1 // SPDX-License-Identifier: (GPL-2.0 OR BSD-3-Clause)
2 /* Copyright(c) 2020 Intel Corporation. All rights reserved. */
3 #include <linux/debugfs.h>
4 #include <linux/delay.h>
5 #include <linux/init.h>
6 #include <linux/interrupt.h>
7 #include <linux/module.h>
9 #include <linux/random.h>
10 #include <linux/slab.h>
11 #include <linux/ntb.h>
12 #include <linux/log2.h>
14 #include "ntb_hw_intel.h"
15 #include "ntb_hw_gen1.h"
16 #include "ntb_hw_gen3.h"
17 #include "ntb_hw_gen4.h"
19 static int gen4_poll_link(struct intel_ntb_dev *ndev);
20 static int gen4_link_is_up(struct intel_ntb_dev *ndev);
22 static const struct intel_ntb_reg gen4_reg = {
23 .poll_link = gen4_poll_link,
24 .link_is_up = gen4_link_is_up,
25 .db_ioread = gen3_db_ioread,
26 .db_iowrite = gen3_db_iowrite,
27 .db_size = sizeof(u32),
28 .ntb_ctl = GEN4_NTBCNTL_OFFSET,
32 static const struct intel_ntb_alt_reg gen4_pri_reg = {
33 .db_clear = GEN4_IM_INT_STATUS_OFFSET,
34 .db_mask = GEN4_IM_INT_DISABLE_OFFSET,
35 .spad = GEN4_IM_SPAD_OFFSET,
38 static const struct intel_ntb_xlat_reg gen4_sec_xlat = {
39 .bar2_limit = GEN4_IM23XLMT_OFFSET,
40 .bar2_xlat = GEN4_IM23XBASE_OFFSET,
41 .bar2_idx = GEN4_IM23XBASEIDX_OFFSET,
44 static const struct intel_ntb_alt_reg gen4_b2b_reg = {
45 .db_bell = GEN4_IM_DOORBELL_OFFSET,
46 .spad = GEN4_EM_SPAD_OFFSET,
49 static int gen4_poll_link(struct intel_ntb_dev *ndev)
54 * We need to write to DLLSCS bit in the SLOTSTS before we
55 * can clear the hardware link interrupt on ICX NTB.
57 iowrite16(GEN4_SLOTSTS_DLLSCS, ndev->self_mmio + GEN4_SLOTSTS);
58 ndev->reg->db_iowrite(ndev->db_link_mask,
60 ndev->self_reg->db_clear);
62 reg_val = ioread16(ndev->self_mmio + GEN4_LINK_STATUS_OFFSET);
63 if (reg_val == ndev->lnk_sta)
66 ndev->lnk_sta = reg_val;
71 static int gen4_link_is_up(struct intel_ntb_dev *ndev)
73 return NTB_LNK_STA_ACTIVE(ndev->lnk_sta);
76 static int gen4_init_isr(struct intel_ntb_dev *ndev)
81 * The MSIX vectors and the interrupt status bits are not lined up
82 * on Gen3 (Skylake) and Gen4. By default the link status bit is bit
83 * 32, however it is by default MSIX vector0. We need to fixup to
84 * line them up. The vectors at reset is 1-32,0. We need to reprogram
87 for (i = 0; i < GEN4_DB_MSIX_VECTOR_COUNT; i++)
88 iowrite8(i, ndev->self_mmio + GEN4_INTVEC_OFFSET + i);
90 return ndev_init_isr(ndev, GEN4_DB_MSIX_VECTOR_COUNT,
91 GEN4_DB_MSIX_VECTOR_COUNT,
92 GEN4_DB_MSIX_VECTOR_SHIFT,
96 static int gen4_setup_b2b_mw(struct intel_ntb_dev *ndev,
97 const struct intel_b2b_addr *addr,
98 const struct intel_b2b_addr *peer_addr)
100 struct pci_dev *pdev;
102 phys_addr_t bar_addr;
104 pdev = ndev->ntb.pdev;
105 mmio = ndev->self_mmio;
107 /* setup incoming bar limits == base addrs (zero length windows) */
108 bar_addr = addr->bar2_addr64;
109 iowrite64(bar_addr, mmio + GEN4_IM23XLMT_OFFSET);
110 bar_addr = ioread64(mmio + GEN4_IM23XLMT_OFFSET);
111 dev_dbg(&pdev->dev, "IM23XLMT %#018llx\n", bar_addr);
113 bar_addr = addr->bar4_addr64;
114 iowrite64(bar_addr, mmio + GEN4_IM45XLMT_OFFSET);
115 bar_addr = ioread64(mmio + GEN4_IM45XLMT_OFFSET);
116 dev_dbg(&pdev->dev, "IM45XLMT %#018llx\n", bar_addr);
118 /* zero incoming translation addrs */
119 iowrite64(0, mmio + GEN4_IM23XBASE_OFFSET);
120 iowrite64(0, mmio + GEN4_IM45XBASE_OFFSET);
122 ndev->peer_mmio = ndev->self_mmio;
127 static int gen4_init_ntb(struct intel_ntb_dev *ndev)
132 ndev->mw_count = XEON_MW_COUNT;
133 ndev->spad_count = GEN4_SPAD_COUNT;
134 ndev->db_count = GEN4_DB_COUNT;
135 ndev->db_link_mask = GEN4_DB_LINK_BIT;
137 ndev->self_reg = &gen4_pri_reg;
138 ndev->xlat_reg = &gen4_sec_xlat;
139 ndev->peer_reg = &gen4_b2b_reg;
141 if (ndev->ntb.topo == NTB_TOPO_B2B_USD)
142 rc = gen4_setup_b2b_mw(ndev, &xeon_b2b_dsd_addr,
145 rc = gen4_setup_b2b_mw(ndev, &xeon_b2b_usd_addr,
150 ndev->db_valid_mask = BIT_ULL(ndev->db_count) - 1;
152 ndev->reg->db_iowrite(ndev->db_valid_mask,
154 ndev->self_reg->db_mask);
159 static enum ntb_topo gen4_ppd_topo(struct intel_ntb_dev *ndev, u32 ppd)
161 switch (ppd & GEN4_PPD_TOPO_MASK) {
162 case GEN4_PPD_TOPO_B2B_USD:
163 return NTB_TOPO_B2B_USD;
164 case GEN4_PPD_TOPO_B2B_DSD:
165 return NTB_TOPO_B2B_DSD;
168 return NTB_TOPO_NONE;
171 int gen4_init_dev(struct intel_ntb_dev *ndev)
173 struct pci_dev *pdev = ndev->ntb.pdev;
178 ndev->reg = &gen4_reg;
180 if (pdev_is_ICX(pdev))
181 ndev->hwerr_flags |= NTB_HWERR_BAR_ALIGN;
183 ppd1 = ioread32(ndev->self_mmio + GEN4_PPD1_OFFSET);
184 ndev->ntb.topo = gen4_ppd_topo(ndev, ppd1);
185 dev_dbg(&pdev->dev, "ppd %#x topo %s\n", ppd1,
186 ntb_topo_string(ndev->ntb.topo));
187 if (ndev->ntb.topo == NTB_TOPO_NONE)
190 rc = gen4_init_ntb(ndev);
194 /* init link setup */
195 lnkctl = ioread16(ndev->self_mmio + GEN4_LINK_CTRL_OFFSET);
196 lnkctl |= GEN4_LINK_CTRL_LINK_DISABLE;
197 iowrite16(lnkctl, ndev->self_mmio + GEN4_LINK_CTRL_OFFSET);
199 return gen4_init_isr(ndev);
202 ssize_t ndev_ntb4_debugfs_read(struct file *filp, char __user *ubuf,
203 size_t count, loff_t *offp)
205 struct intel_ntb_dev *ndev;
210 union { u64 v64; u32 v32; u16 v16; } u;
212 ndev = filp->private_data;
213 mmio = ndev->self_mmio;
215 buf_size = min(count, 0x800ul);
217 buf = kmalloc(buf_size, GFP_KERNEL);
223 off += scnprintf(buf + off, buf_size - off,
224 "NTB Device Information:\n");
226 off += scnprintf(buf + off, buf_size - off,
227 "Connection Topology -\t%s\n",
228 ntb_topo_string(ndev->ntb.topo));
230 off += scnprintf(buf + off, buf_size - off,
231 "NTB CTL -\t\t%#06x\n", ndev->ntb_ctl);
232 off += scnprintf(buf + off, buf_size - off,
233 "LNK STA (cached) -\t\t%#06x\n", ndev->lnk_sta);
235 if (!ndev->reg->link_is_up(ndev))
236 off += scnprintf(buf + off, buf_size - off,
237 "Link Status -\t\tDown\n");
239 off += scnprintf(buf + off, buf_size - off,
240 "Link Status -\t\tUp\n");
241 off += scnprintf(buf + off, buf_size - off,
242 "Link Speed -\t\tPCI-E Gen %u\n",
243 NTB_LNK_STA_SPEED(ndev->lnk_sta));
244 off += scnprintf(buf + off, buf_size - off,
245 "Link Width -\t\tx%u\n",
246 NTB_LNK_STA_WIDTH(ndev->lnk_sta));
249 off += scnprintf(buf + off, buf_size - off,
250 "Memory Window Count -\t%u\n", ndev->mw_count);
251 off += scnprintf(buf + off, buf_size - off,
252 "Scratchpad Count -\t%u\n", ndev->spad_count);
253 off += scnprintf(buf + off, buf_size - off,
254 "Doorbell Count -\t%u\n", ndev->db_count);
255 off += scnprintf(buf + off, buf_size - off,
256 "Doorbell Vector Count -\t%u\n", ndev->db_vec_count);
257 off += scnprintf(buf + off, buf_size - off,
258 "Doorbell Vector Shift -\t%u\n", ndev->db_vec_shift);
260 off += scnprintf(buf + off, buf_size - off,
261 "Doorbell Valid Mask -\t%#llx\n", ndev->db_valid_mask);
262 off += scnprintf(buf + off, buf_size - off,
263 "Doorbell Link Mask -\t%#llx\n", ndev->db_link_mask);
264 off += scnprintf(buf + off, buf_size - off,
265 "Doorbell Mask Cached -\t%#llx\n", ndev->db_mask);
267 u.v64 = ndev_db_read(ndev, mmio + ndev->self_reg->db_mask);
268 off += scnprintf(buf + off, buf_size - off,
269 "Doorbell Mask -\t\t%#llx\n", u.v64);
271 off += scnprintf(buf + off, buf_size - off,
272 "\nNTB Incoming XLAT:\n");
274 u.v64 = ioread64(mmio + GEN4_IM23XBASE_OFFSET);
275 off += scnprintf(buf + off, buf_size - off,
276 "IM23XBASE -\t\t%#018llx\n", u.v64);
278 u.v64 = ioread64(mmio + GEN4_IM45XBASE_OFFSET);
279 off += scnprintf(buf + off, buf_size - off,
280 "IM45XBASE -\t\t%#018llx\n", u.v64);
282 u.v64 = ioread64(mmio + GEN4_IM23XLMT_OFFSET);
283 off += scnprintf(buf + off, buf_size - off,
284 "IM23XLMT -\t\t\t%#018llx\n", u.v64);
286 u.v64 = ioread64(mmio + GEN4_IM45XLMT_OFFSET);
287 off += scnprintf(buf + off, buf_size - off,
288 "IM45XLMT -\t\t\t%#018llx\n", u.v64);
290 off += scnprintf(buf + off, buf_size - off,
291 "\nNTB Statistics:\n");
293 off += scnprintf(buf + off, buf_size - off,
294 "\nNTB Hardware Errors:\n");
296 if (!pci_read_config_word(ndev->ntb.pdev,
297 GEN4_DEVSTS_OFFSET, &u.v16))
298 off += scnprintf(buf + off, buf_size - off,
299 "DEVSTS -\t\t%#06x\n", u.v16);
301 u.v16 = ioread16(mmio + GEN4_LINK_STATUS_OFFSET);
302 off += scnprintf(buf + off, buf_size - off,
303 "LNKSTS -\t\t%#06x\n", u.v16);
305 if (!pci_read_config_dword(ndev->ntb.pdev,
306 GEN4_UNCERRSTS_OFFSET, &u.v32))
307 off += scnprintf(buf + off, buf_size - off,
308 "UNCERRSTS -\t\t%#06x\n", u.v32);
310 if (!pci_read_config_dword(ndev->ntb.pdev,
311 GEN4_CORERRSTS_OFFSET, &u.v32))
312 off += scnprintf(buf + off, buf_size - off,
313 "CORERRSTS -\t\t%#06x\n", u.v32);
315 ret = simple_read_from_buffer(ubuf, count, offp, buf, off);
320 static int intel_ntb4_mw_set_trans(struct ntb_dev *ntb, int pidx, int idx,
321 dma_addr_t addr, resource_size_t size)
323 struct intel_ntb_dev *ndev = ntb_ndev(ntb);
324 unsigned long xlat_reg, limit_reg, idx_reg;
325 unsigned short base_idx, reg_val16;
326 resource_size_t bar_size, mw_size;
328 u64 base, limit, reg_val;
331 if (pidx != NTB_DEF_PEER_IDX)
334 if (idx >= ndev->b2b_idx && !ndev->b2b_off)
337 bar = ndev_mw_to_bar(ndev, idx);
341 bar_size = pci_resource_len(ndev->ntb.pdev, bar);
343 if (idx == ndev->b2b_idx)
344 mw_size = bar_size - ndev->b2b_off;
348 if (ndev->hwerr_flags & NTB_HWERR_BAR_ALIGN) {
349 /* hardware requires that addr is aligned to bar size */
350 if (addr & (bar_size - 1))
353 if (addr & (PAGE_SIZE - 1))
357 /* make sure the range fits in the usable mw size */
361 mmio = ndev->self_mmio;
362 xlat_reg = ndev->xlat_reg->bar2_xlat + (idx * 0x10);
363 limit_reg = ndev->xlat_reg->bar2_limit + (idx * 0x10);
364 base = pci_resource_start(ndev->ntb.pdev, bar);
366 /* Set the limit if supported, if size is not mw_size */
367 if (limit_reg && size != mw_size) {
369 base_idx = __ilog2_u64(size);
371 limit = base + mw_size;
372 base_idx = __ilog2_u64(mw_size);
376 /* set and verify setting the translation address */
377 iowrite64(addr, mmio + xlat_reg);
378 reg_val = ioread64(mmio + xlat_reg);
379 if (reg_val != addr) {
380 iowrite64(0, mmio + xlat_reg);
384 dev_dbg(&ntb->pdev->dev, "BAR %d IMXBASE: %#Lx\n", bar, reg_val);
386 /* set and verify setting the limit */
387 iowrite64(limit, mmio + limit_reg);
388 reg_val = ioread64(mmio + limit_reg);
389 if (reg_val != limit) {
390 iowrite64(base, mmio + limit_reg);
391 iowrite64(0, mmio + xlat_reg);
395 dev_dbg(&ntb->pdev->dev, "BAR %d IMXLMT: %#Lx\n", bar, reg_val);
397 if (ndev->hwerr_flags & NTB_HWERR_BAR_ALIGN) {
398 idx_reg = ndev->xlat_reg->bar2_idx + (idx * 0x2);
399 iowrite16(base_idx, mmio + idx_reg);
400 reg_val16 = ioread16(mmio + idx_reg);
401 if (reg_val16 != base_idx) {
402 iowrite64(base, mmio + limit_reg);
403 iowrite64(0, mmio + xlat_reg);
404 iowrite16(0, mmio + idx_reg);
407 dev_dbg(&ntb->pdev->dev, "BAR %d IMBASEIDX: %#x\n", bar, reg_val16);
414 static int intel_ntb4_link_enable(struct ntb_dev *ntb,
415 enum ntb_speed max_speed, enum ntb_width max_width)
417 struct intel_ntb_dev *ndev;
421 ndev = container_of(ntb, struct intel_ntb_dev, ntb);
423 dev_dbg(&ntb->pdev->dev,
424 "Enabling link with max_speed %d max_width %d\n",
425 max_speed, max_width);
427 if (max_speed != NTB_SPEED_AUTO)
428 dev_dbg(&ntb->pdev->dev,
429 "ignoring max_speed %d\n", max_speed);
430 if (max_width != NTB_WIDTH_AUTO)
431 dev_dbg(&ntb->pdev->dev,
432 "ignoring max_width %d\n", max_width);
434 ntb_ctl = NTB_CTL_E2I_BAR23_SNOOP | NTB_CTL_I2E_BAR23_SNOOP;
435 ntb_ctl |= NTB_CTL_E2I_BAR45_SNOOP | NTB_CTL_I2E_BAR45_SNOOP;
436 iowrite32(ntb_ctl, ndev->self_mmio + ndev->reg->ntb_ctl);
438 lnkctl = ioread16(ndev->self_mmio + GEN4_LINK_CTRL_OFFSET);
439 lnkctl &= ~GEN4_LINK_CTRL_LINK_DISABLE;
440 iowrite16(lnkctl, ndev->self_mmio + GEN4_LINK_CTRL_OFFSET);
442 /* start link training in PPD0 */
443 ppd0 = ioread32(ndev->self_mmio + GEN4_PPD0_OFFSET);
444 ppd0 |= GEN4_PPD_LINKTRN;
445 iowrite32(ppd0, ndev->self_mmio + GEN4_PPD0_OFFSET);
447 /* make sure link training has started */
448 ppd0 = ioread32(ndev->self_mmio + GEN4_PPD0_OFFSET);
449 if (!(ppd0 & GEN4_PPD_LINKTRN)) {
450 dev_warn(&ntb->pdev->dev, "Link is not training\n");
459 static int intel_ntb4_link_disable(struct ntb_dev *ntb)
461 struct intel_ntb_dev *ndev;
465 ndev = container_of(ntb, struct intel_ntb_dev, ntb);
467 dev_dbg(&ntb->pdev->dev, "Disabling link\n");
469 /* clear the snoop bits */
470 ntb_cntl = ioread32(ndev->self_mmio + ndev->reg->ntb_ctl);
471 ntb_cntl &= ~(NTB_CTL_E2I_BAR23_SNOOP | NTB_CTL_I2E_BAR23_SNOOP);
472 ntb_cntl &= ~(NTB_CTL_E2I_BAR45_SNOOP | NTB_CTL_I2E_BAR45_SNOOP);
473 iowrite32(ntb_cntl, ndev->self_mmio + ndev->reg->ntb_ctl);
475 lnkctl = ioread16(ndev->self_mmio + GEN4_LINK_CTRL_OFFSET);
476 lnkctl |= GEN4_LINK_CTRL_LINK_DISABLE;
477 iowrite16(lnkctl, ndev->self_mmio + GEN4_LINK_CTRL_OFFSET);
484 static int intel_ntb4_mw_get_align(struct ntb_dev *ntb, int pidx, int idx,
485 resource_size_t *addr_align,
486 resource_size_t *size_align,
487 resource_size_t *size_max)
489 struct intel_ntb_dev *ndev = ntb_ndev(ntb);
490 resource_size_t bar_size, mw_size;
493 if (pidx != NTB_DEF_PEER_IDX)
496 if (idx >= ndev->b2b_idx && !ndev->b2b_off)
499 bar = ndev_mw_to_bar(ndev, idx);
503 bar_size = pci_resource_len(ndev->ntb.pdev, bar);
505 if (idx == ndev->b2b_idx)
506 mw_size = bar_size - ndev->b2b_off;
511 if (ndev->hwerr_flags & NTB_HWERR_BAR_ALIGN)
512 *addr_align = pci_resource_len(ndev->ntb.pdev, bar);
514 *addr_align = PAGE_SIZE;
526 const struct ntb_dev_ops intel_ntb4_ops = {
527 .mw_count = intel_ntb_mw_count,
528 .mw_get_align = intel_ntb4_mw_get_align,
529 .mw_set_trans = intel_ntb4_mw_set_trans,
530 .peer_mw_count = intel_ntb_peer_mw_count,
531 .peer_mw_get_addr = intel_ntb_peer_mw_get_addr,
532 .link_is_up = intel_ntb_link_is_up,
533 .link_enable = intel_ntb4_link_enable,
534 .link_disable = intel_ntb4_link_disable,
535 .db_valid_mask = intel_ntb_db_valid_mask,
536 .db_vector_count = intel_ntb_db_vector_count,
537 .db_vector_mask = intel_ntb_db_vector_mask,
538 .db_read = intel_ntb3_db_read,
539 .db_clear = intel_ntb3_db_clear,
540 .db_set_mask = intel_ntb_db_set_mask,
541 .db_clear_mask = intel_ntb_db_clear_mask,
542 .peer_db_addr = intel_ntb3_peer_db_addr,
543 .peer_db_set = intel_ntb3_peer_db_set,
544 .spad_is_unsafe = intel_ntb_spad_is_unsafe,
545 .spad_count = intel_ntb_spad_count,
546 .spad_read = intel_ntb_spad_read,
547 .spad_write = intel_ntb_spad_write,
548 .peer_spad_addr = intel_ntb_peer_spad_addr,
549 .peer_spad_read = intel_ntb_peer_spad_read,
550 .peer_spad_write = intel_ntb_peer_spad_write,