![Chris Metcalf](/assets/img/avatar_default.png)
This change is a prerequisite change for TASK_ISOLATION but also
stands on its own for readability and maintainability. The existing
tile do_work_pending() was called in a loop from assembly on
the slow path; this change moves the loop into C code as well.
For the x86 version see commit c5c46f59e4
("x86/entry: Add new,
comprehensible entry and exit handlers written in C").
This change exposes a pre-existing bug on the older tilepro platform;
the singlestep processing is done last, but on tilepro (unlike tilegx)
we enable interrupts while doing that processing, so we could in
theory miss a signal or other asynchronous event. A future change
could fix this by breaking the singlestep work into a "prepare"
step done in the main loop, and a "trigger" step done after exiting
the loop. Since this change is intended as purely a restructuring
change, we call out the bug explicitly now, but don't yet fix it.
Signed-off-by: Chris Metcalf <cmetcalf@ezchip.com>
378 lines
11 KiB
C
378 lines
11 KiB
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.
|
|
*/
|
|
|
|
#ifndef _ASM_TILE_PROCESSOR_H
|
|
#define _ASM_TILE_PROCESSOR_H
|
|
|
|
#include <arch/chip.h>
|
|
|
|
#ifndef __ASSEMBLY__
|
|
|
|
/*
|
|
* NOTE: we don't include <linux/ptrace.h> or <linux/percpu.h> as one
|
|
* normally would, due to #include dependencies.
|
|
*/
|
|
#include <linux/types.h>
|
|
#include <asm/ptrace.h>
|
|
#include <asm/percpu.h>
|
|
|
|
#include <arch/spr_def.h>
|
|
|
|
struct task_struct;
|
|
struct thread_struct;
|
|
|
|
typedef struct {
|
|
unsigned long seg;
|
|
} mm_segment_t;
|
|
|
|
/*
|
|
* Default implementation of macro that returns current
|
|
* instruction pointer ("program counter").
|
|
*/
|
|
void *current_text_addr(void);
|
|
|
|
#if CHIP_HAS_TILE_DMA()
|
|
/* Capture the state of a suspended DMA. */
|
|
struct tile_dma_state {
|
|
int enabled;
|
|
unsigned long src;
|
|
unsigned long dest;
|
|
unsigned long strides;
|
|
unsigned long chunk_size;
|
|
unsigned long src_chunk;
|
|
unsigned long dest_chunk;
|
|
unsigned long byte;
|
|
unsigned long status;
|
|
};
|
|
|
|
/*
|
|
* A mask of the DMA status register for selecting only the 'running'
|
|
* and 'done' bits.
|
|
*/
|
|
#define DMA_STATUS_MASK \
|
|
(SPR_DMA_STATUS__RUNNING_MASK | SPR_DMA_STATUS__DONE_MASK)
|
|
#endif
|
|
|
|
/*
|
|
* Track asynchronous TLB events (faults and access violations)
|
|
* that occur while we are in kernel mode from DMA or the SN processor.
|
|
*/
|
|
struct async_tlb {
|
|
short fault_num; /* original fault number; 0 if none */
|
|
char is_fault; /* was it a fault (vs an access violation) */
|
|
char is_write; /* for fault: was it caused by a write? */
|
|
unsigned long address; /* what address faulted? */
|
|
};
|
|
|
|
#ifdef CONFIG_HARDWALL
|
|
struct hardwall_info;
|
|
struct hardwall_task {
|
|
/* Which hardwall is this task tied to? (or NULL if none) */
|
|
struct hardwall_info *info;
|
|
/* Chains this task into the list at info->task_head. */
|
|
struct list_head list;
|
|
};
|
|
#ifdef __tilepro__
|
|
#define HARDWALL_TYPES 1 /* udn */
|
|
#else
|
|
#define HARDWALL_TYPES 3 /* udn, idn, and ipi */
|
|
#endif
|
|
#endif
|
|
|
|
struct thread_struct {
|
|
/* kernel stack pointer */
|
|
unsigned long ksp;
|
|
/* kernel PC */
|
|
unsigned long pc;
|
|
/* starting user stack pointer (for page migration) */
|
|
unsigned long usp0;
|
|
/* pid of process that created this one */
|
|
pid_t creator_pid;
|
|
#if CHIP_HAS_TILE_DMA()
|
|
/* DMA info for suspended threads (byte == 0 means no DMA state) */
|
|
struct tile_dma_state tile_dma_state;
|
|
#endif
|
|
/* User EX_CONTEXT registers */
|
|
unsigned long ex_context[2];
|
|
/* User SYSTEM_SAVE registers */
|
|
unsigned long system_save[4];
|
|
/* User interrupt mask */
|
|
unsigned long long interrupt_mask;
|
|
/* User interrupt-control 0 state */
|
|
unsigned long intctrl_0;
|
|
/* Any other miscellaneous processor state bits */
|
|
unsigned long proc_status;
|
|
#if !CHIP_HAS_FIXED_INTVEC_BASE()
|
|
/* Interrupt base for PL0 interrupts */
|
|
unsigned long interrupt_vector_base;
|
|
#endif
|
|
/* Tile cache retry fifo high-water mark */
|
|
unsigned long tile_rtf_hwm;
|
|
#if CHIP_HAS_DSTREAM_PF()
|
|
/* Data stream prefetch control */
|
|
unsigned long dstream_pf;
|
|
#endif
|
|
#ifdef CONFIG_HARDWALL
|
|
/* Hardwall information for various resources. */
|
|
struct hardwall_task hardwall[HARDWALL_TYPES];
|
|
#endif
|
|
#if CHIP_HAS_TILE_DMA()
|
|
/* Async DMA TLB fault information */
|
|
struct async_tlb dma_async_tlb;
|
|
#endif
|
|
};
|
|
|
|
#endif /* !__ASSEMBLY__ */
|
|
|
|
/*
|
|
* Start with "sp" this many bytes below the top of the kernel stack.
|
|
* This allows us to be cache-aware when handling the initial save
|
|
* of the pt_regs value to the stack.
|
|
*/
|
|
#define STACK_TOP_DELTA 64
|
|
|
|
/*
|
|
* When entering the kernel via a fault, start with the top of the
|
|
* pt_regs structure this many bytes below the top of the page.
|
|
* This aligns the pt_regs structure optimally for cache-line access.
|
|
*/
|
|
#ifdef __tilegx__
|
|
#define KSTK_PTREGS_GAP 48
|
|
#else
|
|
#define KSTK_PTREGS_GAP 56
|
|
#endif
|
|
|
|
#ifndef __ASSEMBLY__
|
|
|
|
#ifdef __tilegx__
|
|
#define TASK_SIZE_MAX (_AC(1, UL) << (MAX_VA_WIDTH - 1))
|
|
#else
|
|
#define TASK_SIZE_MAX PAGE_OFFSET
|
|
#endif
|
|
|
|
/* TASK_SIZE and related variables are always checked in "current" context. */
|
|
#ifdef CONFIG_COMPAT
|
|
#define COMPAT_TASK_SIZE (1UL << 31)
|
|
#define TASK_SIZE ((current_thread_info()->status & TS_COMPAT) ?\
|
|
COMPAT_TASK_SIZE : TASK_SIZE_MAX)
|
|
#else
|
|
#define TASK_SIZE TASK_SIZE_MAX
|
|
#endif
|
|
|
|
#define VDSO_BASE ((unsigned long)current->active_mm->context.vdso_base)
|
|
#define VDSO_SYM(x) (VDSO_BASE + (unsigned long)(x))
|
|
|
|
#define STACK_TOP TASK_SIZE
|
|
|
|
/* STACK_TOP_MAX is used temporarily in execve and should not check COMPAT. */
|
|
#define STACK_TOP_MAX TASK_SIZE_MAX
|
|
|
|
/*
|
|
* This decides where the kernel will search for a free chunk of vm
|
|
* space during mmap's, if it is using bottom-up mapping.
|
|
*/
|
|
#define TASK_UNMAPPED_BASE (PAGE_ALIGN(TASK_SIZE / 3))
|
|
|
|
#define HAVE_ARCH_PICK_MMAP_LAYOUT
|
|
|
|
#define INIT_THREAD { \
|
|
.ksp = (unsigned long)init_stack + THREAD_SIZE - STACK_TOP_DELTA, \
|
|
.interrupt_mask = -1ULL \
|
|
}
|
|
|
|
/* Kernel stack top for the task that first boots on this cpu. */
|
|
DECLARE_PER_CPU(unsigned long, boot_sp);
|
|
|
|
/* PC to boot from on this cpu. */
|
|
DECLARE_PER_CPU(unsigned long, boot_pc);
|
|
|
|
/* Do necessary setup to start up a newly executed thread. */
|
|
static inline void start_thread(struct pt_regs *regs,
|
|
unsigned long pc, unsigned long usp)
|
|
{
|
|
regs->pc = pc;
|
|
regs->sp = usp;
|
|
single_step_execve();
|
|
}
|
|
|
|
/* Free all resources held by a thread. */
|
|
static inline void release_thread(struct task_struct *dead_task)
|
|
{
|
|
/* Nothing for now */
|
|
}
|
|
|
|
extern void prepare_exit_to_usermode(struct pt_regs *regs, u32 flags);
|
|
|
|
|
|
/*
|
|
* Return saved (kernel) PC of a blocked thread.
|
|
* Only used in a printk() in kernel/sched/core.c, so don't work too hard.
|
|
*/
|
|
#define thread_saved_pc(t) ((t)->thread.pc)
|
|
|
|
unsigned long get_wchan(struct task_struct *p);
|
|
|
|
/* Return initial ksp value for given task. */
|
|
#define task_ksp0(task) \
|
|
((unsigned long)(task)->stack + THREAD_SIZE - STACK_TOP_DELTA)
|
|
|
|
/* Return some info about the user process TASK. */
|
|
#define task_pt_regs(task) \
|
|
((struct pt_regs *)(task_ksp0(task) - KSTK_PTREGS_GAP) - 1)
|
|
#define current_pt_regs() \
|
|
((struct pt_regs *)((stack_pointer | (THREAD_SIZE - 1)) - \
|
|
STACK_TOP_DELTA - (KSTK_PTREGS_GAP - 1)) - 1)
|
|
#define task_sp(task) (task_pt_regs(task)->sp)
|
|
#define task_pc(task) (task_pt_regs(task)->pc)
|
|
/* Aliases for pc and sp (used in fs/proc/array.c) */
|
|
#define KSTK_EIP(task) task_pc(task)
|
|
#define KSTK_ESP(task) task_sp(task)
|
|
|
|
/* Fine-grained unaligned JIT support */
|
|
#define GET_UNALIGN_CTL(tsk, adr) get_unalign_ctl((tsk), (adr))
|
|
#define SET_UNALIGN_CTL(tsk, val) set_unalign_ctl((tsk), (val))
|
|
|
|
extern int get_unalign_ctl(struct task_struct *tsk, unsigned long adr);
|
|
extern int set_unalign_ctl(struct task_struct *tsk, unsigned int val);
|
|
|
|
/* Standard format for printing registers and other word-size data. */
|
|
#ifdef __tilegx__
|
|
# define REGFMT "0x%016lx"
|
|
#else
|
|
# define REGFMT "0x%08lx"
|
|
#endif
|
|
|
|
/*
|
|
* Do some slow action (e.g. read a slow SPR).
|
|
* Note that this must also have compiler-barrier semantics since
|
|
* it may be used in a busy loop reading memory.
|
|
*/
|
|
static inline void cpu_relax(void)
|
|
{
|
|
__insn_mfspr(SPR_PASS);
|
|
barrier();
|
|
}
|
|
|
|
#define cpu_relax_lowlatency() cpu_relax()
|
|
|
|
/* Info on this processor (see fs/proc/cpuinfo.c) */
|
|
struct seq_operations;
|
|
extern const struct seq_operations cpuinfo_op;
|
|
|
|
/* Provide information about the chip model. */
|
|
extern char chip_model[64];
|
|
|
|
/* Data on which physical memory controller corresponds to which NUMA node. */
|
|
extern int node_controller[];
|
|
|
|
/* Does the heap allocator return hash-for-home pages by default? */
|
|
extern int hash_default;
|
|
|
|
/* Should kernel stack pages be hash-for-home? */
|
|
extern int kstack_hash;
|
|
|
|
/* Does MAP_ANONYMOUS return hash-for-home pages by default? */
|
|
#define uheap_hash hash_default
|
|
|
|
|
|
/* Are we using huge pages in the TLB for kernel data? */
|
|
extern int kdata_huge;
|
|
|
|
/* Support standard Linux prefetching. */
|
|
#define ARCH_HAS_PREFETCH
|
|
#define prefetch(x) __builtin_prefetch(x)
|
|
#define PREFETCH_STRIDE CHIP_L2_LINE_SIZE()
|
|
|
|
/* Bring a value into the L1D, faulting the TLB if necessary. */
|
|
#ifdef __tilegx__
|
|
#define prefetch_L1(x) __insn_prefetch_l1_fault((void *)(x))
|
|
#else
|
|
#define prefetch_L1(x) __insn_prefetch_L1((void *)(x))
|
|
#endif
|
|
|
|
#else /* __ASSEMBLY__ */
|
|
|
|
/* Do some slow action (e.g. read a slow SPR). */
|
|
#define CPU_RELAX mfspr zero, SPR_PASS
|
|
|
|
#endif /* !__ASSEMBLY__ */
|
|
|
|
/* Assembly code assumes that the PL is in the low bits. */
|
|
#if SPR_EX_CONTEXT_1_1__PL_SHIFT != 0
|
|
# error Fix assembly assumptions about PL
|
|
#endif
|
|
|
|
/* We sometimes use these macros for EX_CONTEXT_0_1 as well. */
|
|
#if SPR_EX_CONTEXT_1_1__PL_SHIFT != SPR_EX_CONTEXT_0_1__PL_SHIFT || \
|
|
SPR_EX_CONTEXT_1_1__PL_RMASK != SPR_EX_CONTEXT_0_1__PL_RMASK || \
|
|
SPR_EX_CONTEXT_1_1__ICS_SHIFT != SPR_EX_CONTEXT_0_1__ICS_SHIFT || \
|
|
SPR_EX_CONTEXT_1_1__ICS_RMASK != SPR_EX_CONTEXT_0_1__ICS_RMASK
|
|
# error Fix assumptions that EX1 macros work for both PL0 and PL1
|
|
#endif
|
|
|
|
/* Allow pulling apart and recombining the PL and ICS bits in EX_CONTEXT. */
|
|
#define EX1_PL(ex1) \
|
|
(((ex1) >> SPR_EX_CONTEXT_1_1__PL_SHIFT) & SPR_EX_CONTEXT_1_1__PL_RMASK)
|
|
#define EX1_ICS(ex1) \
|
|
(((ex1) >> SPR_EX_CONTEXT_1_1__ICS_SHIFT) & SPR_EX_CONTEXT_1_1__ICS_RMASK)
|
|
#define PL_ICS_EX1(pl, ics) \
|
|
(((pl) << SPR_EX_CONTEXT_1_1__PL_SHIFT) | \
|
|
((ics) << SPR_EX_CONTEXT_1_1__ICS_SHIFT))
|
|
|
|
/*
|
|
* Provide symbolic constants for PLs.
|
|
*/
|
|
#define USER_PL 0
|
|
#if CONFIG_KERNEL_PL == 2
|
|
#define GUEST_PL 1
|
|
#endif
|
|
#define KERNEL_PL CONFIG_KERNEL_PL
|
|
|
|
/* SYSTEM_SAVE_K_0 holds the current cpu number ORed with ksp0. */
|
|
#ifdef __tilegx__
|
|
#define CPU_SHIFT 48
|
|
#if CHIP_VA_WIDTH() > CPU_SHIFT
|
|
# error Too many VA bits!
|
|
#endif
|
|
#define MAX_CPU_ID ((1 << (64 - CPU_SHIFT)) - 1)
|
|
#define raw_smp_processor_id() \
|
|
((int)(__insn_mfspr(SPR_SYSTEM_SAVE_K_0) >> CPU_SHIFT))
|
|
#define get_current_ksp0() \
|
|
((unsigned long)(((long)__insn_mfspr(SPR_SYSTEM_SAVE_K_0) << \
|
|
(64 - CPU_SHIFT)) >> (64 - CPU_SHIFT)))
|
|
#define next_current_ksp0(task) ({ \
|
|
unsigned long __ksp0 = task_ksp0(task) & ((1UL << CPU_SHIFT) - 1); \
|
|
unsigned long __cpu = (long)raw_smp_processor_id() << CPU_SHIFT; \
|
|
__ksp0 | __cpu; \
|
|
})
|
|
#else
|
|
#define LOG2_NR_CPU_IDS 6
|
|
#define MAX_CPU_ID ((1 << LOG2_NR_CPU_IDS) - 1)
|
|
#define raw_smp_processor_id() \
|
|
((int)__insn_mfspr(SPR_SYSTEM_SAVE_K_0) & MAX_CPU_ID)
|
|
#define get_current_ksp0() \
|
|
(__insn_mfspr(SPR_SYSTEM_SAVE_K_0) & ~MAX_CPU_ID)
|
|
#define next_current_ksp0(task) ({ \
|
|
unsigned long __ksp0 = task_ksp0(task); \
|
|
int __cpu = raw_smp_processor_id(); \
|
|
BUG_ON(__ksp0 & MAX_CPU_ID); \
|
|
__ksp0 | __cpu; \
|
|
})
|
|
#endif
|
|
#if CONFIG_NR_CPUS > (MAX_CPU_ID + 1)
|
|
# error Too many cpus!
|
|
#endif
|
|
|
|
#endif /* _ASM_TILE_PROCESSOR_H */
|