZE_4K;
		} while (page < end_addr);
	}

	*data_size = total_sz;
	iser_dbg("page_vec->data_size:%d cur_page %d\n",
		 *data_size, cur_page);
	return cur_page;
}


/**
 * iser_data_buf_aligned_len - Tries to determine the maximal correctly aligned
 * for RDMA sub-list of a scatter-gather list of memory buffers, and  returns
 * the number of entries which are aligned correctly. Supports the case where
 * consecutive SG elements are actually fragments of the same physcial page.
 */
static int iser_data_buf_aligned_len(struct iser_data_buf *data,
				      struct ib_device *ibdev)
{
	struct scatterlist *sgl, *sg, *next_sg = NULL;
	u64 start_addr, end_addr;
	int i, ret_len, start_check = 0;

	if (data->dma_nents == 1)
		return 1;

	sgl = (struct scatterlist *)data->buf;
	start_addr  = ib_sg_dma_address(ibdev, sgl);

	for_each_sg(sgl, sg, data->dma_nents, i) {
		if (start_check && !IS_4K_ALIGNED(start_addr))
			break;

		next_sg = sg_next(sg);
		if (!next_sg)
			break;

		end_addr    = start_addr + ib_sg_dma_len(ibdev, sg);
		start_addr  = ib_sg_dma_address(ibdev, next_sg);

		if (end_addr == start_addr) {
			start_check = 0;
			continue;
		} else
			start_check = 1;

		if (!IS_4K_ALIGNED(end_addr))
			break;
	}
	ret_len = (next_sg) ? i : i+1;
	iser_dbg("Found %d aligned entries out of %d in sg:0x%p\n",
		 ret_len, data->dma_nents, data);
	return ret_len;
}

static void iser_data_buf_dump(struct iser_data_buf *data,
			       struct ib_device *ibdev)
{
	struct scatterlist *sgl = (struct scatterlist *)data->buf;
	struct scatterlist *sg;
	int i;

	for_each_sg(sgl, sg, data->dma_nents, i)
		iser_dbg("sg[%d] dma_addr:0x%lX page:0x%p "
			 "off:0x%x sz:0x%x dma_len:0x%x\n",
			 i, (unsigned long)ib_sg_dma_address(ibdev, sg),
			 sg_page(sg), sg->offset,
			 sg->length, ib_sg_dma_len(ibdev, sg));
}

static void iser_dump_page_vec(struct iser_page_vec *page_vec)
{
	int i;

	iser_err("page vec length %d data size %d\n",
		 page_vec->length, page_vec->data_size);
	for (i = 0; i < page_vec->length; i++)
		iser_err("%d %lx\n",i,(unsigned long)page_vec->pages[i]);
}

static void iser_page_vec_build(struct iser_data_buf *data,
				struct iser_page_vec *page_vec,
				struct ib_device *ibdev)
{
	int page_vec_len = 0;

	page_vec->length = 0;
	page_vec->offset = 0;

	iser_dbg("Translating sg sz: %d\n", data->dma_nents);
	page_vec_len = iser_sg_to_page_vec(data, ibdev, page_vec->pages,
					   &page_vec->offset,
					   &page_vec->data_size);
	iser_dbg("sg len %d page_vec_len %d\n", data->dma_nents, page_vec_len);

	page_vec->length = page_vec_len;

	if (page_vec_len * SIZE_4K < page_vec->data_size) {
		iser_err("page_vec too short to hold this SG\n");
		iser_data_buf_dump(data, ibdev);
		iser_dump_page_vec(page_vec);
		BUG();
	}
}

int iser_dma_map_task_data(struct iscsi_iser_task *iser_task,
			    struct iser_data_buf *data,
			    enum iser_data_dir iser_dir,
			    enum dma_data_direction dma_dir)
{
	struct ib_device *dev;

	iser_task->dir[iser_dir] = 1;
	dev = iser_task->iser_conn->ib_conn->device->ib_device;

	data->dma_nents = ib_dma_map_sg(dev, data->buf, data->size, dma_dir);
	if (data->dma_nents == 0) {
		iser_err("dma_map_sg failed!!!\n");
		return -EINVAL;
	}
	return 0;
}

void iser_dma_unmap_task_data(struct iscsi_iser_task *iser_task)
{
	struct ib_device *dev;
	struct iser_data_buf *data;

	dev = iser_task->iser_conn->ib_conn->device->ib_device;

	if (iser_task->dir[ISER_DIR_IN]) {
		data = &iser_task->data[ISER_DIR_IN];
		ib_dma_unmap_sg(dev, data->buf, data->size, DMA_FROM_DEVICE);
	}

	if (iser_task->dir[ISER_DIR_OUT]) {
		data = &iser_task->data[ISER_DIR_OUT];
		ib_dma_unmap_sg(dev, data->buf, data->size, DMA_TO_DEVICE);
	}
}

static int fall_to_bounce_buf(struct iscsi_iser_task *iser_task,
			      struct ib_device *ibdev,
			      enum iser_data_dir cmd_dir,
			      int aligned_len)
{
	struct iscsi_conn    *iscsi_conn = iser_task->iser_conn->iscsi_conn;
	struct iser_data_buf *mem = &iser_task->data[cmd_dir];

	iscsi_conn->fmr_unalign_cnt++;
	iser_warn("rdma alignment violation (%d/%d aligned) or FMR not supported\n",
		  aligned_len, mem->size);

	if (iser_debug_level > 0)
		iser_data_buf_dump(mem, ibdev);

	/* unmap the command data before accessing it */
	iser_dma_unmap_task_data(iser_task);

	/* allocate copy buf, if we are writing, copy the */
	/* unaligned scatterlist, dma map the copy        */
	if (iser_start_rdma_unaligned_sg(iser_task, cmd_dir) != 0)
			return -ENOMEM;

	return 0;
}

/**
 * iser_reg_rdma_mem_fmr - Registers memory intended for RDMA,
 * using FMR (if possible) obtaining rkey and va
 *
 * returns 0 on success, errno code on failure
 */
int iser_reg_rdma_mem_fmr(struct iscsi_iser_task *iser_task,
			  enum iser_data_dir cmd_dir)
{
	struct iser_conn     *ib_conn = iser_task->iser_conn->ib_conn;
	struct iser_device   *device = ib_conn->device;
	struct ib_device     *ibdev = device->ib_device;
	struct iser_data_buf *mem = &iser_task->data[cmd_dir];
	struct iser_regd_buf *regd_buf;
	int aligned_len;
	int err;
	int i;
	struct scatterlist *sg;

	regd_buf = &iser_task->rdma_regd[cmd_dir];

	aligned_len = iser_data_buf_aligned_len(mem, ibdev);
	if (aligned_len != mem->dma_nents ||
	    (!ib_conn->fastreg.fmr.pool && mem->dma_nents > 1)) {
		err = fall_to_bounce_buf(iser_task, ibdev,
					 cmd_dir, aligned_len);
		if (err) {
			iser_err("failed to allocate bounce buffer\n");
			return err;
		}
		mem = &iser_task->data_copy[cmd_dir];
	}

	/* if there a single dma entry, FMR is not needed */
	if (mem->dma_nents == 1) {
		sg = (struct scatterlist *)mem->buf;

		regd_buf->reg.lkey = device->mr->lkey;
		regd_buf->reg.rkey = device->mr->rkey;
		regd_buf->reg.len  = ib_sg_dma_len(ibdev, &sg[0]);
		regd_buf->reg.va   = ib_sg_dma_address(ibdev, &sg[0]);
		regd_buf->reg.is_fmr = 0;

		iser_dbg("PHYSICAL Mem.register: lkey: 0x%08X rkey: 0x%08X  "
			 "va: 0x%08lX sz: %ld]\n",
			 (unsigned int)regd_buf->reg.lkey,
			 (unsigned int)regd_buf->reg.rkey,
			 (unsigned long)regd_buf->reg.va,
			 (unsigned long)regd_buf->reg.len);
	} else { /* use FMR for multiple dma entries */
		iser_page_vec_build(mem, ib_conn->fastreg.fmr.page_vec, ibdev);
		err = iser_reg_page_vec(ib_conn, ib_conn->fastreg.fmr.page_vec,
					&regd_buf->reg);
		if (err && err != -EAGAIN) {
			iser_data_buf_dump(mem, ibdev);
			iser_err("mem->dma_nents = %d (dlength = 0x%x)\n",
				 mem->dma_nents,
				 ntoh24(iser_task->desc.iscsi_header.dlength));
			iser_err("page_vec: data_size = 0x%x, length = %d, offset = 0x%x\n",
				 ib_conn->fastreg.fmr.page_vec->data_size,
				 ib_conn->fastreg.fmr.page_vec->length,
				 ib_conn->fastreg.fmr.page_vec->offset);
			for (i = 0; i < ib_conn->fastreg.fmr.page_vec->length; i++)
				iser_err("page_vec[%d] = 0x%llx\n", i,
					 (unsigned long long) ib_conn->fastreg.fmr.page_vec->pages[i]);
		}
		if (err)
			return err;
	}
	return 0;
}
Commit	Line	Data
6461f64a OG	1	/*
6461f64a OG	2	* Copyright (c) 2004, 2005, 2006 Voltaire, Inc. All rights reserved.
28f292e8	3	* Copyright (c) 2013 Mellanox Technologies. All rights reserved.
6461f64a OG	4	*
	5	* This software is available to you under a choice of one of two
	6	* licenses. You may choose to be licensed under the terms of the GNU
	7	* General Public License (GPL) Version 2, available from the file
	8	* COPYING in the main directory of this source tree, or the
	9	* OpenIB.org BSD license below:
	10	*
	11	* Redistribution and use in source and binary forms, with or
	12	* without modification, are permitted provided that the following
	13	* conditions are met:
	14	*
	15	* - Redistributions of source code must retain the above
	16	* copyright notice, this list of conditions and the following
	17	* disclaimer.
	18	*
	19	* - Redistributions in binary form must reproduce the above
	20	* copyright notice, this list of conditions and the following
	21	* disclaimer in the documentation and/or other materials
	22	* provided with the distribution.
	23	*
	24	* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
	25	* EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
	26	* MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
	27	* NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
	28	* BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
	29	* ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
	30	* CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
	31	* SOFTWARE.
6461f64a OG	32	*/
	33	#include <linux/module.h>
	34	#include <linux/kernel.h>
	35	#include <linux/slab.h>
	36	#include <linux/mm.h>
a1f8e7f7	37	#include <linux/highmem.h>
6461f64a OG	38	#include <linux/scatterlist.h>
	39
	40	#include "iscsi_iser.h"
	41
	42	#define ISER_KMALLOC_THRESHOLD 0x20000 /* 128K - kmalloc limit */
8dfa0876	43
6461f64a OG	44	/**
	45	* iser_start_rdma_unaligned_sg
	46	*/
2261ec3d	47	static int iser_start_rdma_unaligned_sg(struct iscsi_iser_task *iser_task,
41179e2d	48	enum iser_data_dir cmd_dir)
6461f64a OG	49	{
6461f64a OG	50	int dma_nents;
5180311f	51	struct ib_device *dev;
6461f64a	52	char *mem = NULL;
2261ec3d	53	struct iser_data_buf *data = &iser_task->data[cmd_dir];
6461f64a OG	54	unsigned long cmd_data_len = data->data_len;
	55
	56	if (cmd_data_len > ISER_KMALLOC_THRESHOLD)
528f4e8c	57	mem = (void *)__get_free_pages(GFP_ATOMIC,
f0d1b0b3	58	ilog2(roundup_pow_of_two(cmd_data_len)) - PAGE_SHIFT);
6461f64a	59	else
528f4e8c	60	mem = kmalloc(cmd_data_len, GFP_ATOMIC);
6461f64a OG	61
	62	if (mem == NULL) {
	63	iser_err("Failed to allocate mem size %d %d for copying sglist\n",
	64	data->size,(int)cmd_data_len);
	65	return -ENOMEM;
	66	}
	67
	68	if (cmd_dir == ISER_DIR_OUT) {
	69	/* copy the unaligned sg the buffer which is used for RDMA */
53d412fc JA	70	struct scatterlist sgl = (struct scatterlist )data->buf;
53d412fc JA	71	struct scatterlist *sg;
6461f64a OG	72	int i;
	73	char p, from;
	74
53d412fc JA	75	p = mem;
53d412fc JA	76	for_each_sg(sgl, sg, data->size, i) {
2a156d09	77	from = kmap_atomic(sg_page(sg));
6461f64a	78	memcpy(p,
53d412fc JA	79	from + sg->offset,
53d412fc JA	80	sg->length);
2a156d09	81	kunmap_atomic(from);
53d412fc	82	p += sg->length;
6461f64a OG	83	}
	84	}
	85
2261ec3d MC	86	sg_init_one(&iser_task->data_copy[cmd_dir].sg_single, mem, cmd_data_len);
	87	iser_task->data_copy[cmd_dir].buf =
	88	&iser_task->data_copy[cmd_dir].sg_single;
	89	iser_task->data_copy[cmd_dir].size = 1;
6461f64a	90
2261ec3d	91	iser_task->data_copy[cmd_dir].copy_buf = mem;
6461f64a	92
2261ec3d	93	dev = iser_task->iser_conn->ib_conn->device->ib_device;
5180311f	94	dma_nents = ib_dma_map_sg(dev,
2261ec3d	95	&iser_task->data_copy[cmd_dir].sg_single,
5180311f RC	96	1,
	97	(cmd_dir == ISER_DIR_OUT) ?
	98	DMA_TO_DEVICE : DMA_FROM_DEVICE);
6461f64a OG	99	BUG_ON(dma_nents == 0);
6461f64a OG	100
2261ec3d	101	iser_task->data_copy[cmd_dir].dma_nents = dma_nents;
6461f64a OG	102	return 0;
	103	}
	104
	105	/**
	106	* iser_finalize_rdma_unaligned_sg
	107	*/
2261ec3d	108	void iser_finalize_rdma_unaligned_sg(struct iscsi_iser_task *iser_task,
6461f64a OG	109	enum iser_data_dir cmd_dir)
6461f64a OG	110	{
5180311f	111	struct ib_device *dev;
6461f64a OG	112	struct iser_data_buf *mem_copy;
	113	unsigned long cmd_data_len;
	114
2261ec3d MC	115	dev = iser_task->iser_conn->ib_conn->device->ib_device;
2261ec3d MC	116	mem_copy = &iser_task->data_copy[cmd_dir];
6461f64a	117
5180311f RC	118	ib_dma_unmap_sg(dev, &mem_copy->sg_single, 1,
	119	(cmd_dir == ISER_DIR_OUT) ?
	120	DMA_TO_DEVICE : DMA_FROM_DEVICE);
6461f64a OG	121
	122	if (cmd_dir == ISER_DIR_IN) {
	123	char *mem;
53d412fc	124	struct scatterlist sgl, sg;
6461f64a OG	125	unsigned char p, to;
	126	unsigned int sg_size;
	127	int i;
	128
	129	/* copy back read RDMA to unaligned sg */
	130	mem = mem_copy->copy_buf;
	131
2261ec3d MC	132	sgl = (struct scatterlist *)iser_task->data[ISER_DIR_IN].buf;
2261ec3d MC	133	sg_size = iser_task->data[ISER_DIR_IN].size;
6461f64a	134
53d412fc JA	135	p = mem;
53d412fc JA	136	for_each_sg(sgl, sg, sg_size, i) {
2a156d09	137	to = kmap_atomic(sg_page(sg));
53d412fc	138	memcpy(to + sg->offset,
6461f64a	139	p,
53d412fc	140	sg->length);
2a156d09	141	kunmap_atomic(to);
53d412fc	142	p += sg->length;
6461f64a OG	143	}
	144	}
	145
2261ec3d	146	cmd_data_len = iser_task->data[cmd_dir].data_len;
6461f64a OG	147
	148	if (cmd_data_len > ISER_KMALLOC_THRESHOLD)
	149	free_pages((unsigned long)mem_copy->copy_buf,
f0d1b0b3	150	ilog2(roundup_pow_of_two(cmd_data_len)) - PAGE_SHIFT);
6461f64a OG	151	else
	152	kfree(mem_copy->copy_buf);
	153
	154	mem_copy->copy_buf = NULL;
	155	}
	156
c1ccaf24 OG	157	#define IS_4K_ALIGNED(addr) ((((unsigned long)addr) & ~MASK_4K) == 0)
c1ccaf24 OG	158
6461f64a OG	159	/**
	160	* iser_sg_to_page_vec - Translates scatterlist entries to physical addresses
	161	* and returns the length of resulting physical address array (may be less than
	162	* the original due to possible compaction).
	163	*
	164	* we build a "page vec" under the assumption that the SG meets the RDMA
	165	* alignment requirements. Other then the first and last SG elements, all
	166	* the "internal" elements can be compacted into a list whose elements are
	167	* dma addresses of physical pages. The code supports also the weird case
	168	* where --few fragments of the same page-- are present in the SG as
	169	* consecutive elements. Also, it handles one entry SG.
	170	*/
c1ccaf24	171
6461f64a	172	static int iser_sg_to_page_vec(struct iser_data_buf *data,
919fc274 SG	173	struct ib_device ibdev, u64 pages,
919fc274 SG	174	int offset, int data_size)
6461f64a	175	{
c1ccaf24 OG	176	struct scatterlist sg, sgl = (struct scatterlist *)data->buf;
c1ccaf24 OG	177	u64 start_addr, end_addr, page, chunk_start = 0;
6461f64a	178	unsigned long total_sz = 0;
c1ccaf24 OG	179	unsigned int dma_len;
c1ccaf24 OG	180	int i, new_chunk, cur_page, last_ent = data->dma_nents - 1;
6461f64a OG	181
6461f64a OG	182	/* compute the offset of first element */
919fc274	183	*offset = (u64) sgl[0].offset & ~MASK_4K;
6461f64a	184
c1ccaf24 OG	185	new_chunk = 1;
c1ccaf24 OG	186	cur_page = 0;
53d412fc	187	for_each_sg(sgl, sg, data->dma_nents, i) {
c1ccaf24 OG	188	start_addr = ib_sg_dma_address(ibdev, sg);
	189	if (new_chunk)
	190	chunk_start = start_addr;
	191	dma_len = ib_sg_dma_len(ibdev, sg);
	192	end_addr = start_addr + dma_len;
5180311f	193	total_sz += dma_len;
6461f64a	194
c1ccaf24 OG	195	/* collect page fragments until aligned or end of SG list */
	196	if (!IS_4K_ALIGNED(end_addr) && i < last_ent) {
	197	new_chunk = 0;
	198	continue;
6461f64a	199	}
c1ccaf24 OG	200	new_chunk = 1;
	201
	202	/* address of the first page in the contiguous chunk;
	203	masking relevant for the very first SG entry,
	204	which might be unaligned */
	205	page = chunk_start & MASK_4K;
	206	do {
919fc274	207	pages[cur_page++] = page;
8dfa0876	208	page += SIZE_4K;
c1ccaf24	209	} while (page < end_addr);
6461f64a	210	}
c1ccaf24	211
919fc274 SG	212	*data_size = total_sz;
	213	iser_dbg("page_vec->data_size:%d cur_page %d\n",
	214	*data_size, cur_page);
6461f64a OG	215	return cur_page;
	216	}
	217
6461f64a OG	218
	219	/**
	220	* iser_data_buf_aligned_len - Tries to determine the maximal correctly aligned
	221	* for RDMA sub-list of a scatter-gather list of memory buffers, and returns
	222	* the number of entries which are aligned correctly. Supports the case where
	223	* consecutive SG elements are actually fragments of the same physcial page.
	224	*/
c1ccaf24 OG	225	static int iser_data_buf_aligned_len(struct iser_data_buf *data,
c1ccaf24 OG	226	struct ib_device *ibdev)
6461f64a	227	{
c1ccaf24 OG	228	struct scatterlist sgl, sg, *next_sg = NULL;
	229	u64 start_addr, end_addr;
	230	int i, ret_len, start_check = 0;
	231
	232	if (data->dma_nents == 1)
	233	return 1;
6461f64a	234
53d412fc	235	sgl = (struct scatterlist *)data->buf;
c1ccaf24	236	start_addr = ib_sg_dma_address(ibdev, sgl);
6461f64a	237
53d412fc	238	for_each_sg(sgl, sg, data->dma_nents, i) {
c1ccaf24 OG	239	if (start_check && !IS_4K_ALIGNED(start_addr))
	240	break;
	241
	242	next_sg = sg_next(sg);
	243	if (!next_sg)
	244	break;
	245
	246	end_addr = start_addr + ib_sg_dma_len(ibdev, sg);
	247	start_addr = ib_sg_dma_address(ibdev, next_sg);
	248
	249	if (end_addr == start_addr) {
	250	start_check = 0;
	251	continue;
	252	} else
	253	start_check = 1;
	254
	255	if (!IS_4K_ALIGNED(end_addr))
	256	break;
6461f64a	257	}
c1ccaf24	258	ret_len = (next_sg) ? i : i+1;
6461f64a OG	259	iser_dbg("Found %d aligned entries out of %d in sg:0x%p\n",
	260	ret_len, data->dma_nents, data);
	261	return ret_len;
	262	}
	263
5180311f RC	264	static void iser_data_buf_dump(struct iser_data_buf *data,
5180311f RC	265	struct ib_device *ibdev)
6461f64a	266	{
53d412fc JA	267	struct scatterlist sgl = (struct scatterlist )data->buf;
53d412fc JA	268	struct scatterlist *sg;
6461f64a OG	269	int i;
6461f64a OG	270
53d412fc	271	for_each_sg(sgl, sg, data->dma_nents, i)
f91424cf	272	iser_dbg("sg[%d] dma_addr:0x%lX page:0x%p "
e981f1d4	273	"off:0x%x sz:0x%x dma_len:0x%x\n",
53d412fc	274	i, (unsigned long)ib_sg_dma_address(ibdev, sg),
45711f1a	275	sg_page(sg), sg->offset,
53d412fc	276	sg->length, ib_sg_dma_len(ibdev, sg));
6461f64a OG	277	}
	278
	279	static void iser_dump_page_vec(struct iser_page_vec *page_vec)
	280	{
	281	int i;
	282
	283	iser_err("page vec length %d data size %d\n",
	284	page_vec->length, page_vec->data_size);
	285	for (i = 0; i < page_vec->length; i++)
	286	iser_err("%d %lx\n",i,(unsigned long)page_vec->pages[i]);
	287	}
	288
	289	static void iser_page_vec_build(struct iser_data_buf *data,
5180311f RC	290	struct iser_page_vec *page_vec,
5180311f RC	291	struct ib_device *ibdev)
6461f64a OG	292	{
	293	int page_vec_len = 0;
	294
	295	page_vec->length = 0;
	296	page_vec->offset = 0;
	297
	298	iser_dbg("Translating sg sz: %d\n", data->dma_nents);
919fc274 SG	299	page_vec_len = iser_sg_to_page_vec(data, ibdev, page_vec->pages,
	300	&page_vec->offset,
	301	&page_vec->data_size);
	302	iser_dbg("sg len %d page_vec_len %d\n", data->dma_nents, page_vec_len);
6461f64a OG	303
	304	page_vec->length = page_vec_len;
	305
8dfa0876	306	if (page_vec_len * SIZE_4K < page_vec->data_size) {
6461f64a	307	iser_err("page_vec too short to hold this SG\n");
5180311f	308	iser_data_buf_dump(data, ibdev);
6461f64a OG	309	iser_dump_page_vec(page_vec);
	310	BUG();
	311	}
	312	}
	313
2261ec3d MC	314	int iser_dma_map_task_data(struct iscsi_iser_task *iser_task,
	315	struct iser_data_buf *data,
	316	enum iser_data_dir iser_dir,
	317	enum dma_data_direction dma_dir)
74a20780	318	{
5180311f	319	struct ib_device *dev;
74a20780	320
2261ec3d MC	321	iser_task->dir[iser_dir] = 1;
2261ec3d MC	322	dev = iser_task->iser_conn->ib_conn->device->ib_device;
74a20780	323
5180311f	324	data->dma_nents = ib_dma_map_sg(dev, data->buf, data->size, dma_dir);
74a20780 EZ	325	if (data->dma_nents == 0) {
	326	iser_err("dma_map_sg failed!!!\n");
	327	return -EINVAL;
	328	}
	329	return 0;
	330	}
	331
2261ec3d	332	void iser_dma_unmap_task_data(struct iscsi_iser_task *iser_task)
74a20780	333	{
5180311f	334	struct ib_device *dev;
74a20780 EZ	335	struct iser_data_buf *data;
74a20780 EZ	336
2261ec3d	337	dev = iser_task->iser_conn->ib_conn->device->ib_device;
74a20780	338
2261ec3d MC	339	if (iser_task->dir[ISER_DIR_IN]) {
2261ec3d MC	340	data = &iser_task->data[ISER_DIR_IN];
5180311f	341	ib_dma_unmap_sg(dev, data->buf, data->size, DMA_FROM_DEVICE);
74a20780 EZ	342	}
74a20780 EZ	343
2261ec3d MC	344	if (iser_task->dir[ISER_DIR_OUT]) {
2261ec3d MC	345	data = &iser_task->data[ISER_DIR_OUT];
5180311f	346	ib_dma_unmap_sg(dev, data->buf, data->size, DMA_TO_DEVICE);
74a20780 EZ	347	}
	348	}
	349
919fc274 SG	350	static int fall_to_bounce_buf(struct iscsi_iser_task *iser_task,
	351	struct ib_device *ibdev,
	352	enum iser_data_dir cmd_dir,
	353	int aligned_len)
	354	{
	355	struct iscsi_conn *iscsi_conn = iser_task->iser_conn->iscsi_conn;
	356	struct iser_data_buf *mem = &iser_task->data[cmd_dir];
	357
	358	iscsi_conn->fmr_unalign_cnt++;
	359	iser_warn("rdma alignment violation (%d/%d aligned) or FMR not supported\n",
	360	aligned_len, mem->size);
	361
	362	if (iser_debug_level > 0)
	363	iser_data_buf_dump(mem, ibdev);
	364
	365	/* unmap the command data before accessing it */
	366	iser_dma_unmap_task_data(iser_task);
	367
	368	/* allocate copy buf, if we are writing, copy the */
	369	/* unaligned scatterlist, dma map the copy */
	370	if (iser_start_rdma_unaligned_sg(iser_task, cmd_dir) != 0)
	371	return -ENOMEM;
	372
	373	return 0;
	374	}
	375
6461f64a	376	/**
e657571b SG	377	* iser_reg_rdma_mem_fmr - Registers memory intended for RDMA,
e657571b SG	378	* using FMR (if possible) obtaining rkey and va
6461f64a OG	379	*
	380	* returns 0 on success, errno code on failure
	381	*/
e657571b SG	382	int iser_reg_rdma_mem_fmr(struct iscsi_iser_task *iser_task,
e657571b SG	383	enum iser_data_dir cmd_dir)
6461f64a	384	{
2261ec3d	385	struct iser_conn *ib_conn = iser_task->iser_conn->ib_conn;
d8111028	386	struct iser_device *device = ib_conn->device;
5180311f	387	struct ib_device *ibdev = device->ib_device;
2261ec3d	388	struct iser_data_buf *mem = &iser_task->data[cmd_dir];
6461f64a OG	389	struct iser_regd_buf *regd_buf;
	390	int aligned_len;
	391	int err;
e981f1d4	392	int i;
d8111028	393	struct scatterlist *sg;
6461f64a	394
2261ec3d	395	regd_buf = &iser_task->rdma_regd[cmd_dir];
6461f64a	396
5180311f	397	aligned_len = iser_data_buf_aligned_len(mem, ibdev);
5525d210	398	if (aligned_len != mem->dma_nents \|\|
e657571b	399	(!ib_conn->fastreg.fmr.pool && mem->dma_nents > 1)) {
919fc274 SG	400	err = fall_to_bounce_buf(iser_task, ibdev,
	401	cmd_dir, aligned_len);
	402	if (err) {
	403	iser_err("failed to allocate bounce buffer\n");
	404	return err;
	405	}
2261ec3d	406	mem = &iser_task->data_copy[cmd_dir];
6461f64a OG	407	}
6461f64a OG	408
d8111028 EZ	409	/* if there a single dma entry, FMR is not needed */
	410	if (mem->dma_nents == 1) {
	411	sg = (struct scatterlist *)mem->buf;
	412
	413	regd_buf->reg.lkey = device->mr->lkey;
	414	regd_buf->reg.rkey = device->mr->rkey;
5180311f RC	415	regd_buf->reg.len = ib_sg_dma_len(ibdev, &sg[0]);
5180311f RC	416	regd_buf->reg.va = ib_sg_dma_address(ibdev, &sg[0]);
d8111028 EZ	417	regd_buf->reg.is_fmr = 0;
	418
	419	iser_dbg("PHYSICAL Mem.register: lkey: 0x%08X rkey: 0x%08X "
	420	"va: 0x%08lX sz: %ld]\n",
	421	(unsigned int)regd_buf->reg.lkey,
	422	(unsigned int)regd_buf->reg.rkey,
	423	(unsigned long)regd_buf->reg.va,
	424	(unsigned long)regd_buf->reg.len);
	425	} else { /* use FMR for multiple dma entries */
e657571b SG	426	iser_page_vec_build(mem, ib_conn->fastreg.fmr.page_vec, ibdev);
	427	err = iser_reg_page_vec(ib_conn, ib_conn->fastreg.fmr.page_vec,
	428	&regd_buf->reg);
819a0873	429	if (err && err != -EAGAIN) {
5180311f	430	iser_data_buf_dump(mem, ibdev);
2261ec3d MC	431	iser_err("mem->dma_nents = %d (dlength = 0x%x)\n",
	432	mem->dma_nents,
	433	ntoh24(iser_task->desc.iscsi_header.dlength));
d8111028	434	iser_err("page_vec: data_size = 0x%x, length = %d, offset = 0x%x\n",
e657571b SG	435	ib_conn->fastreg.fmr.page_vec->data_size,
	436	ib_conn->fastreg.fmr.page_vec->length,
	437	ib_conn->fastreg.fmr.page_vec->offset);
	438	for (i = 0; i < ib_conn->fastreg.fmr.page_vec->length; i++)
d8111028	439	iser_err("page_vec[%d] = 0x%llx\n", i,
e657571b	440	(unsigned long long) ib_conn->fastreg.fmr.page_vec->pages[i]);
e981f1d4	441	}
450d1e40 OG	442	if (err)
450d1e40 OG	443	return err;
e981f1d4	444	}
6461f64a OG	445	return 0;
6461f64a OG	446	}