[kernel] Set mxcsr and xcr0 in cpu_early_init

There are some SSE instructions (moveaps, moveups) in userland code that
QEMU software emulation seems to be fine with but generate `#UD` on KVM.
So let's finally get floating-point support working. This is the first
step, just setting the control regs to try to fix that error.

src/kernel/cpu.cpp

@@ -63,6 +63,13 @@ cpu_early_init(cpu_data *cpu)
     cpu->idt->install();
     cpu->gdt->install();
 
+    util::bitset64 cr0_val = 0;
+    asm ("mov %%cr0, %0" : "=r"(cr0_val));
+    cr0_val
+        .set(cr0::WP)
+        .clear(cr0::CD);
+    asm volatile ( "mov %0, %%cr0" :: "r" (cr0_val) );
+
     util::bitset64 cr4_val = 0;
     asm ("mov %%cr4, %0" : "=r"(cr4_val));
     cr4_val
@@ -79,6 +86,11 @@ cpu_early_init(cpu_data *cpu)
         .set(efer::NXE);
     wrmsr(msr::ia32_efer, efer_val);
 
+    util::bitset64 xcr0_val = get_xcr0();
+    xcr0_val
+        .set(xcr0::SSE);
+    set_xcr0(xcr0_val);
+
     // Install the GS base pointint to the cpu_data
     wrmsr(msr::ia32_gs_base, reinterpret_cast<uintptr_t>(cpu));
 }
@@ -119,8 +131,11 @@ bsp_late_init()
     asm ("mov %%cr0, %0" : "=r"(cr0v));
     asm ("mov %%cr4, %0" : "=r"(cr4v));
 
+    uint32_t mxcsrv = get_mxcsr();
+    uint64_t xcr0v = get_xcr0();
+
     uint64_t efer = rdmsr(msr::ia32_efer);
-    log::spam(logs::boot, "Control regs: cr0:%lx cr4:%lx efer:%lx", cr0v, cr4v, efer);
+    log::spam(logs::boot, "Control regs: cr0:%lx cr4:%lx efer:%lx mxcsr:%x xcr0:%x", cr0v, cr4v, efer, mxcsrv, xcr0v);
 }
 
 cpu_data *

src/kernel/cpu.h

@@ -17,9 +17,12 @@ enum class cr0
 {
     PE =  0,    // Protected mode enable
     MP =  1,    // Monitor co-processor
+    EM =  2,    // (FPU) Emulation
+    TS =  3,    // Task switched
     ET =  4,    // Extension type
     NE =  5,    // Numeric error
     WP = 16,    // (ring 0) Write protect
+    CD = 30,    // Cache disable
     PG = 31,    // Paging
 };
 
@@ -58,6 +61,26 @@ enum class efer
     FFXSR = 14, // Fast FXSAVE
 };
 
+enum class mxcsr
+{
+    IE  =  0, // Invalid operation flag
+    DE  =  1, // Denormal flag
+    ZE  =  2, // Divide by zero flag
+    OE  =  3, // Overflow flag
+    UE  =  4, // Underflow flag
+    PE  =  5, // Precision flag
+    DAZ =  6, // Denormals are zero
+    IM  =  7, // Invalid operation mask
+    DM  =  8, // Denormal mask
+    ZM  =  9, // Divide by zero mask
+    OM  = 10, // Overflow mask
+    UM  = 11, // Underflow mask
+    PM  = 12, // Precision mask
+    RC0 = 13, // Rounding control bit 0
+    RC1 = 14, // Rounding control bit 1
+    FTZ = 15, // Flush to zero
+};
+
 struct cpu_state
 {
     uint64_t r15, r14, r13, r12, r11, r10, r9,  r8;
@@ -107,7 +130,15 @@ struct cpu_data
     panic_data *panic;
 };
 
-extern "C" cpu_data * _current_gsbase();
+extern "C" {
+    uint32_t get_mxcsr();
+    uint32_t set_mxcsr(uint32_t val);
+
+    uint64_t get_xcr0();
+    uint64_t set_xcr0(uint64_t val);
+
+    cpu_data * _current_gsbase();
+}
 
 /// Do early initialization of the BSP CPU.
 /// \returns  A pointer to the BSP cpu_data structure

src/kernel/cpu.s

@@ -1,3 +1,34 @@
+global get_mxcsr: function hidden (get_mxcsr.end - get_mxcsr)
+get_mxcsr:
+    push 0
+    stmxcsr [rsp]
+    pop rax
+    ret
+.end:
+
+global set_mxcsr: function hidden (set_mxcsr.end - set_mxcsr)
+set_mxcsr:
+    push rdi
+    ldmxcsr [rsp]
+    pop rax
+    ret
+.end:
+
+global get_xcr0: function hidden (get_xcr0.end - get_xcr0)
+get_xcr0:
+    xor rcx, rcx ; there is no dana there is only xcr0
+    xgetbv
+    ret ; technically edx has the high 32 bits, but bits 10+ are reserved
+.end:
+
+global set_xcr0: function hidden (set_xcr0.end - set_xcr0)
+set_xcr0:
+    xor rcx, rcx ; there is no dana there is only xcr0
+    mov rax, rdi ; technically edx should be or'd into the high bits, but xcr0 bits 10+ are resereved
+    xsetbv
+    ret
+.end:
+
 global get_rsp: function hidden (get_rsp.end - get_rsp)
 get_rsp:
 	mov rax, rsp

src/kernel/kernel.module

@@ -16,7 +16,7 @@ kernel = module("kernel",
         "clock.cpp",
         "cpprt.cpp",
         "cpu.cpp",
-        "debug.s",
+        "cpu.s",
         "device_manager.cpp",
         "frame_allocator.cpp",
         "gdt.cpp",