1// SPDX-License-Identifier: GPL-2.0
2/*
3 * Intel Platform Monitory Technology Telemetry driver
4 *
5 * Copyright (c) 2020, Intel Corporation.
6 * All Rights Reserved.
7 *
8 * Author: "Alexander Duyck" <alexander.h.duyck@linux.intel.com>
9 */
10
11#include <linux/kernel.h>
12#include <linux/io-64-nonatomic-lo-hi.h>
13#include <linux/module.h>
14#include <linux/mm.h>
15#include <linux/pci.h>
16
17#include "../vsec.h"
18#include "class.h"
19
20#define PMT_XA_START		0
21#define PMT_XA_MAX		INT_MAX
22#define PMT_XA_LIMIT		XA_LIMIT(PMT_XA_START, PMT_XA_MAX)
23#define GUID_SPR_PUNIT		0x9956f43f
24
25bool intel_pmt_is_early_client_hw(struct device *dev)
26{
27	struct intel_vsec_device *ivdev = dev_to_ivdev(dev);
28
29	/*
30	 * Early implementations of PMT on client platforms have some
31	 * differences from the server platforms (which use the Out Of Band
32	 * Management Services Module OOBMSM).
33	 */
34	return !!(ivdev->info->quirks & VSEC_QUIRK_EARLY_HW);
35}
36EXPORT_SYMBOL_NS_GPL(intel_pmt_is_early_client_hw, INTEL_PMT);
37
38static inline int
39pmt_memcpy64_fromio(void *to, const u64 __iomem *from, size_t count)
40{
41	int i, remain;
42	u64 *buf = to;
43
44	if (!IS_ALIGNED((unsigned long)from, 8))
45		return -EFAULT;
46
47	for (i = 0; i < count/8; i++)
48		buf[i] = readq(&from[i]);
49
50	/* Copy any remaining bytes */
51	remain = count % 8;
52	if (remain) {
53		u64 tmp = readq(&from[i]);
54
55		memcpy(&buf[i], &tmp, remain);
56	}
57
58	return count;
59}
60
61/*
62 * sysfs
63 */
64static ssize_t
65intel_pmt_read(struct file *filp, struct kobject *kobj,
66	       struct bin_attribute *attr, char *buf, loff_t off,
67	       size_t count)
68{
69	struct intel_pmt_entry *entry = container_of(attr,
70						     struct intel_pmt_entry,
71						     pmt_bin_attr);
72
73	if (off < 0)
74		return -EINVAL;
75
76	if (off >= entry->size)
77		return 0;
78
79	if (count > entry->size - off)
80		count = entry->size - off;
81
82	if (entry->guid == GUID_SPR_PUNIT)
83		/* PUNIT on SPR only supports aligned 64-bit read */
84		count = pmt_memcpy64_fromio(buf, entry->base + off, count);
85	else
86		memcpy_fromio(buf, entry->base + off, count);
87
88	return count;
89}
90
91static int
92intel_pmt_mmap(struct file *filp, struct kobject *kobj,
93		struct bin_attribute *attr, struct vm_area_struct *vma)
94{
95	struct intel_pmt_entry *entry = container_of(attr,
96						     struct intel_pmt_entry,
97						     pmt_bin_attr);
98	unsigned long vsize = vma->vm_end - vma->vm_start;
99	struct device *dev = kobj_to_dev(kobj);
100	unsigned long phys = entry->base_addr;
101	unsigned long pfn = PFN_DOWN(phys);
102	unsigned long psize;
103
104	if (vma->vm_flags & (VM_WRITE | VM_MAYWRITE))
105		return -EROFS;
106
107	psize = (PFN_UP(entry->base_addr + entry->size) - pfn) * PAGE_SIZE;
108	if (vsize > psize) {
109		dev_err(dev, "Requested mmap size is too large\n");
110		return -EINVAL;
111	}
112
113	vma->vm_page_prot = pgprot_noncached(vma->vm_page_prot);
114	if (io_remap_pfn_range(vma, vma->vm_start, pfn,
115		vsize, vma->vm_page_prot))
116		return -EAGAIN;
117
118	return 0;
119}
120
121static ssize_t
122guid_show(struct device *dev, struct device_attribute *attr, char *buf)
123{
124	struct intel_pmt_entry *entry = dev_get_drvdata(dev);
125
126	return sprintf(buf, "0x%x\n", entry->guid);
127}
128static DEVICE_ATTR_RO(guid);
129
130static ssize_t size_show(struct device *dev, struct device_attribute *attr,
131			 char *buf)
132{
133	struct intel_pmt_entry *entry = dev_get_drvdata(dev);
134
135	return sprintf(buf, "%zu\n", entry->size);
136}
137static DEVICE_ATTR_RO(size);
138
139static ssize_t
140offset_show(struct device *dev, struct device_attribute *attr, char *buf)
141{
142	struct intel_pmt_entry *entry = dev_get_drvdata(dev);
143
144	return sprintf(buf, "%lu\n", offset_in_page(entry->base_addr));
145}
146static DEVICE_ATTR_RO(offset);
147
148static struct attribute *intel_pmt_attrs[] = {
149	&dev_attr_guid.attr,
150	&dev_attr_size.attr,
151	&dev_attr_offset.attr,
152	NULL
153};
154ATTRIBUTE_GROUPS(intel_pmt);
155
156static struct class intel_pmt_class = {
157	.name = "intel_pmt",
158	.dev_groups = intel_pmt_groups,
159};
160
161static int intel_pmt_populate_entry(struct intel_pmt_entry *entry,
162				    struct intel_pmt_header *header,
163				    struct device *dev,
164				    struct resource *disc_res)
165{
166	struct pci_dev *pci_dev = to_pci_dev(dev->parent);
167	u8 bir;
168
169	/*
170	 * The base offset should always be 8 byte aligned.
171	 *
172	 * For non-local access types the lower 3 bits of base offset
173	 * contains the index of the base address register where the
174	 * telemetry can be found.
175	 */
176	bir = GET_BIR(header->base_offset);
177
178	/* Local access and BARID only for now */
179	switch (header->access_type) {
180	case ACCESS_LOCAL:
181		if (bir) {
182			dev_err(dev,
183				"Unsupported BAR index %d for access type %d\n",
184				bir, header->access_type);
185			return -EINVAL;
186		}
187		/*
188		 * For access_type LOCAL, the base address is as follows:
189		 * base address = end of discovery region + base offset
190		 */
191		entry->base_addr = disc_res->end + 1 + header->base_offset;
192
193		/*
194		 * Some hardware use a different calculation for the base address
195		 * when access_type == ACCESS_LOCAL. On the these systems
196		 * ACCCESS_LOCAL refers to an address in the same BAR as the
197		 * header but at a fixed offset. But as the header address was
198		 * supplied to the driver, we don't know which BAR it was in.
199		 * So search for the bar whose range includes the header address.
200		 */
201		if (intel_pmt_is_early_client_hw(dev)) {
202			int i;
203
204			entry->base_addr = 0;
205			for (i = 0; i < 6; i++)
206				if (disc_res->start >= pci_resource_start(pci_dev, i) &&
207				   (disc_res->start <= pci_resource_end(pci_dev, i))) {
208					entry->base_addr = pci_resource_start(pci_dev, i) +
209							   header->base_offset;
210					break;
211				}
212			if (!entry->base_addr)
213				return -EINVAL;
214		}
215
216		break;
217	case ACCESS_BARID:
218		/*
219		 * If another BAR was specified then the base offset
220		 * represents the offset within that BAR. SO retrieve the
221		 * address from the parent PCI device and add offset.
222		 */
223		entry->base_addr = pci_resource_start(pci_dev, bir) +
224				   GET_ADDRESS(header->base_offset);
225		break;
226	default:
227		dev_err(dev, "Unsupported access type %d\n",
228			header->access_type);
229		return -EINVAL;
230	}
231
232	entry->guid = header->guid;
233	entry->size = header->size;
234
235	return 0;
236}
237
238static int intel_pmt_dev_register(struct intel_pmt_entry *entry,
239				  struct intel_pmt_namespace *ns,
240				  struct device *parent)
241{
242	struct resource res = {0};
243	struct device *dev;
244	int ret;
245
246	ret = xa_alloc(ns->xa, &entry->devid, entry, PMT_XA_LIMIT, GFP_KERNEL);
247	if (ret)
248		return ret;
249
250	dev = device_create(&intel_pmt_class, parent, MKDEV(0, 0), entry,
251			    "%s%d", ns->name, entry->devid);
252
253	if (IS_ERR(dev)) {
254		dev_err(parent, "Could not create %s%d device node\n",
255			ns->name, entry->devid);
256		ret = PTR_ERR(dev);
257		goto fail_dev_create;
258	}
259
260	entry->kobj = &dev->kobj;
261
262	if (ns->attr_grp) {
263		ret = sysfs_create_group(entry->kobj, ns->attr_grp);
264		if (ret)
265			goto fail_sysfs;
266	}
267
268	/* if size is 0 assume no data buffer, so no file needed */
269	if (!entry->size)
270		return 0;
271
272	res.start = entry->base_addr;
273	res.end = res.start + entry->size - 1;
274	res.flags = IORESOURCE_MEM;
275
276	entry->base = devm_ioremap_resource(dev, &res);
277	if (IS_ERR(entry->base)) {
278		ret = PTR_ERR(entry->base);
279		goto fail_ioremap;
280	}
281
282	sysfs_bin_attr_init(&entry->pmt_bin_attr);
283	entry->pmt_bin_attr.attr.name = ns->name;
284	entry->pmt_bin_attr.attr.mode = 0440;
285	entry->pmt_bin_attr.mmap = intel_pmt_mmap;
286	entry->pmt_bin_attr.read = intel_pmt_read;
287	entry->pmt_bin_attr.size = entry->size;
288
289	ret = sysfs_create_bin_file(&dev->kobj, &entry->pmt_bin_attr);
290	if (!ret)
291		return 0;
292
293fail_ioremap:
294	if (ns->attr_grp)
295		sysfs_remove_group(entry->kobj, ns->attr_grp);
296fail_sysfs:
297	device_unregister(dev);
298fail_dev_create:
299	xa_erase(ns->xa, entry->devid);
300
301	return ret;
302}
303
304int intel_pmt_dev_create(struct intel_pmt_entry *entry, struct intel_pmt_namespace *ns,
305			 struct intel_vsec_device *intel_vsec_dev, int idx)
306{
307	struct device *dev = &intel_vsec_dev->auxdev.dev;
308	struct intel_pmt_header header;
309	struct resource	*disc_res;
310	int ret;
311
312	disc_res = &intel_vsec_dev->resource[idx];
313
314	entry->disc_table = devm_ioremap_resource(dev, disc_res);
315	if (IS_ERR(entry->disc_table))
316		return PTR_ERR(entry->disc_table);
317
318	ret = ns->pmt_header_decode(entry, &header, dev);
319	if (ret)
320		return ret;
321
322	ret = intel_pmt_populate_entry(entry, &header, dev, disc_res);
323	if (ret)
324		return ret;
325
326	return intel_pmt_dev_register(entry, ns, dev);
327
328}
329EXPORT_SYMBOL_NS_GPL(intel_pmt_dev_create, INTEL_PMT);
330
331void intel_pmt_dev_destroy(struct intel_pmt_entry *entry,
332			   struct intel_pmt_namespace *ns)
333{
334	struct device *dev = kobj_to_dev(entry->kobj);
335
336	if (entry->size)
337		sysfs_remove_bin_file(entry->kobj, &entry->pmt_bin_attr);
338
339	if (ns->attr_grp)
340		sysfs_remove_group(entry->kobj, ns->attr_grp);
341
342	device_unregister(dev);
343	xa_erase(ns->xa, entry->devid);
344}
345EXPORT_SYMBOL_NS_GPL(intel_pmt_dev_destroy, INTEL_PMT);
346
347static int __init pmt_class_init(void)
348{
349	return class_register(&intel_pmt_class);
350}
351
352static void __exit pmt_class_exit(void)
353{
354	class_unregister(&intel_pmt_class);
355}
356
357module_init(pmt_class_init);
358module_exit(pmt_class_exit);
359
360MODULE_AUTHOR("Alexander Duyck <alexander.h.duyck@linux.intel.com>");
361MODULE_DESCRIPTION("Intel PMT Class driver");
362MODULE_LICENSE("GPL v2");
363