setscheduler() in Linux Kernel

everyone, I found this inside static int setscheduler(pid_t pid, int policy, struct sched_param *param) :

p->policy = policy;

if (policy != SCHED_OTHER) p->prio = MAX_USER_RT_PRIO-1 - p->rt_priority;
开发者_高级运维

p is a pointer to the task descriptor with current pid (parameter above) so if it's policy is not SCHED_OHTER (it means SCHED_FIFO or SCHED_RR) but why do we change p->prio such way? what exactly does it mean rt_priority? thanks in advance

Short answer: rt_priority means real-time priority and for SCHED_RR and SCHED_FIFO it decides how much time the process will get.

Long answer

First of all Linux implements something called Realtime Process Scheduling (SCHED_RR and SCHED_FIFO). Processes operating under these policies always have priority over other processes. Linux provides 99 realtime priority levels, numbered 1 (lowest) to 99 (highest).

In the kernel, if I recall correctly "smaller numbers mean better preference" - smaller p->prio means better priority.

Here us what sched_setscheduler looks like:

int sched_setscheduler(pid_t pid, int policy, const struct sched_param *param);

And sched_param

struct sched_param {
     int sched_priority;        /* Scheduling priority */
};

The integer sched_priority, for a policy of SCHED_RR or SCHED_FIFO means "realtime priority" which I assume is rt_priority. So the kernel does the right thing

• Take the maximmum priority
• Subtract the real time priority from that value. The bigger the priority (the more it subtracts) the better treatment by the scheduler.