2 * This file is provided under a dual BSD/GPLv2 license. When using or
3 * redistributing this file, you may do so under either license.
7 * Copyright(c) 2012 Intel Corporation. All rights reserved.
8 * Copyright (C) 2015 EMC Corporation. All Rights Reserved.
10 * This program is free software; you can redistribute it and/or modify
11 * it under the terms of version 2 of the GNU General Public License as
12 * published by the Free Software Foundation.
16 * Copyright(c) 2012 Intel Corporation. All rights reserved.
17 * Copyright (C) 2015 EMC Corporation. All Rights Reserved.
19 * Redistribution and use in source and binary forms, with or without
20 * modification, are permitted provided that the following conditions
23 * * Redistributions of source code must retain the above copyright
24 * notice, this list of conditions and the following disclaimer.
25 * * Redistributions in binary form must reproduce the above copy
26 * notice, this list of conditions and the following disclaimer in
27 * the documentation and/or other materials provided with the
29 * * Neither the name of Intel Corporation nor the names of its
30 * contributors may be used to endorse or promote products derived
31 * from this software without specific prior written permission.
33 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
34 * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
35 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
36 * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
37 * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
38 * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
39 * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
40 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
41 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
42 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
43 * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
45 * Intel PCIe NTB Linux driver
47 * Contact Information:
51 #include <linux/debugfs.h>
52 #include <linux/delay.h>
53 #include <linux/init.h>
54 #include <linux/interrupt.h>
55 #include <linux/module.h>
56 #include <linux/pci.h>
57 #include <linux/random.h>
58 #include <linux/slab.h>
59 #include <linux/ntb.h>
61 #include "ntb_hw_intel.h"
63 #define NTB_NAME "ntb_hw_intel"
64 #define NTB_DESC "Intel(R) PCI-E Non-Transparent Bridge Driver"
67 MODULE_DESCRIPTION(NTB_DESC);
68 MODULE_VERSION(NTB_VER);
69 MODULE_LICENSE("Dual BSD/GPL");
70 MODULE_AUTHOR("Intel Corporation");
72 #define bar0_off(base, bar) ((base) + ((bar) << 2))
73 #define bar2_off(base, bar) bar0_off(base, (bar) - 2)
75 static const struct intel_ntb_reg atom_reg;
76 static const struct intel_ntb_alt_reg atom_pri_reg;
77 static const struct intel_ntb_alt_reg atom_sec_reg;
78 static const struct intel_ntb_alt_reg atom_b2b_reg;
79 static const struct intel_ntb_xlat_reg atom_pri_xlat;
80 static const struct intel_ntb_xlat_reg atom_sec_xlat;
81 static const struct intel_ntb_reg xeon_reg;
82 static const struct intel_ntb_alt_reg xeon_pri_reg;
83 static const struct intel_ntb_alt_reg xeon_sec_reg;
84 static const struct intel_ntb_alt_reg xeon_b2b_reg;
85 static const struct intel_ntb_xlat_reg xeon_pri_xlat;
86 static const struct intel_ntb_xlat_reg xeon_sec_xlat;
87 static struct intel_b2b_addr xeon_b2b_usd_addr;
88 static struct intel_b2b_addr xeon_b2b_dsd_addr;
89 static const struct ntb_dev_ops intel_ntb_ops;
91 static const struct file_operations intel_ntb_debugfs_info;
92 static struct dentry *debugfs_dir;
94 static int b2b_mw_idx = -1;
95 module_param(b2b_mw_idx, int, 0644);
96 MODULE_PARM_DESC(b2b_mw_idx, "Use this mw idx to access the peer ntb. A "
97 "value of zero or positive starts from first mw idx, and a "
98 "negative value starts from last mw idx. Both sides MUST "
99 "set the same value here!");
101 static unsigned int b2b_mw_share;
102 module_param(b2b_mw_share, uint, 0644);
103 MODULE_PARM_DESC(b2b_mw_share, "If the b2b mw is large enough, configure the "
104 "ntb so that the peer ntb only occupies the first half of "
105 "the mw, so the second half can still be used as a mw. Both "
106 "sides MUST set the same value here!");
108 module_param_named(xeon_b2b_usd_bar2_addr64,
109 xeon_b2b_usd_addr.bar2_addr64, ullong, 0644);
110 MODULE_PARM_DESC(xeon_b2b_usd_bar2_addr64,
111 "XEON B2B USD BAR 2 64-bit address");
113 module_param_named(xeon_b2b_usd_bar4_addr64,
114 xeon_b2b_usd_addr.bar4_addr64, ullong, 0644);
115 MODULE_PARM_DESC(xeon_b2b_usd_bar2_addr64,
116 "XEON B2B USD BAR 4 64-bit address");
118 module_param_named(xeon_b2b_usd_bar4_addr32,
119 xeon_b2b_usd_addr.bar4_addr32, ullong, 0644);
120 MODULE_PARM_DESC(xeon_b2b_usd_bar2_addr64,
121 "XEON B2B USD split-BAR 4 32-bit address");
123 module_param_named(xeon_b2b_usd_bar5_addr32,
124 xeon_b2b_usd_addr.bar5_addr32, ullong, 0644);
125 MODULE_PARM_DESC(xeon_b2b_usd_bar2_addr64,
126 "XEON B2B USD split-BAR 5 32-bit address");
128 module_param_named(xeon_b2b_dsd_bar2_addr64,
129 xeon_b2b_dsd_addr.bar2_addr64, ullong, 0644);
130 MODULE_PARM_DESC(xeon_b2b_dsd_bar2_addr64,
131 "XEON B2B DSD BAR 2 64-bit address");
133 module_param_named(xeon_b2b_dsd_bar4_addr64,
134 xeon_b2b_dsd_addr.bar4_addr64, ullong, 0644);
135 MODULE_PARM_DESC(xeon_b2b_dsd_bar2_addr64,
136 "XEON B2B DSD BAR 4 64-bit address");
138 module_param_named(xeon_b2b_dsd_bar4_addr32,
139 xeon_b2b_dsd_addr.bar4_addr32, ullong, 0644);
140 MODULE_PARM_DESC(xeon_b2b_dsd_bar2_addr64,
141 "XEON B2B DSD split-BAR 4 32-bit address");
143 module_param_named(xeon_b2b_dsd_bar5_addr32,
144 xeon_b2b_dsd_addr.bar5_addr32, ullong, 0644);
145 MODULE_PARM_DESC(xeon_b2b_dsd_bar2_addr64,
146 "XEON B2B DSD split-BAR 5 32-bit address");
150 #define ioread64 readq
152 #define ioread64 _ioread64
153 static inline u64 _ioread64(void __iomem *mmio)
157 low = ioread32(mmio);
158 high = ioread32(mmio + sizeof(u32));
159 return low | (high << 32);
166 #define iowrite64 writeq
168 #define iowrite64 _iowrite64
169 static inline void _iowrite64(u64 val, void __iomem *mmio)
171 iowrite32(val, mmio);
172 iowrite32(val >> 32, mmio + sizeof(u32));
177 static inline int pdev_is_atom(struct pci_dev *pdev)
179 switch (pdev->device) {
180 case PCI_DEVICE_ID_INTEL_NTB_B2B_BWD:
186 static inline int pdev_is_xeon(struct pci_dev *pdev)
188 switch (pdev->device) {
189 case PCI_DEVICE_ID_INTEL_NTB_SS_JSF:
190 case PCI_DEVICE_ID_INTEL_NTB_SS_SNB:
191 case PCI_DEVICE_ID_INTEL_NTB_SS_IVT:
192 case PCI_DEVICE_ID_INTEL_NTB_SS_HSX:
193 case PCI_DEVICE_ID_INTEL_NTB_PS_JSF:
194 case PCI_DEVICE_ID_INTEL_NTB_PS_SNB:
195 case PCI_DEVICE_ID_INTEL_NTB_PS_IVT:
196 case PCI_DEVICE_ID_INTEL_NTB_PS_HSX:
197 case PCI_DEVICE_ID_INTEL_NTB_B2B_JSF:
198 case PCI_DEVICE_ID_INTEL_NTB_B2B_SNB:
199 case PCI_DEVICE_ID_INTEL_NTB_B2B_IVT:
200 case PCI_DEVICE_ID_INTEL_NTB_B2B_HSX:
206 static inline void ndev_reset_unsafe_flags(struct intel_ntb_dev *ndev)
208 ndev->unsafe_flags = 0;
209 ndev->unsafe_flags_ignore = 0;
211 /* Only B2B has a workaround to avoid SDOORBELL */
212 if (ndev->hwerr_flags & NTB_HWERR_SDOORBELL_LOCKUP)
213 if (!ntb_topo_is_b2b(ndev->ntb.topo))
214 ndev->unsafe_flags |= NTB_UNSAFE_DB;
216 /* No low level workaround to avoid SB01BASE */
217 if (ndev->hwerr_flags & NTB_HWERR_SB01BASE_LOCKUP) {
218 ndev->unsafe_flags |= NTB_UNSAFE_DB;
219 ndev->unsafe_flags |= NTB_UNSAFE_SPAD;
223 static inline int ndev_is_unsafe(struct intel_ntb_dev *ndev,
226 return !!(flag & ndev->unsafe_flags & ~ndev->unsafe_flags_ignore);
229 static inline int ndev_ignore_unsafe(struct intel_ntb_dev *ndev,
232 flag &= ndev->unsafe_flags;
233 ndev->unsafe_flags_ignore |= flag;
238 static int ndev_mw_to_bar(struct intel_ntb_dev *ndev, int idx)
240 if (idx < 0 || idx > ndev->mw_count)
242 return ndev->reg->mw_bar[idx];
245 static inline int ndev_db_addr(struct intel_ntb_dev *ndev,
246 phys_addr_t *db_addr, resource_size_t *db_size,
247 phys_addr_t reg_addr, unsigned long reg)
249 if (ndev_is_unsafe(ndev, NTB_UNSAFE_DB))
250 pr_warn_once("%s: NTB unsafe doorbell access", __func__);
253 *db_addr = reg_addr + reg;
254 dev_dbg(ndev_dev(ndev), "Peer db addr %llx\n", *db_addr);
258 *db_size = ndev->reg->db_size;
259 dev_dbg(ndev_dev(ndev), "Peer db size %llx\n", *db_size);
265 static inline u64 ndev_db_read(struct intel_ntb_dev *ndev,
268 if (ndev_is_unsafe(ndev, NTB_UNSAFE_DB))
269 pr_warn_once("%s: NTB unsafe doorbell access", __func__);
271 return ndev->reg->db_ioread(mmio);
274 static inline int ndev_db_write(struct intel_ntb_dev *ndev, u64 db_bits,
277 if (ndev_is_unsafe(ndev, NTB_UNSAFE_DB))
278 pr_warn_once("%s: NTB unsafe doorbell access", __func__);
280 if (db_bits & ~ndev->db_valid_mask)
283 ndev->reg->db_iowrite(db_bits, mmio);
288 static inline int ndev_db_set_mask(struct intel_ntb_dev *ndev, u64 db_bits,
291 unsigned long irqflags;
293 if (ndev_is_unsafe(ndev, NTB_UNSAFE_DB))
294 pr_warn_once("%s: NTB unsafe doorbell access", __func__);
296 if (db_bits & ~ndev->db_valid_mask)
299 spin_lock_irqsave(&ndev->db_mask_lock, irqflags);
301 ndev->db_mask |= db_bits;
302 ndev->reg->db_iowrite(ndev->db_mask, mmio);
304 spin_unlock_irqrestore(&ndev->db_mask_lock, irqflags);
309 static inline int ndev_db_clear_mask(struct intel_ntb_dev *ndev, u64 db_bits,
312 unsigned long irqflags;
314 if (ndev_is_unsafe(ndev, NTB_UNSAFE_DB))
315 pr_warn_once("%s: NTB unsafe doorbell access", __func__);
317 if (db_bits & ~ndev->db_valid_mask)
320 spin_lock_irqsave(&ndev->db_mask_lock, irqflags);
322 ndev->db_mask &= ~db_bits;
323 ndev->reg->db_iowrite(ndev->db_mask, mmio);
325 spin_unlock_irqrestore(&ndev->db_mask_lock, irqflags);
330 static inline int ndev_vec_mask(struct intel_ntb_dev *ndev, int db_vector)
334 shift = ndev->db_vec_shift;
335 mask = BIT_ULL(shift) - 1;
337 return mask << (shift * db_vector);
340 static inline int ndev_spad_addr(struct intel_ntb_dev *ndev, int idx,
341 phys_addr_t *spad_addr, phys_addr_t reg_addr,
344 if (ndev_is_unsafe(ndev, NTB_UNSAFE_SPAD))
345 pr_warn_once("%s: NTB unsafe scratchpad access", __func__);
347 if (idx < 0 || idx >= ndev->spad_count)
351 *spad_addr = reg_addr + reg + (idx << 2);
352 dev_dbg(ndev_dev(ndev), "Peer spad addr %llx\n", *spad_addr);
358 static inline u32 ndev_spad_read(struct intel_ntb_dev *ndev, int idx,
361 if (ndev_is_unsafe(ndev, NTB_UNSAFE_SPAD))
362 pr_warn_once("%s: NTB unsafe scratchpad access", __func__);
364 if (idx < 0 || idx >= ndev->spad_count)
367 return ioread32(mmio + (idx << 2));
370 static inline int ndev_spad_write(struct intel_ntb_dev *ndev, int idx, u32 val,
373 if (ndev_is_unsafe(ndev, NTB_UNSAFE_SPAD))
374 pr_warn_once("%s: NTB unsafe scratchpad access", __func__);
376 if (idx < 0 || idx >= ndev->spad_count)
379 iowrite32(val, mmio + (idx << 2));
384 static irqreturn_t ndev_interrupt(struct intel_ntb_dev *ndev, int vec)
388 vec_mask = ndev_vec_mask(ndev, vec);
390 dev_dbg(ndev_dev(ndev), "vec %d vec_mask %llx\n", vec, vec_mask);
392 ndev->last_ts = jiffies;
394 if (vec_mask & ndev->db_link_mask) {
395 if (ndev->reg->poll_link(ndev))
396 ntb_link_event(&ndev->ntb);
399 if (vec_mask & ndev->db_valid_mask)
400 ntb_db_event(&ndev->ntb, vec);
405 static irqreturn_t ndev_vec_isr(int irq, void *dev)
407 struct intel_ntb_vec *nvec = dev;
409 return ndev_interrupt(nvec->ndev, nvec->num);
412 static irqreturn_t ndev_irq_isr(int irq, void *dev)
414 struct intel_ntb_dev *ndev = dev;
416 return ndev_interrupt(ndev, irq - ndev_pdev(ndev)->irq);
419 static int ndev_init_isr(struct intel_ntb_dev *ndev,
420 int msix_min, int msix_max,
421 int msix_shift, int total_shift)
423 struct pci_dev *pdev;
424 int rc, i, msix_count, node;
426 pdev = ndev_pdev(ndev);
428 node = dev_to_node(&pdev->dev);
430 /* Mask all doorbell interrupts */
431 ndev->db_mask = ndev->db_valid_mask;
432 ndev->reg->db_iowrite(ndev->db_mask,
434 ndev->self_reg->db_mask);
436 /* Try to set up msix irq */
438 ndev->vec = kzalloc_node(msix_max * sizeof(*ndev->vec),
441 goto err_msix_vec_alloc;
443 ndev->msix = kzalloc_node(msix_max * sizeof(*ndev->msix),
448 for (i = 0; i < msix_max; ++i)
449 ndev->msix[i].entry = i;
451 msix_count = pci_enable_msix_range(pdev, ndev->msix,
454 goto err_msix_enable;
456 for (i = 0; i < msix_count; ++i) {
457 ndev->vec[i].ndev = ndev;
458 ndev->vec[i].num = i;
459 rc = request_irq(ndev->msix[i].vector, ndev_vec_isr, 0,
460 "ndev_vec_isr", &ndev->vec[i]);
462 goto err_msix_request;
465 dev_dbg(ndev_dev(ndev), "Using msix interrupts\n");
466 ndev->db_vec_count = msix_count;
467 ndev->db_vec_shift = msix_shift;
472 free_irq(ndev->msix[i].vector, ndev);
473 pci_disable_msix(pdev);
482 /* Try to set up msi irq */
484 rc = pci_enable_msi(pdev);
488 rc = request_irq(pdev->irq, ndev_irq_isr, 0,
489 "ndev_irq_isr", ndev);
491 goto err_msi_request;
493 dev_dbg(ndev_dev(ndev), "Using msi interrupts\n");
494 ndev->db_vec_count = 1;
495 ndev->db_vec_shift = total_shift;
499 pci_disable_msi(pdev);
502 /* Try to set up intx irq */
506 rc = request_irq(pdev->irq, ndev_irq_isr, IRQF_SHARED,
507 "ndev_irq_isr", ndev);
509 goto err_intx_request;
511 dev_dbg(ndev_dev(ndev), "Using intx interrupts\n");
512 ndev->db_vec_count = 1;
513 ndev->db_vec_shift = total_shift;
520 static void ndev_deinit_isr(struct intel_ntb_dev *ndev)
522 struct pci_dev *pdev;
525 pdev = ndev_pdev(ndev);
527 /* Mask all doorbell interrupts */
528 ndev->db_mask = ndev->db_valid_mask;
529 ndev->reg->db_iowrite(ndev->db_mask,
531 ndev->self_reg->db_mask);
534 i = ndev->db_vec_count;
536 free_irq(ndev->msix[i].vector, &ndev->vec[i]);
537 pci_disable_msix(pdev);
541 free_irq(pdev->irq, ndev);
542 if (pci_dev_msi_enabled(pdev))
543 pci_disable_msi(pdev);
547 static ssize_t ndev_debugfs_read(struct file *filp, char __user *ubuf,
548 size_t count, loff_t *offp)
550 struct intel_ntb_dev *ndev;
555 union { u64 v64; u32 v32; u16 v16; } u;
557 ndev = filp->private_data;
558 mmio = ndev->self_mmio;
560 buf_size = min(count, 0x800ul);
562 buf = kmalloc(buf_size, GFP_KERNEL);
568 off += scnprintf(buf + off, buf_size - off,
569 "NTB Device Information:\n");
571 off += scnprintf(buf + off, buf_size - off,
572 "Connection Topology -\t%s\n",
573 ntb_topo_string(ndev->ntb.topo));
575 off += scnprintf(buf + off, buf_size - off,
576 "B2B Offset -\t\t%#lx\n", ndev->b2b_off);
577 off += scnprintf(buf + off, buf_size - off,
578 "B2B MW Idx -\t\t%d\n", ndev->b2b_idx);
579 off += scnprintf(buf + off, buf_size - off,
580 "BAR4 Split -\t\t%s\n",
581 ndev->bar4_split ? "yes" : "no");
583 off += scnprintf(buf + off, buf_size - off,
584 "NTB CTL -\t\t%#06x\n", ndev->ntb_ctl);
585 off += scnprintf(buf + off, buf_size - off,
586 "LNK STA -\t\t%#06x\n", ndev->lnk_sta);
588 if (!ndev->reg->link_is_up(ndev)) {
589 off += scnprintf(buf + off, buf_size - off,
590 "Link Status -\t\tDown\n");
592 off += scnprintf(buf + off, buf_size - off,
593 "Link Status -\t\tUp\n");
594 off += scnprintf(buf + off, buf_size - off,
595 "Link Speed -\t\tPCI-E Gen %u\n",
596 NTB_LNK_STA_SPEED(ndev->lnk_sta));
597 off += scnprintf(buf + off, buf_size - off,
598 "Link Width -\t\tx%u\n",
599 NTB_LNK_STA_WIDTH(ndev->lnk_sta));
602 off += scnprintf(buf + off, buf_size - off,
603 "Memory Window Count -\t%u\n", ndev->mw_count);
604 off += scnprintf(buf + off, buf_size - off,
605 "Scratchpad Count -\t%u\n", ndev->spad_count);
606 off += scnprintf(buf + off, buf_size - off,
607 "Doorbell Count -\t%u\n", ndev->db_count);
608 off += scnprintf(buf + off, buf_size - off,
609 "Doorbell Vector Count -\t%u\n", ndev->db_vec_count);
610 off += scnprintf(buf + off, buf_size - off,
611 "Doorbell Vector Shift -\t%u\n", ndev->db_vec_shift);
613 off += scnprintf(buf + off, buf_size - off,
614 "Doorbell Valid Mask -\t%#llx\n", ndev->db_valid_mask);
615 off += scnprintf(buf + off, buf_size - off,
616 "Doorbell Link Mask -\t%#llx\n", ndev->db_link_mask);
617 off += scnprintf(buf + off, buf_size - off,
618 "Doorbell Mask Cached -\t%#llx\n", ndev->db_mask);
620 u.v64 = ndev_db_read(ndev, mmio + ndev->self_reg->db_mask);
621 off += scnprintf(buf + off, buf_size - off,
622 "Doorbell Mask -\t\t%#llx\n", u.v64);
624 u.v64 = ndev_db_read(ndev, mmio + ndev->self_reg->db_bell);
625 off += scnprintf(buf + off, buf_size - off,
626 "Doorbell Bell -\t\t%#llx\n", u.v64);
628 off += scnprintf(buf + off, buf_size - off,
629 "\nNTB Incoming XLAT:\n");
631 u.v64 = ioread64(mmio + bar2_off(ndev->xlat_reg->bar2_xlat, 2));
632 off += scnprintf(buf + off, buf_size - off,
633 "XLAT23 -\t\t%#018llx\n", u.v64);
635 if (ndev->bar4_split) {
636 u.v32 = ioread32(mmio + bar2_off(ndev->xlat_reg->bar2_xlat, 4));
637 off += scnprintf(buf + off, buf_size - off,
638 "XLAT4 -\t\t\t%#06x\n", u.v32);
640 u.v32 = ioread32(mmio + bar2_off(ndev->xlat_reg->bar2_xlat, 5));
641 off += scnprintf(buf + off, buf_size - off,
642 "XLAT5 -\t\t\t%#06x\n", u.v32);
644 u.v64 = ioread64(mmio + bar2_off(ndev->xlat_reg->bar2_xlat, 4));
645 off += scnprintf(buf + off, buf_size - off,
646 "XLAT45 -\t\t%#018llx\n", u.v64);
649 u.v64 = ioread64(mmio + bar2_off(ndev->xlat_reg->bar2_limit, 2));
650 off += scnprintf(buf + off, buf_size - off,
651 "LMT23 -\t\t\t%#018llx\n", u.v64);
653 if (ndev->bar4_split) {
654 u.v32 = ioread32(mmio + bar2_off(ndev->xlat_reg->bar2_limit, 4));
655 off += scnprintf(buf + off, buf_size - off,
656 "LMT4 -\t\t\t%#06x\n", u.v32);
657 u.v32 = ioread32(mmio + bar2_off(ndev->xlat_reg->bar2_limit, 5));
658 off += scnprintf(buf + off, buf_size - off,
659 "LMT5 -\t\t\t%#06x\n", u.v32);
661 u.v64 = ioread64(mmio + bar2_off(ndev->xlat_reg->bar2_limit, 4));
662 off += scnprintf(buf + off, buf_size - off,
663 "LMT45 -\t\t\t%#018llx\n", u.v64);
666 if (pdev_is_xeon(ndev->ntb.pdev)) {
667 if (ntb_topo_is_b2b(ndev->ntb.topo)) {
668 off += scnprintf(buf + off, buf_size - off,
669 "\nNTB Outgoing B2B XLAT:\n");
671 u.v64 = ioread64(mmio + XEON_PBAR23XLAT_OFFSET);
672 off += scnprintf(buf + off, buf_size - off,
673 "B2B XLAT23 -\t\t%#018llx\n", u.v64);
675 if (ndev->bar4_split) {
676 u.v32 = ioread32(mmio + XEON_PBAR4XLAT_OFFSET);
677 off += scnprintf(buf + off, buf_size - off,
678 "B2B XLAT4 -\t\t%#06x\n",
680 u.v32 = ioread32(mmio + XEON_PBAR5XLAT_OFFSET);
681 off += scnprintf(buf + off, buf_size - off,
682 "B2B XLAT5 -\t\t%#06x\n",
685 u.v64 = ioread64(mmio + XEON_PBAR45XLAT_OFFSET);
686 off += scnprintf(buf + off, buf_size - off,
687 "B2B XLAT45 -\t\t%#018llx\n",
691 u.v64 = ioread64(mmio + XEON_PBAR23LMT_OFFSET);
692 off += scnprintf(buf + off, buf_size - off,
693 "B2B LMT23 -\t\t%#018llx\n", u.v64);
695 if (ndev->bar4_split) {
696 u.v32 = ioread32(mmio + XEON_PBAR4LMT_OFFSET);
697 off += scnprintf(buf + off, buf_size - off,
698 "B2B LMT4 -\t\t%#06x\n",
700 u.v32 = ioread32(mmio + XEON_PBAR5LMT_OFFSET);
701 off += scnprintf(buf + off, buf_size - off,
702 "B2B LMT5 -\t\t%#06x\n",
705 u.v64 = ioread64(mmio + XEON_PBAR45LMT_OFFSET);
706 off += scnprintf(buf + off, buf_size - off,
707 "B2B LMT45 -\t\t%#018llx\n",
711 off += scnprintf(buf + off, buf_size - off,
712 "\nNTB Secondary BAR:\n");
714 u.v64 = ioread64(mmio + XEON_SBAR0BASE_OFFSET);
715 off += scnprintf(buf + off, buf_size - off,
716 "SBAR01 -\t\t%#018llx\n", u.v64);
718 u.v64 = ioread64(mmio + XEON_SBAR23BASE_OFFSET);
719 off += scnprintf(buf + off, buf_size - off,
720 "SBAR23 -\t\t%#018llx\n", u.v64);
722 if (ndev->bar4_split) {
723 u.v32 = ioread32(mmio + XEON_SBAR4BASE_OFFSET);
724 off += scnprintf(buf + off, buf_size - off,
725 "SBAR4 -\t\t\t%#06x\n", u.v32);
726 u.v32 = ioread32(mmio + XEON_SBAR5BASE_OFFSET);
727 off += scnprintf(buf + off, buf_size - off,
728 "SBAR5 -\t\t\t%#06x\n", u.v32);
730 u.v64 = ioread64(mmio + XEON_SBAR45BASE_OFFSET);
731 off += scnprintf(buf + off, buf_size - off,
732 "SBAR45 -\t\t%#018llx\n",
737 off += scnprintf(buf + off, buf_size - off,
738 "\nXEON NTB Statistics:\n");
740 u.v16 = ioread16(mmio + XEON_USMEMMISS_OFFSET);
741 off += scnprintf(buf + off, buf_size - off,
742 "Upstream Memory Miss -\t%u\n", u.v16);
744 off += scnprintf(buf + off, buf_size - off,
745 "\nXEON NTB Hardware Errors:\n");
747 if (!pci_read_config_word(ndev->ntb.pdev,
748 XEON_DEVSTS_OFFSET, &u.v16))
749 off += scnprintf(buf + off, buf_size - off,
750 "DEVSTS -\t\t%#06x\n", u.v16);
752 if (!pci_read_config_word(ndev->ntb.pdev,
753 XEON_LINK_STATUS_OFFSET, &u.v16))
754 off += scnprintf(buf + off, buf_size - off,
755 "LNKSTS -\t\t%#06x\n", u.v16);
757 if (!pci_read_config_dword(ndev->ntb.pdev,
758 XEON_UNCERRSTS_OFFSET, &u.v32))
759 off += scnprintf(buf + off, buf_size - off,
760 "UNCERRSTS -\t\t%#06x\n", u.v32);
762 if (!pci_read_config_dword(ndev->ntb.pdev,
763 XEON_CORERRSTS_OFFSET, &u.v32))
764 off += scnprintf(buf + off, buf_size - off,
765 "CORERRSTS -\t\t%#06x\n", u.v32);
768 ret = simple_read_from_buffer(ubuf, count, offp, buf, off);
773 static void ndev_init_debugfs(struct intel_ntb_dev *ndev)
776 ndev->debugfs_dir = NULL;
777 ndev->debugfs_info = NULL;
780 debugfs_create_dir(ndev_name(ndev), debugfs_dir);
781 if (!ndev->debugfs_dir)
782 ndev->debugfs_info = NULL;
785 debugfs_create_file("info", S_IRUSR,
786 ndev->debugfs_dir, ndev,
787 &intel_ntb_debugfs_info);
791 static void ndev_deinit_debugfs(struct intel_ntb_dev *ndev)
793 debugfs_remove_recursive(ndev->debugfs_dir);
796 static int intel_ntb_mw_count(struct ntb_dev *ntb)
798 return ntb_ndev(ntb)->mw_count;
801 static int intel_ntb_mw_get_range(struct ntb_dev *ntb, int idx,
803 resource_size_t *size,
804 resource_size_t *align,
805 resource_size_t *align_size)
807 struct intel_ntb_dev *ndev = ntb_ndev(ntb);
810 if (idx >= ndev->b2b_idx && !ndev->b2b_off)
813 bar = ndev_mw_to_bar(ndev, idx);
818 *base = pci_resource_start(ndev->ntb.pdev, bar) +
819 (idx == ndev->b2b_idx ? ndev->b2b_off : 0);
822 *size = pci_resource_len(ndev->ntb.pdev, bar) -
823 (idx == ndev->b2b_idx ? ndev->b2b_off : 0);
826 *align = pci_resource_len(ndev->ntb.pdev, bar);
834 static int intel_ntb_mw_set_trans(struct ntb_dev *ntb, int idx,
835 dma_addr_t addr, resource_size_t size)
837 struct intel_ntb_dev *ndev = ntb_ndev(ntb);
838 unsigned long base_reg, xlat_reg, limit_reg;
839 resource_size_t bar_size, mw_size;
841 u64 base, limit, reg_val;
844 if (idx >= ndev->b2b_idx && !ndev->b2b_off)
847 bar = ndev_mw_to_bar(ndev, idx);
851 bar_size = pci_resource_len(ndev->ntb.pdev, bar);
853 if (idx == ndev->b2b_idx)
854 mw_size = bar_size - ndev->b2b_off;
858 /* hardware requires that addr is aligned to bar size */
859 if (addr & (bar_size - 1))
862 /* make sure the range fits in the usable mw size */
866 mmio = ndev->self_mmio;
867 base_reg = bar0_off(ndev->xlat_reg->bar0_base, bar);
868 xlat_reg = bar2_off(ndev->xlat_reg->bar2_xlat, bar);
869 limit_reg = bar2_off(ndev->xlat_reg->bar2_limit, bar);
871 if (bar < 4 || !ndev->bar4_split) {
872 base = ioread64(mmio + base_reg);
874 /* Set the limit if supported, if size is not mw_size */
875 if (limit_reg && size != mw_size)
880 /* set and verify setting the translation address */
881 iowrite64(addr, mmio + xlat_reg);
882 reg_val = ioread64(mmio + xlat_reg);
883 if (reg_val != addr) {
884 iowrite64(0, mmio + xlat_reg);
888 /* set and verify setting the limit */
889 iowrite64(limit, mmio + limit_reg);
890 reg_val = ioread64(mmio + limit_reg);
891 if (reg_val != limit) {
892 iowrite64(base, mmio + limit_reg);
893 iowrite64(0, mmio + xlat_reg);
897 /* split bar addr range must all be 32 bit */
898 if (addr & (~0ull << 32))
900 if ((addr + size) & (~0ull << 32))
903 base = ioread32(mmio + base_reg);
905 /* Set the limit if supported, if size is not mw_size */
906 if (limit_reg && size != mw_size)
911 /* set and verify setting the translation address */
912 iowrite32(addr, mmio + xlat_reg);
913 reg_val = ioread32(mmio + xlat_reg);
914 if (reg_val != addr) {
915 iowrite32(0, mmio + xlat_reg);
919 /* set and verify setting the limit */
920 iowrite32(limit, mmio + limit_reg);
921 reg_val = ioread32(mmio + limit_reg);
922 if (reg_val != limit) {
923 iowrite32(base, mmio + limit_reg);
924 iowrite32(0, mmio + xlat_reg);
932 static int intel_ntb_link_is_up(struct ntb_dev *ntb,
933 enum ntb_speed *speed,
934 enum ntb_width *width)
936 struct intel_ntb_dev *ndev = ntb_ndev(ntb);
938 if (ndev->reg->link_is_up(ndev)) {
940 *speed = NTB_LNK_STA_SPEED(ndev->lnk_sta);
942 *width = NTB_LNK_STA_WIDTH(ndev->lnk_sta);
945 /* TODO MAYBE: is it possible to observe the link speed and
946 * width while link is training? */
948 *speed = NTB_SPEED_NONE;
950 *width = NTB_WIDTH_NONE;
955 static int intel_ntb_link_enable(struct ntb_dev *ntb,
956 enum ntb_speed max_speed,
957 enum ntb_width max_width)
959 struct intel_ntb_dev *ndev;
962 ndev = container_of(ntb, struct intel_ntb_dev, ntb);
964 if (ndev->ntb.topo == NTB_TOPO_SEC)
967 dev_dbg(ndev_dev(ndev),
968 "Enabling link with max_speed %d max_width %d\n",
969 max_speed, max_width);
970 if (max_speed != NTB_SPEED_AUTO)
971 dev_dbg(ndev_dev(ndev), "ignoring max_speed %d\n", max_speed);
972 if (max_width != NTB_WIDTH_AUTO)
973 dev_dbg(ndev_dev(ndev), "ignoring max_width %d\n", max_width);
975 ntb_ctl = ioread32(ndev->self_mmio + ndev->reg->ntb_ctl);
976 ntb_ctl &= ~(NTB_CTL_DISABLE | NTB_CTL_CFG_LOCK);
977 ntb_ctl |= NTB_CTL_P2S_BAR2_SNOOP | NTB_CTL_S2P_BAR2_SNOOP;
978 ntb_ctl |= NTB_CTL_P2S_BAR4_SNOOP | NTB_CTL_S2P_BAR4_SNOOP;
979 if (ndev->bar4_split)
980 ntb_ctl |= NTB_CTL_P2S_BAR5_SNOOP | NTB_CTL_S2P_BAR5_SNOOP;
981 iowrite32(ntb_ctl, ndev->self_mmio + ndev->reg->ntb_ctl);
986 static int intel_ntb_link_disable(struct ntb_dev *ntb)
988 struct intel_ntb_dev *ndev;
991 ndev = container_of(ntb, struct intel_ntb_dev, ntb);
993 if (ndev->ntb.topo == NTB_TOPO_SEC)
996 dev_dbg(ndev_dev(ndev), "Disabling link\n");
998 /* Bring NTB link down */
999 ntb_cntl = ioread32(ndev->self_mmio + ndev->reg->ntb_ctl);
1000 ntb_cntl &= ~(NTB_CTL_P2S_BAR2_SNOOP | NTB_CTL_S2P_BAR2_SNOOP);
1001 ntb_cntl &= ~(NTB_CTL_P2S_BAR4_SNOOP | NTB_CTL_S2P_BAR4_SNOOP);
1002 if (ndev->bar4_split)
1003 ntb_cntl &= ~(NTB_CTL_P2S_BAR5_SNOOP | NTB_CTL_S2P_BAR5_SNOOP);
1004 ntb_cntl |= NTB_CTL_DISABLE | NTB_CTL_CFG_LOCK;
1005 iowrite32(ntb_cntl, ndev->self_mmio + ndev->reg->ntb_ctl);
1010 static int intel_ntb_db_is_unsafe(struct ntb_dev *ntb)
1012 return ndev_ignore_unsafe(ntb_ndev(ntb), NTB_UNSAFE_DB);
1015 static u64 intel_ntb_db_valid_mask(struct ntb_dev *ntb)
1017 return ntb_ndev(ntb)->db_valid_mask;
1020 static int intel_ntb_db_vector_count(struct ntb_dev *ntb)
1022 struct intel_ntb_dev *ndev;
1024 ndev = container_of(ntb, struct intel_ntb_dev, ntb);
1026 return ndev->db_vec_count;
1029 static u64 intel_ntb_db_vector_mask(struct ntb_dev *ntb, int db_vector)
1031 struct intel_ntb_dev *ndev = ntb_ndev(ntb);
1033 if (db_vector < 0 || db_vector > ndev->db_vec_count)
1036 return ndev->db_valid_mask & ndev_vec_mask(ndev, db_vector);
1039 static u64 intel_ntb_db_read(struct ntb_dev *ntb)
1041 struct intel_ntb_dev *ndev = ntb_ndev(ntb);
1043 return ndev_db_read(ndev,
1045 ndev->self_reg->db_bell);
1048 static int intel_ntb_db_clear(struct ntb_dev *ntb, u64 db_bits)
1050 struct intel_ntb_dev *ndev = ntb_ndev(ntb);
1052 return ndev_db_write(ndev, db_bits,
1054 ndev->self_reg->db_bell);
1057 static int intel_ntb_db_set_mask(struct ntb_dev *ntb, u64 db_bits)
1059 struct intel_ntb_dev *ndev = ntb_ndev(ntb);
1061 return ndev_db_set_mask(ndev, db_bits,
1063 ndev->self_reg->db_mask);
1066 static int intel_ntb_db_clear_mask(struct ntb_dev *ntb, u64 db_bits)
1068 struct intel_ntb_dev *ndev = ntb_ndev(ntb);
1070 return ndev_db_clear_mask(ndev, db_bits,
1072 ndev->self_reg->db_mask);
1075 static int intel_ntb_peer_db_addr(struct ntb_dev *ntb,
1076 phys_addr_t *db_addr,
1077 resource_size_t *db_size)
1079 struct intel_ntb_dev *ndev = ntb_ndev(ntb);
1081 return ndev_db_addr(ndev, db_addr, db_size, ndev->peer_addr,
1082 ndev->peer_reg->db_bell);
1085 static int intel_ntb_peer_db_set(struct ntb_dev *ntb, u64 db_bits)
1087 struct intel_ntb_dev *ndev = ntb_ndev(ntb);
1089 return ndev_db_write(ndev, db_bits,
1091 ndev->peer_reg->db_bell);
1094 static int intel_ntb_spad_is_unsafe(struct ntb_dev *ntb)
1096 return ndev_ignore_unsafe(ntb_ndev(ntb), NTB_UNSAFE_SPAD);
1099 static int intel_ntb_spad_count(struct ntb_dev *ntb)
1101 struct intel_ntb_dev *ndev;
1103 ndev = container_of(ntb, struct intel_ntb_dev, ntb);
1105 return ndev->spad_count;
1108 static u32 intel_ntb_spad_read(struct ntb_dev *ntb, int idx)
1110 struct intel_ntb_dev *ndev = ntb_ndev(ntb);
1112 return ndev_spad_read(ndev, idx,
1114 ndev->self_reg->spad);
1117 static int intel_ntb_spad_write(struct ntb_dev *ntb,
1120 struct intel_ntb_dev *ndev = ntb_ndev(ntb);
1122 return ndev_spad_write(ndev, idx, val,
1124 ndev->self_reg->spad);
1127 static int intel_ntb_peer_spad_addr(struct ntb_dev *ntb, int idx,
1128 phys_addr_t *spad_addr)
1130 struct intel_ntb_dev *ndev = ntb_ndev(ntb);
1132 return ndev_spad_addr(ndev, idx, spad_addr, ndev->peer_addr,
1133 ndev->peer_reg->spad);
1136 static u32 intel_ntb_peer_spad_read(struct ntb_dev *ntb, int idx)
1138 struct intel_ntb_dev *ndev = ntb_ndev(ntb);
1140 return ndev_spad_read(ndev, idx,
1142 ndev->peer_reg->spad);
1145 static int intel_ntb_peer_spad_write(struct ntb_dev *ntb,
1148 struct intel_ntb_dev *ndev = ntb_ndev(ntb);
1150 return ndev_spad_write(ndev, idx, val,
1152 ndev->peer_reg->spad);
1157 static u64 atom_db_ioread(void __iomem *mmio)
1159 return ioread64(mmio);
1162 static void atom_db_iowrite(u64 bits, void __iomem *mmio)
1164 iowrite64(bits, mmio);
1167 static int atom_poll_link(struct intel_ntb_dev *ndev)
1171 ntb_ctl = ioread32(ndev->self_mmio + ATOM_NTBCNTL_OFFSET);
1173 if (ntb_ctl == ndev->ntb_ctl)
1176 ndev->ntb_ctl = ntb_ctl;
1178 ndev->lnk_sta = ioread32(ndev->self_mmio + ATOM_LINK_STATUS_OFFSET);
1183 static int atom_link_is_up(struct intel_ntb_dev *ndev)
1185 return ATOM_NTB_CTL_ACTIVE(ndev->ntb_ctl);
1188 static int atom_link_is_err(struct intel_ntb_dev *ndev)
1190 if (ioread32(ndev->self_mmio + ATOM_LTSSMSTATEJMP_OFFSET)
1191 & ATOM_LTSSMSTATEJMP_FORCEDETECT)
1194 if (ioread32(ndev->self_mmio + ATOM_IBSTERRRCRVSTS0_OFFSET)
1195 & ATOM_IBIST_ERR_OFLOW)
1201 static inline enum ntb_topo atom_ppd_topo(struct intel_ntb_dev *ndev, u32 ppd)
1203 switch (ppd & ATOM_PPD_TOPO_MASK) {
1204 case ATOM_PPD_TOPO_B2B_USD:
1205 dev_dbg(ndev_dev(ndev), "PPD %d B2B USD\n", ppd);
1206 return NTB_TOPO_B2B_USD;
1208 case ATOM_PPD_TOPO_B2B_DSD:
1209 dev_dbg(ndev_dev(ndev), "PPD %d B2B DSD\n", ppd);
1210 return NTB_TOPO_B2B_DSD;
1212 case ATOM_PPD_TOPO_PRI_USD:
1213 case ATOM_PPD_TOPO_PRI_DSD: /* accept bogus PRI_DSD */
1214 case ATOM_PPD_TOPO_SEC_USD:
1215 case ATOM_PPD_TOPO_SEC_DSD: /* accept bogus SEC_DSD */
1216 dev_dbg(ndev_dev(ndev), "PPD %d non B2B disabled\n", ppd);
1217 return NTB_TOPO_NONE;
1220 dev_dbg(ndev_dev(ndev), "PPD %d invalid\n", ppd);
1221 return NTB_TOPO_NONE;
1224 static void atom_link_hb(struct work_struct *work)
1226 struct intel_ntb_dev *ndev = hb_ndev(work);
1227 unsigned long poll_ts;
1231 poll_ts = ndev->last_ts + ATOM_LINK_HB_TIMEOUT;
1233 /* Delay polling the link status if an interrupt was received,
1234 * unless the cached link status says the link is down.
1236 if (time_after(poll_ts, jiffies) && atom_link_is_up(ndev)) {
1237 schedule_delayed_work(&ndev->hb_timer, poll_ts - jiffies);
1241 if (atom_poll_link(ndev))
1242 ntb_link_event(&ndev->ntb);
1244 if (atom_link_is_up(ndev) || !atom_link_is_err(ndev)) {
1245 schedule_delayed_work(&ndev->hb_timer, ATOM_LINK_HB_TIMEOUT);
1249 /* Link is down with error: recover the link! */
1251 mmio = ndev->self_mmio;
1253 /* Driver resets the NTB ModPhy lanes - magic! */
1254 iowrite8(0xe0, mmio + ATOM_MODPHY_PCSREG6);
1255 iowrite8(0x40, mmio + ATOM_MODPHY_PCSREG4);
1256 iowrite8(0x60, mmio + ATOM_MODPHY_PCSREG4);
1257 iowrite8(0x60, mmio + ATOM_MODPHY_PCSREG6);
1259 /* Driver waits 100ms to allow the NTB ModPhy to settle */
1262 /* Clear AER Errors, write to clear */
1263 status32 = ioread32(mmio + ATOM_ERRCORSTS_OFFSET);
1264 dev_dbg(ndev_dev(ndev), "ERRCORSTS = %x\n", status32);
1265 status32 &= PCI_ERR_COR_REP_ROLL;
1266 iowrite32(status32, mmio + ATOM_ERRCORSTS_OFFSET);
1268 /* Clear unexpected electrical idle event in LTSSM, write to clear */
1269 status32 = ioread32(mmio + ATOM_LTSSMERRSTS0_OFFSET);
1270 dev_dbg(ndev_dev(ndev), "LTSSMERRSTS0 = %x\n", status32);
1271 status32 |= ATOM_LTSSMERRSTS0_UNEXPECTEDEI;
1272 iowrite32(status32, mmio + ATOM_LTSSMERRSTS0_OFFSET);
1274 /* Clear DeSkew Buffer error, write to clear */
1275 status32 = ioread32(mmio + ATOM_DESKEWSTS_OFFSET);
1276 dev_dbg(ndev_dev(ndev), "DESKEWSTS = %x\n", status32);
1277 status32 |= ATOM_DESKEWSTS_DBERR;
1278 iowrite32(status32, mmio + ATOM_DESKEWSTS_OFFSET);
1280 status32 = ioread32(mmio + ATOM_IBSTERRRCRVSTS0_OFFSET);
1281 dev_dbg(ndev_dev(ndev), "IBSTERRRCRVSTS0 = %x\n", status32);
1282 status32 &= ATOM_IBIST_ERR_OFLOW;
1283 iowrite32(status32, mmio + ATOM_IBSTERRRCRVSTS0_OFFSET);
1285 /* Releases the NTB state machine to allow the link to retrain */
1286 status32 = ioread32(mmio + ATOM_LTSSMSTATEJMP_OFFSET);
1287 dev_dbg(ndev_dev(ndev), "LTSSMSTATEJMP = %x\n", status32);
1288 status32 &= ~ATOM_LTSSMSTATEJMP_FORCEDETECT;
1289 iowrite32(status32, mmio + ATOM_LTSSMSTATEJMP_OFFSET);
1291 /* There is a potential race between the 2 NTB devices recovering at the
1292 * same time. If the times are the same, the link will not recover and
1293 * the driver will be stuck in this loop forever. Add a random interval
1294 * to the recovery time to prevent this race.
1296 schedule_delayed_work(&ndev->hb_timer, ATOM_LINK_RECOVERY_TIME
1297 + prandom_u32() % ATOM_LINK_RECOVERY_TIME);
1300 static int atom_init_isr(struct intel_ntb_dev *ndev)
1304 rc = ndev_init_isr(ndev, 1, ATOM_DB_MSIX_VECTOR_COUNT,
1305 ATOM_DB_MSIX_VECTOR_SHIFT, ATOM_DB_TOTAL_SHIFT);
1309 /* ATOM doesn't have link status interrupt, poll on that platform */
1310 ndev->last_ts = jiffies;
1311 INIT_DELAYED_WORK(&ndev->hb_timer, atom_link_hb);
1312 schedule_delayed_work(&ndev->hb_timer, ATOM_LINK_HB_TIMEOUT);
1317 static void atom_deinit_isr(struct intel_ntb_dev *ndev)
1319 cancel_delayed_work_sync(&ndev->hb_timer);
1320 ndev_deinit_isr(ndev);
1323 static int atom_init_ntb(struct intel_ntb_dev *ndev)
1325 ndev->mw_count = ATOM_MW_COUNT;
1326 ndev->spad_count = ATOM_SPAD_COUNT;
1327 ndev->db_count = ATOM_DB_COUNT;
1329 switch (ndev->ntb.topo) {
1330 case NTB_TOPO_B2B_USD:
1331 case NTB_TOPO_B2B_DSD:
1332 ndev->self_reg = &atom_pri_reg;
1333 ndev->peer_reg = &atom_b2b_reg;
1334 ndev->xlat_reg = &atom_sec_xlat;
1336 /* Enable Bus Master and Memory Space on the secondary side */
1337 iowrite16(PCI_COMMAND_MEMORY | PCI_COMMAND_MASTER,
1338 ndev->self_mmio + ATOM_SPCICMD_OFFSET);
1346 ndev->db_valid_mask = BIT_ULL(ndev->db_count) - 1;
1351 static int atom_init_dev(struct intel_ntb_dev *ndev)
1356 rc = pci_read_config_dword(ndev->ntb.pdev, ATOM_PPD_OFFSET, &ppd);
1360 ndev->ntb.topo = atom_ppd_topo(ndev, ppd);
1361 if (ndev->ntb.topo == NTB_TOPO_NONE)
1364 rc = atom_init_ntb(ndev);
1368 rc = atom_init_isr(ndev);
1372 if (ndev->ntb.topo != NTB_TOPO_SEC) {
1373 /* Initiate PCI-E link training */
1374 rc = pci_write_config_dword(ndev->ntb.pdev, ATOM_PPD_OFFSET,
1375 ppd | ATOM_PPD_INIT_LINK);
1383 static void atom_deinit_dev(struct intel_ntb_dev *ndev)
1385 atom_deinit_isr(ndev);
1390 static u64 xeon_db_ioread(void __iomem *mmio)
1392 return (u64)ioread16(mmio);
1395 static void xeon_db_iowrite(u64 bits, void __iomem *mmio)
1397 iowrite16((u16)bits, mmio);
1400 static int xeon_poll_link(struct intel_ntb_dev *ndev)
1405 ndev->reg->db_iowrite(ndev->db_link_mask,
1407 ndev->self_reg->db_bell);
1409 rc = pci_read_config_word(ndev->ntb.pdev,
1410 XEON_LINK_STATUS_OFFSET, ®_val);
1414 if (reg_val == ndev->lnk_sta)
1417 ndev->lnk_sta = reg_val;
1422 static int xeon_link_is_up(struct intel_ntb_dev *ndev)
1424 if (ndev->ntb.topo == NTB_TOPO_SEC)
1427 return NTB_LNK_STA_ACTIVE(ndev->lnk_sta);
1430 static inline enum ntb_topo xeon_ppd_topo(struct intel_ntb_dev *ndev, u8 ppd)
1432 switch (ppd & XEON_PPD_TOPO_MASK) {
1433 case XEON_PPD_TOPO_B2B_USD:
1434 return NTB_TOPO_B2B_USD;
1436 case XEON_PPD_TOPO_B2B_DSD:
1437 return NTB_TOPO_B2B_DSD;
1439 case XEON_PPD_TOPO_PRI_USD:
1440 case XEON_PPD_TOPO_PRI_DSD: /* accept bogus PRI_DSD */
1441 return NTB_TOPO_PRI;
1443 case XEON_PPD_TOPO_SEC_USD:
1444 case XEON_PPD_TOPO_SEC_DSD: /* accept bogus SEC_DSD */
1445 return NTB_TOPO_SEC;
1448 return NTB_TOPO_NONE;
1451 static inline int xeon_ppd_bar4_split(struct intel_ntb_dev *ndev, u8 ppd)
1453 if (ppd & XEON_PPD_SPLIT_BAR_MASK) {
1454 dev_dbg(ndev_dev(ndev), "PPD %d split bar\n", ppd);
1460 static int xeon_init_isr(struct intel_ntb_dev *ndev)
1462 return ndev_init_isr(ndev, XEON_DB_MSIX_VECTOR_COUNT,
1463 XEON_DB_MSIX_VECTOR_COUNT,
1464 XEON_DB_MSIX_VECTOR_SHIFT,
1465 XEON_DB_TOTAL_SHIFT);
1468 static void xeon_deinit_isr(struct intel_ntb_dev *ndev)
1470 ndev_deinit_isr(ndev);
1473 static int xeon_setup_b2b_mw(struct intel_ntb_dev *ndev,
1474 const struct intel_b2b_addr *addr,
1475 const struct intel_b2b_addr *peer_addr)
1477 struct pci_dev *pdev;
1479 resource_size_t bar_size;
1480 phys_addr_t bar_addr;
1484 pdev = ndev_pdev(ndev);
1485 mmio = ndev->self_mmio;
1487 if (ndev->b2b_idx >= ndev->mw_count) {
1488 dev_dbg(ndev_dev(ndev), "not using b2b mw\n");
1492 b2b_bar = ndev_mw_to_bar(ndev, ndev->b2b_idx);
1496 dev_dbg(ndev_dev(ndev), "using b2b mw bar %d\n", b2b_bar);
1498 bar_size = pci_resource_len(ndev->ntb.pdev, b2b_bar);
1500 dev_dbg(ndev_dev(ndev), "b2b bar size %#llx\n", bar_size);
1502 if (b2b_mw_share && XEON_B2B_MIN_SIZE <= bar_size >> 1) {
1503 dev_dbg(ndev_dev(ndev),
1504 "b2b using first half of bar\n");
1505 ndev->b2b_off = bar_size >> 1;
1506 } else if (XEON_B2B_MIN_SIZE <= bar_size) {
1507 dev_dbg(ndev_dev(ndev),
1508 "b2b using whole bar\n");
1512 dev_dbg(ndev_dev(ndev),
1513 "b2b bar size is too small\n");
1518 /* Reset the secondary bar sizes to match the primary bar sizes,
1519 * except disable or halve the size of the b2b secondary bar.
1521 * Note: code for each specific bar size register, because the register
1522 * offsets are not in a consistent order (bar5sz comes after ppd, odd).
1524 pci_read_config_byte(pdev, XEON_PBAR23SZ_OFFSET, &bar_sz);
1525 dev_dbg(ndev_dev(ndev), "PBAR23SZ %#x\n", bar_sz);
1532 pci_write_config_byte(pdev, XEON_SBAR23SZ_OFFSET, bar_sz);
1533 pci_read_config_byte(pdev, XEON_SBAR23SZ_OFFSET, &bar_sz);
1534 dev_dbg(ndev_dev(ndev), "SBAR23SZ %#x\n", bar_sz);
1536 if (!ndev->bar4_split) {
1537 pci_read_config_byte(pdev, XEON_PBAR45SZ_OFFSET, &bar_sz);
1538 dev_dbg(ndev_dev(ndev), "PBAR45SZ %#x\n", bar_sz);
1545 pci_write_config_byte(pdev, XEON_SBAR45SZ_OFFSET, bar_sz);
1546 pci_read_config_byte(pdev, XEON_SBAR45SZ_OFFSET, &bar_sz);
1547 dev_dbg(ndev_dev(ndev), "SBAR45SZ %#x\n", bar_sz);
1549 pci_read_config_byte(pdev, XEON_PBAR4SZ_OFFSET, &bar_sz);
1550 dev_dbg(ndev_dev(ndev), "PBAR4SZ %#x\n", bar_sz);
1557 pci_write_config_byte(pdev, XEON_SBAR4SZ_OFFSET, bar_sz);
1558 pci_read_config_byte(pdev, XEON_SBAR4SZ_OFFSET, &bar_sz);
1559 dev_dbg(ndev_dev(ndev), "SBAR4SZ %#x\n", bar_sz);
1561 pci_read_config_byte(pdev, XEON_PBAR5SZ_OFFSET, &bar_sz);
1562 dev_dbg(ndev_dev(ndev), "PBAR5SZ %#x\n", bar_sz);
1569 pci_write_config_byte(pdev, XEON_SBAR5SZ_OFFSET, bar_sz);
1570 pci_read_config_byte(pdev, XEON_SBAR5SZ_OFFSET, &bar_sz);
1571 dev_dbg(ndev_dev(ndev), "SBAR5SZ %#x\n", bar_sz);
1574 /* SBAR01 hit by first part of the b2b bar */
1576 bar_addr = addr->bar0_addr;
1577 else if (b2b_bar == 2)
1578 bar_addr = addr->bar2_addr64;
1579 else if (b2b_bar == 4 && !ndev->bar4_split)
1580 bar_addr = addr->bar4_addr64;
1581 else if (b2b_bar == 4)
1582 bar_addr = addr->bar4_addr32;
1583 else if (b2b_bar == 5)
1584 bar_addr = addr->bar5_addr32;
1588 dev_dbg(ndev_dev(ndev), "SBAR01 %#018llx\n", bar_addr);
1589 iowrite64(bar_addr, mmio + XEON_SBAR0BASE_OFFSET);
1591 /* Other SBAR are normally hit by the PBAR xlat, except for b2b bar.
1592 * The b2b bar is either disabled above, or configured half-size, and
1593 * it starts at the PBAR xlat + offset.
1596 bar_addr = addr->bar2_addr64 + (b2b_bar == 2 ? ndev->b2b_off : 0);
1597 iowrite64(bar_addr, mmio + XEON_SBAR23BASE_OFFSET);
1598 bar_addr = ioread64(mmio + XEON_SBAR23BASE_OFFSET);
1599 dev_dbg(ndev_dev(ndev), "SBAR23 %#018llx\n", bar_addr);
1601 if (!ndev->bar4_split) {
1602 bar_addr = addr->bar4_addr64 +
1603 (b2b_bar == 4 ? ndev->b2b_off : 0);
1604 iowrite64(bar_addr, mmio + XEON_SBAR45BASE_OFFSET);
1605 bar_addr = ioread64(mmio + XEON_SBAR45BASE_OFFSET);
1606 dev_dbg(ndev_dev(ndev), "SBAR45 %#018llx\n", bar_addr);
1608 bar_addr = addr->bar4_addr32 +
1609 (b2b_bar == 4 ? ndev->b2b_off : 0);
1610 iowrite32(bar_addr, mmio + XEON_SBAR4BASE_OFFSET);
1611 bar_addr = ioread32(mmio + XEON_SBAR4BASE_OFFSET);
1612 dev_dbg(ndev_dev(ndev), "SBAR4 %#010llx\n", bar_addr);
1614 bar_addr = addr->bar5_addr32 +
1615 (b2b_bar == 5 ? ndev->b2b_off : 0);
1616 iowrite32(bar_addr, mmio + XEON_SBAR5BASE_OFFSET);
1617 bar_addr = ioread32(mmio + XEON_SBAR5BASE_OFFSET);
1618 dev_dbg(ndev_dev(ndev), "SBAR5 %#010llx\n", bar_addr);
1621 /* setup incoming bar limits == base addrs (zero length windows) */
1623 bar_addr = addr->bar2_addr64 + (b2b_bar == 2 ? ndev->b2b_off : 0);
1624 iowrite64(bar_addr, mmio + XEON_SBAR23LMT_OFFSET);
1625 bar_addr = ioread64(mmio + XEON_SBAR23LMT_OFFSET);
1626 dev_dbg(ndev_dev(ndev), "SBAR23LMT %#018llx\n", bar_addr);
1628 if (!ndev->bar4_split) {
1629 bar_addr = addr->bar4_addr64 +
1630 (b2b_bar == 4 ? ndev->b2b_off : 0);
1631 iowrite64(bar_addr, mmio + XEON_SBAR45LMT_OFFSET);
1632 bar_addr = ioread64(mmio + XEON_SBAR45LMT_OFFSET);
1633 dev_dbg(ndev_dev(ndev), "SBAR45LMT %#018llx\n", bar_addr);
1635 bar_addr = addr->bar4_addr32 +
1636 (b2b_bar == 4 ? ndev->b2b_off : 0);
1637 iowrite32(bar_addr, mmio + XEON_SBAR4LMT_OFFSET);
1638 bar_addr = ioread32(mmio + XEON_SBAR4LMT_OFFSET);
1639 dev_dbg(ndev_dev(ndev), "SBAR4LMT %#010llx\n", bar_addr);
1641 bar_addr = addr->bar5_addr32 +
1642 (b2b_bar == 5 ? ndev->b2b_off : 0);
1643 iowrite32(bar_addr, mmio + XEON_SBAR5LMT_OFFSET);
1644 bar_addr = ioread32(mmio + XEON_SBAR5LMT_OFFSET);
1645 dev_dbg(ndev_dev(ndev), "SBAR5LMT %#05llx\n", bar_addr);
1648 /* zero incoming translation addrs */
1649 iowrite64(0, mmio + XEON_SBAR23XLAT_OFFSET);
1651 if (!ndev->bar4_split) {
1652 iowrite64(0, mmio + XEON_SBAR45XLAT_OFFSET);
1654 iowrite32(0, mmio + XEON_SBAR4XLAT_OFFSET);
1655 iowrite32(0, mmio + XEON_SBAR5XLAT_OFFSET);
1658 /* zero outgoing translation limits (whole bar size windows) */
1659 iowrite64(0, mmio + XEON_PBAR23LMT_OFFSET);
1660 if (!ndev->bar4_split) {
1661 iowrite64(0, mmio + XEON_PBAR45LMT_OFFSET);
1663 iowrite32(0, mmio + XEON_PBAR4LMT_OFFSET);
1664 iowrite32(0, mmio + XEON_PBAR5LMT_OFFSET);
1667 /* set outgoing translation offsets */
1668 bar_addr = peer_addr->bar2_addr64;
1669 iowrite64(bar_addr, mmio + XEON_PBAR23XLAT_OFFSET);
1670 bar_addr = ioread64(mmio + XEON_PBAR23XLAT_OFFSET);
1671 dev_dbg(ndev_dev(ndev), "PBAR23XLAT %#018llx\n", bar_addr);
1673 if (!ndev->bar4_split) {
1674 bar_addr = peer_addr->bar4_addr64;
1675 iowrite64(bar_addr, mmio + XEON_PBAR45XLAT_OFFSET);
1676 bar_addr = ioread64(mmio + XEON_PBAR45XLAT_OFFSET);
1677 dev_dbg(ndev_dev(ndev), "PBAR45XLAT %#018llx\n", bar_addr);
1679 bar_addr = peer_addr->bar4_addr32;
1680 iowrite32(bar_addr, mmio + XEON_PBAR4XLAT_OFFSET);
1681 bar_addr = ioread32(mmio + XEON_PBAR4XLAT_OFFSET);
1682 dev_dbg(ndev_dev(ndev), "PBAR4XLAT %#010llx\n", bar_addr);
1684 bar_addr = peer_addr->bar5_addr32;
1685 iowrite32(bar_addr, mmio + XEON_PBAR5XLAT_OFFSET);
1686 bar_addr = ioread32(mmio + XEON_PBAR5XLAT_OFFSET);
1687 dev_dbg(ndev_dev(ndev), "PBAR5XLAT %#010llx\n", bar_addr);
1690 /* set the translation offset for b2b registers */
1692 bar_addr = peer_addr->bar0_addr;
1693 else if (b2b_bar == 2)
1694 bar_addr = peer_addr->bar2_addr64;
1695 else if (b2b_bar == 4 && !ndev->bar4_split)
1696 bar_addr = peer_addr->bar4_addr64;
1697 else if (b2b_bar == 4)
1698 bar_addr = peer_addr->bar4_addr32;
1699 else if (b2b_bar == 5)
1700 bar_addr = peer_addr->bar5_addr32;
1704 /* B2B_XLAT_OFFSET is 64bit, but can only take 32bit writes */
1705 dev_dbg(ndev_dev(ndev), "B2BXLAT %#018llx\n", bar_addr);
1706 iowrite32(bar_addr, mmio + XEON_B2B_XLAT_OFFSETL);
1707 iowrite32(bar_addr >> 32, mmio + XEON_B2B_XLAT_OFFSETU);
1710 /* map peer ntb mmio config space registers */
1711 ndev->peer_mmio = pci_iomap(pdev, b2b_bar,
1713 if (!ndev->peer_mmio)
1720 static int xeon_init_ntb(struct intel_ntb_dev *ndev)
1725 if (ndev->bar4_split)
1726 ndev->mw_count = HSX_SPLIT_BAR_MW_COUNT;
1728 ndev->mw_count = XEON_MW_COUNT;
1730 ndev->spad_count = XEON_SPAD_COUNT;
1731 ndev->db_count = XEON_DB_COUNT;
1732 ndev->db_link_mask = XEON_DB_LINK_BIT;
1734 switch (ndev->ntb.topo) {
1736 if (ndev->hwerr_flags & NTB_HWERR_SDOORBELL_LOCKUP) {
1737 dev_err(ndev_dev(ndev), "NTB Primary config disabled\n");
1741 /* enable link to allow secondary side device to appear */
1742 ntb_ctl = ioread32(ndev->self_mmio + ndev->reg->ntb_ctl);
1743 ntb_ctl &= ~NTB_CTL_DISABLE;
1744 iowrite32(ntb_ctl, ndev->self_mmio + ndev->reg->ntb_ctl);
1746 /* use half the spads for the peer */
1747 ndev->spad_count >>= 1;
1748 ndev->self_reg = &xeon_pri_reg;
1749 ndev->peer_reg = &xeon_sec_reg;
1750 ndev->xlat_reg = &xeon_sec_xlat;
1754 if (ndev->hwerr_flags & NTB_HWERR_SDOORBELL_LOCKUP) {
1755 dev_err(ndev_dev(ndev), "NTB Secondary config disabled\n");
1758 /* use half the spads for the peer */
1759 ndev->spad_count >>= 1;
1760 ndev->self_reg = &xeon_sec_reg;
1761 ndev->peer_reg = &xeon_pri_reg;
1762 ndev->xlat_reg = &xeon_pri_xlat;
1765 case NTB_TOPO_B2B_USD:
1766 case NTB_TOPO_B2B_DSD:
1767 ndev->self_reg = &xeon_pri_reg;
1768 ndev->peer_reg = &xeon_b2b_reg;
1769 ndev->xlat_reg = &xeon_sec_xlat;
1771 if (ndev->hwerr_flags & NTB_HWERR_SDOORBELL_LOCKUP) {
1772 ndev->peer_reg = &xeon_pri_reg;
1775 ndev->b2b_idx = b2b_mw_idx + ndev->mw_count;
1777 ndev->b2b_idx = b2b_mw_idx;
1779 dev_dbg(ndev_dev(ndev),
1780 "setting up b2b mw idx %d means %d\n",
1781 b2b_mw_idx, ndev->b2b_idx);
1783 } else if (ndev->hwerr_flags & NTB_HWERR_B2BDOORBELL_BIT14) {
1784 dev_warn(ndev_dev(ndev), "Reduce doorbell count by 1\n");
1785 ndev->db_count -= 1;
1788 if (ndev->ntb.topo == NTB_TOPO_B2B_USD) {
1789 rc = xeon_setup_b2b_mw(ndev,
1791 &xeon_b2b_usd_addr);
1793 rc = xeon_setup_b2b_mw(ndev,
1795 &xeon_b2b_dsd_addr);
1800 /* Enable Bus Master and Memory Space on the secondary side */
1801 iowrite16(PCI_COMMAND_MEMORY | PCI_COMMAND_MASTER,
1802 ndev->self_mmio + XEON_SPCICMD_OFFSET);
1810 ndev->db_valid_mask = BIT_ULL(ndev->db_count) - 1;
1812 ndev->reg->db_iowrite(ndev->db_valid_mask,
1814 ndev->self_reg->db_mask);
1819 static int xeon_init_dev(struct intel_ntb_dev *ndev)
1821 struct pci_dev *pdev;
1825 pdev = ndev_pdev(ndev);
1827 switch (pdev->device) {
1828 /* There is a Xeon hardware errata related to writes to SDOORBELL or
1829 * B2BDOORBELL in conjunction with inbound access to NTB MMIO Space,
1830 * which may hang the system. To workaround this use the second memory
1831 * window to access the interrupt and scratch pad registers on the
1834 case PCI_DEVICE_ID_INTEL_NTB_SS_JSF:
1835 case PCI_DEVICE_ID_INTEL_NTB_PS_JSF:
1836 case PCI_DEVICE_ID_INTEL_NTB_B2B_JSF:
1837 case PCI_DEVICE_ID_INTEL_NTB_SS_SNB:
1838 case PCI_DEVICE_ID_INTEL_NTB_PS_SNB:
1839 case PCI_DEVICE_ID_INTEL_NTB_B2B_SNB:
1840 case PCI_DEVICE_ID_INTEL_NTB_SS_IVT:
1841 case PCI_DEVICE_ID_INTEL_NTB_PS_IVT:
1842 case PCI_DEVICE_ID_INTEL_NTB_B2B_IVT:
1843 case PCI_DEVICE_ID_INTEL_NTB_SS_HSX:
1844 case PCI_DEVICE_ID_INTEL_NTB_PS_HSX:
1845 case PCI_DEVICE_ID_INTEL_NTB_B2B_HSX:
1846 ndev->hwerr_flags |= NTB_HWERR_SDOORBELL_LOCKUP;
1850 switch (pdev->device) {
1851 /* There is a hardware errata related to accessing any register in
1852 * SB01BASE in the presence of bidirectional traffic crossing the NTB.
1854 case PCI_DEVICE_ID_INTEL_NTB_SS_IVT:
1855 case PCI_DEVICE_ID_INTEL_NTB_PS_IVT:
1856 case PCI_DEVICE_ID_INTEL_NTB_B2B_IVT:
1857 case PCI_DEVICE_ID_INTEL_NTB_SS_HSX:
1858 case PCI_DEVICE_ID_INTEL_NTB_PS_HSX:
1859 case PCI_DEVICE_ID_INTEL_NTB_B2B_HSX:
1860 ndev->hwerr_flags |= NTB_HWERR_SB01BASE_LOCKUP;
1864 switch (pdev->device) {
1865 /* HW Errata on bit 14 of b2bdoorbell register. Writes will not be
1866 * mirrored to the remote system. Shrink the number of bits by one,
1867 * since bit 14 is the last bit.
1869 case PCI_DEVICE_ID_INTEL_NTB_SS_JSF:
1870 case PCI_DEVICE_ID_INTEL_NTB_PS_JSF:
1871 case PCI_DEVICE_ID_INTEL_NTB_B2B_JSF:
1872 case PCI_DEVICE_ID_INTEL_NTB_SS_SNB:
1873 case PCI_DEVICE_ID_INTEL_NTB_PS_SNB:
1874 case PCI_DEVICE_ID_INTEL_NTB_B2B_SNB:
1875 case PCI_DEVICE_ID_INTEL_NTB_SS_IVT:
1876 case PCI_DEVICE_ID_INTEL_NTB_PS_IVT:
1877 case PCI_DEVICE_ID_INTEL_NTB_B2B_IVT:
1878 case PCI_DEVICE_ID_INTEL_NTB_SS_HSX:
1879 case PCI_DEVICE_ID_INTEL_NTB_PS_HSX:
1880 case PCI_DEVICE_ID_INTEL_NTB_B2B_HSX:
1881 ndev->hwerr_flags |= NTB_HWERR_B2BDOORBELL_BIT14;
1885 ndev->reg = &xeon_reg;
1887 rc = pci_read_config_byte(pdev, XEON_PPD_OFFSET, &ppd);
1891 ndev->ntb.topo = xeon_ppd_topo(ndev, ppd);
1892 dev_dbg(ndev_dev(ndev), "ppd %#x topo %s\n", ppd,
1893 ntb_topo_string(ndev->ntb.topo));
1894 if (ndev->ntb.topo == NTB_TOPO_NONE)
1897 if (ndev->ntb.topo != NTB_TOPO_SEC) {
1898 ndev->bar4_split = xeon_ppd_bar4_split(ndev, ppd);
1899 dev_dbg(ndev_dev(ndev), "ppd %#x bar4_split %d\n",
1900 ppd, ndev->bar4_split);
1902 /* This is a way for transparent BAR to figure out if we are
1903 * doing split BAR or not. There is no way for the hw on the
1904 * transparent side to know and set the PPD.
1906 mem = pci_select_bars(pdev, IORESOURCE_MEM);
1907 ndev->bar4_split = hweight32(mem) ==
1908 HSX_SPLIT_BAR_MW_COUNT + 1;
1909 dev_dbg(ndev_dev(ndev), "mem %#x bar4_split %d\n",
1910 mem, ndev->bar4_split);
1913 rc = xeon_init_ntb(ndev);
1917 return xeon_init_isr(ndev);
1920 static void xeon_deinit_dev(struct intel_ntb_dev *ndev)
1922 xeon_deinit_isr(ndev);
1925 static int intel_ntb_init_pci(struct intel_ntb_dev *ndev, struct pci_dev *pdev)
1929 pci_set_drvdata(pdev, ndev);
1931 rc = pci_enable_device(pdev);
1933 goto err_pci_enable;
1935 rc = pci_request_regions(pdev, NTB_NAME);
1937 goto err_pci_regions;
1939 pci_set_master(pdev);
1941 rc = pci_set_dma_mask(pdev, DMA_BIT_MASK(64));
1943 rc = pci_set_dma_mask(pdev, DMA_BIT_MASK(32));
1946 dev_warn(ndev_dev(ndev), "Cannot DMA highmem\n");
1949 rc = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(64));
1951 rc = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(32));
1954 dev_warn(ndev_dev(ndev), "Cannot DMA consistent highmem\n");
1957 ndev->self_mmio = pci_iomap(pdev, 0, 0);
1958 if (!ndev->self_mmio) {
1962 ndev->peer_mmio = ndev->self_mmio;
1968 pci_clear_master(pdev);
1969 pci_release_regions(pdev);
1971 pci_disable_device(pdev);
1973 pci_set_drvdata(pdev, NULL);
1977 static void intel_ntb_deinit_pci(struct intel_ntb_dev *ndev)
1979 struct pci_dev *pdev = ndev_pdev(ndev);
1981 if (ndev->peer_mmio && ndev->peer_mmio != ndev->self_mmio)
1982 pci_iounmap(pdev, ndev->peer_mmio);
1983 pci_iounmap(pdev, ndev->self_mmio);
1985 pci_clear_master(pdev);
1986 pci_release_regions(pdev);
1987 pci_disable_device(pdev);
1988 pci_set_drvdata(pdev, NULL);
1991 static inline void ndev_init_struct(struct intel_ntb_dev *ndev,
1992 struct pci_dev *pdev)
1994 ndev->ntb.pdev = pdev;
1995 ndev->ntb.topo = NTB_TOPO_NONE;
1996 ndev->ntb.ops = &intel_ntb_ops;
1999 ndev->b2b_idx = INT_MAX;
2001 ndev->bar4_split = 0;
2004 ndev->spad_count = 0;
2006 ndev->db_vec_count = 0;
2007 ndev->db_vec_shift = 0;
2012 ndev->db_valid_mask = 0;
2013 ndev->db_link_mask = 0;
2016 spin_lock_init(&ndev->db_mask_lock);
2019 static int intel_ntb_pci_probe(struct pci_dev *pdev,
2020 const struct pci_device_id *id)
2022 struct intel_ntb_dev *ndev;
2025 node = dev_to_node(&pdev->dev);
2027 if (pdev_is_atom(pdev)) {
2028 ndev = kzalloc_node(sizeof(*ndev), GFP_KERNEL, node);
2034 ndev_init_struct(ndev, pdev);
2036 rc = intel_ntb_init_pci(ndev, pdev);
2040 rc = atom_init_dev(ndev);
2044 } else if (pdev_is_xeon(pdev)) {
2045 ndev = kzalloc_node(sizeof(*ndev), GFP_KERNEL, node);
2051 ndev_init_struct(ndev, pdev);
2053 rc = intel_ntb_init_pci(ndev, pdev);
2057 rc = xeon_init_dev(ndev);
2066 ndev_reset_unsafe_flags(ndev);
2068 ndev->reg->poll_link(ndev);
2070 ndev_init_debugfs(ndev);
2072 rc = ntb_register_device(&ndev->ntb);
2076 dev_info(&pdev->dev, "NTB device registered.\n");
2081 ndev_deinit_debugfs(ndev);
2082 if (pdev_is_atom(pdev))
2083 atom_deinit_dev(ndev);
2084 else if (pdev_is_xeon(pdev))
2085 xeon_deinit_dev(ndev);
2087 intel_ntb_deinit_pci(ndev);
2094 static void intel_ntb_pci_remove(struct pci_dev *pdev)
2096 struct intel_ntb_dev *ndev = pci_get_drvdata(pdev);
2098 ntb_unregister_device(&ndev->ntb);
2099 ndev_deinit_debugfs(ndev);
2100 if (pdev_is_atom(pdev))
2101 atom_deinit_dev(ndev);
2102 else if (pdev_is_xeon(pdev))
2103 xeon_deinit_dev(ndev);
2104 intel_ntb_deinit_pci(ndev);
2108 static const struct intel_ntb_reg atom_reg = {
2109 .poll_link = atom_poll_link,
2110 .link_is_up = atom_link_is_up,
2111 .db_ioread = atom_db_ioread,
2112 .db_iowrite = atom_db_iowrite,
2113 .db_size = sizeof(u64),
2114 .ntb_ctl = ATOM_NTBCNTL_OFFSET,
2118 static const struct intel_ntb_alt_reg atom_pri_reg = {
2119 .db_bell = ATOM_PDOORBELL_OFFSET,
2120 .db_mask = ATOM_PDBMSK_OFFSET,
2121 .spad = ATOM_SPAD_OFFSET,
2124 static const struct intel_ntb_alt_reg atom_b2b_reg = {
2125 .db_bell = ATOM_B2B_DOORBELL_OFFSET,
2126 .spad = ATOM_B2B_SPAD_OFFSET,
2129 static const struct intel_ntb_xlat_reg atom_sec_xlat = {
2130 /* FIXME : .bar0_base = ATOM_SBAR0BASE_OFFSET, */
2131 /* FIXME : .bar2_limit = ATOM_SBAR2LMT_OFFSET, */
2132 .bar2_xlat = ATOM_SBAR2XLAT_OFFSET,
2135 static const struct intel_ntb_reg xeon_reg = {
2136 .poll_link = xeon_poll_link,
2137 .link_is_up = xeon_link_is_up,
2138 .db_ioread = xeon_db_ioread,
2139 .db_iowrite = xeon_db_iowrite,
2140 .db_size = sizeof(u32),
2141 .ntb_ctl = XEON_NTBCNTL_OFFSET,
2142 .mw_bar = {2, 4, 5},
2145 static const struct intel_ntb_alt_reg xeon_pri_reg = {
2146 .db_bell = XEON_PDOORBELL_OFFSET,
2147 .db_mask = XEON_PDBMSK_OFFSET,
2148 .spad = XEON_SPAD_OFFSET,
2151 static const struct intel_ntb_alt_reg xeon_sec_reg = {
2152 .db_bell = XEON_SDOORBELL_OFFSET,
2153 .db_mask = XEON_SDBMSK_OFFSET,
2154 /* second half of the scratchpads */
2155 .spad = XEON_SPAD_OFFSET + (XEON_SPAD_COUNT << 1),
2158 static const struct intel_ntb_alt_reg xeon_b2b_reg = {
2159 .db_bell = XEON_B2B_DOORBELL_OFFSET,
2160 .spad = XEON_B2B_SPAD_OFFSET,
2163 static const struct intel_ntb_xlat_reg xeon_pri_xlat = {
2164 /* Note: no primary .bar0_base visible to the secondary side.
2166 * The secondary side cannot get the base address stored in primary
2167 * bars. The base address is necessary to set the limit register to
2168 * any value other than zero, or unlimited.
2170 * WITHOUT THE BASE ADDRESS, THE SECONDARY SIDE CANNOT DISABLE the
2171 * window by setting the limit equal to base, nor can it limit the size
2172 * of the memory window by setting the limit to base + size.
2174 .bar2_limit = XEON_PBAR23LMT_OFFSET,
2175 .bar2_xlat = XEON_PBAR23XLAT_OFFSET,
2178 static const struct intel_ntb_xlat_reg xeon_sec_xlat = {
2179 .bar0_base = XEON_SBAR0BASE_OFFSET,
2180 .bar2_limit = XEON_SBAR23LMT_OFFSET,
2181 .bar2_xlat = XEON_SBAR23XLAT_OFFSET,
2184 static struct intel_b2b_addr xeon_b2b_usd_addr = {
2185 .bar2_addr64 = XEON_B2B_BAR2_USD_ADDR64,
2186 .bar4_addr64 = XEON_B2B_BAR4_USD_ADDR64,
2187 .bar4_addr32 = XEON_B2B_BAR4_USD_ADDR32,
2188 .bar5_addr32 = XEON_B2B_BAR5_USD_ADDR32,
2191 static struct intel_b2b_addr xeon_b2b_dsd_addr = {
2192 .bar2_addr64 = XEON_B2B_BAR2_DSD_ADDR64,
2193 .bar4_addr64 = XEON_B2B_BAR4_DSD_ADDR64,
2194 .bar4_addr32 = XEON_B2B_BAR4_DSD_ADDR32,
2195 .bar5_addr32 = XEON_B2B_BAR5_DSD_ADDR32,
2198 /* operations for primary side of local ntb */
2199 static const struct ntb_dev_ops intel_ntb_ops = {
2200 .mw_count = intel_ntb_mw_count,
2201 .mw_get_range = intel_ntb_mw_get_range,
2202 .mw_set_trans = intel_ntb_mw_set_trans,
2203 .link_is_up = intel_ntb_link_is_up,
2204 .link_enable = intel_ntb_link_enable,
2205 .link_disable = intel_ntb_link_disable,
2206 .db_is_unsafe = intel_ntb_db_is_unsafe,
2207 .db_valid_mask = intel_ntb_db_valid_mask,
2208 .db_vector_count = intel_ntb_db_vector_count,
2209 .db_vector_mask = intel_ntb_db_vector_mask,
2210 .db_read = intel_ntb_db_read,
2211 .db_clear = intel_ntb_db_clear,
2212 .db_set_mask = intel_ntb_db_set_mask,
2213 .db_clear_mask = intel_ntb_db_clear_mask,
2214 .peer_db_addr = intel_ntb_peer_db_addr,
2215 .peer_db_set = intel_ntb_peer_db_set,
2216 .spad_is_unsafe = intel_ntb_spad_is_unsafe,
2217 .spad_count = intel_ntb_spad_count,
2218 .spad_read = intel_ntb_spad_read,
2219 .spad_write = intel_ntb_spad_write,
2220 .peer_spad_addr = intel_ntb_peer_spad_addr,
2221 .peer_spad_read = intel_ntb_peer_spad_read,
2222 .peer_spad_write = intel_ntb_peer_spad_write,
2225 static const struct file_operations intel_ntb_debugfs_info = {
2226 .owner = THIS_MODULE,
2227 .open = simple_open,
2228 .read = ndev_debugfs_read,
2231 static const struct pci_device_id intel_ntb_pci_tbl[] = {
2232 {PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_NTB_B2B_BWD)},
2233 {PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_NTB_B2B_JSF)},
2234 {PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_NTB_B2B_SNB)},
2235 {PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_NTB_B2B_IVT)},
2236 {PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_NTB_B2B_HSX)},
2237 {PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_NTB_PS_JSF)},
2238 {PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_NTB_PS_SNB)},
2239 {PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_NTB_PS_IVT)},
2240 {PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_NTB_PS_HSX)},
2241 {PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_NTB_SS_JSF)},
2242 {PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_NTB_SS_SNB)},
2243 {PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_NTB_SS_IVT)},
2244 {PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_NTB_SS_HSX)},
2247 MODULE_DEVICE_TABLE(pci, intel_ntb_pci_tbl);
2249 static struct pci_driver intel_ntb_pci_driver = {
2250 .name = KBUILD_MODNAME,
2251 .id_table = intel_ntb_pci_tbl,
2252 .probe = intel_ntb_pci_probe,
2253 .remove = intel_ntb_pci_remove,
2256 static int __init intel_ntb_pci_driver_init(void)
2258 pr_info("%s %s\n", NTB_DESC, NTB_VER);
2260 if (debugfs_initialized())
2261 debugfs_dir = debugfs_create_dir(KBUILD_MODNAME, NULL);
2263 return pci_register_driver(&intel_ntb_pci_driver);
2265 module_init(intel_ntb_pci_driver_init);
2267 static void __exit intel_ntb_pci_driver_exit(void)
2269 pci_unregister_driver(&intel_ntb_pci_driver);
2271 debugfs_remove_recursive(debugfs_dir);
2273 module_exit(intel_ntb_pci_driver_exit);