Re: [PATCH v2] Pre-emption control for userspace

[Andi Kleen <andi@firstfloor.org>]

Khalid Aziz <khalid.aziz@oracle.com> writes:

First it would be nice to have some standard reference lock library that
uses this. What would it take to support this in glibc?

> +==================================
> +Using the preemption delay feature
> +==================================
> +
> +This feature is enabled in the kernel by setting
> +CONFIG_SCHED_PREEMPT_DELAY in kernel configuration. Once this feature is
> +enabled, the userspace process communicates with the kernel using a
> +4-byte memory location in its address space. It first gives the kernel
> +address for this memory location by writing its address to
> +/proc/<tgid>/task/<tid>/sched_preempt_delay. This memory location is
> +interpreted as a sequence of 4 bytes:
> +
> +	byte[0] = flag to request preemption delay
> +	byte[1] = flag from kernel indicating preemption delay was granted
> +	byte[2] = reserved for future use
> +	byte[3] = reserved for future use

Should reserve more bytes (64, 128?) and rename the proc flag
to a more generic name. I could well assume other things
using such a mechanism in the future. Also please add a flag
word with feature bits (similar to the perf mmap page)

How about alignment? x86 will not care, but other architectures
may. 

>  #endif
> +#ifdef CONFIG_SCHED_PREEMPT_DELAY
> +	REG("sched_preempt_delay", S_IRUGO|S_IWUSR,
> proc_tid_preempt_delay_ops),

This shouldn't be readable by group/other, as it exposes the address space,
so could help exploits.

> @@ -2061,6 +2069,13 @@ extern u64 scheduler_tick_max_deferment(void);
>  static inline bool sched_can_stop_tick(void) { return false; }
>  #endif
>  
> +#if defined(CONFIG_SCHED_PREEMPT_DELAY) && defined(CONFIG_PROC_FS)
> +extern void sched_preempt_delay_show(struct seq_file *m,
> +					struct task_struct *task);
> +extern void sched_preempt_delay_set(struct task_struct *task,
> +					unsigned char *val);
> +#endif

Prototypes don't need to be ifdefed.

-Andi

-- 
ak@linux.intel.com -- Speaking for myself only