/* SPDX-License-Identifier: GPL-2.0-only */ #include #include #include #include #include #include #include static unsigned int pciexp_get_ext_cap_offset(const struct device *dev, unsigned int cap, unsigned int offset) { unsigned int this_cap_offset = offset; unsigned int next_cap_offset, this_cap, cafe; while (this_cap_offset != 0) { this_cap = pci_read_config32(dev, this_cap_offset); /* Bail out when this request is unsupported */ if (this_cap == 0xffffffff) break; cafe = pci_read_config32(dev, this_cap_offset + 4); if ((this_cap & 0xffff) == cap) { return this_cap_offset; } else if ((cafe & 0xffff) == cap) { return this_cap_offset + 4; } else { next_cap_offset = this_cap >> 20; this_cap_offset = next_cap_offset; } } return 0; } /* * Search for an extended capability with the ID `cap`. * * Returns the offset of the first matching extended * capability if found, or 0 otherwise. * * A new search is started with `offset == 0`. * To continue a search, the prior return value * should be passed as `offset`. */ unsigned int pciexp_find_extended_cap(const struct device *dev, unsigned int cap, unsigned int offset) { unsigned int next_cap_offset; if (offset) next_cap_offset = pci_read_config32(dev, offset) >> 20; else next_cap_offset = PCIE_EXT_CAP_OFFSET; return pciexp_get_ext_cap_offset(dev, cap, next_cap_offset); } /* * Re-train a PCIe link */ #define PCIE_TRAIN_RETRY 10000 static int pciexp_retrain_link(struct device *dev, unsigned int cap) { unsigned int try; u16 lnk; /* * Implementation note (page 633) in PCIe Specification 3.0 suggests * polling the Link Training bit in the Link Status register until the * value returned is 0 before setting the Retrain Link bit to 1. * This is meant to avoid a race condition when using the * Retrain Link mechanism. */ for (try = PCIE_TRAIN_RETRY; try > 0; try--) { lnk = pci_read_config16(dev, cap + PCI_EXP_LNKSTA); if (!(lnk & PCI_EXP_LNKSTA_LT)) break; udelay(100); } if (try == 0) { printk(BIOS_ERR, "%s: Link Retrain timeout\n", dev_path(dev)); return -1; } /* Start link retraining */ lnk = pci_read_config16(dev, cap + PCI_EXP_LNKCTL); lnk |= PCI_EXP_LNKCTL_RL; pci_write_config16(dev, cap + PCI_EXP_LNKCTL, lnk); /* Wait for training to complete */ for (try = PCIE_TRAIN_RETRY; try > 0; try--) { lnk = pci_read_config16(dev, cap + PCI_EXP_LNKSTA); if (!(lnk & PCI_EXP_LNKSTA_LT)) return 0; udelay(100); } printk(BIOS_ERR, "%s: Link Retrain timeout\n", dev_path(dev)); return -1; } /* * Check the Slot Clock Configuration for root port and endpoint * and enable Common Clock Configuration if possible. If CCC is * enabled the link must be retrained. */ static void pciexp_enable_common_clock(struct device *root, unsigned int root_cap, struct device *endp, unsigned int endp_cap) { u16 root_scc, endp_scc, lnkctl; /* Get Slot Clock Configuration for root port */ root_scc = pci_read_config16(root, root_cap + PCI_EXP_LNKSTA); root_scc &= PCI_EXP_LNKSTA_SLC; /* Get Slot Clock Configuration for endpoint */ endp_scc = pci_read_config16(endp, endp_cap + PCI_EXP_LNKSTA); endp_scc &= PCI_EXP_LNKSTA_SLC; /* Enable Common Clock Configuration and retrain */ if (root_scc && endp_scc) { printk(BIOS_INFO, "Enabling Common Clock Configuration\n"); /* Set in endpoint */ lnkctl = pci_read_config16(endp, endp_cap + PCI_EXP_LNKCTL); lnkctl |= PCI_EXP_LNKCTL_CCC; pci_write_config16(endp, endp_cap + PCI_EXP_LNKCTL, lnkctl); /* Set in root port */ lnkctl = pci_read_config16(root, root_cap + PCI_EXP_LNKCTL); lnkctl |= PCI_EXP_LNKCTL_CCC; pci_write_config16(root, root_cap + PCI_EXP_LNKCTL, lnkctl); /* Retrain link if CCC was enabled */ pciexp_retrain_link(root, root_cap); } } static void pciexp_enable_clock_power_pm(struct device *endp, unsigned int endp_cap) { /* check if per port clk req is supported in device */ u32 endp_ca; u16 lnkctl; endp_ca = pci_read_config32(endp, endp_cap + PCI_EXP_LNKCAP); if ((endp_ca & PCI_EXP_CLK_PM) == 0) { printk(BIOS_INFO, "PCIE CLK PM is not supported by endpoint\n"); return; } lnkctl = pci_read_config16(endp, endp_cap + PCI_EXP_LNKCTL); lnkctl = lnkctl | PCI_EXP_EN_CLK_PM; pci_write_config16(endp, endp_cap + PCI_EXP_LNKCTL, lnkctl); } static bool _pciexp_ltr_supported(struct device *dev, unsigned int cap) { return pci_read_config16(dev, cap + PCI_EXP_DEVCAP2) & PCI_EXP_DEVCAP2_LTR; } static bool _pciexp_ltr_enabled(struct device *dev, unsigned int cap) { return pci_read_config16(dev, cap + PCI_EXP_DEVCTL2) & PCI_EXP_DEV2_LTR; } static bool _pciexp_enable_ltr(struct device *parent, unsigned int parent_cap, struct device *dev, unsigned int cap) { if (!_pciexp_ltr_supported(dev, cap)) { printk(BIOS_DEBUG, "%s: No LTR support\n", dev_path(dev)); return false; } if (_pciexp_ltr_enabled(dev, cap)) return true; if (parent && (parent->path.type != DEVICE_PATH_PCI || !_pciexp_ltr_supported(parent, parent_cap) || !_pciexp_ltr_enabled(parent, parent_cap))) return false; pci_or_config16(dev, cap + PCI_EXP_DEVCTL2, PCI_EXP_DEV2_LTR); printk(BIOS_INFO, "%s: Enabled LTR\n", dev_path(dev)); return true; } static void pciexp_enable_ltr(struct device *dev) { const unsigned int cap = pci_find_capability(dev, PCI_CAP_ID_PCIE); if (!cap) return; /* * If we have get_ltr_max_latencies(), treat `dev` as the root. * If not, let _pciexp_enable_ltr() query the parent's state. */ struct device *parent = NULL; unsigned int parent_cap = 0; if (!dev->ops->ops_pci || !dev->ops->ops_pci->get_ltr_max_latencies) { parent = dev->bus->dev; parent_cap = pci_find_capability(parent, PCI_CAP_ID_PCIE); if (!parent_cap) return; } (void)_pciexp_enable_ltr(parent, parent_cap, dev, cap); } bool pciexp_get_ltr_max_latencies(struct device *dev, u16 *max_snoop, u16 *max_nosnoop) { /* Walk the hierarchy up to find get_ltr_max_latencies(). */ do { if (dev->ops->ops_pci && dev->ops->ops_pci->get_ltr_max_latencies) break; if (dev->bus->dev == dev || dev->bus->dev->path.type != DEVICE_PATH_PCI) return false; dev = dev->bus->dev; } while (true); dev->ops->ops_pci->get_ltr_max_latencies(max_snoop, max_nosnoop); return true; } static void pciexp_configure_ltr(struct device *parent, unsigned int parent_cap, struct device *dev, unsigned int cap) { if (!_pciexp_enable_ltr(parent, parent_cap, dev, cap)) return; const unsigned int ltr_cap = pciexp_find_extended_cap(dev, PCIE_EXT_CAP_LTR_ID, 0); if (!ltr_cap) return; u16 max_snoop, max_nosnoop; if (!pciexp_get_ltr_max_latencies(dev, &max_snoop, &max_nosnoop)) return; pci_write_config16(dev, ltr_cap + PCI_LTR_MAX_SNOOP, max_snoop); pci_write_config16(dev, ltr_cap + PCI_LTR_MAX_NOSNOOP, max_nosnoop); printk(BIOS_INFO, "%s: Programmed LTR max latencies\n", dev_path(dev)); } static unsigned char pciexp_L1_substate_cal(struct device *dev, unsigned int endp_cap, unsigned int *data) { unsigned char mult[4] = {2, 10, 100, 0}; unsigned int L1SubStateSupport = *data & 0xf; unsigned int comm_mode_rst_time = (*data >> 8) & 0xff; unsigned int power_on_scale = (*data >> 16) & 0x3; unsigned int power_on_value = (*data >> 19) & 0x1f; unsigned int endp_data = pci_read_config32(dev, endp_cap + 4); unsigned int endp_L1SubStateSupport = endp_data & 0xf; unsigned int endp_comm_mode_restore_time = (endp_data >> 8) & 0xff; unsigned int endp_power_on_scale = (endp_data >> 16) & 0x3; unsigned int endp_power_on_value = (endp_data >> 19) & 0x1f; L1SubStateSupport &= endp_L1SubStateSupport; if (L1SubStateSupport == 0) return 0; if (power_on_value * mult[power_on_scale] < endp_power_on_value * mult[endp_power_on_scale]) { power_on_value = endp_power_on_value; power_on_scale = endp_power_on_scale; } if (comm_mode_rst_time < endp_comm_mode_restore_time) comm_mode_rst_time = endp_comm_mode_restore_time; *data = (comm_mode_rst_time << 8) | (power_on_scale << 16) | (power_on_value << 19) | L1SubStateSupport; return 1; } static void pciexp_L1_substate_commit(struct device *root, struct device *dev, unsigned int root_cap, unsigned int end_cap) { struct device *dev_t; unsigned char L1_ss_ok; unsigned int rp_L1_support = pci_read_config32(root, root_cap + 4); unsigned int L1SubStateSupport; unsigned int comm_mode_rst_time; unsigned int power_on_scale; unsigned int endp_power_on_value; for (dev_t = dev; dev_t; dev_t = dev_t->sibling) { /* * rp_L1_support is init'd above from root port. * it needs coordination with endpoints to reach in common. * if certain endpoint doesn't support L1 Sub-State, abort * this feature enabling. */ L1_ss_ok = pciexp_L1_substate_cal(dev_t, end_cap, &rp_L1_support); if (!L1_ss_ok) return; } L1SubStateSupport = rp_L1_support & 0xf; comm_mode_rst_time = (rp_L1_support >> 8) & 0xff; power_on_scale = (rp_L1_support >> 16) & 0x3; endp_power_on_value = (rp_L1_support >> 19) & 0x1f; printk(BIOS_INFO, "L1 Sub-State supported from root port %d\n", root->path.pci.devfn >> 3); printk(BIOS_INFO, "L1 Sub-State Support = 0x%x\n", L1SubStateSupport); printk(BIOS_INFO, "CommonModeRestoreTime = 0x%x\n", comm_mode_rst_time); printk(BIOS_INFO, "Power On Value = 0x%x, Power On Scale = 0x%x\n", endp_power_on_value, power_on_scale); pci_update_config32(root, root_cap + 0x08, ~0xff00, (comm_mode_rst_time << 8)); pci_update_config32(root, root_cap + 0x0c, 0xffffff04, (endp_power_on_value << 3) | (power_on_scale)); /* TODO: 0xa0, 2 are values that work on some chipsets but really * should be determined dynamically by looking at downstream devices. */ pci_update_config32(root, root_cap + 0x08, ~(ASPM_LTR_L12_THRESHOLD_VALUE_MASK | ASPM_LTR_L12_THRESHOLD_SCALE_MASK), (0xa0 << ASPM_LTR_L12_THRESHOLD_VALUE_OFFSET) | (2 << ASPM_LTR_L12_THRESHOLD_SCALE_OFFSET)); pci_update_config32(root, root_cap + 0x08, ~0x1f, L1SubStateSupport); for (dev_t = dev; dev_t; dev_t = dev_t->sibling) { pci_update_config32(dev_t, end_cap + 0x0c, 0xffffff04, (endp_power_on_value << 3) | (power_on_scale)); pci_update_config32(dev_t, end_cap + 0x08, ~(ASPM_LTR_L12_THRESHOLD_VALUE_MASK | ASPM_LTR_L12_THRESHOLD_SCALE_MASK), (0xa0 << ASPM_LTR_L12_THRESHOLD_VALUE_OFFSET) | (2 << ASPM_LTR_L12_THRESHOLD_SCALE_OFFSET)); pci_update_config32(dev_t, end_cap + 0x08, ~0x1f, L1SubStateSupport); } } static void pciexp_config_L1_sub_state(struct device *root, struct device *dev) { unsigned int root_cap, end_cap; /* Do it for function 0 only */ if (dev->path.pci.devfn & 0x7) return; root_cap = pciexp_find_extended_cap(root, PCIE_EXT_CAP_L1SS_ID, 0); if (!root_cap) return; end_cap = pciexp_find_extended_cap(dev, PCIE_EXT_CAP_L1SS_ID, 0); if (!end_cap) { end_cap = pciexp_find_extended_cap(dev, 0xcafe, 0); if (!end_cap) return; } pciexp_L1_substate_commit(root, dev, root_cap, end_cap); } /* * Determine the ASPM L0s or L1 exit latency for a link * by checking both root port and endpoint and returning * the highest latency value. */ static int pciexp_aspm_latency(struct device *root, unsigned int root_cap, struct device *endp, unsigned int endp_cap, enum aspm_type type) { int root_lat = 0, endp_lat = 0; u32 root_lnkcap, endp_lnkcap; root_lnkcap = pci_read_config32(root, root_cap + PCI_EXP_LNKCAP); endp_lnkcap = pci_read_config32(endp, endp_cap + PCI_EXP_LNKCAP); /* Make sure the link supports this ASPM type by checking * capability bits 11:10 with aspm_type offset by 1 */ if (!(root_lnkcap & (1 << (type + 9))) || !(endp_lnkcap & (1 << (type + 9)))) return -1; /* Find the one with higher latency */ switch (type) { case PCIE_ASPM_L0S: root_lat = (root_lnkcap & PCI_EXP_LNKCAP_L0SEL) >> 12; endp_lat = (endp_lnkcap & PCI_EXP_LNKCAP_L0SEL) >> 12; break; case PCIE_ASPM_L1: root_lat = (root_lnkcap & PCI_EXP_LNKCAP_L1EL) >> 15; endp_lat = (endp_lnkcap & PCI_EXP_LNKCAP_L1EL) >> 15; break; default: return -1; } return (endp_lat > root_lat) ? endp_lat : root_lat; } /* * Enable ASPM on PCIe root port and endpoint. */ static void pciexp_enable_aspm(struct device *root, unsigned int root_cap, struct device *endp, unsigned int endp_cap) { const char *aspm_type_str[] = { "None", "L0s", "L1", "L0s and L1" }; enum aspm_type apmc = PCIE_ASPM_NONE; int exit_latency, ok_latency; u16 lnkctl; u32 devcap; if (endp->disable_pcie_aspm) return; /* Get endpoint device capabilities for acceptable limits */ devcap = pci_read_config32(endp, endp_cap + PCI_EXP_DEVCAP); /* Enable L0s if it is within endpoint acceptable limit */ ok_latency = (devcap & PCI_EXP_DEVCAP_L0S) >> 6; exit_latency = pciexp_aspm_latency(root, root_cap, endp, endp_cap, PCIE_ASPM_L0S); if (exit_latency >= 0 && exit_latency <= ok_latency) apmc |= PCIE_ASPM_L0S; /* Enable L1 if it is within endpoint acceptable limit */ ok_latency = (devcap & PCI_EXP_DEVCAP_L1) >> 9; exit_latency = pciexp_aspm_latency(root, root_cap, endp, endp_cap, PCIE_ASPM_L1); if (exit_latency >= 0 && exit_latency <= ok_latency) apmc |= PCIE_ASPM_L1; if (apmc != PCIE_ASPM_NONE) { /* Set APMC in root port first */ lnkctl = pci_read_config16(root, root_cap + PCI_EXP_LNKCTL); lnkctl |= apmc; pci_write_config16(root, root_cap + PCI_EXP_LNKCTL, lnkctl); /* Set APMC in endpoint device next */ lnkctl = pci_read_config16(endp, endp_cap + PCI_EXP_LNKCTL); lnkctl |= apmc; pci_write_config16(endp, endp_cap + PCI_EXP_LNKCTL, lnkctl); } printk(BIOS_INFO, "ASPM: Enabled %s\n", aspm_type_str[apmc]); } /* * Set max payload size of endpoint in accordance with max payload size of root port. */ static void pciexp_set_max_payload_size(struct device *root, unsigned int root_cap, struct device *endp, unsigned int endp_cap) { unsigned int endp_max_payload, root_max_payload, max_payload; u16 endp_devctl, root_devctl; u32 endp_devcap, root_devcap; /* Get max payload size supported by endpoint */ endp_devcap = pci_read_config32(endp, endp_cap + PCI_EXP_DEVCAP); endp_max_payload = endp_devcap & PCI_EXP_DEVCAP_PAYLOAD; /* Get max payload size supported by root port */ root_devcap = pci_read_config32(root, root_cap + PCI_EXP_DEVCAP); root_max_payload = root_devcap & PCI_EXP_DEVCAP_PAYLOAD; /* Set max payload to smaller of the reported device capability. */ max_payload = MIN(endp_max_payload, root_max_payload); if (max_payload > 5) { /* Values 6 and 7 are reserved in PCIe 3.0 specs. */ printk(BIOS_ERR, "PCIe: Max_Payload_Size field restricted from %d to 5\n", max_payload); max_payload = 5; } endp_devctl = pci_read_config16(endp, endp_cap + PCI_EXP_DEVCTL); endp_devctl &= ~PCI_EXP_DEVCTL_PAYLOAD; endp_devctl |= max_payload << 5; pci_write_config16(endp, endp_cap + PCI_EXP_DEVCTL, endp_devctl); root_devctl = pci_read_config16(root, root_cap + PCI_EXP_DEVCTL); root_devctl &= ~PCI_EXP_DEVCTL_PAYLOAD; root_devctl |= max_payload << 5; pci_write_config16(root, root_cap + PCI_EXP_DEVCTL, root_devctl); printk(BIOS_INFO, "PCIe: Max_Payload_Size adjusted to %d\n", (1 << (max_payload + 7))); } static void pciexp_tune_dev(struct device *dev) { struct device *root = dev->bus->dev; unsigned int root_cap, cap; cap = pci_find_capability(dev, PCI_CAP_ID_PCIE); if (!cap) return; root_cap = pci_find_capability(root, PCI_CAP_ID_PCIE); if (!root_cap) return; /* Check for and enable Common Clock */ if (CONFIG(PCIEXP_COMMON_CLOCK)) pciexp_enable_common_clock(root, root_cap, dev, cap); /* Check if per port CLK req is supported by endpoint*/ if (CONFIG(PCIEXP_CLK_PM)) pciexp_enable_clock_power_pm(dev, cap); /* Enable L1 Sub-State when both root port and endpoint support */ if (CONFIG(PCIEXP_L1_SUB_STATE)) pciexp_config_L1_sub_state(root, dev); /* Check for and enable ASPM */ if (CONFIG(PCIEXP_ASPM)) pciexp_enable_aspm(root, root_cap, dev, cap); /* Adjust Max_Payload_Size of link ends. */ pciexp_set_max_payload_size(root, root_cap, dev, cap); pciexp_configure_ltr(root, root_cap, dev, cap); } void pciexp_scan_bus(struct bus *bus, unsigned int min_devfn, unsigned int max_devfn) { struct device *child; pciexp_enable_ltr(bus->dev); pci_scan_bus(bus, min_devfn, max_devfn); for (child = bus->children; child; child = child->sibling) { if (child->path.type != DEVICE_PATH_PCI) continue; if ((child->path.pci.devfn < min_devfn) || (child->path.pci.devfn > max_devfn)) { continue; } pciexp_tune_dev(child); } } void pciexp_scan_bridge(struct device *dev) { do_pci_scan_bridge(dev, pciexp_scan_bus); } /** Default device operations for PCI Express bridges */ static struct pci_operations pciexp_bus_ops_pci = { .set_subsystem = 0, }; struct device_operations default_pciexp_ops_bus = { .read_resources = pci_bus_read_resources, .set_resources = pci_dev_set_resources, .enable_resources = pci_bus_enable_resources, .scan_bus = pciexp_scan_bridge, .reset_bus = pci_bus_reset, .ops_pci = &pciexp_bus_ops_pci, }; static void pciexp_hotplug_dummy_read_resources(struct device *dev) { struct resource *resource; /* Add extra memory space */ resource = new_resource(dev, 0x10); resource->size = CONFIG_PCIEXP_HOTPLUG_MEM; resource->align = 12; resource->gran = 12; resource->limit = 0xffffffff; resource->flags |= IORESOURCE_MEM; /* Add extra prefetchable memory space */ resource = new_resource(dev, 0x14); resource->size = CONFIG_PCIEXP_HOTPLUG_PREFETCH_MEM; resource->align = 12; resource->gran = 12; resource->limit = 0xffffffffffffffff; resource->flags |= IORESOURCE_MEM | IORESOURCE_PREFETCH; /* Set resource flag requesting allocation above 4G boundary. */ if (CONFIG(PCIEXP_HOTPLUG_PREFETCH_MEM_ABOVE_4G)) resource->flags |= IORESOURCE_ABOVE_4G; /* Add extra I/O space */ resource = new_resource(dev, 0x18); resource->size = CONFIG_PCIEXP_HOTPLUG_IO; resource->align = 12; resource->gran = 12; resource->limit = 0xffff; resource->flags |= IORESOURCE_IO; } static struct device_operations pciexp_hotplug_dummy_ops = { .read_resources = pciexp_hotplug_dummy_read_resources, .set_resources = noop_set_resources, }; void pciexp_hotplug_scan_bridge(struct device *dev) { dev->hotplug_buses = CONFIG_PCIEXP_HOTPLUG_BUSES; /* Normal PCIe Scan */ pciexp_scan_bridge(dev); /* Add dummy slot to preserve resources, must happen after bus scan */ struct device *dummy; struct device_path dummy_path = { .type = DEVICE_PATH_NONE }; dummy = alloc_dev(dev->link_list, &dummy_path); dummy->ops = &pciexp_hotplug_dummy_ops; } struct device_operations default_pciexp_hotplug_ops_bus = { .read_resources = pci_bus_read_resources, .set_resources = pci_dev_set_resources, .enable_resources = pci_bus_enable_resources, .scan_bus = pciexp_hotplug_scan_bridge, .reset_bus = pci_bus_reset, .ops_pci = &pciexp_bus_ops_pci, };