Re: [PATCH v3] Tracepoint Tapset for Memory Subsystem

[fche at redhat dot com (Frank Ch. Eigler)]

Rajasekhar Duddu <rajduddu@linux.vnet.ibm.com> writes:

> [...]
> +/* Function to convert the GFP_FLAGS . */
> +
> +function gfp_flag_str:string (gfp_flag:long)
> +%{
> +int flags = (int)THIS->gfp_flag;
> +THIS->__retvalue[0] = '\0';
> +
> +#ifdef __GFP_HIGH
> +        if (flags & __GFP_HIGH)
> +                strlcat (THIS->__retvalue, "GFP_HIGH",MAXSTRINGLEN);
> +#endif
> +
> +#ifdef __GFP_WAIT
> +        if (flags & __GFP_WAIT)
> +                strlcat (THIS->__retvalue, "GFP_WAIT",MAXSTRINGLEN);
> +#endif
> +
> +#ifdef __GFP_IO
> +        if (flags & __GFP_IO)
> +                strlcat (THIS->__retvalue, "|GFP_IO",MAXSTRINGLEN);
> +#endif

Why no "|" before GFP_HIGH/GFP_WAIT?
Also, why no "__" before the stringified version?


> +#ifdef __GFP_FS
> +        if (flags & __GFP_FS)
> +                strlcat (THIS->__retvalue, "|GFP_FS",MAXSTRINGLEN);
> +#endif

(How about a macro to generate all these near-identical branches?)


> +%}


> +/**
> + * probe vm.kmalloc - Fires when <command>kmalloc</command> is requested.
> + * @call_site: Address of the caller function.
> + * @caller_function: Name of the caller function.
> + * @bytes_req: Requested Bytes
> + * @bytes_alloc: Allocated Bytes
> + * @gfp_flags: type of kmemory to allocate
> + * @ptr: Pointer to the kmemory allocated
> + */
> +
> +probe vm.kmalloc = kernel.trace("kmalloc") {
> +	name = "kmalloc"
> +	call_site = $call_site
> +	caller_function = symname(call_site)
> +	bytes_req = $bytes_req
> +	bytes_alloc = $bytes_alloc
> +	gfp_flags = gfp_flag_str($gfp_flags)
> +	ptr = $ptr
> +}

Nice.  I thought that the raison d'etre for these aliases was to
abstract the presence or absence of tracepoints, so is there no
fallback kprobe available?  Something like this:


> +probe __vm.kfree.kp = kernel.function("kfree") {
> +	name = "kfree"
> +	call_site = "0"

(Note though that this will fail type checking on a non-tracepoint
kernel -- have you tried it? -- it should be just 0 instead of "0".)

> +	caller_function = "unknown"
> +	ptr = $x
> +}
> +
> +probe __vm.kfree.tp = kernel.trace("kfree") {
> +	name = "kfree"
> +	call_site = $call_site
> +	caller_function = symname(call_site)
> +	ptr = $ptr
> +}
> +
> +/**
> + * probe vm.kfree - Fires when <command>kfree</comand> is requested.
> + * @call_site: Address of the caller function (displayed if available)
> + * @caller_function - Name of the caller function (displayed if available)
> + * @ptr: Pointer to the kmemory allocated which is returned by kmalloc
> + */
> +probe vm.kfree = __vm.kfree.tp !,
> +		   __vm.kfree.kp
> +{}

Right.


> +/**
> + * probe vm.kmalloc_node - Fires when <command>kmalloc_node</command> is requested.
> + * @call_site: Address of the caller function.
> + * @caller_function: Name of the caller function.
> + * @bytes_req: Requested Bytes
> + * @bytes_alloc: Allocated Bytes
> + * @gfp_flags: Type of kmemory to allocate
> + * @ptr: Pointer to the kmemory allocated
> + */

Please, no "<command>" markup in there, it is not valid.


> +probe vm.kmalloc_node = kernel.trace("kmalloc_node")? {
> [...]

Why is this marked with "?"?


> --- a/testsuite/buildok/vm.tracepoints.stp	1969-12-31 19:00:00.000000000 -0500
> +++ b/testsuite/buildok/vm.tracepoints.stp	2009-10-02 10:59:20.000000000 -0400
> @@ -0,0 +1,32 @@
> +#!/usr/bin/stp -up4

Other similar test cases just use

#! stap -up4



- FChE