iser_memory.c 14.4 KB
Newer Older
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37
/*
 * 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>
A
Al Viro 已提交
38
#include <linux/highmem.h>
39 40 41 42 43
#include <linux/scatterlist.h>

#include "iscsi_iser.h"

#define ISER_KMALLOC_THRESHOLD 0x20000 /* 128K - kmalloc limit */
44

45 46 47 48 49 50 51 52
/**
 * 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)
{
53
	struct ib_device *dev;
54 55 56 57

	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 */
58
		if (regd_buf->reg.is_fmr)
59 60 61
			iser_unreg_mem(&regd_buf->reg);

		if (regd_buf->dma_addr) {
62 63
			dev = regd_buf->device->ib_device;
			ib_dma_unmap_single(dev,
64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84
					 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)
{
85
	u64 dma_addr;
86

87 88 89 90
	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));
91 92 93 94

	regd_buf->reg.lkey = device->mr->lkey;
	regd_buf->reg.len  = regd_buf->data_size;
	regd_buf->reg.va   = dma_addr;
95
	regd_buf->reg.is_fmr = 0;
96 97 98 99 100 101 102 103

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

/**
 * iser_start_rdma_unaligned_sg
 */
104 105
static int iser_start_rdma_unaligned_sg(struct iscsi_iser_cmd_task *iser_ctask,
					enum iser_data_dir cmd_dir)
106 107
{
	int dma_nents;
108
	struct ib_device *dev;
109 110 111 112 113 114
	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,
115
		      ilog2(roundup_pow_of_two(cmd_data_len)) - PAGE_SHIFT);
116 117 118 119 120 121 122 123 124 125 126
	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 */
J
Jens Axboe 已提交
127 128
		struct scatterlist *sgl = (struct scatterlist *)data->buf;
		struct scatterlist *sg;
129 130 131
		int i;
		char *p, *from;

J
Jens Axboe 已提交
132 133
		p = mem;
		for_each_sg(sgl, sg, data->size, i) {
J
Jens Axboe 已提交
134
			from = kmap_atomic(sg_page(sg), KM_USER0);
135
			memcpy(p,
J
Jens Axboe 已提交
136 137
			       from + sg->offset,
			       sg->length);
138
			kunmap_atomic(from, KM_USER0);
J
Jens Axboe 已提交
139
			p += sg->length;
140 141 142 143 144 145 146 147 148 149
		}
	}

	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;

150 151 152 153 154 155
	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);
156 157 158 159 160 161 162 163 164 165 166 167
	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)
{
168
	struct ib_device *dev;
169 170 171
	struct iser_data_buf *mem_copy;
	unsigned long  cmd_data_len;

172 173
	dev = iser_ctask->iser_conn->ib_conn->device->ib_device;
	mem_copy = &iser_ctask->data_copy[cmd_dir];
174

175 176 177
	ib_dma_unmap_sg(dev, &mem_copy->sg_single, 1,
			(cmd_dir == ISER_DIR_OUT) ?
			DMA_TO_DEVICE : DMA_FROM_DEVICE);
178 179 180

	if (cmd_dir == ISER_DIR_IN) {
		char *mem;
J
Jens Axboe 已提交
181
		struct scatterlist *sgl, *sg;
182 183 184 185 186 187 188
		unsigned char *p, *to;
		unsigned int sg_size;
		int i;

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

J
Jens Axboe 已提交
189
		sgl	= (struct scatterlist *)iser_ctask->data[ISER_DIR_IN].buf;
190 191
		sg_size = iser_ctask->data[ISER_DIR_IN].size;

J
Jens Axboe 已提交
192 193
		p = mem;
		for_each_sg(sgl, sg, sg_size, i) {
J
Jens Axboe 已提交
194
			to = kmap_atomic(sg_page(sg), KM_SOFTIRQ0);
J
Jens Axboe 已提交
195
			memcpy(to + sg->offset,
196
			       p,
J
Jens Axboe 已提交
197
			       sg->length);
198
			kunmap_atomic(to, KM_SOFTIRQ0);
J
Jens Axboe 已提交
199
			p += sg->length;
200 201 202 203 204 205 206
		}
	}

	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,
207
			   ilog2(roundup_pow_of_two(cmd_data_len)) - PAGE_SHIFT);
208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226
	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,
227 228
			       struct iser_page_vec *page_vec,
			       struct ib_device *ibdev)
229
{
J
Jens Axboe 已提交
230 231
	struct scatterlist *sgl = (struct scatterlist *)data->buf;
	struct scatterlist *sg;
232
	u64 first_addr, last_addr, page;
233
	int end_aligned;
234 235 236 237 238
	unsigned int cur_page = 0;
	unsigned long total_sz = 0;
	int i;

	/* compute the offset of first element */
J
Jens Axboe 已提交
239
	page_vec->offset = (u64) sgl[0].offset & ~MASK_4K;
240

J
Jens Axboe 已提交
241 242
	for_each_sg(sgl, sg, data->dma_nents, i) {
		unsigned int dma_len = ib_sg_dma_len(ibdev, sg);
243 244

		total_sz += dma_len;
245

J
Jens Axboe 已提交
246
		first_addr = ib_sg_dma_address(ibdev, sg);
247
		last_addr  = first_addr + dma_len;
248

249
		end_aligned   = !(last_addr  & ~MASK_4K);
250 251 252

		/* continue to collect page fragments till aligned or SG ends */
		while (!end_aligned && (i + 1 < data->dma_nents)) {
J
Jens Axboe 已提交
253
			sg = sg_next(sg);
254
			i++;
J
Jens Axboe 已提交
255
			dma_len = ib_sg_dma_len(ibdev, sg);
256
			total_sz += dma_len;
J
Jens Axboe 已提交
257
			last_addr = ib_sg_dma_address(ibdev, sg) + dma_len;
258
			end_aligned = !(last_addr  & ~MASK_4K);
259 260
		}

261 262 263 264 265 266 267 268 269 270 271 272 273
		/* 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++;
		}
274 275 276 277 278 279 280

	}
	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;
}

281
#define IS_4K_ALIGNED(addr)	((((unsigned long)addr) & ~MASK_4K) == 0)
282 283 284 285 286 287 288

/**
 * 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.
 */
289 290
static unsigned int iser_data_buf_aligned_len(struct iser_data_buf *data,
					      struct ib_device *ibdev)
291
{
J
Jens Axboe 已提交
292
	struct scatterlist *sgl, *sg;
293
	u64 end_addr, next_addr;
294 295 296
	int i, cnt;
	unsigned int ret_len = 0;

J
Jens Axboe 已提交
297
	sgl = (struct scatterlist *)data->buf;
298

J
Jens Axboe 已提交
299 300
	cnt = 0;
	for_each_sg(sgl, sg, data->dma_nents, i) {
301 302
		/* iser_dbg("Checking sg iobuf [%d]: phys=0x%08lX "
		   "offset: %ld sz: %ld\n", i,
J
Jens Axboe 已提交
303
		   (unsigned long)sg_phys(sg),
J
Jens Axboe 已提交
304 305 306 307
		   (unsigned long)sg->offset,
		   (unsigned long)sg->length); */
		end_addr = ib_sg_dma_address(ibdev, sg) +
			   ib_sg_dma_len(ibdev, sg);
308 309 310
		/* iser_dbg("Checking sg iobuf end address "
		       "0x%08lX\n", end_addr); */
		if (i + 1 < data->dma_nents) {
J
Jens Axboe 已提交
311
			next_addr = ib_sg_dma_address(ibdev, sg_next(sg));
312
			/* are i, i+1 fragments of the same page? */
313 314
			if (end_addr == next_addr) {
				cnt++;
315
				continue;
316
			} else if (!IS_4K_ALIGNED(end_addr)) {
317 318 319 320
				ret_len = cnt + 1;
				break;
			}
		}
321
		cnt++;
322 323 324 325 326 327 328 329
	}
	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;
}

330 331
static void iser_data_buf_dump(struct iser_data_buf *data,
			       struct ib_device *ibdev)
332
{
J
Jens Axboe 已提交
333 334
	struct scatterlist *sgl = (struct scatterlist *)data->buf;
	struct scatterlist *sg;
335 336
	int i;

337 338 339
	if (iser_debug_level == 0)
		return;

J
Jens Axboe 已提交
340
	for_each_sg(sgl, sg, data->dma_nents, i)
341
		iser_warn("sg[%d] dma_addr:0x%lX page:0x%p "
E
Erez Zilber 已提交
342
			 "off:0x%x sz:0x%x dma_len:0x%x\n",
J
Jens Axboe 已提交
343
			 i, (unsigned long)ib_sg_dma_address(ibdev, sg),
J
Jens Axboe 已提交
344
			 sg_page(sg), sg->offset,
J
Jens Axboe 已提交
345
			 sg->length, ib_sg_dma_len(ibdev, sg));
346 347 348 349 350 351 352 353 354 355 356 357 358
}

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,
359 360
				struct iser_page_vec *page_vec,
				struct ib_device *ibdev)
361 362 363 364 365 366 367
{
	int page_vec_len = 0;

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

	iser_dbg("Translating sg sz: %d\n", data->dma_nents);
368
	page_vec_len = iser_sg_to_page_vec(data, page_vec, ibdev);
369 370 371 372
	iser_dbg("sg len %d page_vec_len %d\n", data->dma_nents,page_vec_len);

	page_vec->length = page_vec_len;

373
	if (page_vec_len * SIZE_4K < page_vec->data_size) {
374
		iser_err("page_vec too short to hold this SG\n");
375
		iser_data_buf_dump(data, ibdev);
376 377 378 379 380
		iser_dump_page_vec(page_vec);
		BUG();
	}
}

381 382 383 384 385
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)
{
386
	struct ib_device *dev;
387 388

	iser_ctask->dir[iser_dir] = 1;
389
	dev = iser_ctask->iser_conn->ib_conn->device->ib_device;
390

391
	data->dma_nents = ib_dma_map_sg(dev, data->buf, data->size, dma_dir);
392 393 394 395 396 397 398 399 400
	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)
{
401
	struct ib_device *dev;
402 403
	struct iser_data_buf *data;

404
	dev = iser_ctask->iser_conn->ib_conn->device->ib_device;
405 406 407

	if (iser_ctask->dir[ISER_DIR_IN]) {
		data = &iser_ctask->data[ISER_DIR_IN];
408
		ib_dma_unmap_sg(dev, data->buf, data->size, DMA_FROM_DEVICE);
409 410 411 412
	}

	if (iser_ctask->dir[ISER_DIR_OUT]) {
		data = &iser_ctask->data[ISER_DIR_OUT];
413
		ib_dma_unmap_sg(dev, data->buf, data->size, DMA_TO_DEVICE);
414 415 416
	}
}

417 418 419 420 421 422 423 424 425
/**
 * 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)
{
426
	struct iscsi_conn    *iscsi_conn = iser_ctask->iser_conn->iscsi_conn;
427
	struct iser_conn     *ib_conn = iser_ctask->iser_conn->ib_conn;
428
	struct iser_device   *device = ib_conn->device;
429
	struct ib_device     *ibdev = device->ib_device;
430 431 432 433
	struct iser_data_buf *mem = &iser_ctask->data[cmd_dir];
	struct iser_regd_buf *regd_buf;
	int aligned_len;
	int err;
E
Erez Zilber 已提交
434
	int i;
435
	struct scatterlist *sg;
436 437 438

	regd_buf = &iser_ctask->rdma_regd[cmd_dir];

439
	aligned_len = iser_data_buf_aligned_len(mem, ibdev);
440
	if (aligned_len != mem->dma_nents) {
441
		iscsi_conn->fmr_unalign_cnt++;
442
		iser_warn("rdma alignment violation %d/%d aligned\n",
443
			 aligned_len, mem->size);
444
		iser_data_buf_dump(mem, ibdev);
445 446 447 448

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

449 450 451 452 453 454 455
		/* 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];
	}

456 457 458 459 460 461
	/* 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;
462 463
		regd_buf->reg.len  = ib_sg_dma_len(ibdev, &sg[0]);
		regd_buf->reg.va   = ib_sg_dma_address(ibdev, &sg[0]);
464 465 466 467 468 469 470 471 472
		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 */
473
		iser_page_vec_build(mem, ib_conn->page_vec, ibdev);
474 475
		err = iser_reg_page_vec(ib_conn, ib_conn->page_vec, &regd_buf->reg);
		if (err) {
476
			iser_data_buf_dump(mem, ibdev);
477 478 479 480 481 482 483 484 485
			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;
E
Erez Zilber 已提交
486 487
		}
	}
488 489 490 491 492 493

	/* 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;
}