blob: 1d745b2868aeb18600b5ed190443d8023e94d1a4 [file] [log] [blame]
/*
* Copyright 2017, Data61
* Commonwealth Scientific and Industrial Research Organisation (CSIRO)
* ABN 41 687 119 230.
*
* This software may be distributed and modified according to the terms of
* the BSD 2-Clause license. Note that NO WARRANTY is provided.
* See "LICENSE_BSD2.txt" for details.
*
* @TAG(DATA61_BSD)
*/
/**
*
* Provides basic thread configuration/starting/cleanup functions.
*
* Any other operations (start, stop, resume) should use the seL4 API directly on
* sel4utils_thread_t->tcb.cptr.
*
*/
#pragma once
#include <autoconf.h>
#include <sel4utils/gen_config.h>
#include <sel4/sel4.h>
#include <stdbool.h>
#include <vka/vka.h>
#include <vspace/vspace.h>
#include <sel4utils/thread_config.h>
#ifdef CONFIG_DEBUG_BUILD
#define NAME_THREAD(_tcbcap, _name) seL4_DebugNameThread(_tcbcap, _name);
#else
#define NAME_THREAD(_tcbcap, _name)
#endif
typedef struct sel4utils_thread {
vka_object_t tcb;
vka_object_t sched_context;
void *stack_top;
void *initial_stack_pointer;
size_t stack_size;
seL4_CPtr ipc_buffer;
seL4_Word ipc_buffer_addr;
bool own_sc;
bool own_reply;
vka_object_t reply;
} sel4utils_thread_t;
typedef struct sel4utils_checkpoint {
/* checkpointed stack */
void *stack;
seL4_UserContext regs;
sel4utils_thread_t *thread;
/* stack pointer this checkpoint preserves */
uintptr_t sp;
} sel4utils_checkpoint_t;
typedef void (*sel4utils_thread_entry_fn)(void *arg0, void *arg1, void *ipc_buf);
/**
* Configure a thread, allocating any resources required. The thread will start at priority 0.
*
* If CONFIG_RT is enabled, the thread will not have a scheduling context, so it will not be able to run.
*
* @param vka initialised vka to allocate objects with
* @param parent vspace structure of the thread calling this function, used for temporary mappings
* @param alloc initialised vspace structure to allocate virtual memory with
* @param fault_endpoint endpoint to set as the threads fault endpoint. Can be 0.
* @param cspace the root of the cspace to start the thread in
* @param cspace_root_data data for cspace access
* @param res an uninitialised sel4utils_thread_t data structure that will be initialised
* after this operation.
*
* @return 0 on success, -1 on failure. Use CONFIG_DEBUG to see error messages.
*/
int sel4utils_configure_thread(vka_t *vka, vspace_t *parent, vspace_t *alloc, seL4_CPtr fault_endpoint,
seL4_CNode cspace, seL4_Word cspace_root_data,
sel4utils_thread_t *res);
/**
* As per sel4utils_configure_thread, but using a config struct.
*/
int sel4utils_configure_thread_config(vka_t *vka, vspace_t *parent, vspace_t *alloc,
sel4utils_thread_config_t config, sel4utils_thread_t *res);
/**
* Start a thread, allocating any resources required.
* The third argument to the thread (in r2 for arm, on stack for ia32) will be the
* address of the ipc buffer.
*
* @param thread thread data structure that has been initialised with sel4utils_configure_thread
* @param entry_point the address that the thread will start at
*
* NOTE: In order for the on-stack argument passing to work for ia32,
* entry points must be functions.
*
* ie. jumping to this start symbol will work:
*
* void _start(int argc, char **argv) {
* int ret = main(argc, argv);
* exit(ret);
* }
*
*
* However, jumping to a start symbol like this:
*
* _start:
* call main
*
* will NOT work, as call pushes an extra value (the return value)
* onto the stack. If you really require an assembler stub, it should
* decrement the stack value to account for this.
*
* ie.
*
* _start:
* popl %eax
* call main
*
* This does not apply for arm, as arguments are passed in registers.
*
*
* @param arg0 a pointer to the arguments for this thread. User decides the protocol.
* @param arg1 another pointer. User decides the protocol. Note that there are two args here
* to easily support C standard: int main(int argc, char **argv).
* @param resume 1 to start the thread immediately, 0 otherwise.
*
* @return 0 on success, -1 on failure.
*/
int sel4utils_start_thread(sel4utils_thread_t *thread, sel4utils_thread_entry_fn entry_point,
void *arg0, void *arg1, int resume);
/**
* Release any resources used by this thread. The thread data structure will not be usable
* until sel4utils_thread_configure is called again.
*
* @param vka the vka interface that this thread was initialised with
* @param alloc the allocation interface that this thread was initialised with
* @param thread the thread structure that was returned when the thread started
*/
void sel4utils_clean_up_thread(vka_t *vka, vspace_t *alloc, sel4utils_thread_t *thread);
/**
* Checkpoint a thread at its current state, storing its current register set and stack.
*
* Note that the heap state is not saved, so threads intending to use this functionality
* should not mutate the heap or other state beyond the checkpoint, unless extra functionality
* is included to roll these back.
*
* This should not be called on a currently running thread, and is designed to be called on
* threads which are known to be blocked on an seL4_Recv, for checkpointing passive threads on
* the mcs kernel (threads without scheduling contexts). The checkpoint is set up that such a
* thread can be restarted successfully at the instruction which enters the kernel, with
* register state set up specifically for that.
*
* @param thread the thread to checkpoint
* @param checkpoint pointer to uninitialised checkpoint struct
* @param suspend true if the thread should be suspended
*
* @return 0 on success.
*/
int sel4utils_checkpoint_thread(sel4utils_thread_t *thread, sel4utils_checkpoint_t *checkpoint, bool suspend);
/**
* Rollback a thread to a previous checkpoint, restoring its register set and stack.
*
* This is not atomic and callers should make sure the target thread is stopped or that the
* caller is higher priority such that the target is not switched to by the kernel mid-restore.
*
* @param checkpoint the previously saved checkpoint to restore.
* @param free true if this checkpoint should free all memory allocated, i.e if the checkpoint
* will not be used again.
* @param resume true if the thread should be resumed immediately.
*
* @return 0 on success.
*/
int sel4utils_checkpoint_restore(sel4utils_checkpoint_t *checkpoint, bool free, bool resume);
/**
* Clean up a previously allocated checkpoint.
*/
void sel4utils_free_checkpoint(sel4utils_checkpoint_t *checkpoint);
/**
* Start a fault handling thread that will print the name of the thread that faulted
* as well as debugging information. The thread will start at priority 0.
*
* If CONFIG_RT it be passive (not have a scheulding context) and will run on the SC of the faulter.
*
* @param fault_endpoint the fault_endpoint to wait on
* @param vka allocator
* @param vspace vspace (this library must be mapped into that vspace).
* @param cspace the cspace that the fault_endpoint is in
* @param data the cspace_data for that cspace (with correct guard)
* @param name the name of the thread to print if it faults
* @param thread the thread data structure to populate
*
* @return 0 on success.
*/
int sel4utils_start_fault_handler(seL4_CPtr fault_endpoint, vka_t *vka, vspace_t *vspace,
seL4_CPtr cspace, seL4_Word data, char *name, sel4utils_thread_t *res);
/**
* Pretty print a fault message.
*
* @param tag the message info tag delivered by the fault.
* @param name thread name
*/
void sel4utils_print_fault_message(seL4_MessageInfo_t tag, const char *name);
/* Set the affinity of a thread, which will cause migration if the thread
* is running on a different core.
*
* On master, only use sched_params_set_core to set the core then call this function.
*
* If CONFIG_KERNEL_MCS is set, the sched params must be fully populated or the scheduling
* context will be empty when it changes core as scheduling parameters of scheduling
* contexts are not maintained across migrations.
*/
int sel4utils_set_sched_affinity(sel4utils_thread_t *thread, sched_params_t params);
static inline seL4_TCB sel4utils_get_tcb(sel4utils_thread_t *thread)
{
return thread->tcb.cptr;
}
static inline int sel4utils_suspend_thread(sel4utils_thread_t *thread)
{
return seL4_TCB_Suspend(thread->tcb.cptr);
}