py/persistentcode: Add ability to relocate loaded native code.

Implements text, rodata and bss generalised relocations, as well as generic
qstr-object linking.  This allows importing dynamic native modules on all
supported architectures in a unified way.

ports/esp32/main.c

@@ -47,6 +47,7 @@
 #include "py/nlr.h"
 #include "py/compile.h"
 #include "py/runtime.h"
+#include "py/persistentcode.h"
 #include "py/repl.h"
 #include "py/gc.h"
 #include "py/mphal.h"
@@ -173,12 +174,15 @@ void mbedtls_debug_set_threshold(int threshold) {
     (void)threshold;
 }
 
-void *esp_native_code_commit(void *buf, size_t len) {
+void *esp_native_code_commit(void *buf, size_t len, void *reloc) {
     len = (len + 3) & ~3;
     uint32_t *p = heap_caps_malloc(len, MALLOC_CAP_EXEC);
     if (p == NULL) {
         m_malloc_fail(len);
     }
+    if (reloc) {
+        mp_native_relocate(reloc, buf, (uintptr_t)p);
+    }
     memcpy(p, buf, len);
     return p;
 }

ports/esp32/mpconfigport.h

@@ -21,8 +21,6 @@
 // emitters
 #define MICROPY_PERSISTENT_CODE_LOAD        (1)
 #define MICROPY_EMIT_XTENSAWIN              (1)
-void *esp_native_code_commit(void*, size_t);
-#define MP_PLAT_COMMIT_EXEC(buf, len) esp_native_code_commit(buf, len)
 
 // compiler configuration
 #define MICROPY_COMP_MODULE_CONST           (1)
@@ -225,6 +223,8 @@ struct mp_bluetooth_nimble_root_pointers_t;
 #define BYTES_PER_WORD (4)
 #define MICROPY_MAKE_POINTER_CALLABLE(p) ((void*)((mp_uint_t)(p)))
 #define MP_PLAT_PRINT_STRN(str, len) mp_hal_stdout_tx_strn_cooked(str, len)
+void *esp_native_code_commit(void*, size_t, void*);
+#define MP_PLAT_COMMIT_EXEC(buf, len, reloc) esp_native_code_commit(buf, len, reloc)
 #define MP_SSIZE_MAX (0x7fffffff)
 
 // Note: these "critical nested" macros do not ensure cross-CPU exclusion,

ports/esp8266/modesp.c

@@ -28,6 +28,7 @@
 
 #include "py/gc.h"
 #include "py/runtime.h"
+#include "py/persistentcode.h"
 #include "py/mperrno.h"
 #include "py/mphal.h"
 #include "drivers/dht/dht.h"
@@ -282,7 +283,7 @@ void esp_native_code_init(void) {
     esp_native_code_erased = 0;
 }
 
-void *esp_native_code_commit(void *buf, size_t len) {
+void *esp_native_code_commit(void *buf, size_t len, void *reloc) {
     //printf("COMMIT(buf=%p, len=%u, start=%08x, cur=%08x, end=%08x, erased=%08x)\n", buf, len, esp_native_code_start, esp_native_code_cur, esp_native_code_end, esp_native_code_erased);
 
     len = (len + 3) & ~3;
@@ -294,6 +295,14 @@ void *esp_native_code_commit(void *buf, size_t len) {
     void *dest;
     if (esp_native_code_location == ESP_NATIVE_CODE_IRAM1) {
         dest = (void*)esp_native_code_cur;
+    } else {
+        dest = (void*)(FLASH_START + esp_native_code_cur);
+    }
+    if (reloc) {
+        mp_native_relocate(reloc, buf, (uintptr_t)dest);
+    }
+
+    if (esp_native_code_location == ESP_NATIVE_CODE_IRAM1) {
         memcpy(dest, buf, len);
     } else {
         SpiFlashOpResult res;
@@ -313,7 +322,6 @@ void *esp_native_code_commit(void *buf, size_t len) {
         if (res != SPI_FLASH_RESULT_OK) {
             mp_raise_OSError(res == SPI_FLASH_RESULT_TIMEOUT ? MP_ETIMEDOUT : MP_EIO);
         }
-        dest = (void*)(FLASH_START + esp_native_code_cur);
     }
 
     esp_native_code_cur += len;

ports/esp8266/mpconfigport.h

@@ -131,8 +131,8 @@ typedef uint32_t sys_prot_t; // for modlwip
 #include <sys/types.h>
 
 #define MP_PLAT_PRINT_STRN(str, len) mp_hal_stdout_tx_strn_cooked(str, len)
-void *esp_native_code_commit(void*, size_t);
-#define MP_PLAT_COMMIT_EXEC(buf, len) esp_native_code_commit(buf, len)
+void *esp_native_code_commit(void*, size_t, void*);
+#define MP_PLAT_COMMIT_EXEC(buf, len, reloc) esp_native_code_commit(buf, len, reloc)
 
 // printer for debugging output, goes to UART only
 extern const struct _mp_print_t mp_debug_print;

py/asmbase.h

@@ -60,7 +60,7 @@ static inline size_t mp_asm_base_get_code_size(mp_asm_base_t *as) {
 
 static inline void *mp_asm_base_get_code(mp_asm_base_t *as) {
     #if defined(MP_PLAT_COMMIT_EXEC)
-    return MP_PLAT_COMMIT_EXEC(as->code_base, as->code_size);
+    return MP_PLAT_COMMIT_EXEC(as->code_base, as->code_size, NULL);
     #else
     return as->code_base;
     #endif

py/bc.h

@@ -74,7 +74,7 @@
 #define MP_BC_PRELUDE_SIG_ENCODE(S, E, scope, out_byte, out_env) \
 do {                                                            \
     /*// Get values to store in prelude */                      \
-    size_t F = scope->scope_flags & 0x0f; /* only need to store lower 4 flag bits */ \
+    size_t F = scope->scope_flags & MP_SCOPE_FLAG_ALL_SIG;      \
     size_t A = scope->num_pos_args;                             \
     size_t K = scope->num_kwonly_args;                          \
     size_t D = scope->num_def_pos_args;                         \

py/persistentcode.c

@@ -33,6 +33,7 @@
 #include "py/emitglue.h"
 #include "py/persistentcode.h"
 #include "py/bc0.h"
+#include "py/objstr.h"
 
 #if MICROPY_PERSISTENT_CODE_LOAD || MICROPY_PERSISTENT_CODE_SAVE
 
@@ -145,8 +146,16 @@ STATIC byte *extract_prelude(const byte **ip, bytecode_prelude_t *prelude) {
 
 #include "py/parsenum.h"
 
+STATIC int read_byte(mp_reader_t *reader);
+STATIC size_t read_uint(mp_reader_t *reader, byte **out);
+
 #if MICROPY_EMIT_MACHINE_CODE
 
+typedef struct _reloc_info_t {
+    mp_reader_t *reader;
+    mp_uint_t *const_table;
+} reloc_info_t;
+
 #if MICROPY_EMIT_THUMB
 STATIC void asm_thumb_rewrite_mov(uint8_t *pc, uint16_t val) {
     // high part
@@ -179,6 +188,52 @@ STATIC void arch_link_qstr(uint8_t *pc, bool is_obj, qstr qst) {
     #endif
 }
 
+void mp_native_relocate(void *ri_in, uint8_t *text, uintptr_t reloc_text) {
+    // Relocate native code
+    reloc_info_t *ri = ri_in;
+    uint8_t op;
+    uintptr_t *addr_to_adjust = NULL;
+    while ((op = read_byte(ri->reader)) != 0xff) {
+        if (op & 1) {
+            // Point to new location to make adjustments
+            size_t addr = read_uint(ri->reader, NULL);
+            if ((addr & 1) == 0) {
+                // Point to somewhere in text
+                addr_to_adjust = &((uintptr_t*)text)[addr >> 1];
+            } else {
+                // Point to somewhere in rodata
+                addr_to_adjust = &((uintptr_t*)ri->const_table[1])[addr >> 1];
+            }
+        }
+        op >>= 1;
+        uintptr_t dest;
+        size_t n = 1;
+        if (op <= 5) {
+            if (op & 1) {
+                // Read in number of adjustments to make
+                n = read_uint(ri->reader, NULL);
+            }
+            op >>= 1;
+            if (op == 0) {
+                // Destination is text
+                dest = reloc_text;
+            } else {
+                // Destination is rodata (op=1) or bss (op=1 if no rodata, else op=2)
+                dest = ri->const_table[op];
+            }
+        } else if (op == 6) {
+            // Destination is mp_fun_table itself
+            dest = (uintptr_t)&mp_fun_table;
+        } else {
+            // Destination is an entry in mp_fun_table
+            dest = ((uintptr_t*)&mp_fun_table)[op - 7];
+        }
+        while (n--) {
+            *addr_to_adjust++ += dest;
+        }
+    }
+}
+
 #endif
 
 STATIC int read_byte(mp_reader_t *reader) {
@@ -340,6 +395,9 @@ STATIC mp_raw_code_t *load_raw_code(mp_reader_t *reader, qstr_window_t *qw) {
                     // Generic 16-bit link
                     dest[0] = qst & 0xff;
                     dest[1] = (qst >> 8) & 0xff;
+                } else if ((off & 3) == 3) {
+                    // Generic, aligned qstr-object link
+                    *(mp_obj_t*)dest = MP_OBJ_NEW_QSTR(qst);
                 } else {
                     // Architecture-specific link
                     arch_link_qstr(dest, (off & 3) == 2, qst);
@@ -423,8 +481,26 @@ STATIC mp_raw_code_t *load_raw_code(mp_reader_t *reader, qstr_window_t *qw) {
 
     #if MICROPY_EMIT_MACHINE_CODE
     } else {
+        mp_uint_t *ct = &const_table[1];
+        if (prelude.scope_flags & MP_SCOPE_FLAG_VIPERRODATA) {
+            size_t size = read_uint(reader, NULL);
+            uint8_t *rodata = m_new(uint8_t, size);
+            read_bytes(reader, rodata, size);
+            *ct++ = (uintptr_t)rodata;
+        }
+        if (prelude.scope_flags & MP_SCOPE_FLAG_VIPERBSS) {
+            size_t size = read_uint(reader, NULL);
+            uint8_t *bss = m_new0(uint8_t, size);
+            *ct++ = (uintptr_t)bss;
+        }
+        reloc_info_t ri = {reader, const_table};
         #if defined(MP_PLAT_COMMIT_EXEC)
-        fun_data = MP_PLAT_COMMIT_EXEC(fun_data, fun_data_len);
+        void *opt_ri = (prelude.scope_flags & MP_SCOPE_FLAG_VIPERRELOC) ? &ri : NULL;
+        fun_data = MP_PLAT_COMMIT_EXEC(fun_data, fun_data_len, opt_ri);
+        #else
+        if (prelude.scope_flags & MP_SCOPE_FLAG_VIPERRELOC) {
+            mp_native_relocate(&ri, fun_data, (uintptr_t)fun_data);
+        }
         #endif
 
         mp_emit_glue_assign_native(rc, kind,
@@ -624,7 +700,7 @@ STATIC void save_raw_code(mp_print_t *print, mp_raw_code_t *rc, qstr_window_t *q
             save_prelude_qstrs(print, qstr_window, ip_info);
         } else {
             // Save basic scope info for viper and asm
-            mp_print_uint(print, rc->scope_flags);
+            mp_print_uint(print, rc->scope_flags & MP_SCOPE_FLAG_ALL_SIG);
             prelude.n_pos_args = 0;
             prelude.n_kwonly_args = 0;
             if (rc->kind == MP_CODE_NATIVE_ASM) {

py/persistentcode.h

@@ -95,4 +95,6 @@ mp_raw_code_t *mp_raw_code_load_file(const char *filename);
 void mp_raw_code_save(mp_raw_code_t *rc, mp_print_t *print);
 void mp_raw_code_save_file(mp_raw_code_t *rc, const char *filename);
 
+void mp_native_relocate(void *reloc, uint8_t *text, uintptr_t reloc_text);
+
 #endif // MICROPY_INCLUDED_PY_PERSISTENTCODE_H

py/runtime0.h

@@ -28,13 +28,17 @@
 
 // The first four must fit in 8 bits, see emitbc.c
 // The remaining must fit in 16 bits, see scope.h
+#define MP_SCOPE_FLAG_ALL_SIG      (0x0f)
 #define MP_SCOPE_FLAG_GENERATOR    (0x01)
 #define MP_SCOPE_FLAG_VARKEYWORDS  (0x02)
 #define MP_SCOPE_FLAG_VARARGS      (0x04)
 #define MP_SCOPE_FLAG_DEFKWARGS    (0x08)
 #define MP_SCOPE_FLAG_REFGLOBALS   (0x10) // used only if native emitter enabled
 #define MP_SCOPE_FLAG_HASCONSTS    (0x20) // used only if native emitter enabled
-#define MP_SCOPE_FLAG_VIPERRET_POS    (6) // 3 bits used for viper return type
+#define MP_SCOPE_FLAG_VIPERRET_POS    (6) // 3 bits used for viper return type, to pass from compiler to native emitter
+#define MP_SCOPE_FLAG_VIPERRELOC   (0x10) // used only when loading viper from .mpy
+#define MP_SCOPE_FLAG_VIPERRODATA  (0x20) // used only when loading viper from .mpy
+#define MP_SCOPE_FLAG_VIPERBSS     (0x40) // used only when loading viper from .mpy
 
 // types for native (viper) function signature
 #define MP_NATIVE_TYPE_OBJ  (0x00)