path: root/arch/tile/kernel/process.c

/*
 * Copyright 2010 Tilera Corporation. All Rights Reserved.
 *
 *   This program is free software; you can redistribute it and/or
 *   modify it under the terms of the GNU General Public License
 *   as published by the Free Software Foundation, version 2.
 *
 *   This program is distributed in the hope that it will be useful, but
 *   WITHOUT ANY WARRANTY; without even the implied warranty of
 *   MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE, GOOD TITLE or
 *   NON INFRINGEMENT.  See the GNU General Public License for
 *   more details.
 */

#include <linux/sched.h>
#include <linux/preempt.h>
#include <linux/module.h>
#include <linux/fs.h>
#include <linux/kprobes.h>
#include <linux/elfcore.h>
#include <linux/tick.h>
#include <linux/init.h>
#include <linux/mm.h>
#include <linux/compat.h>
#include <linux/hardirq.h>
#include <linux/syscalls.h>
#include <linux/kernel.h>
#include <linux/tracehook.h>
#include <linux/signal.h>
#include <asm/stack.h>
#include <asm/switch_to.h>
#include <asm/homecache.h>
#include <asm/syscalls.h>
#include <asm/traps.h>
#include <asm/setup.h>
#ifdef CONFIG_HARDWALL
#include <asm/hardwall.h>
#endif
#include <arch/chip.h>
#include <arch/abi.h>
#include <arch/sim_def.h>

/*
 * Use the (x86) "idle=poll" option to prefer low latency when leaving the
 * idle loop over low power while in the idle loop, e.g. if we have
 * one thread per core and we want to get threads out of futex waits fast.
 */
static int __init idle_setup(char *str)
{
	if (!str)
		return -EINVAL;

	if (!strcmp(str, "poll")) {
		pr_info("using polling idle threads.\n");
		cpu_idle_poll_ctrl(true);
		return 0;
	} else if (!strcmp(str, "halt")) {
		return 0;
	}
	return -1;
}
early_param("idle", idle_setup);

void arch_cpu_idle(void)
{
	__get_cpu_var(irq_stat).idle_timestamp = jiffies;
	_cpu_idle();
}

/*
 * Release a thread_info structure
 */
void arch_release_thread_info(struct thread_info *info)
{
	struct single_step_state *step_state = info->step_state;

#ifdef CONFIG_HARDWALL
	/*
	 * We free a thread_info from the context of the task that has
	 * been scheduled next, so the original task is already dead.
	 * Calling deactivate here just frees up the data structures.
	 * If the task we're freeing held the last reference to a
	 * hardwall fd, it would have been released prior to this point
	 * anyway via exit_files(), and the hardwall_task.info pointers
	 * would be NULL by now.
	 */
	hardwall_deactivate_all(info->task);
#endif

	if (step_state) {

		/*
		 * FIXME: we don't munmap step_state->buffer
		 * because the mm_struct for this process (info->task->mm)
		 * has already been zeroed in exit_mm().  Keeping a
		 * reference to it here seems like a bad move, so this
		 * means we can't munmap() the buffer, and therefore if we
		 * ptrace multiple threads in a process, we will slowly
		 * leak user memory.  (Note that as soon as the last
		 * thread in a process dies, we will reclaim all user
		 * memory including single-step buffers in the usual way.)
		 * We should either assign a kernel VA to this buffer
		 * somehow, or we should associate the buffer(s) with the
		 * mm itself so we can clean them up that way.
		 */
		kfree(step_state);
	}
}

static void save_arch_state(struct thread_struct *t);

int copy_thread(unsigned long clone_flags, unsigned long sp,
		unsigned long arg, struct task_struct *p)
{
	struct pt_regs *childregs = task_pt_regs(p);
	unsigned long ksp;
	unsigned long *callee_regs;

	/*
	 * Set up the stack and stack pointer appropriately for the
	 * new child to find itself woken up in __switch_to().
	 * The callee-saved registers must be on the stack to be read;
	 * the new task will then jump to assembly support to handle
	 * calling schedule_tail(), etc., and (for userspace tasks)
	 * returning to the context set up in the pt_regs.
	 */
	ksp = (unsigned long) childregs;
	ksp -= C_ABI_SAVE_AREA_SIZE;   /* interrupt-entry save area */
	((long *)ksp)[0] = ((long *)ksp)[1] = 0;
	ksp -= CALLEE_SAVED_REGS_COUNT * sizeof(unsigned long);
	callee_regs = (unsigned long *)ksp;
	ksp -= C_ABI_SAVE_AREA_SIZE;   /* __switch_to() save area */
	((long *)ksp)[0] = ((long *)ksp)[1] = 0;
	p->thread.ksp = ksp;

	/* Record the pid of the task that created this one. */
	p->thread.creator_pid = current->pid;

	if (unlikely(p->flags & PF_KTHREAD)) {
		/* kernel thread */
		memset(childregs, 0, sizeof(struct pt_regs));
		memset(&callee_regs[2], 0,
		       (CALLEE_SAVED_REGS_COUNT - 2) * sizeof(unsigned long));
		callee_regs[0] = sp;   /* r30 = function */
		callee_regs[1] = arg;  /* r31 = arg */
		childregs->ex1 = PL_ICS_EX1(KERNEL_PL, 0);
		p->thread.pc = (unsigned long) ret_from_kernel_thread;
		return 0;
	}

	/*
	 * Start new thread in ret_from_fork so it schedules properly
	 * and then return from interrupt like the parent.
	 */
	p->thread.pc = (unsigned long) ret_from_fork;

	/*
	 * Do not clone step state from the parent; each thread
	 * must make its own lazily.
	 */
	task_thread_info(p)->step_state = NULL;

	/*
	 * Copy the registers onto the kernel stack so the
	 * return-from-interrupt code will reload it into registers.
	 */
	*childregs = *current_pt_regs();
	childregs->regs[0] = 0;         /* return value is zero */
	if (sp)
		childregs->sp = sp;  /* override with new user stack pointer */
	memcpy(callee_regs, &childregs->regs[CALLEE_SAVED_FIRST_REG],
	       CALLEE_SAVED_REGS_COUNT * sizeof(unsigned long));

	/* Save user stack top pointer so we can ID the stack vm area later. */
	p->thread.usp0 = childregs->sp;

	/*
	 * If CLONE_SETTLS is set, set "tp" in the new task to "r4",
	 * which is passed in as arg #5 to sys_clone().
	 */
	if (clone_flags & CLONE_SETTLS)
		childregs->tp = childregs->regs[4];


#if CHIP_HAS_TILE_DMA()
	/*
	 * No DMA in the new thread.  We model this on the fact that
	 * fork() clears the pending signals, alarms, and aio for the child.
	 */
	memset(&p->thread.tile_dma_state, 0, sizeof(struct tile_dma_state));
	memset(&p->thread.dma_async_tlb, 0, sizeof(struct async_tlb));
#endif

#if CHIP_HAS_SN_PROC()
	/* Likewise, the new thread is not running static processor code. */
	p->thread.sn_proc_running = 0;
	memset(&p->thread.sn_async_tlb, 0, sizeof(struct async_tlb));
#endif

#if CHIP_HAS_PROC_STATUS_SPR()
	/* New thread has its miscellaneous processor state bits clear. */
	p->thread.proc_status = 0;
#endif

#ifdef CONFIG_HARDWALL
	/* New thread does not own any networks. */
	memset(&p->thread.hardwall[0], 0,
	       sizeof(struct hardwall_task) * HARDWALL_TYPES);
#endif


	/*
	 * Start the new thread with the current architecture state
	 * (user interrupt masks, etc.).
	 */
	save_arch_state(&p->thread);

	return 0;
}

/*
 * Return "current" if it looks plausible, or else a pointer to a dummy.
 * This can be helpful if we are just trying to emit a clean panic.
 */
struct task_struct *validate_current(void)
{
	static struct task_struct corrupt = { .comm = "<corrupt>" };
	struct task_struct *tsk = current;
	if (unlikely((unsigned long)tsk < PAGE_OFFSET ||
		     (high_memory && (void *)tsk > high_memory) ||
		     ((unsigned long)tsk & (__alignof__(*tsk) - 1)) != 0)) {
		pr_err("Corrupt 'current' %p (sp %#lx)\n", tsk, stack_pointer);
		tsk = &corrupt;
	}
	return tsk;
}

/* Take and return the pointer to the previous task, for schedule_tail(). */
struct task_struct *sim_notify_fork(struct task_struct *prev)
{
	struct task_struct *tsk = current;
	__insn_mtspr(SPR_SIM_CONTROL, SIM_CONTROL_OS_FORK_PARENT |
		     (tsk->thread.creator_pid << _SIM_CONTROL_OPERATOR_BITS));
	__insn_mtspr(SPR_SIM_CONTROL, SIM_CONTROL_OS_FORK |
		     (tsk->pid << _SIM_CONTROL_OPERATOR_BITS));
	return prev;
}

int dump_task_regs(struct task_struct *tsk, elf_gregset_t *regs)
{
	struct pt_regs *ptregs = task_pt_regs(tsk);
	elf_core_copy_regs(regs, ptregs);
	return 1;
}

#if CHIP_HAS_TILE_DMA()

/* Allow user processes to access the DMA SPRs */
void grant_dma_mpls(void)
{
#if CONFIG_KERNEL_PL == 2
	__insn_mtspr(SPR_MPL_DMA_CPL_SET_1,