diff --git a/memory.x b/memory.x
index e6b80c473..add5c902f 100644
--- a/memory.x
+++ b/memory.x
@@ -34,3 +34,38 @@ SECTIONS {
         KEEP(*(.boot2));
     } > BOOT2
 } INSERT BEFORE .text;
+
+/* Per-core (thread) data into flash */
+SECTIONS {
+    .tdata : ALIGN(4)
+    {
+        . = ALIGN(4);
+        PROVIDE(__tdata_start = .);
+        *(.tdata .tdata.*);
+        . = ALIGN(4);
+        PROVIDE(__tdata_end = .);
+    } > FLASH
+    PROVIDE(__tdata_len = __tdata_end - __tdata_start);
+} INSERT AFTER .data;
+
+/* Size per-core state and allocate bss space for each core */
+SECTIONS {
+    .tbss (NOLOAD) : ALIGN(4)
+    {
+        . = ALIGN(4);
+        PROVIDE(__tbss_start = .);
+        *(.tbss .tbss.*);
+        *(.tcommon);
+        . = ALIGN(4);
+        PROVIDE(__tbss_end = .);
+    } > RAM
+    PROVIDE(__tbss_len = __tbss_end - __tbss_start);
+
+    .tls_state (NOLOAD) : ALIGN(4) {
+        PROVIDE(TLS_CORE_0 = ALIGN(4));
+        . += __tdata_len + __tbss_len;
+        PROVIDE(TLS_CORE_1 = ALIGN(4));
+        . += __tdata_len + __tbss_len;
+    } > RAM
+} INSERT AFTER .bss;
+
diff --git a/rp2040-hal/Cargo.toml b/rp2040-hal/Cargo.toml
index 0718629ff..fec750101 100644
--- a/rp2040-hal/Cargo.toml
+++ b/rp2040-hal/Cargo.toml
@@ -106,6 +106,9 @@ rtic-monotonic = ["dep:rtic-monotonic"]
 # Implement `i2c-write-iter` traits
 i2c-write-iter = ["dep:i2c-write-iter"]
 
+# Enable use of thread-local variables for multicore state
+thread_local = []
+
 [[example]]
 # irq example uses cortex-m-rt::interrupt, need rt feature for that
 name = "gpio_irq_example"
diff --git a/rp2040-hal/src/multicore.rs b/rp2040-hal/src/multicore.rs
index d1b018402..afe809cf4 100644
--- a/rp2040-hal/src/multicore.rs
+++ b/rp2040-hal/src/multicore.rs
@@ -290,3 +290,68 @@ impl<'p> Core<'p> {
         }
     }
 }
+
+#[cfg(all(target_arch = "arm", feature = "thread_local"))]
+mod thread_local {
+    use core::arch::global_asm;
+    use core::ptr::addr_of;
+
+    extern "C" {
+        static TLS_CORE_0: u8;
+        static TLS_CORE_1: u8;
+    }
+    // Not really a const pointer, but we reform it into mut in the asm
+    static mut TLS_STATE: [*const u8; 2] = [
+        // Point to linker-allocated space in .bss
+        unsafe { addr_of!(TLS_CORE_0) },
+        unsafe { addr_of!(TLS_CORE_1) },
+    ];
+
+    // Define `__aeabi_read_tp` called by the compiler to get access to
+    // thread-local storage.
+    global_asm! {
+        ".pushsection .text.__aeabi_read_tp",
+        ".align 4",
+        ".p2align 4,,15",
+        ".global __aeabi_read_tp",
+        ".type __aeabi_read_tp,%function",
+
+        "__aeabi_read_tp:",
+        "    push {{r1, lr}}",
+        "    ldr r1, =0xd0000000",      // Load SIO CPUID addr
+        "    ldr r1, [r1]",             // Get current CPUID
+        "    lsls r1, r1, #2",          // Scale by 4
+        "    ldr r0, ={tls_state}",     // Load TLS_STATE base addr
+        "    ldr r0, [r0, r1]",         // Load CPU per-thread
+        "    pop {{r1, pc}}",
+
+        ".popsection",
+        tls_state = sym TLS_STATE,
+    }
+
+    // Intercept __pre_init to hook into the startup code to copy the tdata into
+    // TLS_CORE_[01].
+    global_asm! {
+        ".pushsection .text.__pre_init",
+        ".align 4",
+        ".p2align 4,,15",
+        ".global __pre_init",
+        ".type __pre_init,%function",
+
+        "__pre_init:",
+        "    push {{lr}}",
+        "    ldr r0, ={tls_core_0}",
+        "    ldr r1, =__tdata_start",
+        "    ldr r2, =__tdata_len",
+        "    bl __aeabi_memcpy",
+        "    ldr r0, ={tls_core_1}",
+        "    ldr r1, =__tdata_start",
+        "    ldr r2, =__tdata_len",
+        "    bl __aeabi_memcpy",
+        "    pop {{pc}}",
+
+        ".popsection",
+        tls_core_0 = sym TLS_CORE_0,
+        tls_core_1 = sym TLS_CORE_1,
+    }
+}