ACPR/android_system_core
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

merge intel x86 patches into gingerbread branch

Browse source
This commit is contained in:
David 'Digit' Turner 2011-07-08 18:19:06 +02:00
commit 8f1f5b061f
42 changed files with 4015 additions and 346 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

12
debuggerd/Android.mk
View file

@ -1,11 +1,15 @@
# Copyright 2005 The Android Open Source Project
ifeq ($(TARGET_ARCH),arm)
ifneq ($(filter arm x86,$(TARGET_ARCH)),)
LOCAL_PATH:= $(call my-dir)
include $(CLEAR_VARS)
LOCAL_SRC_FILES:= debuggerd.c getevent.c unwind-arm.c pr-support.c utility.c
LOCAL_SRC_FILES:= debuggerd.c utility.c getevent.c $(TARGET_ARCH)/machine.c $(TARGET_ARCH)/unwind.c symbol_table.c
ifeq ($(TARGET_ARCH),arm)
LOCAL_SRC_FILES += $(TARGET_ARCH)/pr-support.c
endif
LOCAL_CFLAGS := -Wall
LOCAL_MODULE := debuggerd
@ -22,7 +26,7 @@ include $(BUILD_EXECUTABLE)
include $(CLEAR_VARS)
LOCAL_SRC_FILES := crasher.c
LOCAL_SRC_FILES += crashglue.S
LOCAL_SRC_FILES += $(TARGET_ARCH)/crashglue.S
LOCAL_MODULE := crasher
LOCAL_MODULE_PATH := $(TARGET_OUT_OPTIONAL_EXECUTABLES)
LOCAL_MODULE_TAGS := eng
@ -46,4 +50,4 @@ LOCAL_SHARED_LIBRARIES := libcutils libc
include $(BUILD_EXECUTABLE)
endif # ARCH_ARM_HAVE_VFP == true
endif # TARGET_ARCH == arm
endif # arm or x86 in TARGET_ARCH

2
debuggerd/crashglue.S → debuggerd/arm/crashglue.S
View file

@ -2,7 +2,7 @@
.type crash1, %function
.globl crashnostack
.type crashnostack, %function
crash1:
ldr r0, =0xa5a50000
ldr r1, =0xa5a50001

229
debuggerd/arm/machine.c Normal file
View file

@ -0,0 +1,229 @@
/* system/debuggerd/debuggerd.c
**
** Copyright 2006, The Android Open Source Project
**
** Licensed under the Apache License, Version 2.0 (the "License");
** you may not use this file except in compliance with the License.
** You may obtain a copy of the License at
**
** http://www.apache.org/licenses/LICENSE-2.0
**
** Unless required by applicable law or agreed to in writing, software
** distributed under the License is distributed on an "AS IS" BASIS,
** WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
** See the License for the specific language governing permissions and
** limitations under the License.
*/
#include <stdio.h>
#include <errno.h>
#include <signal.h>
#include <pthread.h>
#include <fcntl.h>
#include <sys/types.h>
#include <dirent.h>
#include <sys/ptrace.h>
#include <sys/wait.h>
#include <sys/exec_elf.h>
#include <sys/stat.h>
#include <cutils/sockets.h>
#include <cutils/properties.h>
#include <linux/input.h>
#include "utility.h"
#ifdef WITH_VFP
#ifdef WITH_VFP_D32
#define NUM_VFP_REGS 32
#else
#define NUM_VFP_REGS 16
#endif
#endif
/* Main entry point to get the backtrace from the crashing process */
extern int unwind_backtrace_with_ptrace(int tfd, pid_t pid, mapinfo *map,
unsigned int sp_list[],
int *frame0_pc_sane,
bool at_fault);
void dump_stack_and_code(int tfd, int pid, mapinfo *map,
int unwind_depth, unsigned int sp_list[],
bool at_fault)
{
unsigned int sp, pc, p, end, data;
struct pt_regs r;
int sp_depth;
bool only_in_tombstone = !at_fault;
char code_buffer[80];
if(ptrace(PTRACE_GETREGS, pid, 0, &r)) return;
sp = r.ARM_sp;
pc = r.ARM_pc;
_LOG(tfd, only_in_tombstone, "\ncode around pc:\n");
end = p = pc & ~3;
p -= 32;
end += 32;
/* Dump the code around PC as:
* addr contents
* 00008d34 fffffcd0 4c0eb530 b0934a0e 1c05447c
* 00008d44 f7ff18a0 490ced94 68035860 d0012b00
*/
while (p <= end) {
int i;
sprintf(code_buffer, "%08x ", p);
for (i = 0; i < 4; i++) {
data = ptrace(PTRACE_PEEKTEXT, pid, (void*)p, NULL);
sprintf(code_buffer + strlen(code_buffer), "%08x ", data);
p += 4;
}
_LOG(tfd, only_in_tombstone, "%s\n", code_buffer);
}
if ((unsigned) r.ARM_lr != pc) {
_LOG(tfd, only_in_tombstone, "\ncode around lr:\n");
end = p = r.ARM_lr & ~3;
p -= 32;
end += 32;
/* Dump the code around LR as:
* addr contents
* 00008d34 fffffcd0 4c0eb530 b0934a0e 1c05447c
* 00008d44 f7ff18a0 490ced94 68035860 d0012b00
*/
while (p <= end) {
int i;
sprintf(code_buffer, "%08x ", p);
for (i = 0; i < 4; i++) {
data = ptrace(PTRACE_PEEKTEXT, pid, (void*)p, NULL);
sprintf(code_buffer + strlen(code_buffer), "%08x ", data);
p += 4;
}
_LOG(tfd, only_in_tombstone, "%s\n", code_buffer);
}
}
p = sp - 64;
p &= ~3;
if (unwind_depth != 0) {
if (unwind_depth < STACK_CONTENT_DEPTH) {
end = sp_list[unwind_depth-1];
}
else {
end = sp_list[STACK_CONTENT_DEPTH-1];
}
}
else {
end = sp | 0x000000ff;
end += 0xff;
}
_LOG(tfd, only_in_tombstone, "\nstack:\n");
/* If the crash is due to PC == 0, there will be two frames that
* have identical SP value.
*/
if (sp_list[0] == sp_list[1]) {
sp_depth = 1;
}
else {
sp_depth = 0;
}
while (p <= end) {
char *prompt;
char level[16];
data = ptrace(PTRACE_PEEKTEXT, pid, (void*)p, NULL);
if (p == sp_list[sp_depth]) {
sprintf(level, "#%02d", sp_depth++);
prompt = level;
}
else {
prompt = " ";
}
/* Print the stack content in the log for the first 3 frames. For the
* rest only print them in the tombstone file.
*/
_LOG(tfd, (sp_depth > 2) || only_in_tombstone,
"%s %08x %08x %s\n", prompt, p, data,
map_to_name(map, data, ""));
p += 4;
}
/* print another 64-byte of stack data after the last frame */
end = p+64;
while (p <= end) {
data = ptrace(PTRACE_PEEKTEXT, pid, (void*)p, NULL);
_LOG(tfd, (sp_depth > 2) || only_in_tombstone,
" %08x %08x %s\n", p, data,
map_to_name(map, data, ""));
p += 4;
}
}
void dump_pc_and_lr(int tfd, int pid, mapinfo *map, int unwound_level,
bool at_fault)
{
struct pt_regs r;
if(ptrace(PTRACE_GETREGS, pid, 0, &r)) {
_LOG(tfd, !at_fault, "tid %d not responding!\n", pid);
return;
}
if (unwound_level == 0) {
_LOG(tfd, !at_fault, " #%02d pc %08x %s\n", 0, r.ARM_pc,
map_to_name(map, r.ARM_pc, "<unknown>"));
}
_LOG(tfd, !at_fault, " #%02d lr %08x %s\n", 1, r.ARM_lr,
map_to_name(map, r.ARM_lr, "<unknown>"));
}
void dump_registers(int tfd, int pid, bool at_fault)
{
struct pt_regs r;
bool only_in_tombstone = !at_fault;
if(ptrace(PTRACE_GETREGS, pid, 0, &r)) {
_LOG(tfd, only_in_tombstone,
"cannot get registers: %s\n", strerror(errno));
return;
}
_LOG(tfd, only_in_tombstone, " r0 %08x r1 %08x r2 %08x r3 %08x\n",
r.ARM_r0, r.ARM_r1, r.ARM_r2, r.ARM_r3);
_LOG(tfd, only_in_tombstone, " r4 %08x r5 %08x r6 %08x r7 %08x\n",
r.ARM_r4, r.ARM_r5, r.ARM_r6, r.ARM_r7);
_LOG(tfd, only_in_tombstone, " r8 %08x r9 %08x 10 %08x fp %08x\n",
r.ARM_r8, r.ARM_r9, r.ARM_r10, r.ARM_fp);
_LOG(tfd, only_in_tombstone,
" ip %08x sp %08x lr %08x pc %08x cpsr %08x\n",
r.ARM_ip, r.ARM_sp, r.ARM_lr, r.ARM_pc, r.ARM_cpsr);
#ifdef WITH_VFP
struct user_vfp vfp_regs;
int i;
if(ptrace(PTRACE_GETVFPREGS, pid, 0, &vfp_regs)) {
_LOG(tfd, only_in_tombstone,
"cannot get registers: %s\n", strerror(errno));
return;
}
for (i = 0; i < NUM_VFP_REGS; i += 2) {
_LOG(tfd, only_in_tombstone,
" d%-2d %016llx d%-2d %016llx\n",
i, vfp_regs.fpregs[i], i+1, vfp_regs.fpregs[i+1]);
}
_LOG(tfd, only_in_tombstone, " scr %08lx\n\n", vfp_regs.fpscr);
#endif
}

0
debuggerd/pr-support.c → debuggerd/arm/pr-support.c
View file

23
debuggerd/unwind-arm.c → debuggerd/arm/unwind.c
View file

@ -37,6 +37,8 @@
#include <unwind.h>
#include "utility.h"
#include "symbol_table.h"
typedef struct _ZSt9type_info type_info; /* This names C++ type_info type */
void __attribute__((weak)) __cxa_call_unexpected(_Unwind_Control_Block *ucbp);
@ -393,6 +395,7 @@ static _Unwind_Reason_Code log_function(_Unwind_Context *context, pid_t pid,
phase2_vrs *vrs = (phase2_vrs*) context;
const mapinfo *mi;
bool only_in_tombstone = !at_fault;
const struct symbol* sym = 0;
if (stack_level < STACK_CONTENT_DEPTH) {
sp_list[stack_level] = vrs->core.r[R_SP];
@ -427,7 +430,7 @@ static _Unwind_Reason_Code log_function(_Unwind_Context *context, pid_t pid,
if (pc & 1) {
_uw prev_word;
pc = (pc & ~1);
prev_word = get_remote_word(pid, (void *) pc-4);
prev_word = get_remote_word(pid, (char *) pc-4);
// Long offset
if ((prev_word & 0xf0000000) == 0xf0000000 &&
(prev_word & 0x0000e000) == 0x0000e000) {
@ -448,12 +451,22 @@ static _Unwind_Reason_Code log_function(_Unwind_Context *context, pid_t pid,
* 1MB boundaries, and the library may be larger than 1MB. So for .so
* addresses we print the relative offset in back trace.
*/
rel_pc = pc;
mi = pc_to_mapinfo(map, pc, &rel_pc);
_LOG(tfd, only_in_tombstone,
" #%02d pc %08x %s\n", stack_level, rel_pc,
mi ? mi->name : "");
/* See if we can determine what symbol this stack frame resides in */
if (mi != 0 && mi->symbols != 0) {
sym = symbol_table_lookup(mi->symbols, rel_pc);
}
if (sym) {
_LOG(tfd, only_in_tombstone,
" #%02d pc %08x %s (%s)\n", stack_level, rel_pc,
mi ? mi->name : "", sym->name);
} else {
_LOG(tfd, only_in_tombstone,
" #%02d pc %08x %s\n", stack_level, rel_pc,
mi ? mi->name : "");
}
return _URC_NO_REASON;
}

1
debuggerd/crasher.c
View file

@ -19,6 +19,7 @@
void crash1(void);
void crashnostack(void);
void maybeabort(void);
static void debuggerd_connect()
{

235
debuggerd/debuggerd.c
View file

@ -16,8 +16,6 @@
*/
#include <stdio.h>
#include <stdlib.h>
#include <unistd.h>
#include <errno.h>
#include <signal.h>
#include <pthread.h>
@ -33,37 +31,21 @@
#include <cutils/sockets.h>
#include <cutils/logd.h>
#include <cutils/sockets.h>
#include <cutils/properties.h>
#include <linux/input.h>
#include <private/android_filesystem_config.h>
#include "debuggerd.h"
#include "utility.h"
#ifdef WITH_VFP
#ifdef WITH_VFP_D32
#define NUM_VFP_REGS 32
#else
#define NUM_VFP_REGS 16
#endif
#endif
/* Main entry point to get the backtrace from the crashing process */
extern int unwind_backtrace_with_ptrace(int tfd, pid_t pid, mapinfo *map,
unsigned int sp_list[],
int *frame0_pc_sane,
bool at_fault);
static int logsocket = -1;
#define ANDROID_LOG_INFO 4
/* Log information onto the tombstone */
void _LOG(int tfd, bool in_tombstone_only, const char *fmt, ...)
{
char buf[128];
char buf[512];
va_list ap;
va_start(ap, fmt);
@ -79,13 +61,6 @@ void _LOG(int tfd, bool in_tombstone_only, const char *fmt, ...)
__android_log_vprint(ANDROID_LOG_INFO, "DEBUG", fmt, ap);
}
#define LOG(fmt...) _LOG(-1, 0, fmt)
#if 0
#define XLOG(fmt...) _LOG(-1, 0, fmt)
#else
#define XLOG(fmt...) do {} while(0)
#endif
// 6f000000-6f01e000 rwxp 00000000 00:0c 16389419 /system/lib/libcomposer.so
// 012345678901234567890123456789012345678901234567890123456789
// 0 1 2 3 4 5
@ -106,10 +81,11 @@ mapinfo *parse_maps_line(char *line)
mi->start = strtoul(line, 0, 16);
mi->end = strtoul(line + 9, 0, 16);
/* To be filled in parse_exidx_info if the mapped section starts with
/* To be filled in parse_elf_info if the mapped section starts with
* elf_header
*/
mi->exidx_start = mi->exidx_end = 0;
mi->symbols = 0;
mi->next = 0;
strcpy(mi->name, line + 49);
@ -125,186 +101,6 @@ void dump_build_info(int tfd)
_LOG(tfd, false, "Build fingerprint: '%s'\n", fingerprint);
}
void dump_stack_and_code(int tfd, int pid, mapinfo *map,
int unwind_depth, unsigned int sp_list[],
bool at_fault)
{
unsigned int sp, pc, p, end, data;
struct pt_regs r;
int sp_depth;
bool only_in_tombstone = !at_fault;
char code_buffer[80];
if(ptrace(PTRACE_GETREGS, pid, 0, &r)) return;
sp = r.ARM_sp;
pc = r.ARM_pc;
_LOG(tfd, only_in_tombstone, "\ncode around pc:\n");
end = p = pc & ~3;
p -= 32;
end += 32;
/* Dump the code around PC as:
* addr contents
* 00008d34 fffffcd0 4c0eb530 b0934a0e 1c05447c
* 00008d44 f7ff18a0 490ced94 68035860 d0012b00
*/
while (p <= end) {
int i;
sprintf(code_buffer, "%08x ", p);
for (i = 0; i < 4; i++) {
data = ptrace(PTRACE_PEEKTEXT, pid, (void*)p, NULL);
sprintf(code_buffer + strlen(code_buffer), "%08x ", data);
p += 4;
}
_LOG(tfd, only_in_tombstone, "%s\n", code_buffer);
}
if ((unsigned) r.ARM_lr != pc) {
_LOG(tfd, only_in_tombstone, "\ncode around lr:\n");
end = p = r.ARM_lr & ~3;
p -= 32;
end += 32;
/* Dump the code around LR as:
* addr contents
* 00008d34 fffffcd0 4c0eb530 b0934a0e 1c05447c
* 00008d44 f7ff18a0 490ced94 68035860 d0012b00
*/
while (p <= end) {
int i;
sprintf(code_buffer, "%08x ", p);
for (i = 0; i < 4; i++) {
data = ptrace(PTRACE_PEEKTEXT, pid, (void*)p, NULL);
sprintf(code_buffer + strlen(code_buffer), "%08x ", data);
p += 4;
}
_LOG(tfd, only_in_tombstone, "%s\n", code_buffer);
}
}
p = sp - 64;
p &= ~3;
if (unwind_depth != 0) {
if (unwind_depth < STACK_CONTENT_DEPTH) {
end = sp_list[unwind_depth-1];
}
else {
end = sp_list[STACK_CONTENT_DEPTH-1];
}
}
else {
end = sp | 0x000000ff;
end += 0xff;
}
_LOG(tfd, only_in_tombstone, "\nstack:\n");
/* If the crash is due to PC == 0, there will be two frames that
* have identical SP value.
*/
if (sp_list[0] == sp_list[1]) {
sp_depth = 1;
}
else {
sp_depth = 0;
}
while (p <= end) {
char *prompt;
char level[16];
data = ptrace(PTRACE_PEEKTEXT, pid, (void*)p, NULL);
if (p == sp_list[sp_depth]) {
sprintf(level, "#%02d", sp_depth++);
prompt = level;
}
else {
prompt = " ";
}
/* Print the stack content in the log for the first 3 frames. For the
* rest only print them in the tombstone file.
*/
_LOG(tfd, (sp_depth > 2) || only_in_tombstone,
"%s %08x %08x %s\n", prompt, p, data,
map_to_name(map, data, ""));
p += 4;
}
/* print another 64-byte of stack data after the last frame */
end = p+64;
while (p <= end) {
data = ptrace(PTRACE_PEEKTEXT, pid, (void*)p, NULL);
_LOG(tfd, (sp_depth > 2) || only_in_tombstone,
" %08x %08x %s\n", p, data,
map_to_name(map, data, ""));
p += 4;
}
}
void dump_pc_and_lr(int tfd, int pid, mapinfo *map, int unwound_level,
bool at_fault)
{
struct pt_regs r;
if(ptrace(PTRACE_GETREGS, pid, 0, &r)) {
_LOG(tfd, !at_fault, "tid %d not responding!\n", pid);
return;
}
if (unwound_level == 0) {
_LOG(tfd, !at_fault, " #%02d pc %08x %s\n", 0, r.ARM_pc,
map_to_name(map, r.ARM_pc, "<unknown>"));
}
_LOG(tfd, !at_fault, " #%02d lr %08x %s\n", 1, r.ARM_lr,
map_to_name(map, r.ARM_lr, "<unknown>"));
}
void dump_registers(int tfd, int pid, bool at_fault)
{
struct pt_regs r;
bool only_in_tombstone = !at_fault;
if(ptrace(PTRACE_GETREGS, pid, 0, &r)) {
_LOG(tfd, only_in_tombstone,
"cannot get registers: %s\n", strerror(errno));
return;
}
_LOG(tfd, only_in_tombstone, " r0 %08x r1 %08x r2 %08x r3 %08x\n",
r.ARM_r0, r.ARM_r1, r.ARM_r2, r.ARM_r3);
_LOG(tfd, only_in_tombstone, " r4 %08x r5 %08x r6 %08x r7 %08x\n",
r.ARM_r4, r.ARM_r5, r.ARM_r6, r.ARM_r7);
_LOG(tfd, only_in_tombstone, " r8 %08x r9 %08x 10 %08x fp %08x\n",
r.ARM_r8, r.ARM_r9, r.ARM_r10, r.ARM_fp);
_LOG(tfd, only_in_tombstone,
" ip %08x sp %08x lr %08x pc %08x cpsr %08x\n",
r.ARM_ip, r.ARM_sp, r.ARM_lr, r.ARM_pc, r.ARM_cpsr);
#ifdef WITH_VFP
struct user_vfp vfp_regs;
int i;
if(ptrace(PTRACE_GETVFPREGS, pid, 0, &vfp_regs)) {
_LOG(tfd, only_in_tombstone,
"cannot get registers: %s\n", strerror(errno));
return;
}
for (i = 0; i < NUM_VFP_REGS; i += 2) {
_LOG(tfd, only_in_tombstone,
" d%-2d %016llx d%-2d %016llx\n",
i, vfp_regs.fpregs[i], i+1, vfp_regs.fpregs[i+1]);
}
_LOG(tfd, only_in_tombstone, " scr %08lx\n\n", vfp_regs.fpscr);
#endif
}
const char *get_signame(int sig)
{
switch(sig) {
@ -399,7 +195,7 @@ void dump_crash_banner(int tfd, unsigned pid, unsigned tid, int sig)
if(sig) dump_fault_addr(tfd, tid, sig);
}
static void parse_exidx_info(mapinfo *milist, pid_t pid)
static void parse_elf_info(mapinfo *milist, pid_t pid)
{
mapinfo *mi;
for (mi = milist; mi != NULL; mi = mi->next) {
@ -422,13 +218,18 @@ static void parse_exidx_info(mapinfo *milist, pid_t pid)
/* Parse the program header */
get_remote_struct(pid, (char *) (ptr+i), &phdr,
sizeof(Elf32_Phdr));
#ifdef __arm__
/* Found a EXIDX segment? */
if (phdr.p_type == PT_ARM_EXIDX) {
mi->exidx_start = mi->start + phdr.p_offset;
mi->exidx_end = mi->exidx_start + phdr.p_filesz;
break;
}
#endif
}
/* Try to load symbols from this file */
mi->symbols = symbol_table_create(mi->name);
}
}
}
@ -440,7 +241,9 @@ void dump_crash_report(int tfd, unsigned pid, unsigned tid, bool at_fault)
mapinfo *milist = 0;
unsigned int sp_list[STACK_CONTENT_DEPTH];
int stack_depth;
#ifdef __arm__
int frame0_pc_sane = 1;
#endif
if (!at_fault) {
_LOG(tfd, true,
@ -466,8 +269,9 @@ void dump_crash_report(int tfd, unsigned pid, unsigned tid, bool at_fault)
fclose(fp);
}
parse_exidx_info(milist, tid);
parse_elf_info(milist, tid);
#if __arm__
/* If stack unwinder fails, use the default solution to dump the stack
* content.
*/
@ -482,9 +286,18 @@ void dump_crash_report(int tfd, unsigned pid, unsigned tid, bool at_fault)
}
dump_stack_and_code(tfd, tid, milist, stack_depth, sp_list, at_fault);
#elif __i386__
/* If stack unwinder fails, use the default solution to dump the stack
* content.
*/
stack_depth = unwind_backtrace_with_ptrace_x86(tfd, tid, milist,at_fault);
#else
#error "Unsupported architecture"
#endif
while(milist) {
mapinfo *next = milist->next;
symbol_table_free(milist->symbols);
free(milist);
milist = next;
}
@ -863,10 +676,12 @@ done:
if(fd != -1) close(fd);
}
int main()
{
int s;
struct sigaction act;
int logsocket = -1;
logsocket = socket_local_client("logd",
ANDROID_SOCKET_NAMESPACE_ABSTRACT, SOCK_DGRAM);

911
debuggerd/debuggerd.c.orig Normal file
View file

@ -0,0 +1,911 @@
/* system/debuggerd/debuggerd.c
**
** Copyright 2006, The Android Open Source Project
**
** Licensed under the Apache License, Version 2.0 (the "License");
** you may not use this file except in compliance with the License.
** You may obtain a copy of the License at
**
** http://www.apache.org/licenses/LICENSE-2.0
**
** Unless required by applicable law or agreed to in writing, software
** distributed under the License is distributed on an "AS IS" BASIS,
** WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
** See the License for the specific language governing permissions and
** limitations under the License.
*/
#include <stdio.h>
#include <stdlib.h>
#include <unistd.h>
#include <errno.h>
#include <signal.h>
#include <pthread.h>
#include <stdarg.h>
#include <fcntl.h>
#include <sys/types.h>
#include <dirent.h>
#include <sys/ptrace.h>
#include <sys/wait.h>
#include <sys/exec_elf.h>
#include <sys/stat.h>
#include <cutils/sockets.h>
#include <cutils/logd.h>
#include <cutils/sockets.h>
#include <cutils/properties.h>
#include <linux/input.h>
#include <private/android_filesystem_config.h>
#include "utility.h"
#ifdef WITH_VFP
#ifdef WITH_VFP_D32
#define NUM_VFP_REGS 32
#else
#define NUM_VFP_REGS 16
#endif
#endif
/* Main entry point to get the backtrace from the crashing process */
extern int unwind_backtrace_with_ptrace(int tfd, pid_t pid, mapinfo *map,
unsigned int sp_list[],
int *frame0_pc_sane,
bool at_fault);
static int logsocket = -1;
#define ANDROID_LOG_INFO 4
/* Log information onto the tombstone */
void _LOG(int tfd, bool in_tombstone_only, const char *fmt, ...)
{
char buf[512];
va_list ap;
va_start(ap, fmt);
if (tfd >= 0) {
int len;
vsnprintf(buf, sizeof(buf), fmt, ap);
len = strlen(buf);
if(tfd >= 0) write(tfd, buf, len);
}
if (!in_tombstone_only)
__android_log_vprint(ANDROID_LOG_INFO, "DEBUG", fmt, ap);
}
#define LOG(fmt...) _LOG(-1, 0, fmt)
#if 0
#define XLOG(fmt...) _LOG(-1, 0, fmt)
#else
#define XLOG(fmt...) do {} while(0)
#endif
// 6f000000-6f01e000 rwxp 00000000 00:0c 16389419 /system/lib/libcomposer.so
// 012345678901234567890123456789012345678901234567890123456789
// 0 1 2 3 4 5
mapinfo *parse_maps_line(char *line)
{
mapinfo *mi;
int len = strlen(line);
if(len < 1) return 0;
line[--len] = 0;
if(len < 50) return 0;
if(line[20] != 'x') return 0;
mi = malloc(sizeof(mapinfo) + (len - 47));
if(mi == 0) return 0;
mi->start = strtoul(line, 0, 16);
mi->end = strtoul(line + 9, 0, 16);
/* To be filled in parse_elf_info if the mapped section starts with
* elf_header
*/
mi->exidx_start = mi->exidx_end = 0;
mi->symbols = 0;
mi->next = 0;
strcpy(mi->name, line + 49);
return mi;
}
void dump_build_info(int tfd)
{
char fingerprint[PROPERTY_VALUE_MAX];
property_get("ro.build.fingerprint", fingerprint, "unknown");
_LOG(tfd, false, "Build fingerprint: '%s'\n", fingerprint);
}
void dump_stack_and_code(int tfd, int pid, mapinfo *map,
int unwind_depth, unsigned int sp_list[],
bool at_fault)
{
unsigned int sp, pc, p, end, data;
struct pt_regs r;
int sp_depth;
bool only_in_tombstone = !at_fault;
char code_buffer[80];
if(ptrace(PTRACE_GETREGS, pid, 0, &r)) return;
sp = r.ARM_sp;
pc = r.ARM_pc;
_LOG(tfd, only_in_tombstone, "\ncode around pc:\n");
end = p = pc & ~3;
p -= 32;
end += 32;
/* Dump the code around PC as:
* addr contents
* 00008d34 fffffcd0 4c0eb530 b0934a0e 1c05447c
* 00008d44 f7ff18a0 490ced94 68035860 d0012b00
*/
while (p <= end) {
int i;
sprintf(code_buffer, "%08x ", p);
for (i = 0; i < 4; i++) {
data = ptrace(PTRACE_PEEKTEXT, pid, (void*)p, NULL);
sprintf(code_buffer + strlen(code_buffer), "%08x ", data);
p += 4;
}
_LOG(tfd, only_in_tombstone, "%s\n", code_buffer);
}
if ((unsigned) r.ARM_lr != pc) {
_LOG(tfd, only_in_tombstone, "\ncode around lr:\n");
end = p = r.ARM_lr & ~3;
p -= 32;
end += 32;
/* Dump the code around LR as:
* addr contents
* 00008d34 fffffcd0 4c0eb530 b0934a0e 1c05447c
* 00008d44 f7ff18a0 490ced94 68035860 d0012b00
*/
while (p <= end) {
int i;
sprintf(code_buffer, "%08x ", p);
for (i = 0; i < 4; i++) {
data = ptrace(PTRACE_PEEKTEXT, pid, (void*)p, NULL);
sprintf(code_buffer + strlen(code_buffer), "%08x ", data);
p += 4;
}
_LOG(tfd, only_in_tombstone, "%s\n", code_buffer);
}
}
p = sp - 64;
p &= ~3;
if (unwind_depth != 0) {
if (unwind_depth < STACK_CONTENT_DEPTH) {
end = sp_list[unwind_depth-1];
}
else {
end = sp_list[STACK_CONTENT_DEPTH-1];
}
}
else {
end = sp | 0x000000ff;
end += 0xff;
}
_LOG(tfd, only_in_tombstone, "\nstack:\n");
/* If the crash is due to PC == 0, there will be two frames that
* have identical SP value.
*/
if (sp_list[0] == sp_list[1]) {
sp_depth = 1;
}
else {
sp_depth = 0;
}
while (p <= end) {
char *prompt;
char level[16];
data = ptrace(PTRACE_PEEKTEXT, pid, (void*)p, NULL);
if (p == sp_list[sp_depth]) {
sprintf(level, "#%02d", sp_depth++);
prompt = level;
}
else {
prompt = " ";
}
/* Print the stack content in the log for the first 3 frames. For the
* rest only print them in the tombstone file.
*/
_LOG(tfd, (sp_depth > 2) || only_in_tombstone,
"%s %08x %08x %s\n", prompt, p, data,
map_to_name(map, data, ""));
p += 4;
}
/* print another 64-byte of stack data after the last frame */
end = p+64;
while (p <= end) {
data = ptrace(PTRACE_PEEKTEXT, pid, (void*)p, NULL);
_LOG(tfd, (sp_depth > 2) || only_in_tombstone,
" %08x %08x %s\n", p, data,
map_to_name(map, data, ""));
p += 4;
}
}
void dump_pc_and_lr(int tfd, int pid, mapinfo *map, int unwound_level,
bool at_fault)
{
struct pt_regs r;
if(ptrace(PTRACE_GETREGS, pid, 0, &r)) {
_LOG(tfd, !at_fault, "tid %d not responding!\n", pid);
return;
}
if (unwound_level == 0) {
_LOG(tfd, !at_fault, " #%02d pc %08x %s\n", 0, r.ARM_pc,
map_to_name(map, r.ARM_pc, "<unknown>"));
}
_LOG(tfd, !at_fault, " #%02d lr %08x %s\n", 1, r.ARM_lr,
map_to_name(map, r.ARM_lr, "<unknown>"));
}
void dump_registers(int tfd, int pid, bool at_fault)
{
struct pt_regs r;
bool only_in_tombstone = !at_fault;
if(ptrace(PTRACE_GETREGS, pid, 0, &r)) {
_LOG(tfd, only_in_tombstone,
"cannot get registers: %s\n", strerror(errno));
return;
}
_LOG(tfd, only_in_tombstone, " r0 %08x r1 %08x r2 %08x r3 %08x\n",
r.ARM_r0, r.ARM_r1, r.ARM_r2, r.ARM_r3);
_LOG(tfd, only_in_tombstone, " r4 %08x r5 %08x r6 %08x r7 %08x\n",
r.ARM_r4, r.ARM_r5, r.ARM_r6, r.ARM_r7);
_LOG(tfd, only_in_tombstone, " r8 %08x r9 %08x 10 %08x fp %08x\n",
r.ARM_r8, r.ARM_r9, r.ARM_r10, r.ARM_fp);
_LOG(tfd, only_in_tombstone,
" ip %08x sp %08x lr %08x pc %08x cpsr %08x\n",
r.ARM_ip, r.ARM_sp, r.ARM_lr, r.ARM_pc, r.ARM_cpsr);
#ifdef WITH_VFP
struct user_vfp vfp_regs;
int i;
if(ptrace(PTRACE_GETVFPREGS, pid, 0, &vfp_regs)) {
_LOG(tfd, only_in_tombstone,
"cannot get registers: %s\n", strerror(errno));
return;
}
for (i = 0; i < NUM_VFP_REGS; i += 2) {
_LOG(tfd, only_in_tombstone,
" d%-2d %016llx d%-2d %016llx\n",
i, vfp_regs.fpregs[i], i+1, vfp_regs.fpregs[i+1]);
}
_LOG(tfd, only_in_tombstone, " scr %08lx\n\n", vfp_regs.fpscr);
#endif
}
const char *get_signame(int sig)
{
switch(sig) {
case SIGILL: return "SIGILL";
case SIGABRT: return "SIGABRT";
case SIGBUS: return "SIGBUS";
case SIGFPE: return "SIGFPE";
case SIGSEGV: return "SIGSEGV";
case SIGSTKFLT: return "SIGSTKFLT";
default: return "?";
}
}
const char *get_sigcode(int signo, int code)
{
switch (signo) {
case SIGILL:
switch (code) {
case ILL_ILLOPC: return "ILL_ILLOPC";
case ILL_ILLOPN: return "ILL_ILLOPN";
case ILL_ILLADR: return "ILL_ILLADR";
case ILL_ILLTRP: return "ILL_ILLTRP";
case ILL_PRVOPC: return "ILL_PRVOPC";
case ILL_PRVREG: return "ILL_PRVREG";
case ILL_COPROC: return "ILL_COPROC";
case ILL_BADSTK: return "ILL_BADSTK";
}
break;
case SIGBUS:
switch (code) {
case BUS_ADRALN: return "BUS_ADRALN";
case BUS_ADRERR: return "BUS_ADRERR";
case BUS_OBJERR: return "BUS_OBJERR";
}
break;
case SIGFPE:
switch (code) {
case FPE_INTDIV: return "FPE_INTDIV";
case FPE_INTOVF: return "FPE_INTOVF";
case FPE_FLTDIV: return "FPE_FLTDIV";
case FPE_FLTOVF: return "FPE_FLTOVF";
case FPE_FLTUND: return "FPE_FLTUND";
case FPE_FLTRES: return "FPE_FLTRES";
case FPE_FLTINV: return "FPE_FLTINV";
case FPE_FLTSUB: return "FPE_FLTSUB";
}
break;
case SIGSEGV:
switch (code) {
case SEGV_MAPERR: return "SEGV_MAPERR";
case SEGV_ACCERR: return "SEGV_ACCERR";
}
break;
}
return "?";
}
void dump_fault_addr(int tfd, int pid, int sig)
{
siginfo_t si;
memset(&si, 0, sizeof(si));
if(ptrace(PTRACE_GETSIGINFO, pid, 0, &si)){
_LOG(tfd, false, "cannot get siginfo: %s\n", strerror(errno));
} else {
_LOG(tfd, false, "signal %d (%s), code %d (%s), fault addr %08x\n",
sig, get_signame(sig),
si.si_code, get_sigcode(sig, si.si_code),
si.si_addr);
}
}
void dump_crash_banner(int tfd, unsigned pid, unsigned tid, int sig)
{
char data[1024];
char *x = 0;
FILE *fp;
sprintf(data, "/proc/%d/cmdline", pid);
fp = fopen(data, "r");
if(fp) {
x = fgets(data, 1024, fp);
fclose(fp);
}
_LOG(tfd, false,
"*** *** *** *** *** *** *** *** *** *** *** *** *** *** *** ***\n");
dump_build_info(tfd);
_LOG(tfd, false, "pid: %d, tid: %d >>> %s <<<\n",
pid, tid, x ? x : "UNKNOWN");
if(sig) dump_fault_addr(tfd, tid, sig);
}
static void parse_elf_info(mapinfo *milist, pid_t pid)
{
mapinfo *mi;
for (mi = milist; mi != NULL; mi = mi->next) {
Elf32_Ehdr ehdr;
memset(&ehdr, 0, sizeof(Elf32_Ehdr));
/* Read in sizeof(Elf32_Ehdr) worth of data from the beginning of
* mapped section.
*/
get_remote_struct(pid, (void *) (mi->start), &ehdr,
sizeof(Elf32_Ehdr));
/* Check if it has the matching magic words */
if (IS_ELF(ehdr)) {
Elf32_Phdr phdr;
Elf32_Phdr *ptr;
int i;
ptr = (Elf32_Phdr *) (mi->start + ehdr.e_phoff);
for (i = 0; i < ehdr.e_phnum; i++) {
/* Parse the program header */
get_remote_struct(pid, (char *) (ptr+i), &phdr,
sizeof(Elf32_Phdr));
/* Found a EXIDX segment? */
if (phdr.p_type == PT_ARM_EXIDX) {
mi->exidx_start = mi->start + phdr.p_offset;
mi->exidx_end = mi->exidx_start + phdr.p_filesz;
break;
}
}
/* Try to load symbols from this file */
mi->symbols = symbol_table_create(mi->name);
}
}
}
void dump_crash_report(int tfd, unsigned pid, unsigned tid, bool at_fault)
{
char data[1024];
FILE *fp;
mapinfo *milist = 0;
unsigned int sp_list[STACK_CONTENT_DEPTH];
int stack_depth;
int frame0_pc_sane = 1;
if (!at_fault) {
_LOG(tfd, true,
"--- --- --- --- --- --- --- --- --- --- --- --- --- --- --- ---\n");
_LOG(tfd, true, "pid: %d, tid: %d\n", pid, tid);
}
dump_registers(tfd, tid, at_fault);
/* Clear stack pointer records */
memset(sp_list, 0, sizeof(sp_list));
sprintf(data, "/proc/%d/maps", pid);
fp = fopen(data, "r");
if(fp) {
while(fgets(data, 1024, fp)) {
mapinfo *mi = parse_maps_line(data);
if(mi) {
mi->next = milist;
milist = mi;
}
}
fclose(fp);
}
parse_elf_info(milist, tid);
/* If stack unwinder fails, use the default solution to dump the stack
* content.
*/
stack_depth = unwind_backtrace_with_ptrace(tfd, tid, milist, sp_list,
&frame0_pc_sane, at_fault);
/* The stack unwinder should at least unwind two levels of stack. If less
* level is seen we make sure at lease pc and lr are dumped.
*/
if (stack_depth < 2) {
dump_pc_and_lr(tfd, tid, milist, stack_depth, at_fault);
}
dump_stack_and_code(tfd, tid, milist, stack_depth, sp_list, at_fault);
while(milist) {
mapinfo *next = milist->next;
symbol_table_free(milist->symbols);
free(milist);
milist = next;
}
}
#define MAX_TOMBSTONES 10
#define typecheck(x,y) { \
typeof(x) __dummy1; \
typeof(y) __dummy2; \
(void)(&__dummy1 == &__dummy2); }
#define TOMBSTONE_DIR "/data/tombstones"
/*
* find_and_open_tombstone - find an available tombstone slot, if any, of the
* form tombstone_XX where XX is 00 to MAX_TOMBSTONES-1, inclusive. If no
* file is available, we reuse the least-recently-modified file.
*/
static int find_and_open_tombstone(void)
{
unsigned long mtime = ULONG_MAX;
struct stat sb;
char path[128];
int fd, i, oldest = 0;
/*
* XXX: Our stat.st_mtime isn't time_t. If it changes, as it probably ought
* to, our logic breaks. This check will generate a warning if that happens.
*/
typecheck(mtime, sb.st_mtime);
/*
* In a single wolf-like pass, find an available slot and, in case none
* exist, find and record the least-recently-modified file.
*/
for (i = 0; i < MAX_TOMBSTONES; i++) {
snprintf(path, sizeof(path), TOMBSTONE_DIR"/tombstone_%02d", i);
if (!stat(path, &sb)) {
if (sb.st_mtime < mtime) {
oldest = i;
mtime = sb.st_mtime;
}
continue;
}
if (errno != ENOENT)
continue;
fd = open(path, O_CREAT | O_EXCL | O_WRONLY, 0600);
if (fd < 0)
continue; /* raced ? */
fchown(fd, AID_SYSTEM, AID_SYSTEM);
return fd;
}
/* we didn't find an available file, so we clobber the oldest one */
snprintf(path, sizeof(path), TOMBSTONE_DIR"/tombstone_%02d", oldest);
fd = open(path, O_CREAT | O_TRUNC | O_WRONLY, 0600);
fchown(fd, AID_SYSTEM, AID_SYSTEM);
return fd;
}
/* Return true if some thread is not detached cleanly */
static bool dump_sibling_thread_report(int tfd, unsigned pid, unsigned tid)
{
char task_path[1024];
sprintf(task_path, "/proc/%d/task", pid);
DIR *d;
struct dirent *de;
int need_cleanup = 0;
d = opendir(task_path);
/* Bail early if cannot open the task directory */
if (d == NULL) {
XLOG("Cannot open /proc/%d/task\n", pid);
return false;
}
while ((de = readdir(d)) != NULL) {
unsigned new_tid;
/* Ignore "." and ".." */
if (!strcmp(de->d_name, ".") || !strcmp(de->d_name, ".."))
continue;
new_tid = atoi(de->d_name);
/* The main thread at fault has been handled individually */
if (new_tid == tid)
continue;
/* Skip this thread if cannot ptrace it */
if (ptrace(PTRACE_ATTACH, new_tid, 0, 0) < 0)
continue;
dump_crash_report(tfd, pid, new_tid, false);
need_cleanup |= ptrace(PTRACE_DETACH, new_tid, 0, 0);
}
closedir(d);
return need_cleanup != 0;
}
/* Return true if some thread is not detached cleanly */
static bool engrave_tombstone(unsigned pid, unsigned tid, int debug_uid,
int signal)
{
int fd;
bool need_cleanup = false;
mkdir(TOMBSTONE_DIR, 0755);
chown(TOMBSTONE_DIR, AID_SYSTEM, AID_SYSTEM);
fd = find_and_open_tombstone();
if (fd < 0)
return need_cleanup;
dump_crash_banner(fd, pid, tid, signal);
dump_crash_report(fd, pid, tid, true);
/*
* If the user has requested to attach gdb, don't collect the per-thread
* information as it increases the chance to lose track of the process.
*/
if ((signed)pid > debug_uid) {
need_cleanup = dump_sibling_thread_report(fd, pid, tid);
}
close(fd);
return need_cleanup;
}
static int
write_string(const char* file, const char* string)
{
int len;
int fd;
ssize_t amt;
fd = open(file, O_RDWR);
len = strlen(string);
if (fd < 0)
return -errno;
amt = write(fd, string, len);
close(fd);
return amt >= 0 ? 0 : -errno;
}
static
void init_debug_led(void)
{
// trout leds
write_string("/sys/class/leds/red/brightness", "0");
write_string("/sys/class/leds/green/brightness", "0");
write_string("/sys/class/leds/blue/brightness", "0");
write_string("/sys/class/leds/red/device/blink", "0");
// sardine leds
write_string("/sys/class/leds/left/cadence", "0,0");
}
static
void enable_debug_led(void)
{
// trout leds
write_string("/sys/class/leds/red/brightness", "255");
// sardine leds
write_string("/sys/class/leds/left/cadence", "1,0");
}
static
void disable_debug_led(void)
{
// trout leds
write_string("/sys/class/leds/red/brightness", "0");
// sardine leds
write_string("/sys/class/leds/left/cadence", "0,0");
}
extern int init_getevent();
extern void uninit_getevent();
extern int get_event(struct input_event* event, int timeout);
static void wait_for_user_action(unsigned tid, struct ucred* cr)
{
(void)tid;
/* First log a helpful message */
LOG( "********************************************************\n"
"* Process %d has been suspended while crashing. To\n"
"* attach gdbserver for a gdb connection on port 5039:\n"
"*\n"
"* adb shell gdbserver :5039 --attach %d &\n"
"*\n"
"* Press HOME key to let the process continue crashing.\n"
"********************************************************\n",
cr->pid, cr->pid);
/* wait for HOME key (TODO: something useful for devices w/o HOME key) */
if (init_getevent() == 0) {
int ms = 1200 / 10;
int dit = 1;
int dah = 3*dit;
int _ = -dit;
int ___ = 3*_;
int _______ = 7*_;
const signed char codes[] = {
dit,_,dit,_,dit,___,dah,_,dah,_,dah,___,dit,_,dit,_,dit,_______
};
size_t s = 0;
struct input_event e;
int home = 0;
init_debug_led();
enable_debug_led();
do {
int timeout = abs((int)(codes[s])) * ms;
int res = get_event(&e, timeout);
if (res == 0) {
if (e.type==EV_KEY && e.code==KEY_HOME && e.value==0)
home = 1;
} else if (res == 1) {
if (++s >= sizeof(codes)/sizeof(*codes))
s = 0;
if (codes[s] > 0) {
enable_debug_led();
} else {
disable_debug_led();
}
}
} while (!home);
uninit_getevent();
}
/* don't forget to turn debug led off */
disable_debug_led();
/* close filedescriptor */
LOG("debuggerd resuming process %d", cr->pid);
}
static void handle_crashing_process(int fd)
{
char buf[64];
struct stat s;
unsigned tid;
struct ucred cr;
int n, len, status;
int tid_attach_status = -1;
unsigned retry = 30;
bool need_cleanup = false;
char value[PROPERTY_VALUE_MAX];
property_get("debug.db.uid", value, "-1");
int debug_uid = atoi(value);
XLOG("handle_crashing_process(%d)\n", fd);
len = sizeof(cr);
n = getsockopt(fd, SOL_SOCKET, SO_PEERCRED, &cr, &len);
if(n != 0) {
LOG("cannot get credentials\n");
goto done;
}
XLOG("reading tid\n");
fcntl(fd, F_SETFL, O_NONBLOCK);
while((n = read(fd, &tid, sizeof(unsigned))) != sizeof(unsigned)) {
if(errno == EINTR) continue;
if(errno == EWOULDBLOCK) {
if(retry-- > 0) {
usleep(100 * 1000);
continue;
}
LOG("timed out reading tid\n");
goto done;
}
LOG("read failure? %s\n", strerror(errno));
goto done;
}
sprintf(buf,"/proc/%d/task/%d", cr.pid, tid);
if(stat(buf, &s)) {
LOG("tid %d does not exist in pid %d. ignoring debug request\n",
tid, cr.pid);
close(fd);
return;
}
XLOG("BOOM: pid=%d uid=%d gid=%d tid=%d\n", cr.pid, cr.uid, cr.gid, tid);
tid_attach_status = ptrace(PTRACE_ATTACH, tid, 0, 0);
if(tid_attach_status < 0) {
LOG("ptrace attach failed: %s\n", strerror(errno));
goto done;
}
close(fd);
fd = -1;
for(;;) {
n = waitpid(tid, &status, __WALL);
if(n < 0) {
if(errno == EAGAIN) continue;
LOG("waitpid failed: %s\n", strerror(errno));
goto done;
}
XLOG("waitpid: n=%d status=%08x\n", n, status);
if(WIFSTOPPED(status)){
n = WSTOPSIG(status);
switch(n) {
case SIGSTOP:
XLOG("stopped -- continuing\n");
n = ptrace(PTRACE_CONT, tid, 0, 0);
if(n) {
LOG("ptrace failed: %s\n", strerror(errno));
goto done;
}
continue;
case SIGILL:
case SIGABRT:
case SIGBUS:
case SIGFPE:
case SIGSEGV:
case SIGSTKFLT: {
XLOG("stopped -- fatal signal\n");
need_cleanup = engrave_tombstone(cr.pid, tid, debug_uid, n);
kill(tid, SIGSTOP);
goto done;
}
default:
XLOG("stopped -- unexpected signal\n");
goto done;
}
} else {
XLOG("unexpected waitpid response\n");
goto done;
}
}
done:
XLOG("detaching\n");
/* stop the process so we can debug */
kill(cr.pid, SIGSTOP);
/*
* If a thread has been attached by ptrace, make sure it is detached
* successfully otherwise we will get a zombie.
*/
if (tid_attach_status == 0) {
int detach_status;
/* detach so we can attach gdbserver */
detach_status = ptrace(PTRACE_DETACH, tid, 0, 0);
need_cleanup |= (detach_status != 0);
}
/*
* if debug.db.uid is set, its value indicates if we should wait
* for user action for the crashing process.
* in this case, we log a message and turn the debug LED on
* waiting for a gdb connection (for instance)
*/
if ((signed)cr.uid <= debug_uid) {
wait_for_user_action(tid, &cr);
}
/* resume stopped process (so it can crash in peace) */
kill(cr.pid, SIGCONT);
if (need_cleanup) {
LOG("debuggerd committing suicide to free the zombie!\n");
kill(getpid(), SIGKILL);
}
if(fd != -1) close(fd);
}
int main()
{
int s;
struct sigaction act;
logsocket = socket_local_client("logd",
ANDROID_SOCKET_NAMESPACE_ABSTRACT, SOCK_DGRAM);
if(logsocket < 0) {
logsocket = -1;
} else {
fcntl(logsocket, F_SETFD, FD_CLOEXEC);
}
act.sa_handler = SIG_DFL;
sigemptyset(&act.sa_mask);
sigaddset(&act.sa_mask,SIGCHLD);
act.sa_flags = SA_NOCLDWAIT;
sigaction(SIGCHLD, &act, 0);
s = socket_local_server("android:debuggerd",
ANDROID_SOCKET_NAMESPACE_ABSTRACT, SOCK_STREAM);
if(s < 0) return -1;
fcntl(s, F_SETFD, FD_CLOEXEC);
LOG("debuggerd: " __DATE__ " " __TIME__ "\n");
for(;;) {
struct sockaddr addr;
socklen_t alen;
int fd;
alen = sizeof(addr);
fd = accept(s, &addr, &alen);
if(fd < 0) continue;
fcntl(fd, F_SETFD, FD_CLOEXEC);
handle_crashing_process(fd);
}
return 0;
}

39
debuggerd/debuggerd.h Normal file
View file

@ -0,0 +1,39 @@
/* system/debuggerd/debuggerd.h
**
** Copyright 2006, The Android Open Source Project
**
** Licensed under the Apache License, Version 2.0 (the "License");
** you may not use this file except in compliance with the License.
** You may obtain a copy of the License at
**
** http://www.apache.org/licenses/LICENSE-2.0
**
** Unless required by applicable law or agreed to in writing, software
** distributed under the License is distributed on an "AS IS" BASIS,
** WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
** See the License for the specific language governing permissions and
** limitations under the License.
*/
#include <cutils/logd.h>
#include <sys/ptrace.h>
#include <unwind.h>
#include "utility.h"
#include "symbol_table.h"
/* Main entry point to get the backtrace from the crashing process */
extern int unwind_backtrace_with_ptrace(int tfd, pid_t pid, mapinfo *map,
unsigned int sp_list[],
int *frame0_pc_sane,
bool at_fault);
extern void dump_registers(int tfd, int pid, bool at_fault);
extern int unwind_backtrace_with_ptrace_x86(int tfd, pid_t pid, mapinfo *map, bool at_fault);
void dump_pc_and_lr(int tfd, int pid, mapinfo *map, int unwound_level, bool at_fault);
void dump_stack_and_code(int tfd, int pid, mapinfo *map,
int unwind_depth, unsigned int sp_list[],
bool at_fault);

231
debuggerd/symbol_table.c Normal file
View file

@ -0,0 +1,231 @@
#include <stdlib.h>
#include <fcntl.h>
#include <string.h>
#include <sys/stat.h>
#include <sys/mman.h>
#include "symbol_table.h"
#include "utility.h"
#include <linux/elf.h>
// Compare func for qsort
static int qcompar(const void *a, const void *b)
{
return ((struct symbol*)a)->addr - ((struct symbol*)b)->addr;
}
// Compare func for bsearch
static int bcompar(const void *addr, const void *element)
{
struct symbol *symbol = (struct symbol*)element;
if((unsigned int)addr < symbol->addr) {
return -1;
}
if((unsigned int)addr - symbol->addr >= symbol->size) {
return 1;
}
return 0;
}
/*
* Create a symbol table from a given file
*
* Parameters:
* filename - Filename to process
*
* Returns:
* A newly-allocated SymbolTable structure, or NULL if error.
* Free symbol table with symbol_table_free()
*/
struct symbol_table *symbol_table_create(const char *filename)
{
struct symbol_table *table = NULL;
// Open the file, and map it into memory
struct stat sb;
int length;
char *base;
XLOG("Creating symbol table for %s\n", filename);
int fd = open(filename, O_RDONLY);
if(fd < 0) {
goto out;
}
fstat(fd, &sb);
length = sb.st_size;
base = mmap(NULL, length, PROT_READ, MAP_PRIVATE, fd, 0);
if(!base) {
goto out_close;
}
// Parse the file header
Elf32_Ehdr *hdr = (Elf32_Ehdr*)base;
Elf32_Shdr *shdr = (Elf32_Shdr*)(base + hdr->e_shoff);
// Search for the dynamic symbols section
int sym_idx = -1;
int dynsym_idx = -1;
int i;
for(i = 0; i < hdr->e_shnum; i++) {
if(shdr[i].sh_type == SHT_SYMTAB ) {
sym_idx = i;
}
if(shdr[i].sh_type == SHT_DYNSYM ) {
dynsym_idx = i;
}
}
if ((dynsym_idx == -1) && (sym_idx == -1)) {
goto out_unmap;
}
table = malloc(sizeof(struct symbol_table));
if(!table) {
goto out_unmap;
}
table->name = strdup(filename);
table->num_symbols = 0;
Elf32_Sym *dynsyms = (Elf32_Sym*)(base + shdr[dynsym_idx].sh_offset);
Elf32_Sym *syms = (Elf32_Sym*)(base + shdr[sym_idx].sh_offset);
int dynnumsyms = shdr[dynsym_idx].sh_size / shdr[dynsym_idx].sh_entsize;
int numsyms = shdr[sym_idx].sh_size / shdr[sym_idx].sh_entsize;
int dynstr_idx = shdr[dynsym_idx].sh_link;
int str_idx = shdr[sym_idx].sh_link;
char *dynstr = base + shdr[dynstr_idx].sh_offset;
char *str = base + shdr[str_idx].sh_offset;
int symbol_count = 0;
int dynsymbol_count = 0;
if (dynsym_idx != -1) {
// Iterate through the dynamic symbol table, and count how many symbols
// are actually defined
for(i = 0; i < dynnumsyms; i++) {
if(dynsyms[i].st_shndx != SHN_UNDEF) {
dynsymbol_count++;
}
}
XLOG("Dynamic Symbol count: %d\n", dynsymbol_count);
}
if (sym_idx != -1) {
// Iterate through the symbol table, and count how many symbols
// are actually defined
for(i = 0; i < numsyms; i++) {
if((syms[i].st_shndx != SHN_UNDEF) &&
(strlen(str+syms[i].st_name)) &&
(syms[i].st_value != 0) && (syms[i].st_size != 0)) {
symbol_count++;
}
}
XLOG("Symbol count: %d\n", symbol_count);
}
// Now, create an entry in our symbol table structure for each symbol...
table->num_symbols += symbol_count + dynsymbol_count;;
table->symbols = malloc(table->num_symbols * sizeof(struct symbol));
if(!table->symbols) {
free(table);
table = NULL;
goto out_unmap;
}
int j = 0;
if (dynsym_idx != -1) {
// ...and populate them
for(i = 0; i < dynnumsyms; i++) {
if(dynsyms[i].st_shndx != SHN_UNDEF) {
table->symbols[j].name = strdup(dynstr + dynsyms[i].st_name);
table->symbols[j].addr = dynsyms[i].st_value;
table->symbols[j].size = dynsyms[i].st_size;
XLOG("name: %s, addr: %x, size: %x\n",
table->symbols[j].name, table->symbols[j].addr, table->symbols[j].size);
j++;
}
}
}
if (sym_idx != -1) {
// ...and populate them
for(i = 0; i < numsyms; i++) {
if((syms[i].st_shndx != SHN_UNDEF) &&
(strlen(str+syms[i].st_name)) &&
(syms[i].st_value != 0) && (syms[i].st_size != 0)) {
table->symbols[j].name = strdup(str + syms[i].st_name);
table->symbols[j].addr = syms[i].st_value;
table->symbols[j].size = syms[i].st_size;
XLOG("name: %s, addr: %x, size: %x\n",
table->symbols[j].name, table->symbols[j].addr, table->symbols[j].size);
j++;
}
}
}
// Sort the symbol table entries, so they can be bsearched later
qsort(table->symbols, table->num_symbols, sizeof(struct symbol), qcompar);
out_unmap:
munmap(base, length);
out_close:
close(fd);
out:
return table;
}
/*
* Free a symbol table
*
* Parameters:
* table - Table to free
*/
void symbol_table_free(struct symbol_table *table)
{
int i;
if(!table) {
return;
}
for(i=0; i<table->num_symbols; i++) {
free(table->symbols[i].name);
}
free(table->symbols);
free(table);
}
/*
* Search for an address in the symbol table
*
* Parameters:
* table - Table to search in
* addr - Address to search for.
*
* Returns:
* A pointer to the Symbol structure corresponding to the
* symbol which contains this address, or NULL if no symbol
* contains it.
*/
const struct symbol *symbol_table_lookup(struct symbol_table *table, unsigned int addr)
{
if(!table) {
return NULL;
}
return bsearch((void*)addr, table->symbols, table->num_symbols, sizeof(struct symbol), bcompar);
}

20
debuggerd/symbol_table.h Normal file
View file

@ -0,0 +1,20 @@
#ifndef SYMBOL_TABLE_H
#define SYMBOL_TABLE_H
struct symbol {
unsigned int addr;
unsigned int size;
char *name;
};
struct symbol_table {
struct symbol *symbols;
int num_symbols;
char *name;
};
struct symbol_table *symbol_table_create(const char *filename);
void symbol_table_free(struct symbol_table *table);
const struct symbol *symbol_table_lookup(struct symbol_table *table, unsigned int addr);
#endif

7
debuggerd/utility.c
View file

@ -38,14 +38,14 @@ void get_remote_struct(int pid, void *src, void *dst, size_t size)
unsigned int i;
for (i = 0; i+4 <= size; i+=4) {
*(int *)(dst+i) = ptrace(PTRACE_PEEKTEXT, pid, src+i, NULL);
*(int *)((char *)dst+i) = ptrace(PTRACE_PEEKTEXT, pid, (char *)src+i, NULL);
}
if (i < size) {
int val;
assert((size - i) < 4);
val = ptrace(PTRACE_PEEKTEXT, pid, src+i, NULL);
val = ptrace(PTRACE_PEEKTEXT, pid, (char *)src+i, NULL);
while (i < size) {
((unsigned char *)dst)[i] = val & 0xff;
i++;
@ -69,11 +69,12 @@ const char *map_to_name(mapinfo *mi, unsigned pc, const char* def)
/* Find the containing map info for the pc */
const mapinfo *pc_to_mapinfo(mapinfo *mi, unsigned pc, unsigned *rel_pc)
{
*rel_pc = pc;
while(mi) {
if((pc >= mi->start) && (pc < mi->end)){
// Only calculate the relative offset for shared libraries
if (strstr(mi->name, ".so")) {
*rel_pc = pc - mi->start;
*rel_pc -= mi->start;
}
return mi;
}

10
debuggerd/utility.h
View file

@ -21,6 +21,8 @@
#include <stddef.h>
#include <stdbool.h>
#include "symbol_table.h"
#ifndef PT_ARM_EXIDX
#define PT_ARM_EXIDX 0x70000001 /* .ARM.exidx segment */
#endif
@ -33,6 +35,7 @@ typedef struct mapinfo {
unsigned end;
unsigned exidx_start;
unsigned exidx_end;
struct symbol_table *symbols;
char name[];
} mapinfo;
@ -53,4 +56,11 @@ const char *map_to_name(mapinfo *mi, unsigned pc, const char* def);
/* Log information onto the tombstone */
extern void _LOG(int tfd, bool in_tombstone_only, const char *fmt, ...);
#define LOG(fmt...) _LOG(-1, 0, fmt)
#if 0
#define XLOG(fmt...) _LOG(-1, 0, fmt)
#else
#define XLOG(fmt...) do {} while(0)
#endif
#endif

15
debuggerd/x86/crashglue.S Normal file
View file

@ -0,0 +1,15 @@
.globl crash1
.globl crashnostack
crash1:
movl $0xa5a50000, %eax
movl $0xa5a50001, %ebx
movl $0xa5a50002, %ecx
movl $0, %edx
jmp *%edx
crashnostack:
movl $0, %ebp
jmp *%ebp

61
debuggerd/x86/machine.c Normal file
View file

@ -0,0 +1,61 @@
/* system/debuggerd/debuggerd.c
**
** Copyright 2006, The Android Open Source Project
**
** Licensed under the Apache License, Version 2.0 (the "License");
** you may not use this file except in compliance with the License.
** You may obtain a copy of the License at
**
** http://www.apache.org/licenses/LICENSE-2.0
**
** Unless required by applicable law or agreed to in writing, software
** distributed under the License is distributed on an "AS IS" BASIS,
** WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
** See the License for the specific language governing permissions and
** limitations under the License.
*/
#include <stdio.h>
#include <errno.h>
#include <signal.h>
#include <pthread.h>
#include <fcntl.h>
#include <sys/types.h>
#include <dirent.h>
#include <sys/ptrace.h>
#include <sys/wait.h>
#include <sys/exec_elf.h>
#include <sys/stat.h>
#include <cutils/sockets.h>
#include <cutils/properties.h>
#include <linux/input.h>
#include "../utility.h"
#include "x86_utility.h"
void dump_registers(int tfd, int pid, bool at_fault)
{
struct pt_regs_x86 r;
bool only_in_tombstone = !at_fault;
if(ptrace(PTRACE_GETREGS, pid, 0, &r)) {
_LOG(tfd, only_in_tombstone,
"cannot get registers: %s\n", strerror(errno));
return;
}
//if there is no stack, no print just like arm
if(!r.ebp)
return;
_LOG(tfd, only_in_tombstone, " eax %08x ebx %08x ecx %08x edx %08x\n",
r.eax, r.ebx, r.ecx, r.edx);
_LOG(tfd, only_in_tombstone, " esi %08x edi %08x\n",
r.esi, r.edi);
_LOG(tfd, only_in_tombstone, " xcs %08x xds %08x xes %08x xfs %08x xss %08x\n",
r.xcs, r.xds, r.xes, r.xfs, r.xss);
_LOG(tfd, only_in_tombstone,
" eip %08x ebp %08x esp %08x flags %08x\n",
r.eip, r.ebp, r.esp, r.eflags);
}

85
debuggerd/x86/unwind.c Normal file
View file

@ -0,0 +1,85 @@
#include <cutils/logd.h>
#include <sys/ptrace.h>
#include "../utility.h"
#include "x86_utility.h"
int unwind_backtrace_with_ptrace_x86(int tfd, pid_t pid, mapinfo *map,
bool at_fault)
{
struct pt_regs_x86 r;
unsigned int stack_level = 0;
unsigned int stack_depth = 0;
unsigned int rel_pc;
unsigned int stack_ptr;
unsigned int stack_content;
if(ptrace(PTRACE_GETREGS, pid, 0, &r)) return 0;
unsigned int eip = (unsigned int)r.eip;
unsigned int ebp = (unsigned int)r.ebp;
unsigned int cur_sp = (unsigned int)r.esp;
const mapinfo *mi;
const struct symbol* sym = 0;
//ebp==0, it indicates that the stack is poped to the bottom or there is no stack at all.
while (ebp) {
_LOG(tfd, !at_fault, "#0%d ",stack_level);
mi = pc_to_mapinfo(map, eip, &rel_pc);
/* See if we can determine what symbol this stack frame resides in */
if (mi != 0 && mi->symbols != 0) {
sym = symbol_table_lookup(mi->symbols, rel_pc);
}
if (sym) {
_LOG(tfd, !at_fault, " eip: %08x %s (%s)\n", eip, mi ? mi->name : "", sym->name);
} else {
_LOG(tfd, !at_fault, " eip: %08x %s\n", eip, mi ? mi->name : "");
}
stack_level++;
if (stack_level >= STACK_DEPTH || eip == 0)
break;
eip = ptrace(PTRACE_PEEKTEXT, pid, (void*)(ebp + 4), NULL);
ebp = ptrace(PTRACE_PEEKTEXT, pid, (void*)ebp, NULL);
}
ebp = (unsigned int)r.ebp;
stack_depth = stack_level;
stack_level = 0;
if (ebp)
_LOG(tfd, !at_fault, "stack: \n");
while (ebp) {
_LOG(tfd, !at_fault, "#0%d \n",stack_level);
stack_ptr = cur_sp;
while((int)(ebp - stack_ptr) >= 0) {
stack_content = ptrace(PTRACE_PEEKTEXT, pid, (void*)stack_ptr, NULL);
mi = pc_to_mapinfo(map, stack_content, &rel_pc);
/* See if we can determine what symbol this stack frame resides in */
if (mi != 0 && mi->symbols != 0) {
sym = symbol_table_lookup(mi->symbols, rel_pc);
}
if (sym) {
_LOG(tfd, !at_fault, " %08x %08x %s (%s)\n",
stack_ptr, stack_content, mi ? mi->name : "", sym->name);
} else {
_LOG(tfd, !at_fault, " %08x %08x %s\n", stack_ptr, stack_content, mi ? mi->name : "");
}
stack_ptr = stack_ptr + 4;
//the stack frame may be very deep.
if((int)(stack_ptr - cur_sp) >= STACK_FRAME_DEPTH) {
_LOG(tfd, !at_fault, " ...... ...... \n");
break;
}
}
cur_sp = ebp + 4;
stack_level++;
if (stack_level >= STACK_DEPTH || stack_level >= stack_depth)
break;
ebp = ptrace(PTRACE_PEEKTEXT, pid, (void*)ebp, NULL);
}
return stack_depth;
}

40
debuggerd/x86/x86_utility.h Normal file
View file

@ -0,0 +1,40 @@
/*
**
** Copyright 2006, The Android Open Source Project
**
** Licensed under the Apache License, Version 2.0 (the "License");
** you may not use this file except in compliance with the License.
** You may obtain a copy of the License at
**
** http://www.apache.org/licenses/LICENSE-2.0
**
** Unless required by applicable law or agreed to in writing, software
** distributed under the License is distributed on an "AS IS" BASIS,
** WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
** See the License for the specific language governing permissions and
** limitations under the License.
*/
#define STACK_DEPTH 8
#define STACK_FRAME_DEPTH 64
typedef struct pt_regs_x86 {
long ebx;
long ecx;
long edx;
long esi;
long edi;
long ebp;
long eax;
int xds;
int xes;
int xfs;
int xgs;
long orig_eax;
long eip;
int xcs;
long eflags;
long esp;
int xss;
}pt_regs_x86;

2
include/arch/target_linux-x86/AndroidConfig.h
View file

@ -227,7 +227,7 @@
/*
* Define if we have Linux's dbus
*/
#define HAVE_DBUS 1
/* #define HAVE_DBUS 1 */
/*
* Define if tm struct has tm_gmtoff field

1
include/private/android_filesystem_config.h
View file

@ -173,6 +173,7 @@ static struct fs_path_config android_files[] = {
{ 00640, AID_SYSTEM, AID_SYSTEM, "system/etc/bluetooth/auto_pairing.conf" },
{ 00444, AID_RADIO, AID_AUDIO, "system/etc/AudioPara4.csv" },
{ 00555, AID_ROOT, AID_ROOT, "system/etc/ppp/*" },
{ 00555, AID_ROOT, AID_ROOT, "system/etc/rc.*" },
{ 00644, AID_SYSTEM, AID_SYSTEM, "data/app/*" },
{ 00644, AID_SYSTEM, AID_SYSTEM, "data/app-private/*" },
{ 00644, AID_APP, AID_APP, "data/data/*" },

7
libcutils/Android.mk
View file

@ -112,12 +112,17 @@ LOCAL_MODULE := libcutils
LOCAL_SRC_FILES := $(commonSources) ashmem-dev.c mq.c
ifeq ($(TARGET_ARCH),arm)
LOCAL_SRC_FILES += memset32.S
LOCAL_SRC_FILES += arch-arm/memset32.S
else # !arm
ifeq ($(TARGET_ARCH),sh)
LOCAL_SRC_FILES += memory.c atomic-android-sh.c
else # !sh
ifeq ($(TARGET_ARCH_VARIANT),x86-atom)
LOCAL_CFLAGS += -DHAVE_MEMSET16 -DHAVE_MEMSET32
LOCAL_SRC_FILES += arch-x86/android_memset16.S arch-x86/android_memset32.S memory.c
else # !x86-atom
LOCAL_SRC_FILES += memory.c
endif # !x86-atom
endif # !sh
endif # !arm

0
libcutils/memset32.S → libcutils/arch-arm/memset32.S
View file

32
libcutils/arch-x86/android_memset16.S Normal file
View file

@ -0,0 +1,32 @@
/*
* Copyright (C) 2010 The Android Open Source Project
*
* Licensed under the Apache License, Version 2.0 (the "License");
* you may not use this file except in compliance with the License.
* You may obtain a copy of the License at
*
* http://www.apache.org/licenses/LICENSE-2.0
*
* Unless required by applicable law or agreed to in writing, software
* distributed under the License is distributed on an "AS IS" BASIS,
* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
* See the License for the specific language governing permissions and
* limitations under the License.
*/
/*
* Contributed by: Intel Corporation
*/
#if defined(USE_SSE2)
# include "cache_wrapper.S"
# undef __i686
# define USE_AS_ANDROID
# define sse2_memset16_atom android_memset16
# include "sse2-memset16-atom.S"
#else
# include "memset16.S"
#endif

33
libcutils/arch-x86/android_memset32.S Normal file
View file

@ -0,0 +1,33 @@
/*
* Copyright (C) 2010 The Android Open Source Project
*
* Licensed under the Apache License, Version 2.0 (the "License");
* you may not use this file except in compliance with the License.
* You may obtain a copy of the License at
*
* http://www.apache.org/licenses/LICENSE-2.0
*
* Unless required by applicable law or agreed to in writing, software
* distributed under the License is distributed on an "AS IS" BASIS,
* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
* See the License for the specific language governing permissions and
* limitations under the License.
*/
/*
* Contributed by: Intel Corporation
*/
#if defined(USE_SSE2)
# include "cache_wrapper.S"
# undef __i686
# define USE_AS_ANDROID
# define sse2_memset32_atom android_memset32
# include "sse2-memset32-atom.S"
#else
# include "memset32.S"
#endif

24
libcutils/arch-x86/cache_wrapper.S Normal file
View file

@ -0,0 +1,24 @@
/*
* Copyright (C) 2010 The Android Open Source Project
*
* Licensed under the Apache License, Version 2.0 (the "License");
* you may not use this file except in compliance with the License.
* You may obtain a copy of the License at
*
* http://www.apache.org/licenses/LICENSE-2.0
*
* Unless required by applicable law or agreed to in writing, software
* distributed under the License is distributed on an "AS IS" BASIS,
* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
* See the License for the specific language governing permissions and
* limitations under the License.
*/
/*
* Contributed by: Intel Corporation
*/
/* Values are optimized for Atom */
#define SHARED_CACHE_SIZE (512*1024) /* Atom L2 Cache */
#define DATA_CACHE_SIZE (24*1024) /* Atom L1 Data Cache */
#define SHARED_CACHE_SIZE_HALF (SHARED_CACHE_SIZE / 2)
#define DATA_CACHE_SIZE_HALF (DATA_CACHE_SIZE / 2)

722
libcutils/arch-x86/sse2-memset16-atom.S Normal file
View file

@ -0,0 +1,722 @@
/*
* Copyright (C) 2010 The Android Open Source Project
*
* Licensed under the Apache License, Version 2.0 (the "License");
* you may not use this file except in compliance with the License.
* You may obtain a copy of the License at
*
* http://www.apache.org/licenses/LICENSE-2.0
*
* Unless required by applicable law or agreed to in writing, software
* distributed under the License is distributed on an "AS IS" BASIS,
* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
* See the License for the specific language governing permissions and
* limitations under the License.
*/
/*
* Contributed by: Intel Corporation
*/
#ifndef L
# define L(label) .L##label
#endif
#ifndef ALIGN
# define ALIGN(n) .p2align n
#endif
#ifndef cfi_startproc
# define cfi_startproc .cfi_startproc
#endif
#ifndef cfi_endproc
# define cfi_endproc .cfi_endproc
#endif
#ifndef cfi_rel_offset
# define cfi_rel_offset(reg, off) .cfi_rel_offset reg, off
#endif
#ifndef cfi_restore
# define cfi_restore(reg) .cfi_restore reg
#endif
#ifndef cfi_adjust_cfa_offset
# define cfi_adjust_cfa_offset(off) .cfi_adjust_cfa_offset off
#endif
#ifndef ENTRY
# define ENTRY(name) \
.type name, @function; \
.globl name; \
.p2align 4; \
name: \
cfi_startproc
#endif
#ifndef END
# define END(name) \
cfi_endproc; \
.size name, .-name
#endif
#define CFI_PUSH(REG) \
cfi_adjust_cfa_offset (4); \
cfi_rel_offset (REG, 0)
#define CFI_POP(REG) \
cfi_adjust_cfa_offset (-4); \
cfi_restore (REG)
#define PUSH(REG) pushl REG; CFI_PUSH (REG)
#define POP(REG) popl REG; CFI_POP (REG)
#ifdef USE_AS_BZERO16
# define DEST PARMS
# define LEN DEST+4
#else
# define DEST PARMS
# define CHR DEST+4
# define LEN CHR+4
#endif
#if 1
# define SETRTNVAL
#else
# define SETRTNVAL movl DEST(%esp), %eax
#endif
#ifdef SHARED
# define ENTRANCE PUSH (%ebx);
# define RETURN_END POP (%ebx); ret
# define RETURN RETURN_END; CFI_PUSH (%ebx)
# define PARMS 8 /* Preserve EBX. */
# define JMPTBL(I, B) I - B
/* Load an entry in a jump table into EBX and branch to it. TABLE is a
jump table with relative offsets. */
# define BRANCH_TO_JMPTBL_ENTRY(TABLE) \
/* We first load PC into EBX. */ \
call __i686.get_pc_thunk.bx; \
/* Get the address of the jump table. */ \
add $(TABLE - .), %ebx; \
/* Get the entry and convert the relative offset to the \
absolute address. */ \
add (%ebx,%ecx,4), %ebx; \
/* We loaded the jump table and adjuested EDX. Go. */ \
jmp *%ebx
.section .gnu.linkonce.t.__i686.get_pc_thunk.bx,"ax",@progbits
.globl __i686.get_pc_thunk.bx
.hidden __i686.get_pc_thunk.bx
ALIGN (4)
.type __i686.get_pc_thunk.bx,@function
__i686.get_pc_thunk.bx:
movl (%esp), %ebx
ret
#else
# define ENTRANCE
# define RETURN_END ret
# define RETURN RETURN_END
# define PARMS 4
# define JMPTBL(I, B) I
/* Branch to an entry in a jump table. TABLE is a jump table with
absolute offsets. */
# define BRANCH_TO_JMPTBL_ENTRY(TABLE) \
jmp *TABLE(,%ecx,4)
#endif
.section .text.sse2,"ax",@progbits
ALIGN (4)
ENTRY (sse2_memset16_atom)
ENTRANCE
movl LEN(%esp), %ecx
#ifdef USE_AS_ANDROID
shr $1, %ecx
#endif
#ifdef USE_AS_BZERO16
xor %eax, %eax
#else
movzwl CHR(%esp), %eax
mov %eax, %edx
shl $16, %eax
or %edx, %eax
#endif
movl DEST(%esp), %edx
cmp $32, %ecx
jae L(32wordsormore)
L(write_less32words):
lea (%edx, %ecx, 2), %edx
BRANCH_TO_JMPTBL_ENTRY (L(table_less32words))
.pushsection .rodata.sse2,"a",@progbits
ALIGN (2)
L(table_less32words):
.int JMPTBL (L(write_0words), L(table_less32words))
.int JMPTBL (L(write_1words), L(table_less32words))
.int JMPTBL (L(write_2words), L(table_less32words))
.int JMPTBL (L(write_3words), L(table_less32words))
.int JMPTBL (L(write_4words), L(table_less32words))
.int JMPTBL (L(write_5words), L(table_less32words))
.int JMPTBL (L(write_6words), L(table_less32words))
.int JMPTBL (L(write_7words), L(table_less32words))
.int JMPTBL (L(write_8words), L(table_less32words))
.int JMPTBL (L(write_9words), L(table_less32words))
.int JMPTBL (L(write_10words), L(table_less32words))
.int JMPTBL (L(write_11words), L(table_less32words))
.int JMPTBL (L(write_12words), L(table_less32words))
.int JMPTBL (L(write_13words), L(table_less32words))
.int JMPTBL (L(write_14words), L(table_less32words))
.int JMPTBL (L(write_15words), L(table_less32words))
.int JMPTBL (L(write_16words), L(table_less32words))
.int JMPTBL (L(write_17words), L(table_less32words))
.int JMPTBL (L(write_18words), L(table_less32words))
.int JMPTBL (L(write_19words), L(table_less32words))
.int JMPTBL (L(write_20words), L(table_less32words))
.int JMPTBL (L(write_21words), L(table_less32words))
.int JMPTBL (L(write_22words), L(table_less32words))
.int JMPTBL (L(write_23words), L(table_less32words))
.int JMPTBL (L(write_24words), L(table_less32words))
.int JMPTBL (L(write_25words), L(table_less32words))
.int JMPTBL (L(write_26words), L(table_less32words))
.int JMPTBL (L(write_27words), L(table_less32words))
.int JMPTBL (L(write_28words), L(table_less32words))
.int JMPTBL (L(write_29words), L(table_less32words))
.int JMPTBL (L(write_30words), L(table_less32words))
.int JMPTBL (L(write_31words), L(table_less32words))
.popsection
ALIGN (4)
L(write_28words):
movl %eax, -56(%edx)
movl %eax, -52(%edx)
L(write_24words):
movl %eax, -48(%edx)
movl %eax, -44(%edx)
L(write_20words):
movl %eax, -40(%edx)
movl %eax, -36(%edx)
L(write_16words):
movl %eax, -32(%edx)
movl %eax, -28(%edx)
L(write_12words):
movl %eax, -24(%edx)
movl %eax, -20(%edx)
L(write_8words):
movl %eax, -16(%edx)
movl %eax, -12(%edx)
L(write_4words):
movl %eax, -8(%edx)
movl %eax, -4(%edx)
L(write_0words):
SETRTNVAL
RETURN
ALIGN (4)
L(write_29words):
movl %eax, -58(%edx)
movl %eax, -54(%edx)
L(write_25words):
movl %eax, -50(%edx)
movl %eax, -46(%edx)
L(write_21words):
movl %eax, -42(%edx)
movl %eax, -38(%edx)
L(write_17words):
movl %eax, -34(%edx)
movl %eax, -30(%edx)
L(write_13words):
movl %eax, -26(%edx)
movl %eax, -22(%edx)
L(write_9words):
movl %eax, -18(%edx)
movl %eax, -14(%edx)
L(write_5words):
movl %eax, -10(%edx)
movl %eax, -6(%edx)
L(write_1words):
mov %ax, -2(%edx)
SETRTNVAL
RETURN
ALIGN (4)
L(write_30words):
movl %eax, -60(%edx)
movl %eax, -56(%edx)
L(write_26words):
movl %eax, -52(%edx)
movl %eax, -48(%edx)
L(write_22words):
movl %eax, -44(%edx)
movl %eax, -40(%edx)
L(write_18words):
movl %eax, -36(%edx)
movl %eax, -32(%edx)
L(write_14words):
movl %eax, -28(%edx)
movl %eax, -24(%edx)
L(write_10words):
movl %eax, -20(%edx)
movl %eax, -16(%edx)
L(write_6words):
movl %eax, -12(%edx)
movl %eax, -8(%edx)
L(write_2words):
movl %eax, -4(%edx)
SETRTNVAL
RETURN
ALIGN (4)
L(write_31words):
movl %eax, -62(%edx)
movl %eax, -58(%edx)
L(write_27words):
movl %eax, -54(%edx)
movl %eax, -50(%edx)
L(write_23words):
movl %eax, -46(%edx)
movl %eax, -42(%edx)
L(write_19words):
movl %eax, -38(%edx)
movl %eax, -34(%edx)
L(write_15words):
movl %eax, -30(%edx)
movl %eax, -26(%edx)
L(write_11words):
movl %eax, -22(%edx)
movl %eax, -18(%edx)
L(write_7words):
movl %eax, -14(%edx)
movl %eax, -10(%edx)
L(write_3words):
movl %eax, -6(%edx)
movw %ax, -2(%edx)
SETRTNVAL
RETURN
ALIGN (4)
L(32wordsormore):
shl $1, %ecx
test $0x01, %edx
jz L(aligned2bytes)
mov %eax, (%edx)
mov %eax, -4(%edx, %ecx)
sub $2, %ecx
add $1, %edx
rol $8, %eax
L(aligned2bytes):
#ifdef USE_AS_BZERO16
pxor %xmm0, %xmm0
#else
movd %eax, %xmm0
pshufd $0, %xmm0, %xmm0
#endif
testl $0xf, %edx
jz L(aligned_16)
/* ECX > 32 and EDX is not 16 byte aligned. */
L(not_aligned_16):
movdqu %xmm0, (%edx)
movl %edx, %eax
and $-16, %edx
add $16, %edx
sub %edx, %eax
add %eax, %ecx
movd %xmm0, %eax
ALIGN (4)
L(aligned_16):
cmp $128, %ecx
jae L(128bytesormore)
L(aligned_16_less128bytes):
add %ecx, %edx
shr $1, %ecx
BRANCH_TO_JMPTBL_ENTRY (L(table_16_128bytes))
ALIGN (4)
L(128bytesormore):
#ifdef SHARED_CACHE_SIZE
PUSH (%ebx)
mov $SHARED_CACHE_SIZE, %ebx
#else
# ifdef SHARED
call __i686.get_pc_thunk.bx
add $_GLOBAL_OFFSET_TABLE_, %ebx
mov __x86_shared_cache_size@GOTOFF(%ebx), %ebx
# else
PUSH (%ebx)
mov __x86_shared_cache_size, %ebx
# endif
#endif
cmp %ebx, %ecx
jae L(128bytesormore_nt_start)
#ifdef DATA_CACHE_SIZE
POP (%ebx)
# define RESTORE_EBX_STATE CFI_PUSH (%ebx)
cmp $DATA_CACHE_SIZE, %ecx
#else
# ifdef SHARED
# define RESTORE_EBX_STATE
call __i686.get_pc_thunk.bx
add $_GLOBAL_OFFSET_TABLE_, %ebx
cmp __x86_data_cache_size@GOTOFF(%ebx), %ecx
# else
POP (%ebx)
# define RESTORE_EBX_STATE CFI_PUSH (%ebx)
cmp __x86_data_cache_size, %ecx
# endif
#endif
jae L(128bytes_L2_normal)
subl $128, %ecx
L(128bytesormore_normal):
sub $128, %ecx
movdqa %xmm0, (%edx)
movdqa %xmm0, 0x10(%edx)
movdqa %xmm0, 0x20(%edx)
movdqa %xmm0, 0x30(%edx)
movdqa %xmm0, 0x40(%edx)
movdqa %xmm0, 0x50(%edx)
movdqa %xmm0, 0x60(%edx)
movdqa %xmm0, 0x70(%edx)
lea 128(%edx), %edx
jb L(128bytesless_normal)
sub $128, %ecx
movdqa %xmm0, (%edx)
movdqa %xmm0, 0x10(%edx)
movdqa %xmm0, 0x20(%edx)
movdqa %xmm0, 0x30(%edx)
movdqa %xmm0, 0x40(%edx)
movdqa %xmm0, 0x50(%edx)
movdqa %xmm0, 0x60(%edx)
movdqa %xmm0, 0x70(%edx)
lea 128(%edx), %edx
jae L(128bytesormore_normal)
L(128bytesless_normal):
lea 128(%ecx), %ecx
add %ecx, %edx
shr $1, %ecx
BRANCH_TO_JMPTBL_ENTRY (L(table_16_128bytes))
ALIGN (4)
L(128bytes_L2_normal):
prefetcht0 0x380(%edx)
prefetcht0 0x3c0(%edx)
sub $128, %ecx
movdqa %xmm0, (%edx)
movaps %xmm0, 0x10(%edx)
movaps %xmm0, 0x20(%edx)
movaps %xmm0, 0x30(%edx)
movaps %xmm0, 0x40(%edx)
movaps %xmm0, 0x50(%edx)
movaps %xmm0, 0x60(%edx)
movaps %xmm0, 0x70(%edx)
add $128, %edx
cmp $128, %ecx
jae L(128bytes_L2_normal)
L(128bytesless_L2_normal):
add %ecx, %edx
shr $1, %ecx
BRANCH_TO_JMPTBL_ENTRY (L(table_16_128bytes))
RESTORE_EBX_STATE
L(128bytesormore_nt_start):
sub %ebx, %ecx
mov %ebx, %eax
and $0x7f, %eax
add %eax, %ecx
movd %xmm0, %eax
ALIGN (4)
L(128bytesormore_shared_cache_loop):
prefetcht0 0x3c0(%edx)
prefetcht0 0x380(%edx)
sub $0x80, %ebx
movdqa %xmm0, (%edx)
movdqa %xmm0, 0x10(%edx)
movdqa %xmm0, 0x20(%edx)
movdqa %xmm0, 0x30(%edx)
movdqa %xmm0, 0x40(%edx)
movdqa %xmm0, 0x50(%edx)
movdqa %xmm0, 0x60(%edx)
movdqa %xmm0, 0x70(%edx)
add $0x80, %edx
cmp $0x80, %ebx
jae L(128bytesormore_shared_cache_loop)
cmp $0x80, %ecx
jb L(shared_cache_loop_end)
ALIGN (4)
L(128bytesormore_nt):
sub $0x80, %ecx
movntdq %xmm0, (%edx)
movntdq %xmm0, 0x10(%edx)
movntdq %xmm0, 0x20(%edx)
movntdq %xmm0, 0x30(%edx)
movntdq %xmm0, 0x40(%edx)
movntdq %xmm0, 0x50(%edx)
movntdq %xmm0, 0x60(%edx)
movntdq %xmm0, 0x70(%edx)
add $0x80, %edx
cmp $0x80, %ecx
jae L(128bytesormore_nt)
sfence
L(shared_cache_loop_end):
#if defined DATA_CACHE_SIZE || !defined SHARED
POP (%ebx)
#endif
add %ecx, %edx
shr $1, %ecx
BRANCH_TO_JMPTBL_ENTRY (L(table_16_128bytes))
.pushsection .rodata.sse2,"a",@progbits
ALIGN (2)
L(table_16_128bytes):
.int JMPTBL (L(aligned_16_0bytes), L(table_16_128bytes))
.int JMPTBL (L(aligned_16_2bytes), L(table_16_128bytes))
.int JMPTBL (L(aligned_16_4bytes), L(table_16_128bytes))
.int JMPTBL (L(aligned_16_6bytes), L(table_16_128bytes))
.int JMPTBL (L(aligned_16_8bytes), L(table_16_128bytes))
.int JMPTBL (L(aligned_16_10bytes), L(table_16_128bytes))
.int JMPTBL (L(aligned_16_12bytes), L(table_16_128bytes))
.int JMPTBL (L(aligned_16_14bytes), L(table_16_128bytes))
.int JMPTBL (L(aligned_16_16bytes), L(table_16_128bytes))
.int JMPTBL (L(aligned_16_18bytes), L(table_16_128bytes))
.int JMPTBL (L(aligned_16_20bytes), L(table_16_128bytes))
.int JMPTBL (L(aligned_16_22bytes), L(table_16_128bytes))
.int JMPTBL (L(aligned_16_24bytes), L(table_16_128bytes))
.int JMPTBL (L(aligned_16_26bytes), L(table_16_128bytes))
.int JMPTBL (L(aligned_16_28bytes), L(table_16_128bytes))
.int JMPTBL (L(aligned_16_30bytes), L(table_16_128bytes))
.int JMPTBL (L(aligned_16_32bytes), L(table_16_128bytes))
.int JMPTBL (L(aligned_16_34bytes), L(table_16_128bytes))
.int JMPTBL (L(aligned_16_36bytes), L(table_16_128bytes))
.int JMPTBL (L(aligned_16_38bytes), L(table_16_128bytes))
.int JMPTBL (L(aligned_16_40bytes), L(table_16_128bytes))
.int JMPTBL (L(aligned_16_42bytes), L(table_16_128bytes))
.int JMPTBL (L(aligned_16_44bytes), L(table_16_128bytes))
.int JMPTBL (L(aligned_16_46bytes), L(table_16_128bytes))
.int JMPTBL (L(aligned_16_48bytes), L(table_16_128bytes))
.int JMPTBL (L(aligned_16_50bytes), L(table_16_128bytes))
.int JMPTBL (L(aligned_16_52bytes), L(table_16_128bytes))
.int JMPTBL (L(aligned_16_54bytes), L(table_16_128bytes))
.int JMPTBL (L(aligned_16_56bytes), L(table_16_128bytes))
.int JMPTBL (L(aligned_16_58bytes), L(table_16_128bytes))
.int JMPTBL (L(aligned_16_60bytes), L(table_16_128bytes))
.int JMPTBL (L(aligned_16_62bytes), L(table_16_128bytes))
.int JMPTBL (L(aligned_16_64bytes), L(table_16_128bytes))
.int JMPTBL (L(aligned_16_66bytes), L(table_16_128bytes))
.int JMPTBL (L(aligned_16_68bytes), L(table_16_128bytes))
.int JMPTBL (L(aligned_16_70bytes), L(table_16_128bytes))
.int JMPTBL (L(aligned_16_72bytes), L(table_16_128bytes))
.int JMPTBL (L(aligned_16_74bytes), L(table_16_128bytes))
.int JMPTBL (L(aligned_16_76bytes), L(table_16_128bytes))
.int JMPTBL (L(aligned_16_78bytes), L(table_16_128bytes))
.int JMPTBL (L(aligned_16_80bytes), L(table_16_128bytes))
.int JMPTBL (L(aligned_16_82bytes), L(table_16_128bytes))
.int JMPTBL (L(aligned_16_84bytes), L(table_16_128bytes))
.int JMPTBL (L(aligned_16_86bytes), L(table_16_128bytes))
.int JMPTBL (L(aligned_16_88bytes), L(table_16_128bytes))
.int JMPTBL (L(aligned_16_90bytes), L(table_16_128bytes))
.int JMPTBL (L(aligned_16_92bytes), L(table_16_128bytes))
.int JMPTBL (L(aligned_16_94bytes), L(table_16_128bytes))
.int JMPTBL (L(aligned_16_96bytes), L(table_16_128bytes))
.int JMPTBL (L(aligned_16_98bytes), L(table_16_128bytes))
.int JMPTBL (L(aligned_16_100bytes), L(table_16_128bytes))
.int JMPTBL (L(aligned_16_102bytes), L(table_16_128bytes))
.int JMPTBL (L(aligned_16_104bytes), L(table_16_128bytes))
.int JMPTBL (L(aligned_16_106bytes), L(table_16_128bytes))
.int JMPTBL (L(aligned_16_108bytes), L(table_16_128bytes))
.int JMPTBL (L(aligned_16_110bytes), L(table_16_128bytes))
.int JMPTBL (L(aligned_16_112bytes), L(table_16_128bytes))
.int JMPTBL (L(aligned_16_114bytes), L(table_16_128bytes))
.int JMPTBL (L(aligned_16_116bytes), L(table_16_128bytes))
.int JMPTBL (L(aligned_16_118bytes), L(table_16_128bytes))
.int JMPTBL (L(aligned_16_120bytes), L(table_16_128bytes))
.int JMPTBL (L(aligned_16_122bytes), L(table_16_128bytes))
.int JMPTBL (L(aligned_16_124bytes), L(table_16_128bytes))
.int JMPTBL (L(aligned_16_126bytes), L(table_16_128bytes))
.popsection
ALIGN (4)
L(aligned_16_112bytes):
movdqa %xmm0, -112(%edx)
L(aligned_16_96bytes):
movdqa %xmm0, -96(%edx)
L(aligned_16_80bytes):
movdqa %xmm0, -80(%edx)
L(aligned_16_64bytes):
movdqa %xmm0, -64(%edx)
L(aligned_16_48bytes):
movdqa %xmm0, -48(%edx)
L(aligned_16_32bytes):
movdqa %xmm0, -32(%edx)
L(aligned_16_16bytes):
movdqa %xmm0, -16(%edx)
L(aligned_16_0bytes):
SETRTNVAL
RETURN
ALIGN (4)
L(aligned_16_114bytes):
movdqa %xmm0, -114(%edx)
L(aligned_16_98bytes):
movdqa %xmm0, -98(%edx)
L(aligned_16_82bytes):
movdqa %xmm0, -82(%edx)
L(aligned_16_66bytes):
movdqa %xmm0, -66(%edx)
L(aligned_16_50bytes):
movdqa %xmm0, -50(%edx)
L(aligned_16_34bytes):
movdqa %xmm0, -34(%edx)
L(aligned_16_18bytes):
movdqa %xmm0, -18(%edx)
L(aligned_16_2bytes):
movw %ax, -2(%edx)
SETRTNVAL
RETURN
ALIGN (4)
L(aligned_16_116bytes):
movdqa %xmm0, -116(%edx)
L(aligned_16_100bytes):
movdqa %xmm0, -100(%edx)
L(aligned_16_84bytes):
movdqa %xmm0, -84(%edx)
L(aligned_16_68bytes):
movdqa %xmm0, -68(%edx)
L(aligned_16_52bytes):
movdqa %xmm0, -52(%edx)
L(aligned_16_36bytes):
movdqa %xmm0, -36(%edx)
L(aligned_16_20bytes):
movdqa %xmm0, -20(%edx)
L(aligned_16_4bytes):
movl %eax, -4(%edx)
SETRTNVAL
RETURN
ALIGN (4)
L(aligned_16_118bytes):
movdqa %xmm0, -118(%edx)
L(aligned_16_102bytes):
movdqa %xmm0, -102(%edx)
L(aligned_16_86bytes):
movdqa %xmm0, -86(%edx)
L(aligned_16_70bytes):
movdqa %xmm0, -70(%edx)
L(aligned_16_54bytes):
movdqa %xmm0, -54(%edx)
L(aligned_16_38bytes):
movdqa %xmm0, -38(%edx)
L(aligned_16_22bytes):
movdqa %xmm0, -22(%edx)
L(aligned_16_6bytes):
movl %eax, -6(%edx)
movw %ax, -2(%edx)
SETRTNVAL
RETURN
ALIGN (4)
L(aligned_16_120bytes):
movdqa %xmm0, -120(%edx)
L(aligned_16_104bytes):
movdqa %xmm0, -104(%edx)
L(aligned_16_88bytes):
movdqa %xmm0, -88(%edx)
L(aligned_16_72bytes):
movdqa %xmm0, -72(%edx)
L(aligned_16_56bytes):
movdqa %xmm0, -56(%edx)
L(aligned_16_40bytes):
movdqa %xmm0, -40(%edx)
L(aligned_16_24bytes):
movdqa %xmm0, -24(%edx)
L(aligned_16_8bytes):
movq %xmm0, -8(%edx)
SETRTNVAL
RETURN
ALIGN (4)
L(aligned_16_122bytes):
movdqa %xmm0, -122(%edx)
L(aligned_16_106bytes):
movdqa %xmm0, -106(%edx)
L(aligned_16_90bytes):
movdqa %xmm0, -90(%edx)
L(aligned_16_74bytes):
movdqa %xmm0, -74(%edx)
L(aligned_16_58bytes):
movdqa %xmm0, -58(%edx)
L(aligned_16_42bytes):
movdqa %xmm0, -42(%edx)
L(aligned_16_26bytes):
movdqa %xmm0, -26(%edx)
L(aligned_16_10bytes):
movq %xmm0, -10(%edx)
movw %ax, -2(%edx)
SETRTNVAL
RETURN
ALIGN (4)
L(aligned_16_124bytes):
movdqa %xmm0, -124(%edx)
L(aligned_16_108bytes):
movdqa %xmm0, -108(%edx)
L(aligned_16_92bytes):
movdqa %xmm0, -92(%edx)
L(aligned_16_76bytes):
movdqa %xmm0, -76(%edx)
L(aligned_16_60bytes):
movdqa %xmm0, -60(%edx)
L(aligned_16_44bytes):
movdqa %xmm0, -44(%edx)
L(aligned_16_28bytes):
movdqa %xmm0, -28(%edx)
L(aligned_16_12bytes):
movq %xmm0, -12(%edx)
movl %eax, -4(%edx)
SETRTNVAL
RETURN
ALIGN (4)
L(aligned_16_126bytes):
movdqa %xmm0, -126(%edx)
L(aligned_16_110bytes):
movdqa %xmm0, -110(%edx)
L(aligned_16_94bytes):
movdqa %xmm0, -94(%edx)
L(aligned_16_78bytes):
movdqa %xmm0, -78(%edx)
L(aligned_16_62bytes):
movdqa %xmm0, -62(%edx)
L(aligned_16_46bytes):
movdqa %xmm0, -46(%edx)
L(aligned_16_30bytes):
movdqa %xmm0, -30(%edx)
L(aligned_16_14bytes):
movq %xmm0, -14(%edx)
movl %eax, -6(%edx)
movw %ax, -2(%edx)
SETRTNVAL
RETURN
END (sse2_memset16_atom)

513
libcutils/arch-x86/sse2-memset32-atom.S Normal file
View file

@ -0,0 +1,513 @@
/*
* Copyright (C) 2010 The Android Open Source Project
*
* Licensed under the Apache License, Version 2.0 (the "License");
* you may not use this file except in compliance with the License.
* You may obtain a copy of the License at
*
* http://www.apache.org/licenses/LICENSE-2.0
*
* Unless required by applicable law or agreed to in writing, software
* distributed under the License is distributed on an "AS IS" BASIS,
* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
* See the License for the specific language governing permissions and
* limitations under the License.
*/
/*
* Contributed by: Intel Corporation
*/
#ifndef L
# define L(label) .L##label
#endif
#ifndef ALIGN
# define ALIGN(n) .p2align n
#endif
#ifndef cfi_startproc
# define cfi_startproc .cfi_startproc
#endif
#ifndef cfi_endproc
# define cfi_endproc .cfi_endproc
#endif
#ifndef cfi_rel_offset
# define cfi_rel_offset(reg, off) .cfi_rel_offset reg, off
#endif
#ifndef cfi_restore
# define cfi_restore(reg) .cfi_restore reg
#endif
#ifndef cfi_adjust_cfa_offset
# define cfi_adjust_cfa_offset(off) .cfi_adjust_cfa_offset off
#endif
#ifndef ENTRY
# define ENTRY(name) \
.type name, @function; \
.globl name; \
.p2align 4; \
name: \
cfi_startproc
#endif
#ifndef END
# define END(name) \
cfi_endproc; \
.size name, .-name
#endif
#define CFI_PUSH(REG) \
cfi_adjust_cfa_offset (4); \
cfi_rel_offset (REG, 0)
#define CFI_POP(REG) \
cfi_adjust_cfa_offset (-4); \
cfi_restore (REG)
#define PUSH(REG) pushl REG; CFI_PUSH (REG)
#define POP(REG) popl REG; CFI_POP (REG)
#ifdef USE_AS_BZERO32
# define DEST PARMS
# define LEN DEST+4
#else
# define DEST PARMS
# define DWDS DEST+4
# define LEN DWDS+4
#endif
#ifdef USE_AS_WMEMSET32
# define SETRTNVAL movl DEST(%esp), %eax
#else
# define SETRTNVAL
#endif
#ifdef SHARED
# define ENTRANCE PUSH (%ebx);
# define RETURN_END POP (%ebx); ret
# define RETURN RETURN_END; CFI_PUSH (%ebx)
# define PARMS 8 /* Preserve EBX. */
# define JMPTBL(I, B) I - B
/* Load an entry in a jump table into EBX and branch to it. TABLE is a
jump table with relative offsets. */
# define BRANCH_TO_JMPTBL_ENTRY(TABLE) \
/* We first load PC into EBX. */ \
call __i686.get_pc_thunk.bx; \
/* Get the address of the jump table. */ \
add $(TABLE - .), %ebx; \
/* Get the entry and convert the relative offset to the \
absolute address. */ \
add (%ebx,%ecx,4), %ebx; \
/* We loaded the jump table and adjuested EDX. Go. */ \
jmp *%ebx
.section .gnu.linkonce.t.__i686.get_pc_thunk.bx,"ax",@progbits
.globl __i686.get_pc_thunk.bx
.hidden __i686.get_pc_thunk.bx
ALIGN (4)
.type __i686.get_pc_thunk.bx,@function
__i686.get_pc_thunk.bx:
movl (%esp), %ebx
ret
#else
# define ENTRANCE
# define RETURN_END ret
# define RETURN RETURN_END
# define PARMS 4
# define JMPTBL(I, B) I
/* Branch to an entry in a jump table. TABLE is a jump table with
absolute offsets. */
# define BRANCH_TO_JMPTBL_ENTRY(TABLE) \
jmp *TABLE(,%ecx,4)
#endif
.section .text.sse2,"ax",@progbits
ALIGN (4)
ENTRY (sse2_memset32_atom)
ENTRANCE
movl LEN(%esp), %ecx
#ifdef USE_AS_ANDROID
shr $2, %ecx
#endif
#ifdef USE_AS_BZERO32
xor %eax, %eax
#else
mov DWDS(%esp), %eax
mov %eax, %edx
#endif
movl DEST(%esp), %edx
cmp $16, %ecx
jae L(16dbwordsormore)
L(write_less16dbwords):
lea (%edx, %ecx, 4), %edx
BRANCH_TO_JMPTBL_ENTRY (L(table_less16dbwords))
.pushsection .rodata.sse2,"a",@progbits
ALIGN (2)
L(table_less16dbwords):
.int JMPTBL (L(write_0dbwords), L(table_less16dbwords))
.int JMPTBL (L(write_1dbwords), L(table_less16dbwords))
.int JMPTBL (L(write_2dbwords), L(table_less16dbwords))
.int JMPTBL (L(write_3dbwords), L(table_less16dbwords))
.int JMPTBL (L(write_4dbwords), L(table_less16dbwords))
.int JMPTBL (L(write_5dbwords), L(table_less16dbwords))
.int JMPTBL (L(write_6dbwords), L(table_less16dbwords))
.int JMPTBL (L(write_7dbwords), L(table_less16dbwords))
.int JMPTBL (L(write_8dbwords), L(table_less16dbwords))
.int JMPTBL (L(write_9dbwords), L(table_less16dbwords))
.int JMPTBL (L(write_10dbwords), L(table_less16dbwords))
.int JMPTBL (L(write_11dbwords), L(table_less16dbwords))
.int JMPTBL (L(write_12dbwords), L(table_less16dbwords))
.int JMPTBL (L(write_13dbwords), L(table_less16dbwords))
.int JMPTBL (L(write_14dbwords), L(table_less16dbwords))
.int JMPTBL (L(write_15dbwords), L(table_less16dbwords))
.popsection
ALIGN (4)
L(write_15dbwords):
movl %eax, -60(%edx)
L(write_14dbwords):
movl %eax, -56(%edx)
L(write_13dbwords):
movl %eax, -52(%edx)
L(write_12dbwords):
movl %eax, -48(%edx)
L(write_11dbwords):
movl %eax, -44(%edx)
L(write_10dbwords):
movl %eax, -40(%edx)
L(write_9dbwords):
movl %eax, -36(%edx)
L(write_8dbwords):
movl %eax, -32(%edx)
L(write_7dbwords):
movl %eax, -28(%edx)
L(write_6dbwords):
movl %eax, -24(%edx)
L(write_5dbwords):
movl %eax, -20(%edx)
L(write_4dbwords):
movl %eax, -16(%edx)
L(write_3dbwords):
movl %eax, -12(%edx)
L(write_2dbwords):
movl %eax, -8(%edx)
L(write_1dbwords):
movl %eax, -4(%edx)
L(write_0dbwords):
SETRTNVAL
RETURN
ALIGN (4)
L(16dbwordsormore):
test $3, %edx
jz L(aligned4bytes)
mov %eax, (%edx)
mov %eax, -4(%edx, %ecx, 4)
sub $1, %ecx
rol $24, %eax
add $1, %edx
test $3, %edx
jz L(aligned4bytes)
ror $8, %eax
add $1, %edx
test $3, %edx
jz L(aligned4bytes)
ror $8, %eax
add $1, %edx
L(aligned4bytes):
shl $2, %ecx
#ifdef USE_AS_BZERO32
pxor %xmm0, %xmm0
#else
movd %eax, %xmm0
pshufd $0, %xmm0, %xmm0
#endif
testl $0xf, %edx
jz L(aligned_16)
/* ECX > 32 and EDX is not 16 byte aligned. */
L(not_aligned_16):
movdqu %xmm0, (%edx)
movl %edx, %eax
and $-16, %edx
add $16, %edx
sub %edx, %eax
add %eax, %ecx
movd %xmm0, %eax
ALIGN (4)
L(aligned_16):
cmp $128, %ecx
jae L(128bytesormore)
L(aligned_16_less128bytes):
add %ecx, %edx
shr $2, %ecx
BRANCH_TO_JMPTBL_ENTRY (L(table_16_128bytes))
ALIGN (4)
L(128bytesormore):
#ifdef SHARED_CACHE_SIZE
PUSH (%ebx)
mov $SHARED_CACHE_SIZE, %ebx
#else
# ifdef SHARED
call __i686.get_pc_thunk.bx
add $_GLOBAL_OFFSET_TABLE_, %ebx
mov __x86_shared_cache_size@GOTOFF(%ebx), %ebx
# else
PUSH (%ebx)
mov __x86_shared_cache_size, %ebx
# endif
#endif
cmp %ebx, %ecx
jae L(128bytesormore_nt_start)
#ifdef DATA_CACHE_SIZE
POP (%ebx)
# define RESTORE_EBX_STATE CFI_PUSH (%ebx)
cmp $DATA_CACHE_SIZE, %ecx
#else
# ifdef SHARED
# define RESTORE_EBX_STATE
call __i686.get_pc_thunk.bx
add $_GLOBAL_OFFSET_TABLE_, %ebx
cmp __x86_data_cache_size@GOTOFF(%ebx), %ecx
# else
POP (%ebx)
# define RESTORE_EBX_STATE CFI_PUSH (%ebx)
cmp __x86_data_cache_size, %ecx
# endif
#endif
jae L(128bytes_L2_normal)
subl $128, %ecx
L(128bytesormore_normal):
sub $128, %ecx
movdqa %xmm0, (%edx)
movdqa %xmm0, 0x10(%edx)
movdqa %xmm0, 0x20(%edx)
movdqa %xmm0, 0x30(%edx)
movdqa %xmm0, 0x40(%edx)
movdqa %xmm0, 0x50(%edx)
movdqa %xmm0, 0x60(%edx)
movdqa %xmm0, 0x70(%edx)
lea 128(%edx), %edx
jb L(128bytesless_normal)
sub $128, %ecx
movdqa %xmm0, (%edx)
movdqa %xmm0, 0x10(%edx)
movdqa %xmm0, 0x20(%edx)
movdqa %xmm0, 0x30(%edx)
movdqa %xmm0, 0x40(%edx)
movdqa %xmm0, 0x50(%edx)
movdqa %xmm0, 0x60(%edx)
movdqa %xmm0, 0x70(%edx)
lea 128(%edx), %edx
jae L(128bytesormore_normal)
L(128bytesless_normal):
lea 128(%ecx), %ecx
add %ecx, %edx
shr $2, %ecx
BRANCH_TO_JMPTBL_ENTRY (L(table_16_128bytes))
ALIGN (4)
L(128bytes_L2_normal):
prefetcht0 0x380(%edx)
prefetcht0 0x3c0(%edx)
sub $128, %ecx
movdqa %xmm0, (%edx)
movaps %xmm0, 0x10(%edx)
movaps %xmm0, 0x20(%edx)
movaps %xmm0, 0x30(%edx)
movaps %xmm0, 0x40(%edx)
movaps %xmm0, 0x50(%edx)
movaps %xmm0, 0x60(%edx)
movaps %xmm0, 0x70(%edx)
add $128, %edx
cmp $128, %ecx
jae L(128bytes_L2_normal)
L(128bytesless_L2_normal):
add %ecx, %edx
shr $2, %ecx
BRANCH_TO_JMPTBL_ENTRY (L(table_16_128bytes))
RESTORE_EBX_STATE
L(128bytesormore_nt_start):
sub %ebx, %ecx
mov %ebx, %eax
and $0x7f, %eax
add %eax, %ecx
movd %xmm0, %eax
ALIGN (4)
L(128bytesormore_shared_cache_loop):
prefetcht0 0x3c0(%edx)
prefetcht0 0x380(%edx)
sub $0x80, %ebx
movdqa %xmm0, (%edx)
movdqa %xmm0, 0x10(%edx)
movdqa %xmm0, 0x20(%edx)
movdqa %xmm0, 0x30(%edx)
movdqa %xmm0, 0x40(%edx)
movdqa %xmm0, 0x50(%edx)
movdqa %xmm0, 0x60(%edx)
movdqa %xmm0, 0x70(%edx)
add $0x80, %edx
cmp $0x80, %ebx
jae L(128bytesormore_shared_cache_loop)
cmp $0x80, %ecx
jb L(shared_cache_loop_end)
ALIGN (4)
L(128bytesormore_nt):
sub $0x80, %ecx
movntdq %xmm0, (%edx)
movntdq %xmm0, 0x10(%edx)
movntdq %xmm0, 0x20(%edx)
movntdq %xmm0, 0x30(%edx)
movntdq %xmm0, 0x40(%edx)
movntdq %xmm0, 0x50(%edx)
movntdq %xmm0, 0x60(%edx)
movntdq %xmm0, 0x70(%edx)
add $0x80, %edx
cmp $0x80, %ecx
jae L(128bytesormore_nt)
sfence
L(shared_cache_loop_end):
#if defined DATA_CACHE_SIZE || !defined SHARED
POP (%ebx)
#endif
add %ecx, %edx
shr $2, %ecx
BRANCH_TO_JMPTBL_ENTRY (L(table_16_128bytes))
.pushsection .rodata.sse2,"a",@progbits
ALIGN (2)
L(table_16_128bytes):
.int JMPTBL (L(aligned_16_0bytes), L(table_16_128bytes))
.int JMPTBL (L(aligned_16_4bytes), L(table_16_128bytes))
.int JMPTBL (L(aligned_16_8bytes), L(table_16_128bytes))
.int JMPTBL (L(aligned_16_12bytes), L(table_16_128bytes))
.int JMPTBL (L(aligned_16_16bytes), L(table_16_128bytes))
.int JMPTBL (L(aligned_16_20bytes), L(table_16_128bytes))
.int JMPTBL (L(aligned_16_24bytes), L(table_16_128bytes))
.int JMPTBL (L(aligned_16_28bytes), L(table_16_128bytes))
.int JMPTBL (L(aligned_16_32bytes), L(table_16_128bytes))
.int JMPTBL (L(aligned_16_36bytes), L(table_16_128bytes))
.int JMPTBL (L(aligned_16_40bytes), L(table_16_128bytes))
.int JMPTBL (L(aligned_16_44bytes), L(table_16_128bytes))
.int JMPTBL (L(aligned_16_48bytes), L(table_16_128bytes))
.int JMPTBL (L(aligned_16_52bytes), L(table_16_128bytes))
.int JMPTBL (L(aligned_16_56bytes), L(table_16_128bytes))
.int JMPTBL (L(aligned_16_60bytes), L(table_16_128bytes))
.int JMPTBL (L(aligned_16_64bytes), L(table_16_128bytes))
.int JMPTBL (L(aligned_16_68bytes), L(table_16_128bytes))
.int JMPTBL (L(aligned_16_72bytes), L(table_16_128bytes))
.int JMPTBL (L(aligned_16_76bytes), L(table_16_128bytes))
.int JMPTBL (L(aligned_16_80bytes), L(table_16_128bytes))
.int JMPTBL (L(aligned_16_84bytes), L(table_16_128bytes))
.int JMPTBL (L(aligned_16_88bytes), L(table_16_128bytes))
.int JMPTBL (L(aligned_16_92bytes), L(table_16_128bytes))
.int JMPTBL (L(aligned_16_96bytes), L(table_16_128bytes))
.int JMPTBL (L(aligned_16_100bytes), L(table_16_128bytes))
.int JMPTBL (L(aligned_16_104bytes), L(table_16_128bytes))
.int JMPTBL (L(aligned_16_108bytes), L(table_16_128bytes))
.int JMPTBL (L(aligned_16_112bytes), L(table_16_128bytes))
.int JMPTBL (L(aligned_16_116bytes), L(table_16_128bytes))
.int JMPTBL (L(aligned_16_120bytes), L(table_16_128bytes))
.int JMPTBL (L(aligned_16_124bytes), L(table_16_128bytes))
.popsection
ALIGN (4)
L(aligned_16_112bytes):
movdqa %xmm0, -112(%edx)
L(aligned_16_96bytes):
movdqa %xmm0, -96(%edx)
L(aligned_16_80bytes):
movdqa %xmm0, -80(%edx)
L(aligned_16_64bytes):
movdqa %xmm0, -64(%edx)
L(aligned_16_48bytes):
movdqa %xmm0, -48(%edx)
L(aligned_16_32bytes):
movdqa %xmm0, -32(%edx)
L(aligned_16_16bytes):
movdqa %xmm0, -16(%edx)
L(aligned_16_0bytes):
SETRTNVAL
RETURN
ALIGN (4)
L(aligned_16_116bytes):
movdqa %xmm0, -116(%edx)
L(aligned_16_100bytes):
movdqa %xmm0, -100(%edx)
L(aligned_16_84bytes):
movdqa %xmm0, -84(%edx)
L(aligned_16_68bytes):
movdqa %xmm0, -68(%edx)
L(aligned_16_52bytes):
movdqa %xmm0, -52(%edx)
L(aligned_16_36bytes):
movdqa %xmm0, -36(%edx)
L(aligned_16_20bytes):
movdqa %xmm0, -20(%edx)
L(aligned_16_4bytes):
movl %eax, -4(%edx)
SETRTNVAL
RETURN
ALIGN (4)
L(aligned_16_120bytes):
movdqa %xmm0, -120(%edx)
L(aligned_16_104bytes):
movdqa %xmm0, -104(%edx)
L(aligned_16_88bytes):
movdqa %xmm0, -88(%edx)
L(aligned_16_72bytes):
movdqa %xmm0, -72(%edx)
L(aligned_16_56bytes):
movdqa %xmm0, -56(%edx)
L(aligned_16_40bytes):
movdqa %xmm0, -40(%edx)
L(aligned_16_24bytes):
movdqa %xmm0, -24(%edx)
L(aligned_16_8bytes):
movq %xmm0, -8(%edx)
SETRTNVAL
RETURN
ALIGN (4)
L(aligned_16_124bytes):
movdqa %xmm0, -124(%edx)
L(aligned_16_108bytes):
movdqa %xmm0, -108(%edx)
L(aligned_16_92bytes):
movdqa %xmm0, -92(%edx)
L(aligned_16_76bytes):
movdqa %xmm0, -76(%edx)
L(aligned_16_60bytes):
movdqa %xmm0, -60(%edx)
L(aligned_16_44bytes):
movdqa %xmm0, -44(%edx)
L(aligned_16_28bytes):
movdqa %xmm0, -28(%edx)
L(aligned_16_12bytes):
movq %xmm0, -12(%edx)
movl %eax, -4(%edx)
SETRTNVAL
RETURN
END (sse2_memset32_atom)

4
libcutils/memory.c
View file

@ -16,6 +16,7 @@
#include <cutils/memory.h>
#if !HAVE_MEMSET16
void android_memset16(uint16_t* dst, uint16_t value, size_t size)
{
size >>= 1;
@ -23,7 +24,9 @@ void android_memset16(uint16_t* dst, uint16_t value, size_t size)
*dst++ = value;
}
}
#endif
#if !HAVE_MEMSET32
void android_memset32(uint32_t* dst, uint32_t value, size_t size)
{
size >>= 2;
@ -31,6 +34,7 @@ void android_memset32(uint32_t* dst, uint32_t value, size_t size)
*dst++ = value;
}
}
#endif
#if !HAVE_STRLCPY
/*

18
libpixelflinger/Android.mk
View file

@ -1,17 +1,6 @@
LOCAL_PATH:= $(call my-dir)
include $(CLEAR_VARS)
#
# ARMv6 specific objects
#
ifeq ($(TARGET_ARCH),arm)
LOCAL_ASFLAGS := -march=armv6
LOCAL_SRC_FILES := rotate90CW_4x4_16v6.S
LOCAL_MODULE := libpixelflinger_armv6
include $(BUILD_STATIC_LIBRARY)
endif
#
# C/C++ and ARMv5 objects
#
@ -77,10 +66,6 @@ ifneq ($(BUILD_TINY_ANDROID),true)
LOCAL_SHARED_LIBRARIES += libhardware_legacy
LOCAL_CFLAGS += -DWITH_LIB_HARDWARE
endif
ifeq ($(TARGET_ARCH),arm)
LOCAL_WHOLE_STATIC_LIBRARIES := libpixelflinger_armv6
endif
include $(BUILD_SHARED_LIBRARY)
#
@ -91,9 +76,6 @@ include $(CLEAR_VARS)
LOCAL_MODULE:= libpixelflinger_static
LOCAL_SRC_FILES := $(PIXELFLINGER_SRC_FILES)
LOCAL_CFLAGS := $(PIXELFLINGER_CFLAGS)
ifeq ($(TARGET_ARCH),arm)
LOCAL_WHOLE_STATIC_LIBRARIES := libpixelflinger_armv6
endif
include $(BUILD_STATIC_LIBRARY)

12
libpixelflinger/codeflinger/ARMAssembler.cpp
View file

@ -334,7 +334,7 @@ void ARMAssembler::LDM(int cc, int dir,
void ARMAssembler::STM(int cc, int dir,
int Rn, int W, uint32_t reg_list)
{ // FA EA FD ED IB IA DB DA
{ // ED FD EA FA IB IA DB DA
const uint8_t P[8] = { 0, 1, 0, 1, 1, 0, 1, 0 };
const uint8_t U[8] = { 0, 0, 1, 1, 1, 1, 0, 0 };
*mPC++ = (cc<<28) | (4<<25) | (uint32_t(P[dir])<<24) |
@ -433,6 +433,16 @@ void ARMAssembler::UXTB16(int cc, int Rd, int Rm, int rotate)
{
*mPC++ = (cc<<28) | 0x6CF0070 | (Rd<<12) | ((rotate >> 3) << 10) | Rm;
}
#if 0
#pragma mark -
#pragma mark Bit manipulation (ARMv7+ only)...
#endif
// Bit manipulation (ARMv7+ only)...
void ARMAssembler::UBFX(int cc, int Rd, int Rn, int lsb, int width)
{
*mPC++ = (cc<<28) | 0x7E00000 | ((width-1)<<16) | (Rd<<12) | (lsb<<7) | 0x50 | Rn;
}
}; // namespace android

1
libpixelflinger/codeflinger/ARMAssembler.h
View file

@ -124,6 +124,7 @@ public:
virtual void SMLAW(int cc, int y,
int Rd, int Rm, int Rs, int Rn);
virtual void UXTB16(int cc, int Rd, int Rm, int rotate);
virtual void UBFX(int cc, int Rd, int Rn, int lsb, int width);
private:
ARMAssembler(const ARMAssembler& rhs);

3
libpixelflinger/codeflinger/ARMAssemblerInterface.h
View file

@ -206,6 +206,9 @@ public:
// byte/half word extract...
virtual void UXTB16(int cc, int Rd, int Rm, int rotate) = 0;
// bit manipulation...
virtual void UBFX(int cc, int Rd, int Rn, int lsb, int width) = 0;
// -----------------------------------------------------------------------
// convenience...
// -----------------------------------------------------------------------

4
libpixelflinger/codeflinger/ARMAssemblerProxy.cpp
View file

@ -199,5 +199,9 @@ void ARMAssemblerProxy::UXTB16(int cc, int Rd, int Rm, int rotate) {
mTarget->UXTB16(cc, Rd, Rm, rotate);
}
void ARMAssemblerProxy::UBFX(int cc, int Rd, int Rn, int lsb, int width) {
mTarget->UBFX(cc, Rd, Rn, lsb, width);
}
}; // namespace android

1
libpixelflinger/codeflinger/ARMAssemblerProxy.h
View file

@ -115,6 +115,7 @@ public:
int Rd, int Rm, int Rs, int Rn);
virtual void UXTB16(int cc, int Rd, int Rm, int rotate);
virtual void UBFX(int cc, int Rd, int Rn, int lsb, int width);
private:
ARMAssemblerInterface* mTarget;

11
libpixelflinger/codeflinger/disassem.c
View file

@ -81,6 +81,8 @@
* g - 2nd fp operand (register) (bits 16-18)
* h - 3rd fp operand (register/immediate) (bits 0-4)
* j - xtb rotate literal (bits 10-11)
* i - bfx lsb literal (bits 7-11)
* w - bfx width literal (bits 16-20)
* b - branch address
* t - thumb branch address (bits 24, 0-23)
* k - breakpoint comment (bits 0-3, 8-19)
@ -124,6 +126,7 @@ static const struct arm32_insn arm32_i[] = {
{ 0x0fe000f0, 0x00a00090, "umlal", "Sdnms" },
{ 0x0fe000f0, 0x00e00090, "smlal", "Sdnms" },
{ 0x0fff03f0, 0x06cf0070, "uxtb16", "dmj" },
{ 0x0fe00070, 0x07e00050, "ubfx", "dmiw" },
{ 0x0d700000, 0x04200000, "strt", "daW" },
{ 0x0d700000, 0x04300000, "ldrt", "daW" },
{ 0x0d700000, 0x04600000, "strbt", "daW" },
@ -412,6 +415,14 @@ disasm(const disasm_interface_t *di, u_int loc, int altfmt)
case 'j':
di->di_printf("ror #%d", ((insn >> 10) & 3) << 3);
break;
/* i - bfx lsb literal (bits 7-11) */
case 'i':
di->di_printf("#%d", (insn >> 7) & 31);
break;
/* w - bfx width literal (bits 16-20) */
case 'w':
di->di_printf("#%d", 1 + ((insn >> 16) & 31));
break;
/* b - branch address */
case 'b':
branch = ((insn << 2) & 0x03ffffff);

20
libpixelflinger/codeflinger/load_store.cpp
View file

@ -18,9 +18,12 @@
#include <assert.h>
#include <stdio.h>
#include <cutils/log.h>
#include "codeflinger/GGLAssembler.h"
#ifdef __ARM_ARCH__
#include <machine/cpu-features.h>
#endif
namespace android {
// ----------------------------------------------------------------------------
@ -110,6 +113,20 @@ void GGLAssembler::extract(integer_t& d, int s, int h, int l, int bits)
assert(maskLen<=8);
assert(h);
#if __ARM_ARCH__ >= 7
const int mask = (1<<maskLen)-1;
if ((h == bits) && !l && (s != d.reg)) {
MOV(AL, 0, d.reg, s); // component = packed;
} else if ((h == bits) && l) {
MOV(AL, 0, d.reg, reg_imm(s, LSR, l)); // component = packed >> l;
} else if (!l && isValidImmediate(mask)) {
AND(AL, 0, d.reg, s, imm(mask)); // component = packed & mask;
} else if (!l && isValidImmediate(~mask)) {
BIC(AL, 0, d.reg, s, imm(~mask)); // component = packed & mask;
} else {
UBFX(AL, d.reg, s, l, maskLen); // component = (packed & mask) >> l;
}
#else
if (h != bits) {
const int mask = ((1<<maskLen)-1) << l;
if (isValidImmediate(mask)) {
@ -132,6 +149,7 @@ void GGLAssembler::extract(integer_t& d, int s, int h, int l, int bits)
if (s != d.reg) {
MOV(AL, 0, d.reg, s);
}
#endif
d.s = maskLen;
}

31
libpixelflinger/col32cb16blend.S
View file

@ -1,20 +1,19 @@
/* libs/pixelflinger/col32cb16blend.S
**
** (C) COPYRIGHT 2009 ARM Limited.
**
** Licensed under the Apache License, Version 2.0 (the "License");
** you may not use this file except in compliance with the License.
** You may obtain a copy of the License at
**
** http://www.apache.org/licenses/LICENSE-2.0
**
** Unless required by applicable law or agreed to in writing, software
** distributed under the License is distributed on an "AS IS" BASIS,
** WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
** See the License for the specific language governing permissions and
** limitations under the License.
**
*/
*
* Copyright (C) 2009 The Android Open Source Project
*
* Licensed under the Apache License, Version 2.0 (the "License");
* you may not use this file except in compliance with the License.
* You may obtain a copy of the License at
*
* http://www.apache.org/licenses/LICENSE-2.0
*
* Unless required by applicable law or agreed to in writing, software
* distributed under the License is distributed on an "AS IS" BASIS,
* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
* See the License for the specific language governing permissions and
* limitations under the License.
*/
.text
.align

32
libpixelflinger/col32cb16blend_neon.S
View file

@ -1,20 +1,20 @@
/* libs/pixelflinger/col32cb16blend_neon.S
**
** (C) COPYRIGHT 2009 ARM Limited.
**
** Licensed under the Apache License, Version 2.0 (the "License");
** you may not use this file except in compliance with the License.
** You may obtain a copy of the License at
**
** http://www.apache.org/licenses/LICENSE-2.0
**
** Unless required by applicable law or agreed to in writing, software
** distributed under the License is distributed on an "AS IS" BASIS,
** WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
** See the License for the specific language governing permissions and
** limitations under the License.
**
*/
*
* Copyright (C) 2009 The Android Open Source Project
*
* Licensed under the Apache License, Version 2.0 (the "License");
* you may not use this file except in compliance with the License.
* You may obtain a copy of the License at
*
* http://www.apache.org/licenses/LICENSE-2.0
*
* Unless required by applicable law or agreed to in writing, software
* distributed under the License is distributed on an "AS IS" BASIS,
* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
* See the License for the specific language governing permissions and
* limitations under the License.
*/
.text
.align

2
libpixelflinger/raster.cpp
View file

@ -143,7 +143,7 @@ void ggl_copyPixels(void* con, GGLint xs, GGLint ys,
using namespace android;
GGLint gglBitBlti(GGLContext* con, int tmu, GGLint crop[4], GGLint where[4])
GGLint gglBitBlit(GGLContext* con, int tmu, GGLint crop[4], GGLint where[4])
{
GGL_CONTEXT(c, (void*)con);

909
libpixelflinger/scanline.cpp
View file

@ -1,6 +1,6 @@
/* libs/pixelflinger/scanline.cpp
**
** Copyright 2006, The Android Open Source Project
** Copyright 2006-2011, The Android Open Source Project
**
** Licensed under the Apache License, Version 2.0 (the "License");
** you may not use this file except in compliance with the License.
@ -57,6 +57,11 @@
#define DEBUG__CODEGEN_ONLY 0
/* Set to 1 to dump to the log the states that need a new
* code-generated scanline callback, i.e. those that don't
* have a corresponding shortcut function.
*/
#define DEBUG_NEEDS 0
#define ASSEMBLY_SCRATCH_SIZE 2048
@ -79,8 +84,21 @@ static void scanline(context_t* c);
static void scanline_perspective(context_t* c);
static void scanline_perspective_single(context_t* c);
static void scanline_t32cb16blend(context_t* c);
static void scanline_t32cb16blend_dither(context_t* c);
static void scanline_t32cb16blend_srca(context_t* c);
static void scanline_t32cb16blend_clamp(context_t* c);
static void scanline_t32cb16blend_clamp_dither(context_t* c);
static void scanline_t32cb16blend_clamp_mod(context_t* c);
static void scanline_x32cb16blend_clamp_mod(context_t* c);
static void scanline_t32cb16blend_clamp_mod_dither(context_t* c);
static void scanline_x32cb16blend_clamp_mod_dither(context_t* c);
static void scanline_t32cb16(context_t* c);
static void scanline_t32cb16_dither(context_t* c);
static void scanline_t32cb16_clamp(context_t* c);
static void scanline_t32cb16_clamp_dither(context_t* c);
static void scanline_col32cb16blend(context_t* c);
static void scanline_t16cb16_clamp(context_t* c);
static void scanline_t16cb16blend_clamp_mod(context_t* c);
static void scanline_memcpy(context_t* c);
static void scanline_memset8(context_t* c);
static void scanline_memset16(context_t* c);
@ -99,6 +117,13 @@ extern "C" void scanline_col32cb16blend_arm(uint16_t *dst, uint32_t col, size_t
// ----------------------------------------------------------------------------
static inline uint16_t convertAbgr8888ToRgb565(uint32_t pix)
{
return uint16_t( ((pix << 8) & 0xf800) |
((pix >> 5) & 0x07e0) |
((pix >> 19) & 0x001f) );
}
struct shortcut_t {
needs_filter_t filter;
const char* desc;
@ -107,13 +132,95 @@ struct shortcut_t {
};
// Keep in sync with needs
/* To understand the values here, have a look at:
* system/core/include/private/pixelflinger/ggl_context.h
*
* Especially the lines defining and using GGL_RESERVE_NEEDS
*
* Quick reminders:
* - the last nibble of the first value is the destination buffer format.
* - the last nibble of the third value is the source texture format
* - formats: 4=rgb565 1=abgr8888 2=xbgr8888
*
* In the descriptions below:
*
* SRC means we copy the source pixels to the destination
*
* SRC_OVER means we blend the source pixels to the destination
* with dstFactor = 1-srcA, srcFactor=1 (premultiplied source).
* This mode is otherwise called 'blend'.
*
* SRCA_OVER means we blend the source pixels to the destination
* with dstFactor=srcA*(1-srcA) srcFactor=srcA (non-premul source).
* This mode is otherwise called 'blend_srca'
*
* clamp means we fetch source pixels from a texture with u/v clamping
*
* mod means the source pixels are modulated (multiplied) by the
* a/r/g/b of the current context's color. Typically used for
* fade-in / fade-out.
*
* dither means we dither 32 bit values to 16 bits
*/
static shortcut_t shortcuts[] = {
{ { { 0x03515104, 0x00000077, { 0x00000A01, 0x00000000 } },
{ 0xFFFFFFFF, 0xFFFFFFFF, { 0xFFFFFFFF, 0x0000003F } } },
"565 fb, 8888 tx, blend", scanline_t32cb16blend, init_y_noop },
"565 fb, 8888 tx, blend SRC_OVER", scanline_t32cb16blend, init_y_noop },
{ { { 0x03010104, 0x00000077, { 0x00000A01, 0x00000000 } },
{ 0xFFFFFFFF, 0xFFFFFFFF, { 0xFFFFFFFF, 0x0000003F } } },
"565 fb, 8888 tx", scanline_t32cb16, init_y_noop },
"565 fb, 8888 tx, SRC", scanline_t32cb16, init_y_noop },
/* same as first entry, but with dithering */
{ { { 0x03515104, 0x00000177, { 0x00000A01, 0x00000000 } },
{ 0xFFFFFFFF, 0xFFFFFFFF, { 0xFFFFFFFF, 0x0000003F } } },
"565 fb, 8888 tx, blend SRC_OVER dither", scanline_t32cb16blend_dither, init_y_noop },
/* same as second entry, but with dithering */
{ { { 0x03010104, 0x00000177, { 0x00000A01, 0x00000000 } },
{ 0xFFFFFFFF, 0xFFFFFFFF, { 0xFFFFFFFF, 0x0000003F } } },
"565 fb, 8888 tx, SRC dither", scanline_t32cb16_dither, init_y_noop },
/* this is used during the boot animation - CHEAT: ignore dithering */
{ { { 0x03545404, 0x00000077, { 0x00000A01, 0x00000000 } },
{ 0xFFFFFFFF, 0xFFFFFEFF, { 0xFFFFFFFF, 0x0000003F } } },
"565 fb, 8888 tx, blend dst:ONE_MINUS_SRCA src:SRCA", scanline_t32cb16blend_srca, init_y_noop },
/* special case for arbitrary texture coordinates (think scaling) */
{ { { 0x03515104, 0x00000077, { 0x00000001, 0x00000000 } },
{ 0xFFFFFFFF, 0xFFFFFFFF, { 0xFFFFFFFF, 0x0000003F } } },
"565 fb, 8888 tx, SRC_OVER clamp", scanline_t32cb16blend_clamp, init_y },
{ { { 0x03515104, 0x00000177, { 0x00000001, 0x00000000 } },
{ 0xFFFFFFFF, 0xFFFFFFFF, { 0xFFFFFFFF, 0x0000003F } } },
"565 fb, 8888 tx, SRC_OVER clamp dither", scanline_t32cb16blend_clamp_dither, init_y },
/* another case used during emulation */
{ { { 0x03515104, 0x00000077, { 0x00001001, 0x00000000 } },
{ 0xFFFFFFFF, 0xFFFFFFFF, { 0xFFFFFFFF, 0x0000003F } } },
"565 fb, 8888 tx, SRC_OVER clamp modulate", scanline_t32cb16blend_clamp_mod, init_y },
/* and this */
{ { { 0x03515104, 0x00000077, { 0x00001002, 0x00000000 } },
{ 0xFFFFFFFF, 0xFFFFFFFF, { 0xFFFFFFFF, 0x0000003F } } },
"565 fb, x888 tx, SRC_OVER clamp modulate", scanline_x32cb16blend_clamp_mod, init_y },
{ { { 0x03515104, 0x00000177, { 0x00001001, 0x00000000 } },
{ 0xFFFFFFFF, 0xFFFFFFFF, { 0xFFFFFFFF, 0x0000003F } } },
"565 fb, 8888 tx, SRC_OVER clamp modulate dither", scanline_t32cb16blend_clamp_mod_dither, init_y },
{ { { 0x03515104, 0x00000177, { 0x00001002, 0x00000000 } },
{ 0xFFFFFFFF, 0xFFFFFFFF, { 0xFFFFFFFF, 0x0000003F } } },
"565 fb, x888 tx, SRC_OVER clamp modulate dither", scanline_x32cb16blend_clamp_mod_dither, init_y },
{ { { 0x03010104, 0x00000077, { 0x00000001, 0x00000000 } },
{ 0xFFFFFFFF, 0xFFFFFFFF, { 0xFFFFFFFF, 0x0000003F } } },
"565 fb, 8888 tx, SRC clamp", scanline_t32cb16_clamp, init_y },
{ { { 0x03010104, 0x00000077, { 0x00000002, 0x00000000 } },
{ 0xFFFFFFFF, 0xFFFFFFFF, { 0xFFFFFFFF, 0x0000003F } } },
"565 fb, x888 tx, SRC clamp", scanline_t32cb16_clamp, init_y },
{ { { 0x03010104, 0x00000177, { 0x00000001, 0x00000000 } },
{ 0xFFFFFFFF, 0xFFFFFFFF, { 0xFFFFFFFF, 0x0000003F } } },
"565 fb, 8888 tx, SRC clamp dither", scanline_t32cb16_clamp_dither, init_y },
{ { { 0x03010104, 0x00000177, { 0x00000002, 0x00000000 } },
{ 0xFFFFFFFF, 0xFFFFFFFF, { 0xFFFFFFFF, 0x0000003F } } },
"565 fb, x888 tx, SRC clamp dither", scanline_t32cb16_clamp_dither, init_y },
{ { { 0x03010104, 0x00000077, { 0x00000004, 0x00000000 } },
{ 0xFFFFFFFF, 0xFFFFFFFF, { 0xFFFFFFFF, 0x0000003F } } },
"565 fb, 565 tx, SRC clamp", scanline_t16cb16_clamp, init_y },
{ { { 0x03515104, 0x00000077, { 0x00001004, 0x00000000 } },
{ 0xFFFFFFFF, 0xFFFFFFFF, { 0xFFFFFFFF, 0x0000003F } } },
"565 fb, 565 tx, SRC_OVER clamp", scanline_t16cb16blend_clamp_mod, init_y },
{ { { 0x03515104, 0x00000077, { 0x00000000, 0x00000000 } },
{ 0xFFFFFFFF, 0xFFFFFFFF, { 0xFFFFFFFF, 0xFFFFFFFF } } },
"565 fb, 8888 fixed color", scanline_col32cb16blend, init_y_packed },
@ -243,6 +350,12 @@ static void pick_scanline(context_t* c)
}
}
#ifdef DEBUG_NEEDS
LOGI("Needs: n=0x%08x p=0x%08x t0=0x%08x t1=0x%08x",
c->state.needs.n, c->state.needs.p,
c->state.needs.t[0], c->state.needs.t[1]);
#endif
#endif // DEBUG__CODEGEN_ONLY
c->init_y = init_y;
@ -797,6 +910,678 @@ discard:
#pragma mark Scanline
#endif
/* Used to parse a 32-bit source texture linearly. Usage is:
*
* horz_iterator32 hi(context);
* while (...) {
* uint32_t src_pixel = hi.get_pixel32();
* ...
* }
*
* Use only for one-to-one texture mapping.
*/
struct horz_iterator32 {
horz_iterator32(context_t* c) {
const int x = c->iterators.xl;
const int y = c->iterators.y;
texture_t& tx = c->state.texture[0];
const int32_t u = (tx.shade.is0>>16) + x;
const int32_t v = (tx.shade.it0>>16) + y;
m_src = reinterpret_cast<uint32_t*>(tx.surface.data)+(u+(tx.surface.stride*v));
}
uint32_t get_pixel32() {
return *m_src++;
}
protected:
uint32_t* m_src;
};
/* A variant for 16-bit source textures. */
struct horz_iterator16 {
horz_iterator16(context_t* c) {
const int x = c->iterators.xl;
const int y = c->iterators.y;
texture_t& tx = c->state.texture[0];
const int32_t u = (tx.shade.is0>>16) + x;
const int32_t v = (tx.shade.it0>>16) + y;
m_src = reinterpret_cast<uint16_t*>(tx.surface.data)+(u+(tx.surface.stride*v));
}
uint16_t get_pixel16() {
return *m_src++;
}
protected:
uint16_t* m_src;
};
/* A clamp iterator is used to iterate inside a texture with GGL_CLAMP.
* After initialization, call get_src16() or get_src32() to get the current
* texture pixel value.
*/
struct clamp_iterator {
clamp_iterator(context_t* c) {
const int xs = c->iterators.xl;
texture_t& tx = c->state.texture[0];
texture_iterators_t& ti = tx.iterators;
m_s = (xs * ti.dsdx) + ti.ydsdy;
m_t = (xs * ti.dtdx) + ti.ydtdy;
m_ds = ti.dsdx;
m_dt = ti.dtdx;
m_width_m1 = tx.surface.width - 1;
m_height_m1 = tx.surface.height - 1;
m_data = tx.surface.data;
m_stride = tx.surface.stride;
}
uint16_t get_pixel16() {
int u, v;
get_uv(u, v);
uint16_t* src = reinterpret_cast<uint16_t*>(m_data) + (u + (m_stride*v));
return src[0];
}
uint32_t get_pixel32() {
int u, v;
get_uv(u, v);
uint32_t* src = reinterpret_cast<uint32_t*>(m_data) + (u + (m_stride*v));
return src[0];
}
private:
void get_uv(int& u, int& v) {
int uu = m_s >> 16;
int vv = m_t >> 16;
if (uu < 0)
uu = 0;
if (uu > m_width_m1)
uu = m_width_m1;
if (vv < 0)
vv = 0;
if (vv > m_height_m1)
vv = m_height_m1;
u = uu;
v = vv;
m_s += m_ds;
m_t += m_dt;
}
GGLfixed m_s, m_t;
GGLfixed m_ds, m_dt;
int m_width_m1, m_height_m1;
uint8_t* m_data;
int m_stride;
};
/*
* The 'horizontal clamp iterator' variant corresponds to the case where
* the 'v' coordinate doesn't change. This is useful to avoid one mult and
* extra adds / checks per pixels, if the blending/processing operation after
* this is very fast.
*/
static int is_context_horizontal(const context_t* c) {
return (c->state.texture[0].iterators.dtdx == 0);
}
struct horz_clamp_iterator {
uint16_t get_pixel16() {
int u = m_s >> 16;
m_s += m_ds;
if (u < 0)
u = 0;
if (u > m_width_m1)
u = m_width_m1;
const uint16_t* src = reinterpret_cast<const uint16_t*>(m_data);
return src[u];
}
uint32_t get_pixel32() {
int u = m_s >> 16;
m_s += m_ds;
if (u < 0)
u = 0;
if (u > m_width_m1)
u = m_width_m1;
const uint32_t* src = reinterpret_cast<const uint32_t*>(m_data);
return src[u];
}
protected:
void init(const context_t* c, int shift);
GGLfixed m_s;
GGLfixed m_ds;
int m_width_m1;
const uint8_t* m_data;
};
void horz_clamp_iterator::init(const context_t* c, int shift)
{
const int xs = c->iterators.xl;
const texture_t& tx = c->state.texture[0];
const texture_iterators_t& ti = tx.iterators;
m_s = (xs * ti.dsdx) + ti.ydsdy;
m_ds = ti.dsdx;
m_width_m1 = tx.surface.width-1;
m_data = tx.surface.data;
GGLfixed t = (xs * ti.dtdx) + ti.ydtdy;
int v = t >> 16;
if (v < 0)
v = 0;
else if (v >= (int)tx.surface.height)
v = (int)tx.surface.height-1;
m_data += (tx.surface.stride*v) << shift;
}
struct horz_clamp_iterator16 : horz_clamp_iterator {
horz_clamp_iterator16(const context_t* c) {
init(c,1);
};
};
struct horz_clamp_iterator32 : horz_clamp_iterator {
horz_clamp_iterator32(context_t* c) {
init(c,2);
};
};
/* This is used to perform dithering operations.
*/
struct ditherer {
ditherer(const context_t* c) {
const int x = c->iterators.xl;
const int y = c->iterators.y;
m_line = &c->ditherMatrix[ ((y & GGL_DITHER_MASK)<<GGL_DITHER_ORDER_SHIFT) ];
m_index = x & GGL_DITHER_MASK;
}
void step(void) {
m_index++;
}
int get_value(void) {
int ret = m_line[m_index & GGL_DITHER_MASK];
m_index++;
return ret;
}
uint16_t abgr8888ToRgb565(uint32_t s) {
uint32_t r = s & 0xff;
uint32_t g = (s >> 8) & 0xff;
uint32_t b = (s >> 16) & 0xff;
return rgb888ToRgb565(r,g,b);
}
/* The following assumes that r/g/b are in the 0..255 range each */
uint16_t rgb888ToRgb565(uint32_t& r, uint32_t& g, uint32_t &b) {
int threshold = get_value();
/* dither in on GGL_DITHER_BITS, and each of r, g, b is on 8 bits */
r += (threshold >> (GGL_DITHER_BITS-8 +5));
g += (threshold >> (GGL_DITHER_BITS-8 +6));
b += (threshold >> (GGL_DITHER_BITS-8 +5));
if (r > 0xff)
r = 0xff;
if (g > 0xff)
g = 0xff;
if (b > 0xff)
b = 0xff;
return uint16_t(((r & 0xf8) << 8) | ((g & 0xfc) << 3) | (b >> 3));
}
protected:
const uint8_t* m_line;
int m_index;
};
/* This structure is used to blend (SRC_OVER) 32-bit source pixels
* onto 16-bit destination ones. Usage is simply:
*
* blender.blend(<32-bit-src-pixel-value>,<ptr-to-16-bit-dest-pixel>)
*/
struct blender_32to16 {
blender_32to16(context_t* c) { }
void write(uint32_t s, uint16_t* dst) {
if (s == 0)
return;
s = GGL_RGBA_TO_HOST(s);
int sA = (s>>24);
if (sA == 0xff) {
*dst = convertAbgr8888ToRgb565(s);
} else {
int f = 0x100 - (sA + (sA>>7));
int sR = (s >> ( 3))&0x1F;
int sG = (s >> ( 8+2))&0x3F;
int sB = (s >> (16+3))&0x1F;
uint16_t d = *dst;
int dR = (d>>11)&0x1f;
int dG = (d>>5)&0x3f;
int dB = (d)&0x1f;
sR += (f*dR)>>8;
sG += (f*dG)>>8;
sB += (f*dB)>>8;
*dst = uint16_t((sR<<11)|(sG<<5)|sB);
}
}
void write(uint32_t s, uint16_t* dst, ditherer& di) {
if (s == 0) {
di.step();
return;
}
s = GGL_RGBA_TO_HOST(s);
int sA = (s>>24);
if (sA == 0xff) {
*dst = di.abgr8888ToRgb565(s);
} else {
int threshold = di.get_value() << (8 - GGL_DITHER_BITS);
int f = 0x100 - (sA + (sA>>7));
int sR = (s >> ( 3))&0x1F;
int sG = (s >> ( 8+2))&0x3F;
int sB = (s >> (16+3))&0x1F;
uint16_t d = *dst;
int dR = (d>>11)&0x1f;
int dG = (d>>5)&0x3f;
int dB = (d)&0x1f;
sR = ((sR << 8) + f*dR + threshold)>>8;
sG = ((sG << 8) + f*dG + threshold)>>8;
sB = ((sB << 8) + f*dB + threshold)>>8;
if (sR > 0x1f) sR = 0x1f;
if (sG > 0x3f) sG = 0x3f;
if (sB > 0x1f) sB = 0x1f;
*dst = uint16_t((sR<<11)|(sG<<5)|sB);
}
}
};
/* This blender does the same for the 'blend_srca' operation.
* where dstFactor=srcA*(1-srcA) srcFactor=srcA
*/
struct blender_32to16_srcA {
blender_32to16_srcA(const context_t* c) { }
void write(uint32_t s, uint16_t* dst) {
if (!s) {
return;
}
uint16_t d = *dst;
s = GGL_RGBA_TO_HOST(s);
int sR = (s >> ( 3))&0x1F;
int sG = (s >> ( 8+2))&0x3F;
int sB = (s >> (16+3))&0x1F;
int sA = (s>>24);
int f1 = (sA + (sA>>7));
int f2 = 0x100-f1;
int dR = (d>>11)&0x1f;
int dG = (d>>5)&0x3f;
int dB = (d)&0x1f;
sR = (f1*sR + f2*dR)>>8;
sG = (f1*sG + f2*dG)>>8;
sB = (f1*sB + f2*dB)>>8;
*dst = uint16_t((sR<<11)|(sG<<5)|sB);
}
};
/* Common init code the modulating blenders */
struct blender_modulate {
void init(const context_t* c) {
const int r = c->iterators.ydrdy >> (GGL_COLOR_BITS-8);
const int g = c->iterators.ydgdy >> (GGL_COLOR_BITS-8);
const int b = c->iterators.ydbdy >> (GGL_COLOR_BITS-8);
const int a = c->iterators.ydady >> (GGL_COLOR_BITS-8);
m_r = r + (r >> 7);
m_g = g + (g >> 7);
m_b = b + (b >> 7);
m_a = a + (a >> 7);
}
protected:
int m_r, m_g, m_b, m_a;
};
/* This blender does a normal blend after modulation.
*/
struct blender_32to16_modulate : blender_modulate {
blender_32to16_modulate(const context_t* c) {
init(c);
}
void write(uint32_t s, uint16_t* dst) {
// blend source and destination
if (!s) {
return;
}
s = GGL_RGBA_TO_HOST(s);
/* We need to modulate s */
uint32_t sA = (s >> 24);
uint32_t sB = (s >> 16) & 0xff;
uint32_t sG = (s >> 8) & 0xff;
uint32_t sR = s & 0xff;
sA = (sA*m_a) >> 8;
/* Keep R/G/B scaled to 5.8 or 6.8 fixed float format */
sR = (sR*m_r) >> (8 - 5);
sG = (sG*m_g) >> (8 - 6);
sB = (sB*m_b) >> (8 - 5);
/* Now do a normal blend */
int f = 0x100 - (sA + (sA>>7));
uint16_t d = *dst;
int dR = (d>>11)&0x1f;
int dG = (d>>5)&0x3f;
int dB = (d)&0x1f;
sR = (sR + f*dR)>>8;
sG = (sG + f*dG)>>8;
sB = (sB + f*dB)>>8;
*dst = uint16_t((sR<<11)|(sG<<5)|sB);
}
void write(uint32_t s, uint16_t* dst, ditherer& di) {
// blend source and destination
if (!s) {
di.step();
return;
}
s = GGL_RGBA_TO_HOST(s);
/* We need to modulate s */
uint32_t sA = (s >> 24);
uint32_t sB = (s >> 16) & 0xff;
uint32_t sG = (s >> 8) & 0xff;
uint32_t sR = s & 0xff;
sA = (sA*m_a) >> 8;
/* keep R/G/B scaled to 5.8 or 6.8 fixed float format */
sR = (sR*m_r) >> (8 - 5);
sG = (sG*m_g) >> (8 - 6);
sB = (sB*m_b) >> (8 - 5);
/* Scale threshold to 0.8 fixed float format */
int threshold = di.get_value() << (8 - GGL_DITHER_BITS);
int f = 0x100 - (sA + (sA>>7));
uint16_t d = *dst;
int dR = (d>>11)&0x1f;
int dG = (d>>5)&0x3f;
int dB = (d)&0x1f;
sR = (sR + f*dR + threshold)>>8;
sG = (sG + f*dG + threshold)>>8;
sB = (sB + f*dB + threshold)>>8;
if (sR > 0x1f) sR = 0x1f;
if (sG > 0x3f) sG = 0x3f;
if (sB > 0x1f) sB = 0x1f;
*dst = uint16_t((sR<<11)|(sG<<5)|sB);
}
};
/* same as 32to16_modulate, except that the input is xRGB, instead of ARGB */
struct blender_x32to16_modulate : blender_modulate {
blender_x32to16_modulate(const context_t* c) {
init(c);
}
void write(uint32_t s, uint16_t* dst) {
s = GGL_RGBA_TO_HOST(s);
uint32_t sB = (s >> 16) & 0xff;
uint32_t sG = (s >> 8) & 0xff;
uint32_t sR = s & 0xff;
/* Keep R/G/B in 5.8 or 6.8 format */
sR = (sR*m_r) >> (8 - 5);
sG = (sG*m_g) >> (8 - 6);
sB = (sB*m_b) >> (8 - 5);
int f = 0x100 - m_a;
uint16_t d = *dst;
int dR = (d>>11)&0x1f;
int dG = (d>>5)&0x3f;
int dB = (d)&0x1f;
sR = (sR + f*dR)>>8;
sG = (sG + f*dG)>>8;
sB = (sB + f*dB)>>8;
*dst = uint16_t((sR<<11)|(sG<<5)|sB);
}
void write(uint32_t s, uint16_t* dst, ditherer& di) {
s = GGL_RGBA_TO_HOST(s);
uint32_t sB = (s >> 16) & 0xff;
uint32_t sG = (s >> 8) & 0xff;
uint32_t sR = s & 0xff;
sR = (sR*m_r) >> (8 - 5);
sG = (sG*m_g) >> (8 - 6);
sB = (sB*m_b) >> (8 - 5);
/* Now do a normal blend */
int threshold = di.get_value() << (8 - GGL_DITHER_BITS);
int f = 0x100 - m_a;
uint16_t d = *dst;
int dR = (d>>11)&0x1f;
int dG = (d>>5)&0x3f;
int dB = (d)&0x1f;
sR = (sR + f*dR + threshold)>>8;
sG = (sG + f*dG + threshold)>>8;
sB = (sB + f*dB + threshold)>>8;
if (sR > 0x1f) sR = 0x1f;
if (sG > 0x3f) sG = 0x3f;
if (sB > 0x1f) sB = 0x1f;
*dst = uint16_t((sR<<11)|(sG<<5)|sB);
}
};
/* Same as above, but source is 16bit rgb565 */
struct blender_16to16_modulate : blender_modulate {
blender_16to16_modulate(const context_t* c) {
init(c);
}
void write(uint16_t s16, uint16_t* dst) {
uint32_t s = s16;
uint32_t sR = s >> 11;
uint32_t sG = (s >> 5) & 0x3f;
uint32_t sB = s & 0x1f;
sR = (sR*m_r);
sG = (sG*m_g);
sB = (sB*m_b);
int f = 0x100 - m_a;
uint16_t d = *dst;
int dR = (d>>11)&0x1f;
int dG = (d>>5)&0x3f;
int dB = (d)&0x1f;
sR = (sR + f*dR)>>8;
sG = (sG + f*dG)>>8;
sB = (sB + f*dB)>>8;
*dst = uint16_t((sR<<11)|(sG<<5)|sB);
}
};
/* This is used to iterate over a 16-bit destination color buffer.
* Usage is:
*
* dst_iterator16 di(context);
* while (di.count--) {
* <do stuff with dest pixel at di.dst>
* di.dst++;
* }
*/
struct dst_iterator16 {
dst_iterator16(const context_t* c) {
const int x = c->iterators.xl;
const int width = c->iterators.xr - x;
const int32_t y = c->iterators.y;
const surface_t* cb = &(c->state.buffers.color);
count = width;
dst = reinterpret_cast<uint16_t*>(cb->data) + (x+(cb->stride*y));
}
int count;
uint16_t* dst;
};
static void scanline_t32cb16_clamp(context_t* c)
{
dst_iterator16 di(c);
if (is_context_horizontal(c)) {
/* Special case for simple horizontal scaling */
horz_clamp_iterator32 ci(c);
while (di.count--) {
uint32_t s = ci.get_pixel32();
*di.dst++ = convertAbgr8888ToRgb565(s);
}
} else {
/* General case */
clamp_iterator ci(c);
while (di.count--) {
uint32_t s = ci.get_pixel32();
*di.dst++ = convertAbgr8888ToRgb565(s);
}
}
}
static void scanline_t32cb16_dither(context_t* c)
{
horz_iterator32 si(c);
dst_iterator16 di(c);
ditherer dither(c);
while (di.count--) {
uint32_t s = si.get_pixel32();
*di.dst++ = dither.abgr8888ToRgb565(s);
}
}
static void scanline_t32cb16_clamp_dither(context_t* c)
{
dst_iterator16 di(c);
ditherer dither(c);
if (is_context_horizontal(c)) {
/* Special case for simple horizontal scaling */
horz_clamp_iterator32 ci(c);
while (di.count--) {
uint32_t s = ci.get_pixel32();
*di.dst++ = dither.abgr8888ToRgb565(s);
}
} else {
/* General case */
clamp_iterator ci(c);
while (di.count--) {
uint32_t s = ci.get_pixel32();
*di.dst++ = dither.abgr8888ToRgb565(s);
}
}
}
static void scanline_t32cb16blend_dither(context_t* c)
{
dst_iterator16 di(c);
ditherer dither(c);
blender_32to16 bl(c);
horz_iterator32 hi(c);
while (di.count--) {
uint32_t s = hi.get_pixel32();
bl.write(s, di.dst, dither);
di.dst++;
}
}
static void scanline_t32cb16blend_clamp(context_t* c)
{
dst_iterator16 di(c);
blender_32to16 bl(c);
if (is_context_horizontal(c)) {
horz_clamp_iterator32 ci(c);
while (di.count--) {
uint32_t s = ci.get_pixel32();
bl.write(s, di.dst);
di.dst++;
}
} else {
clamp_iterator ci(c);
while (di.count--) {
uint32_t s = ci.get_pixel32();
bl.write(s, di.dst);
di.dst++;
}
}
}
static void scanline_t32cb16blend_clamp_dither(context_t* c)
{
dst_iterator16 di(c);
ditherer dither(c);
blender_32to16 bl(c);
clamp_iterator ci(c);
while (di.count--) {
uint32_t s = ci.get_pixel32();
bl.write(s, di.dst, dither);
di.dst++;
}
}
void scanline_t32cb16blend_clamp_mod(context_t* c)
{
dst_iterator16 di(c);
blender_32to16_modulate bl(c);
clamp_iterator ci(c);
while (di.count--) {
uint32_t s = ci.get_pixel32();
bl.write(s, di.dst);
di.dst++;
}
}
void scanline_t32cb16blend_clamp_mod_dither(context_t* c)
{
dst_iterator16 di(c);
blender_32to16_modulate bl(c);
ditherer dither(c);
clamp_iterator ci(c);
while (di.count--) {
uint32_t s = ci.get_pixel32();
bl.write(s, di.dst, dither);
di.dst++;
}
}
/* Variant of scanline_t32cb16blend_clamp_mod with a xRGB texture */
void scanline_x32cb16blend_clamp_mod(context_t* c)
{
dst_iterator16 di(c);
blender_x32to16_modulate bl(c);
clamp_iterator ci(c);
while (di.count--) {
uint32_t s = ci.get_pixel32();
bl.write(s, di.dst);
di.dst++;
}
}
void scanline_x32cb16blend_clamp_mod_dither(context_t* c)
{
dst_iterator16 di(c);
blender_x32to16_modulate bl(c);
ditherer dither(c);
clamp_iterator ci(c);
while (di.count--) {
uint32_t s = ci.get_pixel32();
bl.write(s, di.dst, dither);
di.dst++;
}
}
void scanline_t16cb16_clamp(context_t* c)
{
dst_iterator16 di(c);
/* Special case for simple horizontal scaling */
if (is_context_horizontal(c)) {
horz_clamp_iterator16 ci(c);
while (di.count--) {
*di.dst++ = ci.get_pixel16();
}
} else {
clamp_iterator ci(c);
while (di.count--) {
*di.dst++ = ci.get_pixel16();
}
}
}
template <typename T, typename U>
static inline __attribute__((const))
T interpolate(int y, T v0, U dvdx, U dvdy) {
@ -1322,30 +2107,24 @@ void scanline_t32cb16(context_t* c)
if (ct==1 || uint32_t(dst)&2) {
last_one:
s = GGL_RGBA_TO_HOST( *src++ );
sR = (s >> ( 3))&0x1F;
sG = (s >> ( 8+2))&0x3F;
sB = (s >> (16+3))&0x1F;
*dst++ = uint16_t((sR<<11)|(sG<<5)|sB);
*dst++ = convertAbgr8888ToRgb565(s);
ct--;
}
while (ct >= 2) {
s = GGL_RGBA_TO_HOST( *src++ );
sR = (s >> ( 3))&0x1F;
sG = (s >> ( 8+2))&0x3F;
sB = (s >> (16+3))&0x1F;
d = (sR<<11)|(sG<<5)|sB;
s = GGL_RGBA_TO_HOST( *src++ );
sR = (s >> ( 3))&0x1F;
sG = (s >> ( 8+2))&0x3F;
sB = (s >> (16+3))&0x1F;
d |= ((sR<<11)|(sG<<5)|sB)<<16;
#if BYTE_ORDER == BIG_ENDIAN
d = (d>>16) | (d<<16);
#endif
s = GGL_RGBA_TO_HOST( *src++ );
d = convertAbgr8888ToRgb565_hi16(s);
s = GGL_RGBA_TO_HOST( *src++ );
d |= convertAbgr8888ToRgb565(s);
#else
s = GGL_RGBA_TO_HOST( *src++ );
d = convertAbgr8888ToRgb565(s);
s = GGL_RGBA_TO_HOST( *src++ );
d |= convertAbgr8888ToRgb565(s) << 16;
#endif
*dst32++ = d;
ct -= 2;
}
@ -1357,6 +2136,7 @@ last_one:
void scanline_t32cb16blend(context_t* c)
{
#if ((ANDROID_CODEGEN >= ANDROID_CODEGEN_ASM) && defined(__arm__))
int32_t x = c->iterators.xl;
size_t ct = c->iterators.xr - x;
int32_t y = c->iterators.y;
@ -1368,33 +2148,55 @@ void scanline_t32cb16blend(context_t* c)
const int32_t v = (c->state.texture[0].shade.it0>>16) + y;
uint32_t *src = reinterpret_cast<uint32_t*>(tex->data)+(u+(tex->stride*v));
#if ((ANDROID_CODEGEN >= ANDROID_CODEGEN_ASM) && defined(__arm__))
scanline_t32cb16blend_arm(dst, src, ct);
#else
while (ct--) {
uint32_t s = *src++;
if (!s) {
dst++;
continue;
}
uint16_t d = *dst;
s = GGL_RGBA_TO_HOST(s);
int sR = (s >> ( 3))&0x1F;
int sG = (s >> ( 8+2))&0x3F;
int sB = (s >> (16+3))&0x1F;
int sA = (s>>24);
int f = 0x100 - (sA + (sA>>7));
int dR = (d>>11)&0x1f;
int dG = (d>>5)&0x3f;
int dB = (d)&0x1f;
sR += (f*dR)>>8;
sG += (f*dG)>>8;
sB += (f*dB)>>8;
*dst++ = uint16_t((sR<<11)|(sG<<5)|sB);
dst_iterator16 di(c);
horz_iterator32 hi(c);
blender_32to16 bl(c);
while (di.count--) {
uint32_t s = hi.get_pixel32();
bl.write(s, di.dst);
di.dst++;
}
#endif
}
void scanline_t32cb16blend_srca(context_t* c)
{
dst_iterator16 di(c);
horz_iterator32 hi(c);
blender_32to16_srcA blender(c);
while (di.count--) {
uint32_t s = hi.get_pixel32();
blender.write(s,di.dst);
di.dst++;
}
}
void scanline_t16cb16blend_clamp_mod(context_t* c)
{
const int a = c->iterators.ydady >> (GGL_COLOR_BITS-8);
if (a == 0) {
return;
}
if (a == 255) {
scanline_t16cb16_clamp(c);
return;
}
dst_iterator16 di(c);
blender_16to16_modulate blender(c);
clamp_iterator ci(c);
while (di.count--) {
uint16_t s = ci.get_pixel16();
blender.write(s, di.dst);
di.dst++;
}
}
void scanline_memcpy(context_t* c)
{
int32_t x = c->iterators.xl;
@ -1518,26 +2320,3 @@ void rect_memcpy(context_t* c, size_t yc)
// ----------------------------------------------------------------------------
}; // namespace android
using namespace android;
extern "C" void ggl_test_codegen(uint32_t n, uint32_t p, uint32_t t0, uint32_t t1)
{
#if ANDROID_ARM_CODEGEN
GGLContext* c;
gglInit(&c);
needs_t needs;
needs.n = n;
needs.p = p;
needs.t[0] = t0;
needs.t[1] = t1;
sp<ScanlineAssembly> a(new ScanlineAssembly(needs, ASSEMBLY_SCRATCH_SIZE));
GGLAssembler assembler( new ARMAssembler(a) );
int err = assembler.scanline(needs, (context_t*)c);
if (err != 0) {
printf("error %08x (%s)\n", err, strerror(-err));
}
gglUninit(c);
#else
printf("This test runs only on ARM\n");
#endif
}

5
libpixelflinger/tests/codegen/Android.mk
View file

@ -2,12 +2,15 @@ LOCAL_PATH:= $(call my-dir)
include $(CLEAR_VARS)
LOCAL_SRC_FILES:= \
codegen.cpp
codegen.cpp.arm
LOCAL_SHARED_LIBRARIES := \
libcutils \
libpixelflinger
LOCAL_C_INCLUDES := \
system/core/libpixelflinger
LOCAL_MODULE:= test-opengl-codegen
LOCAL_MODULE_TAGS := tests

49
libpixelflinger/tests/codegen/codegen.cpp
View file

@ -1,9 +1,54 @@
#include <stdio.h>
#include <stdint.h>
extern "C" void ggl_test_codegen(
uint32_t n, uint32_t p, uint32_t t0, uint32_t t1);
#include "private/pixelflinger/ggl_context.h"
#include "buffer.h"
#include "scanline.h"
#include "codeflinger/CodeCache.h"
#include "codeflinger/GGLAssembler.h"
#include "codeflinger/ARMAssembler.h"
#if defined(__arm__)
# define ANDROID_ARM_CODEGEN 1
#else
# define ANDROID_ARM_CODEGEN 0
#endif
#define ASSEMBLY_SCRATCH_SIZE 2048
using namespace android;
class ScanlineAssembly : public Assembly {
AssemblyKey<needs_t> mKey;
public:
ScanlineAssembly(needs_t needs, size_t size)
: Assembly(size), mKey(needs) { }
const AssemblyKey<needs_t>& key() const { return mKey; }
};
static void ggl_test_codegen(uint32_t n, uint32_t p, uint32_t t0, uint32_t t1)
{
#if ANDROID_ARM_CODEGEN
GGLContext* c;
gglInit(&c);
needs_t needs;
needs.n = n;
needs.p = p;
needs.t[0] = t0;
needs.t[1] = t1;
sp<ScanlineAssembly> a(new ScanlineAssembly(needs, ASSEMBLY_SCRATCH_SIZE));
GGLAssembler assembler( new ARMAssembler(a) );
int err = assembler.scanline(needs, (context_t*)c);
if (err != 0) {
printf("error %08x (%s)\n", err, strerror(-err));
}
gglUninit(c);
#else
printf("This test runs only on ARM\n");
#endif
}
int main(int argc, char** argv)
{

4
rootdir/Android.mk
View file

@ -11,6 +11,10 @@ ifeq ($(TARGET_PRODUCT),generic)
copy_from += etc/vold.fstab
endif
ifeq ($(TARGET_PRODUCT),full_x86)
copy_from += etc/vold.fstab
endif
# the /system/etc/init.goldfish.sh is needed to enable emulator support
# in the system image. In theory, we don't need these for -user builds
# which are device-specific. However, these builds require at the moment