redo boot around an explicit system volume and fixed kernel/HAL images, redo volumes and filesystem paths around capabilities instead of drive numbers and kernel CWD state, add a capability/security model with derived and revocable handles and security contexts, add port IPC and rssrv to manage the userspace resource namespace and publish the system volume, move process, dll and path loading to userspace and add process builder syscalls, init now supervises rssrv and the session process, redo the syscall ABI around request structs and four arguments, add a volume manager and proper partition arrival/removal handling so FAT and NTFS mount dynamically, clean up the HAL/DDK boundary so drivers don't depend on HAL internals, save fpu state properly now, the serial driver doesn't hang on a missing port, fix AHCI startup races, fix timeout accounting, fix shared mmap writeback, more smoke tests

code/kernel/ke_syscall.asm

@@ -112,22 +112,18 @@ KeSyscallEntry PROC
     call    KiEnterTrapFrame
     add     rsp, 32
 
-    ; Re-order original user arguments into KeSyscallDispatch(nr,a0..a5).
+    ; Re-order original user arguments into KeSyscallDispatch(nr,a0..a3).
     ; Original registers were saved in the trap frame before any C calls.
-    sub     rsp, 64
+    sub     rsp, 48
     mov     rax, [rbx + TF_R9]
     mov     [rsp + 32], rax     ; a3
-    mov     rax, [rbx + TF_R10]
-    mov     [rsp + 40], rax     ; a4
-    mov     rax, [rbx + TF_R11]
-    mov     [rsp + 48], rax     ; a5
 
     mov     r9,  [rbx + TF_R8]  ; a2
     mov     r8,  [rbx + TF_Rdx] ; a1
     mov     rdx, [rbx + TF_Rcx] ; a0
     mov     rcx, [rbx + TF_Rax] ; nr
     call    KeSyscallDispatch
-    add     rsp, 64
+    add     rsp, 48
 
     mov     [rbx + TF_Rax], rax
 
@@ -199,7 +195,22 @@ KF_R13       EQU 0A8h
 KF_R14       EQU 0B0h
 KF_R15       EQU 0B8h
 
-FAULT_CALL_SHADOW EQU 28h
+; Asynchronous kernel entry must preserve the full interrupted x87/SSE state.
+; MS x64 calls may freely clobber XMM0-XMM5, and an interrupt or fault can
+; arrive between any two user or kernel instructions rather than at a call ABI.
+; Reserve one 512-byte FXSAVE image plus alignment slack on the interrupted stack.
+ASYNC_FPU_SCRATCH EQU 528
+
+ASYNC_FPU_SAVE MACRO
+    sub  rsp, ASYNC_FPU_SCRATCH
+    lea  r12, [rsp + 15]
+    and  r12, -16
+    fxsave64 [r12]
+ENDM
+
+ASYNC_FPU_RESTORE MACRO
+    fxrstor64 [r12]
+ENDM
 
 FAULT_RESUME_NOERR MACRO
     LOCAL noerr_skip_user_slots, noerr_no_swapgs, noerr_unhandled_halt
@@ -372,11 +383,13 @@ de_kernel_frame:
     mov  qword ptr [rax + KF_Ss], 0
 
 de_call:
-    and  rsp, -16
-    sub  rsp, FAULT_CALL_SHADOW
     mov  rbx, rax
-    mov  rcx, rax
+    ASYNC_FPU_SAVE
+    and  rsp, -16
+    sub  rsp, 32
+    mov  rcx, rbx
     call KeFaultDispatchFrame
+    ASYNC_FPU_RESTORE
     FAULT_RESUME_NOERR
 KeFaultDE ENDP
 
@@ -440,11 +453,13 @@ gp_kernel_frame:
     mov  qword ptr [rax + KF_Ss], 0
 
 gp_call:
-    and  rsp, -16
-    sub  rsp, FAULT_CALL_SHADOW
     mov  rbx, rax
-    mov  rcx, rax
+    ASYNC_FPU_SAVE
+    and  rsp, -16
+    sub  rsp, 32
+    mov  rcx, rbx
     call KeFaultDispatchFrame
+    ASYNC_FPU_RESTORE
     FAULT_RESUME_ERR
 KeFaultGP ENDP
 
@@ -509,11 +524,13 @@ pf_kernel_frame:
     mov  qword ptr [rax + KF_Ss], 0
 
 pf_call:
-    and  rsp, -16
-    sub  rsp, FAULT_CALL_SHADOW
     mov  rbx, rax               ; Keep the KFAULT_FRAME pointer across C.
-    mov  rcx, rax
+    ASYNC_FPU_SAVE
+    and  rsp, -16
+    sub  rsp, 32
+    mov  rcx, rbx
     call KeFaultDispatchFrame
+    ASYNC_FPU_RESTORE
     FAULT_RESUME_ERR
 KeFaultPF ENDP
 
@@ -579,11 +596,13 @@ noerr_kernel_frame:
     mov  qword ptr [rax + KF_Ss], 0
 
 noerr_call:
-    and  rsp, -16
-    sub  rsp, FAULT_CALL_SHADOW
     mov  rbx, rax
-    mov  rcx, rax
+    ASYNC_FPU_SAVE
+    and  rsp, -16
+    sub  rsp, 32
+    mov  rcx, rbx
     call KeFaultDispatchFrame
+    ASYNC_FPU_RESTORE
     FAULT_RESUME_NOERR
 StubName ENDP
 ENDM
@@ -646,11 +665,13 @@ err_kernel_frame:
     mov  qword ptr [rax + KF_Ss], 0
 
 err_call:
-    and  rsp, -16
-    sub  rsp, FAULT_CALL_SHADOW
     mov  rbx, rax
-    mov  rcx, rax
+    ASYNC_FPU_SAVE
+    and  rsp, -16
+    sub  rsp, 32
+    mov  rcx, rbx
     call KeFaultDispatchFrame
+    ASYNC_FPU_RESTORE
     FAULT_RESUME_ERR
 StubName ENDP
 ENDM
@@ -669,14 +690,22 @@ KeNmiIpiStub PROC
     push    r9
     push    r10
     push    r11
+    push    r12
+    push    r13
+    mov     r13, rsp
 
+    ASYNC_FPU_SAVE
+    and     rsp, -16
     sub     rsp, 32
     call    KeHandleNmiIpi
-    add     rsp, 32
+    ASYNC_FPU_RESTORE
+    mov     rsp, r13
 
     test    eax, eax
     jz      short NmiUnhandled
 
+    pop     r13
+    pop     r12
     pop     r11
     pop     r10
     pop     r9
@@ -687,6 +716,8 @@ KeNmiIpiStub PROC
     iretq
 
 NmiUnhandled:
+    pop     r13
+    pop     r12
     pop     r11
     pop     r10
     pop     r9
@@ -857,7 +888,9 @@ IrqKernelModeFrame:
     mov     qword ptr [rbx + TF_SsSlot], 0
 
 IrqFrameReady:
-    sub     rsp, 40         ; 32-byte shadow space + alignment/local slack
+    ASYNC_FPU_SAVE
+    and     rsp, -16
+    sub     rsp, 32
     mov     rcx, rbx
     call    KiEnterTrapFrame
     call    KiEnterInterrupt
@@ -866,14 +899,15 @@ IrqFrameReady:
     call    KiExitInterrupt
     mov     rcx, rbx
     call    KiExitTrapFrame
-    add     rsp, 40
+    ASYNC_FPU_RESTORE
 
     mov     eax, [rbx + TF_PreviousMode]
     cmp     eax, KTRAP_MODE_USER
     jne     short IrqNoSwapGsOut
     swapgs             ; returning to ring 3: restore user GS/TEB
 IrqNoSwapGsOut:
-    add     rsp, KTRAP_FRAME_SIZE
+    mov     r12, [rbx + TF_R12]
+    lea     rsp, [rbx + KTRAP_FRAME_SIZE]
     pop     rbx
     pop     r11
     pop     r10