[deliverable/linux.git] / net / sunrpc / xprtrdma / fmr_ops.c

/*
 * Copyright (c) 2015 Oracle.  All rights reserved.
 * Copyright (c) 2003-2007 Network Appliance, Inc. All rights reserved.
 */

/* Lightweight memory registration using Fast Memory Regions (FMR).
 * Referred to sometimes as MTHCAFMR mode.
 *
 * FMR uses synchronous memory registration and deregistration.
 * FMR registration is known to be fast, but FMR deregistration
 * can take tens of usecs to complete.
 */

/* Normal operation
 *
 * A Memory Region is prepared for RDMA READ or WRITE using the
 * ib_map_phys_fmr verb (fmr_op_map). When the RDMA operation is
 * finished, the Memory Region is unmapped using the ib_unmap_fmr
 * verb (fmr_op_unmap).
 */

/* Transport recovery
 *
 * After a transport reconnect, fmr_op_map re-uses the MR already
 * allocated for the RPC, but generates a fresh rkey then maps the
 * MR again. This process is synchronous.
 */

#include "xprt_rdma.h"

#if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
# define RPCDBG_FACILITY	RPCDBG_TRANS
#endif

/* Maximum scatter/gather per FMR */
#define RPCRDMA_MAX_FMR_SGES	(64)

static int
fmr_op_open(struct rpcrdma_ia *ia, struct rpcrdma_ep *ep,
	    struct rpcrdma_create_data_internal *cdata)
{
	struct ib_device_attr *devattr = &ia->ri_devattr;
	struct ib_mr *mr;

	/* Obtain an lkey to use for the regbufs, which are
	 * protected from remote access.
	 */
	if (devattr->device_cap_flags & IB_DEVICE_LOCAL_DMA_LKEY) {
		ia->ri_dma_lkey = ia->ri_device->local_dma_lkey;
	} else {
		mr = ib_get_dma_mr(ia->ri_pd, IB_ACCESS_LOCAL_WRITE);
		if (IS_ERR(mr)) {
			pr_err("%s: ib_get_dma_mr for failed with %lX\n",
			       __func__, PTR_ERR(mr));
			return -ENOMEM;
		}
		ia->ri_dma_lkey = ia->ri_dma_mr->lkey;
		ia->ri_dma_mr = mr;
	}

	return 0;
}

/* FMR mode conveys up to 64 pages of payload per chunk segment.
 */
static size_t
fmr_op_maxpages(struct rpcrdma_xprt *r_xprt)
{
	return min_t(unsigned int, RPCRDMA_MAX_DATA_SEGS,
		     rpcrdma_max_segments(r_xprt) * RPCRDMA_MAX_FMR_SGES);
}

static int
fmr_op_init(struct rpcrdma_xprt *r_xprt)
{
	struct rpcrdma_buffer *buf = &r_xprt->rx_buf;
	int mr_access_flags = IB_ACCESS_REMOTE_WRITE | IB_ACCESS_REMOTE_READ;
	struct ib_fmr_attr fmr_attr = {
		.max_pages	= RPCRDMA_MAX_FMR_SGES,
		.max_maps	= 1,
		.page_shift	= PAGE_SHIFT
	};
	struct ib_pd *pd = r_xprt->rx_ia.ri_pd;
	struct rpcrdma_mw *r;
	int i, rc;

	spin_lock_init(&buf->rb_mwlock);
	INIT_LIST_HEAD(&buf->rb_mws);
	INIT_LIST_HEAD(&buf->rb_all);

	i = max_t(int, RPCRDMA_MAX_DATA_SEGS / RPCRDMA_MAX_FMR_SGES, 1);
	i += 2;				/* head + tail */
	i *= buf->rb_max_requests;	/* one set for each RPC slot */
	dprintk("RPC:       %s: initalizing %d FMRs\n", __func__, i);

	rc = -ENOMEM;
	while (i--) {
		r = kzalloc(sizeof(*r), GFP_KERNEL);
		if (!r)
			goto out;

		r->r.fmr.physaddrs = kmalloc(RPCRDMA_MAX_FMR_SGES *
					     sizeof(u64), GFP_KERNEL);
		if (!r->r.fmr.physaddrs)
			goto out_free;

		r->r.fmr.fmr = ib_alloc_fmr(pd, mr_access_flags, &fmr_attr);
		if (IS_ERR(r->r.fmr.fmr))
			goto out_fmr_err;

		list_add(&r->mw_list, &buf->rb_mws);
		list_add(&r->mw_all, &buf->rb_all);
	}
	return 0;

out_fmr_err:
	rc = PTR_ERR(r->r.fmr.fmr);
	dprintk("RPC:       %s: ib_alloc_fmr status %i\n", __func__, rc);
	kfree(r->r.fmr.physaddrs);
out_free:
	kfree(r);
out:
	return rc;
}

static int
__fmr_unmap(struct rpcrdma_mw *r)
{
	LIST_HEAD(l);

	list_add(&r->r.fmr.fmr->list, &l);
	return ib_unmap_fmr(&l);
}

/* Use the ib_map_phys_fmr() verb to register a memory region
 * for remote access via RDMA READ or RDMA WRITE.
 */
static int
fmr_op_map(struct rpcrdma_xprt *r_xprt, struct rpcrdma_mr_seg *seg,
	   int nsegs, bool writing)
{
	struct rpcrdma_ia *ia = &r_xprt->rx_ia;
	struct ib_device *device = ia->ri_device;
	enum dma_data_direction direction = rpcrdma_data_dir(writing);
	struct rpcrdma_mr_seg *seg1 = seg;
	int len, pageoff, i, rc;
	struct rpcrdma_mw *mw;

	mw = seg1->rl_mw;
	seg1->rl_mw = NULL;
	if (!mw) {
		mw = rpcrdma_get_mw(r_xprt);
		if (!mw)
			return -ENOMEM;
	} else {
		/* this is a retransmit; generate a fresh rkey */
		rc = __fmr_unmap(mw);
		if (rc)
			return rc;
	}

	pageoff = offset_in_page(seg1->mr_offset);
	seg1->mr_offset -= pageoff;	/* start of page */
	seg1->mr_len += pageoff;
	len = -pageoff;
	if (nsegs > RPCRDMA_MAX_FMR_SGES)
		nsegs = RPCRDMA_MAX_FMR_SGES;
	for (i = 0; i < nsegs;) {
		rpcrdma_map_one(device, seg, direction);
		mw->r.fmr.physaddrs[i] = seg->mr_dma;
		len += seg->mr_len;
		++seg;
		++i;
		/* Check for holes */
		if ((i < nsegs && offset_in_page(seg->mr_offset)) ||
		    offset_in_page((seg-1)->mr_offset + (seg-1)->mr_len))
			break;
	}

	rc = ib_map_phys_fmr(mw->r.fmr.fmr, mw->r.fmr.physaddrs,
			     i, seg1->mr_dma);
	if (rc)
		goto out_maperr;

	seg1->rl_mw = mw;
	seg1->mr_rkey = mw->r.fmr.fmr->rkey;
	seg1->mr_base = seg1->mr_dma + pageoff;
	seg1->mr_nsegs = i;
	seg1->mr_len = len;
	return i;

out_maperr:
	dprintk("RPC:       %s: ib_map_phys_fmr %u@0x%llx+%i (%d) status %i\n",
		__func__, len, (unsigned long long)seg1->mr_dma,
		pageoff, i, rc);
	while (i--)
		rpcrdma_unmap_one(device, --seg);
	return rc;
}

/* Use the ib_unmap_fmr() verb to prevent further remote
 * access via RDMA READ or RDMA WRITE.
 */
static int
fmr_op_unmap(struct rpcrdma_xprt *r_xprt, struct rpcrdma_mr_seg *seg)
{
	struct rpcrdma_ia *ia = &r_xprt->rx_ia;
	struct rpcrdma_mr_seg *seg1 = seg;
	struct rpcrdma_mw *mw = seg1->rl_mw;
	int rc, nsegs = seg->mr_nsegs;

	dprintk("RPC:       %s: FMR %p\n", __func__, mw);

	seg1->rl_mw = NULL;
	while (seg1->mr_nsegs--)
		rpcrdma_unmap_one(ia->ri_device, seg++);
	rc = __fmr_unmap(mw);
	if (rc)
		goto out_err;
	rpcrdma_put_mw(r_xprt, mw);
	return nsegs;

out_err:
	/* The FMR is abandoned, but remains in rb_all. fmr_op_destroy
	 * will attempt to release it when the transport is destroyed.
	 */
	dprintk("RPC:       %s: ib_unmap_fmr status %i\n", __func__, rc);
	return nsegs;
}

static void
fmr_op_destroy(struct rpcrdma_buffer *buf)
{
	struct rpcrdma_mw *r;
	int rc;

	while (!list_empty(&buf->rb_all)) {
		r = list_entry(buf->rb_all.next, struct rpcrdma_mw, mw_all);
		list_del(&r->mw_all);
		kfree(r->r.fmr.physaddrs);

		rc = ib_dealloc_fmr(r->r.fmr.fmr);
		if (rc)
			dprintk("RPC:       %s: ib_dealloc_fmr failed %i\n",
				__func__, rc);

		kfree(r);
	}
}

const struct rpcrdma_memreg_ops rpcrdma_fmr_memreg_ops = {
	.ro_map				= fmr_op_map,
	.ro_unmap			= fmr_op_unmap,
	.ro_open			= fmr_op_open,
	.ro_maxpages			= fmr_op_maxpages,
	.ro_init			= fmr_op_init,
	.ro_destroy			= fmr_op_destroy,
	.ro_displayname			= "fmr",
};
Commit	Line	Data
a0ce85f5 CL	1	/*
	2	* Copyright (c) 2015 Oracle. All rights reserved.
	3	* Copyright (c) 2003-2007 Network Appliance, Inc. All rights reserved.
	4	*/
	5
	6	/* Lightweight memory registration using Fast Memory Regions (FMR).
	7	* Referred to sometimes as MTHCAFMR mode.
	8	*
	9	* FMR uses synchronous memory registration and deregistration.
	10	* FMR registration is known to be fast, but FMR deregistration
	11	* can take tens of usecs to complete.
	12	*/
	13
fc7fbb59 CL	14	/* Normal operation
	15	*
	16	* A Memory Region is prepared for RDMA READ or WRITE using the
	17	* ib_map_phys_fmr verb (fmr_op_map). When the RDMA operation is
	18	* finished, the Memory Region is unmapped using the ib_unmap_fmr
	19	* verb (fmr_op_unmap).
	20	*/
	21
	22	/* Transport recovery
	23	*
	24	* After a transport reconnect, fmr_op_map re-uses the MR already
	25	* allocated for the RPC, but generates a fresh rkey then maps the
	26	* MR again. This process is synchronous.
	27	*/
	28
a0ce85f5 CL	29	#include "xprt_rdma.h"
	30
	31	#if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
	32	# define RPCDBG_FACILITY RPCDBG_TRANS
	33	#endif
	34
1c9351ee CL	35	/* Maximum scatter/gather per FMR */
	36	#define RPCRDMA_MAX_FMR_SGES (64)
	37
3968cb58 CL	38	static int
	39	fmr_op_open(struct rpcrdma_ia ia, struct rpcrdma_ep ep,
	40	struct rpcrdma_create_data_internal *cdata)
	41	{
d1ed857e CL	42	struct ib_device_attr *devattr = &ia->ri_devattr;
	43	struct ib_mr *mr;
	44
	45	/* Obtain an lkey to use for the regbufs, which are
	46	* protected from remote access.
	47	*/
	48	if (devattr->device_cap_flags & IB_DEVICE_LOCAL_DMA_LKEY) {
	49	ia->ri_dma_lkey = ia->ri_device->local_dma_lkey;
	50	} else {
	51	mr = ib_get_dma_mr(ia->ri_pd, IB_ACCESS_LOCAL_WRITE);
	52	if (IS_ERR(mr)) {
	53	pr_err("%s: ib_get_dma_mr for failed with %lX\n",
	54	__func__, PTR_ERR(mr));
	55	return -ENOMEM;
	56	}
	57	ia->ri_dma_lkey = ia->ri_dma_mr->lkey;
	58	ia->ri_dma_mr = mr;
	59	}
	60
3968cb58 CL	61	return 0;
	62	}
	63
1c9351ee CL	64	/* FMR mode conveys up to 64 pages of payload per chunk segment.
	65	*/
	66	static size_t
	67	fmr_op_maxpages(struct rpcrdma_xprt *r_xprt)
	68	{
	69	return min_t(unsigned int, RPCRDMA_MAX_DATA_SEGS,
	70	rpcrdma_max_segments(r_xprt) * RPCRDMA_MAX_FMR_SGES);
	71	}
	72
91e70e70 CL	73	static int
	74	fmr_op_init(struct rpcrdma_xprt *r_xprt)
	75	{
	76	struct rpcrdma_buffer *buf = &r_xprt->rx_buf;
	77	int mr_access_flags = IB_ACCESS_REMOTE_WRITE \| IB_ACCESS_REMOTE_READ;
	78	struct ib_fmr_attr fmr_attr = {
	79	.max_pages = RPCRDMA_MAX_FMR_SGES,
	80	.max_maps = 1,
	81	.page_shift = PAGE_SHIFT
	82	};
	83	struct ib_pd *pd = r_xprt->rx_ia.ri_pd;
	84	struct rpcrdma_mw *r;
	85	int i, rc;
	86
58d1dcf5	87	spin_lock_init(&buf->rb_mwlock);
91e70e70 CL	88	INIT_LIST_HEAD(&buf->rb_mws);
	89	INIT_LIST_HEAD(&buf->rb_all);
	90
40c6ed0c CL	91	i = max_t(int, RPCRDMA_MAX_DATA_SEGS / RPCRDMA_MAX_FMR_SGES, 1);
	92	i += 2; /* head + tail */
	93	i = buf->rb_max_requests; / one set for each RPC slot */
	94	dprintk("RPC: %s: initalizing %d FMRs\n", __func__, i);
91e70e70	95
acb9da7a	96	rc = -ENOMEM;
91e70e70 CL	97	while (i--) {
	98	r = kzalloc(sizeof(*r), GFP_KERNEL);
	99	if (!r)
acb9da7a CL	100	goto out;
	101
	102	r->r.fmr.physaddrs = kmalloc(RPCRDMA_MAX_FMR_SGES *
	103	sizeof(u64), GFP_KERNEL);
	104	if (!r->r.fmr.physaddrs)
	105	goto out_free;
91e70e70	106
acb9da7a CL	107	r->r.fmr.fmr = ib_alloc_fmr(pd, mr_access_flags, &fmr_attr);
acb9da7a CL	108	if (IS_ERR(r->r.fmr.fmr))
91e70e70 CL	109	goto out_fmr_err;
	110
	111	list_add(&r->mw_list, &buf->rb_mws);
	112	list_add(&r->mw_all, &buf->rb_all);
	113	}
	114	return 0;
	115
	116	out_fmr_err:
acb9da7a	117	rc = PTR_ERR(r->r.fmr.fmr);
91e70e70	118	dprintk("RPC: %s: ib_alloc_fmr status %i\n", __func__, rc);
acb9da7a CL	119	kfree(r->r.fmr.physaddrs);
acb9da7a CL	120	out_free:
91e70e70	121	kfree(r);
acb9da7a	122	out:
91e70e70 CL	123	return rc;
	124	}
	125
fc7fbb59 CL	126	static int
	127	__fmr_unmap(struct rpcrdma_mw *r)
	128	{
	129	LIST_HEAD(l);
	130
acb9da7a	131	list_add(&r->r.fmr.fmr->list, &l);
fc7fbb59 CL	132	return ib_unmap_fmr(&l);
	133	}
	134
9c1b4d77 CL	135	/* Use the ib_map_phys_fmr() verb to register a memory region
	136	* for remote access via RDMA READ or RDMA WRITE.
	137	*/
	138	static int
	139	fmr_op_map(struct rpcrdma_xprt r_xprt, struct rpcrdma_mr_seg seg,
	140	int nsegs, bool writing)
	141	{
	142	struct rpcrdma_ia *ia = &r_xprt->rx_ia;
89e0d112	143	struct ib_device *device = ia->ri_device;
d654788e	144	enum dma_data_direction direction = rpcrdma_data_dir(writing);
9c1b4d77	145	struct rpcrdma_mr_seg *seg1 = seg;
9c1b4d77	146	int len, pageoff, i, rc;
fc7fbb59 CL	147	struct rpcrdma_mw *mw;
	148
	149	mw = seg1->rl_mw;
	150	seg1->rl_mw = NULL;
	151	if (!mw) {
	152	mw = rpcrdma_get_mw(r_xprt);
	153	if (!mw)
	154	return -ENOMEM;
	155	} else {
	156	/* this is a retransmit; generate a fresh rkey */
	157	rc = __fmr_unmap(mw);
	158	if (rc)
	159	return rc;
	160	}
9c1b4d77 CL	161
	162	pageoff = offset_in_page(seg1->mr_offset);
	163	seg1->mr_offset -= pageoff; /* start of page */
	164	seg1->mr_len += pageoff;
	165	len = -pageoff;
	166	if (nsegs > RPCRDMA_MAX_FMR_SGES)
	167	nsegs = RPCRDMA_MAX_FMR_SGES;
	168	for (i = 0; i < nsegs;) {
d654788e	169	rpcrdma_map_one(device, seg, direction);
acb9da7a	170	mw->r.fmr.physaddrs[i] = seg->mr_dma;
9c1b4d77 CL	171	len += seg->mr_len;
	172	++seg;
	173	++i;
	174	/* Check for holes */
	175	if ((i < nsegs && offset_in_page(seg->mr_offset)) \|\|
	176	offset_in_page((seg-1)->mr_offset + (seg-1)->mr_len))
	177	break;
	178	}
	179
acb9da7a CL	180	rc = ib_map_phys_fmr(mw->r.fmr.fmr, mw->r.fmr.physaddrs,
acb9da7a CL	181	i, seg1->mr_dma);
9c1b4d77 CL	182	if (rc)
	183	goto out_maperr;
	184
fc7fbb59	185	seg1->rl_mw = mw;
acb9da7a	186	seg1->mr_rkey = mw->r.fmr.fmr->rkey;
9c1b4d77 CL	187	seg1->mr_base = seg1->mr_dma + pageoff;
	188	seg1->mr_nsegs = i;
	189	seg1->mr_len = len;
	190	return i;
	191
	192	out_maperr:
	193	dprintk("RPC: %s: ib_map_phys_fmr %u@0x%llx+%i (%d) status %i\n",
	194	__func__, len, (unsigned long long)seg1->mr_dma,
	195	pageoff, i, rc);
	196	while (i--)
d654788e	197	rpcrdma_unmap_one(device, --seg);
9c1b4d77 CL	198	return rc;
	199	}
	200
6814baea CL	201	/* Use the ib_unmap_fmr() verb to prevent further remote
	202	* access via RDMA READ or RDMA WRITE.
	203	*/
	204	static int
	205	fmr_op_unmap(struct rpcrdma_xprt r_xprt, struct rpcrdma_mr_seg seg)
	206	{
	207	struct rpcrdma_ia *ia = &r_xprt->rx_ia;
	208	struct rpcrdma_mr_seg *seg1 = seg;
fc7fbb59	209	struct rpcrdma_mw *mw = seg1->rl_mw;
6814baea	210	int rc, nsegs = seg->mr_nsegs;
6814baea	211
fc7fbb59 CL	212	dprintk("RPC: %s: FMR %p\n", __func__, mw);
	213
	214	seg1->rl_mw = NULL;
6814baea	215	while (seg1->mr_nsegs--)
89e0d112	216	rpcrdma_unmap_one(ia->ri_device, seg++);
fc7fbb59	217	rc = __fmr_unmap(mw);
6814baea CL	218	if (rc)
6814baea CL	219	goto out_err;
fc7fbb59	220	rpcrdma_put_mw(r_xprt, mw);
6814baea CL	221	return nsegs;
	222
	223	out_err:
fc7fbb59 CL	224	/* The FMR is abandoned, but remains in rb_all. fmr_op_destroy
	225	* will attempt to release it when the transport is destroyed.
	226	*/
6814baea CL	227	dprintk("RPC: %s: ib_unmap_fmr status %i\n", __func__, rc);
	228	return nsegs;
	229	}
	230
4561f347 CL	231	static void
	232	fmr_op_destroy(struct rpcrdma_buffer *buf)
	233	{
	234	struct rpcrdma_mw *r;
	235	int rc;
	236
	237	while (!list_empty(&buf->rb_all)) {
	238	r = list_entry(buf->rb_all.next, struct rpcrdma_mw, mw_all);
	239	list_del(&r->mw_all);
acb9da7a CL	240	kfree(r->r.fmr.physaddrs);
	241
	242	rc = ib_dealloc_fmr(r->r.fmr.fmr);
4561f347 CL	243	if (rc)
	244	dprintk("RPC: %s: ib_dealloc_fmr failed %i\n",
	245	__func__, rc);
acb9da7a	246
4561f347 CL	247	kfree(r);
	248	}
	249	}
	250
a0ce85f5	251	const struct rpcrdma_memreg_ops rpcrdma_fmr_memreg_ops = {
9c1b4d77	252	.ro_map = fmr_op_map,
6814baea	253	.ro_unmap = fmr_op_unmap,
3968cb58	254	.ro_open = fmr_op_open,
1c9351ee	255	.ro_maxpages = fmr_op_maxpages,
91e70e70	256	.ro_init = fmr_op_init,
4561f347	257	.ro_destroy = fmr_op_destroy,
a0ce85f5 CL	258	.ro_displayname = "fmr",
a0ce85f5 CL	259	};