[deliverable/linux.git] / drivers / infiniband / ulp / iser / iser_memory.c

/*
 * Copyright (c) 2004, 2005, 2006 Voltaire, 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.
 *
 * $Id: iser_memory.c 6964 2006-05-07 11:11:43Z ogerlitz $
 */
#include <linux/module.h>
#include <linux/kernel.h>
#include <linux/slab.h>
#include <linux/mm.h>
#include <linux/highmem.h>
#include <asm/io.h>
#include <asm/scatterlist.h>
#include <linux/scatterlist.h>

#include "iscsi_iser.h"

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

/**
 * Decrements the reference count for the
 * registered buffer & releases it
 *
 * returns 0 if released, 1 if deferred
 */
int iser_regd_buff_release(struct iser_regd_buf *regd_buf)
{
	struct ib_device *dev;

	if ((atomic_read(&regd_buf->ref_count) == 0) ||
	    atomic_dec_and_test(&regd_buf->ref_count)) {
		/* if we used the dma mr, unreg is just NOP */
		if (regd_buf->reg.is_fmr)
			iser_unreg_mem(&regd_buf->reg);

		if (regd_buf->dma_addr) {
			dev = regd_buf->device->ib_device;
			ib_dma_unmap_single(dev,
					 regd_buf->dma_addr,
					 regd_buf->data_size,
					 regd_buf->direction);
		}
		/* else this regd buf is associated with task which we */
		/* dma_unmap_single/sg later */
		return 0;
	} else {
		iser_dbg("Release deferred, regd.buff: 0x%p\n", regd_buf);
		return 1;
	}
}

/**
 * iser_reg_single - fills registered buffer descriptor with
 *		     registration information
 */
void iser_reg_single(struct iser_device *device,
		     struct iser_regd_buf *regd_buf,
		     enum dma_data_direction direction)
{
	u64 dma_addr;

	dma_addr = ib_dma_map_single(device->ib_device,
				     regd_buf->virt_addr,
				     regd_buf->data_size, direction);
	BUG_ON(ib_dma_mapping_error(device->ib_device, dma_addr));

	regd_buf->reg.lkey = device->mr->lkey;
	regd_buf->reg.len  = regd_buf->data_size;
	regd_buf->reg.va   = dma_addr;
	regd_buf->reg.is_fmr = 0;

	regd_buf->dma_addr  = dma_addr;
	regd_buf->direction = direction;
}

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

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

	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 *sg = (struct scatterlist *)data->buf;
		int i;
		char *p, *from;

		for (p = mem, i = 0; i < data->size; i++) {
			from = kmap_atomic(sg[i].page, KM_USER0);
			memcpy(p,
			       from + sg[i].offset,
			       sg[i].length);
			kunmap_atomic(from, KM_USER0);
			p += sg[i].length;
		}
	}

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

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

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

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

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

	dev = iser_ctask->iser_conn->ib_conn->device->ib_device;
	mem_copy = &iser_ctask->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 *sg;
		unsigned char *p, *to;
		unsigned int sg_size;
		int i;

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

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

		for (p = mem, i = 0; i < sg_size; i++){
			to = kmap_atomic(sg[i].page, KM_SOFTIRQ0);
			memcpy(to + sg[i].offset,
			       p,
			       sg[i].length);
			kunmap_atomic(to, KM_SOFTIRQ0);
			p += sg[i].length;
		}
	}

	cmd_data_len = iser_ctask->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;
}

/**
 * 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 iser_page_vec *page_vec,
			       struct ib_device *ibdev)
{
	struct scatterlist *sg = (struct scatterlist *)data->buf;
	u64 first_addr, last_addr, page;
	int end_aligned;
	unsigned int cur_page = 0;
	unsigned long total_sz = 0;
	int i;

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

	for (i = 0; i < data->dma_nents; i++) {
		unsigned int dma_len = ib_sg_dma_len(ibdev, &sg[i]);

		total_sz += dma_len;

		first_addr = ib_sg_dma_address(ibdev, &sg[i]);
		last_addr  = first_addr + dma_len;

		end_aligned   = !(last_addr  & ~MASK_4K);

		/* continue to collect page fragments till aligned or SG ends */
		while (!end_aligned && (i + 1 < data->dma_nents)) {
			i++;
			dma_len = ib_sg_dma_len(ibdev, &sg[i]);
			total_sz += dma_len;
			last_addr = ib_sg_dma_address(ibdev, &sg[i]) + dma_len;
			end_aligned = !(last_addr  & ~MASK_4K);
		}

		/* handle the 1st page in the 1st DMA element */
		if (cur_page == 0) {
			page = first_addr & MASK_4K;
			page_vec->pages[cur_page] = page;
			cur_page++;
			page += SIZE_4K;
		} else
			page = first_addr;

		for (; page < last_addr; page += SIZE_4K) {
			page_vec->pages[cur_page] = page;
			cur_page++;
		}

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

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

/**
 * 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 unsigned int iser_data_buf_aligned_len(struct iser_data_buf *data,
					      struct ib_device *ibdev)
{
	struct scatterlist *sg;
	u64 end_addr, next_addr;
	int i, cnt;
	unsigned int ret_len = 0;

	sg = (struct scatterlist *)data->buf;

	for (cnt = 0, i = 0; i < data->dma_nents; i++, cnt++) {
		/* iser_dbg("Checking sg iobuf [%d]: phys=0x%08lX "
		   "offset: %ld sz: %ld\n", i,
		   (unsigned long)page_to_phys(sg[i].page),
		   (unsigned long)sg[i].offset,
		   (unsigned long)sg[i].length); */
		end_addr = ib_sg_dma_address(ibdev, &sg[i]) +
			   ib_sg_dma_len(ibdev, &sg[i]);
		/* iser_dbg("Checking sg iobuf end address "
		       "0x%08lX\n", end_addr); */
		if (i + 1 < data->dma_nents) {
			next_addr = ib_sg_dma_address(ibdev, &sg[i+1]);
			/* are i, i+1 fragments of the same page? */
			if (end_addr == next_addr)
				continue;
			else if (!IS_4K_ALIGNED(end_addr)) {
				ret_len = cnt + 1;
				break;
			}
		}
	}
	if (i == data->dma_nents)
		ret_len = cnt;	/* loop ended */
	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 *sg = (struct scatterlist *)data->buf;
	int i;

	for (i = 0; i < data->dma_nents; i++)
		iser_err("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[i]),
			 sg[i].page, sg[i].offset,
			 sg[i].length, ib_sg_dma_len(ibdev, &sg[i]));
}

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, page_vec, ibdev);
	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_cmd_task *iser_ctask,
			    struct iser_data_buf       *data,
			    enum   iser_data_dir       iser_dir,
			    enum   dma_data_direction  dma_dir)
{
	struct ib_device *dev;

	iser_ctask->dir[iser_dir] = 1;
	dev = iser_ctask->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_cmd_task *iser_ctask)
{
	struct ib_device *dev;
	struct iser_data_buf *data;

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

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

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

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

	regd_buf = &iser_ctask->rdma_regd[cmd_dir];

	aligned_len = iser_data_buf_aligned_len(mem, ibdev);
	if (aligned_len != mem->dma_nents) {
		iser_err("rdma alignment violation %d/%d aligned\n",
			 aligned_len, mem->size);
		iser_data_buf_dump(mem, ibdev);

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

		/* allocate copy buf, if we are writing, copy the */
		/* unaligned scatterlist, dma map the copy        */
		if (iser_start_rdma_unaligned_sg(iser_ctask, cmd_dir) != 0)
				return -ENOMEM;
		mem = &iser_ctask->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->page_vec, ibdev);
		err = iser_reg_page_vec(ib_conn, ib_conn->page_vec, &regd_buf->reg);
		if (err) {
			iser_data_buf_dump(mem, ibdev);
			iser_err("mem->dma_nents = %d (dlength = 0x%x)\n", mem->dma_nents,
				 ntoh24(iser_ctask->desc.iscsi_header.dlength));
			iser_err("page_vec: data_size = 0x%x, length = %d, offset = 0x%x\n",
				 ib_conn->page_vec->data_size, ib_conn->page_vec->length,
				 ib_conn->page_vec->offset);
			for (i=0 ; i<ib_conn->page_vec->length ; i++)
				iser_err("page_vec[%d] = 0x%llx\n", i,
					 (unsigned long long) ib_conn->page_vec->pages[i]);
			return err;
		}
	}

	/* take a reference on this regd buf such that it will not be released *
	 * (eg in send dto completion) before we get the scsi response         */
	atomic_inc(&regd_buf->ref_count);
	return 0;
}
Commit	Line	Data
6461f64a OG	1	/*
	2	* Copyright (c) 2004, 2005, 2006 Voltaire, Inc. 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	* $Id: iser_memory.c 6964 2006-05-07 11:11:43Z ogerlitz $
	33	*/
	34	#include <linux/module.h>
	35	#include <linux/kernel.h>
	36	#include <linux/slab.h>
	37	#include <linux/mm.h>
a1f8e7f7	38	#include <linux/highmem.h>
6461f64a OG	39	#include <asm/io.h>
	40	#include <asm/scatterlist.h>
	41	#include <linux/scatterlist.h>
	42
	43	#include "iscsi_iser.h"
	44
	45	#define ISER_KMALLOC_THRESHOLD 0x20000 /* 128K - kmalloc limit */
8dfa0876	46
6461f64a OG	47	/**
	48	* Decrements the reference count for the
	49	* registered buffer & releases it
	50	*
	51	* returns 0 if released, 1 if deferred
	52	*/
	53	int iser_regd_buff_release(struct iser_regd_buf *regd_buf)
	54	{
5180311f	55	struct ib_device *dev;
6461f64a OG	56
	57	if ((atomic_read(&regd_buf->ref_count) == 0) \|\|
	58	atomic_dec_and_test(&regd_buf->ref_count)) {
	59	/* if we used the dma mr, unreg is just NOP */
d8111028	60	if (regd_buf->reg.is_fmr)
6461f64a OG	61	iser_unreg_mem(&regd_buf->reg);
	62
	63	if (regd_buf->dma_addr) {
5180311f RC	64	dev = regd_buf->device->ib_device;
5180311f RC	65	ib_dma_unmap_single(dev,
6461f64a OG	66	regd_buf->dma_addr,
	67	regd_buf->data_size,
	68	regd_buf->direction);
	69	}
	70	/* else this regd buf is associated with task which we */
	71	/* dma_unmap_single/sg later */
	72	return 0;
	73	} else {
	74	iser_dbg("Release deferred, regd.buff: 0x%p\n", regd_buf);
	75	return 1;
	76	}
	77	}
	78
	79	/**
	80	* iser_reg_single - fills registered buffer descriptor with
	81	* registration information
	82	*/
	83	void iser_reg_single(struct iser_device *device,
	84	struct iser_regd_buf *regd_buf,
	85	enum dma_data_direction direction)
	86	{
5180311f	87	u64 dma_addr;
6461f64a	88
5180311f RC	89	dma_addr = ib_dma_map_single(device->ib_device,
	90	regd_buf->virt_addr,
	91	regd_buf->data_size, direction);
	92	BUG_ON(ib_dma_mapping_error(device->ib_device, dma_addr));
6461f64a OG	93
6461f64a OG	94	regd_buf->reg.lkey = device->mr->lkey;
6461f64a OG	95	regd_buf->reg.len = regd_buf->data_size;
6461f64a OG	96	regd_buf->reg.va = dma_addr;
d8111028	97	regd_buf->reg.is_fmr = 0;
6461f64a OG	98
	99	regd_buf->dma_addr = dma_addr;
	100	regd_buf->direction = direction;
	101	}
	102
	103	/**
	104	* iser_start_rdma_unaligned_sg
	105	*/
	106	int iser_start_rdma_unaligned_sg(struct iscsi_iser_cmd_task *iser_ctask,
	107	enum iser_data_dir cmd_dir)
	108	{
	109	int dma_nents;
5180311f	110	struct ib_device *dev;
6461f64a OG	111	char *mem = NULL;
	112	struct iser_data_buf *data = &iser_ctask->data[cmd_dir];
	113	unsigned long cmd_data_len = data->data_len;
	114
	115	if (cmd_data_len > ISER_KMALLOC_THRESHOLD)
	116	mem = (void *)__get_free_pages(GFP_NOIO,
f0d1b0b3	117	ilog2(roundup_pow_of_two(cmd_data_len)) - PAGE_SHIFT);
6461f64a OG	118	else
	119	mem = kmalloc(cmd_data_len, GFP_NOIO);
	120
	121	if (mem == NULL) {
	122	iser_err("Failed to allocate mem size %d %d for copying sglist\n",
	123	data->size,(int)cmd_data_len);
	124	return -ENOMEM;
	125	}
	126
	127	if (cmd_dir == ISER_DIR_OUT) {
	128	/* copy the unaligned sg the buffer which is used for RDMA */
	129	struct scatterlist sg = (struct scatterlist )data->buf;
	130	int i;
	131	char p, from;
	132
	133	for (p = mem, i = 0; i < data->size; i++) {
	134	from = kmap_atomic(sg[i].page, KM_USER0);
	135	memcpy(p,
	136	from + sg[i].offset,
	137	sg[i].length);
	138	kunmap_atomic(from, KM_USER0);
	139	p += sg[i].length;
	140	}
	141	}
	142
	143	sg_init_one(&iser_ctask->data_copy[cmd_dir].sg_single, mem, cmd_data_len);
	144	iser_ctask->data_copy[cmd_dir].buf =
	145	&iser_ctask->data_copy[cmd_dir].sg_single;
	146	iser_ctask->data_copy[cmd_dir].size = 1;
	147
	148	iser_ctask->data_copy[cmd_dir].copy_buf = mem;
	149
5180311f RC	150	dev = iser_ctask->iser_conn->ib_conn->device->ib_device;
	151	dma_nents = ib_dma_map_sg(dev,
	152	&iser_ctask->data_copy[cmd_dir].sg_single,
	153	1,
	154	(cmd_dir == ISER_DIR_OUT) ?
	155	DMA_TO_DEVICE : DMA_FROM_DEVICE);
6461f64a OG	156	BUG_ON(dma_nents == 0);
	157
	158	iser_ctask->data_copy[cmd_dir].dma_nents = dma_nents;
	159	return 0;
	160	}
	161
	162	/**
	163	* iser_finalize_rdma_unaligned_sg
	164	*/
	165	void iser_finalize_rdma_unaligned_sg(struct iscsi_iser_cmd_task *iser_ctask,
	166	enum iser_data_dir cmd_dir)
	167	{
5180311f	168	struct ib_device *dev;
6461f64a OG	169	struct iser_data_buf *mem_copy;
	170	unsigned long cmd_data_len;
	171
5180311f RC	172	dev = iser_ctask->iser_conn->ib_conn->device->ib_device;
5180311f RC	173	mem_copy = &iser_ctask->data_copy[cmd_dir];
6461f64a	174
5180311f RC	175	ib_dma_unmap_sg(dev, &mem_copy->sg_single, 1,
	176	(cmd_dir == ISER_DIR_OUT) ?
	177	DMA_TO_DEVICE : DMA_FROM_DEVICE);
6461f64a OG	178
	179	if (cmd_dir == ISER_DIR_IN) {
	180	char *mem;
	181	struct scatterlist *sg;
	182	unsigned char p, to;
	183	unsigned int sg_size;
	184	int i;
	185
	186	/* copy back read RDMA to unaligned sg */
	187	mem = mem_copy->copy_buf;
	188
	189	sg = (struct scatterlist *)iser_ctask->data[ISER_DIR_IN].buf;
	190	sg_size = iser_ctask->data[ISER_DIR_IN].size;
	191
	192	for (p = mem, i = 0; i < sg_size; i++){
	193	to = kmap_atomic(sg[i].page, KM_SOFTIRQ0);
	194	memcpy(to + sg[i].offset,
	195	p,
	196	sg[i].length);
	197	kunmap_atomic(to, KM_SOFTIRQ0);
	198	p += sg[i].length;
	199	}
	200	}
	201
	202	cmd_data_len = iser_ctask->data[cmd_dir].data_len;
	203
	204	if (cmd_data_len > ISER_KMALLOC_THRESHOLD)
	205	free_pages((unsigned long)mem_copy->copy_buf,
f0d1b0b3	206	ilog2(roundup_pow_of_two(cmd_data_len)) - PAGE_SHIFT);
6461f64a OG	207	else
	208	kfree(mem_copy->copy_buf);
	209
	210	mem_copy->copy_buf = NULL;
	211	}
	212
	213	/**
	214	* iser_sg_to_page_vec - Translates scatterlist entries to physical addresses
	215	* and returns the length of resulting physical address array (may be less than
	216	* the original due to possible compaction).
	217	*
	218	* we build a "page vec" under the assumption that the SG meets the RDMA
	219	* alignment requirements. Other then the first and last SG elements, all
	220	* the "internal" elements can be compacted into a list whose elements are
	221	* dma addresses of physical pages. The code supports also the weird case
	222	* where --few fragments of the same page-- are present in the SG as
	223	* consecutive elements. Also, it handles one entry SG.
	224	*/
	225	static int iser_sg_to_page_vec(struct iser_data_buf *data,
5180311f RC	226	struct iser_page_vec *page_vec,
5180311f RC	227	struct ib_device *ibdev)
6461f64a OG	228	{
6461f64a OG	229	struct scatterlist sg = (struct scatterlist )data->buf;
5180311f	230	u64 first_addr, last_addr, page;
dee234f4	231	int end_aligned;
6461f64a OG	232	unsigned int cur_page = 0;
	233	unsigned long total_sz = 0;
	234	int i;
	235
	236	/* compute the offset of first element */
8dfa0876	237	page_vec->offset = (u64) sg[0].offset & ~MASK_4K;
6461f64a OG	238
6461f64a OG	239	for (i = 0; i < data->dma_nents; i++) {
5180311f RC	240	unsigned int dma_len = ib_sg_dma_len(ibdev, &sg[i]);
	241
	242	total_sz += dma_len;
6461f64a	243
5180311f RC	244	first_addr = ib_sg_dma_address(ibdev, &sg[i]);
5180311f RC	245	last_addr = first_addr + dma_len;
6461f64a	246
8dfa0876	247	end_aligned = !(last_addr & ~MASK_4K);
6461f64a OG	248
	249	/* continue to collect page fragments till aligned or SG ends */
	250	while (!end_aligned && (i + 1 < data->dma_nents)) {
	251	i++;
5180311f RC	252	dma_len = ib_sg_dma_len(ibdev, &sg[i]);
	253	total_sz += dma_len;
	254	last_addr = ib_sg_dma_address(ibdev, &sg[i]) + dma_len;
8dfa0876	255	end_aligned = !(last_addr & ~MASK_4K);
6461f64a OG	256	}
6461f64a OG	257
8dfa0876 EZ	258	/* handle the 1st page in the 1st DMA element */
	259	if (cur_page == 0) {
	260	page = first_addr & MASK_4K;
	261	page_vec->pages[cur_page] = page;
	262	cur_page++;
	263	page += SIZE_4K;
	264	} else
	265	page = first_addr;
	266
	267	for (; page < last_addr; page += SIZE_4K) {
	268	page_vec->pages[cur_page] = page;
	269	cur_page++;
	270	}
6461f64a OG	271
	272	}
	273	page_vec->data_size = total_sz;
	274	iser_dbg("page_vec->data_size:%d cur_page %d\n", page_vec->data_size,cur_page);
	275	return cur_page;
	276	}
	277
8dfa0876	278	#define IS_4K_ALIGNED(addr) ((((unsigned long)addr) & ~MASK_4K) == 0)
6461f64a OG	279
	280	/**
	281	* iser_data_buf_aligned_len - Tries to determine the maximal correctly aligned
	282	* for RDMA sub-list of a scatter-gather list of memory buffers, and returns
	283	* the number of entries which are aligned correctly. Supports the case where
	284	* consecutive SG elements are actually fragments of the same physcial page.
	285	*/
5180311f RC	286	static unsigned int iser_data_buf_aligned_len(struct iser_data_buf *data,
5180311f RC	287	struct ib_device *ibdev)
6461f64a OG	288	{
6461f64a OG	289	struct scatterlist *sg;
5180311f	290	u64 end_addr, next_addr;
6461f64a OG	291	int i, cnt;
	292	unsigned int ret_len = 0;
	293
	294	sg = (struct scatterlist *)data->buf;
	295
	296	for (cnt = 0, i = 0; i < data->dma_nents; i++, cnt++) {
	297	/* iser_dbg("Checking sg iobuf [%d]: phys=0x%08lX "
	298	"offset: %ld sz: %ld\n", i,
	299	(unsigned long)page_to_phys(sg[i].page),
	300	(unsigned long)sg[i].offset,
	301	(unsigned long)sg[i].length); */
5180311f RC	302	end_addr = ib_sg_dma_address(ibdev, &sg[i]) +
5180311f RC	303	ib_sg_dma_len(ibdev, &sg[i]);
6461f64a OG	304	/* iser_dbg("Checking sg iobuf end address "
	305	"0x%08lX\n", end_addr); */
	306	if (i + 1 < data->dma_nents) {
5180311f	307	next_addr = ib_sg_dma_address(ibdev, &sg[i+1]);
6461f64a OG	308	/* are i, i+1 fragments of the same page? */
	309	if (end_addr == next_addr)
	310	continue;
	311	else if (!IS_4K_ALIGNED(end_addr)) {
	312	ret_len = cnt + 1;
	313	break;
	314	}
	315	}
	316	}
	317	if (i == data->dma_nents)
	318	ret_len = cnt; /* loop ended */
	319	iser_dbg("Found %d aligned entries out of %d in sg:0x%p\n",
	320	ret_len, data->dma_nents, data);
	321	return ret_len;
	322	}
	323
5180311f RC	324	static void iser_data_buf_dump(struct iser_data_buf *data,
5180311f RC	325	struct ib_device *ibdev)
6461f64a OG	326	{
	327	struct scatterlist sg = (struct scatterlist )data->buf;
	328	int i;
	329
e981f1d4	330	for (i = 0; i < data->dma_nents; i++)
6461f64a	331	iser_err("sg[%d] dma_addr:0x%lX page:0x%p "
e981f1d4	332	"off:0x%x sz:0x%x dma_len:0x%x\n",
5180311f	333	i, (unsigned long)ib_sg_dma_address(ibdev, &sg[i]),
6461f64a	334	sg[i].page, sg[i].offset,
5180311f	335	sg[i].length, ib_sg_dma_len(ibdev, &sg[i]));
6461f64a OG	336	}
	337
	338	static void iser_dump_page_vec(struct iser_page_vec *page_vec)
	339	{
	340	int i;
	341
	342	iser_err("page vec length %d data size %d\n",
	343	page_vec->length, page_vec->data_size);
	344	for (i = 0; i < page_vec->length; i++)
	345	iser_err("%d %lx\n",i,(unsigned long)page_vec->pages[i]);
	346	}
	347
	348	static void iser_page_vec_build(struct iser_data_buf *data,
5180311f RC	349	struct iser_page_vec *page_vec,
5180311f RC	350	struct ib_device *ibdev)
6461f64a OG	351	{
	352	int page_vec_len = 0;
	353
	354	page_vec->length = 0;
	355	page_vec->offset = 0;
	356
	357	iser_dbg("Translating sg sz: %d\n", data->dma_nents);
5180311f	358	page_vec_len = iser_sg_to_page_vec(data, page_vec, ibdev);
6461f64a OG	359	iser_dbg("sg len %d page_vec_len %d\n", data->dma_nents,page_vec_len);
	360
	361	page_vec->length = page_vec_len;
	362
8dfa0876	363	if (page_vec_len * SIZE_4K < page_vec->data_size) {
6461f64a	364	iser_err("page_vec too short to hold this SG\n");
5180311f	365	iser_data_buf_dump(data, ibdev);
6461f64a OG	366	iser_dump_page_vec(page_vec);
	367	BUG();
	368	}
	369	}
	370
74a20780 EZ	371	int iser_dma_map_task_data(struct iscsi_iser_cmd_task *iser_ctask,
	372	struct iser_data_buf *data,
	373	enum iser_data_dir iser_dir,
	374	enum dma_data_direction dma_dir)
	375	{
5180311f	376	struct ib_device *dev;
74a20780 EZ	377
74a20780 EZ	378	iser_ctask->dir[iser_dir] = 1;
5180311f	379	dev = iser_ctask->iser_conn->ib_conn->device->ib_device;
74a20780	380
5180311f	381	data->dma_nents = ib_dma_map_sg(dev, data->buf, data->size, dma_dir);
74a20780 EZ	382	if (data->dma_nents == 0) {
	383	iser_err("dma_map_sg failed!!!\n");
	384	return -EINVAL;
	385	}
	386	return 0;
	387	}
	388
	389	void iser_dma_unmap_task_data(struct iscsi_iser_cmd_task *iser_ctask)
	390	{
5180311f	391	struct ib_device *dev;
74a20780 EZ	392	struct iser_data_buf *data;
74a20780 EZ	393
5180311f	394	dev = iser_ctask->iser_conn->ib_conn->device->ib_device;
74a20780 EZ	395
	396	if (iser_ctask->dir[ISER_DIR_IN]) {
	397	data = &iser_ctask->data[ISER_DIR_IN];
5180311f	398	ib_dma_unmap_sg(dev, data->buf, data->size, DMA_FROM_DEVICE);
74a20780 EZ	399	}
	400
	401	if (iser_ctask->dir[ISER_DIR_OUT]) {
	402	data = &iser_ctask->data[ISER_DIR_OUT];
5180311f	403	ib_dma_unmap_sg(dev, data->buf, data->size, DMA_TO_DEVICE);
74a20780 EZ	404	}
	405	}
	406
6461f64a OG	407	/**
	408	* iser_reg_rdma_mem - Registers memory intended for RDMA,
	409	* obtaining rkey and va
	410	*
	411	* returns 0 on success, errno code on failure
	412	*/
	413	int iser_reg_rdma_mem(struct iscsi_iser_cmd_task *iser_ctask,
	414	enum iser_data_dir cmd_dir)
	415	{
	416	struct iser_conn *ib_conn = iser_ctask->iser_conn->ib_conn;
d8111028	417	struct iser_device *device = ib_conn->device;
5180311f	418	struct ib_device *ibdev = device->ib_device;
6461f64a OG	419	struct iser_data_buf *mem = &iser_ctask->data[cmd_dir];
	420	struct iser_regd_buf *regd_buf;
	421	int aligned_len;
	422	int err;
e981f1d4	423	int i;
d8111028	424	struct scatterlist *sg;
6461f64a OG	425
	426	regd_buf = &iser_ctask->rdma_regd[cmd_dir];
	427
5180311f	428	aligned_len = iser_data_buf_aligned_len(mem, ibdev);
777a71dd	429	if (aligned_len != mem->dma_nents) {
6461f64a OG	430	iser_err("rdma alignment violation %d/%d aligned\n",
6461f64a OG	431	aligned_len, mem->size);
5180311f	432	iser_data_buf_dump(mem, ibdev);
74a20780 EZ	433
	434	/* unmap the command data before accessing it */
	435	iser_dma_unmap_task_data(iser_ctask);
	436
6461f64a OG	437	/* allocate copy buf, if we are writing, copy the */
	438	/* unaligned scatterlist, dma map the copy */
	439	if (iser_start_rdma_unaligned_sg(iser_ctask, cmd_dir) != 0)
	440	return -ENOMEM;
	441	mem = &iser_ctask->data_copy[cmd_dir];
	442	}
	443
d8111028 EZ	444	/* if there a single dma entry, FMR is not needed */
	445	if (mem->dma_nents == 1) {
	446	sg = (struct scatterlist *)mem->buf;
	447
	448	regd_buf->reg.lkey = device->mr->lkey;
	449	regd_buf->reg.rkey = device->mr->rkey;
5180311f RC	450	regd_buf->reg.len = ib_sg_dma_len(ibdev, &sg[0]);
5180311f RC	451	regd_buf->reg.va = ib_sg_dma_address(ibdev, &sg[0]);
d8111028 EZ	452	regd_buf->reg.is_fmr = 0;
	453
	454	iser_dbg("PHYSICAL Mem.register: lkey: 0x%08X rkey: 0x%08X "
	455	"va: 0x%08lX sz: %ld]\n",
	456	(unsigned int)regd_buf->reg.lkey,
	457	(unsigned int)regd_buf->reg.rkey,
	458	(unsigned long)regd_buf->reg.va,
	459	(unsigned long)regd_buf->reg.len);
	460	} else { /* use FMR for multiple dma entries */
5180311f	461	iser_page_vec_build(mem, ib_conn->page_vec, ibdev);
d8111028 EZ	462	err = iser_reg_page_vec(ib_conn, ib_conn->page_vec, &regd_buf->reg);
d8111028 EZ	463	if (err) {
5180311f	464	iser_data_buf_dump(mem, ibdev);
d8111028 EZ	465	iser_err("mem->dma_nents = %d (dlength = 0x%x)\n", mem->dma_nents,
	466	ntoh24(iser_ctask->desc.iscsi_header.dlength));
	467	iser_err("page_vec: data_size = 0x%x, length = %d, offset = 0x%x\n",
	468	ib_conn->page_vec->data_size, ib_conn->page_vec->length,
	469	ib_conn->page_vec->offset);
	470	for (i=0 ; i<ib_conn->page_vec->length ; i++)
	471	iser_err("page_vec[%d] = 0x%llx\n", i,
	472	(unsigned long long) ib_conn->page_vec->pages[i]);
	473	return err;
e981f1d4	474	}
e981f1d4	475	}
6461f64a OG	476
	477	/* take a reference on this regd buf such that it will not be released *
	478	* (eg in send dto completion) before we get the scsi response */
	479	atomic_inc(&regd_buf->ref_count);
	480	return 0;
	481	}