soc/intel/xeon_sp/cpx: add chip operation and PCIe enumeration

Add PCIe enumeration and resource assignment/allocation.

Xeon-SP processor family has split IIO design, where PCIe domain 0 is
split into multiple stacks. Each stack has its own resource ranges (eg.
IO resource, mem32 resource, mem64 resource). The stack itself is not
PCIe device, it does not have config space to be probed/programmed.

The stack is programmed by FSP. coreboot needs to take into account of
stack when doing PCIe enumeration and resource allocation.

Current coreboot PCIe resource allocator does not support the concept of
split IIO stack, thus entire support is done locally in this patch.

In near future, improvements will be done, first generalize for xeon-sp,
then generalize for coreboot PCIe device code.

Signed-off-by: Jonathan Zhang <jonzhang@fb.com>
Signed-off-by: Reddy Chagam <anjaneya.chagam@intel.com>
Change-Id: If461b1dc1f313d98b676dc9e91d08a1dbb9cb388
Reviewed-on: https://review.coreboot.org/c/coreboot/+/40110
Tested-by: build bot (Jenkins) <no-reply@coreboot.org>
Reviewed-by: Aaron Durbin <adurbin@chromium.org>
Reviewed-by: Philipp Deppenwiese <zaolin.daisuki@gmail.com>

1 files changed, 523 insertions, 9 deletions

diff --git a/src/soc/intel/xeon_sp/cpx/chip.c b/src/soc/intel/xeon_sp/cpx/chip.c
index c19c0cf93c..19ebe4786d 100644
--- a/src/soc/intel/xeon_sp/cpx/chip.c
+++ b/src/soc/intel/xeon_sp/cpx/chip.c
@@ -1,18 +1,492 @@
 /* SPDX-License-Identifier: GPL-2.0-only */
 
 #include <arch/ioapic.h>
+#include <assert.h>
 #include <console/console.h>
 #include <cpu/x86/lapic.h>
 #include <device/pci.h>
 #include <fsp/api.h>
 #include <intelblocks/p2sb.h>
+#include <post.h>
 #include <soc/cpu.h>
 #include <soc/ramstage.h>
 #include <soc/pm.h>
+#include <soc/soc_util.h>
+#include <stdlib.h>
 
 /* C620 IOAPIC has 120 redirection entries */
 #define C620_IOAPIC_REDIR_ENTRIES		120
 
+struct pci_resource {
+	struct device        *dev;
+	struct resource      *res;
+	struct pci_resource  *next;
+};
+
+struct stack_dev_resource {
+	uint8_t                   align;
+	struct pci_resource       *children;
+	struct stack_dev_resource *next;
+};
+
+typedef enum {
+	RES_TYPE_IO = 0,
+	RES_TYPE_NONPREF_MEM,
+	RES_TYPE_PREF_MEM,
+	MAX_RES_TYPES
+} ResType;
+
+static ResType get_res_type(uint64_t flags)
+{
+	if (flags & IORESOURCE_IO)
+		return RES_TYPE_IO;
+	if (flags & IORESOURCE_MEM) {
+		if (flags & IORESOURCE_PREFETCH) {
+			printk(BIOS_DEBUG, "%s:%d flags: 0x%llx\n", __func__, __LINE__, flags);
+			return RES_TYPE_PREF_MEM;
+		}
+		/* both 64-bit and 32-bit use below 4GB address space */
+		return RES_TYPE_NONPREF_MEM;
+	}
+	printk(BIOS_ERR, "Invalid resource type 0x%llx\n", flags);
+	die("Invalida resource type");
+}
+
+static bool need_assignment(uint64_t flags)
+{
+	if (flags & (IORESOURCE_STORED | IORESOURCE_RESERVE | IORESOURCE_FIXED |
+		IORESOURCE_ASSIGNED))
+		return false;
+	else
+		return true;
+}
+
+static uint64_t get_resource_base(STACK_RES *stack, ResType res_type)
+{
+	if (res_type == RES_TYPE_IO) {
+		assert(stack->PciResourceIoBase <= stack->PciResourceIoLimit);
+		return stack->PciResourceIoBase;
+	}
+	if (res_type == RES_TYPE_NONPREF_MEM) {
+		assert(stack->PciResourceMem32Base <= stack->PciResourceMem32Limit);
+		return stack->PciResourceMem32Base;
+	}
+	assert(stack->PciResourceMem64Base <= stack->PciResourceMem64Limit);
+	return stack->PciResourceMem64Base;
+}
+
+static void set_resource_base(STACK_RES *stack, ResType res_type, uint64_t base)
+{
+	if (res_type == RES_TYPE_IO) {
+		assert(base <= (stack->PciResourceIoLimit + 1));
+		stack->PciResourceIoBase = base;
+	} else if (res_type == RES_TYPE_NONPREF_MEM) {
+		assert(base <= (stack->PciResourceMem32Limit + 1));
+		stack->PciResourceMem32Base = base;
+	} else {
+		assert(base <= (stack->PciResourceMem64Limit + 1));
+		stack->PciResourceMem64Base = base;
+	}
+}
+
+static void assign_stack_resources(struct iiostack_resource *stack_list,
+	struct device *dev, struct resource *bridge);
+
+static void xeonsp_cpx_pci_domain_scan_bus(struct device *dev)
+{
+	DEV_FUNC_ENTER(dev);
+	struct bus *link = dev->link_list;
+
+	printk(BIOS_SPEW, "%s:%s scanning buses under device %s\n",
+		__FILE__, __func__, dev_path(dev));
+	while (link != NULL) {
+		if (link->secondary == 0)  { // scan only PSTACK buses
+			struct device *d;
+			for (d = link->children; d; d = d->sibling)
+				pci_probe_dev(d, link, d->path.pci.devfn);
+			scan_bridges(link);
+		} else {
+			pci_scan_bus(link, PCI_DEVFN(0, 0), 0xff);
+		}
+		link = link->next;
+	}
+	DEV_FUNC_EXIT(dev);
+}
+
+static void xeonsp_pci_dev_iterator(struct bus *bus,
+		void (*dev_iterator)(struct device *, void *),
+		void (*res_iterator)(struct device *, struct resource *, void *),
+		void *data)
+{
+	struct device *curdev;
+	struct resource *res;
+
+	/* Walk through all devices and find which resources they need. */
+	for (curdev = bus->children; curdev; curdev = curdev->sibling) {
+		struct bus *link;
+
+		if (!curdev->enabled)
+			continue;
+
+		if (!curdev->ops || !curdev->ops->read_resources) {
+			if (curdev->path.type != DEVICE_PATH_APIC)
+				printk(BIOS_ERR, "%s missing read_resources\n",
+					dev_path(curdev));
+			continue;
+		}
+
+		if (dev_iterator)
+			dev_iterator(curdev, data);
+
+		if (res_iterator) {
+			for (res = curdev->resource_list; res; res = res->next)
+				res_iterator(curdev, res, data);
+		}
+
+		/* Read in the resources behind the current device's links. */
+		for (link = curdev->link_list; link; link = link->next)
+			xeonsp_pci_dev_iterator(link, dev_iterator, res_iterator, data);
+	}
+}
+
+static void xeonsp_pci_dev_read_resources(struct device *dev, void *data)
+{
+	post_log_path(dev);
+	dev->ops->read_resources(dev);
+}
+
+static void xeonsp_pci_dev_dummy_func(struct device *dev)
+{
+}
+
+static void xeonsp_reset_pci_op(struct device *dev, void *data)
+{
+	if (dev->ops)
+		dev->ops->read_resources = xeonsp_pci_dev_dummy_func;
+}
+
+static STACK_RES *find_stack_for_bus(struct iiostack_resource *info, uint8_t bus)
+{
+	for (int i = 0; i < info->no_of_stacks; ++i) {
+		if (bus >= info->res[i].BusBase && bus <= info->res[i].BusLimit)
+			return &info->res[i];
+	}
+	return NULL;
+}
+
+static void add_res_to_stack(struct stack_dev_resource **root,
+	struct device *dev, struct resource *res)
+{
+	struct stack_dev_resource *cur = *root;
+	while (cur) {
+		if (cur->align == res->align || cur->next == NULL) /* equal or last record */
+			break;
+		else if (cur->align > res->align) {
+			if (cur->next->align < res->align) /* need to insert new record here */
+				break;
+			cur = cur->next;
+		} else {
+			break;
+		}
+	}
+
+	struct stack_dev_resource *nr;
+	if (!cur || cur->align != res->align) { /* need to add new record */
+		nr = malloc(sizeof(struct stack_dev_resource));
+		if (nr == 0)
+			die("assign_resource_to_stack(): out of memory.\n");
+		memset(nr, 0, sizeof(struct stack_dev_resource));
+		nr->align = res->align;
+		if (!cur) {
+			*root = nr; /* head node */
+		} else if (cur->align > nr->align) {
+			if (cur->next == NULL) {
+				cur->next = nr;
+			} else {
+				nr->next = cur->next;
+				cur->next = nr;
+			}
+		} else { /* insert in the beginning */
+			nr->next = cur;
+			*root = nr;
+		}
+	} else {
+		nr = cur;
+	}
+
+	assert(nr != NULL && nr->align == res->align);
+
+	struct pci_resource *npr = malloc(sizeof(struct pci_resource));
+	if (npr == NULL)
+		die("%s: out of memory.\n", __func__);
+	npr->res = res;
+	npr->dev = dev;
+	npr->next = NULL;
+
+	if (nr->children == NULL) {
+		nr->children = npr;
+	} else {
+		struct pci_resource *pr = nr->children;
+		while (pr->next != NULL)
+			pr = pr->next;
+		pr->next = npr;
+	}
+}
+
+static void reserve_dev_resources(STACK_RES *stack, ResType res_type,
+	struct stack_dev_resource *res_root, struct resource *bridge)
+{
+	uint8_t align;
+	uint64_t orig_base, base;
+
+	orig_base = get_resource_base(stack, res_type);
+
+	align = 0;
+	base = orig_base;
+	int first = 1;
+	while (res_root) { /* loop through all devices grouped by alignment requirements */
+		struct pci_resource *pr = res_root->children;
+		while (pr) {
+			if (first) {
+				if (bridge) { /* takes highest alignment */
+					if (bridge->align < pr->res->align)
+						bridge->align = pr->res->align;
+					orig_base = ALIGN_UP(orig_base, 1 << bridge->align);
+				} else {
+					orig_base = ALIGN_UP(orig_base, 1 << pr->res->align);
+				}
+				base = orig_base;
+
+				if (bridge)
+					bridge->base = base;
+				pr->res->base = base;
+				first = 0;
+			} else {
+				pr->res->base = ALIGN_UP(base, 1 << pr->res->align);
+			}
+			pr->res->limit = pr->res->base + pr->res->size - 1;
+			base = pr->res->limit + 1;
+			pr->res->flags |= (IORESOURCE_ASSIGNED);
+			pr = pr->next;
+		}
+		res_root = res_root->next;
+	}
+
+	if (bridge) {
+		/* this bridge doesn't have any resources, will set it to default window */
+		if (first) {
+			orig_base = ALIGN_UP(orig_base, 1 << bridge->align);
+			bridge->base = orig_base;
+			base = orig_base + (1ULL << bridge->gran);
+		}
+
+		bridge->size = ALIGN_UP(base, 1 << bridge->align) - bridge->base;
+
+		bridge->limit = bridge->base + bridge->size - 1;
+		bridge->flags |= (IORESOURCE_ASSIGNED);
+		base = bridge->limit + 1;
+	}
+
+	set_resource_base(stack, res_type, base);
+}
+
+static void reclaim_resource_mem(struct stack_dev_resource *res_root)
+{
+	while (res_root) { /* loop through all devices grouped by alignment requirements */
+		/* free pci_resource */
+		struct pci_resource *pr = res_root->children;
+		while (pr) {
+			struct pci_resource *dpr = pr;
+			pr = pr->next;
+			free(dpr);
+		}
+
+		/* free stack_dev_resource */
+		struct stack_dev_resource *ddr = res_root;
+		res_root = res_root->next;
+		free(ddr);
+	}
+}
+
+static void assign_bridge_resources(struct iiostack_resource *stack_list,
+	struct device *dev, struct resource *bridge)
+{
+	struct resource *res;
+	if (!dev->enabled)
+		return;
+
+	for (res = dev->resource_list; res; res = res->next) {
+		if (!(res->flags & IORESOURCE_BRIDGE) ||
+		    (bridge && (get_res_type(bridge->flags) != get_res_type(res->flags))))
+			continue;
+
+		assign_stack_resources(stack_list, dev, res);
+
+		if (!bridge)
+			continue;
+
+		/* for 1st time update, overlading IORESOURCE_ASSIGNED */
+		if (!(bridge->flags & IORESOURCE_ASSIGNED)) {
+			bridge->base = res->base;
+			bridge->limit = res->limit;
+			bridge->flags |= (IORESOURCE_ASSIGNED);
+		} else {
+			/* update bridge range from child bridge range */
+			if (res->base < bridge->base)
+				bridge->base = res->base;
+			if (res->limit > bridge->limit)
+				bridge->limit = res->limit;
+		}
+		bridge->size = (bridge->limit - bridge->base + 1);
+	}
+}
+
+static void assign_stack_resources(struct iiostack_resource *stack_list,
+	struct device *dev, struct resource *bridge)
+{
+	struct bus *bus;
+
+	/* Read in the resources behind the current device's links. */
+	for (bus = dev->link_list; bus; bus = bus->next) {
+		struct device *curdev;
+		STACK_RES *stack;
+
+		/* get IIO stack for this bus */
+		stack = find_stack_for_bus(stack_list, bus->secondary);
+		assert(stack != NULL);
+
+		/* Assign resources to bridge */
+		for (curdev = bus->children; curdev; curdev = curdev->sibling)
+			assign_bridge_resources(stack_list, curdev, bridge);
+
+		/* Pick non-bridged resources for resource allocation for each resource type */
+		ResType res_types[MAX_RES_TYPES] = {
+			RES_TYPE_IO,
+			RES_TYPE_NONPREF_MEM,
+			RES_TYPE_PREF_MEM
+		};
+
+		uint8_t no_res_types = MAX_RES_TYPES;
+
+		/* if it is a bridge, only process matching brigge resource type */
+		if (bridge) {
+			res_types[0] = get_res_type(bridge->flags);
+			no_res_types = 1;
+		}
+
+		printk(BIOS_DEBUG, "%s:%d no_res_types: %d\n", __func__, __LINE__,
+			no_res_types);
+
+		/* Process each resource type */
+		for (int rt = 0; rt < no_res_types; ++rt) {
+			struct stack_dev_resource *res_root = NULL;
+			printk(BIOS_DEBUG, "%s:%d rt: %d\n", __func__, __LINE__, rt);
+			for (curdev = bus->children; curdev; curdev = curdev->sibling) {
+				struct resource *res;
+				printk(BIOS_DEBUG, "%s:%d dev: %s\n",
+					__func__, __LINE__, dev_path(curdev));
+				if (!curdev->enabled)
+					continue;
+
+				for (res = curdev->resource_list; res; res = res->next) {
+					printk(BIOS_DEBUG, "%s:%d dev: %s, flags: 0x%lx\n",
+						__func__, __LINE__,
+						dev_path(curdev), res->flags);
+					if (res->size == 0                            ||
+					    get_res_type(res->flags) != res_types[rt] ||
+					    (res->flags & IORESOURCE_BRIDGE)          ||
+					    !need_assignment(res->flags))
+						continue;
+					else
+						add_res_to_stack(&res_root, curdev, res);
+				}
+			}
+
+			/* Allocate resources and update bridge range */
+			if (res_root || (bridge && !(bridge->flags & IORESOURCE_ASSIGNED))) {
+				reserve_dev_resources(stack, res_types[rt], res_root, bridge);
+				reclaim_resource_mem(res_root);
+			}
+		}
+	}
+}
+
+static void xeonsp_pci_domain_read_resources(struct device *dev)
+{
+	struct bus *link;
+
+	DEV_FUNC_ENTER(dev);
+
+	pci_domain_read_resources(dev);
+
+	/*
+	 * Walk through all devices in this domain and read resources.
+	 * Since there is no callback when read resource operation is
+	 * complete for all devices, domain read resource function initiates
+	 * read resources for all devices and swaps read resource operation
+	 * with dummy function to avoid warning.
+	 */
+	for (link = dev->link_list; link; link = link->next)
+		xeonsp_pci_dev_iterator(link, xeonsp_pci_dev_read_resources, NULL, NULL);
+
+	for (link = dev->link_list; link; link = link->next)
+		xeonsp_pci_dev_iterator(link, xeonsp_reset_pci_op, NULL, NULL);
+
+	struct iiostack_resource stack_info = {0};
+	uint8_t pci64bit_alloc_flag = get_iiostack_info(&stack_info);
+	if (!pci64bit_alloc_flag) {
+		/*
+		 * Split 32 bit address space between prefetchable and
+		 * non-prefetchable windows
+		 */
+		for (int s = 0; s < stack_info.no_of_stacks; ++s) {
+			STACK_RES *res = &stack_info.res[s];
+			uint64_t length = (res->PciResourceMem32Limit -
+				res->PciResourceMem32Base + 1)/2;
+			res->PciResourceMem64Limit = res->PciResourceMem32Limit;
+			res->PciResourceMem32Limit = (res->PciResourceMem32Base + length - 1);
+			res->PciResourceMem64Base = res->PciResourceMem32Limit + 1;
+		}
+	}
+
+
+	/* assign resources */
+	assign_stack_resources(&stack_info, dev, NULL);
+
+	DEV_FUNC_EXIT(dev);
+}
+
+static void reset_resource_to_unassigned(struct device *dev, struct resource *res, void *data)
+{
+	if ((res->flags & (IORESOURCE_IO | IORESOURCE_MEM)) &&
+		!(res->flags & (IORESOURCE_FIXED | IORESOURCE_RESERVE))) {
+		res->flags &= ~IORESOURCE_ASSIGNED;
+	}
+}
+
+static void xeonsp_cpx_pci_domain_set_resources(struct device *dev)
+{
+	DEV_FUNC_ENTER(dev);
+
+	print_resource_tree(dev, BIOS_SPEW, "Before xeonsp pci domain set resource");
+
+	/* reset bus 0 dev resource assignment - need to change them to FSP IIOStack window */
+	xeonsp_pci_dev_iterator(dev->link_list, NULL, reset_resource_to_unassigned, NULL);
+
+	/* update dev resources based on IIOStack IO/Mem32/Mem64 windows */
+	xeonsp_pci_domain_read_resources(dev);
+
+	struct bus *link = dev->link_list;
+	while (link != NULL) {
+		assign_resources(link);
+		link = link->next;
+	}
+
+	print_resource_tree(dev, BIOS_SPEW, "After xeonsp pci domain set resource");
+
+	DEV_FUNC_EXIT(dev);
+}
+
 void platform_fsp_silicon_init_params_cb(FSPS_UPD *silupd)
 {
 	/* not implemented yet */
@@ -20,8 +494,8 @@ void platform_fsp_silicon_init_params_cb(FSPS_UPD *silupd)
 
 static struct device_operations pci_domain_ops = {
 	.read_resources = &pci_domain_read_resources,
-	.set_resources = &pci_domain_set_resources,
-	.scan_bus = &pci_domain_scan_bus,
+	.set_resources = &xeonsp_cpx_pci_domain_set_resources,
+	.scan_bus = &xeonsp_cpx_pci_domain_scan_bus,
 };
 
 static struct device_operations cpu_bus_ops = {
@@ -30,14 +504,43 @@ static struct device_operations cpu_bus_ops = {
 	.init = cpx_init_cpus,
 };
 
-static void chip_enable_dev(struct device *dev)
+/* Attach IIO stack bus numbers with dummy device to PCI DOMAIN 0000 device */
+static void attach_iio_stacks(struct device *dev)
 {
-	/* Set the operations if it is a special bus type */
-	if (dev->path.type == DEVICE_PATH_DOMAIN) {
-		dev->ops = &pci_domain_ops;
-	} else if (dev->path.type == DEVICE_PATH_CPU_CLUSTER) {
-		dev->ops = &cpu_bus_ops;
+	struct bus *iiostack_bus;
+	struct iiostack_resource stack_info = {0};
+
+	DEV_FUNC_ENTER(dev);
+
+	get_iiostack_info(&stack_info);
+	for (int s = 0; s < stack_info.no_of_stacks; ++s) {
+		/* only non zero bus no. needs to be enumerated */
+		if (stack_info.res[s].BusBase == 0)
+			continue;
+
+		iiostack_bus = malloc(sizeof(struct bus));
+		if (iiostack_bus == NULL)
+			die("%s: out of memory.\n", __func__);
+		memset(iiostack_bus, 0, sizeof(*iiostack_bus));
+		memcpy(iiostack_bus, dev->bus, sizeof(*iiostack_bus));
+		iiostack_bus->secondary = stack_info.res[s].BusBase;
+		iiostack_bus->subordinate = stack_info.res[s].BusBase;
+		iiostack_bus->dev = NULL;
+		iiostack_bus->children = NULL;
+		iiostack_bus->next = NULL;
+		iiostack_bus->link_num = 1;
+
+		if (dev->link_list == NULL) {
+			dev->link_list = iiostack_bus;
+		} else {
+			struct bus *nlink = dev->link_list;
+			while (nlink->next != NULL)
+				nlink = nlink->next;
+			nlink->next = iiostack_bus;
+		}
 	}
+
+	DEV_FUNC_EXIT(dev);
 }
 
 static void pch_enable_ioapic(const struct device *dev)
@@ -65,6 +568,17 @@ struct pci_operations soc_pci_ops = {
 	.set_subsystem = pci_dev_set_subsystem,
 };
 
+static void chip_enable_dev(struct device *dev)
+{
+	/* Set the operations if it is a special bus type */
+	if (dev->path.type == DEVICE_PATH_DOMAIN) {
+		dev->ops = &pci_domain_ops;
+		attach_iio_stacks(dev);
+	} else if (dev->path.type == DEVICE_PATH_CPU_CLUSTER) {
+		dev->ops = &cpu_bus_ops;
+	}
+}
+
 static void chip_final(void *data)
 {
 	p2sb_hide();
@@ -83,5 +597,5 @@ struct chip_operations soc_intel_xeon_sp_cpx_ops = {
 	CHIP_NAME("Intel Cooperlake-SP")
 	.enable_dev = chip_enable_dev,
 	.init = chip_init,
-	.final = chip_final
+	.final = chip_final,
 };