#define BWD_LINK_RECOVERY_TIME 500
-/* Translate memory window 0,1 to BAR 2,4 */
-#define MW_TO_BAR(mw) (mw * NTB_MAX_NUM_MW + 2)
+/* Translate memory window 0,1,2 to BAR 2,4,5 */
+#define MW_TO_BAR(mw) (mw == 0 ? 2 : (mw == 1 ? 4 : 5))
static const struct pci_device_id ntb_pci_tbl[] = {
{PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_NTB_B2B_BWD)},
case NTB_BAR_23:
writeq(addr, ndev->reg_ofs.bar2_xlat);
break;
- case NTB_BAR_45:
- writeq(addr, ndev->reg_ofs.bar4_xlat);
+ case NTB_BAR_4:
+ if (ndev->split_bar)
+ writel(addr, ndev->reg_ofs.bar4_xlat);
+ else
+ writeq(addr, ndev->reg_ofs.bar4_xlat);
+ break;
+ case NTB_BAR_5:
+ writel(addr, ndev->reg_ofs.bar5_xlat);
break;
}
}
ndev->reg_ofs.spad_read = ndev->reg_base + SNB_SPAD_OFFSET;
ndev->reg_ofs.bar2_xlat = ndev->reg_base + SNB_SBAR2XLAT_OFFSET;
ndev->reg_ofs.bar4_xlat = ndev->reg_base + SNB_SBAR4XLAT_OFFSET;
+ if (ndev->split_bar)
+ ndev->reg_ofs.bar5_xlat =
+ ndev->reg_base + SNB_SBAR5XLAT_OFFSET;
ndev->limits.max_spads = SNB_MAX_B2B_SPADS;
/* There is a Xeon hardware errata related to writes to
* scratch pad registers on the remote system.
*/
if (ndev->wa_flags & WA_SNB_ERR) {
- if (!ndev->mw[1].bar_sz)
+ if (!ndev->mw[ndev->limits.max_mw - 1].bar_sz)
return -EINVAL;
- ndev->limits.max_mw = SNB_ERRATA_MAX_MW;
ndev->limits.max_db_bits = SNB_MAX_DB_BITS;
- ndev->reg_ofs.spad_write = ndev->mw[1].vbase +
- SNB_SPAD_OFFSET;
- ndev->reg_ofs.rdb = ndev->mw[1].vbase +
- SNB_PDOORBELL_OFFSET;
+ ndev->reg_ofs.spad_write =
+ ndev->mw[ndev->limits.max_mw - 1].vbase +
+ SNB_SPAD_OFFSET;
+ ndev->reg_ofs.rdb =
+ ndev->mw[ndev->limits.max_mw - 1].vbase +
+ SNB_PDOORBELL_OFFSET;
/* Set the Limit register to 4k, the minimum size, to
* prevent an illegal access
* the driver defaults, but write the Limit registers
* first just in case.
*/
- } else {
- ndev->limits.max_mw = SNB_MAX_MW;
+ ndev->limits.max_mw = SNB_ERRATA_MAX_MW;
+ } else {
/* HW Errata on bit 14 of b2bdoorbell register. Writes
* will not be mirrored to the remote system. Shrink
* the number of bits by one, since bit 14 is the last
SNB_B2B_DOORBELL_OFFSET;
/* Disable the Limit register, just incase it is set to
- * something silly
+ * something silly. A 64bit write should handle it
+ * regardless of whether it has a split BAR or not.
*/
writeq(0, ndev->reg_base + SNB_PBAR4LMT_OFFSET);
/* HW errata on the Limit registers. They can only be
* the driver defaults, but write the Limit registers
* first just in case.
*/
+ if (ndev->split_bar)
+ ndev->limits.max_mw = HSX_SPLITBAR_MAX_MW;
+ else
+ ndev->limits.max_mw = SNB_MAX_MW;
}
/* The Xeon errata workaround requires setting SBAR Base
writeq(SNB_MBAR01_DSD_ADDR, ndev->reg_base +
SNB_PBAR4XLAT_OFFSET);
else {
- writeq(SNB_MBAR45_DSD_ADDR, ndev->reg_base +
- SNB_PBAR4XLAT_OFFSET);
+ if (ndev->split_bar) {
+ writel(SNB_MBAR4_DSD_ADDR,
+ ndev->reg_base +
+ SNB_PBAR4XLAT_OFFSET);
+ writel(SNB_MBAR5_DSD_ADDR,
+ ndev->reg_base +
+ SNB_PBAR5XLAT_OFFSET);
+ } else
+ writeq(SNB_MBAR4_DSD_ADDR,
+ ndev->reg_base +
+ SNB_PBAR4XLAT_OFFSET);
+
/* B2B_XLAT_OFFSET is a 64bit register, but can
* only take 32bit writes
*/
SNB_SBAR0BASE_OFFSET);
writeq(SNB_MBAR23_USD_ADDR, ndev->reg_base +
SNB_SBAR2BASE_OFFSET);
- writeq(SNB_MBAR45_USD_ADDR, ndev->reg_base +
- SNB_SBAR4BASE_OFFSET);
+ if (ndev->split_bar) {
+ writel(SNB_MBAR4_USD_ADDR, ndev->reg_base +
+ SNB_SBAR4BASE_OFFSET);
+ writel(SNB_MBAR5_USD_ADDR, ndev->reg_base +
+ SNB_SBAR5BASE_OFFSET);
+ } else
+ writeq(SNB_MBAR4_USD_ADDR, ndev->reg_base +
+ SNB_SBAR4BASE_OFFSET);
} else {
writeq(SNB_MBAR23_USD_ADDR, ndev->reg_base +
SNB_PBAR2XLAT_OFFSET);
writeq(SNB_MBAR01_USD_ADDR, ndev->reg_base +
SNB_PBAR4XLAT_OFFSET);
else {
- writeq(SNB_MBAR45_USD_ADDR, ndev->reg_base +
- SNB_PBAR4XLAT_OFFSET);
- /* B2B_XLAT_OFFSET is a 64bit register, but can
+ if (ndev->split_bar) {
+ writel(SNB_MBAR4_USD_ADDR,
+ ndev->reg_base +
+ SNB_PBAR4XLAT_OFFSET);
+ writel(SNB_MBAR5_USD_ADDR,
+ ndev->reg_base +
+ SNB_PBAR5XLAT_OFFSET);
+ } else
+ writeq(SNB_MBAR4_USD_ADDR,
+ ndev->reg_base +
+ SNB_PBAR4XLAT_OFFSET);
+
+ /*
+ * B2B_XLAT_OFFSET is a 64bit register, but can
* only take 32bit writes
*/
writel(SNB_MBAR01_USD_ADDR & 0xffffffff,
SNB_SBAR0BASE_OFFSET);
writeq(SNB_MBAR23_DSD_ADDR, ndev->reg_base +
SNB_SBAR2BASE_OFFSET);
- writeq(SNB_MBAR45_DSD_ADDR, ndev->reg_base +
- SNB_SBAR4BASE_OFFSET);
+ if (ndev->split_bar) {
+ writel(SNB_MBAR4_DSD_ADDR, ndev->reg_base +
+ SNB_SBAR4BASE_OFFSET);
+ writel(SNB_MBAR5_DSD_ADDR, ndev->reg_base +
+ SNB_SBAR5BASE_OFFSET);
+ } else
+ writeq(SNB_MBAR4_DSD_ADDR, ndev->reg_base +
+ SNB_SBAR4BASE_OFFSET);
+
}
break;
case NTB_CONN_RP:
ndev->reg_ofs.spad_read = ndev->reg_base + SNB_SPAD_OFFSET;
ndev->reg_ofs.bar2_xlat = ndev->reg_base + SNB_SBAR2XLAT_OFFSET;
ndev->reg_ofs.bar4_xlat = ndev->reg_base + SNB_SBAR4XLAT_OFFSET;
- ndev->limits.max_mw = SNB_MAX_MW;
+ if (ndev->split_bar) {
+ ndev->reg_ofs.bar5_xlat =
+ ndev->reg_base + SNB_SBAR5XLAT_OFFSET;
+ ndev->limits.max_mw = HSX_SPLITBAR_MAX_MW;
+ } else
+ ndev->limits.max_mw = SNB_MAX_MW;
break;
case NTB_CONN_TRANSPARENT:
if (ndev->wa_flags & WA_SNB_ERR) {
ndev->reg_ofs.bar2_xlat = ndev->reg_base + SNB_PBAR2XLAT_OFFSET;
ndev->reg_ofs.bar4_xlat = ndev->reg_base + SNB_PBAR4XLAT_OFFSET;
- ndev->limits.max_mw = SNB_MAX_MW;
+ if (ndev->split_bar) {
+ ndev->reg_ofs.bar5_xlat =
+ ndev->reg_base + SNB_PBAR5XLAT_OFFSET;
+ ndev->limits.max_mw = HSX_SPLITBAR_MAX_MW;
+ } else
+ ndev->limits.max_mw = SNB_MAX_MW;
break;
default:
/*
ntb_cntl = readl(ndev->reg_ofs.lnk_cntl);
ntb_cntl &= ~(NTB_CNTL_LINK_DISABLE | NTB_CNTL_CFG_LOCK);
ntb_cntl |= NTB_CNTL_P2S_BAR23_SNOOP | NTB_CNTL_S2P_BAR23_SNOOP;
- ntb_cntl |= NTB_CNTL_P2S_BAR45_SNOOP | NTB_CNTL_S2P_BAR45_SNOOP;
+ ntb_cntl |= NTB_CNTL_P2S_BAR4_SNOOP | NTB_CNTL_S2P_BAR4_SNOOP;
+ if (ndev->split_bar)
+ ntb_cntl |= NTB_CNTL_P2S_BAR5_SNOOP |
+ NTB_CNTL_S2P_BAR5_SNOOP;
+
writel(ntb_cntl, ndev->reg_ofs.lnk_cntl);
}
}
/* Bring NTB link down */
ntb_cntl = readl(ndev->reg_ofs.lnk_cntl);
ntb_cntl &= ~(NTB_CNTL_P2S_BAR23_SNOOP | NTB_CNTL_S2P_BAR23_SNOOP);
- ntb_cntl &= ~(NTB_CNTL_P2S_BAR45_SNOOP | NTB_CNTL_S2P_BAR45_SNOOP);
+ ntb_cntl &= ~(NTB_CNTL_P2S_BAR4_SNOOP | NTB_CNTL_S2P_BAR4_SNOOP);
+ if (ndev->split_bar)
+ ntb_cntl &= ~(NTB_CNTL_P2S_BAR5_SNOOP |
+ NTB_CNTL_S2P_BAR5_SNOOP);
ntb_cntl |= NTB_CNTL_LINK_DISABLE | NTB_CNTL_CFG_LOCK;
writel(ntb_cntl, ndev->reg_ofs.lnk_cntl);
}
+static void ntb_max_mw_detect(struct ntb_device *ndev)
+{
+ if (ndev->split_bar)
+ ndev->limits.max_mw = HSX_SPLITBAR_MAX_MW;
+ else
+ ndev->limits.max_mw = SNB_MAX_MW;
+}
+
static int ntb_xeon_detect(struct ntb_device *ndev)
{
- int rc;
+ int rc, bars_mask;
+ u32 bars;
u8 ppd;
ndev->hw_type = SNB_HW;
else
ndev->dev_type = NTB_DEV_DSD;
+ ndev->split_bar = (ppd & SNB_PPD_SPLIT_BAR) ? 1 : 0;
+
switch (ppd & SNB_PPD_CONN_TYPE) {
case NTB_CONN_B2B:
dev_info(&ndev->pdev->dev, "Conn Type = B2B\n");
* NTB. We will just force correct here.
*/
ndev->dev_type = NTB_DEV_USD;
+
+ /*
+ * This is a way for transparent BAR to figure out if we
+ * are doing split BAR or not. There is no way for the hw
+ * on the transparent side to know and set the PPD.
+ */
+ bars_mask = pci_select_bars(ndev->pdev, IORESOURCE_MEM);
+ bars = hweight32(bars_mask);
+ if (bars == (HSX_SPLITBAR_MAX_MW + 1))
+ ndev->split_bar = 1;
+
break;
default:
dev_err(&ndev->pdev->dev, "Unknown PPD %x\n", ppd);
return -ENODEV;
}
+ ntb_max_mw_detect(ndev);
+
return 0;
}
if (rc)
goto err;
- rc = pci_request_selected_regions(pdev, NTB_BAR_MASK, KBUILD_MODNAME);
- if (rc)
+ ndev->mw = kcalloc(ndev->limits.max_mw, sizeof(struct ntb_mw),
+ GFP_KERNEL);
+ if (!ndev->mw) {
+ rc = -ENOMEM;
goto err1;
+ }
+
+ if (ndev->split_bar)
+ rc = pci_request_selected_regions(pdev, NTB_SPLITBAR_MASK,
+ KBUILD_MODNAME);
+ else
+ rc = pci_request_selected_regions(pdev, NTB_BAR_MASK,
+ KBUILD_MODNAME);
+
+ if (rc)
+ goto err2;
ndev->reg_base = pci_ioremap_bar(pdev, NTB_BAR_MMIO);
if (!ndev->reg_base) {
dev_warn(&pdev->dev, "Cannot remap BAR 0\n");
rc = -EIO;
- goto err2;
+ goto err3;
}
- for (i = 0; i < NTB_MAX_NUM_MW; i++) {
+ for (i = 0; i < ndev->limits.max_mw; i++) {
ndev->mw[i].bar_sz = pci_resource_len(pdev, MW_TO_BAR(i));
- ndev->mw[i].vbase =
- ioremap_wc(pci_resource_start(pdev, MW_TO_BAR(i)),
- ndev->mw[i].bar_sz);
+
+ /*
+ * with the errata we need to steal last of the memory
+ * windows for workarounds and they point to MMIO registers.
+ */
+ if ((ndev->wa_flags & WA_SNB_ERR) &&
+ (i == (ndev->limits.max_mw - 1))) {
+ ndev->mw[i].vbase =
+ ioremap_nocache(pci_resource_start(pdev,
+ MW_TO_BAR(i)),
+ ndev->mw[i].bar_sz);
+ } else {
+ ndev->mw[i].vbase =
+ ioremap_wc(pci_resource_start(pdev,
+ MW_TO_BAR(i)),
+ ndev->mw[i].bar_sz);
+ }
+
dev_info(&pdev->dev, "MW %d size %llu\n", i,
(unsigned long long) ndev->mw[i].bar_sz);
if (!ndev->mw[i].vbase) {
if (rc) {
rc = pci_set_dma_mask(pdev, DMA_BIT_MASK(32));
if (rc)
- goto err3;
+ goto err4;
dev_warn(&pdev->dev, "Cannot DMA highmem\n");
}
if (rc) {
rc = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(32));
if (rc)
- goto err3;
+ goto err4;
dev_warn(&pdev->dev, "Cannot DMA consistent highmem\n");
}
rc = ntb_device_setup(ndev);
if (rc)
- goto err3;
+ goto err4;
rc = ntb_create_callbacks(ndev);
if (rc)
- goto err4;
+ goto err5;
rc = ntb_setup_interrupts(ndev);
if (rc)
- goto err5;
+ goto err6;
/* The scratchpad registers keep the values between rmmod/insmod,
* blast them now
rc = ntb_transport_init(pdev);
if (rc)
- goto err6;
+ goto err7;
ntb_hw_link_up(ndev);
return 0;
-err6:
+err7:
ntb_free_interrupts(ndev);
-err5:
+err6:
ntb_free_callbacks(ndev);
-err4:
+err5:
ntb_device_free(ndev);
-err3:
+err4:
for (i--; i >= 0; i--)
iounmap(ndev->mw[i].vbase);
iounmap(ndev->reg_base);
+err3:
+ if (ndev->split_bar)
+ pci_release_selected_regions(pdev, NTB_SPLITBAR_MASK);
+ else
+ pci_release_selected_regions(pdev, NTB_BAR_MASK);
err2:
- pci_release_selected_regions(pdev, NTB_BAR_MASK);
+ kfree(ndev->mw);
err1:
pci_disable_device(pdev);
err:
ntb_free_callbacks(ndev);
ntb_device_free(ndev);
- for (i = 0; i < NTB_MAX_NUM_MW; i++)
+ /* need to reset max_mw limits so we can unmap properly */
+ if (ndev->hw_type == SNB_HW)
+ ntb_max_mw_detect(ndev);
+
+ for (i = 0; i < ndev->limits.max_mw; i++)
iounmap(ndev->mw[i].vbase);
+ kfree(ndev->mw);
iounmap(ndev->reg_base);
- pci_release_selected_regions(pdev, NTB_BAR_MASK);
+ if (ndev->split_bar)
+ pci_release_selected_regions(pdev, NTB_SPLITBAR_MASK);
+ else
+ pci_release_selected_regions(pdev, NTB_BAR_MASK);
pci_disable_device(pdev);
ntb_free_debugfs(ndev);
kfree(ndev);
#define SNB_MAX_DB_BITS 15
#define SNB_LINK_DB 15
#define SNB_DB_BITS_PER_VEC 5
+#define HSX_SPLITBAR_MAX_MW 3
#define SNB_MAX_MW 2
#define SNB_ERRATA_MAX_MW 1
#define SNB_PBAR2LMT_OFFSET 0x0000
#define SNB_PBAR4LMT_OFFSET 0x0008
+#define SNB_PBAR5LMT_OFFSET 0x000C
#define SNB_PBAR2XLAT_OFFSET 0x0010
#define SNB_PBAR4XLAT_OFFSET 0x0018
+#define SNB_PBAR5XLAT_OFFSET 0x001C
#define SNB_SBAR2LMT_OFFSET 0x0020
#define SNB_SBAR4LMT_OFFSET 0x0028
+#define SNB_SBAR5LMT_OFFSET 0x002C
#define SNB_SBAR2XLAT_OFFSET 0x0030
#define SNB_SBAR4XLAT_OFFSET 0x0038
+#define SNB_SBAR5XLAT_OFFSET 0x003C
#define SNB_SBAR0BASE_OFFSET 0x0040
#define SNB_SBAR2BASE_OFFSET 0x0048
#define SNB_SBAR4BASE_OFFSET 0x0050
+#define SNB_SBAR5BASE_OFFSET 0x0054
#define SNB_NTBCNTL_OFFSET 0x0058
#define SNB_SBDF_OFFSET 0x005C
#define SNB_PDOORBELL_OFFSET 0x0060
#define SNB_B2B_XLAT_OFFSETL 0x0144
#define SNB_B2B_XLAT_OFFSETU 0x0148
-#define SNB_MBAR01_USD_ADDR 0x000000210000000CULL
-#define SNB_MBAR23_USD_ADDR 0x000000410000000CULL
-#define SNB_MBAR45_USD_ADDR 0x000000810000000CULL
-#define SNB_MBAR01_DSD_ADDR 0x000000200000000CULL
-#define SNB_MBAR23_DSD_ADDR 0x000000400000000CULL
-#define SNB_MBAR45_DSD_ADDR 0x000000800000000CULL
+/*
+ * The addresses are setup so the 32bit BARs can function. Thus
+ * the addresses are all in 32bit space
+ */
+#define SNB_MBAR01_USD_ADDR 0x000000002100000CULL
+#define SNB_MBAR23_USD_ADDR 0x000000004100000CULL
+#define SNB_MBAR4_USD_ADDR 0x000000008100000CULL
+#define SNB_MBAR5_USD_ADDR 0x00000000A100000CULL
+#define SNB_MBAR01_DSD_ADDR 0x000000002000000CULL
+#define SNB_MBAR23_DSD_ADDR 0x000000004000000CULL
+#define SNB_MBAR4_DSD_ADDR 0x000000008000000CULL
+#define SNB_MBAR5_DSD_ADDR 0x00000000A000000CULL
#define BWD_MSIX_CNT 34
#define BWD_MAX_SPADS 16
#define NTB_CNTL_LINK_DISABLE (1 << 1)
#define NTB_CNTL_S2P_BAR23_SNOOP (1 << 2)
#define NTB_CNTL_P2S_BAR23_SNOOP (1 << 4)
-#define NTB_CNTL_S2P_BAR45_SNOOP (1 << 6)
-#define NTB_CNTL_P2S_BAR45_SNOOP (1 << 8)
+#define NTB_CNTL_S2P_BAR4_SNOOP (1 << 6)
+#define NTB_CNTL_P2S_BAR4_SNOOP (1 << 8)
+#define NTB_CNTL_S2P_BAR5_SNOOP (1 << 12)
+#define NTB_CNTL_P2S_BAR5_SNOOP (1 << 14)
#define BWD_CNTL_LINK_DOWN (1 << 16)
#define NTB_PPD_OFFSET 0x00D4
#define SNB_PPD_CONN_TYPE 0x0003
#define SNB_PPD_DEV_TYPE 0x0010
+#define SNB_PPD_SPLIT_BAR (1 << 6)
#define BWD_PPD_INIT_LINK 0x0008
#define BWD_PPD_CONN_TYPE 0x0300
#define BWD_PPD_DEV_TYPE 0x1000