[Patch]Fix the bug of missing definition of argstr and retstr

[Patch]Fix the bug of missing definition of argstr and retstr

[Cai Fei]

Hi, all

After adding definition of pid and argstr for syscall getpgid, I checked 
the tapset
for other syscalls. The same problems of missing definition of argstr 
and retstr exist.
So I added them all:

diff -Nur systemtap-20070811/tapset/syscalls2.stp systemtap-20070811new/tapset/syscalls2.stp
--- systemtap-20070811/tapset/syscalls2.stp	2007-08-17 10:21:34.000000000 +0900
+++ systemtap-20070811new/tapset/syscalls2.stp	2007-08-17 10:25:10.000000000 +0900
@@ -2147,6 +2147,7 @@
 		kernel.function("sys32_sigreturn").return ?
 {
 	name = "sigreturn"
+	retstr = returnstr(1)
 }
 
 # sigsuspend _________________________________________________
diff -Nur systemtap-20070811/tapset/syscalls.stp systemtap-20070811new/tapset/syscalls.stp
--- systemtap-20070811/tapset/syscalls.stp	2007-08-17 10:18:12.000000000 +0900
+++ systemtap-20070811new/tapset/syscalls.stp	2007-08-17 10:24:46.000000000 +0900
@@ -1116,6 +1116,7 @@
 		kernel.function("sys_getegid")
 {
 	name = "getegid"
+	argstr = ""
 }
 probe syscall.getegid.return =
 		kernel.function("sys_getegid16").return ?,
@@ -1136,6 +1137,7 @@
 		kernel.function("sys_geteuid")
 {
 	name = "geteuid"
+	argstr = ""
 }
 probe syscall.geteuid.return = 
 		kernel.function("sys_geteuid16").return ?,
@@ -1156,6 +1158,7 @@
 		kernel.function("sys_getgid") 
 {
 	name = "getgid"
+	argstr = ""
 }
 probe syscall.getgid.return = 
 		kernel.function("sys_getgid16").return ?,
@@ -1282,6 +1285,7 @@
 # long sys_getpgrp(void)
 probe syscall.getpgrp = kernel.function("sys_getpgrp") ? {
 	name = "getpgrp"
+	argstr = ""
 }
 probe syscall.getpgrp.return = kernel.function("sys_getpgrp").return ? {
 	name = "getpgrp"
@@ -1292,6 +1296,7 @@
 # long sys_getpid(void)
 probe syscall.getpid = kernel.function("sys_getpid") {
 	name = "getpid"
+	argstr = ""
 }
 probe syscall.getpid.return = kernel.function("sys_getpid").return {
 	name = "getpid"
@@ -1302,9 +1307,11 @@
 # long sys_getppid(void)
 probe syscall.getppid = kernel.function("sys_getppid") {
 	name = "getppid"
+	argstr = ""
 }
 probe syscall.getppid.return = kernel.function("sys_getppid").return {
 	name = "getppid"
+	retstr = returnstr(1)
 }
 
 # getpriority ________________________________________________
@@ -1466,6 +1473,7 @@
 # long sys_gettid(void)
 probe syscall.gettid = kernel.function("sys_gettid") {
 	name = "gettid"
+	argstr = ""
 }
 probe syscall.gettid.return = kernel.function("sys_gettid").return {
 	name = "gettid"
@@ -1510,6 +1518,7 @@
 		kernel.function("sys_getuid")
 {
 	name = "getuid"
+	argstr = ""
 }
 probe syscall.getuid.return =
 		kernel.function("sys_getuid16").return ?,
@@ -2546,6 +2555,7 @@
 # long sys_munlockall(void)
 probe syscall.munlockall = kernel.function("sys_munlockall") {
 	name = "munlockall"
+	argstr = ""
 }
 probe syscall.munlockall.return = kernel.function("sys_munlockall").return {
 	name = "munlockall"


Regards, Cai Fei

Re: [Patch]Fix the bug of missing definition of argstr and retstr

[Frank Ch. Eigler]

Cai Fei <caifei@cn.fujitsu.com> writes:

> [...]
> After adding definition of pid and argstr for syscall getpgid, I
> checked the tapset
> for other syscalls. The same problems of missing definition of argstr
> and retstr exist.

There is generally no need to create variables that simply carry empty
strings.  The translator will do that automatically if necessary.

>
> 		kernel.function("sys_getegid")
> {
> 	name = "getegid"
> +	argstr = ""
> }

The only substantial effect of this is to prevent a probe based on
this alias from using "argstr" as a *numeric* variable.


- FChE

Re: [Patch]Fix the bug of missing definition of argstr and retstr

[Cai Fei]

Frank Ch. Eigler wrote:
> Cai Fei <caifei@cn.fujitsu.com> writes:
>
>   
>> [...]
>> After adding definition of pid and argstr for syscall getpgid, I
>> checked the tapset
>> for other syscalls. The same problems of missing definition of argstr
>> and retstr exist.
>>     
>
> There is generally no need to create variables that simply carry empty
> strings.  The translator will do that automatically if necessary.
>
>   
My probe definition is :
probe syscall.getpid {
log(sprint(argstr))
}
I used it for tracing syscall getpid, and the following error occured:
Pass 1: parsed user script and 54 library script(s) in 
397usr/2sys/399real ms.
semantic error: probe_1526 with unresolved type for identifier 'argstr' 
at getpid.stp:2:13
......

But it works well if I change it to
probe syscall.getpid {
log(sprintf("%s\n", argstr))
}

So I think it's better to add all these argstr definitions.
>> 		kernel.function("sys_getegid")
>> {
>> 	name = "getegid"
>> +	argstr = ""
>> }
>>     
>
> The only substantial effect of this is to prevent a probe based on
> this alias from using "argstr" as a *numeric* variable.
>
>
> - FChE
>
>
>   


-- 
A new email address of FJWAN is launched from Apr.1 2007.
The updated address is: caifei@cn.fujitsu.com
--------------------------------------------------
Cai Fei
Development Dept.I
Nanjing Fujitsu Nanda Software Tech. Co., Ltd.(FNST)
caifei@cn.fujitsu.com  (内7551-5875)
--------------------------------------------------

Re: [Patch]Fix the bug of missing definition of argstr and retstr

[Frank Ch. Eigler]

Hi -

On Fri, Aug 17, 2007 at 12:09:59PM +0900, Cai Fei wrote:
> My probe definition is :
> probe syscall.getpid {
> log(sprint(argstr))
> }

Why not just "log(argstr)" ?

> [...]
> So I think it's better to add all these argstr definitions.

Sure, there is no problem adding them, except that they make the
tapset file larger.

- FChE

Re: [Patch]Fix the bug of missing definition of argstr and retstr

[Cai Fei]

Frank Ch. Eigler wrote:
> Hi -
>
> On Fri, Aug 17, 2007 at 12:09:59PM +0900, Cai Fei wrote:
>   
>> My probe definition is :
>> probe syscall.getpid {
>> log(sprint(argstr))
>> }
>>     
>
> Why not just "log(argstr)" ?
>
>   
I used sprint so that numeric and non-numeric arguments can both be 
outputted
in the same way.
>> [...]
>> So I think it's better to add all these argstr definitions.
>>     
>
> Sure, there is no problem adding them, except that they make the
> tapset file larger.
>
> - FChE
>
>
>   

RE: [Patch]Fix the bug of missing definition of argstr and retstr

[Stone, Joshua I]

Cai Fei wrote:
> Frank Ch. Eigler wrote:
>> Hi -
>> 
>> On Fri, Aug 17, 2007 at 12:09:59PM +0900, Cai Fei wrote:
>> 
>>> My probe definition is :
>>> probe syscall.getpid {
>>> log(sprint(argstr))
>>> }
>>> 
>> 
>> Why not just "log(argstr)" ?
>> 
>> 
> I used sprint so that numeric and non-numeric arguments can both be
> outputted in the same way.

With bug #4591 resolved, you can now use println() for this.

Josh

Re: [Patch]Fix the bug of missing definition of argstr and retstr

[Cai Fei]

Hi, Frank.

Frank Ch. Eigler wrote:
> Hi -
>
>   
>> [...]
>> So I think it's better to add all these argstr definitions.
>>     
>
> Sure, there is no problem adding them, except that they make the
> tapset file larger.
>   
Does that mean you will add them? Or not? Sorry for the language question.