[kernel] Add MXCSR handling, mask SIMD exceptions

Now threads inherit their MXCSR (sans exception state bits) SIMD
settings from their creator. By default, all exceptions are masked, and
both "to zero" flags are set.

src/kernel/cpu.cpp

@@ -118,6 +118,19 @@ cpu_early_init(cpu_data *cpu)
         .set(xcr0::SSE);
     set_xcr0(xcr0_val);
 
+    // Set initial floating point state
+    const util::bitset32 mxcsr_val {
+        mxcsr::DAZ,
+        mxcsr::IM,
+        mxcsr::DM,
+        mxcsr::ZM,
+        mxcsr::OM,
+        mxcsr::UM,
+        mxcsr::PM,
+        mxcsr::FTZ,
+    };
+    asm ( "ldmxcsr %0" :: "m"(mxcsr_val) );
+
     // Install the GS base pointint to the cpu_data
     wrmsr(msr::ia32_gs_base, reinterpret_cast<uintptr_t>(cpu));
 }
@@ -234,3 +247,13 @@ cpu_init(cpu_data *cpu, bool bsp)
 
     xsave_enable();
 }
+
+/// Set up initial per-thread CPU state. Called once from initialize_user_cpu in
+/// syscall.s, the first code run after a thread comes out of task_switch for the
+/// very first time.
+extern "C" void
+cpu_initialize_thread_state()
+{
+    const util::bitset32 &mxcsr_val = obj::thread::current().m_mxcsr;
+    asm ( "ldmxcsr %0" :: "m"(mxcsr_val) );
+}

src/kernel/cpu.h

@@ -142,6 +142,8 @@ extern "C" {
     uint64_t set_xcr0(uint64_t val);
 
     cpu_data * _current_gsbase();
+
+    void cpu_initialize_thread_state();
 }
 
 /// Do early initialization of the BSP CPU.

src/kernel/interrupts.cpp

@@ -145,6 +145,30 @@ isr_handler(cpu_state *regs)
         }
         break;
 
+    case isr::isrSIMDFPE: {
+            uint32_t mxcsr = 0;
+            asm volatile ("stmxcsr %0" : "=m"(mxcsr));
+            util::format({message, sizeof(message)},
+                "SIMD Exception; MXCSR[%s%s%s%s%s%s%s%s%s%s%s%s%s%s rc:%d]",
+                (mxcsr & 0x0001) ? " IE" : "",
+                (mxcsr & 0x0002) ? " DE" : "",
+                (mxcsr & 0x0004) ? " ZE" : "",
+                (mxcsr & 0x0008) ? " OE" : "",
+                (mxcsr & 0x0010) ? " UE" : "",
+                (mxcsr & 0x0020) ? " PE" : "",
+                (mxcsr & 0x0040) ? " DAZ" : "",
+                (mxcsr & 0x0080) ? " IM" : "",
+                (mxcsr & 0x0100) ? " DM" : "",
+                (mxcsr & 0x0200) ? " ZM" : "",
+                (mxcsr & 0x0400) ? " OM" : "",
+                (mxcsr & 0x0800) ? " UM" : "",
+                (mxcsr & 0x1000) ? " PM" : "",
+                (mxcsr & 0x8000) ? " FTZ" : "",
+                ((mxcsr >> 13) & 0x3));
+            kassert(false, message, regs);
+        }
+        break;
+ 
     case isr::isrSpurious:
         // No EOI for the spurious interrupt
         return;

src/kernel/objects/thread.cpp

@@ -34,6 +34,15 @@ thread::thread(process &parent, uint8_t pri, uintptr_t rsp0) :
         m_tcb.rsp0 = rsp0;
 
     m_creator = current_cpu().thread;
+
+    asm volatile ( "stmxcsr %0" : "=m"(m_mxcsr) );
+    m_mxcsr
+        .clear(mxcsr::IE)
+        .clear(mxcsr::DE)
+        .clear(mxcsr::ZE)
+        .clear(mxcsr::OE)
+        .clear(mxcsr::UE)
+        .clear(mxcsr::PE);
 }
 
 thread::~thread()

src/kernel/objects/thread.h

@@ -7,10 +7,10 @@
 #include <util/linked_list.h>
 #include <util/spinlock.h>
 
+#include "cpu.h"
 #include "objects/kobject.h"
 #include "wait_queue.h"
 
-struct cpu_data;
 struct page_table;
 
 namespace obj {
@@ -176,6 +176,7 @@ private:
     thread(const thread &other) = delete;
     thread(const thread &&other) = delete;
     friend class process;
+    friend void ::cpu_initialize_thread_state();
 
     /// Constructor. Used when a kernel stack already exists.
     /// \arg parent  The process which owns this thread
@@ -195,6 +196,7 @@ private:
     thread *m_creator;
 
     state m_state;
+    util::bitset32 m_mxcsr;
 
     uint64_t m_wake_value;
     uint64_t m_wake_timeout;

src/kernel/syscall.s

@@ -16,10 +16,11 @@ STAR_HIGH  equ \
 ; IA32_FMASK - Mask off interrupts in syscalls
 FMASK_VAL  equ 0x200
 
-extern __counter_syscall_enter
+extern __counter_syscall_enter          ; syscall.cpp.cog
-extern __counter_syscall_sysret
+extern __counter_syscall_sysret         ;
-extern syscall_registry
+extern syscall_registry                 ;
-extern syscall_invalid
+extern syscall_invalid                  ;
+extern cpu_initialize_thread_state      ; cpu.cpp
 
 
 global syscall_handler_prelude: function hidden (syscall_handler_prelude.end - syscall_handler_prelude)
@@ -79,6 +80,8 @@ syscall_handler_prelude:
 
 global initialize_user_cpu: function hidden (initialize_user_cpu.end - initialize_user_cpu)
 initialize_user_cpu:
+    call cpu_initialize_thread_state
+
     mov rax, 0xaaaaaaaa
     mov rdx, 0xdddddddd
     mov r8,  0x08080808

src/libraries/util/util/bitset.h

@@ -66,8 +66,6 @@ namespace {
 template <>
 class bitset<64>
 {
-    static constexpr unsigned num_elems = 1;
-
     template <typename T>
     static constexpr uint64_t bit_or(T b) { return 1ull << uint64_t(b); }
 
@@ -127,8 +125,6 @@ private:
 template <>
 class bitset<32>
 {
-    static constexpr unsigned num_elems = 1;
-
     template <typename T>
     static constexpr uint32_t bit_or(T b) { return 1u << uint32_t(b); }