aboutsummaryrefslogtreecommitdiffstats
diff options
context:
space:
mode:
authorDave Jiang <dave.jiang@intel.com>2014-08-28 16:53:23 -0400
committerJon Mason <jdmason@kudzu.us>2014-10-17 07:08:51 -0400
commitab760a0c5667519b375ea9c5ab3a23501c4817ef (patch)
tree29d2b7dd165287dd635ff4b320fcb6a9dc5a7eae
parent069684e888da73f175da0f10fe26da4f450d8c18 (diff)
ntb: Adding split BAR support for Haswell platforms
On the Haswell platform, a split BAR option to allow creation of 2 32bit BARs (4 and 5) from the 64bit BAR 4. Adding support for this new option. Signed-off-by: Dave Jiang <dave.jiang@intel.com> Signed-off-by: Jon Mason <jdmason@kudzu.us>
-rw-r--r--drivers/ntb/ntb_hw.c225
-rw-r--r--drivers/ntb/ntb_hw.h14
-rw-r--r--drivers/ntb/ntb_regs.h31
3 files changed, 210 insertions, 60 deletions
diff --git a/drivers/ntb/ntb_hw.c b/drivers/ntb/ntb_hw.c
index 53b739df8cd5..cd29b1038c5e 100644
--- a/drivers/ntb/ntb_hw.c
+++ b/drivers/ntb/ntb_hw.c
@@ -84,8 +84,8 @@ static struct dentry *debugfs_dir;
84 84
85#define BWD_LINK_RECOVERY_TIME 500 85#define BWD_LINK_RECOVERY_TIME 500
86 86
87/* Translate memory window 0,1 to BAR 2,4 */ 87/* Translate memory window 0,1,2 to BAR 2,4,5 */
88#define MW_TO_BAR(mw) (mw * NTB_MAX_NUM_MW + 2) 88#define MW_TO_BAR(mw) (mw == 0 ? 2 : (mw == 1 ? 4 : 5))
89 89
90static const struct pci_device_id ntb_pci_tbl[] = { 90static const struct pci_device_id ntb_pci_tbl[] = {
91 {PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_NTB_B2B_BWD)}, 91 {PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_NTB_B2B_BWD)},
@@ -506,8 +506,14 @@ void ntb_set_mw_addr(struct ntb_device *ndev, unsigned int mw, u64 addr)
506 case NTB_BAR_23: 506 case NTB_BAR_23:
507 writeq(addr, ndev->reg_ofs.bar2_xlat); 507 writeq(addr, ndev->reg_ofs.bar2_xlat);
508 break; 508 break;
509 case NTB_BAR_45: 509 case NTB_BAR_4:
510 writeq(addr, ndev->reg_ofs.bar4_xlat); 510 if (ndev->split_bar)
511 writel(addr, ndev->reg_ofs.bar4_xlat);
512 else
513 writeq(addr, ndev->reg_ofs.bar4_xlat);
514 break;
515 case NTB_BAR_5:
516 writel(addr, ndev->reg_ofs.bar5_xlat);
511 break; 517 break;
512 } 518 }
513} 519}
@@ -729,6 +735,9 @@ static int ntb_xeon_setup(struct ntb_device *ndev)
729 ndev->reg_ofs.spad_read = ndev->reg_base + SNB_SPAD_OFFSET; 735 ndev->reg_ofs.spad_read = ndev->reg_base + SNB_SPAD_OFFSET;
730 ndev->reg_ofs.bar2_xlat = ndev->reg_base + SNB_SBAR2XLAT_OFFSET; 736 ndev->reg_ofs.bar2_xlat = ndev->reg_base + SNB_SBAR2XLAT_OFFSET;
731 ndev->reg_ofs.bar4_xlat = ndev->reg_base + SNB_SBAR4XLAT_OFFSET; 737 ndev->reg_ofs.bar4_xlat = ndev->reg_base + SNB_SBAR4XLAT_OFFSET;
738 if (ndev->split_bar)
739 ndev->reg_ofs.bar5_xlat =
740 ndev->reg_base + SNB_SBAR5XLAT_OFFSET;
732 ndev->limits.max_spads = SNB_MAX_B2B_SPADS; 741 ndev->limits.max_spads = SNB_MAX_B2B_SPADS;
733 742
734 /* There is a Xeon hardware errata related to writes to 743 /* There is a Xeon hardware errata related to writes to
@@ -738,15 +747,16 @@ static int ntb_xeon_setup(struct ntb_device *ndev)
738 * scratch pad registers on the remote system. 747 * scratch pad registers on the remote system.
739 */ 748 */
740 if (ndev->wa_flags & WA_SNB_ERR) { 749 if (ndev->wa_flags & WA_SNB_ERR) {
741 if (!ndev->mw[1].bar_sz) 750 if (!ndev->mw[ndev->limits.max_mw - 1].bar_sz)
742 return -EINVAL; 751 return -EINVAL;
743 752
744 ndev->limits.max_mw = SNB_ERRATA_MAX_MW;
745 ndev->limits.max_db_bits = SNB_MAX_DB_BITS; 753 ndev->limits.max_db_bits = SNB_MAX_DB_BITS;
746 ndev->reg_ofs.spad_write = ndev->mw[1].vbase + 754 ndev->reg_ofs.spad_write =
747 SNB_SPAD_OFFSET; 755 ndev->mw[ndev->limits.max_mw - 1].vbase +
748 ndev->reg_ofs.rdb = ndev->mw[1].vbase + 756 SNB_SPAD_OFFSET;
749 SNB_PDOORBELL_OFFSET; 757 ndev->reg_ofs.rdb =
758 ndev->mw[ndev->limits.max_mw - 1].vbase +
759 SNB_PDOORBELL_OFFSET;
750 760
751 /* Set the Limit register to 4k, the minimum size, to 761 /* Set the Limit register to 4k, the minimum size, to
752 * prevent an illegal access 762 * prevent an illegal access
@@ -759,9 +769,9 @@ static int ntb_xeon_setup(struct ntb_device *ndev)
759 * the driver defaults, but write the Limit registers 769 * the driver defaults, but write the Limit registers
760 * first just in case. 770 * first just in case.
761 */ 771 */
762 } else {
763 ndev->limits.max_mw = SNB_MAX_MW;
764 772
773 ndev->limits.max_mw = SNB_ERRATA_MAX_MW;
774 } else {
765 /* HW Errata on bit 14 of b2bdoorbell register. Writes 775 /* HW Errata on bit 14 of b2bdoorbell register. Writes
766 * will not be mirrored to the remote system. Shrink 776 * will not be mirrored to the remote system. Shrink
767 * the number of bits by one, since bit 14 is the last 777 * the number of bits by one, since bit 14 is the last
@@ -774,7 +784,8 @@ static int ntb_xeon_setup(struct ntb_device *ndev)
774 SNB_B2B_DOORBELL_OFFSET; 784 SNB_B2B_DOORBELL_OFFSET;
775 785
776 /* Disable the Limit register, just incase it is set to 786 /* Disable the Limit register, just incase it is set to
777 * something silly 787 * something silly. A 64bit write should handle it
788 * regardless of whether it has a split BAR or not.
778 */ 789 */
779 writeq(0, ndev->reg_base + SNB_PBAR4LMT_OFFSET); 790 writeq(0, ndev->reg_base + SNB_PBAR4LMT_OFFSET);
780 /* HW errata on the Limit registers. They can only be 791 /* HW errata on the Limit registers. They can only be
@@ -783,6 +794,10 @@ static int ntb_xeon_setup(struct ntb_device *ndev)
783 * the driver defaults, but write the Limit registers 794 * the driver defaults, but write the Limit registers
784 * first just in case. 795 * first just in case.
785 */ 796 */
797 if (ndev->split_bar)
798 ndev->limits.max_mw = HSX_SPLITBAR_MAX_MW;
799 else
800 ndev->limits.max_mw = SNB_MAX_MW;
786 } 801 }
787 802
788 /* The Xeon errata workaround requires setting SBAR Base 803 /* The Xeon errata workaround requires setting SBAR Base
@@ -796,8 +811,18 @@ static int ntb_xeon_setup(struct ntb_device *ndev)
796 writeq(SNB_MBAR01_DSD_ADDR, ndev->reg_base + 811 writeq(SNB_MBAR01_DSD_ADDR, ndev->reg_base +
797 SNB_PBAR4XLAT_OFFSET); 812 SNB_PBAR4XLAT_OFFSET);
798 else { 813 else {
799 writeq(SNB_MBAR45_DSD_ADDR, ndev->reg_base + 814 if (ndev->split_bar) {
800 SNB_PBAR4XLAT_OFFSET); 815 writel(SNB_MBAR4_DSD_ADDR,
816 ndev->reg_base +
817 SNB_PBAR4XLAT_OFFSET);
818 writel(SNB_MBAR5_DSD_ADDR,
819 ndev->reg_base +
820 SNB_PBAR5XLAT_OFFSET);
821 } else
822 writeq(SNB_MBAR4_DSD_ADDR,
823 ndev->reg_base +
824 SNB_PBAR4XLAT_OFFSET);
825
801 /* B2B_XLAT_OFFSET is a 64bit register, but can 826 /* B2B_XLAT_OFFSET is a 64bit register, but can
802 * only take 32bit writes 827 * only take 32bit writes
803 */ 828 */
@@ -811,8 +836,14 @@ static int ntb_xeon_setup(struct ntb_device *ndev)
811 SNB_SBAR0BASE_OFFSET); 836 SNB_SBAR0BASE_OFFSET);
812 writeq(SNB_MBAR23_USD_ADDR, ndev->reg_base + 837 writeq(SNB_MBAR23_USD_ADDR, ndev->reg_base +
813 SNB_SBAR2BASE_OFFSET); 838 SNB_SBAR2BASE_OFFSET);
814 writeq(SNB_MBAR45_USD_ADDR, ndev->reg_base + 839 if (ndev->split_bar) {
815 SNB_SBAR4BASE_OFFSET); 840 writel(SNB_MBAR4_USD_ADDR, ndev->reg_base +
841 SNB_SBAR4BASE_OFFSET);
842 writel(SNB_MBAR5_USD_ADDR, ndev->reg_base +
843 SNB_SBAR5BASE_OFFSET);
844 } else
845 writeq(SNB_MBAR4_USD_ADDR, ndev->reg_base +
846 SNB_SBAR4BASE_OFFSET);
816 } else { 847 } else {
817 writeq(SNB_MBAR23_USD_ADDR, ndev->reg_base + 848 writeq(SNB_MBAR23_USD_ADDR, ndev->reg_base +
818 SNB_PBAR2XLAT_OFFSET); 849 SNB_PBAR2XLAT_OFFSET);
@@ -820,9 +851,20 @@ static int ntb_xeon_setup(struct ntb_device *ndev)
820 writeq(SNB_MBAR01_USD_ADDR, ndev->reg_base + 851 writeq(SNB_MBAR01_USD_ADDR, ndev->reg_base +
821 SNB_PBAR4XLAT_OFFSET); 852 SNB_PBAR4XLAT_OFFSET);
822 else { 853 else {
823 writeq(SNB_MBAR45_USD_ADDR, ndev->reg_base + 854 if (ndev->split_bar) {
824 SNB_PBAR4XLAT_OFFSET); 855 writel(SNB_MBAR4_USD_ADDR,
825 /* B2B_XLAT_OFFSET is a 64bit register, but can 856 ndev->reg_base +
857 SNB_PBAR4XLAT_OFFSET);
858 writel(SNB_MBAR5_USD_ADDR,
859 ndev->reg_base +
860 SNB_PBAR5XLAT_OFFSET);
861 } else
862 writeq(SNB_MBAR4_USD_ADDR,
863 ndev->reg_base +
864 SNB_PBAR4XLAT_OFFSET);
865
866 /*
867 * B2B_XLAT_OFFSET is a 64bit register, but can
826 * only take 32bit writes 868 * only take 32bit writes
827 */ 869 */
828 writel(SNB_MBAR01_USD_ADDR & 0xffffffff, 870 writel(SNB_MBAR01_USD_ADDR & 0xffffffff,
@@ -834,8 +876,15 @@ static int ntb_xeon_setup(struct ntb_device *ndev)
834 SNB_SBAR0BASE_OFFSET); 876 SNB_SBAR0BASE_OFFSET);
835 writeq(SNB_MBAR23_DSD_ADDR, ndev->reg_base + 877 writeq(SNB_MBAR23_DSD_ADDR, ndev->reg_base +
836 SNB_SBAR2BASE_OFFSET); 878 SNB_SBAR2BASE_OFFSET);
837 writeq(SNB_MBAR45_DSD_ADDR, ndev->reg_base + 879 if (ndev->split_bar) {
838 SNB_SBAR4BASE_OFFSET); 880 writel(SNB_MBAR4_DSD_ADDR, ndev->reg_base +
881 SNB_SBAR4BASE_OFFSET);
882 writel(SNB_MBAR5_DSD_ADDR, ndev->reg_base +
883 SNB_SBAR5BASE_OFFSET);
884 } else
885 writeq(SNB_MBAR4_DSD_ADDR, ndev->reg_base +
886 SNB_SBAR4BASE_OFFSET);
887
839 } 888 }
840 break; 889 break;
841 case NTB_CONN_RP: 890 case NTB_CONN_RP:
@@ -865,7 +914,12 @@ static int ntb_xeon_setup(struct ntb_device *ndev)
865 ndev->reg_ofs.spad_read = ndev->reg_base + SNB_SPAD_OFFSET; 914 ndev->reg_ofs.spad_read = ndev->reg_base + SNB_SPAD_OFFSET;
866 ndev->reg_ofs.bar2_xlat = ndev->reg_base + SNB_SBAR2XLAT_OFFSET; 915 ndev->reg_ofs.bar2_xlat = ndev->reg_base + SNB_SBAR2XLAT_OFFSET;
867 ndev->reg_ofs.bar4_xlat = ndev->reg_base + SNB_SBAR4XLAT_OFFSET; 916 ndev->reg_ofs.bar4_xlat = ndev->reg_base + SNB_SBAR4XLAT_OFFSET;
868 ndev->limits.max_mw = SNB_MAX_MW; 917 if (ndev->split_bar) {
918 ndev->reg_ofs.bar5_xlat =
919 ndev->reg_base + SNB_SBAR5XLAT_OFFSET;
920 ndev->limits.max_mw = HSX_SPLITBAR_MAX_MW;
921 } else
922 ndev->limits.max_mw = SNB_MAX_MW;
869 break; 923 break;
870 case NTB_CONN_TRANSPARENT: 924 case NTB_CONN_TRANSPARENT:
871 if (ndev->wa_flags & WA_SNB_ERR) { 925 if (ndev->wa_flags & WA_SNB_ERR) {
@@ -892,7 +946,12 @@ static int ntb_xeon_setup(struct ntb_device *ndev)
892 ndev->reg_ofs.bar2_xlat = ndev->reg_base + SNB_PBAR2XLAT_OFFSET; 946 ndev->reg_ofs.bar2_xlat = ndev->reg_base + SNB_PBAR2XLAT_OFFSET;
893 ndev->reg_ofs.bar4_xlat = ndev->reg_base + SNB_PBAR4XLAT_OFFSET; 947 ndev->reg_ofs.bar4_xlat = ndev->reg_base + SNB_PBAR4XLAT_OFFSET;
894 948
895 ndev->limits.max_mw = SNB_MAX_MW; 949 if (ndev->split_bar) {
950 ndev->reg_ofs.bar5_xlat =
951 ndev->reg_base + SNB_PBAR5XLAT_OFFSET;
952 ndev->limits.max_mw = HSX_SPLITBAR_MAX_MW;
953 } else
954 ndev->limits.max_mw = SNB_MAX_MW;
896 break; 955 break;
897 default: 956 default:
898 /* 957 /*
@@ -1499,7 +1558,11 @@ static void ntb_hw_link_up(struct ntb_device *ndev)
1499 ntb_cntl = readl(ndev->reg_ofs.lnk_cntl); 1558 ntb_cntl = readl(ndev->reg_ofs.lnk_cntl);
1500 ntb_cntl &= ~(NTB_CNTL_LINK_DISABLE | NTB_CNTL_CFG_LOCK); 1559 ntb_cntl &= ~(NTB_CNTL_LINK_DISABLE | NTB_CNTL_CFG_LOCK);
1501 ntb_cntl |= NTB_CNTL_P2S_BAR23_SNOOP | NTB_CNTL_S2P_BAR23_SNOOP; 1560 ntb_cntl |= NTB_CNTL_P2S_BAR23_SNOOP | NTB_CNTL_S2P_BAR23_SNOOP;
1502 ntb_cntl |= NTB_CNTL_P2S_BAR45_SNOOP | NTB_CNTL_S2P_BAR45_SNOOP; 1561 ntb_cntl |= NTB_CNTL_P2S_BAR4_SNOOP | NTB_CNTL_S2P_BAR4_SNOOP;
1562 if (ndev->split_bar)
1563 ntb_cntl |= NTB_CNTL_P2S_BAR5_SNOOP |
1564 NTB_CNTL_S2P_BAR5_SNOOP;
1565
1503 writel(ntb_cntl, ndev->reg_ofs.lnk_cntl); 1566 writel(ntb_cntl, ndev->reg_ofs.lnk_cntl);
1504 } 1567 }
1505} 1568}
@@ -1516,14 +1579,26 @@ static void ntb_hw_link_down(struct ntb_device *ndev)
1516 /* Bring NTB link down */ 1579 /* Bring NTB link down */
1517 ntb_cntl = readl(ndev->reg_ofs.lnk_cntl); 1580 ntb_cntl = readl(ndev->reg_ofs.lnk_cntl);
1518 ntb_cntl &= ~(NTB_CNTL_P2S_BAR23_SNOOP | NTB_CNTL_S2P_BAR23_SNOOP); 1581 ntb_cntl &= ~(NTB_CNTL_P2S_BAR23_SNOOP | NTB_CNTL_S2P_BAR23_SNOOP);
1519 ntb_cntl &= ~(NTB_CNTL_P2S_BAR45_SNOOP | NTB_CNTL_S2P_BAR45_SNOOP); 1582 ntb_cntl &= ~(NTB_CNTL_P2S_BAR4_SNOOP | NTB_CNTL_S2P_BAR4_SNOOP);
1583 if (ndev->split_bar)
1584 ntb_cntl &= ~(NTB_CNTL_P2S_BAR5_SNOOP |
1585 NTB_CNTL_S2P_BAR5_SNOOP);
1520 ntb_cntl |= NTB_CNTL_LINK_DISABLE | NTB_CNTL_CFG_LOCK; 1586 ntb_cntl |= NTB_CNTL_LINK_DISABLE | NTB_CNTL_CFG_LOCK;
1521 writel(ntb_cntl, ndev->reg_ofs.lnk_cntl); 1587 writel(ntb_cntl, ndev->reg_ofs.lnk_cntl);
1522} 1588}
1523 1589
1590static void ntb_max_mw_detect(struct ntb_device *ndev)
1591{
1592 if (ndev->split_bar)
1593 ndev->limits.max_mw = HSX_SPLITBAR_MAX_MW;
1594 else
1595 ndev->limits.max_mw = SNB_MAX_MW;
1596}
1597
1524static int ntb_xeon_detect(struct ntb_device *ndev) 1598static int ntb_xeon_detect(struct ntb_device *ndev)
1525{ 1599{
1526 int rc; 1600 int rc, bars_mask;
1601 u32 bars;
1527 u8 ppd; 1602 u8 ppd;
1528 1603
1529 ndev->hw_type = SNB_HW; 1604 ndev->hw_type = SNB_HW;
@@ -1537,6 +1612,8 @@ static int ntb_xeon_detect(struct ntb_device *ndev)
1537 else 1612 else
1538 ndev->dev_type = NTB_DEV_DSD; 1613 ndev->dev_type = NTB_DEV_DSD;
1539 1614
1615 ndev->split_bar = (ppd & SNB_PPD_SPLIT_BAR) ? 1 : 0;
1616
1540 switch (ppd & SNB_PPD_CONN_TYPE) { 1617 switch (ppd & SNB_PPD_CONN_TYPE) {
1541 case NTB_CONN_B2B: 1618 case NTB_CONN_B2B:
1542 dev_info(&ndev->pdev->dev, "Conn Type = B2B\n"); 1619 dev_info(&ndev->pdev->dev, "Conn Type = B2B\n");
@@ -1555,12 +1632,25 @@ static int ntb_xeon_detect(struct ntb_device *ndev)
1555 * NTB. We will just force correct here. 1632 * NTB. We will just force correct here.
1556 */ 1633 */
1557 ndev->dev_type = NTB_DEV_USD; 1634 ndev->dev_type = NTB_DEV_USD;
1635
1636 /*
1637 * This is a way for transparent BAR to figure out if we
1638 * are doing split BAR or not. There is no way for the hw
1639 * on the transparent side to know and set the PPD.
1640 */
1641 bars_mask = pci_select_bars(ndev->pdev, IORESOURCE_MEM);
1642 bars = hweight32(bars_mask);
1643 if (bars == (HSX_SPLITBAR_MAX_MW + 1))
1644 ndev->split_bar = 1;
1645
1558 break; 1646 break;
1559 default: 1647 default:
1560 dev_err(&ndev->pdev->dev, "Unknown PPD %x\n", ppd); 1648 dev_err(&ndev->pdev->dev, "Unknown PPD %x\n", ppd);
1561 return -ENODEV; 1649 return -ENODEV;
1562 } 1650 }
1563 1651
1652 ntb_max_mw_detect(ndev);
1653
1564 return 0; 1654 return 0;
1565} 1655}
1566 1656
@@ -1638,22 +1728,50 @@ static int ntb_pci_probe(struct pci_dev *pdev, const struct pci_device_id *id)
1638 if (rc) 1728 if (rc)
1639 goto err; 1729 goto err;
1640 1730
1641 rc = pci_request_selected_regions(pdev, NTB_BAR_MASK, KBUILD_MODNAME); 1731 ndev->mw = kcalloc(ndev->limits.max_mw, sizeof(struct ntb_mw),
1642 if (rc) 1732 GFP_KERNEL);
1733 if (!ndev->mw) {
1734 rc = -ENOMEM;
1643 goto err1; 1735 goto err1;
1736 }
1737
1738 if (ndev->split_bar)
1739 rc = pci_request_selected_regions(pdev, NTB_SPLITBAR_MASK,
1740 KBUILD_MODNAME);
1741 else
1742 rc = pci_request_selected_regions(pdev, NTB_BAR_MASK,
1743 KBUILD_MODNAME);
1744
1745 if (rc)
1746 goto err2;
1644 1747
1645 ndev->reg_base = pci_ioremap_bar(pdev, NTB_BAR_MMIO); 1748 ndev->reg_base = pci_ioremap_bar(pdev, NTB_BAR_MMIO);
1646 if (!ndev->reg_base) { 1749 if (!ndev->reg_base) {
1647 dev_warn(&pdev->dev, "Cannot remap BAR 0\n"); 1750 dev_warn(&pdev->dev, "Cannot remap BAR 0\n");
1648 rc = -EIO; 1751 rc = -EIO;
1649 goto err2; 1752 goto err3;
1650 } 1753 }
1651 1754
1652 for (i = 0; i < NTB_MAX_NUM_MW; i++) { 1755 for (i = 0; i < ndev->limits.max_mw; i++) {
1653 ndev->mw[i].bar_sz = pci_resource_len(pdev, MW_TO_BAR(i)); 1756 ndev->mw[i].bar_sz = pci_resource_len(pdev, MW_TO_BAR(i));
1654 ndev->mw[i].vbase = 1757
1655 ioremap_wc(pci_resource_start(pdev, MW_TO_BAR(i)), 1758 /*
1656 ndev->mw[i].bar_sz); 1759 * with the errata we need to steal last of the memory
1760 * windows for workarounds and they point to MMIO registers.
1761 */
1762 if ((ndev->wa_flags & WA_SNB_ERR) &&
1763 (i == (ndev->limits.max_mw - 1))) {
1764 ndev->mw[i].vbase =
1765 ioremap_nocache(pci_resource_start(pdev,
1766 MW_TO_BAR(i)),
1767 ndev->mw[i].bar_sz);
1768 } else {
1769 ndev->mw[i].vbase =
1770 ioremap_wc(pci_resource_start(pdev,
1771 MW_TO_BAR(i)),
1772 ndev->mw[i].bar_sz);
1773 }
1774
1657 dev_info(&pdev->dev, "MW %d size %llu\n", i, 1775 dev_info(&pdev->dev, "MW %d size %llu\n", i,
1658 (unsigned long long) ndev->mw[i].bar_sz); 1776 (unsigned long long) ndev->mw[i].bar_sz);
1659 if (!ndev->mw[i].vbase) { 1777 if (!ndev->mw[i].vbase) {
@@ -1668,7 +1786,7 @@ static int ntb_pci_probe(struct pci_dev *pdev, const struct pci_device_id *id)
1668 if (rc) { 1786 if (rc) {
1669 rc = pci_set_dma_mask(pdev, DMA_BIT_MASK(32)); 1787 rc = pci_set_dma_mask(pdev, DMA_BIT_MASK(32));
1670 if (rc) 1788 if (rc)
1671 goto err3; 1789 goto err4;
1672 1790
1673 dev_warn(&pdev->dev, "Cannot DMA highmem\n"); 1791 dev_warn(&pdev->dev, "Cannot DMA highmem\n");
1674 } 1792 }
@@ -1677,22 +1795,22 @@ static int ntb_pci_probe(struct pci_dev *pdev, const struct pci_device_id *id)
1677 if (rc) { 1795 if (rc) {
1678 rc = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(32)); 1796 rc = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(32));
1679 if (rc) 1797 if (rc)
1680 goto err3; 1798 goto err4;
1681 1799
1682 dev_warn(&pdev->dev, "Cannot DMA consistent highmem\n"); 1800 dev_warn(&pdev->dev, "Cannot DMA consistent highmem\n");
1683 } 1801 }
1684 1802
1685 rc = ntb_device_setup(ndev); 1803 rc = ntb_device_setup(ndev);
1686 if (rc) 1804 if (rc)
1687 goto err3; 1805 goto err4;
1688 1806
1689 rc = ntb_create_callbacks(ndev); 1807 rc = ntb_create_callbacks(ndev);
1690 if (rc) 1808 if (rc)
1691 goto err4; 1809 goto err5;
1692 1810
1693 rc = ntb_setup_interrupts(ndev); 1811 rc = ntb_setup_interrupts(ndev);
1694 if (rc) 1812 if (rc)
1695 goto err5; 1813 goto err6;
1696 1814
1697 /* The scratchpad registers keep the values between rmmod/insmod, 1815 /* The scratchpad registers keep the values between rmmod/insmod,
1698 * blast them now 1816 * blast them now
@@ -1704,24 +1822,29 @@ static int ntb_pci_probe(struct pci_dev *pdev, const struct pci_device_id *id)
1704 1822
1705 rc = ntb_transport_init(pdev); 1823 rc = ntb_transport_init(pdev);
1706 if (rc) 1824 if (rc)
1707 goto err6; 1825 goto err7;
1708 1826
1709 ntb_hw_link_up(ndev); 1827 ntb_hw_link_up(ndev);
1710 1828
1711 return 0; 1829 return 0;
1712 1830
1713err6: 1831err7:
1714 ntb_free_interrupts(ndev); 1832 ntb_free_interrupts(ndev);
1715err5: 1833err6:
1716 ntb_free_callbacks(ndev); 1834 ntb_free_callbacks(ndev);
1717err4: 1835err5:
1718 ntb_device_free(ndev); 1836 ntb_device_free(ndev);
1719err3: 1837err4:
1720 for (i--; i >= 0; i--) 1838 for (i--; i >= 0; i--)
1721 iounmap(ndev->mw[i].vbase); 1839 iounmap(ndev->mw[i].vbase);
1722 iounmap(ndev->reg_base); 1840 iounmap(ndev->reg_base);
1841err3:
1842 if (ndev->split_bar)
1843 pci_release_selected_regions(pdev, NTB_SPLITBAR_MASK);
1844 else
1845 pci_release_selected_regions(pdev, NTB_BAR_MASK);
1723err2: 1846err2:
1724 pci_release_selected_regions(pdev, NTB_BAR_MASK); 1847 kfree(ndev->mw);
1725err1: 1848err1:
1726 pci_disable_device(pdev); 1849 pci_disable_device(pdev);
1727err: 1850err:
@@ -1745,11 +1868,19 @@ static void ntb_pci_remove(struct pci_dev *pdev)
1745 ntb_free_callbacks(ndev); 1868 ntb_free_callbacks(ndev);
1746 ntb_device_free(ndev); 1869 ntb_device_free(ndev);
1747 1870
1748 for (i = 0; i < NTB_MAX_NUM_MW; i++) 1871 /* need to reset max_mw limits so we can unmap properly */
1872 if (ndev->hw_type == SNB_HW)
1873 ntb_max_mw_detect(ndev);
1874
1875 for (i = 0; i < ndev->limits.max_mw; i++)
1749 iounmap(ndev->mw[i].vbase); 1876 iounmap(ndev->mw[i].vbase);
1750 1877
1878 kfree(ndev->mw);
1751 iounmap(ndev->reg_base); 1879 iounmap(ndev->reg_base);
1752 pci_release_selected_regions(pdev, NTB_BAR_MASK); 1880 if (ndev->split_bar)
1881 pci_release_selected_regions(pdev, NTB_SPLITBAR_MASK);
1882 else
1883 pci_release_selected_regions(pdev, NTB_BAR_MASK);
1753 pci_disable_device(pdev); 1884 pci_disable_device(pdev);
1754 ntb_free_debugfs(ndev); 1885 ntb_free_debugfs(ndev);
1755 kfree(ndev); 1886 kfree(ndev);
diff --git a/drivers/ntb/ntb_hw.h b/drivers/ntb/ntb_hw.h
index 5380ca16198a..96de5fc95f90 100644
--- a/drivers/ntb/ntb_hw.h
+++ b/drivers/ntb/ntb_hw.h
@@ -78,14 +78,16 @@ static inline void writeq(u64 val, void __iomem *addr)
78 78
79#define NTB_BAR_MMIO 0 79#define NTB_BAR_MMIO 0
80#define NTB_BAR_23 2 80#define NTB_BAR_23 2
81#define NTB_BAR_45 4 81#define NTB_BAR_4 4
82#define NTB_BAR_5 5
83
82#define NTB_BAR_MASK ((1 << NTB_BAR_MMIO) | (1 << NTB_BAR_23) |\ 84#define NTB_BAR_MASK ((1 << NTB_BAR_MMIO) | (1 << NTB_BAR_23) |\
83 (1 << NTB_BAR_45)) 85 (1 << NTB_BAR_4))
86#define NTB_SPLITBAR_MASK ((1 << NTB_BAR_MMIO) | (1 << NTB_BAR_23) |\
87 (1 << NTB_BAR_4) | (1 << NTB_BAR_5))
84 88
85#define NTB_HB_TIMEOUT msecs_to_jiffies(1000) 89#define NTB_HB_TIMEOUT msecs_to_jiffies(1000)
86 90
87#define NTB_MAX_NUM_MW 2
88
89enum ntb_hw_event { 91enum ntb_hw_event {
90 NTB_EVENT_SW_EVENT0 = 0, 92 NTB_EVENT_SW_EVENT0 = 0,
91 NTB_EVENT_SW_EVENT1, 93 NTB_EVENT_SW_EVENT1,
@@ -115,7 +117,7 @@ struct ntb_device {
115 struct pci_dev *pdev; 117 struct pci_dev *pdev;
116 struct msix_entry *msix_entries; 118 struct msix_entry *msix_entries;
117 void __iomem *reg_base; 119 void __iomem *reg_base;
118 struct ntb_mw mw[NTB_MAX_NUM_MW]; 120 struct ntb_mw *mw;
119 struct { 121 struct {
120 unsigned char max_mw; 122 unsigned char max_mw;
121 unsigned char max_spads; 123 unsigned char max_spads;
@@ -128,6 +130,7 @@ struct ntb_device {
128 void __iomem *rdb; 130 void __iomem *rdb;
129 void __iomem *bar2_xlat; 131 void __iomem *bar2_xlat;
130 void __iomem *bar4_xlat; 132 void __iomem *bar4_xlat;
133 void __iomem *bar5_xlat;
131 void __iomem *spad_write; 134 void __iomem *spad_write;
132 void __iomem *spad_read; 135 void __iomem *spad_read;
133 void __iomem *lnk_cntl; 136 void __iomem *lnk_cntl;
@@ -147,6 +150,7 @@ struct ntb_device {
147 unsigned char link_width; 150 unsigned char link_width;
148 unsigned char link_speed; 151 unsigned char link_speed;
149 unsigned char link_status; 152 unsigned char link_status;
153 unsigned char split_bar;
150 154
151 struct delayed_work hb_timer; 155 struct delayed_work hb_timer;
152 unsigned long last_ts; 156 unsigned long last_ts;
diff --git a/drivers/ntb/ntb_regs.h b/drivers/ntb/ntb_regs.h
index 07872057c76e..f028ff81fd77 100644
--- a/drivers/ntb/ntb_regs.h
+++ b/drivers/ntb/ntb_regs.h
@@ -57,6 +57,7 @@
57#define SNB_MAX_DB_BITS 15 57#define SNB_MAX_DB_BITS 15
58#define SNB_LINK_DB 15 58#define SNB_LINK_DB 15
59#define SNB_DB_BITS_PER_VEC 5 59#define SNB_DB_BITS_PER_VEC 5
60#define HSX_SPLITBAR_MAX_MW 3
60#define SNB_MAX_MW 2 61#define SNB_MAX_MW 2
61#define SNB_ERRATA_MAX_MW 1 62#define SNB_ERRATA_MAX_MW 1
62 63
@@ -72,15 +73,20 @@
72 73
73#define SNB_PBAR2LMT_OFFSET 0x0000 74#define SNB_PBAR2LMT_OFFSET 0x0000
74#define SNB_PBAR4LMT_OFFSET 0x0008 75#define SNB_PBAR4LMT_OFFSET 0x0008
76#define SNB_PBAR5LMT_OFFSET 0x000C
75#define SNB_PBAR2XLAT_OFFSET 0x0010 77#define SNB_PBAR2XLAT_OFFSET 0x0010
76#define SNB_PBAR4XLAT_OFFSET 0x0018 78#define SNB_PBAR4XLAT_OFFSET 0x0018
79#define SNB_PBAR5XLAT_OFFSET 0x001C
77#define SNB_SBAR2LMT_OFFSET 0x0020 80#define SNB_SBAR2LMT_OFFSET 0x0020
78#define SNB_SBAR4LMT_OFFSET 0x0028 81#define SNB_SBAR4LMT_OFFSET 0x0028
82#define SNB_SBAR5LMT_OFFSET 0x002C
79#define SNB_SBAR2XLAT_OFFSET 0x0030 83#define SNB_SBAR2XLAT_OFFSET 0x0030
80#define SNB_SBAR4XLAT_OFFSET 0x0038 84#define SNB_SBAR4XLAT_OFFSET 0x0038
85#define SNB_SBAR5XLAT_OFFSET 0x003C
81#define SNB_SBAR0BASE_OFFSET 0x0040 86#define SNB_SBAR0BASE_OFFSET 0x0040
82#define SNB_SBAR2BASE_OFFSET 0x0048 87#define SNB_SBAR2BASE_OFFSET 0x0048
83#define SNB_SBAR4BASE_OFFSET 0x0050 88#define SNB_SBAR4BASE_OFFSET 0x0050
89#define SNB_SBAR5BASE_OFFSET 0x0054
84#define SNB_NTBCNTL_OFFSET 0x0058 90#define SNB_NTBCNTL_OFFSET 0x0058
85#define SNB_SBDF_OFFSET 0x005C 91#define SNB_SBDF_OFFSET 0x005C
86#define SNB_PDOORBELL_OFFSET 0x0060 92#define SNB_PDOORBELL_OFFSET 0x0060
@@ -96,12 +102,18 @@
96#define SNB_B2B_XLAT_OFFSETL 0x0144 102#define SNB_B2B_XLAT_OFFSETL 0x0144
97#define SNB_B2B_XLAT_OFFSETU 0x0148 103#define SNB_B2B_XLAT_OFFSETU 0x0148
98 104
99#define SNB_MBAR01_USD_ADDR 0x000000210000000CULL 105/*
100#define SNB_MBAR23_USD_ADDR 0x000000410000000CULL 106 * The addresses are setup so the 32bit BARs can function. Thus
101#define SNB_MBAR45_USD_ADDR 0x000000810000000CULL 107 * the addresses are all in 32bit space
102#define SNB_MBAR01_DSD_ADDR 0x000000200000000CULL 108 */
103#define SNB_MBAR23_DSD_ADDR 0x000000400000000CULL 109#define SNB_MBAR01_USD_ADDR 0x000000002100000CULL
104#define SNB_MBAR45_DSD_ADDR 0x000000800000000CULL 110#define SNB_MBAR23_USD_ADDR 0x000000004100000CULL
111#define SNB_MBAR4_USD_ADDR 0x000000008100000CULL
112#define SNB_MBAR5_USD_ADDR 0x00000000A100000CULL
113#define SNB_MBAR01_DSD_ADDR 0x000000002000000CULL
114#define SNB_MBAR23_DSD_ADDR 0x000000004000000CULL
115#define SNB_MBAR4_DSD_ADDR 0x000000008000000CULL
116#define SNB_MBAR5_DSD_ADDR 0x00000000A000000CULL
105 117
106#define BWD_MSIX_CNT 34 118#define BWD_MSIX_CNT 34
107#define BWD_MAX_SPADS 16 119#define BWD_MAX_SPADS 16
@@ -150,13 +162,16 @@
150#define NTB_CNTL_LINK_DISABLE (1 << 1) 162#define NTB_CNTL_LINK_DISABLE (1 << 1)
151#define NTB_CNTL_S2P_BAR23_SNOOP (1 << 2) 163#define NTB_CNTL_S2P_BAR23_SNOOP (1 << 2)
152#define NTB_CNTL_P2S_BAR23_SNOOP (1 << 4) 164#define NTB_CNTL_P2S_BAR23_SNOOP (1 << 4)
153#define NTB_CNTL_S2P_BAR45_SNOOP (1 << 6) 165#define NTB_CNTL_S2P_BAR4_SNOOP (1 << 6)
154#define NTB_CNTL_P2S_BAR45_SNOOP (1 << 8) 166#define NTB_CNTL_P2S_BAR4_SNOOP (1 << 8)
167#define NTB_CNTL_S2P_BAR5_SNOOP (1 << 12)
168#define NTB_CNTL_P2S_BAR5_SNOOP (1 << 14)
155#define BWD_CNTL_LINK_DOWN (1 << 16) 169#define BWD_CNTL_LINK_DOWN (1 << 16)
156 170
157#define NTB_PPD_OFFSET 0x00D4 171#define NTB_PPD_OFFSET 0x00D4
158#define SNB_PPD_CONN_TYPE 0x0003 172#define SNB_PPD_CONN_TYPE 0x0003
159#define SNB_PPD_DEV_TYPE 0x0010 173#define SNB_PPD_DEV_TYPE 0x0010
174#define SNB_PPD_SPLIT_BAR (1 << 6)
160#define BWD_PPD_INIT_LINK 0x0008 175#define BWD_PPD_INIT_LINK 0x0008
161#define BWD_PPD_CONN_TYPE 0x0300 176#define BWD_PPD_CONN_TYPE 0x0300
162#define BWD_PPD_DEV_TYPE 0x1000 177#define BWD_PPD_DEV_TYPE 0x1000