Commit | Line | Data |
---|---|---|
2ed6dc34 SN |
1 | /* |
2 | * Intel I/OAT DMA Linux driver | |
3 | * Copyright(c) 2007 Intel Corporation. | |
4 | * | |
5 | * This program is free software; you can redistribute it and/or modify it | |
6 | * under the terms and conditions of the GNU General Public License, | |
7 | * version 2, as published by the Free Software Foundation. | |
8 | * | |
9 | * This program is distributed in the hope that it will be useful, but WITHOUT | |
10 | * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or | |
11 | * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for | |
12 | * more details. | |
13 | * | |
14 | * You should have received a copy of the GNU General Public License along with | |
15 | * this program; if not, write to the Free Software Foundation, Inc., | |
16 | * 51 Franklin St - Fifth Floor, Boston, MA 02110-1301 USA. | |
17 | * | |
18 | * The full GNU General Public License is included in this distribution in | |
19 | * the file called "COPYING". | |
20 | * | |
21 | */ | |
22 | ||
23 | #include <linux/kernel.h> | |
24 | #include <linux/pci.h> | |
25 | #include <linux/smp.h> | |
26 | #include <linux/interrupt.h> | |
27 | #include <linux/dca.h> | |
28 | ||
29 | /* either a kernel change is needed, or we need something like this in kernel */ | |
30 | #ifndef CONFIG_SMP | |
31 | #include <asm/smp.h> | |
32 | #undef cpu_physical_id | |
33 | #define cpu_physical_id(cpu) (cpuid_ebx(1) >> 24) | |
34 | #endif | |
35 | ||
36 | #include "ioatdma.h" | |
37 | #include "ioatdma_registers.h" | |
38 | ||
39 | /* | |
40 | * Bit 16 of a tag map entry is the "valid" bit, if it is set then bits 0:15 | |
41 | * contain the bit number of the APIC ID to map into the DCA tag. If the valid | |
42 | * bit is not set, then the value must be 0 or 1 and defines the bit in the tag. | |
43 | */ | |
44 | #define DCA_TAG_MAP_VALID 0x80 | |
45 | ||
46 | /* | |
47 | * "Legacy" DCA systems do not implement the DCA register set in the | |
48 | * I/OAT device. Software needs direct support for their tag mappings. | |
49 | */ | |
50 | ||
51 | #define APICID_BIT(x) (DCA_TAG_MAP_VALID | (x)) | |
52 | #define IOAT_TAG_MAP_LEN 8 | |
53 | ||
54 | static u8 ioat_tag_map_BNB[IOAT_TAG_MAP_LEN] = { | |
55 | 1, APICID_BIT(1), APICID_BIT(2), APICID_BIT(2), }; | |
56 | static u8 ioat_tag_map_SCNB[IOAT_TAG_MAP_LEN] = { | |
57 | 1, APICID_BIT(1), APICID_BIT(2), APICID_BIT(2), }; | |
58 | static u8 ioat_tag_map_CNB[IOAT_TAG_MAP_LEN] = { | |
59 | 1, APICID_BIT(1), APICID_BIT(3), APICID_BIT(4), APICID_BIT(2), }; | |
60 | static u8 ioat_tag_map_UNISYS[IOAT_TAG_MAP_LEN] = { 0 }; | |
61 | ||
62 | /* pack PCI B/D/F into a u16 */ | |
63 | static inline u16 dcaid_from_pcidev(struct pci_dev *pci) | |
64 | { | |
65 | return (pci->bus->number << 8) | pci->devfn; | |
66 | } | |
67 | ||
5149fd01 | 68 | static int dca_enabled_in_bios(struct pci_dev *pdev) |
2ed6dc34 SN |
69 | { |
70 | /* CPUID level 9 returns DCA configuration */ | |
71 | /* Bit 0 indicates DCA enabled by the BIOS */ | |
72 | unsigned long cpuid_level_9; | |
73 | int res; | |
74 | ||
75 | cpuid_level_9 = cpuid_eax(9); | |
76 | res = test_bit(0, &cpuid_level_9); | |
77 | if (!res) | |
5149fd01 | 78 | dev_err(&pdev->dev, "DCA is disabled in BIOS\n"); |
2ed6dc34 SN |
79 | |
80 | return res; | |
81 | } | |
82 | ||
5149fd01 | 83 | static int system_has_dca_enabled(struct pci_dev *pdev) |
2ed6dc34 SN |
84 | { |
85 | if (boot_cpu_has(X86_FEATURE_DCA)) | |
5149fd01 | 86 | return dca_enabled_in_bios(pdev); |
2ed6dc34 | 87 | |
5149fd01 | 88 | dev_err(&pdev->dev, "boot cpu doesn't have X86_FEATURE_DCA\n"); |
2ed6dc34 SN |
89 | return 0; |
90 | } | |
91 | ||
92 | struct ioat_dca_slot { | |
93 | struct pci_dev *pdev; /* requester device */ | |
94 | u16 rid; /* requester id, as used by IOAT */ | |
95 | }; | |
96 | ||
97 | #define IOAT_DCA_MAX_REQ 6 | |
98 | ||
99 | struct ioat_dca_priv { | |
100 | void __iomem *iobase; | |
53a0c98e | 101 | void __iomem *dca_base; |
2ed6dc34 SN |
102 | int max_requesters; |
103 | int requester_count; | |
104 | u8 tag_map[IOAT_TAG_MAP_LEN]; | |
105 | struct ioat_dca_slot req_slots[0]; | |
106 | }; | |
107 | ||
108 | /* 5000 series chipset DCA Port Requester ID Table Entry Format | |
109 | * [15:8] PCI-Express Bus Number | |
110 | * [7:3] PCI-Express Device Number | |
111 | * [2:0] PCI-Express Function Number | |
112 | * | |
113 | * 5000 series chipset DCA control register format | |
114 | * [7:1] Reserved (0) | |
115 | * [0] Ignore Function Number | |
116 | */ | |
117 | ||
118 | static int ioat_dca_add_requester(struct dca_provider *dca, struct device *dev) | |
119 | { | |
120 | struct ioat_dca_priv *ioatdca = dca_priv(dca); | |
121 | struct pci_dev *pdev; | |
122 | int i; | |
123 | u16 id; | |
124 | ||
125 | /* This implementation only supports PCI-Express */ | |
126 | if (dev->bus != &pci_bus_type) | |
127 | return -ENODEV; | |
128 | pdev = to_pci_dev(dev); | |
129 | id = dcaid_from_pcidev(pdev); | |
130 | ||
131 | if (ioatdca->requester_count == ioatdca->max_requesters) | |
132 | return -ENODEV; | |
133 | ||
134 | for (i = 0; i < ioatdca->max_requesters; i++) { | |
135 | if (ioatdca->req_slots[i].pdev == NULL) { | |
136 | /* found an empty slot */ | |
137 | ioatdca->requester_count++; | |
138 | ioatdca->req_slots[i].pdev = pdev; | |
139 | ioatdca->req_slots[i].rid = id; | |
140 | writew(id, ioatdca->dca_base + (i * 4)); | |
141 | /* make sure the ignore function bit is off */ | |
142 | writeb(0, ioatdca->dca_base + (i * 4) + 2); | |
143 | return i; | |
144 | } | |
145 | } | |
146 | /* Error, ioatdma->requester_count is out of whack */ | |
147 | return -EFAULT; | |
148 | } | |
149 | ||
150 | static int ioat_dca_remove_requester(struct dca_provider *dca, | |
151 | struct device *dev) | |
152 | { | |
153 | struct ioat_dca_priv *ioatdca = dca_priv(dca); | |
154 | struct pci_dev *pdev; | |
155 | int i; | |
156 | ||
157 | /* This implementation only supports PCI-Express */ | |
158 | if (dev->bus != &pci_bus_type) | |
159 | return -ENODEV; | |
160 | pdev = to_pci_dev(dev); | |
161 | ||
162 | for (i = 0; i < ioatdca->max_requesters; i++) { | |
163 | if (ioatdca->req_slots[i].pdev == pdev) { | |
164 | writew(0, ioatdca->dca_base + (i * 4)); | |
165 | ioatdca->req_slots[i].pdev = NULL; | |
166 | ioatdca->req_slots[i].rid = 0; | |
167 | ioatdca->requester_count--; | |
168 | return i; | |
169 | } | |
170 | } | |
171 | return -ENODEV; | |
172 | } | |
173 | ||
174 | static u8 ioat_dca_get_tag(struct dca_provider *dca, int cpu) | |
175 | { | |
176 | struct ioat_dca_priv *ioatdca = dca_priv(dca); | |
177 | int i, apic_id, bit, value; | |
178 | u8 entry, tag; | |
179 | ||
180 | tag = 0; | |
181 | apic_id = cpu_physical_id(cpu); | |
182 | ||
183 | for (i = 0; i < IOAT_TAG_MAP_LEN; i++) { | |
184 | entry = ioatdca->tag_map[i]; | |
185 | if (entry & DCA_TAG_MAP_VALID) { | |
186 | bit = entry & ~DCA_TAG_MAP_VALID; | |
187 | value = (apic_id & (1 << bit)) ? 1 : 0; | |
188 | } else { | |
189 | value = entry ? 1 : 0; | |
190 | } | |
191 | tag |= (value << i); | |
192 | } | |
193 | return tag; | |
194 | } | |
195 | ||
196 | static struct dca_ops ioat_dca_ops = { | |
197 | .add_requester = ioat_dca_add_requester, | |
198 | .remove_requester = ioat_dca_remove_requester, | |
199 | .get_tag = ioat_dca_get_tag, | |
200 | }; | |
201 | ||
202 | ||
203 | struct dca_provider *ioat_dca_init(struct pci_dev *pdev, void __iomem *iobase) | |
204 | { | |
205 | struct dca_provider *dca; | |
206 | struct ioat_dca_priv *ioatdca; | |
207 | u8 *tag_map = NULL; | |
208 | int i; | |
209 | int err; | |
210 | ||
5149fd01 | 211 | if (!system_has_dca_enabled(pdev)) |
2ed6dc34 SN |
212 | return NULL; |
213 | ||
214 | /* I/OAT v1 systems must have a known tag_map to support DCA */ | |
215 | switch (pdev->vendor) { | |
216 | case PCI_VENDOR_ID_INTEL: | |
217 | switch (pdev->device) { | |
218 | case PCI_DEVICE_ID_INTEL_IOAT: | |
219 | tag_map = ioat_tag_map_BNB; | |
220 | break; | |
221 | case PCI_DEVICE_ID_INTEL_IOAT_CNB: | |
222 | tag_map = ioat_tag_map_CNB; | |
223 | break; | |
224 | case PCI_DEVICE_ID_INTEL_IOAT_SCNB: | |
225 | tag_map = ioat_tag_map_SCNB; | |
226 | break; | |
227 | } | |
228 | break; | |
229 | case PCI_VENDOR_ID_UNISYS: | |
230 | switch (pdev->device) { | |
231 | case PCI_DEVICE_ID_UNISYS_DMA_DIRECTOR: | |
232 | tag_map = ioat_tag_map_UNISYS; | |
233 | break; | |
234 | } | |
235 | break; | |
236 | } | |
237 | if (tag_map == NULL) | |
238 | return NULL; | |
239 | ||
240 | dca = alloc_dca_provider(&ioat_dca_ops, | |
241 | sizeof(*ioatdca) + | |
242 | (sizeof(struct ioat_dca_slot) * IOAT_DCA_MAX_REQ)); | |
243 | if (!dca) | |
244 | return NULL; | |
245 | ||
246 | ioatdca = dca_priv(dca); | |
247 | ioatdca->max_requesters = IOAT_DCA_MAX_REQ; | |
248 | ||
249 | ioatdca->dca_base = iobase + 0x54; | |
250 | ||
251 | /* copy over the APIC ID to DCA tag mapping */ | |
252 | for (i = 0; i < IOAT_TAG_MAP_LEN; i++) | |
253 | ioatdca->tag_map[i] = tag_map[i]; | |
254 | ||
255 | err = register_dca_provider(dca, &pdev->dev); | |
256 | if (err) { | |
257 | free_dca_provider(dca); | |
258 | return NULL; | |
259 | } | |
260 | ||
261 | return dca; | |
262 | } | |
263 | ||
7bb67c14 SN |
264 | |
265 | static int ioat2_dca_add_requester(struct dca_provider *dca, struct device *dev) | |
266 | { | |
267 | struct ioat_dca_priv *ioatdca = dca_priv(dca); | |
268 | struct pci_dev *pdev; | |
269 | int i; | |
270 | u16 id; | |
271 | u16 global_req_table; | |
272 | ||
273 | /* This implementation only supports PCI-Express */ | |
274 | if (dev->bus != &pci_bus_type) | |
275 | return -ENODEV; | |
276 | pdev = to_pci_dev(dev); | |
277 | id = dcaid_from_pcidev(pdev); | |
278 | ||
279 | if (ioatdca->requester_count == ioatdca->max_requesters) | |
280 | return -ENODEV; | |
281 | ||
282 | for (i = 0; i < ioatdca->max_requesters; i++) { | |
283 | if (ioatdca->req_slots[i].pdev == NULL) { | |
284 | /* found an empty slot */ | |
285 | ioatdca->requester_count++; | |
286 | ioatdca->req_slots[i].pdev = pdev; | |
287 | ioatdca->req_slots[i].rid = id; | |
288 | global_req_table = | |
289 | readw(ioatdca->dca_base + IOAT_DCA_GREQID_OFFSET); | |
290 | writel(id | IOAT_DCA_GREQID_VALID, | |
291 | ioatdca->iobase + global_req_table + (i * 4)); | |
292 | return i; | |
293 | } | |
294 | } | |
295 | /* Error, ioatdma->requester_count is out of whack */ | |
296 | return -EFAULT; | |
297 | } | |
298 | ||
299 | static int ioat2_dca_remove_requester(struct dca_provider *dca, | |
300 | struct device *dev) | |
301 | { | |
302 | struct ioat_dca_priv *ioatdca = dca_priv(dca); | |
303 | struct pci_dev *pdev; | |
304 | int i; | |
305 | u16 global_req_table; | |
306 | ||
307 | /* This implementation only supports PCI-Express */ | |
308 | if (dev->bus != &pci_bus_type) | |
309 | return -ENODEV; | |
310 | pdev = to_pci_dev(dev); | |
311 | ||
312 | for (i = 0; i < ioatdca->max_requesters; i++) { | |
313 | if (ioatdca->req_slots[i].pdev == pdev) { | |
314 | global_req_table = | |
315 | readw(ioatdca->dca_base + IOAT_DCA_GREQID_OFFSET); | |
316 | writel(0, ioatdca->iobase + global_req_table + (i * 4)); | |
317 | ioatdca->req_slots[i].pdev = NULL; | |
318 | ioatdca->req_slots[i].rid = 0; | |
319 | ioatdca->requester_count--; | |
320 | return i; | |
321 | } | |
322 | } | |
323 | return -ENODEV; | |
324 | } | |
325 | ||
326 | static u8 ioat2_dca_get_tag(struct dca_provider *dca, int cpu) | |
327 | { | |
328 | u8 tag; | |
329 | ||
330 | tag = ioat_dca_get_tag(dca, cpu); | |
331 | tag = (~tag) & 0x1F; | |
332 | return tag; | |
333 | } | |
334 | ||
335 | static struct dca_ops ioat2_dca_ops = { | |
336 | .add_requester = ioat2_dca_add_requester, | |
337 | .remove_requester = ioat2_dca_remove_requester, | |
338 | .get_tag = ioat2_dca_get_tag, | |
339 | }; | |
340 | ||
53a0c98e | 341 | static int ioat2_dca_count_dca_slots(void __iomem *iobase, u16 dca_offset) |
7bb67c14 SN |
342 | { |
343 | int slots = 0; | |
344 | u32 req; | |
345 | u16 global_req_table; | |
346 | ||
347 | global_req_table = readw(iobase + dca_offset + IOAT_DCA_GREQID_OFFSET); | |
348 | if (global_req_table == 0) | |
349 | return 0; | |
350 | do { | |
351 | req = readl(iobase + global_req_table + (slots * sizeof(u32))); | |
352 | slots++; | |
353 | } while ((req & IOAT_DCA_GREQID_LASTID) == 0); | |
354 | ||
355 | return slots; | |
356 | } | |
357 | ||
358 | struct dca_provider *ioat2_dca_init(struct pci_dev *pdev, void __iomem *iobase) | |
359 | { | |
360 | struct dca_provider *dca; | |
361 | struct ioat_dca_priv *ioatdca; | |
362 | int slots; | |
363 | int i; | |
364 | int err; | |
365 | u32 tag_map; | |
366 | u16 dca_offset; | |
367 | u16 csi_fsb_control; | |
368 | u16 pcie_control; | |
369 | u8 bit; | |
370 | ||
371 | if (!system_has_dca_enabled(pdev)) | |
372 | return NULL; | |
373 | ||
374 | dca_offset = readw(iobase + IOAT_DCAOFFSET_OFFSET); | |
375 | if (dca_offset == 0) | |
376 | return NULL; | |
377 | ||
378 | slots = ioat2_dca_count_dca_slots(iobase, dca_offset); | |
379 | if (slots == 0) | |
380 | return NULL; | |
381 | ||
382 | dca = alloc_dca_provider(&ioat2_dca_ops, | |
383 | sizeof(*ioatdca) | |
384 | + (sizeof(struct ioat_dca_slot) * slots)); | |
385 | if (!dca) | |
386 | return NULL; | |
387 | ||
388 | ioatdca = dca_priv(dca); | |
389 | ioatdca->iobase = iobase; | |
390 | ioatdca->dca_base = iobase + dca_offset; | |
391 | ioatdca->max_requesters = slots; | |
392 | ||
393 | /* some bios might not know to turn these on */ | |
394 | csi_fsb_control = readw(ioatdca->dca_base + IOAT_FSB_CAP_ENABLE_OFFSET); | |
395 | if ((csi_fsb_control & IOAT_FSB_CAP_ENABLE_PREFETCH) == 0) { | |
396 | csi_fsb_control |= IOAT_FSB_CAP_ENABLE_PREFETCH; | |
397 | writew(csi_fsb_control, | |
398 | ioatdca->dca_base + IOAT_FSB_CAP_ENABLE_OFFSET); | |
399 | } | |
400 | pcie_control = readw(ioatdca->dca_base + IOAT_PCI_CAP_ENABLE_OFFSET); | |
401 | if ((pcie_control & IOAT_PCI_CAP_ENABLE_MEMWR) == 0) { | |
402 | pcie_control |= IOAT_PCI_CAP_ENABLE_MEMWR; | |
403 | writew(pcie_control, | |
404 | ioatdca->dca_base + IOAT_PCI_CAP_ENABLE_OFFSET); | |
405 | } | |
406 | ||
407 | ||
408 | /* TODO version, compatibility and configuration checks */ | |
409 | ||
410 | /* copy out the APIC to DCA tag map */ | |
411 | tag_map = readl(ioatdca->dca_base + IOAT_APICID_TAG_MAP_OFFSET); | |
412 | for (i = 0; i < 5; i++) { | |
413 | bit = (tag_map >> (4 * i)) & 0x0f; | |
414 | if (bit < 8) | |
415 | ioatdca->tag_map[i] = bit | DCA_TAG_MAP_VALID; | |
416 | else | |
417 | ioatdca->tag_map[i] = 0; | |
418 | } | |
419 | ||
420 | err = register_dca_provider(dca, &pdev->dev); | |
421 | if (err) { | |
422 | free_dca_provider(dca); | |
423 | return NULL; | |
424 | } | |
425 | ||
426 | return dca; | |
427 | } |