xref: /linux/drivers/nvme/host/zns.c (revision 24168c5e6dfbdd5b414f048f47f75d64533296ca)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * Copyright (C) 2020 Western Digital Corporation or its affiliates.
4  */
5 
6 #include <linux/blkdev.h>
7 #include <linux/vmalloc.h>
8 #include "nvme.h"
9 
10 static int nvme_set_max_append(struct nvme_ctrl *ctrl)
11 {
12 	struct nvme_command c = { };
13 	struct nvme_id_ctrl_zns *id;
14 	int status;
15 
16 	id = kzalloc(sizeof(*id), GFP_KERNEL);
17 	if (!id)
18 		return -ENOMEM;
19 
20 	c.identify.opcode = nvme_admin_identify;
21 	c.identify.cns = NVME_ID_CNS_CS_CTRL;
22 	c.identify.csi = NVME_CSI_ZNS;
23 
24 	status = nvme_submit_sync_cmd(ctrl->admin_q, &c, id, sizeof(*id));
25 	if (status) {
26 		kfree(id);
27 		return status;
28 	}
29 
30 	if (id->zasl)
31 		ctrl->max_zone_append = 1 << (id->zasl + 3);
32 	else
33 		ctrl->max_zone_append = ctrl->max_hw_sectors;
34 	kfree(id);
35 	return 0;
36 }
37 
38 int nvme_query_zone_info(struct nvme_ns *ns, unsigned lbaf,
39 		struct nvme_zone_info *zi)
40 {
41 	struct nvme_effects_log *log = ns->head->effects;
42 	struct nvme_command c = { };
43 	struct nvme_id_ns_zns *id;
44 	int status;
45 
46 	/* Driver requires zone append support */
47 	if ((le32_to_cpu(log->iocs[nvme_cmd_zone_append]) &
48 			NVME_CMD_EFFECTS_CSUPP)) {
49 		if (test_and_clear_bit(NVME_NS_FORCE_RO, &ns->flags))
50 			dev_warn(ns->ctrl->device,
51 				 "Zone Append supported for zoned namespace:%d. Remove read-only mode\n",
52 				 ns->head->ns_id);
53 	} else {
54 		set_bit(NVME_NS_FORCE_RO, &ns->flags);
55 		dev_warn(ns->ctrl->device,
56 			 "Zone Append not supported for zoned namespace:%d. Forcing to read-only mode\n",
57 			 ns->head->ns_id);
58 	}
59 
60 	/* Lazily query controller append limit for the first zoned namespace */
61 	if (!ns->ctrl->max_zone_append) {
62 		status = nvme_set_max_append(ns->ctrl);
63 		if (status)
64 			return status;
65 	}
66 
67 	id = kzalloc(sizeof(*id), GFP_KERNEL);
68 	if (!id)
69 		return -ENOMEM;
70 
71 	c.identify.opcode = nvme_admin_identify;
72 	c.identify.nsid = cpu_to_le32(ns->head->ns_id);
73 	c.identify.cns = NVME_ID_CNS_CS_NS;
74 	c.identify.csi = NVME_CSI_ZNS;
75 
76 	status = nvme_submit_sync_cmd(ns->ctrl->admin_q, &c, id, sizeof(*id));
77 	if (status)
78 		goto free_data;
79 
80 	/*
81 	 * We currently do not handle devices requiring any of the zoned
82 	 * operation characteristics.
83 	 */
84 	if (id->zoc) {
85 		dev_warn(ns->ctrl->device,
86 			"zone operations:%x not supported for namespace:%u\n",
87 			le16_to_cpu(id->zoc), ns->head->ns_id);
88 		status = -ENODEV;
89 		goto free_data;
90 	}
91 
92 	zi->zone_size = le64_to_cpu(id->lbafe[lbaf].zsze);
93 	if (!is_power_of_2(zi->zone_size)) {
94 		dev_warn(ns->ctrl->device,
95 			"invalid zone size: %llu for namespace: %u\n",
96 			zi->zone_size, ns->head->ns_id);
97 		status = -ENODEV;
98 		goto free_data;
99 	}
100 	zi->max_open_zones = le32_to_cpu(id->mor) + 1;
101 	zi->max_active_zones = le32_to_cpu(id->mar) + 1;
102 
103 free_data:
104 	kfree(id);
105 	return status;
106 }
107 
108 void nvme_update_zone_info(struct nvme_ns *ns, struct queue_limits *lim,
109 		struct nvme_zone_info *zi)
110 {
111 	lim->zoned = 1;
112 	lim->max_open_zones = zi->max_open_zones;
113 	lim->max_active_zones = zi->max_active_zones;
114 	lim->max_zone_append_sectors = ns->ctrl->max_zone_append;
115 	lim->chunk_sectors = ns->head->zsze =
116 		nvme_lba_to_sect(ns->head, zi->zone_size);
117 	blk_queue_flag_set(QUEUE_FLAG_ZONE_RESETALL, ns->queue);
118 }
119 
120 static void *nvme_zns_alloc_report_buffer(struct nvme_ns *ns,
121 					  unsigned int nr_zones, size_t *buflen)
122 {
123 	struct request_queue *q = ns->disk->queue;
124 	size_t bufsize;
125 	void *buf;
126 
127 	const size_t min_bufsize = sizeof(struct nvme_zone_report) +
128 				   sizeof(struct nvme_zone_descriptor);
129 
130 	nr_zones = min_t(unsigned int, nr_zones,
131 			 get_capacity(ns->disk) >> ilog2(ns->head->zsze));
132 
133 	bufsize = sizeof(struct nvme_zone_report) +
134 		nr_zones * sizeof(struct nvme_zone_descriptor);
135 	bufsize = min_t(size_t, bufsize,
136 			queue_max_hw_sectors(q) << SECTOR_SHIFT);
137 	bufsize = min_t(size_t, bufsize, queue_max_segments(q) << PAGE_SHIFT);
138 
139 	while (bufsize >= min_bufsize) {
140 		buf = __vmalloc(bufsize, GFP_KERNEL | __GFP_NORETRY);
141 		if (buf) {
142 			*buflen = bufsize;
143 			return buf;
144 		}
145 		bufsize >>= 1;
146 	}
147 	return NULL;
148 }
149 
150 static int nvme_zone_parse_entry(struct nvme_ctrl *ctrl,
151 				 struct nvme_ns_head *head,
152 				 struct nvme_zone_descriptor *entry,
153 				 unsigned int idx, report_zones_cb cb,
154 				 void *data)
155 {
156 	struct blk_zone zone = { };
157 
158 	if ((entry->zt & 0xf) != NVME_ZONE_TYPE_SEQWRITE_REQ) {
159 		dev_err(ctrl->device, "invalid zone type %#x\n",
160 				entry->zt);
161 		return -EINVAL;
162 	}
163 
164 	zone.type = BLK_ZONE_TYPE_SEQWRITE_REQ;
165 	zone.cond = entry->zs >> 4;
166 	zone.len = head->zsze;
167 	zone.capacity = nvme_lba_to_sect(head, le64_to_cpu(entry->zcap));
168 	zone.start = nvme_lba_to_sect(head, le64_to_cpu(entry->zslba));
169 	if (zone.cond == BLK_ZONE_COND_FULL)
170 		zone.wp = zone.start + zone.len;
171 	else
172 		zone.wp = nvme_lba_to_sect(head, le64_to_cpu(entry->wp));
173 
174 	return cb(&zone, idx, data);
175 }
176 
177 int nvme_ns_report_zones(struct nvme_ns *ns, sector_t sector,
178 		unsigned int nr_zones, report_zones_cb cb, void *data)
179 {
180 	struct nvme_zone_report *report;
181 	struct nvme_command c = { };
182 	int ret, zone_idx = 0;
183 	unsigned int nz, i;
184 	size_t buflen;
185 
186 	if (ns->head->ids.csi != NVME_CSI_ZNS)
187 		return -EINVAL;
188 
189 	report = nvme_zns_alloc_report_buffer(ns, nr_zones, &buflen);
190 	if (!report)
191 		return -ENOMEM;
192 
193 	c.zmr.opcode = nvme_cmd_zone_mgmt_recv;
194 	c.zmr.nsid = cpu_to_le32(ns->head->ns_id);
195 	c.zmr.numd = cpu_to_le32(nvme_bytes_to_numd(buflen));
196 	c.zmr.zra = NVME_ZRA_ZONE_REPORT;
197 	c.zmr.zrasf = NVME_ZRASF_ZONE_REPORT_ALL;
198 	c.zmr.pr = NVME_REPORT_ZONE_PARTIAL;
199 
200 	sector &= ~(ns->head->zsze - 1);
201 	while (zone_idx < nr_zones && sector < get_capacity(ns->disk)) {
202 		memset(report, 0, buflen);
203 
204 		c.zmr.slba = cpu_to_le64(nvme_sect_to_lba(ns->head, sector));
205 		ret = nvme_submit_sync_cmd(ns->queue, &c, report, buflen);
206 		if (ret) {
207 			if (ret > 0)
208 				ret = -EIO;
209 			goto out_free;
210 		}
211 
212 		nz = min((unsigned int)le64_to_cpu(report->nr_zones), nr_zones);
213 		if (!nz)
214 			break;
215 
216 		for (i = 0; i < nz && zone_idx < nr_zones; i++) {
217 			ret = nvme_zone_parse_entry(ns->ctrl, ns->head,
218 						    &report->entries[i],
219 						    zone_idx, cb, data);
220 			if (ret)
221 				goto out_free;
222 			zone_idx++;
223 		}
224 
225 		sector += ns->head->zsze * nz;
226 	}
227 
228 	if (zone_idx > 0)
229 		ret = zone_idx;
230 	else
231 		ret = -EINVAL;
232 out_free:
233 	kvfree(report);
234 	return ret;
235 }
236 
237 blk_status_t nvme_setup_zone_mgmt_send(struct nvme_ns *ns, struct request *req,
238 		struct nvme_command *c, enum nvme_zone_mgmt_action action)
239 {
240 	memset(c, 0, sizeof(*c));
241 
242 	c->zms.opcode = nvme_cmd_zone_mgmt_send;
243 	c->zms.nsid = cpu_to_le32(ns->head->ns_id);
244 	c->zms.slba = cpu_to_le64(nvme_sect_to_lba(ns->head, blk_rq_pos(req)));
245 	c->zms.zsa = action;
246 
247 	if (req_op(req) == REQ_OP_ZONE_RESET_ALL)
248 		c->zms.select_all = 1;
249 
250 	return BLK_STS_OK;
251 }
252