Re: [patch 20/24] perfmon: system calls interface

[Ingo Molnar <mingo@elte.hu>]

* eranian@googlemail.com <eranian@googlemail.com> wrote:

> +/*
> + * unlike the other perfmon system calls, this one returns a file descriptor
> + * or a value < 0 in case of error, very much like open() or socket()
> + */
> +asmlinkage long sys_pfm_create(int flags, struct pfarg_sinfo __user *ureq)
> +{
> +	struct pfm_context *new_ctx;
> +	struct pfarg_sinfo sif;
> +	int ret;
> +
> +	PFM_DBG("flags=0x%x sif=%p", flags, ureq);
> +
> +	if (perfmon_disabled)
> +		return -ENOSYS;

uhm. So we have a dynamic 'we dont support perfmon' flag. Which is 
global and defined as:

  +int perfmon_disabled;  /* >0 if perfmon is disabled */

(sidenote: that should be __read_mostly)

then we go:

> +	if (flags) {
> +		PFM_DBG("no flags accepted yet");
> +		return -EINVAL;
> +	}

that should be if (unlikely())

then:

> +	ret = __pfm_create_context(flags, &sif, &new_ctx);

where we get:

+int __pfm_create_context(__u32 ctx_flags,
+                        struct pfarg_sinfo *sif,
+                        struct pfm_context **new_ctx)
+{
+       struct pfm_context *ctx;
+       struct file *filp = NULL;
+       int fd = 0, ret = -EINVAL;
+
+       if (!pfm_pmu_conf)
+               return -ENOSYS;
+

_ANOTHER_ global dynamic flag to tell us that ... in essence 'we dont 
support perfmon'. Which flag is again:

+struct pfm_pmu_config  *pfm_pmu_conf;

... which should be __read_mostly at minimum.

+EXPORT_SYMBOL(pfm_pmu_conf);

and _MUST_ be exported as EXPORT_SYMBOL_GPL. If exported at all. Why 
are any symbols exported here? perfmom does core kernel system calls 
and is non-modular:

+config PERFMON
+       bool "Perfmon2 performance monitoring interface"

it needs _zero_ exports.

	Ingo