koffi 1.1.5 → 1.2.0-alpha.3

package/src/abi_arm32_fwd.S

@@ -1,31 +1,32 @@
-// This program is free software: you can redistribute it and/or modify
-// it under the terms of the GNU Affero General Public License as published by
-// the Free Software Foundation, either version 3 of the License, or
-// (at your option) any later version.
-//
-// This program is distributed in the hope that it will be useful,
-// but WITHOUT ANY WARRANTY; without even the implied warranty of
-// MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
-// GNU Affero General Public License for more details.
-//
-// You should have received a copy of the GNU Affero General Public License
-// along with this program. If not, see https://www.gnu.org/licenses/.
+# This program is free software: you can redistribute it and/or modify
+# it under the terms of the GNU Affero General Public License as published by
+# the Free Software Foundation, either version 3 of the License, or
+# (at your option) any later version.
+#
+# This program is distributed in the hope that it will be useful,
+# but WITHOUT ANY WARRANTY; without even the implied warranty of
+# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+# GNU Affero General Public License for more details.
+#
+# You should have received a copy of the GNU Affero General Public License
+# along with this program. If not, see https://www.gnu.org/licenses/.
 
 .syntax unified
 
-// These three are the same, but they differ (in the C side) by their return type.
-// Unlike the three next functions, these ones don't forward XMM argument registers.
+# These three are the same, but they differ (in the C side) by their return type.
+# Unlike the three next functions, these ones don't forward XMM argument registers.
 .global ForwardCallGG
 .global ForwardCallF
 .global ForwardCallDDDD
 
-// The X variants are slightly slower, and are used when XMM arguments must be forwarded.
+# The X variants are slightly slower, and are used when XMM arguments must be forwarded.
 .global ForwardCallXGG
 .global ForwardCallXF
 .global ForwardCallXDDDD
 
-// Copy function pointer to r12, in order to save it through argument forwarding.
-// Save RSP in fp (non-volatile), and use carefully assembled stack provided by caller.
+# Copy function pointer to r12, in order to save it through argument forwarding.
+# Also make a copy of the SP to CallData::old_sp because the callback system might need it.
+# Save RSP in fp (non-volatile), and use carefully assembled stack provided by caller.
 .macro prologue
     .cfi_startproc
     push {fp, lr}
@@ -34,14 +35,15 @@
     .cfi_offset 14, 8
     mov fp, sp
     .cfi_def_cfa fp, 8
+    str fp, [r2, 0]
     mov r12, r0
     mov sp, r1
     add sp, sp, #80
 .endm
 
-// Call native function.
-// Once done, restore normal stack pointer and return.
-// The return value is passed untouched through r0, r1, and or FP registers
+# Call native function.
+# Once done, restore normal stack pointer and return.
+# The return value is passed untouched through r0, r1, and or FP registers
 .macro epilogue
     blx r12
     mov sp, fp
@@ -54,7 +56,7 @@
     .cfi_endproc
 .endm
 
-// Prepare general purpose argument registers from array passed by caller.
+# Prepare general purpose argument registers from array passed by caller.
 .macro forward_int
     ldr r3, [r1, 76]
     ldr r2, [r1, 72]
@@ -62,7 +64,7 @@
     ldr r1, [r1, 68]
 .endm
 
-// Prepare vector argument registers from array passed by caller.
+# Prepare vector argument registers from array passed by caller.
 .macro forward_vec
     vldr d7, [r1, 56]
     vldr d6, [r1, 48]
@@ -106,3 +108,216 @@ ForwardCallXDDDD:
     forward_vec
     forward_int
     epilogue
+
+# Callback trampolines
+# ----------------------------
+
+.global Trampoline0
+.global Trampoline1
+.global Trampoline2
+.global Trampoline3
+.global Trampoline4
+.global Trampoline5
+.global Trampoline6
+.global Trampoline7
+.global Trampoline8
+.global Trampoline9
+.global Trampoline10
+.global Trampoline11
+.global Trampoline12
+.global Trampoline13
+.global Trampoline14
+.global Trampoline15
+.global TrampolineX0
+.global TrampolineX1
+.global TrampolineX2
+.global TrampolineX3
+.global TrampolineX4
+.global TrampolineX5
+.global TrampolineX6
+.global TrampolineX7
+.global TrampolineX8
+.global TrampolineX9
+.global TrampolineX10
+.global TrampolineX11
+.global TrampolineX12
+.global TrampolineX13
+.global TrampolineX14
+.global TrampolineX15
+.global RelayCallBack
+.global CallSwitchStack
+
+# First, make a copy of the GPR argument registers (r0 to r7).
+# Then call the C function RelayCallBack with the following arguments:
+# static trampoline ID, a pointer to the saved GPR array, a pointer to the stack
+# arguments of this call, and a pointer to a struct that will contain the result registers.
+# After the call, simply load these registers from the output struct.
+.macro trampoline id
+    .cfi_startproc
+    push {fp, lr}
+    .cfi_def_cfa sp, 8
+    .cfi_offset 11, 4
+    .cfi_offset 14, 8
+    sub sp, sp, #120
+    .cfi_def_cfa sp, 128
+    str r0, [sp, 64]
+    str r1, [sp, 68]
+    str r2, [sp, 72]
+    str r3, [sp, 76]
+    mov r0, \id
+    mov r1, sp
+    add r2, sp, #128
+    add r3, sp, #80
+    bl RelayCallBack
+    ldr r0, [sp, 80]
+    ldr r1, [sp, 84]
+    add sp, sp, #120
+    .cfi_def_cfa sp, 8
+    pop {fp, lr}
+    .cfi_def_cfa sp, 0
+    .cfi_restore 11
+    .cfi_restore 14
+    bx lr
+    .cfi_endproc
+.endm
+
+# Same thing, but also forwards the floating-point argument registers and loads them at the end.
+.macro trampoline_vec id
+    .cfi_startproc
+    push {fp, lr}
+    .cfi_def_cfa sp, 8
+    .cfi_offset 11, 4
+    .cfi_offset 14, 8
+    sub sp, sp, #120
+    .cfi_def_cfa sp, 128
+    str r0, [sp, 64]
+    str r1, [sp, 68]
+    str r2, [sp, 72]
+    str r3, [sp, 76]
+    vstr d0, [sp, 0]
+    vstr d1, [sp, 8]
+    vstr d2, [sp, 16]
+    vstr d3, [sp, 24]
+    vstr d4, [sp, 32]
+    vstr d5, [sp, 40]
+    vstr d6, [sp, 48]
+    vstr d7, [sp, 56]
+    mov r0, \id
+    mov r1, sp
+    add r2, sp, #128
+    add r3, sp, #80
+    bl RelayCallBack
+    ldr r0, [sp, 80]
+    ldr r1, [sp, 84]
+    vldr d0, [sp, 88]
+    vldr d1, [sp, 96]
+    vldr d2, [sp, 104]
+    vldr d3, [sp, 112]
+    add sp, sp, #120
+    .cfi_def_cfa sp, 8
+    pop {fp, lr}
+    .cfi_def_cfa sp, 0
+    .cfi_restore 11
+    .cfi_restore 14
+    bx lr
+    .cfi_endproc
+.endm
+
+Trampoline0:
+    trampoline 0
+Trampoline1:
+    trampoline 1
+Trampoline2:
+    trampoline 2
+Trampoline3:
+    trampoline 3
+Trampoline4:
+    trampoline 4
+Trampoline5:
+    trampoline 5
+Trampoline6:
+    trampoline 6
+Trampoline7:
+    trampoline 7
+Trampoline8:
+    trampoline 8
+Trampoline9:
+    trampoline 9
+Trampoline10:
+    trampoline 10
+Trampoline11:
+    trampoline 11
+Trampoline12:
+    trampoline 12
+Trampoline13:
+    trampoline 13
+Trampoline14:
+    trampoline 14
+Trampoline15:
+    trampoline 15
+
+TrampolineX0:
+    trampoline_vec 0
+TrampolineX1:
+    trampoline_vec 1
+TrampolineX2:
+    trampoline_vec 2
+TrampolineX3:
+    trampoline_vec 3
+TrampolineX4:
+    trampoline_vec 4
+TrampolineX5:
+    trampoline_vec 5
+TrampolineX6:
+    trampoline_vec 6
+TrampolineX7:
+    trampoline_vec 7
+TrampolineX8:
+    trampoline_vec 8
+TrampolineX9:
+    trampoline_vec 9
+TrampolineX10:
+    trampoline_vec 10
+TrampolineX11:
+    trampoline_vec 11
+TrampolineX12:
+    trampoline_vec 12
+TrampolineX13:
+    trampoline_vec 13
+TrampolineX14:
+    trampoline_vec 14
+TrampolineX15:
+    trampoline_vec 15
+
+# When a callback is relayed, Koffi will call into Node.js and V8 to execute Javascript.
+# The propblem is that we're still running on the separate Koffi stack, and V8 will
+# preobably misdetect this as a "stack overflow". We have to restore the old
+# stack pointer, call Node.js/V8 and go back to ours.
+# The first three parameters (r0, r1, r2) are passed through untouched.
+CallSwitchStack:
+    .cfi_startproc
+    push {fp, lr}
+    .cfi_def_cfa sp, 8
+    push {r4, r5}
+    .cfi_def_cfa sp, 16
+    .cfi_offset 11, 4
+    .cfi_offset 14, 8
+    mov fp, sp
+    ldr r4, [sp, 16]
+    ldr r5, [r4, 0]
+    sub r5, sp, r5
+    and r5, r5, #-16
+    str r5, [r4, 4]
+    ldr r4, [sp, 20]
+    mov sp, r3
+    blx r4
+    mov sp, fp
+    .cfi_def_cfa sp, 16
+    pop {r4, r5}
+    .cfi_def_cfa sp, 8
+    pop {fp, lr}
+    .cfi_def_cfa sp, 0
+    .cfi_restore 11
+    .cfi_restore 14
+    bx lr
+    .cfi_endproc