diff --git a/local/recipes/drivers/linux-kpi/source/src/c_headers/linux/amdgpu_stubs.h b/local/recipes/drivers/linux-kpi/source/src/c_headers/linux/amdgpu_stubs.h
new file mode 100644
index 0000000000..234afdfa1a
--- /dev/null
+++ b/local/recipes/drivers/linux-kpi/source/src/c_headers/linux/amdgpu_stubs.h
@@ -0,0 +1,141 @@
+#ifndef _REDOX_AMDGPU_STUBS_H
+#define _REDOX_AMDGPU_STUBS_H
+
+#include <stdarg.h>
+
+/* Already in linux-kpi but may be included out-of-order */
+#include <linux/spinlock.h>
+#include <linux/bug.h>
+#include <linux/lockdep.h>
+#include <linux/mutex.h>
+#include <linux/completion.h>
+#include <linux/idr.h>
+#include <linux/bitops.h>
+#include <linux/delay.h>
+
+#define _THIS_IP_ ((void *)0)
+
+#ifndef __counted_by
+#define __counted_by(x)
+#endif
+
+#define compiletime_assert(cond, msg) do { (void)sizeof(char[1 - 2*!(cond)]); } while (0)
+#define __compiletime_assert(cond, msg, prefix, suffix) compiletime_assert(cond, msg)
+
+struct raw_spinlock {
+    spinlock_t lock;
+};
+#define raw_spin_lock_init(l) spin_lock_init(&(l)->lock)
+#define raw_spin_lock(l) spin_lock(&(l)->lock)
+#define raw_spin_unlock(l) spin_unlock(&(l)->lock)
+
+static inline u64 ktime_get_raw_ns(void) { return 0; }
+static inline u64 ktime_get_ns(void) { return 0; }
+static inline u64 ktime_get_real_ns(void) { return 0; }
+static inline u64 ktime_get_boottime_ns(void) { return 0; }
+static inline s64 ktime_us_delta(u64 later, u64 earlier) { return (s64)(later - earlier) / 1000; }
+
+/* math64 */
+static inline u64 div_u64_rem(u64 dividend, u32 divisor, u32 *remainder) {
+    *remainder = (u32)(dividend % divisor);
+    return dividend / divisor;
+}
+static inline u64 div_u64(u64 dividend, u32 divisor) { return dividend / divisor; }
+static inline u64 div64_u64(u64 dividend, u64 divisor) { return dividend / divisor; }
+static inline u64 div64_u64_rem(u64 dividend, u64 divisor, u64 *remainder) {
+    *remainder = dividend % divisor;
+    return dividend / divisor;
+}
+static inline s64 div64_s64(s64 dividend, s64 divisor) { return dividend / divisor; }
+static inline u32 div64_u64_rem_u32(u64 dividend, u64 divisor, u32 *remainder) {
+    *remainder = (u32)(dividend % divisor);
+    return (u32)(dividend / divisor);
+}
+
+#define _THIS_IP_ ((void *)0)
+struct va_format {
+    const char *fmt;
+    va_list *va;
+};
+
+/* Additional types not in linux-kpi yet */
+struct srcu_struct { int dummy; };
+#define DEFINE_SRCU(name) struct srcu_struct name
+#define srcu_read_lock(s) 0
+#define srcu_read_unlock(s, idx) do {} while (0)
+
+/* hdmi / audio stubs */
+enum hdmi_infoframe_type { HDMI_INFOFRAME_TYPE_MAX };
+#define drm_hdmi_infoframe_set_hdr_metadata(...) (-22)
+#define drm_hdmi_avi_infoframe_colorimetry(...) do {} while (0)
+#define drm_hdmi_avi_infoframe_bars(...) do {} while (0)
+
+/* i2c stubs */
+struct i2c_adapter;
+struct i2c_msg { int dummy; };
+#define i2c_transfer(a, m, n) 0
+
+/* pci stubs that conflict with redox glue */
+#define pci_read_config_byte(d, w, v) ({ *(v) = 0; 0; })
+#define pci_write_config_byte(d, w, v) 0
+#define pci_read_config_word(d, w, v) ({ *(v) = 0; 0; })
+#define pci_write_config_word(d, w, v) 0
+#define pci_read_config_dword(d, w, v) ({ *(v) = 0; 0; })
+#define pci_write_config_dword(d, w, v) 0
+#define pci_get_drvdata(pdev) NULL
+#define pci_set_drvdata(pdev, data) do {} while (0)
+
+/* power management */
+#define SET_SYSTEM_SLEEP_PM_OPS(s, r) 
+#define SIMPLE_DEV_PM_OPS(n, s, r) 
+#define DEFINE_SIMPLE_DEV_PM_OPS(n, s, r) 
+#define pm_ptr(x) NULL
+#define pm_sleep_ptr(x) NULL
+#define SYSTEM_SLEEP_PM_OPS(s, r) 
+#define RUNTIME_PM_OPS(s, r) 
+#define UNIVERSAL_DEV_PM_OPS(s, r, i) 
+
+/* devres / device managed resources */
+#define devm_kzalloc(d, s, f) kzalloc(s, f)
+#define devm_kmalloc(d, s, f) kmalloc(s, f)
+#define devm_kcalloc(d, n, s, f) kzalloc((n)*(s), f)
+#define devm_kmemdup(d, src, len, gfp) ({ void *p = kmalloc(len, gfp); if(p) memcpy(p, src, len); p; })
+#define devm_request_threaded_irq(d, i, h, th, f, n, dev) 0
+#define devm_free_irq(d, i, dev) do {} while (0)
+#define devm_ioremap(d, o, s) ioremap(o, s)
+#define devm_iounmap(d, a) iounmap(a)
+#define devm_pci_remap_cfgspace(d, o, s) ioremap(o, s)
+
+/* backlight */
+#define backlight_device_register(n, d, dev, ops, props) NULL
+#define backlight_device_unregister(bd) do {} while (0)
+#define backlight_device_set_brightness(bd, b) 0
+#define backlight_device_get_brightness(bd) 0
+
+/* firmware / efivars */
+#define efivar_entry_size() 0
+#define efivar_entry_iter_begin() do {} while (0)
+#define efivar_entry_iter_end() do {} while (0)
+
+/* cpumask / topology */
+#define cpumask_of(cpu) NULL
+#define topology_physical_package_id(cpu) 0
+
+/* HDA / sound */
+struct hda_codec;
+struct hda_nid_item;
+
+/* KMS property helpers */
+#define drm_object_attach_property(obj, prop, val) 0
+
+/* trace */
+#define trace_drm_atomic_state_init(state) do {} while (0)
+
+/* hdmi infoframe pack */
+#define hdmi_infoframe_pack(f, b, s) (-22)
+#define hdmi_infoframe_pack_only(f, b, s) (-22)
+
+/* input */
+#define input_free_device(d) do {} while (0)
+
+#endif
diff --git a/local/recipes/drivers/linux-kpi/source/src/c_headers/linux/byteorder/generic.h b/local/recipes/drivers/linux-kpi/source/src/c_headers/linux/byteorder/generic.h
new file mode 100644
index 0000000000..9f5270ea7c
--- /dev/null
+++ b/local/recipes/drivers/linux-kpi/source/src/c_headers/linux/byteorder/generic.h
@@ -0,0 +1,53 @@
+#ifndef _LINUX_BYTEORDER_GENERIC_H
+#define _LINUX_BYTEORDER_GENERIC_H
+
+#include <linux/types.h>
+#include <stdint.h>
+#include <endian.h>
+
+#define cpu_to_le16(x)  ((__le16)(uint16_t)(x))
+#define cpu_to_le32(x)  ((__le32)(uint32_t)(x))
+#define cpu_to_le64(x)  ((__le64)(uint64_t)(x))
+#define le16_to_cpu(x)  ((uint16_t)(__le16)(x))
+#define le32_to_cpu(x)  ((uint32_t)(__le32)(x))
+#define le64_to_cpu(x)  ((uint64_t)(__le64)(x))
+
+#define cpu_to_be16(x)  ((__be16)__builtin_bswap16((uint16_t)(x)))
+#define cpu_to_be32(x)  ((__be32)__builtin_bswap32((uint32_t)(x)))
+#define cpu_to_be64(x)  ((__be64)__builtin_bswap64((uint64_t)(x)))
+#define be16_to_cpu(x)  __builtin_bswap16((uint16_t)(__be16)(x))
+#define be32_to_cpu(x)  __builtin_bswap32((uint32_t)(__be32)(x))
+#define be64_to_cpu(x)  __builtin_bswap64((uint64_t)(__be64)(x))
+
+static inline void __cpu_to_le16s(__le16 *p) { *p = cpu_to_le16(*p); }
+static inline void __cpu_to_le32s(__le32 *p) { *p = cpu_to_le32(*p); }
+static inline void __cpu_to_le64s(__le64 *p) { *p = cpu_to_le64(*p); }
+static inline void __le32_to_cpus(__le32 *p) { *p = le32_to_cpu(*p); }
+static inline void __le64_to_cpus(__le64 *p) { *p = le64_to_cpu(*p); }
+static inline void __cpu_to_be16s(__be16 *p) { *p = cpu_to_be16(*p); }
+static inline void __cpu_to_be32s(__be32 *p) { *p = cpu_to_be32(*p); }
+static inline void __cpu_to_be64s(__be64 *p) { *p = cpu_to_be64(*p); }
+static inline void __be32_to_cpus(__be32 *p) { *p = be32_to_cpu(*p); }
+static inline void __be64_to_cpus(__be64 *p) { *p = be64_to_cpu(*p); }
+
+static inline __le16 __cpu_to_le16(uint16_t x) { return cpu_to_le16(x); }
+static inline __le32 __cpu_to_le32(uint32_t x) { return cpu_to_le32(x); }
+static inline __le64 __cpu_to_le64(uint64_t x) { return cpu_to_le64(x); }
+static inline uint16_t __le16_to_cpu(__le16 x) { return le16_to_cpu(x); }
+static inline uint32_t __le32_to_cpu(__le32 x) { return le32_to_cpu(x); }
+static inline uint64_t __le64_to_cpu(__le64 x) { return le64_to_cpu(x); }
+static inline __be16 __cpu_to_be16(uint16_t x) { return cpu_to_be16(x); }
+static inline __be32 __cpu_to_be32(uint32_t x) { return cpu_to_be32(x); }
+static inline __be64 __cpu_to_be64(uint64_t x) { return cpu_to_be64(x); }
+static inline uint16_t __be16_to_cpu(__be16 x) { return be16_to_cpu(x); }
+static inline uint32_t __be32_to_cpu(__be32 x) { return be32_to_cpu(x); }
+static inline uint64_t __be64_to_cpu(__be64 x) { return be64_to_cpu(x); }
+
+static inline void le16_add_cpu(__le16 *var, u16 val) { *var = cpu_to_le16(le16_to_cpu(*var) + val); }
+static inline void le32_add_cpu(__le32 *var, u32 val) { *var = cpu_to_le32(le32_to_cpu(*var) + val); }
+static inline void le64_add_cpu(__le64 *var, u64 val) { *var = cpu_to_le64(le64_to_cpu(*var) + val); }
+static inline void be16_add_cpu(__be16 *var, u16 val) { *var = cpu_to_be16(be16_to_cpu(*var) + val); }
+static inline void be32_add_cpu(__be32 *var, u32 val) { *var = cpu_to_be32(be32_to_cpu(*var) + val); }
+static inline void be64_add_cpu(__be64 *var, u64 val) { *var = cpu_to_be64(be64_to_cpu(*var) + val); }
+
+#endif
diff --git a/local/recipes/drivers/linux-kpi/source/src/c_headers/linux/idr.h b/local/recipes/drivers/linux-kpi/source/src/c_headers/linux/idr.h
index 13fd1283c2..05b7e37e40 100644
--- a/local/recipes/drivers/linux-kpi/source/src/c_headers/linux/idr.h
+++ b/local/recipes/drivers/linux-kpi/source/src/c_headers/linux/idr.h
@@ -40,6 +40,18 @@ static inline void idr_destroy(struct idr *idr)
     (void)idr;
 }
 
+struct ida {
+    struct idr idr;
+};
+
+#define DEFINE_IDA(name) struct ida name = { .idr = { .__opaque = {0} } }
+#define ida_init(ida) idr_init(&(ida)->idr)
+#define ida_alloc(ida, gfp) idr_alloc(&(ida)->idr, NULL, 0, 0, (u32)(gfp))
+#define ida_alloc_range(ida, start, end, gfp) idr_alloc(&(ida)->idr, NULL, (start), (end), (u32)(gfp))
+#define ida_alloc_min(ida, min, gfp) ida_alloc_range(ida, min, INT_MAX, gfp)
+#define ida_free(ida, id) idr_remove(&(ida)->idr, (id))
+#define ida_destroy(ida) idr_destroy(&(ida)->idr)
+
 #define idr_for_each_entry(idr, entry, id) \
     for ((id) = 0, (entry) = (void *)0; (entry); (id)++)
 
diff --git a/local/recipes/drivers/linux-kpi/source/src/c_headers/linux/spinlock_types.h b/local/recipes/drivers/linux-kpi/source/src/c_headers/linux/spinlock_types.h
new file mode 100644
index 0000000000..2cc0ad6d7e
--- /dev/null
+++ b/local/recipes/drivers/linux-kpi/source/src/c_headers/linux/spinlock_types.h
@@ -0,0 +1,6 @@
+#ifndef _LINUX_SPINLOCK_TYPES_H
+#define _LINUX_SPINLOCK_TYPES_H
+
+#include "spinlock.h"
+
+#endif
diff --git a/local/recipes/drivers/linux-kpi/source/src/c_headers/linux/stddef.h b/local/recipes/drivers/linux-kpi/source/src/c_headers/linux/stddef.h
new file mode 100644
index 0000000000..6b7ec1f3bc
--- /dev/null
+++ b/local/recipes/drivers/linux-kpi/source/src/c_headers/linux/stddef.h
@@ -0,0 +1,16 @@
+#ifndef _LINUX_STDDEF_H
+#define _LINUX_STDDEF_H
+
+#include <stddef.h>
+#include <stdbool.h>
+
+#undef NULL
+#define NULL ((void *)0)
+
+#undef offsetof
+#define offsetof(TYPE, MEMBER) __builtin_offsetof(TYPE, MEMBER)
+
+#define sizeof_field(TYPE, MEMBER) sizeof((((TYPE *)0)->MEMBER))
+#define offsetofend(TYPE, MEMBER) (offsetof(TYPE, MEMBER) + sizeof_field(TYPE, MEMBER))
+
+#endif
diff --git a/local/recipes/drivers/linux-kpi/source/src/c_headers/linux/types.h b/local/recipes/drivers/linux-kpi/source/src/c_headers/linux/types.h
index 392fca236b..e993d9a431 100644
--- a/local/recipes/drivers/linux-kpi/source/src/c_headers/linux/types.h
+++ b/local/recipes/drivers/linux-kpi/source/src/c_headers/linux/types.h
@@ -28,6 +28,14 @@ typedef u16 __u16;
 typedef u8  __u8;
 typedef s8  __s8;
 typedef s16 __s16;
+
+typedef u16 __le16;
+typedef u32 __le32;
+typedef u64 __le64;
+typedef u16 __be16;
+typedef u32 __be32;
+typedef u64 __be64;
+
 typedef size_t __kernel_size_t;
 #define __bitwise
 
@@ -43,18 +51,6 @@ typedef unsigned __bitwise gfp_t;
 
 #define BITS_PER_LONG (8 * sizeof(long))
 
-typedef struct { long long counter; } atomic_long_t;
-
-#define ATOMIC_LONG_INIT(v) { .counter = (v) }
-#define atomic_long_read(v) ((v)->counter)
-#define atomic_long_set(v, i) do { (v)->counter = (i); } while (0)
-#define atomic_long_inc(v) __sync_fetch_and_add(&(v)->counter, 1)
-#define atomic_long_dec(v) __sync_fetch_and_sub(&(v)->counter, 1)
-#define atomic_long_add(i, v) __sync_fetch_and_add(&(v)->counter, (i))
-#define atomic_long_sub(i, v) __sync_fetch_and_sub(&(v)->counter, (i))
-#define atomic_long_inc_return(v) __sync_add_and_fetch(&(v)->counter, 1)
-#define atomic_long_xchg(v, n) __sync_lock_test_and_set(&(v)->counter, (n))
-
 typedef unsigned int gfp_t_long_deprecated;
 
 typedef struct {
diff --git a/local/recipes/gpu/amdgpu/recipe.toml b/local/recipes/gpu/amdgpu/recipe.toml
index 9a0fbd94c9..bc29b5ee9d 100644
--- a/local/recipes/gpu/amdgpu/recipe.toml
+++ b/local/recipes/gpu/amdgpu/recipe.toml
@@ -32,6 +32,7 @@ TARGET_CC="${TARGET}-gcc"
 export CFLAGS="-std=gnu11 -D__redox__ -D__KERNEL__ -DCONFIG_DRM_AMDGPU -DCONFIG_DRM_AMD_DC \
     -DCONFIG_DRM_AMD_DC_DML2=1 \
     -DCONFIG_DRM_AMD_DC_FP -DCONFIG_DRM_AMD_ACP \
+    -include linux/amdgpu_stubs.h \
     -I${LINUX_KPI} \
     -I${REDOX_GLUE} \
     -I${INCLUDES} \
@@ -40,6 +41,10 @@ export CFLAGS="-std=gnu11 -D__redox__ -D__KERNEL__ -DCONFIG_DRM_AMDGPU -DCONFIG_
     -I${AMD_SRC}/include/asic_reg \
     -I${AMD_SRC}/display \
     -I${AMD_SRC}/display/dc \
+    -I${AMD_SRC}/display/include \
+    -I${AMD_SRC}/display/dc/inc \
+    -I${AMD_SRC}/display/dc/inc/hw \
+    -I${AMD_SRC}/display/dmub/inc \
     -I${AMD_SRC}/display/dc/dml \
     -I${AMD_SRC}/display/dc/dml2_0 \
     -I${AMD_SRC}/display/dc/dml2_0/dml21 \
@@ -49,8 +54,6 @@ export CFLAGS="-std=gnu11 -D__redox__ -D__KERNEL__ -DCONFIG_DRM_AMDGPU -DCONFIG_
     -I${AMD_SRC}/display/dc/dcn301 \
     -I${AMD_SRC}/display/dc/dcn31 \
     -I${AMD_SRC}/display/dc/dcn32 \
-    -I${AMD_SRC}/display/dc/dcn35 \
-    -I${AMD_SRC}/display/dmub \
     -I${AMD_SRC}/display/modules \
     -I${AMD_SRC}/display/modules/freesync \
     -I${AMD_SRC}/display/modules/color \
@@ -68,17 +71,16 @@ export CFLAGS="-std=gnu11 -D__redox__ -D__KERNEL__ -DCONFIG_DRM_AMDGPU -DCONFIG_
 "${TARGET_CC}" -c ${CFLAGS} "${REDOX_GLUE}/amdgpu_redox_main.c" -o amdgpu_redox_main.o
 "${TARGET_CC}" -c ${CFLAGS} "${REDOX_GLUE}/redox_stubs.c" -o redox_stubs.o
 
-# Stage 2: Bounded first-display path
+# Stage 2: DCN31 hardware display files (RDNA2 GPU — Navi21/22/23/24)
 #
-# The current Red Bear AMD display bring-up path does not call into the imported
-# Linux AMD Display Core tree directly. The live FFI surface comes from the
-# Red Bear glue layer (`amdgpu_redox_main.c` / `redox_stubs.c`), while the
-# broad `display/*.c` compile currently drags in optional and unsupported
-# subtrees such as freesync before the retained path is even proven.
-#
-# Keep Stage 2 explicit and intentionally empty until a retained imported
-# display-source subset is proven necessary by bounded compile triage.
-DISPLAY_SRCS=""
+# These DCN 3.1 hardware backend files have been proven to compile against
+# the linux-kpi compatibility headers. They form the minimal verified subset
+# of the imported Linux Display Core tree. Additional DCN revisions (dcn20,
+# dcn30, dcn32) will be added as their compile triage completes.
+DISPLAY_SRCS="${AMD_SRC}/display/dc/dcn31/dcn31_afmt.c
+${AMD_SRC}/display/dc/dcn31/dcn31_vpg.c
+${AMD_SRC}/display/dc/dcn31/dcn31_apg.c
+${AMD_SRC}/display/dc/dcn31/dcn31_panel_cntl.c"
 success=0
 failed=0
 for src in $DISPLAY_SRCS; do