/* * 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/system.h>#include <asm/stack.h>#include <asm/homecache.h>#include <asm/syscalls.h>#include <asm/traps.h>#ifdef CONFIG_HARDWALL#include <asm/hardwall.h>#endif#include <arch/chip.h>#include <arch/abi.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 no_idle_nap;static int __init idle_setup(char *str){ if (!str) return -EINVAL; if (!strcmp(str, "poll")) { pr_info("using polling idle threads.\n"); no_idle_nap = 1; } else if (!strcmp(str, "halt")) no_idle_nap = 0; else return -1; return 0;}early_param("idle", idle_setup); /* * The idle thread. There's no useful work to be * done, so just try to conserve power and have a * low exit latency (ie sit in a loop waiting for * somebody to say that they'd like to reschedule) */void cpu_idle(void){ int cpu = smp_processor_id(); current_thread_info()->status |= TS_POLLING; if (no_idle_nap) { while (1) { while (!need_resched()) cpu_relax(); schedule(); } } /* endless idle loop with no priority at all */ while (1) { tick_nohz_idle_enter(); rcu_idle_enter(); while (!need_resched()) { if (cpu_is_offline(cpu)) BUG(); /* no HOTPLUG_CPU */ local_irq_disable(); __get_cpu_var(irq_stat).idle_timestamp = jiffies; current_thread_info()->status &= ~TS_POLLING; /* * TS_POLLING-cleared state must be visible before we * test NEED_RESCHED: */ smp_mb(); if (!need_resched()) _cpu_idle(); else local_irq_enable(); current_thread_info()->status |= TS_POLLING; } rcu_idle_exit(); tick_nohz_idle_exit(); preempt_enable_no_resched(); schedule(); preempt_disable(); }} struct thread_info *alloc_thread_info_node(struct task_struct *task, int node){ struct page *page; gfp_t flags = GFP_KERNEL; #ifdef CONFIG_DEBUG_STACK_USAGE flags |= __GFP_ZERO;#endif page = alloc_pages_node(node, flags, THREAD_SIZE_ORDER); if (!page) return NULL; return (struct thread_info *)page_address(page);} /* * Free a thread_info node, and all of its derivative * data structures. */void free_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 "hardwall" would be NULL by now. */ if (info->task->thread.hardwall) hardwall_deactivate(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); } free_pages((unsigned long)info, THREAD_SIZE_ORDER);} static void save_arch_state(struct thread_struct *t); int copy_thread(unsigned long clone_flags, unsigned long sp, unsigned long stack_size, struct task_struct *p, struct pt_regs *regs){ struct pt_regs *childregs; unsigned long ksp; /* * When creating a new kernel thread we pass sp as zero. * Assign it to a reasonable value now that we have the stack. */ if (sp == 0 && regs->ex1 == PL_ICS_EX1(KERNEL_PL, 0)) sp = KSTK_TOP(p); /* * Do not clone step state from the parent; each thread * must make its own lazily. */ task_thread_info(p)->step_state = NULL; /* * 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; /* Save user stack top pointer so we can ID the stack vm area later. */ p->thread.usp0 = sp; /* Record the pid of the process that created this one. */ p->thread.creator_pid = current->pid; /* * Copy the registers onto the kernel stack so the * return-from-interrupt code will reload it into registers. */ childregs = task_pt_regs(p); *childregs = *regs; childregs->regs[0] = 0; /* return value is zero */ childregs->sp = sp; /* override with new user stack pointer */ /* * 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 = regs->regs[4]; /* * Copy the callee-saved registers from the passed pt_regs struct * into the context-switch callee-saved registers area. * This way when we start the interrupt-return sequence, the * callee-save registers will be correctly in registers, which * is how we assume the compiler leaves them as we start doing * the normal return-from-interrupt path after calling C code. * Zero out the C ABI save area to mark the top of the stack. */ 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); memcpy((void *)ksp, ®s->regs[CALLEE_SAVED_FIRST_REG], CALLEE_SAVED_REGS_COUNT * sizeof(unsigned long)); ksp -= C_ABI_SAVE_AREA_SIZE; /* __switch_to() save area */ ((long *)ksp)[0] = ((long *)ksp)[1] = 0; p->thread.ksp = ksp; #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. */ p->thread.hardwall = NULL;#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 || (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, 1); __insn_mtspr(SPR_MPL_DMA_NOTIFY_SET_1, 1);#else __insn_mtspr(SPR_MPL_DMA_CPL_SET_0, 1); __insn_mtspr(SPR_MPL_DMA_NOTIFY_SET_0, 1);#endif} /* Forbid user processes from accessing the DMA SPRs */void restrict_dma_mpls(void){#if CONFIG_KERNEL_PL == 2 __insn_mtspr(SPR_MPL_DMA_CPL_SET_2, 1); __insn_mtspr(SPR_MPL_DMA_NOTIFY_SET_2, 1);#else __insn_mtspr(SPR_MPL_DMA_CPL_SET_1, 1); __insn_mtspr(SPR_MPL_DMA_NOTIFY_SET_1, 1);#endif} /* Pause the DMA engine, then save off its state registers. */static void save_tile_dma_state(struct tile_dma_state *dma){ unsigned long state = __insn_mfspr(SPR_DMA_USER_STATUS); unsigned long post_suspend_state; /* If we're running, suspend the engine. */ if ((state & DMA_STATUS_MASK) == SPR_DMA_STATUS__RUNNING_MASK) __insn_mtspr(SPR_DMA_CTR, SPR_DMA_CTR__SUSPEND_MASK); /* * Wait for the engine to idle, then save regs. Note that we * want to record the "running" bit from before suspension, * and the "done" bit from after, so that we can properly * distinguish a case where the user suspended the engine from * the case where the kernel suspended as part of the context * swap. */ do { post_suspend_state = __insn_mfspr(SPR_DMA_USER_STATUS); } while (post_suspend_state & SPR_DMA_STATUS__BUSY_MASK); dma->src = __insn_mfspr(SPR_DMA_SRC_ADDR); dma->src_chunk = __insn_mfspr(SPR_DMA_SRC_CHUNK_ADDR); dma->dest = __insn_mfspr(SPR_DMA_DST_ADDR); dma->dest_chunk = __insn_mfspr(SPR_DMA_DST_CHUNK_ADDR); dma->strides = __insn_mfspr(SPR_DMA_STRIDE); dma->chunk_size = __insn_mfspr(SPR_DMA_CHUNK_SIZE); dma->byte = __insn_mfspr(SPR_DMA_BYTE); dma->status = (state & SPR_DMA_STATUS__RUNNING_MASK) | (post_suspend_state & SPR_DMA_STATUS__DONE_MASK);} /* Restart a DMA that was running before we were context-switched out. */static void restore_tile_dma_state(struct thread_struct *t){ const struct tile_dma_state *dma = &t->tile_dma_state; /* * The only way to restore the done bit is to run a zero * length transaction. */ if ((dma->status & SPR_DMA_STATUS__DONE_MASK) && !(__insn_mfspr(SPR_DMA_USER_STATUS) & SPR_DMA_STATUS__DONE_MASK)) { __insn_mtspr(SPR_DMA_BYTE, 0); __insn_mtspr(SPR_DMA_CTR, SPR_DMA_CTR__REQUEST_MASK); while (__insn_mfspr(SPR_DMA_USER_STATUS) & SPR_DMA_STATUS__BUSY_MASK) ; } __insn_mtspr(SPR_DMA_SRC_ADDR, dma->src); __insn_mtspr(SPR_DMA_SRC_CHUNK_ADDR, dma->src_chunk); __insn_mtspr(SPR_DMA_DST_ADDR, dma->dest); __insn_mtspr(SPR_DMA_DST_CHUNK_ADDR, dma->dest_chunk); __insn_mtspr(SPR_DMA_STRIDE, dma->strides); __insn_mtspr(SPR_DMA_CHUNK_SIZE, dma->chunk_size); __insn_mtspr(SPR_DMA_BYTE, dma->byte); /* * Restart the engine if we were running and not done. * Clear a pending async DMA fault that we were waiting on return * to user space to execute, since we expect the DMA engine * to regenerate those faults for us now. Note that we don't * try to clear the TIF_ASYNC_TLB flag, since it's relatively * harmless if set, and it covers both DMA and the SN processor. */ if ((dma->status & DMA_STATUS_MASK) == SPR_DMA_STATUS__RUNNING_MASK) { t->dma_async_tlb.fault_num = 0; __insn_mtspr(SPR_DMA_CTR, SPR_DMA_CTR__REQUEST_MASK); }} #endif static void save_arch_state(struct thread_struct *t){#if CHIP_HAS_SPLIT_INTR_MASK() t->interrupt_mask = __insn_mfspr(SPR_INTERRUPT_MASK_0_0) | ((u64)__insn_mfspr(SPR_INTERRUPT_MASK_0_1) << 32);#else t->interrupt_mask = __insn_mfspr(SPR_INTERRUPT_MASK_0);#endif t->ex_context[0] = __insn_mfspr(SPR_EX_CONTEXT_0_0); t->ex_context[1] = __insn_mfspr(SPR_EX_CONTEXT_0_1); t->system_save[0] = __insn_mfspr(SPR_SYSTEM_SAVE_0_0); t->system_save[1] = __insn_mfspr(SPR_SYSTEM_SAVE_0_1); t->system_save[2] = __insn_mfspr(SPR_SYSTEM_SAVE_0_2); t->system_save[3] = __insn_mfspr(SPR_SYSTEM_SAVE_0_3); t->intctrl_0 = __insn_mfspr(SPR_INTCTRL_0_STATUS);#if CHIP_HAS_PROC_STATUS_SPR() t->proc_status = __insn_mfspr(SPR_PROC_STATUS);#endif#if !CHIP_HAS_FIXED_INTVEC_BASE() t->interrupt_vector_base = __insn_mfspr(SPR_INTERRUPT_VECTOR_BASE_0);#endif#if CHIP_HAS_TILE_RTF_HWM() t->tile_rtf_hwm = __insn_mfspr(SPR_TILE_RTF_HWM);#endif#if CHIP_HAS_DSTREAM_PF() t->dstream_pf = __insn_mfspr(SPR_DSTREAM_PF);#endif} static void restore_arch_state(const struct thread_struct *t){#if CHIP_HAS_SPLIT_INTR_MASK() __insn_mtspr(SPR_INTERRUPT_MASK_0_0, (u32) t->interrupt_mask); __insn_mtspr(SPR_INTERRUPT_MASK_0_1, t->interrupt_mask >> 32);#else __insn_mtspr(SPR_INTERRUPT_MASK_0, t->interrupt_mask);#endif __insn_mtspr(SPR_EX_CONTEXT_0_0, t->ex_context[0]); __insn_mtspr(SPR_EX_CONTEXT_0_1, t->ex_context[1]); __insn_mtspr(SPR_SYSTEM_SAVE_0_0, t->system_save[0]); __insn_mtspr(SPR_SYSTEM_SAVE_0_1, t->system_save[1]); __insn_mtspr(SPR_SYSTEM_SAVE_0_2, t->system_save[2]); __insn_mtspr(SPR_SYSTEM_SAVE_0_3, t->system_save[3]); __insn_mtspr(SPR_INTCTRL_0_STATUS, t->intctrl_0);#if CHIP_HAS_PROC_STATUS_SPR() __insn_mtspr(SPR_PROC_STATUS, t->proc_status);#endif#if !CHIP_HAS_FIXED_INTVEC_BASE() __insn_mtspr(SPR_INTERRUPT_VECTOR_BASE_0, t->interrupt_vector_base);#endif#if CHIP_HAS_TILE_RTF_HWM() __insn_mtspr(SPR_TILE_RTF_HWM, t->tile_rtf_hwm);#endif#if CHIP_HAS_DSTREAM_PF() __insn_mtspr(SPR_DSTREAM_PF, t->dstream_pf);#endif} void _prepare_arch_switch(struct task_struct *next){#if CHIP_HAS_SN_PROC() int snctl;#endif#if CHIP_HAS_TILE_DMA() struct tile_dma_state *dma = ¤t->thread.tile_dma_state; if (dma->enabled) save_tile_dma_state(dma);#endif#if CHIP_HAS_SN_PROC() /* * Suspend the static network processor if it was running. * We do not suspend the fabric itself, just like we don't * try to suspend the UDN. */ snctl = __insn_mfspr(SPR_SNCTL); current->thread.sn_proc_running = (snctl & SPR_SNCTL__FRZPROC_MASK) == 0; if (current->thread.sn_proc_running) __insn_mtspr(SPR_SNCTL, snctl | SPR_SNCTL__FRZPROC_MASK);#endif} struct task_struct *__sched _switch_to(struct task_struct *prev,Lines 1–500 of 746. Load more
/* * 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/system.h> #include <asm/stack.h> #include <asm/homecache.h> #include <asm/syscalls.h> #include <asm/traps.h> #ifdef CONFIG_HARDWALL #include <asm/hardwall.h> #endif #include <arch/chip.h> #include <arch/abi.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 no_idle_nap; static int __init idle_setup(char *str) { if (!str) return -EINVAL; if (!strcmp(str, "poll")) { pr_info("using polling idle threads.\n"); no_idle_nap = 1; } else if (!strcmp(str, "halt")) no_idle_nap = 0; else return -1; return 0; } early_param("idle", idle_setup); /* * The idle thread. There's no useful work to be * done, so just try to conserve power and have a * low exit latency (ie sit in a loop waiting for * somebody to say that they'd like to reschedule) */ void cpu_idle(void) { int cpu = smp_processor_id(); current_thread_info()->status |= TS_POLLING; if (no_idle_nap) { while (1) { while (!need_resched()) cpu_relax(); schedule(); } } /* endless idle loop with no priority at all */ while (1) { tick_nohz_idle_enter(); rcu_idle_enter(); while (!need_resched()) { if (cpu_is_offline(cpu)) BUG(); /* no HOTPLUG_CPU */ local_irq_disable(); __get_cpu_var(irq_stat).idle_timestamp = jiffies; current_thread_info()->status &= ~TS_POLLING; /* * TS_POLLING-cleared state must be visible before we * test NEED_RESCHED: */ smp_mb(); if (!need_resched()) _cpu_idle(); else local_irq_enable(); current_thread_info()->status |= TS_POLLING; } rcu_idle_exit(); tick_nohz_idle_exit(); preempt_enable_no_resched(); schedule(); preempt_disable(); } } struct thread_info *alloc_thread_info_node(struct task_struct *task, int node) { struct page *page; gfp_t flags = GFP_KERNEL; #ifdef CONFIG_DEBUG_STACK_USAGE flags |= __GFP_ZERO; #endif page = alloc_pages_node(node, flags, THREAD_SIZE_ORDER); if (!page) return NULL; return (struct thread_info *)page_address(page); } /* * Free a thread_info node, and all of its derivative * data structures. */ void free_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 "hardwall" would be NULL by now. */ if (info->task->thread.hardwall) hardwall_deactivate(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); } free_pages((unsigned long)info, THREAD_SIZE_ORDER); } static void save_arch_state(struct thread_struct *t); int copy_thread(unsigned long clone_flags, unsigned long sp, unsigned long stack_size, struct task_struct *p, struct pt_regs *regs) { struct pt_regs *childregs; unsigned long ksp; /* * When creating a new kernel thread we pass sp as zero. * Assign it to a reasonable value now that we have the stack. */ if (sp == 0 && regs->ex1 == PL_ICS_EX1(KERNEL_PL, 0)) sp = KSTK_TOP(p); /* * Do not clone step state from the parent; each thread * must make its own lazily. */ task_thread_info(p)->step_state = NULL; /* * 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; /* Save user stack top pointer so we can ID the stack vm area later. */ p->thread.usp0 = sp; /* Record the pid of the process that created this one. */ p->thread.creator_pid = current->pid; /* * Copy the registers onto the kernel stack so the * return-from-interrupt code will reload it into registers. */ childregs = task_pt_regs(p); *childregs = *regs; childregs->regs[0] = 0; /* return value is zero */ childregs->sp = sp; /* override with new user stack pointer */ /* * 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 = regs->regs[4]; /* * Copy the callee-saved registers from the passed pt_regs struct * into the context-switch callee-saved registers area. * This way when we start the interrupt-return sequence, the * callee-save registers will be correctly in registers, which * is how we assume the compiler leaves them as we start doing * the normal return-from-interrupt path after calling C code. * Zero out the C ABI save area to mark the top of the stack. */ 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); memcpy((void *)ksp, ®s->regs[CALLEE_SAVED_FIRST_REG], CALLEE_SAVED_REGS_COUNT * sizeof(unsigned long)); ksp -= C_ABI_SAVE_AREA_SIZE; /* __switch_to() save area */ ((long *)ksp)[0] = ((long *)ksp)[1] = 0; p->thread.ksp = ksp; #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. */ p->thread.hardwall = NULL; #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 || (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, 1); __insn_mtspr(SPR_MPL_DMA_NOTIFY_SET_1, 1); #else __insn_mtspr(SPR_MPL_DMA_CPL_SET_0, 1); __insn_mtspr(SPR_MPL_DMA_NOTIFY_SET_0, 1); #endif } /* Forbid user processes from accessing the DMA SPRs */ void restrict_dma_mpls(void) { #if CONFIG_KERNEL_PL == 2 __insn_mtspr(SPR_MPL_DMA_CPL_SET_2, 1); __insn_mtspr(SPR_MPL_DMA_NOTIFY_SET_2, 1); #else __insn_mtspr(SPR_MPL_DMA_CPL_SET_1, 1); __insn_mtspr(SPR_MPL_DMA_NOTIFY_SET_1, 1); #endif } /* Pause the DMA engine, then save off its state registers. */ static void save_tile_dma_state(struct tile_dma_state *dma) { unsigned long state = __insn_mfspr(SPR_DMA_USER_STATUS); unsigned long post_suspend_state; /* If we're running, suspend the engine. */ if ((state & DMA_STATUS_MASK) == SPR_DMA_STATUS__RUNNING_MASK) __insn_mtspr(SPR_DMA_CTR, SPR_DMA_CTR__SUSPEND_MASK); /* * Wait for the engine to idle, then save regs. Note that we * want to record the "running" bit from before suspension, * and the "done" bit from after, so that we can properly * distinguish a case where the user suspended the engine from * the case where the kernel suspended as part of the context * swap. */ do { post_suspend_state = __insn_mfspr(SPR_DMA_USER_STATUS); } while (post_suspend_state & SPR_DMA_STATUS__BUSY_MASK); dma->src = __insn_mfspr(SPR_DMA_SRC_ADDR); dma->src_chunk = __insn_mfspr(SPR_DMA_SRC_CHUNK_ADDR); dma->dest = __insn_mfspr(SPR_DMA_DST_ADDR); dma->dest_chunk = __insn_mfspr(SPR_DMA_DST_CHUNK_ADDR); dma->strides = __insn_mfspr(SPR_DMA_STRIDE); dma->chunk_size = __insn_mfspr(SPR_DMA_CHUNK_SIZE); dma->byte = __insn_mfspr(SPR_DMA_BYTE); dma->status = (state & SPR_DMA_STATUS__RUNNING_MASK) | (post_suspend_state & SPR_DMA_STATUS__DONE_MASK); } /* Restart a DMA that was running before we were context-switched out. */ static void restore_tile_dma_state(struct thread_struct *t) { const struct tile_dma_state *dma = &t->tile_dma_state; /* * The only way to restore the done bit is to run a zero * length transaction. */ if ((dma->status & SPR_DMA_STATUS__DONE_MASK) && !(__insn_mfspr(SPR_DMA_USER_STATUS) & SPR_DMA_STATUS__DONE_MASK)) { __insn_mtspr(SPR_DMA_BYTE, 0); __insn_mtspr(SPR_DMA_CTR, SPR_DMA_CTR__REQUEST_MASK); while (__insn_mfspr(SPR_DMA_USER_STATUS) & SPR_DMA_STATUS__BUSY_MASK) ; } __insn_mtspr(SPR_DMA_SRC_ADDR, dma->src); __insn_mtspr(SPR_DMA_SRC_CHUNK_ADDR, dma->src_chunk); __insn_mtspr(SPR_DMA_DST_ADDR, dma->dest); __insn_mtspr(SPR_DMA_DST_CHUNK_ADDR, dma->dest_chunk); __insn_mtspr(SPR_DMA_STRIDE, dma->strides); __insn_mtspr(SPR_DMA_CHUNK_SIZE, dma->chunk_size); __insn_mtspr(SPR_DMA_BYTE, dma->byte); /* * Restart the engine if we were running and not done. * Clear a pending async DMA fault that we were waiting on return * to user space to execute, since we expect the DMA engine * to regenerate those faults for us now. Note that we don't * try to clear the TIF_ASYNC_TLB flag, since it's relatively * harmless if set, and it covers both DMA and the SN processor. */ if ((dma->status & DMA_STATUS_MASK) == SPR_DMA_STATUS__RUNNING_MASK) { t->dma_async_tlb.fault_num = 0; __insn_mtspr(SPR_DMA_CTR, SPR_DMA_CTR__REQUEST_MASK); } } #endif static void save_arch_state(struct thread_struct *t) { #if CHIP_HAS_SPLIT_INTR_MASK() t->interrupt_mask = __insn_mfspr(SPR_INTERRUPT_MASK_0_0) | ((u64)__insn_mfspr(SPR_INTERRUPT_MASK_0_1) << 32); #else t->interrupt_mask = __insn_mfspr(SPR_INTERRUPT_MASK_0); #endif t->ex_context[0] = __insn_mfspr(SPR_EX_CONTEXT_0_0); t->ex_context[1] = __insn_mfspr(SPR_EX_CONTEXT_0_1); t->system_save[0] = __insn_mfspr(SPR_SYSTEM_SAVE_0_0); t->system_save[1] = __insn_mfspr(SPR_SYSTEM_SAVE_0_1); t->system_save[2] = __insn_mfspr(SPR_SYSTEM_SAVE_0_2); t->system_save[3] = __insn_mfspr(SPR_SYSTEM_SAVE_0_3); t->intctrl_0 = __insn_mfspr(SPR_INTCTRL_0_STATUS); #if CHIP_HAS_PROC_STATUS_SPR() t->proc_status = __insn_mfspr(SPR_PROC_STATUS); #endif #if !CHIP_HAS_FIXED_INTVEC_BASE() t->interrupt_vector_base = __insn_mfspr(SPR_INTERRUPT_VECTOR_BASE_0); #endif #if CHIP_HAS_TILE_RTF_HWM() t->tile_rtf_hwm = __insn_mfspr(SPR_TILE_RTF_HWM); #endif #if CHIP_HAS_DSTREAM_PF() t->dstream_pf = __insn_mfspr(SPR_DSTREAM_PF); #endif } static void restore_arch_state(const struct thread_struct *t) { #if CHIP_HAS_SPLIT_INTR_MASK() __insn_mtspr(SPR_INTERRUPT_MASK_0_0, (u32) t->interrupt_mask); __insn_mtspr(SPR_INTERRUPT_MASK_0_1, t->interrupt_mask >> 32); #else __insn_mtspr(SPR_INTERRUPT_MASK_0, t->interrupt_mask); #endif __insn_mtspr(SPR_EX_CONTEXT_0_0, t->ex_context[0]); __insn_mtspr(SPR_EX_CONTEXT_0_1, t->ex_context[1]); __insn_mtspr(SPR_SYSTEM_SAVE_0_0, t->system_save[0]); __insn_mtspr(SPR_SYSTEM_SAVE_0_1, t->system_save[1]); __insn_mtspr(SPR_SYSTEM_SAVE_0_2, t->system_save[2]); __insn_mtspr(SPR_SYSTEM_SAVE_0_3, t->system_save[3]); __insn_mtspr(SPR_INTCTRL_0_STATUS, t->intctrl_0); #if CHIP_HAS_PROC_STATUS_SPR() __insn_mtspr(SPR_PROC_STATUS, t->proc_status); #endif #if !CHIP_HAS_FIXED_INTVEC_BASE() __insn_mtspr(SPR_INTERRUPT_VECTOR_BASE_0, t->interrupt_vector_base); #endif #if CHIP_HAS_TILE_RTF_HWM() __insn_mtspr(SPR_TILE_RTF_HWM, t->tile_rtf_hwm); #endif #if CHIP_HAS_DSTREAM_PF() __insn_mtspr(SPR_DSTREAM_PF, t->dstream_pf); #endif } void _prepare_arch_switch(struct task_struct *next) { #if CHIP_HAS_SN_PROC() int snctl; #endif #if CHIP_HAS_TILE_DMA() struct tile_dma_state *dma = ¤t->thread.tile_dma_state; if (dma->enabled) save_tile_dma_state(dma); #endif #if CHIP_HAS_SN_PROC() /* * Suspend the static network processor if it was running. * We do not suspend the fabric itself, just like we don't * try to suspend the UDN. */ snctl = __insn_mfspr(SPR_SNCTL); current->thread.sn_proc_running = (snctl & SPR_SNCTL__FRZPROC_MASK) == 0; if (current->thread.sn_proc_running) __insn_mtspr(SPR_SNCTL, snctl | SPR_SNCTL__FRZPROC_MASK); #endif } struct task_struct *__sched _switch_to(struct task_struct *prev,