efi: Rename arm-init to efi-init common for all arch
authorAtish Patra <atish.patra@wdc.com>
Wed, 19 Aug 2020 22:24:23 +0000 (15:24 -0700)
committerArd Biesheuvel <ardb@kernel.org>
Fri, 11 Sep 2020 06:31:07 +0000 (09:31 +0300)
arm-init is responsible for setting up efi runtime and doesn't actually
do any ARM specific stuff. RISC-V can use the same source code as it is.

Rename it to efi-init so that RISC-V can use it.

Signed-off-by: Atish Patra <atish.patra@wdc.com>
Link: https://lore.kernel.org/r/20200819222425.30721-8-atish.patra@wdc.com
Signed-off-by: Ard Biesheuvel <ardb@kernel.org>
drivers/firmware/efi/Makefile
drivers/firmware/efi/arm-init.c [deleted file]
drivers/firmware/efi/efi-init.c [new file with mode: 0644]

index 7a21698..61fd1e8 100644 (file)
@@ -32,7 +32,7 @@ obj-$(CONFIG_EFI_EMBEDDED_FIRMWARE)   += embedded-firmware.o
 fake_map-y                             += fake_mem.o
 fake_map-$(CONFIG_X86)                 += x86_fake_mem.o
 
-arm-obj-$(CONFIG_EFI)                  := arm-init.o arm-runtime.o
+arm-obj-$(CONFIG_EFI)                  := efi-init.o arm-runtime.o
 obj-$(CONFIG_ARM)                      += $(arm-obj-y)
 obj-$(CONFIG_ARM64)                    += $(arm-obj-y)
 obj-$(CONFIG_EFI_CAPSULE_LOADER)       += capsule-loader.o
diff --git a/drivers/firmware/efi/arm-init.c b/drivers/firmware/efi/arm-init.c
deleted file mode 100644 (file)
index 71c445d..0000000
+++ /dev/null
@@ -1,386 +0,0 @@
-// SPDX-License-Identifier: GPL-2.0
-/*
- * Extensible Firmware Interface
- *
- * Based on Extensible Firmware Interface Specification version 2.4
- *
- * Copyright (C) 2013 - 2015 Linaro Ltd.
- */
-
-#define pr_fmt(fmt)    "efi: " fmt
-
-#include <linux/efi.h>
-#include <linux/fwnode.h>
-#include <linux/init.h>
-#include <linux/memblock.h>
-#include <linux/mm_types.h>
-#include <linux/of.h>
-#include <linux/of_address.h>
-#include <linux/of_fdt.h>
-#include <linux/platform_device.h>
-#include <linux/screen_info.h>
-
-#include <asm/efi.h>
-
-static int __init is_memory(efi_memory_desc_t *md)
-{
-       if (md->attribute & (EFI_MEMORY_WB|EFI_MEMORY_WT|EFI_MEMORY_WC))
-               return 1;
-       return 0;
-}
-
-/*
- * Translate a EFI virtual address into a physical address: this is necessary,
- * as some data members of the EFI system table are virtually remapped after
- * SetVirtualAddressMap() has been called.
- */
-static phys_addr_t __init efi_to_phys(unsigned long addr)
-{
-       efi_memory_desc_t *md;
-
-       for_each_efi_memory_desc(md) {
-               if (!(md->attribute & EFI_MEMORY_RUNTIME))
-                       continue;
-               if (md->virt_addr == 0)
-                       /* no virtual mapping has been installed by the stub */
-                       break;
-               if (md->virt_addr <= addr &&
-                   (addr - md->virt_addr) < (md->num_pages << EFI_PAGE_SHIFT))
-                       return md->phys_addr + addr - md->virt_addr;
-       }
-       return addr;
-}
-
-static __initdata unsigned long screen_info_table = EFI_INVALID_TABLE_ADDR;
-static __initdata unsigned long cpu_state_table = EFI_INVALID_TABLE_ADDR;
-
-static const efi_config_table_type_t arch_tables[] __initconst = {
-       {LINUX_EFI_ARM_SCREEN_INFO_TABLE_GUID, &screen_info_table},
-       {LINUX_EFI_ARM_CPU_STATE_TABLE_GUID, &cpu_state_table},
-       {}
-};
-
-static void __init init_screen_info(void)
-{
-       struct screen_info *si;
-
-       if (IS_ENABLED(CONFIG_ARM) &&
-           screen_info_table != EFI_INVALID_TABLE_ADDR) {
-               si = early_memremap_ro(screen_info_table, sizeof(*si));
-               if (!si) {
-                       pr_err("Could not map screen_info config table\n");
-                       return;
-               }
-               screen_info = *si;
-               early_memunmap(si, sizeof(*si));
-
-               /* dummycon on ARM needs non-zero values for columns/lines */
-               screen_info.orig_video_cols = 80;
-               screen_info.orig_video_lines = 25;
-       }
-
-       if (screen_info.orig_video_isVGA == VIDEO_TYPE_EFI &&
-           memblock_is_map_memory(screen_info.lfb_base))
-               memblock_mark_nomap(screen_info.lfb_base, screen_info.lfb_size);
-}
-
-static int __init uefi_init(u64 efi_system_table)
-{
-       efi_config_table_t *config_tables;
-       efi_system_table_t *systab;
-       size_t table_size;
-       int retval;
-
-       systab = early_memremap_ro(efi_system_table, sizeof(efi_system_table_t));
-       if (systab == NULL) {
-               pr_warn("Unable to map EFI system table.\n");
-               return -ENOMEM;
-       }
-
-       set_bit(EFI_BOOT, &efi.flags);
-       if (IS_ENABLED(CONFIG_64BIT))
-               set_bit(EFI_64BIT, &efi.flags);
-
-       retval = efi_systab_check_header(&systab->hdr, 2);
-       if (retval)
-               goto out;
-
-       efi.runtime = systab->runtime;
-       efi.runtime_version = systab->hdr.revision;
-
-       efi_systab_report_header(&systab->hdr, efi_to_phys(systab->fw_vendor));
-
-       table_size = sizeof(efi_config_table_t) * systab->nr_tables;
-       config_tables = early_memremap_ro(efi_to_phys(systab->tables),
-                                         table_size);
-       if (config_tables == NULL) {
-               pr_warn("Unable to map EFI config table array.\n");
-               retval = -ENOMEM;
-               goto out;
-       }
-       retval = efi_config_parse_tables(config_tables, systab->nr_tables,
-                                        IS_ENABLED(CONFIG_ARM) ? arch_tables
-                                                               : NULL);
-
-       early_memunmap(config_tables, table_size);
-out:
-       early_memunmap(systab, sizeof(efi_system_table_t));
-       return retval;
-}
-
-/*
- * Return true for regions that can be used as System RAM.
- */
-static __init int is_usable_memory(efi_memory_desc_t *md)
-{
-       switch (md->type) {
-       case EFI_LOADER_CODE:
-       case EFI_LOADER_DATA:
-       case EFI_ACPI_RECLAIM_MEMORY:
-       case EFI_BOOT_SERVICES_CODE:
-       case EFI_BOOT_SERVICES_DATA:
-       case EFI_CONVENTIONAL_MEMORY:
-       case EFI_PERSISTENT_MEMORY:
-               /*
-                * Special purpose memory is 'soft reserved', which means it
-                * is set aside initially, but can be hotplugged back in or
-                * be assigned to the dax driver after boot.
-                */
-               if (efi_soft_reserve_enabled() &&
-                   (md->attribute & EFI_MEMORY_SP))
-                       return false;
-
-               /*
-                * According to the spec, these regions are no longer reserved
-                * after calling ExitBootServices(). However, we can only use
-                * them as System RAM if they can be mapped writeback cacheable.
-                */
-               return (md->attribute & EFI_MEMORY_WB);
-       default:
-               break;
-       }
-       return false;
-}
-
-static __init void reserve_regions(void)
-{
-       efi_memory_desc_t *md;
-       u64 paddr, npages, size;
-
-       if (efi_enabled(EFI_DBG))
-               pr_info("Processing EFI memory map:\n");
-
-       /*
-        * Discard memblocks discovered so far: if there are any at this
-        * point, they originate from memory nodes in the DT, and UEFI
-        * uses its own memory map instead.
-        */
-       memblock_dump_all();
-       memblock_remove(0, PHYS_ADDR_MAX);
-
-       for_each_efi_memory_desc(md) {
-               paddr = md->phys_addr;
-               npages = md->num_pages;
-
-               if (efi_enabled(EFI_DBG)) {
-                       char buf[64];
-
-                       pr_info("  0x%012llx-0x%012llx %s\n",
-                               paddr, paddr + (npages << EFI_PAGE_SHIFT) - 1,
-                               efi_md_typeattr_format(buf, sizeof(buf), md));
-               }
-
-               memrange_efi_to_native(&paddr, &npages);
-               size = npages << PAGE_SHIFT;
-
-               if (is_memory(md)) {
-                       early_init_dt_add_memory_arch(paddr, size);
-
-                       if (!is_usable_memory(md))
-                               memblock_mark_nomap(paddr, size);
-
-                       /* keep ACPI reclaim memory intact for kexec etc. */
-                       if (md->type == EFI_ACPI_RECLAIM_MEMORY)
-                               memblock_reserve(paddr, size);
-               }
-       }
-}
-
-void __init efi_init(void)
-{
-       struct efi_memory_map_data data;
-       u64 efi_system_table;
-
-       /* Grab UEFI information placed in FDT by stub */
-       efi_system_table = efi_get_fdt_params(&data);
-       if (!efi_system_table)
-               return;
-
-       if (efi_memmap_init_early(&data) < 0) {
-               /*
-               * If we are booting via UEFI, the UEFI memory map is the only
-               * description of memory we have, so there is little point in
-               * proceeding if we cannot access it.
-               */
-               panic("Unable to map EFI memory map.\n");
-       }
-
-       WARN(efi.memmap.desc_version != 1,
-            "Unexpected EFI_MEMORY_DESCRIPTOR version %ld",
-             efi.memmap.desc_version);
-
-       if (uefi_init(efi_system_table) < 0) {
-               efi_memmap_unmap();
-               return;
-       }
-
-       reserve_regions();
-       efi_esrt_init();
-
-       memblock_reserve(data.phys_map & PAGE_MASK,
-                        PAGE_ALIGN(data.size + (data.phys_map & ~PAGE_MASK)));
-
-       init_screen_info();
-
-#ifdef CONFIG_ARM
-       /* ARM does not permit early mappings to persist across paging_init() */
-       efi_memmap_unmap();
-
-       if (cpu_state_table != EFI_INVALID_TABLE_ADDR) {
-               struct efi_arm_entry_state *state;
-               bool dump_state = true;
-
-               state = early_memremap_ro(cpu_state_table,
-                                         sizeof(struct efi_arm_entry_state));
-               if (state == NULL) {
-                       pr_warn("Unable to map CPU entry state table.\n");
-                       return;
-               }
-
-               if ((state->sctlr_before_ebs & 1) == 0)
-                       pr_warn(FW_BUG "EFI stub was entered with MMU and Dcache disabled, please fix your firmware!\n");
-               else if ((state->sctlr_after_ebs & 1) == 0)
-                       pr_warn(FW_BUG "ExitBootServices() returned with MMU and Dcache disabled, please fix your firmware!\n");
-               else
-                       dump_state = false;
-
-               if (dump_state || efi_enabled(EFI_DBG)) {
-                       pr_info("CPSR at EFI stub entry        : 0x%08x\n", state->cpsr_before_ebs);
-                       pr_info("SCTLR at EFI stub entry       : 0x%08x\n", state->sctlr_before_ebs);
-                       pr_info("CPSR after ExitBootServices() : 0x%08x\n", state->cpsr_after_ebs);
-                       pr_info("SCTLR after ExitBootServices(): 0x%08x\n", state->sctlr_after_ebs);
-               }
-               early_memunmap(state, sizeof(struct efi_arm_entry_state));
-       }
-#endif
-}
-
-static bool efifb_overlaps_pci_range(const struct of_pci_range *range)
-{
-       u64 fb_base = screen_info.lfb_base;
-
-       if (screen_info.capabilities & VIDEO_CAPABILITY_64BIT_BASE)
-               fb_base |= (u64)(unsigned long)screen_info.ext_lfb_base << 32;
-
-       return fb_base >= range->cpu_addr &&
-              fb_base < (range->cpu_addr + range->size);
-}
-
-static struct device_node *find_pci_overlap_node(void)
-{
-       struct device_node *np;
-
-       for_each_node_by_type(np, "pci") {
-               struct of_pci_range_parser parser;
-               struct of_pci_range range;
-               int err;
-
-               err = of_pci_range_parser_init(&parser, np);
-               if (err) {
-                       pr_warn("of_pci_range_parser_init() failed: %d\n", err);
-                       continue;
-               }
-
-               for_each_of_pci_range(&parser, &range)
-                       if (efifb_overlaps_pci_range(&range))
-                               return np;
-       }
-       return NULL;
-}
-
-/*
- * If the efifb framebuffer is backed by a PCI graphics controller, we have
- * to ensure that this relation is expressed using a device link when
- * running in DT mode, or the probe order may be reversed, resulting in a
- * resource reservation conflict on the memory window that the efifb
- * framebuffer steals from the PCIe host bridge.
- */
-static int efifb_add_links(const struct fwnode_handle *fwnode,
-                          struct device *dev)
-{
-       struct device_node *sup_np;
-       struct device *sup_dev;
-
-       sup_np = find_pci_overlap_node();
-
-       /*
-        * If there's no PCI graphics controller backing the efifb, we are
-        * done here.
-        */
-       if (!sup_np)
-               return 0;
-
-       sup_dev = get_dev_from_fwnode(&sup_np->fwnode);
-       of_node_put(sup_np);
-
-       /*
-        * Return -ENODEV if the PCI graphics controller device hasn't been
-        * registered yet.  This ensures that efifb isn't allowed to probe
-        * and this function is retried again when new devices are
-        * registered.
-        */
-       if (!sup_dev)
-               return -ENODEV;
-
-       /*
-        * If this fails, retrying this function at a later point won't
-        * change anything. So, don't return an error after this.
-        */
-       if (!device_link_add(dev, sup_dev, fw_devlink_get_flags()))
-               dev_warn(dev, "device_link_add() failed\n");
-
-       put_device(sup_dev);
-
-       return 0;
-}
-
-static const struct fwnode_operations efifb_fwnode_ops = {
-       .add_links = efifb_add_links,
-};
-
-static struct fwnode_handle efifb_fwnode = {
-       .ops = &efifb_fwnode_ops,
-};
-
-static int __init register_gop_device(void)
-{
-       struct platform_device *pd;
-       int err;
-
-       if (screen_info.orig_video_isVGA != VIDEO_TYPE_EFI)
-               return 0;
-
-       pd = platform_device_alloc("efi-framebuffer", 0);
-       if (!pd)
-               return -ENOMEM;
-
-       if (IS_ENABLED(CONFIG_PCI))
-               pd->dev.fwnode = &efifb_fwnode;
-
-       err = platform_device_add_data(pd, &screen_info, sizeof(screen_info));
-       if (err)
-               return err;
-
-       return platform_device_add(pd);
-}
-subsys_initcall(register_gop_device);
diff --git a/drivers/firmware/efi/efi-init.c b/drivers/firmware/efi/efi-init.c
new file mode 100644 (file)
index 0000000..71c445d
--- /dev/null
@@ -0,0 +1,386 @@
+// SPDX-License-Identifier: GPL-2.0
+/*
+ * Extensible Firmware Interface
+ *
+ * Based on Extensible Firmware Interface Specification version 2.4
+ *
+ * Copyright (C) 2013 - 2015 Linaro Ltd.
+ */
+
+#define pr_fmt(fmt)    "efi: " fmt
+
+#include <linux/efi.h>
+#include <linux/fwnode.h>
+#include <linux/init.h>
+#include <linux/memblock.h>
+#include <linux/mm_types.h>
+#include <linux/of.h>
+#include <linux/of_address.h>
+#include <linux/of_fdt.h>
+#include <linux/platform_device.h>
+#include <linux/screen_info.h>
+
+#include <asm/efi.h>
+
+static int __init is_memory(efi_memory_desc_t *md)
+{
+       if (md->attribute & (EFI_MEMORY_WB|EFI_MEMORY_WT|EFI_MEMORY_WC))
+               return 1;
+       return 0;
+}
+
+/*
+ * Translate a EFI virtual address into a physical address: this is necessary,
+ * as some data members of the EFI system table are virtually remapped after
+ * SetVirtualAddressMap() has been called.
+ */
+static phys_addr_t __init efi_to_phys(unsigned long addr)
+{
+       efi_memory_desc_t *md;
+
+       for_each_efi_memory_desc(md) {
+               if (!(md->attribute & EFI_MEMORY_RUNTIME))
+                       continue;
+               if (md->virt_addr == 0)
+                       /* no virtual mapping has been installed by the stub */
+                       break;
+               if (md->virt_addr <= addr &&
+                   (addr - md->virt_addr) < (md->num_pages << EFI_PAGE_SHIFT))
+                       return md->phys_addr + addr - md->virt_addr;
+       }
+       return addr;
+}
+
+static __initdata unsigned long screen_info_table = EFI_INVALID_TABLE_ADDR;
+static __initdata unsigned long cpu_state_table = EFI_INVALID_TABLE_ADDR;
+
+static const efi_config_table_type_t arch_tables[] __initconst = {
+       {LINUX_EFI_ARM_SCREEN_INFO_TABLE_GUID, &screen_info_table},
+       {LINUX_EFI_ARM_CPU_STATE_TABLE_GUID, &cpu_state_table},
+       {}
+};
+
+static void __init init_screen_info(void)
+{
+       struct screen_info *si;
+
+       if (IS_ENABLED(CONFIG_ARM) &&
+           screen_info_table != EFI_INVALID_TABLE_ADDR) {
+               si = early_memremap_ro(screen_info_table, sizeof(*si));
+               if (!si) {
+                       pr_err("Could not map screen_info config table\n");
+                       return;
+               }
+               screen_info = *si;
+               early_memunmap(si, sizeof(*si));
+
+               /* dummycon on ARM needs non-zero values for columns/lines */
+               screen_info.orig_video_cols = 80;
+               screen_info.orig_video_lines = 25;
+       }
+
+       if (screen_info.orig_video_isVGA == VIDEO_TYPE_EFI &&
+           memblock_is_map_memory(screen_info.lfb_base))
+               memblock_mark_nomap(screen_info.lfb_base, screen_info.lfb_size);
+}
+
+static int __init uefi_init(u64 efi_system_table)
+{
+       efi_config_table_t *config_tables;
+       efi_system_table_t *systab;
+       size_t table_size;
+       int retval;
+
+       systab = early_memremap_ro(efi_system_table, sizeof(efi_system_table_t));
+       if (systab == NULL) {
+               pr_warn("Unable to map EFI system table.\n");
+               return -ENOMEM;
+       }
+
+       set_bit(EFI_BOOT, &efi.flags);
+       if (IS_ENABLED(CONFIG_64BIT))
+               set_bit(EFI_64BIT, &efi.flags);
+
+       retval = efi_systab_check_header(&systab->hdr, 2);
+       if (retval)
+               goto out;
+
+       efi.runtime = systab->runtime;
+       efi.runtime_version = systab->hdr.revision;
+
+       efi_systab_report_header(&systab->hdr, efi_to_phys(systab->fw_vendor));
+
+       table_size = sizeof(efi_config_table_t) * systab->nr_tables;
+       config_tables = early_memremap_ro(efi_to_phys(systab->tables),
+                                         table_size);
+       if (config_tables == NULL) {
+               pr_warn("Unable to map EFI config table array.\n");
+               retval = -ENOMEM;
+               goto out;
+       }
+       retval = efi_config_parse_tables(config_tables, systab->nr_tables,
+                                        IS_ENABLED(CONFIG_ARM) ? arch_tables
+                                                               : NULL);
+
+       early_memunmap(config_tables, table_size);
+out:
+       early_memunmap(systab, sizeof(efi_system_table_t));
+       return retval;
+}
+
+/*
+ * Return true for regions that can be used as System RAM.
+ */
+static __init int is_usable_memory(efi_memory_desc_t *md)
+{
+       switch (md->type) {
+       case EFI_LOADER_CODE:
+       case EFI_LOADER_DATA:
+       case EFI_ACPI_RECLAIM_MEMORY:
+       case EFI_BOOT_SERVICES_CODE:
+       case EFI_BOOT_SERVICES_DATA:
+       case EFI_CONVENTIONAL_MEMORY:
+       case EFI_PERSISTENT_MEMORY:
+               /*
+                * Special purpose memory is 'soft reserved', which means it
+                * is set aside initially, but can be hotplugged back in or
+                * be assigned to the dax driver after boot.
+                */
+               if (efi_soft_reserve_enabled() &&
+                   (md->attribute & EFI_MEMORY_SP))
+                       return false;
+
+               /*
+                * According to the spec, these regions are no longer reserved
+                * after calling ExitBootServices(). However, we can only use
+                * them as System RAM if they can be mapped writeback cacheable.
+                */
+               return (md->attribute & EFI_MEMORY_WB);
+       default:
+               break;
+       }
+       return false;
+}
+
+static __init void reserve_regions(void)
+{
+       efi_memory_desc_t *md;
+       u64 paddr, npages, size;
+
+       if (efi_enabled(EFI_DBG))
+               pr_info("Processing EFI memory map:\n");
+
+       /*
+        * Discard memblocks discovered so far: if there are any at this
+        * point, they originate from memory nodes in the DT, and UEFI
+        * uses its own memory map instead.
+        */
+       memblock_dump_all();
+       memblock_remove(0, PHYS_ADDR_MAX);
+
+       for_each_efi_memory_desc(md) {
+               paddr = md->phys_addr;
+               npages = md->num_pages;
+
+               if (efi_enabled(EFI_DBG)) {
+                       char buf[64];
+
+                       pr_info("  0x%012llx-0x%012llx %s\n",
+                               paddr, paddr + (npages << EFI_PAGE_SHIFT) - 1,
+                               efi_md_typeattr_format(buf, sizeof(buf), md));
+               }
+
+               memrange_efi_to_native(&paddr, &npages);
+               size = npages << PAGE_SHIFT;
+
+               if (is_memory(md)) {
+                       early_init_dt_add_memory_arch(paddr, size);
+
+                       if (!is_usable_memory(md))
+                               memblock_mark_nomap(paddr, size);
+
+                       /* keep ACPI reclaim memory intact for kexec etc. */
+                       if (md->type == EFI_ACPI_RECLAIM_MEMORY)
+                               memblock_reserve(paddr, size);
+               }
+       }
+}
+
+void __init efi_init(void)
+{
+       struct efi_memory_map_data data;
+       u64 efi_system_table;
+
+       /* Grab UEFI information placed in FDT by stub */
+       efi_system_table = efi_get_fdt_params(&data);
+       if (!efi_system_table)
+               return;
+
+       if (efi_memmap_init_early(&data) < 0) {
+               /*
+               * If we are booting via UEFI, the UEFI memory map is the only
+               * description of memory we have, so there is little point in
+               * proceeding if we cannot access it.
+               */
+               panic("Unable to map EFI memory map.\n");
+       }
+
+       WARN(efi.memmap.desc_version != 1,
+            "Unexpected EFI_MEMORY_DESCRIPTOR version %ld",
+             efi.memmap.desc_version);
+
+       if (uefi_init(efi_system_table) < 0) {
+               efi_memmap_unmap();
+               return;
+       }
+
+       reserve_regions();
+       efi_esrt_init();
+
+       memblock_reserve(data.phys_map & PAGE_MASK,
+                        PAGE_ALIGN(data.size + (data.phys_map & ~PAGE_MASK)));
+
+       init_screen_info();
+
+#ifdef CONFIG_ARM
+       /* ARM does not permit early mappings to persist across paging_init() */
+       efi_memmap_unmap();
+
+       if (cpu_state_table != EFI_INVALID_TABLE_ADDR) {
+               struct efi_arm_entry_state *state;
+               bool dump_state = true;
+
+               state = early_memremap_ro(cpu_state_table,
+                                         sizeof(struct efi_arm_entry_state));
+               if (state == NULL) {
+                       pr_warn("Unable to map CPU entry state table.\n");
+                       return;
+               }
+
+               if ((state->sctlr_before_ebs & 1) == 0)
+                       pr_warn(FW_BUG "EFI stub was entered with MMU and Dcache disabled, please fix your firmware!\n");
+               else if ((state->sctlr_after_ebs & 1) == 0)
+                       pr_warn(FW_BUG "ExitBootServices() returned with MMU and Dcache disabled, please fix your firmware!\n");
+               else
+                       dump_state = false;
+
+               if (dump_state || efi_enabled(EFI_DBG)) {
+                       pr_info("CPSR at EFI stub entry        : 0x%08x\n", state->cpsr_before_ebs);
+                       pr_info("SCTLR at EFI stub entry       : 0x%08x\n", state->sctlr_before_ebs);
+                       pr_info("CPSR after ExitBootServices() : 0x%08x\n", state->cpsr_after_ebs);
+                       pr_info("SCTLR after ExitBootServices(): 0x%08x\n", state->sctlr_after_ebs);
+               }
+               early_memunmap(state, sizeof(struct efi_arm_entry_state));
+       }
+#endif
+}
+
+static bool efifb_overlaps_pci_range(const struct of_pci_range *range)
+{
+       u64 fb_base = screen_info.lfb_base;
+
+       if (screen_info.capabilities & VIDEO_CAPABILITY_64BIT_BASE)
+               fb_base |= (u64)(unsigned long)screen_info.ext_lfb_base << 32;
+
+       return fb_base >= range->cpu_addr &&
+              fb_base < (range->cpu_addr + range->size);
+}
+
+static struct device_node *find_pci_overlap_node(void)
+{
+       struct device_node *np;
+
+       for_each_node_by_type(np, "pci") {
+               struct of_pci_range_parser parser;
+               struct of_pci_range range;
+               int err;
+
+               err = of_pci_range_parser_init(&parser, np);
+               if (err) {
+                       pr_warn("of_pci_range_parser_init() failed: %d\n", err);
+                       continue;
+               }
+
+               for_each_of_pci_range(&parser, &range)
+                       if (efifb_overlaps_pci_range(&range))
+                               return np;
+       }
+       return NULL;
+}
+
+/*
+ * If the efifb framebuffer is backed by a PCI graphics controller, we have
+ * to ensure that this relation is expressed using a device link when
+ * running in DT mode, or the probe order may be reversed, resulting in a
+ * resource reservation conflict on the memory window that the efifb
+ * framebuffer steals from the PCIe host bridge.
+ */
+static int efifb_add_links(const struct fwnode_handle *fwnode,
+                          struct device *dev)
+{
+       struct device_node *sup_np;
+       struct device *sup_dev;
+
+       sup_np = find_pci_overlap_node();
+
+       /*
+        * If there's no PCI graphics controller backing the efifb, we are
+        * done here.
+        */
+       if (!sup_np)
+               return 0;
+
+       sup_dev = get_dev_from_fwnode(&sup_np->fwnode);
+       of_node_put(sup_np);
+
+       /*
+        * Return -ENODEV if the PCI graphics controller device hasn't been
+        * registered yet.  This ensures that efifb isn't allowed to probe
+        * and this function is retried again when new devices are
+        * registered.
+        */
+       if (!sup_dev)
+               return -ENODEV;
+
+       /*
+        * If this fails, retrying this function at a later point won't
+        * change anything. So, don't return an error after this.
+        */
+       if (!device_link_add(dev, sup_dev, fw_devlink_get_flags()))
+               dev_warn(dev, "device_link_add() failed\n");
+
+       put_device(sup_dev);
+
+       return 0;
+}
+
+static const struct fwnode_operations efifb_fwnode_ops = {
+       .add_links = efifb_add_links,
+};
+
+static struct fwnode_handle efifb_fwnode = {
+       .ops = &efifb_fwnode_ops,
+};
+
+static int __init register_gop_device(void)
+{
+       struct platform_device *pd;
+       int err;
+
+       if (screen_info.orig_video_isVGA != VIDEO_TYPE_EFI)
+               return 0;
+
+       pd = platform_device_alloc("efi-framebuffer", 0);
+       if (!pd)
+               return -ENOMEM;
+
+       if (IS_ENABLED(CONFIG_PCI))
+               pd->dev.fwnode = &efifb_fwnode;
+
+       err = platform_device_add_data(pd, &screen_info, sizeof(screen_info));
+       if (err)
+               return err;
+
+       return platform_device_add(pd);
+}
+subsys_initcall(register_gop_device);