xref: /linux/drivers/nvme/host/zns.c (revision f82811e22b480a203a438d8e1f29af9c93ccbb0c)
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 int nvme_revalidate_zones(struct nvme_ns *ns)
11 {
12 	struct request_queue *q = ns->queue;
13 
14 	blk_queue_chunk_sectors(q, ns->head->zsze);
15 	blk_queue_max_zone_append_sectors(q, ns->ctrl->max_zone_append);
16 
17 	return blk_revalidate_disk_zones(ns->disk, NULL);
18 }
19 
20 static int nvme_set_max_append(struct nvme_ctrl *ctrl)
21 {
22 	struct nvme_command c = { };
23 	struct nvme_id_ctrl_zns *id;
24 	int status;
25 
26 	id = kzalloc(sizeof(*id), GFP_KERNEL);
27 	if (!id)
28 		return -ENOMEM;
29 
30 	c.identify.opcode = nvme_admin_identify;
31 	c.identify.cns = NVME_ID_CNS_CS_CTRL;
32 	c.identify.csi = NVME_CSI_ZNS;
33 
34 	status = nvme_submit_sync_cmd(ctrl->admin_q, &c, id, sizeof(*id));
35 	if (status) {
36 		kfree(id);
37 		return status;
38 	}
39 
40 	if (id->zasl)
41 		ctrl->max_zone_append = 1 << (id->zasl + 3);
42 	else
43 		ctrl->max_zone_append = ctrl->max_hw_sectors;
44 	kfree(id);
45 	return 0;
46 }
47 
48 int nvme_update_zone_info(struct nvme_ns *ns, unsigned lbaf)
49 {
50 	struct nvme_effects_log *log = ns->head->effects;
51 	struct request_queue *q = ns->queue;
52 	struct nvme_command c = { };
53 	struct nvme_id_ns_zns *id;
54 	int status;
55 
56 	/* Driver requires zone append support */
57 	if ((le32_to_cpu(log->iocs[nvme_cmd_zone_append]) &
58 			NVME_CMD_EFFECTS_CSUPP)) {
59 		if (test_and_clear_bit(NVME_NS_FORCE_RO, &ns->flags))
60 			dev_warn(ns->ctrl->device,
61 				 "Zone Append supported for zoned namespace:%d. Remove read-only mode\n",
62 				 ns->head->ns_id);
63 	} else {
64 		set_bit(NVME_NS_FORCE_RO, &ns->flags);
65 		dev_warn(ns->ctrl->device,
66 			 "Zone Append not supported for zoned namespace:%d. Forcing to read-only mode\n",
67 			 ns->head->ns_id);
68 	}
69 
70 	/* Lazily query controller append limit for the first zoned namespace */
71 	if (!ns->ctrl->max_zone_append) {
72 		status = nvme_set_max_append(ns->ctrl);
73 		if (status)
74 			return status;
75 	}
76 
77 	id = kzalloc(sizeof(*id), GFP_KERNEL);
78 	if (!id)
79 		return -ENOMEM;
80 
81 	c.identify.opcode = nvme_admin_identify;
82 	c.identify.nsid = cpu_to_le32(ns->head->ns_id);
83 	c.identify.cns = NVME_ID_CNS_CS_NS;
84 	c.identify.csi = NVME_CSI_ZNS;
85 
86 	status = nvme_submit_sync_cmd(ns->ctrl->admin_q, &c, id, sizeof(*id));
87 	if (status)
88 		goto free_data;
89 
90 	/*
91 	 * We currently do not handle devices requiring any of the zoned
92 	 * operation characteristics.
93 	 */
94 	if (id->zoc) {
95 		dev_warn(ns->ctrl->device,
96 			"zone operations:%x not supported for namespace:%u\n",
97 			le16_to_cpu(id->zoc), ns->head->ns_id);
98 		status = -ENODEV;
99 		goto free_data;
100 	}
101 
102 	ns->head->zsze =
103 		nvme_lba_to_sect(ns->head, le64_to_cpu(id->lbafe[lbaf].zsze));
104 	if (!is_power_of_2(ns->head->zsze)) {
105 		dev_warn(ns->ctrl->device,
106 			"invalid zone size:%llu for namespace:%u\n",
107 			ns->head->zsze, ns->head->ns_id);
108 		status = -ENODEV;
109 		goto free_data;
110 	}
111 
112 	disk_set_zoned(ns->disk);
113 	blk_queue_flag_set(QUEUE_FLAG_ZONE_RESETALL, q);
114 	disk_set_max_open_zones(ns->disk, le32_to_cpu(id->mor) + 1);
115 	disk_set_max_active_zones(ns->disk, le32_to_cpu(id->mar) + 1);
116 free_data:
117 	kfree(id);
118 	return status;
119 }
120 
121 static void *nvme_zns_alloc_report_buffer(struct nvme_ns *ns,
122 					  unsigned int nr_zones, size_t *buflen)
123 {
124 	struct request_queue *q = ns->disk->queue;
125 	size_t bufsize;
126 	void *buf;
127 
128 	const size_t min_bufsize = sizeof(struct nvme_zone_report) +
129 				   sizeof(struct nvme_zone_descriptor);
130 
131 	nr_zones = min_t(unsigned int, nr_zones,
132 			 get_capacity(ns->disk) >> ilog2(ns->head->zsze));
133 
134 	bufsize = sizeof(struct nvme_zone_report) +
135 		nr_zones * sizeof(struct nvme_zone_descriptor);
136 	bufsize = min_t(size_t, bufsize,
137 			queue_max_hw_sectors(q) << SECTOR_SHIFT);
138 	bufsize = min_t(size_t, bufsize, queue_max_segments(q) << PAGE_SHIFT);
139 
140 	while (bufsize >= min_bufsize) {
141 		buf = __vmalloc(bufsize, GFP_KERNEL | __GFP_NORETRY);
142 		if (buf) {
143 			*buflen = bufsize;
144 			return buf;
145 		}
146 		bufsize >>= 1;
147 	}
148 	return NULL;
149 }
150 
151 static int nvme_zone_parse_entry(struct nvme_ctrl *ctrl,
152 				 struct nvme_ns_head *head,
153 				 struct nvme_zone_descriptor *entry,
154 				 unsigned int idx, report_zones_cb cb,
155 				 void *data)
156 {
157 	struct blk_zone zone = { };
158 
159 	if ((entry->zt & 0xf) != NVME_ZONE_TYPE_SEQWRITE_REQ) {
160 		dev_err(ctrl->device, "invalid zone type %#x\n",
161 				entry->zt);
162 		return -EINVAL;
163 	}
164 
165 	zone.type = BLK_ZONE_TYPE_SEQWRITE_REQ;
166 	zone.cond = entry->zs >> 4;
167 	zone.len = head->zsze;
168 	zone.capacity = nvme_lba_to_sect(head, le64_to_cpu(entry->zcap));
169 	zone.start = nvme_lba_to_sect(head, le64_to_cpu(entry->zslba));
170 	if (zone.cond == BLK_ZONE_COND_FULL)
171 		zone.wp = zone.start + zone.len;
172 	else
173 		zone.wp = nvme_lba_to_sect(head, le64_to_cpu(entry->wp));
174 
175 	return cb(&zone, idx, data);
176 }
177 
178 int nvme_ns_report_zones(struct nvme_ns *ns, sector_t sector,
179 		unsigned int nr_zones, report_zones_cb cb, void *data)
180 {
181 	struct nvme_zone_report *report;
182 	struct nvme_command c = { };
183 	int ret, zone_idx = 0;
184 	unsigned int nz, i;
185 	size_t buflen;
186 
187 	if (ns->head->ids.csi != NVME_CSI_ZNS)
188 		return -EINVAL;
189 
190 	report = nvme_zns_alloc_report_buffer(ns, nr_zones, &buflen);
191 	if (!report)
192 		return -ENOMEM;
193 
194 	c.zmr.opcode = nvme_cmd_zone_mgmt_recv;
195 	c.zmr.nsid = cpu_to_le32(ns->head->ns_id);
196 	c.zmr.numd = cpu_to_le32(nvme_bytes_to_numd(buflen));
197 	c.zmr.zra = NVME_ZRA_ZONE_REPORT;
198 	c.zmr.zrasf = NVME_ZRASF_ZONE_REPORT_ALL;
199 	c.zmr.pr = NVME_REPORT_ZONE_PARTIAL;
200 
201 	sector &= ~(ns->head->zsze - 1);
202 	while (zone_idx < nr_zones && sector < get_capacity(ns->disk)) {
203 		memset(report, 0, buflen);
204 
205 		c.zmr.slba = cpu_to_le64(nvme_sect_to_lba(ns->head, sector));
206 		ret = nvme_submit_sync_cmd(ns->queue, &c, report, buflen);
207 		if (ret) {
208 			if (ret > 0)
209 				ret = -EIO;
210 			goto out_free;
211 		}
212 
213 		nz = min((unsigned int)le64_to_cpu(report->nr_zones), nr_zones);
214 		if (!nz)
215 			break;
216 
217 		for (i = 0; i < nz && zone_idx < nr_zones; i++) {
218 			ret = nvme_zone_parse_entry(ns->ctrl, ns->head,
219 						    &report->entries[i],
220 						    zone_idx, cb, data);
221 			if (ret)
222 				goto out_free;
223 			zone_idx++;
224 		}
225 
226 		sector += ns->head->zsze * nz;
227 	}
228 
229 	if (zone_idx > 0)
230 		ret = zone_idx;
231 	else
232 		ret = -EINVAL;
233 out_free:
234 	kvfree(report);
235 	return ret;
236 }
237 
238 blk_status_t nvme_setup_zone_mgmt_send(struct nvme_ns *ns, struct request *req,
239 		struct nvme_command *c, enum nvme_zone_mgmt_action action)
240 {
241 	memset(c, 0, sizeof(*c));
242 
243 	c->zms.opcode = nvme_cmd_zone_mgmt_send;
244 	c->zms.nsid = cpu_to_le32(ns->head->ns_id);
245 	c->zms.slba = cpu_to_le64(nvme_sect_to_lba(ns->head, blk_rq_pos(req)));
246 	c->zms.zsa = action;
247 
248 	if (req_op(req) == REQ_OP_ZONE_RESET_ALL)
249 		c->zms.select_all = 1;
250 
251 	return BLK_STS_OK;
252 }
253