CSDN开源夏令营 - 第四周工作总结
本周主要是在回顾“set trace-buffer-size”完成通信的基础上，对于指定参数size的处理，比如指定-1或者unlimited的时候，在KGTP端该如何做解析，以及该如何处理此时的buffer size。同时，根据“开题报告"的进度，中期检查的任务已经完成。对Trace Buffer的实现，做了初步的调研。
1. GDB和KGTP通信实现的回顾
利用GDBRSP，即GDB Remote Serial Protocol，GDB远程串行通信协议。简单来说，
用于GDB远程调试发包，但是也支持File I/O和Console，而KGTP正是利用了GDBRSP对于File I/O的支持，完成GDB和KGTP的通信，在接受到GDB发送的Packet后，KGTP进行解析。而通信的介质则是通过debugfs或者其他kernel space和user space进行数据交换的接口进行的。
具体实现，可以参考上一篇博文[1]。
2. GDB的Packet格式
GDB向target发送的Packet必须遵守提前设定的约定，才能被target按照相应的规则解析处理。关于GDB的Packet格式，可以参考[2]，这里详细列出了所有的预定义的指令和它们相应的响应数据。如果想进一步了解GDBRSP，可以阅读[3]。由于KGTP对Trace Point最感兴趣，因此我们也最关注Tracepoint Packets，只要有两种，分别以Q和q开头，其中Q表示设置命令，即set，而q表示查询命令，即query，具体看[4]。关于如何开始和停止Trace Experiments，看[5]。GDB对于用户输入的commands，会转换后封装成Packet，通过GDBRSP发给target，因此需要结合着[4]和[5]才能把对某一个命令如何进行包解析搞清楚，比如用户输入”tstart“，则会发送一个”QTStart“的包。
3. KGTP对GDB的Packet的处理实现
有了GDB和KGTP的通信的基础设施，比如DebugFS和GDBRSP，正常的文件读写即可完成GDB和KGTP的数据交换。而为了完成相应的任务，KGTP必须对接收到的GDB的Packet进行解析，而我们又知道了GDB得Packet格式，因此只需要对接收到的数据按照格式做相应的解析即可。
在gtp.c的gtp_write函数，针对接收到的GDB包做了具体的解析，具体代码如下：
static ssize_t
gtp_write(struct file *file, const char __user *buf, size_t size,
loff_t *ppos)
{
char *rsppkg = NULL;
int i, ret;
unsigned char csum;
int is_reverse;
if (down_interruptible(>p_rw_lock))
return -EINTR;
if (size == 0) {
#ifdef GTP_DEBUG
printk(GTP_DEBUG "gtp_write: try write 0 size.\n");
#endif
goto error_out;
}
size = min_t(size_t, size, GTP_RW_MAX);
if (copy_from_user(gtp_rw_buf, buf, size)) {
size = -EFAULT;
goto error_out;
}
if (gtp_rw_buf[0] == '+' || gtp_rw_buf[0] == '-'
|| gtp_rw_buf[0] == '\3' || gtp_rw_buf[0] == '\n') {
if (gtp_rw_buf[0] == '+')
gtp_rw_size = 0;
size = 1;
goto out;
}
if (size < 4) {
size = -EINVAL;
goto error_out;
}
/* Check format and get the rsppkg. */
for (i = 0; i < size - 2; i++) {
if (gtp_rw_buf[i] == '$')
rsppkg = gtp_rw_buf + i + 1;
else if (gtp_rw_buf[i] == '#')
break;
}
if (rsppkg && gtp_rw_buf[i] == '#') {
/* Format is OK. Check crc. */
if (gtp_noack_mode < 1)
gtp_read_ack = 1;
size = i + 3;
gtp_rw_buf[i] = '\0';
} else {
printk(KERN_WARNING "gtp_write: format error\n");
size = -EINVAL;
goto error_out;
}
wake_up_interruptible_nr(>p_rw_wq, 1);
up(>p_rw_lock);
if (down_interruptible(>p_rw_lock))
return -EINTR;
#ifdef GTP_DEBUG
printk(GTP_DEBUG "gtp_write: %s\n", rsppkg);
#endif
/* Handle rsppkg and put return to gtp_rw_buf. */
gtp_rw_buf[0] = '$';
gtp_rw_bufp = gtp_rw_buf + 1;
gtp_rw_size = 0;
ret = 1;
is_reverse = 0;
switch (rsppkg[0]) {
case '?':
if (gtp_current_pid == 0)
snprintf(gtp_rw_bufp, GTP_RW_BUFP_MAX, "S05");
else
snprintf(gtp_rw_bufp, GTP_RW_BUFP_MAX, "T05;thread:p%x.%x;",
gtp_current_pid, gtp_current_pid);
gtp_rw_size += strlen(gtp_rw_bufp);
gtp_rw_bufp += strlen(gtp_rw_bufp);
break;
case 'g':
ret = gtp_gdbrsp_g();
break;
case 'm':
ret = gtp_gdbrsp_m(rsppkg + 1);
break;
case 'Q':
#ifdef GTP_RB
/* This check for "tfind -1" and let GDB into step replay.
XXX: just test on X86_64. */
if (gtp_replay_step_id) {
if (strcmp("QTFrame:ffffffff", rsppkg) == 0) {
ret = 0;
goto switch_done;
} else
gtp_replay_reset();
}
#endif
if (rsppkg[1] == 'T')
ret = gtp_gdbrsp_QT(rsppkg + 2);
else if (strncmp("QStartNoAckMode", rsppkg, 15) == 0) {
ret = 0;
gtp_noack_mode = -1;
}
break;
case 'q':
if (rsppkg[1] == 'T')
ret = gtp_gdbrsp_qT(rsppkg + 2);
else if (rsppkg[1] == 'C') {
snprintf(gtp_rw_bufp, GTP_RW_BUFP_MAX, "QC%x",
gtp_current_pid);
gtp_rw_size += strlen(gtp_rw_bufp);
gtp_rw_bufp += strlen(gtp_rw_bufp);
ret = 1;
} else if (strncmp("qSupported", rsppkg, 10) == 0) {
#ifdef GTP_RB
snprintf(gtp_rw_bufp, GTP_RW_BUFP_MAX,
"QStartNoAckMode+;ConditionalTracepoints+;"
"TracepointSource+;DisconnectedTracing+;"
"ReverseContinue+;ReverseStep+;"
#if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,30))
"EnableDisableTracepoints+;"
#endif
"qXfer:traceframe-info:read+;");
#endif
#if defined(GTP_FRAME_SIMPLE) || defined(GTP_FTRACE_RING_BUFFER)
snprintf(gtp_rw_bufp, GTP_RW_BUFP_MAX,
"QStartNoAckMode+;ConditionalTracepoints+;"
#if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,30))
"EnableDisableTracepoints+;"
#endif
"TracepointSource+;DisconnectedTracing+;");
#endif
gtp_rw_size += strlen(gtp_rw_bufp);
gtp_rw_bufp += strlen(gtp_rw_bufp);
ret = 1;
}
#ifdef GTP_RB
else if (strncmp("qXfer:traceframe-info:read::",
rsppkg, 28) == 0)
ret = gtp_gdbrsp_qxfer_traceframe_info_read(rsppkg
+ 28);
#endif
else if (strncmp("qRcmd,", rsppkg, 6) == 0)
ret = gtp_gdbrsp_qRcmd(rsppkg + 6);
else if (strncmp("qAttached", rsppkg, 9) == 0) {
snprintf(gtp_rw_bufp, GTP_RW_BUFP_MAX, "1");
gtp_rw_size += 1;
gtp_rw_bufp += 1;
}
break;
case 'S':
case 'C':
ret = -1;
break;
case 'b':
rsppkg[0] = rsppkg[1];
is_reverse = 1;
case 's':
case 'c':
ret = gtp_gdbrsp_resume (rsppkg[0] == 's', is_reverse);
break;
case 'v':
if (strncmp("vAttach;", rsppkg, 8) == 0) {
#ifdef GTP_RB
if (gtp_replay_step_id)
gtp_replay_reset();
#endif
ret = gtp_gdbrsp_vAttach(rsppkg + 8);
} else if (strncmp("vKill;", rsppkg, 7) == 0) {
#ifdef GTP_RB
if (gtp_replay_step_id)
gtp_replay_reset();
#endif
/* XXX: When we add more code to support trace
user space program. We need add more release
code to this part.
Release tracepoint for this tracepoint. */
ret = 0;
}
break;
case 'D':
#ifdef GTP_RB
if (gtp_replay_step_id)
gtp_replay_reset();
#endif
gtp_gdbrsp_D(rsppkg + 1);
ret = 0;
break;
case 'H':
ret = gtp_gdbrsp_H(rsppkg + 1);
break;
case 'Z':
case 'z':
if (rsppkg[1] == '0')
ret = gtp_gdbrsp_breakpoint(rsppkg + 3,
(rsppkg[0] == 'Z'));
break;
}
switch_done:
if (ret == 0) {
snprintf(gtp_rw_bufp, GTP_RW_BUFP_MAX, "OK");
gtp_rw_bufp += 2;
gtp_rw_size += 2;
} else if (ret < 0) {
snprintf(gtp_rw_bufp, GTP_RW_BUFP_MAX, "E%02x", -ret);
gtp_rw_bufp += 3;
gtp_rw_size += 3;
}
gtp_rw_bufp[0] = '#';
csum = 0;
for (i = 1; i < gtp_rw_size + 1; i++)
csum += gtp_rw_buf[i];
gtp_rw_bufp[1] = INT2CHAR(csum >> 4);
gtp_rw_bufp[2] = INT2CHAR(csum & 0x0f);
gtp_rw_bufp = gtp_rw_buf;
gtp_rw_size += 4;
out:
wake_up_interruptible_nr(>p_rw_wq, 1);
error_out:
up(>p_rw_lock);
return size;
}
我们可以看到根据接收到的Packet，分别调用了不同的handler来做处理，这里我们重点关注Tracepoint Packets，关于以QT开头的包，处理代码如下：
static int
gtp_gdbrsp_QT(char *pkg)
{
int ret = 1;
#ifdef GTP_DEBUG
printk(GTP_DEBUG "gtp_gdbrsp_QT: %s\n", pkg);
#endif
if (strcmp("init", pkg) == 0)
ret = gtp_gdbrsp_qtinit();
else if (strcmp("Stop", pkg) == 0)
ret = gtp_gdbrsp_qtstop();
else if (strcmp("Start", pkg) == 0)
ret = gtp_gdbrsp_qtstart();
else if (strncmp("DP:", pkg, 3) == 0)
ret = gtp_gdbrsp_qtdp(pkg + 3);
else if (strncmp("DPsrc:", pkg, 6) == 0)
ret = gtp_gdbrsp_qtdpsrc(pkg + 6);
else if (strncmp("Disconnected:", pkg, 13) == 0)
ret = gtp_gdbrsp_qtdisconnected(pkg + 13);
else if (strncmp("Buffer:", pkg, 7) == 0)
ret = gtp_gdbrsp_qtbuffer(pkg + 7);
else if (strncmp("Frame:", pkg, 6) == 0)
ret = gtp_gdbrsp_qtframe(pkg + 6);
else if (strncmp("ro:", pkg, 3) == 0)
ret = gtp_gdbrsp_qtro(pkg + 3);
else if (strncmp("DV:", pkg, 3) == 0)
ret = gtp_gdbrsp_qtdv(pkg + 3);
#if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,30))
else if (strncmp("Enable:", pkg, 7) == 0)
ret = gtp_gdbrsp_qtenable_qtdisable(pkg + 7, 1);
else if (strncmp("Disable:", pkg, 8) == 0)
ret = gtp_gdbrsp_qtenable_qtdisable(pkg + 8, 0);
#endif
#ifdef GTP_DEBUG
printk(GTP_DEBUG "gtp_gdbrsp_QT: return %d\n", ret);
#endif
return ret;
}
而以qT打头的包，处理的接口如下：
static int
gtp_gdbrsp_qT(char *pkg)
{
int ret = 1;
#ifdef GTP_DEBUG
printk(GTP_DEBUG "gtp_gdbrsp_qT: %s\n", pkg);
#endif
if (strcmp("Status", pkg) == 0)
ret = gtp_gdbrsp_qtstatus();
else if (strcmp("fP", pkg) == 0)
ret = gtp_gdbrsp_qtfp();
else if (strcmp("sP", pkg) == 0)
ret = gtp_gdbrsp_qtsp();
else if (strcmp("fV", pkg) == 0)
ret = gtp_gdbrsp_qtfsv(1);
else if (strcmp("sV", pkg) == 0)
ret = gtp_gdbrsp_qtfsv(0);
else if (strncmp("V:", pkg, 2) == 0)
ret = gtp_gdbrsp_qtv(pkg + 2);
return ret;
<h2>}</h2>
从以上接口可以看到，是一个dispatch的过程，KGTP首先判断接收到的GDB包属于哪一类包，之后分发，而对于一个类型的包，又细分很多子包，最终分发给具体的接口实现，拿上面为例，有如下一个流程：
gtp_write -> gtp_gdbrsp_QT -> gtp_gdbrsp_qtbuffer
通过以上分析，我们已经在Packet层面对GDB和KGTP的通信有了深入的理解。
4. KGTP中set-buffer-size的处理逻辑
有了以上分析，在用户输入了以下指令：
set remote trace buffer-size on
set trace-buffer-size xxx
之后会发生什么呢？
让我们分别测试一下。
（1）set trace-buffer-size 100
Aug 03 11:00:08 localhost.localdomain kernel: gtp_write: QTBuffer:size:64
Aug 03 11:00:08 localhost.localdomain kernel: gtp_gdbrsp_QT: Buffer:size:64
Aug 03 11:00:08 localhost.localdomain kernel: gtp_gdbrsp_qtbuffer
Aug 03 11:00:08 localhost.localdomain kernel: gtp_gdbrsp_qtbuffer:setting buffer size to 100
Aug 03 11:00:08 localhost.localdomain kernel: gtp_gdbrsp_QT: return 0
Aug 03 11:00:08 localhost.localdomain kernel: gtp_read
（2）set trace-buffer-size 1000000
Aug 03 11:00:33 localhost.localdomain kernel: gtp_write: QTBuffer:size:f4240
Aug 03 11:00:33 localhost.localdomain kernel: gtp_gdbrsp_QT: Buffer:size:f4240
Aug 03 11:00:33 localhost.localdomain kernel: gtp_gdbrsp_qtbuffer
Aug 03 11:00:33 localhost.localdomain kernel: gtp_gdbrsp_qtbuffer:setting buffer size to 1000000
Aug 03 11:00:33 localhost.localdomain kernel: gtp_gdbrsp_QT: return 0
Aug 03 11:00:33 localhost.localdomain kernel: gtp_read
（3）set trace-buffer-size -1
Aug 03 11:00:18 localhost.localdomain kernel: gtp_write: QTBuffer:size:-1
Aug 03 11:00:18 localhost.localdomain kernel: gtp_gdbrsp_QT: Buffer:size:-1
Aug 03 11:00:18 localhost.localdomain kernel: gtp_gdbrsp_qtbuffer
Aug 03 11:00:18 localhost.localdomain kernel: gtp_gdbrsp_qtbuffer:setting buffer size to 0
Aug 03 11:00:18 localhost.localdomain kernel: gtp_gdbrsp_QT: return 0
Aug 03 11:00:18 localhost.localdomain kernel: gtp_read
（4）set trace-buffer-size unlimited
可以看到， This packet directs the target to make the trace buffer be of size size if possible. A value of
Aug 03 11:00:44 localhost.localdomain kernel: gtp_write: QTBuffer:size:-1
Aug 03 11:00:44 localhost.localdomain kernel: gtp_gdbrsp_QT: Buffer:size:-1
Aug 03 11:00:44 localhost.localdomain kernel: gtp_gdbrsp_qtbuffer
Aug 03 11:00:44 localhost.localdomain kernel: gtp_gdbrsp_qtbuffer:setting buffer size to 0
Aug 03 11:00:44 localhost.localdomain kernel: gtp_gdbrsp_QT: return 0
Aug 03 11:00:44 localhost.localdomain kernel: gtp_read
可以看到， This packet directs the target to make the trace buffer be of size size if possible. A value of
-1 tells the target to use whatever size it prefers.
因此，当我们接到GDB发来的size为-1的包时，可以直接忽略掉，而收到normal size的包时，需要跟当前使用的buffer size作对比，然后做相应处理。
5. KGTP中set-buffer-size的具体实现
目前KGTP的set-buffer-size分支，已经实现了对size处理的逻辑，参考[6]，具体代码如下：
static int
gtp_gdbrsp_qtbuffer(char *pkg)
{
#ifdef GTP_DEBUG
printk(GTP_DEBUG "gtp_gdbrsp_qtbuffer\n");
#endif
// Handle QTBuffer:circular:value
if (strncmp("circular:", pkg, 9) == 0) {
ULONGEST setting;
pkg += 9;
if (pkg[0] == '\0')
return -EINVAL;
hex2ulongest(pkg, &setting);
#ifdef GTP_FTRACE_RING_BUFFER
#if (LINUX_VERSION_CODE > KERNEL_VERSION(2,6,38)) \
|| defined(GTP_SELF_RING_BUFFER)
gtp_circular = (int)setting;
if (gtp_frame)
ring_buffer_change_overwrite(gtp_frame, (int)setting);
#else
if (gtp_circular != (int)setting)
gtp_circular_is_changed = 1;
#endif
#endif
gtp_circular = (int)setting;
return 0;
}
// Handle QTBuffer:size:size
else if (strncmp("size:", pkg, 5) == 0) {
ULONGEST size;
int unlimited_or_not;
pkg += 5;
if (pkg[0] == '\0')
return -EINVAL;
// The size may equal to -1, so compare the pkg with "-1"
if((strncmp("-1", pkg, 2) == 0) && (*(pkg + 2) == '\0')) {
unlimited_or_not = 1;
}
if(unlimited_or_not == 1) {
// The user wants KGTP to choose the right size
// So we just ignore the request
#ifdef GTP_DEBUG
printk(GTP_DEBUG "gtp_gdbrsp_qtbuffer:keep buffer size as the user tells KGTP to use unlimited size");
#endif
return 0;
}
// User wants to change the buffer size
hex2ulongest(pkg, &size);
#ifdef GTP_DEBUG
printk(GTP_DEBUG "gtp_gdbrsp_qtbuffer:setting buffer size to %ld\n", size);
#endif
// Handle the new ringbuffer size blow
// gtp_set_trace_buffer_size(size);
return 0;
}
return 1;
}
让我们测试一下：
（1）set trace-buffer-size -1
Aug 03 12:37:28 localhost.localdomain kernel: gtp_write: QTBuffer:size:-1
Aug 03 12:37:28 localhost.localdomain kernel: gtp_gdbrsp_QT: Buffer:size:-1
Aug 03 12:37:28 localhost.localdomain kernel: gtp_gdbrsp_qtbuffer
Aug 03 12:37:28 localhost.localdomain kernel: gtp_gdbrsp_qtbuffer:keep buffer size as the user tells KGTP to use unlimited size
Aug 03 12:37:28 localhost.localdomain kernel: gtp_gdbrsp_QT: return 0
Aug 03 12:37:28 localhost.localdomain kernel: gtp_read
（2）set trace-buffer-size unlimited
Bingo！目前已经正确地实现了对size的解析处理，接下来的工作是继续调研KGTP trace buffer的实现，并根据size大小做出调整。
Aug 03 12:37:57 localhost.localdomain kernel: gtp_write: QTBuffer:size:-1
Aug 03 12:37:57 localhost.localdomain kernel: gtp_gdbrsp_QT: Buffer:size:-1
Aug 03 12:37:57 localhost.localdomain kernel: gtp_gdbrsp_qtbuffer
Aug 03 12:37:57 localhost.localdomain kernel: gtp_gdbrsp_qtbuffer:keep buffer size as the user tells KGTP to use unlimited size
Aug 03 12:37:57 localhost.localdomain kernel: gtp_gdbrsp_QT: return 0
Aug 03 12:37:57 localhost.localdomain kernel: gtp_read
Bingo！目前已经正确地实现了对size的解析处理，接下来的工作是继续调研KGTP trace buffer的实现，并根据size大小做出调整。
6. 参考链接
[1] http://blog.csdn.net/calmdownba/article/details/38174759
[2] https://sourceware.org/gdb/current/onlinedocs/gdb/Packets.html#Packets
[3] https://sourceware.org/gdb/current/onlinedocs/gdb/Remote-Protocol.html#Remote-Protocol
[4] https://sourceware.org/gdb/current/onlinedocs/gdb/Tracepoint-Packets.html
[5] https://sourceware.org/gdb/current/onlinedocs/gdb/Starting-and-Stopping-Trace-Experiments.html
[6] https://code.csdn.net/Calmdownba/kgtp/tree/set-buffer-size
1万+
被折叠的 条评论
为什么被折叠?
到【灌水乐园】发言
