1// SPDX-License-Identifier: GPL-2.0
2/*
3 * Driver for handling the PCIe controller errors on
4 * HiSilicon HIP SoCs.
5 *
6 * Copyright (c) 2020 HiSilicon Limited.
7 */
8
9#include <linux/acpi.h>
10#include <acpi/ghes.h>
11#include <linux/bitops.h>
12#include <linux/delay.h>
13#include <linux/pci.h>
14#include <linux/platform_device.h>
15#include <linux/kfifo.h>
16#include <linux/spinlock.h>
17
18/* HISI PCIe controller error definitions */
19#define HISI_PCIE_ERR_MISC_REGS 33
20
21#define HISI_PCIE_LOCAL_VALID_VERSION BIT(0)
22#define HISI_PCIE_LOCAL_VALID_SOC_ID BIT(1)
23#define HISI_PCIE_LOCAL_VALID_SOCKET_ID BIT(2)
24#define HISI_PCIE_LOCAL_VALID_NIMBUS_ID BIT(3)
25#define HISI_PCIE_LOCAL_VALID_SUB_MODULE_ID BIT(4)
26#define HISI_PCIE_LOCAL_VALID_CORE_ID BIT(5)
27#define HISI_PCIE_LOCAL_VALID_PORT_ID BIT(6)
28#define HISI_PCIE_LOCAL_VALID_ERR_TYPE BIT(7)
29#define HISI_PCIE_LOCAL_VALID_ERR_SEVERITY BIT(8)
30#define HISI_PCIE_LOCAL_VALID_ERR_MISC 9
31
32static guid_t hisi_pcie_sec_guid =
33 GUID_INIT(0xB2889FC9, 0xE7D7, 0x4F9D,
34 0xA8, 0x67, 0xAF, 0x42, 0xE9, 0x8B, 0xE7, 0x72);
35
36/*
37 * Firmware reports the socket port ID where the error occurred. These
38 * macros convert that to the core ID and core port ID required by the
39 * ACPI reset method.
40 */
41#define HISI_PCIE_PORT_ID(core, v) (((v) >> 1) + ((core) << 3))
42#define HISI_PCIE_CORE_ID(v) ((v) >> 3)
43#define HISI_PCIE_CORE_PORT_ID(v) (((v) & 7) << 1)
44
45struct hisi_pcie_error_data {
46 u64 val_bits;
47 u8 version;
48 u8 soc_id;
49 u8 socket_id;
50 u8 nimbus_id;
51 u8 sub_module_id;
52 u8 core_id;
53 u8 port_id;
54 u8 err_severity;
55 u16 err_type;
56 u8 reserv[2];
57 u32 err_misc[HISI_PCIE_ERR_MISC_REGS];
58};
59
60struct hisi_pcie_error_private {
61 struct notifier_block nb;
62 struct device *dev;
63};
64
65enum hisi_pcie_submodule_id {
66 HISI_PCIE_SUB_MODULE_ID_AP,
67 HISI_PCIE_SUB_MODULE_ID_TL,
68 HISI_PCIE_SUB_MODULE_ID_MAC,
69 HISI_PCIE_SUB_MODULE_ID_DL,
70 HISI_PCIE_SUB_MODULE_ID_SDI,
71};
72
73static const char * const hisi_pcie_sub_module[] = {
74 [HISI_PCIE_SUB_MODULE_ID_AP] = "AP Layer",
75 [HISI_PCIE_SUB_MODULE_ID_TL] = "TL Layer",
76 [HISI_PCIE_SUB_MODULE_ID_MAC] = "MAC Layer",
77 [HISI_PCIE_SUB_MODULE_ID_DL] = "DL Layer",
78 [HISI_PCIE_SUB_MODULE_ID_SDI] = "SDI Layer",
79};
80
81enum hisi_pcie_err_severity {
82 HISI_PCIE_ERR_SEV_RECOVERABLE,
83 HISI_PCIE_ERR_SEV_FATAL,
84 HISI_PCIE_ERR_SEV_CORRECTED,
85 HISI_PCIE_ERR_SEV_NONE,
86};
87
88static const char * const hisi_pcie_error_sev[] = {
89 [HISI_PCIE_ERR_SEV_RECOVERABLE] = "recoverable",
90 [HISI_PCIE_ERR_SEV_FATAL] = "fatal",
91 [HISI_PCIE_ERR_SEV_CORRECTED] = "corrected",
92 [HISI_PCIE_ERR_SEV_NONE] = "none",
93};
94
95static const char *hisi_pcie_get_string(const char * const *array,
96 size_t n, u32 id)
97{
98 u32 index;
99
100 for (index = 0; index < n; index++) {
101 if (index == id && array[index])
102 return array[index];
103 }
104
105 return "unknown";
106}
107
108static int hisi_pcie_port_reset(struct platform_device *pdev,
109 u32 chip_id, u32 port_id)
110{
111 struct device *dev = &pdev->dev;
112 acpi_handle handle = ACPI_HANDLE(dev);
113 union acpi_object arg[3];
114 struct acpi_object_list arg_list;
115 acpi_status s;
116 unsigned long long data = 0;
117
118 arg[0].type = ACPI_TYPE_INTEGER;
119 arg[0].integer.value = chip_id;
120 arg[1].type = ACPI_TYPE_INTEGER;
121 arg[1].integer.value = HISI_PCIE_CORE_ID(port_id);
122 arg[2].type = ACPI_TYPE_INTEGER;
123 arg[2].integer.value = HISI_PCIE_CORE_PORT_ID(port_id);
124
125 arg_list.count = 3;
126 arg_list.pointer = arg;
127
128 s = acpi_evaluate_integer(handle, pathname: "RST", arguments: &arg_list, data: &data);
129 if (ACPI_FAILURE(s)) {
130 dev_err(dev, "No RST method\n");
131 return -EIO;
132 }
133
134 if (data) {
135 dev_err(dev, "Failed to Reset\n");
136 return -EIO;
137 }
138
139 return 0;
140}
141
142static int hisi_pcie_port_do_recovery(struct platform_device *dev,
143 u32 chip_id, u32 port_id)
144{
145 acpi_status s;
146 struct device *device = &dev->dev;
147 acpi_handle root_handle = ACPI_HANDLE(device);
148 struct acpi_pci_root *pci_root;
149 struct pci_bus *root_bus;
150 struct pci_dev *pdev;
151 u32 domain, busnr, devfn;
152
153 s = acpi_get_parent(object: root_handle, out_handle: &root_handle);
154 if (ACPI_FAILURE(s))
155 return -ENODEV;
156 pci_root = acpi_pci_find_root(handle: root_handle);
157 if (!pci_root)
158 return -ENODEV;
159 root_bus = pci_root->bus;
160 domain = pci_root->segment;
161
162 busnr = root_bus->number;
163 devfn = PCI_DEVFN(port_id, 0);
164 pdev = pci_get_domain_bus_and_slot(domain, bus: busnr, devfn);
165 if (!pdev) {
166 dev_info(device, "Fail to get root port %04x:%02x:%02x.%d device\n",
167 domain, busnr, PCI_SLOT(devfn), PCI_FUNC(devfn));
168 return -ENODEV;
169 }
170
171 pci_stop_and_remove_bus_device_locked(dev: pdev);
172 pci_dev_put(dev: pdev);
173
174 if (hisi_pcie_port_reset(pdev: dev, chip_id, port_id))
175 return -EIO;
176
177 /*
178 * The initialization time of subordinate devices after
179 * hot reset is no more than 1s, which is required by
180 * the PCI spec v5.0 sec 6.6.1. The time will shorten
181 * if Readiness Notifications mechanisms are used. But
182 * wait 1s here to adapt any conditions.
183 */
184 ssleep(seconds: 1UL);
185
186 /* add root port and downstream devices */
187 pci_lock_rescan_remove();
188 pci_rescan_bus(bus: root_bus);
189 pci_unlock_rescan_remove();
190
191 return 0;
192}
193
194static void hisi_pcie_handle_error(struct platform_device *pdev,
195 const struct hisi_pcie_error_data *edata)
196{
197 struct device *dev = &pdev->dev;
198 int idx, rc;
199 const unsigned long valid_bits[] = {BITMAP_FROM_U64(edata->val_bits)};
200
201 if (edata->val_bits == 0) {
202 dev_warn(dev, "%s: no valid error information\n", __func__);
203 return;
204 }
205
206 dev_info(dev, "\nHISI : HIP : PCIe controller error\n");
207 if (edata->val_bits & HISI_PCIE_LOCAL_VALID_SOC_ID)
208 dev_info(dev, "Table version = %d\n", edata->version);
209 if (edata->val_bits & HISI_PCIE_LOCAL_VALID_SOCKET_ID)
210 dev_info(dev, "Socket ID = %d\n", edata->socket_id);
211 if (edata->val_bits & HISI_PCIE_LOCAL_VALID_NIMBUS_ID)
212 dev_info(dev, "Nimbus ID = %d\n", edata->nimbus_id);
213 if (edata->val_bits & HISI_PCIE_LOCAL_VALID_SUB_MODULE_ID)
214 dev_info(dev, "Sub Module = %s\n",
215 hisi_pcie_get_string(hisi_pcie_sub_module,
216 ARRAY_SIZE(hisi_pcie_sub_module),
217 edata->sub_module_id));
218 if (edata->val_bits & HISI_PCIE_LOCAL_VALID_CORE_ID)
219 dev_info(dev, "Core ID = core%d\n", edata->core_id);
220 if (edata->val_bits & HISI_PCIE_LOCAL_VALID_PORT_ID)
221 dev_info(dev, "Port ID = port%d\n", edata->port_id);
222 if (edata->val_bits & HISI_PCIE_LOCAL_VALID_ERR_SEVERITY)
223 dev_info(dev, "Error severity = %s\n",
224 hisi_pcie_get_string(hisi_pcie_error_sev,
225 ARRAY_SIZE(hisi_pcie_error_sev),
226 edata->err_severity));
227 if (edata->val_bits & HISI_PCIE_LOCAL_VALID_ERR_TYPE)
228 dev_info(dev, "Error type = 0x%x\n", edata->err_type);
229
230 dev_info(dev, "Reg Dump:\n");
231 idx = HISI_PCIE_LOCAL_VALID_ERR_MISC;
232 for_each_set_bit_from(idx, valid_bits,
233 HISI_PCIE_LOCAL_VALID_ERR_MISC + HISI_PCIE_ERR_MISC_REGS)
234 dev_info(dev, "ERR_MISC_%d = 0x%x\n", idx - HISI_PCIE_LOCAL_VALID_ERR_MISC,
235 edata->err_misc[idx - HISI_PCIE_LOCAL_VALID_ERR_MISC]);
236
237 if (edata->err_severity != HISI_PCIE_ERR_SEV_RECOVERABLE)
238 return;
239
240 /* Recovery for the PCIe controller errors, try reset
241 * PCI port for the error recovery
242 */
243 rc = hisi_pcie_port_do_recovery(dev: pdev, chip_id: edata->socket_id,
244 HISI_PCIE_PORT_ID(edata->core_id, edata->port_id));
245 if (rc)
246 dev_info(dev, "fail to do hisi pcie port reset\n");
247}
248
249static int hisi_pcie_notify_error(struct notifier_block *nb,
250 unsigned long event, void *data)
251{
252 struct acpi_hest_generic_data *gdata = data;
253 const struct hisi_pcie_error_data *error_data = acpi_hest_get_payload(gdata);
254 struct hisi_pcie_error_private *priv;
255 struct device *dev;
256 struct platform_device *pdev;
257 guid_t err_sec_guid;
258 u8 socket;
259
260 import_guid(dst: &err_sec_guid, src: gdata->section_type);
261 if (!guid_equal(u1: &err_sec_guid, u2: &hisi_pcie_sec_guid))
262 return NOTIFY_DONE;
263
264 priv = container_of(nb, struct hisi_pcie_error_private, nb);
265 dev = priv->dev;
266
267 if (device_property_read_u8(dev, propname: "socket", val: &socket))
268 return NOTIFY_DONE;
269
270 if (error_data->socket_id != socket)
271 return NOTIFY_DONE;
272
273 pdev = container_of(dev, struct platform_device, dev);
274 hisi_pcie_handle_error(pdev, edata: error_data);
275
276 return NOTIFY_OK;
277}
278
279static int hisi_pcie_error_handler_probe(struct platform_device *pdev)
280{
281 struct hisi_pcie_error_private *priv;
282 int ret;
283
284 priv = devm_kzalloc(dev: &pdev->dev, size: sizeof(*priv), GFP_KERNEL);
285 if (!priv)
286 return -ENOMEM;
287
288 priv->nb.notifier_call = hisi_pcie_notify_error;
289 priv->dev = &pdev->dev;
290 ret = ghes_register_vendor_record_notifier(nb: &priv->nb);
291 if (ret) {
292 dev_err(&pdev->dev,
293 "Failed to register hisi pcie controller error handler with apei\n");
294 return ret;
295 }
296
297 platform_set_drvdata(pdev, data: priv);
298
299 return 0;
300}
301
302static void hisi_pcie_error_handler_remove(struct platform_device *pdev)
303{
304 struct hisi_pcie_error_private *priv = platform_get_drvdata(pdev);
305
306 ghes_unregister_vendor_record_notifier(nb: &priv->nb);
307}
308
309static const struct acpi_device_id hisi_pcie_acpi_match[] = {
310 { "HISI0361", 0 },
311 { }
312};
313
314static struct platform_driver hisi_pcie_error_handler_driver = {
315 .driver = {
316 .name = "hisi-pcie-error-handler",
317 .acpi_match_table = hisi_pcie_acpi_match,
318 },
319 .probe = hisi_pcie_error_handler_probe,
320 .remove_new = hisi_pcie_error_handler_remove,
321};
322module_platform_driver(hisi_pcie_error_handler_driver);
323
324MODULE_DESCRIPTION("HiSilicon HIP PCIe controller error handling driver");
325

source code of linux/drivers/pci/controller/pcie-hisi-error.c