diff --git a/assets/build/base.yaml b/assets/build/base.yaml
index 114c4fc..3c7ea0e 100644
--- a/assets/build/base.yaml
+++ b/assets/build/base.yaml
@@ -22,7 +22,7 @@ variables:
     '-DGIT_VERSION=\"${version_major}.${version_minor}.${version_patch}+${version_sha}\"',
     '-DGIT_VERSION_WIDE=L\"${version_major}.${version_minor}.${version_patch}+${version_sha}\"',
 
-    "-Wformat=2", "-Winit-self", "-Wfloat-equal", "-Winline", "-Wmissing-format-attribute",
+    "-Wformat=2", "-Winit-self", "-Winline", "-Wmissing-format-attribute",
     "-Wmissing-include-dirs", "-Wswitch", "-Wundef", "-Wdisabled-optimization",
     "-Wpointer-arith", "-Wno-attributes", "-Wno-sign-compare", "-Wno-multichar",
     "-Wno-div-by-zero", "-Wno-endif-labels", "-Wno-pragmas", "-Wno-format-extra-args",
diff --git a/src/kernel/cpu.cpp b/src/kernel/cpu.cpp
index e46be42..71fc366 100644
--- a/src/kernel/cpu.cpp
+++ b/src/kernel/cpu.cpp
@@ -14,6 +14,7 @@
 #include "objects/thread.h"
 #include "syscall.h"
 #include "tss.h"
+#include "xsave.h"
 
 unsigned g_num_cpus = 1;
 
@@ -140,6 +141,7 @@ bsp_early_init()
     cpu->gdt = new (&g_bsp_gdt) GDT {cpu->tss};
     cpu->rsp0 = reinterpret_cast<uintptr_t>(&idle_stack_end);
     cpu_early_init(cpu);
+    xsave_init();
 
     return cpu;
 }
@@ -229,4 +231,6 @@ cpu_init(cpu_data *cpu, bool bsp)
         cpu->id = apic->get_id();
         apic->calibrate_timer();
     }
+
+    xsave_enable();
 }
diff --git a/src/kernel/cpu.h b/src/kernel/cpu.h
index 5d4e996..8912bb1 100644
--- a/src/kernel/cpu.h
+++ b/src/kernel/cpu.h
@@ -51,6 +51,8 @@ enum class xcr0
     ZMM_Hi256,
     ZMM_Hi16,
     PKRU = 9,
+
+    J6_SUPPORTED = X87 | SSE | AVX | BINDREG | BINDCSR | OPMASK | ZMM_Hi16 | ZMM_Hi256,
 };
 
 enum class efer
diff --git a/src/kernel/kernel.module b/src/kernel/kernel.module
index fa5fb33..6680182 100644
--- a/src/kernel/kernel.module
+++ b/src/kernel/kernel.module
@@ -65,6 +65,7 @@ kernel = module("kernel",
         "tss.cpp",
         "vm_space.cpp",
         "wait_queue.cpp",
+        "xsave.cpp",
     ])
 
 if config == "debug":
diff --git a/src/kernel/objects/process.cpp b/src/kernel/objects/process.cpp
index b99a3ce..b2e3e9e 100644
--- a/src/kernel/objects/process.cpp
+++ b/src/kernel/objects/process.cpp
@@ -88,6 +88,9 @@ process::create_thread(uintptr_t rsp3, uint8_t priority)
     if (rsp3)
         th->tcb()->rsp3 = rsp3;
 
+    if (this != &g_kernel_process)
+        th->init_xsave_area();
+
     m_threads.append(th);
     scheduler::get().add_thread(th->tcb());
     return th;
diff --git a/src/kernel/objects/thread.cpp b/src/kernel/objects/thread.cpp
index 6d6ddad..1a58f2f 100644
--- a/src/kernel/objects/thread.cpp
+++ b/src/kernel/objects/thread.cpp
@@ -8,6 +8,7 @@
 #include "objects/process.h"
 #include "objects/vm_area.h"
 #include "scheduler.h"
+#include "xsave.h"
 
 extern "C" void initialize_user_cpu();
 extern obj::vm_area_guarded &g_kernel_stacks;
@@ -37,6 +38,9 @@ thread::thread(process &parent, uint8_t pri, uintptr_t rsp0) :
 
 thread::~thread()
 {
+    if (m_tcb.xsave)
+        delete [] reinterpret_cast<uint8_t*>(m_tcb.xsave);
+
     g_kernel_stacks.return_section(m_tcb.kernel_stack);
     m_parent.handle_release();
 }
@@ -158,6 +162,14 @@ thread::add_thunk_user(uintptr_t rip3, uint64_t arg0, uint64_t arg1, uintptr_t r
     add_thunk_kernel(rip0 ? rip0 : trampoline);
 }
 
+void
+thread::init_xsave_area()
+{
+    void *xsave_area = new uint8_t [xsave_size];
+    memset(xsave_area, 0, xsave_size);
+    m_tcb.xsave = reinterpret_cast<uintptr_t>(xsave_area);
+}
+
 void
 thread::setup_kernel_stack()
 {
diff --git a/src/kernel/objects/thread.h b/src/kernel/objects/thread.h
index 22d5c60..52c685a 100644
--- a/src/kernel/objects/thread.h
+++ b/src/kernel/objects/thread.h
@@ -26,6 +26,7 @@ struct TCB
     uintptr_t rsp3;
     uintptr_t rflags3;
     uintptr_t pml4;
+    uintptr_t xsave;
     // End of area used by asembly
 
     obj::thread* thread;
@@ -182,6 +183,9 @@ private:
     /// \arg rsp0    The existing kernel stack rsp, 0 for none
     thread(process &parent, uint8_t pri, uintptr_t rsp0 = 0);
 
+    /// Set up the XSAVE saved processor state area for this thread
+    void init_xsave_area();
+
     /// Set up a new empty kernel stack for this thread.
     void setup_kernel_stack();
 
diff --git a/src/kernel/task.s b/src/kernel/task.s
index 9941db4..704b242 100644
--- a/src/kernel/task.s
+++ b/src/kernel/task.s
@@ -1,5 +1,7 @@
 %include "tasking.inc"
 
+extern xcr0_val
+
 global task_switch: function hidden (task_switch.end - task_switch)
 task_switch:
 	push rbp
@@ -13,22 +15,33 @@ task_switch:
 	push r15
 
 	; Update previous task's TCB
-	mov rax, [gs:CPU_DATA.tcb]     ; rax: current task TCB
-	mov [rax + TCB.rsp], rsp
+	mov r15, [gs:CPU_DATA.tcb]     ; r15: current task TCB
+	mov [r15 + TCB.rsp], rsp
 
 	; Copy off saved user rsp
 	mov rcx, [gs:CPU_DATA.rsp3]    ; rcx: current task's saved user rsp
-	mov [rax + TCB.rsp3], rcx
+	mov [r15 + TCB.rsp3], rcx
 
 	; Copy off saved user rflags
 	mov rcx, [gs:CPU_DATA.rflags3] ; rcx: current task's saved user rflags
-	mov [rax + TCB.rflags3], rcx
+	mov [r15 + TCB.rflags3], rcx
+
+    ; Save processor extended state
+    mov rcx, [r15 + TCB.xsave]     ; rcx: current task's XSAVE area
+    cmp rcx, 0
+    jz .xsave_done
+
+    mov rax, [rel xcr0_val]
+    mov rdx, rax
+    shl rdx, 32
+    xsave [rcx]
+.xsave_done:
 
 	; Install next task's TCB
 	mov [gs:CPU_DATA.tcb], rdi     ; rdi: next TCB (function param)
 	mov rsp, [rdi + TCB.rsp]       ; next task's stack pointer
-	mov rax, 0x00003fffffffffff
-	and rax, [rdi + TCB.pml4]      ; rax: next task's pml4 (phys portion of address)
+	mov r14, 0x00003fffffffffff
+	and r14, [rdi + TCB.pml4]      ; r14: next task's pml4 (phys portion of address)
 
 	; Update syscall/interrupt rsp
 	mov rcx, [rdi + TCB.rsp0]      ; rcx: top of next task's kernel stack
@@ -41,15 +54,26 @@ task_switch:
 	mov rcx, [rdi + TCB.rsp3]      ; rcx: new task's saved user rsp
 	mov [gs:CPU_DATA.rsp3], rcx
 
+    ; Load processor extended state
+    mov rcx, [rdi + TCB.xsave]     ; rcx: new task's XSAVE area
+    cmp rcx, 0
+    jz .xrstor_done
+
+    mov rax, [rel xcr0_val]
+    mov rdx, rax
+    shl rdx, 32
+    xrstor [rcx]
+.xrstor_done:
+
 	; Update saved user rflags
 	mov rcx, [rdi + TCB.rflags3]   ; rcx: new task's saved user rflags
 	mov [gs:CPU_DATA.rflags3], rcx
 
 	; check if we need to update CR3
 	mov rdx, cr3                   ; rdx: old CR3
-	cmp rax, rdx
+	cmp r14, rdx
 	je .no_cr3
-	mov cr3, rax
+	mov cr3, r14
 .no_cr3:
 
 	pop r15
diff --git a/src/kernel/tasking.inc b/src/kernel/tasking.inc
index f70002c..56922e8 100644
--- a/src/kernel/tasking.inc
+++ b/src/kernel/tasking.inc
@@ -4,6 +4,7 @@ struc TCB
 .rsp3:         resq 1
 .rflags3:      resq 1
 .pml4:         resq 1
+.xsave:        resq 1
 endstruc
 
 struc CPU_DATA
diff --git a/src/kernel/xsave.cpp b/src/kernel/xsave.cpp
new file mode 100644
index 0000000..3c6470a
--- /dev/null
+++ b/src/kernel/xsave.cpp
@@ -0,0 +1,30 @@
+#include <stdint.h>
+#include <cpu/cpu_id.h>
+
+#include "cpu.h"
+#include "xsave.h"
+
+uint64_t xcr0_val = 0;
+static size_t xsave_size_val = 0;
+const size_t &xsave_size = xsave_size_val;
+
+void
+xsave_init()
+{
+    cpu::cpu_id cpuid;
+    const auto regs = cpuid.get(0x0d);
+    const uint64_t cpu_supported = 
+        (static_cast<uint64_t>(regs.edx) << 32) |
+        static_cast<uint64_t>(regs.eax);
+
+    xcr0_val = static_cast<uint64_t>(xcr0::J6_SUPPORTED) & cpu_supported;
+    xsave_size_val = regs.ebx;
+}
+
+void
+xsave_enable()
+{
+    const uint64_t rax = (xcr0_val & 0xFFFFFFFF);
+    const uint64_t rdx = (xcr0_val >> 32);
+    asm volatile ( "xsetbv" :: "c"(0), "d"(xcr0_val >> 32), "a"(xcr0_val) );
+}
diff --git a/src/kernel/xsave.h b/src/kernel/xsave.h
new file mode 100644
index 0000000..1520126
--- /dev/null
+++ b/src/kernel/xsave.h
@@ -0,0 +1,10 @@
+#pragma once
+/// \file xsave.h 
+/// XSAVE operations
+
+#include <stddef.h>
+
+extern const size_t &xsave_size;
+
+void xsave_init();
+void xsave_enable();
diff --git a/src/user/testapp/main.cpp b/src/user/testapp/main.cpp
index 576313b..bf655b9 100644
--- a/src/user/testapp/main.cpp
+++ b/src/user/testapp/main.cpp
@@ -5,6 +5,7 @@
 #include <j6/errors.h>
 #include <j6/flags.h>
 #include <j6/syscalls.h>
+#include <j6/syslog.hh>
 #include <j6/thread.hh>
 #include <j6/types.h>
 
@@ -17,11 +18,42 @@ extern j6_handle_t __handle_self;
 constexpr uintptr_t stack_top = 0xf80000000;
 uint32_t flipflop = 0;
 
+bool
+test_floats()
+{
+    static constexpr int len = 30;
+    double as[len];
+    double bs[len];
+
+    double orig = 345.72;
+    double mult = 3.21;
+    for (int i = 0; i < len * 100; ++i) {
+        int idx = i % len;
+        as[idx] = orig * idx;
+    }
+    for (int i = 0; i < len * 100; ++i) {
+        int idx = i % len;
+        bs[idx] = as[idx] * mult;
+    }
+
+    for (int i = 0; i < len; ++i) {
+        if (bs[i] != orig * i * mult) {
+            j6::syslog("ERROR: floating point discrepency");
+            return false;
+        }
+    }
+
+    return true;
+}
+
 void
 thread_proc(void* channelp)
 {
     j6_log("sub thread starting");
 
+    for (int i = 0; i < 100; ++i)
+        if (!test_floats()) break;
+
     j6::channel *chan = reinterpret_cast<j6::channel*>(channelp);
 
     char buffer[512];
@@ -86,6 +118,9 @@ main(int argc, const char **argv)
 
     j6_log("main thread created sub thread");
 
+    for (int i = 0; i < 100; ++i)
+        if (!test_floats()) break;
+
     char message[] = "MAIN THREAD SUCCESSFULLY CALLED SEND AND RECEIVE IF THIS IS LOWERCASE";
     size_t size = sizeof(message);