[mirror_ubuntu-eoan-kernel.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 <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
 */
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_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 *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), KM_USER0);
			memcpy(p,
			       from + sg->offset,
			       sg->length);
			kunmap_atomic(from, KM_USER0);
			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), KM_SOFTIRQ0);
			memcpy(to + sg->offset,
			       p,
			       sg->length);
			kunmap_atomic(to, KM_SOFTIRQ0);
			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;
}

/**
 * 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 *sgl = (struct scatterlist *)data->buf;
	struct scatterlist *sg;
	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) sgl[0].offset & ~MASK_4K;

	for_each_sg(sgl, sg, data->dma_nents, i) {
		unsigned int dma_len = ib_sg_dma_len(ibdev, sg);

		total_sz += dma_len;

		first_addr = ib_sg_dma_address(ibdev, sg);
		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)) {
			sg = sg_next(sg);
			i++;
			dma_len = ib_sg_dma_len(ibdev, sg);
			total_sz += dma_len;
			last_addr = ib_sg_dma_address(ibdev, sg) + 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 *sgl, *sg;
	u64 end_addr, next_addr;
	int i, cnt;
	unsigned int ret_len = 0;

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

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

	if (iser_debug_level == 0)
		return;

	for_each_sg(sgl, sg, data->dma_nents, i)
		iser_warn("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, 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_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);
	}
}

/**
 * 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_task *iser_task,
		      enum   iser_data_dir        cmd_dir)
{
	struct iscsi_conn    *iscsi_conn = iser_task->iser_conn->iscsi_conn;
	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) {
		iscsi_conn->fmr_unalign_cnt++;
		iser_warn("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_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;
		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->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_task->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 <linux/scatterlist.h>
	40
	41	#include "iscsi_iser.h"
	42
	43	#define ISER_KMALLOC_THRESHOLD 0x20000 /* 128K - kmalloc limit */
8dfa0876	44
6461f64a OG	45	/**
	46	* Decrements the reference count for the
	47	* registered buffer & releases it
	48	*
	49	* returns 0 if released, 1 if deferred
	50	*/
	51	int iser_regd_buff_release(struct iser_regd_buf *regd_buf)
	52	{
5180311f	53	struct ib_device *dev;
6461f64a OG	54
	55	if ((atomic_read(&regd_buf->ref_count) == 0) \|\|
	56	atomic_dec_and_test(&regd_buf->ref_count)) {
	57	/* if we used the dma mr, unreg is just NOP */
d8111028	58	if (regd_buf->reg.is_fmr)
6461f64a OG	59	iser_unreg_mem(&regd_buf->reg);
	60
	61	if (regd_buf->dma_addr) {
5180311f RC	62	dev = regd_buf->device->ib_device;
5180311f RC	63	ib_dma_unmap_single(dev,
6461f64a OG	64	regd_buf->dma_addr,
	65	regd_buf->data_size,
	66	regd_buf->direction);
	67	}
	68	/* else this regd buf is associated with task which we */
	69	/* dma_unmap_single/sg later */
	70	return 0;
	71	} else {
	72	iser_dbg("Release deferred, regd.buff: 0x%p\n", regd_buf);
	73	return 1;
	74	}
	75	}
	76
	77	/**
	78	* iser_reg_single - fills registered buffer descriptor with
	79	* registration information
	80	*/
	81	void iser_reg_single(struct iser_device *device,
	82	struct iser_regd_buf *regd_buf,
	83	enum dma_data_direction direction)
	84	{
5180311f	85	u64 dma_addr;
6461f64a	86
5180311f RC	87	dma_addr = ib_dma_map_single(device->ib_device,
	88	regd_buf->virt_addr,
	89	regd_buf->data_size, direction);
	90	BUG_ON(ib_dma_mapping_error(device->ib_device, dma_addr));
6461f64a OG	91
6461f64a OG	92	regd_buf->reg.lkey = device->mr->lkey;
6461f64a OG	93	regd_buf->reg.len = regd_buf->data_size;
6461f64a OG	94	regd_buf->reg.va = dma_addr;
d8111028	95	regd_buf->reg.is_fmr = 0;
6461f64a OG	96
	97	regd_buf->dma_addr = dma_addr;
	98	regd_buf->direction = direction;
	99	}
	100
	101	/**
	102	* iser_start_rdma_unaligned_sg
	103	*/
2261ec3d	104	static int iser_start_rdma_unaligned_sg(struct iscsi_iser_task *iser_task,
41179e2d	105	enum iser_data_dir cmd_dir)
6461f64a OG	106	{
6461f64a OG	107	int dma_nents;
5180311f	108	struct ib_device *dev;
6461f64a	109	char *mem = NULL;
2261ec3d	110	struct iser_data_buf *data = &iser_task->data[cmd_dir];
6461f64a OG	111	unsigned long cmd_data_len = data->data_len;
	112
	113	if (cmd_data_len > ISER_KMALLOC_THRESHOLD)
	114	mem = (void *)__get_free_pages(GFP_NOIO,
f0d1b0b3	115	ilog2(roundup_pow_of_two(cmd_data_len)) - PAGE_SHIFT);
6461f64a OG	116	else
	117	mem = kmalloc(cmd_data_len, GFP_NOIO);
	118
	119	if (mem == NULL) {
	120	iser_err("Failed to allocate mem size %d %d for copying sglist\n",
	121	data->size,(int)cmd_data_len);
	122	return -ENOMEM;
	123	}
	124
	125	if (cmd_dir == ISER_DIR_OUT) {
	126	/* copy the unaligned sg the buffer which is used for RDMA */
53d412fc JA	127	struct scatterlist sgl = (struct scatterlist )data->buf;
53d412fc JA	128	struct scatterlist *sg;
6461f64a OG	129	int i;
	130	char p, from;
	131
53d412fc JA	132	p = mem;
53d412fc JA	133	for_each_sg(sgl, sg, data->size, i) {
45711f1a	134	from = kmap_atomic(sg_page(sg), KM_USER0);
6461f64a	135	memcpy(p,
53d412fc JA	136	from + sg->offset,
53d412fc JA	137	sg->length);
6461f64a	138	kunmap_atomic(from, KM_USER0);
53d412fc	139	p += sg->length;
6461f64a OG	140	}
	141	}
	142
2261ec3d MC	143	sg_init_one(&iser_task->data_copy[cmd_dir].sg_single, mem, cmd_data_len);
	144	iser_task->data_copy[cmd_dir].buf =
	145	&iser_task->data_copy[cmd_dir].sg_single;
	146	iser_task->data_copy[cmd_dir].size = 1;
6461f64a	147
2261ec3d	148	iser_task->data_copy[cmd_dir].copy_buf = mem;
6461f64a	149
2261ec3d	150	dev = iser_task->iser_conn->ib_conn->device->ib_device;
5180311f	151	dma_nents = ib_dma_map_sg(dev,
2261ec3d	152	&iser_task->data_copy[cmd_dir].sg_single,
5180311f RC	153	1,
	154	(cmd_dir == ISER_DIR_OUT) ?
	155	DMA_TO_DEVICE : DMA_FROM_DEVICE);
6461f64a OG	156	BUG_ON(dma_nents == 0);
6461f64a OG	157
2261ec3d	158	iser_task->data_copy[cmd_dir].dma_nents = dma_nents;
6461f64a OG	159	return 0;
	160	}
	161
	162	/**
	163	* iser_finalize_rdma_unaligned_sg
	164	*/
2261ec3d	165	void iser_finalize_rdma_unaligned_sg(struct iscsi_iser_task *iser_task,
6461f64a OG	166	enum iser_data_dir cmd_dir)
6461f64a OG	167	{
5180311f	168	struct ib_device *dev;
6461f64a OG	169	struct iser_data_buf *mem_copy;
	170	unsigned long cmd_data_len;
	171
2261ec3d MC	172	dev = iser_task->iser_conn->ib_conn->device->ib_device;
2261ec3d MC	173	mem_copy = &iser_task->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;
53d412fc	181	struct scatterlist sgl, sg;
6461f64a OG	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
2261ec3d MC	189	sgl = (struct scatterlist *)iser_task->data[ISER_DIR_IN].buf;
2261ec3d MC	190	sg_size = iser_task->data[ISER_DIR_IN].size;
6461f64a	191
53d412fc JA	192	p = mem;
53d412fc JA	193	for_each_sg(sgl, sg, sg_size, i) {
45711f1a	194	to = kmap_atomic(sg_page(sg), KM_SOFTIRQ0);
53d412fc	195	memcpy(to + sg->offset,
6461f64a	196	p,
53d412fc	197	sg->length);
6461f64a	198	kunmap_atomic(to, KM_SOFTIRQ0);
53d412fc	199	p += sg->length;
6461f64a OG	200	}
	201	}
	202
2261ec3d	203	cmd_data_len = iser_task->data[cmd_dir].data_len;
6461f64a OG	204
	205	if (cmd_data_len > ISER_KMALLOC_THRESHOLD)
	206	free_pages((unsigned long)mem_copy->copy_buf,
f0d1b0b3	207	ilog2(roundup_pow_of_two(cmd_data_len)) - PAGE_SHIFT);
6461f64a OG	208	else
	209	kfree(mem_copy->copy_buf);
	210
	211	mem_copy->copy_buf = NULL;
	212	}
	213
	214	/**
	215	* iser_sg_to_page_vec - Translates scatterlist entries to physical addresses
	216	* and returns the length of resulting physical address array (may be less than
	217	* the original due to possible compaction).
	218	*
	219	* we build a "page vec" under the assumption that the SG meets the RDMA
	220	* alignment requirements. Other then the first and last SG elements, all
	221	* the "internal" elements can be compacted into a list whose elements are
	222	* dma addresses of physical pages. The code supports also the weird case
	223	* where --few fragments of the same page-- are present in the SG as
	224	* consecutive elements. Also, it handles one entry SG.
	225	*/
	226	static int iser_sg_to_page_vec(struct iser_data_buf *data,
5180311f RC	227	struct iser_page_vec *page_vec,
5180311f RC	228	struct ib_device *ibdev)
6461f64a	229	{
53d412fc JA	230	struct scatterlist sgl = (struct scatterlist )data->buf;
53d412fc JA	231	struct scatterlist *sg;
5180311f	232	u64 first_addr, last_addr, page;
dee234f4	233	int end_aligned;
6461f64a OG	234	unsigned int cur_page = 0;
	235	unsigned long total_sz = 0;
	236	int i;
	237
	238	/* compute the offset of first element */
53d412fc	239	page_vec->offset = (u64) sgl[0].offset & ~MASK_4K;
6461f64a	240
53d412fc JA	241	for_each_sg(sgl, sg, data->dma_nents, i) {
53d412fc JA	242	unsigned int dma_len = ib_sg_dma_len(ibdev, sg);
5180311f RC	243
5180311f RC	244	total_sz += dma_len;
6461f64a	245
53d412fc	246	first_addr = ib_sg_dma_address(ibdev, sg);
5180311f	247	last_addr = first_addr + dma_len;
6461f64a	248
8dfa0876	249	end_aligned = !(last_addr & ~MASK_4K);
6461f64a OG	250
	251	/* continue to collect page fragments till aligned or SG ends */
	252	while (!end_aligned && (i + 1 < data->dma_nents)) {
53d412fc	253	sg = sg_next(sg);
6461f64a	254	i++;
53d412fc	255	dma_len = ib_sg_dma_len(ibdev, sg);
5180311f	256	total_sz += dma_len;
53d412fc	257	last_addr = ib_sg_dma_address(ibdev, sg) + dma_len;
8dfa0876	258	end_aligned = !(last_addr & ~MASK_4K);
6461f64a OG	259	}
6461f64a OG	260
8dfa0876 EZ	261	/* handle the 1st page in the 1st DMA element */
	262	if (cur_page == 0) {
	263	page = first_addr & MASK_4K;
	264	page_vec->pages[cur_page] = page;
	265	cur_page++;
	266	page += SIZE_4K;
	267	} else
	268	page = first_addr;
	269
	270	for (; page < last_addr; page += SIZE_4K) {
	271	page_vec->pages[cur_page] = page;
	272	cur_page++;
	273	}
6461f64a OG	274
	275	}
	276	page_vec->data_size = total_sz;
	277	iser_dbg("page_vec->data_size:%d cur_page %d\n", page_vec->data_size,cur_page);
	278	return cur_page;
	279	}
	280
8dfa0876	281	#define IS_4K_ALIGNED(addr) ((((unsigned long)addr) & ~MASK_4K) == 0)
6461f64a OG	282
	283	/**
	284	* iser_data_buf_aligned_len - Tries to determine the maximal correctly aligned
	285	* for RDMA sub-list of a scatter-gather list of memory buffers, and returns
	286	* the number of entries which are aligned correctly. Supports the case where
	287	* consecutive SG elements are actually fragments of the same physcial page.
	288	*/
5180311f RC	289	static unsigned int iser_data_buf_aligned_len(struct iser_data_buf *data,
5180311f RC	290	struct ib_device *ibdev)
6461f64a	291	{
53d412fc	292	struct scatterlist sgl, sg;
5180311f	293	u64 end_addr, next_addr;
6461f64a OG	294	int i, cnt;
	295	unsigned int ret_len = 0;
	296
53d412fc	297	sgl = (struct scatterlist *)data->buf;
6461f64a	298
53d412fc JA	299	cnt = 0;
53d412fc JA	300	for_each_sg(sgl, sg, data->dma_nents, i) {
6461f64a OG	301	/* iser_dbg("Checking sg iobuf [%d]: phys=0x%08lX "
6461f64a OG	302	"offset: %ld sz: %ld\n", i,
45711f1a	303	(unsigned long)sg_phys(sg),
53d412fc JA	304	(unsigned long)sg->offset,
	305	(unsigned long)sg->length); */
	306	end_addr = ib_sg_dma_address(ibdev, sg) +
	307	ib_sg_dma_len(ibdev, sg);
6461f64a OG	308	/* iser_dbg("Checking sg iobuf end address "
	309	"0x%08lX\n", end_addr); */
	310	if (i + 1 < data->dma_nents) {
53d412fc	311	next_addr = ib_sg_dma_address(ibdev, sg_next(sg));
6461f64a	312	/* are i, i+1 fragments of the same page? */
a316b79c EZ	313	if (end_addr == next_addr) {
a316b79c EZ	314	cnt++;
6461f64a	315	continue;
a316b79c	316	} else if (!IS_4K_ALIGNED(end_addr)) {
6461f64a OG	317	ret_len = cnt + 1;
	318	break;
	319	}
	320	}
a316b79c	321	cnt++;
6461f64a OG	322	}
	323	if (i == data->dma_nents)
	324	ret_len = cnt; /* loop ended */
	325	iser_dbg("Found %d aligned entries out of %d in sg:0x%p\n",
	326	ret_len, data->dma_nents, data);
	327	return ret_len;
	328	}
	329
5180311f RC	330	static void iser_data_buf_dump(struct iser_data_buf *data,
5180311f RC	331	struct ib_device *ibdev)
6461f64a	332	{
53d412fc JA	333	struct scatterlist sgl = (struct scatterlist )data->buf;
53d412fc JA	334	struct scatterlist *sg;
6461f64a OG	335	int i;
6461f64a OG	336
6f735e36 ED	337	if (iser_debug_level == 0)
	338	return;
	339
53d412fc	340	for_each_sg(sgl, sg, data->dma_nents, i)
6f735e36	341	iser_warn("sg[%d] dma_addr:0x%lX page:0x%p "
e981f1d4	342	"off:0x%x sz:0x%x dma_len:0x%x\n",
53d412fc	343	i, (unsigned long)ib_sg_dma_address(ibdev, sg),
45711f1a	344	sg_page(sg), sg->offset,
53d412fc	345	sg->length, ib_sg_dma_len(ibdev, sg));
6461f64a OG	346	}
	347
	348	static void iser_dump_page_vec(struct iser_page_vec *page_vec)
	349	{
	350	int i;
	351
	352	iser_err("page vec length %d data size %d\n",
	353	page_vec->length, page_vec->data_size);
	354	for (i = 0; i < page_vec->length; i++)
	355	iser_err("%d %lx\n",i,(unsigned long)page_vec->pages[i]);
	356	}
	357
	358	static void iser_page_vec_build(struct iser_data_buf *data,
5180311f RC	359	struct iser_page_vec *page_vec,
5180311f RC	360	struct ib_device *ibdev)
6461f64a OG	361	{
	362	int page_vec_len = 0;
	363
	364	page_vec->length = 0;
	365	page_vec->offset = 0;
	366
	367	iser_dbg("Translating sg sz: %d\n", data->dma_nents);
5180311f	368	page_vec_len = iser_sg_to_page_vec(data, page_vec, ibdev);
6461f64a OG	369	iser_dbg("sg len %d page_vec_len %d\n", data->dma_nents,page_vec_len);
	370
	371	page_vec->length = page_vec_len;
	372
8dfa0876	373	if (page_vec_len * SIZE_4K < page_vec->data_size) {
6461f64a	374	iser_err("page_vec too short to hold this SG\n");
5180311f	375	iser_data_buf_dump(data, ibdev);
6461f64a OG	376	iser_dump_page_vec(page_vec);
	377	BUG();
	378	}
	379	}
	380
2261ec3d MC	381	int iser_dma_map_task_data(struct iscsi_iser_task *iser_task,
	382	struct iser_data_buf *data,
	383	enum iser_data_dir iser_dir,
	384	enum dma_data_direction dma_dir)
74a20780	385	{
5180311f	386	struct ib_device *dev;
74a20780	387
2261ec3d MC	388	iser_task->dir[iser_dir] = 1;
2261ec3d MC	389	dev = iser_task->iser_conn->ib_conn->device->ib_device;
74a20780	390
5180311f	391	data->dma_nents = ib_dma_map_sg(dev, data->buf, data->size, dma_dir);
74a20780 EZ	392	if (data->dma_nents == 0) {
	393	iser_err("dma_map_sg failed!!!\n");
	394	return -EINVAL;
	395	}
	396	return 0;
	397	}
	398
2261ec3d	399	void iser_dma_unmap_task_data(struct iscsi_iser_task *iser_task)
74a20780	400	{
5180311f	401	struct ib_device *dev;
74a20780 EZ	402	struct iser_data_buf *data;
74a20780 EZ	403
2261ec3d	404	dev = iser_task->iser_conn->ib_conn->device->ib_device;
74a20780	405
2261ec3d MC	406	if (iser_task->dir[ISER_DIR_IN]) {
2261ec3d MC	407	data = &iser_task->data[ISER_DIR_IN];
5180311f	408	ib_dma_unmap_sg(dev, data->buf, data->size, DMA_FROM_DEVICE);
74a20780 EZ	409	}
74a20780 EZ	410
2261ec3d MC	411	if (iser_task->dir[ISER_DIR_OUT]) {
2261ec3d MC	412	data = &iser_task->data[ISER_DIR_OUT];
5180311f	413	ib_dma_unmap_sg(dev, data->buf, data->size, DMA_TO_DEVICE);
74a20780 EZ	414	}
	415	}
	416
6461f64a OG	417	/**
	418	* iser_reg_rdma_mem - Registers memory intended for RDMA,
	419	* obtaining rkey and va
	420	*
	421	* returns 0 on success, errno code on failure
	422	*/
2261ec3d	423	int iser_reg_rdma_mem(struct iscsi_iser_task *iser_task,
6461f64a OG	424	enum iser_data_dir cmd_dir)
6461f64a OG	425	{
2261ec3d MC	426	struct iscsi_conn *iscsi_conn = iser_task->iser_conn->iscsi_conn;
2261ec3d MC	427	struct iser_conn *ib_conn = iser_task->iser_conn->ib_conn;
d8111028	428	struct iser_device *device = ib_conn->device;
5180311f	429	struct ib_device *ibdev = device->ib_device;
2261ec3d	430	struct iser_data_buf *mem = &iser_task->data[cmd_dir];
6461f64a OG	431	struct iser_regd_buf *regd_buf;
	432	int aligned_len;
	433	int err;
e981f1d4	434	int i;
d8111028	435	struct scatterlist *sg;
6461f64a	436
2261ec3d	437	regd_buf = &iser_task->rdma_regd[cmd_dir];
6461f64a	438
5180311f	439	aligned_len = iser_data_buf_aligned_len(mem, ibdev);
777a71dd	440	if (aligned_len != mem->dma_nents) {
87528227	441	iscsi_conn->fmr_unalign_cnt++;
6f735e36	442	iser_warn("rdma alignment violation %d/%d aligned\n",
6461f64a	443	aligned_len, mem->size);
5180311f	444	iser_data_buf_dump(mem, ibdev);
74a20780 EZ	445
74a20780 EZ	446	/* unmap the command data before accessing it */
2261ec3d	447	iser_dma_unmap_task_data(iser_task);
74a20780	448
6461f64a OG	449	/* allocate copy buf, if we are writing, copy the */
6461f64a OG	450	/* unaligned scatterlist, dma map the copy */
2261ec3d	451	if (iser_start_rdma_unaligned_sg(iser_task, cmd_dir) != 0)
6461f64a	452	return -ENOMEM;
2261ec3d	453	mem = &iser_task->data_copy[cmd_dir];
6461f64a OG	454	}
6461f64a OG	455
d8111028 EZ	456	/* if there a single dma entry, FMR is not needed */
	457	if (mem->dma_nents == 1) {
	458	sg = (struct scatterlist *)mem->buf;
	459
	460	regd_buf->reg.lkey = device->mr->lkey;
	461	regd_buf->reg.rkey = device->mr->rkey;
5180311f RC	462	regd_buf->reg.len = ib_sg_dma_len(ibdev, &sg[0]);
5180311f RC	463	regd_buf->reg.va = ib_sg_dma_address(ibdev, &sg[0]);
d8111028 EZ	464	regd_buf->reg.is_fmr = 0;
	465
	466	iser_dbg("PHYSICAL Mem.register: lkey: 0x%08X rkey: 0x%08X "
	467	"va: 0x%08lX sz: %ld]\n",
	468	(unsigned int)regd_buf->reg.lkey,
	469	(unsigned int)regd_buf->reg.rkey,
	470	(unsigned long)regd_buf->reg.va,
	471	(unsigned long)regd_buf->reg.len);
	472	} else { /* use FMR for multiple dma entries */
5180311f	473	iser_page_vec_build(mem, ib_conn->page_vec, ibdev);
d8111028 EZ	474	err = iser_reg_page_vec(ib_conn, ib_conn->page_vec, &regd_buf->reg);
d8111028 EZ	475	if (err) {
5180311f	476	iser_data_buf_dump(mem, ibdev);
2261ec3d MC	477	iser_err("mem->dma_nents = %d (dlength = 0x%x)\n",
	478	mem->dma_nents,
	479	ntoh24(iser_task->desc.iscsi_header.dlength));
d8111028 EZ	480	iser_err("page_vec: data_size = 0x%x, length = %d, offset = 0x%x\n",
	481	ib_conn->page_vec->data_size, ib_conn->page_vec->length,
	482	ib_conn->page_vec->offset);
	483	for (i=0 ; i<ib_conn->page_vec->length ; i++)
	484	iser_err("page_vec[%d] = 0x%llx\n", i,
	485	(unsigned long long) ib_conn->page_vec->pages[i]);
	486	return err;
e981f1d4	487	}
e981f1d4	488	}
6461f64a OG	489
	490	/* take a reference on this regd buf such that it will not be released *
	491	* (eg in send dto completion) before we get the scsi response */
	492	atomic_inc(&regd_buf->ref_count);
	493	return 0;
	494	}