linux/drivers/acpi/acpica/dbxface.c

484 lines
12 KiB
C
Raw Normal View History

// SPDX-License-Identifier: BSD-3-Clause OR GPL-2.0
/*******************************************************************************
*
* Module Name: dbxface - AML Debugger external interfaces
*
******************************************************************************/
#include <acpi/acpi.h>
#include "accommon.h"
#include "amlcode.h"
#include "acdebug.h"
#include "acinterp.h"
#define _COMPONENT ACPI_CA_DEBUGGER
ACPI_MODULE_NAME("dbxface")
/* Local prototypes */
static acpi_status
acpi_db_start_command(struct acpi_walk_state *walk_state,
union acpi_parse_object *op);
#ifdef ACPI_OBSOLETE_FUNCTIONS
void acpi_db_method_end(struct acpi_walk_state *walk_state);
#endif
/*******************************************************************************
*
* FUNCTION: acpi_db_start_command
*
* PARAMETERS: walk_state - Current walk
* op - Current executing Op, from AML interpreter
*
* RETURN: Status
*
* DESCRIPTION: Enter debugger command loop
*
******************************************************************************/
static acpi_status
acpi_db_start_command(struct acpi_walk_state *walk_state,
union acpi_parse_object *op)
{
acpi_status status;
/* TBD: [Investigate] are there namespace locking issues here? */
/* acpi_ut_release_mutex (ACPI_MTX_NAMESPACE); */
/* Go into the command loop and await next user command */
acpi_gbl_method_executing = TRUE;
status = AE_CTRL_TRUE;
ACPICA: Debugger: Convert some mechanisms to OSPM specific The following mechanisms are OSPM specific: 1. Redirect output destination to console: no file redirection will be needed by an in-kernel debugger, there is even no file can be accessed when the debugger is running in the kernel mode. 2. Output command prompts: programs other than acpiexec can have different prompt characters and the prompt characters may be implemented as a special character sequence to form a char device IO protocol. 3. Command ready/complete handshake: OSPM debugger may wait more conditions to implement OSPM specific semantics (for example, FIFO full/empty conditions for O_NONBLOCK or IO open/close conditions). Leaving such OSPM specific stuffs in the ACPICA debugger core blocks Linux debugger IO driver implementation. Several new OSL APIs are provided by this patch: 1. acpi_os_initialize_command_signals: initialize command handshake mechanism or any other OSPM specific stuffs. 2. acpi_os_terminate_command_signals: reversal of acpi_os_initialize_command_signals. 3. acpi_os_wait_command_ready: putting debugger task into wait state when a command is not ready. OSPMs can terminate command loop by returning AE_CTRL_TERMINATE from this API. Normally, wait_event() or wait_for_multiple_object() may be used to implement this API. 4. acpi_os_notify_command_complete: putting user task into running state when a command has been completed. OSPMs can terminate command loop by returning AE_CTRL_TERMINATE from this API. Normally, wake_up() or set_event() may be used to implement this API. This patch also converts current command signaling implementation into a generic debugger layer (osgendbg.c) to be used by the existing OSPMs or acpiexec, in return, Linux can have chance to implement its own command handshake mechanism. This patch also implements acpiexec batch mode in a multi-threading mode comaptible style as a demo (this can be confirmed by configuring acpiexec into DEBUGGER_MULTI_THREADED mode where the batch mode is still working). Lv Zheng. Note that the OSPM specific command handshake mechanism is required by Linux kernel because: 1. Linux kernel trends to use wait queue to synchronize two threads, using mutexes to achieve that will cause false "dead lock" warnings. 2. The command handshake mechanism implemented by ACPICA is implemented in this way because of a design issue in debugger IO streaming. Debugger IO outputs are simply cached using a giant buffer, this should be tuned by Linux in the future. Signed-off-by: Lv Zheng <lv.zheng@intel.com> Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
2015-12-03 02:42:46 +00:00
while (status == AE_CTRL_TRUE) {
ACPICA: Debugger: Convert some mechanisms to OSPM specific The following mechanisms are OSPM specific: 1. Redirect output destination to console: no file redirection will be needed by an in-kernel debugger, there is even no file can be accessed when the debugger is running in the kernel mode. 2. Output command prompts: programs other than acpiexec can have different prompt characters and the prompt characters may be implemented as a special character sequence to form a char device IO protocol. 3. Command ready/complete handshake: OSPM debugger may wait more conditions to implement OSPM specific semantics (for example, FIFO full/empty conditions for O_NONBLOCK or IO open/close conditions). Leaving such OSPM specific stuffs in the ACPICA debugger core blocks Linux debugger IO driver implementation. Several new OSL APIs are provided by this patch: 1. acpi_os_initialize_command_signals: initialize command handshake mechanism or any other OSPM specific stuffs. 2. acpi_os_terminate_command_signals: reversal of acpi_os_initialize_command_signals. 3. acpi_os_wait_command_ready: putting debugger task into wait state when a command is not ready. OSPMs can terminate command loop by returning AE_CTRL_TERMINATE from this API. Normally, wait_event() or wait_for_multiple_object() may be used to implement this API. 4. acpi_os_notify_command_complete: putting user task into running state when a command has been completed. OSPMs can terminate command loop by returning AE_CTRL_TERMINATE from this API. Normally, wake_up() or set_event() may be used to implement this API. This patch also converts current command signaling implementation into a generic debugger layer (osgendbg.c) to be used by the existing OSPMs or acpiexec, in return, Linux can have chance to implement its own command handshake mechanism. This patch also implements acpiexec batch mode in a multi-threading mode comaptible style as a demo (this can be confirmed by configuring acpiexec into DEBUGGER_MULTI_THREADED mode where the batch mode is still working). Lv Zheng. Note that the OSPM specific command handshake mechanism is required by Linux kernel because: 1. Linux kernel trends to use wait queue to synchronize two threads, using mutexes to achieve that will cause false "dead lock" warnings. 2. The command handshake mechanism implemented by ACPICA is implemented in this way because of a design issue in debugger IO streaming. Debugger IO outputs are simply cached using a giant buffer, this should be tuned by Linux in the future. Signed-off-by: Lv Zheng <lv.zheng@intel.com> Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
2015-12-03 02:42:46 +00:00
/* Notify the completion of the command */
ACPICA: Debugger: Convert some mechanisms to OSPM specific The following mechanisms are OSPM specific: 1. Redirect output destination to console: no file redirection will be needed by an in-kernel debugger, there is even no file can be accessed when the debugger is running in the kernel mode. 2. Output command prompts: programs other than acpiexec can have different prompt characters and the prompt characters may be implemented as a special character sequence to form a char device IO protocol. 3. Command ready/complete handshake: OSPM debugger may wait more conditions to implement OSPM specific semantics (for example, FIFO full/empty conditions for O_NONBLOCK or IO open/close conditions). Leaving such OSPM specific stuffs in the ACPICA debugger core blocks Linux debugger IO driver implementation. Several new OSL APIs are provided by this patch: 1. acpi_os_initialize_command_signals: initialize command handshake mechanism or any other OSPM specific stuffs. 2. acpi_os_terminate_command_signals: reversal of acpi_os_initialize_command_signals. 3. acpi_os_wait_command_ready: putting debugger task into wait state when a command is not ready. OSPMs can terminate command loop by returning AE_CTRL_TERMINATE from this API. Normally, wait_event() or wait_for_multiple_object() may be used to implement this API. 4. acpi_os_notify_command_complete: putting user task into running state when a command has been completed. OSPMs can terminate command loop by returning AE_CTRL_TERMINATE from this API. Normally, wake_up() or set_event() may be used to implement this API. This patch also converts current command signaling implementation into a generic debugger layer (osgendbg.c) to be used by the existing OSPMs or acpiexec, in return, Linux can have chance to implement its own command handshake mechanism. This patch also implements acpiexec batch mode in a multi-threading mode comaptible style as a demo (this can be confirmed by configuring acpiexec into DEBUGGER_MULTI_THREADED mode where the batch mode is still working). Lv Zheng. Note that the OSPM specific command handshake mechanism is required by Linux kernel because: 1. Linux kernel trends to use wait queue to synchronize two threads, using mutexes to achieve that will cause false "dead lock" warnings. 2. The command handshake mechanism implemented by ACPICA is implemented in this way because of a design issue in debugger IO streaming. Debugger IO outputs are simply cached using a giant buffer, this should be tuned by Linux in the future. Signed-off-by: Lv Zheng <lv.zheng@intel.com> Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
2015-12-03 02:42:46 +00:00
status = acpi_os_notify_command_complete();
if (ACPI_FAILURE(status)) {
goto error_exit;
}
ACPICA: Debugger: Convert some mechanisms to OSPM specific The following mechanisms are OSPM specific: 1. Redirect output destination to console: no file redirection will be needed by an in-kernel debugger, there is even no file can be accessed when the debugger is running in the kernel mode. 2. Output command prompts: programs other than acpiexec can have different prompt characters and the prompt characters may be implemented as a special character sequence to form a char device IO protocol. 3. Command ready/complete handshake: OSPM debugger may wait more conditions to implement OSPM specific semantics (for example, FIFO full/empty conditions for O_NONBLOCK or IO open/close conditions). Leaving such OSPM specific stuffs in the ACPICA debugger core blocks Linux debugger IO driver implementation. Several new OSL APIs are provided by this patch: 1. acpi_os_initialize_command_signals: initialize command handshake mechanism or any other OSPM specific stuffs. 2. acpi_os_terminate_command_signals: reversal of acpi_os_initialize_command_signals. 3. acpi_os_wait_command_ready: putting debugger task into wait state when a command is not ready. OSPMs can terminate command loop by returning AE_CTRL_TERMINATE from this API. Normally, wait_event() or wait_for_multiple_object() may be used to implement this API. 4. acpi_os_notify_command_complete: putting user task into running state when a command has been completed. OSPMs can terminate command loop by returning AE_CTRL_TERMINATE from this API. Normally, wake_up() or set_event() may be used to implement this API. This patch also converts current command signaling implementation into a generic debugger layer (osgendbg.c) to be used by the existing OSPMs or acpiexec, in return, Linux can have chance to implement its own command handshake mechanism. This patch also implements acpiexec batch mode in a multi-threading mode comaptible style as a demo (this can be confirmed by configuring acpiexec into DEBUGGER_MULTI_THREADED mode where the batch mode is still working). Lv Zheng. Note that the OSPM specific command handshake mechanism is required by Linux kernel because: 1. Linux kernel trends to use wait queue to synchronize two threads, using mutexes to achieve that will cause false "dead lock" warnings. 2. The command handshake mechanism implemented by ACPICA is implemented in this way because of a design issue in debugger IO streaming. Debugger IO outputs are simply cached using a giant buffer, this should be tuned by Linux in the future. Signed-off-by: Lv Zheng <lv.zheng@intel.com> Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
2015-12-03 02:42:46 +00:00
/* Wait the readiness of the command */
ACPICA: Debugger: Convert some mechanisms to OSPM specific The following mechanisms are OSPM specific: 1. Redirect output destination to console: no file redirection will be needed by an in-kernel debugger, there is even no file can be accessed when the debugger is running in the kernel mode. 2. Output command prompts: programs other than acpiexec can have different prompt characters and the prompt characters may be implemented as a special character sequence to form a char device IO protocol. 3. Command ready/complete handshake: OSPM debugger may wait more conditions to implement OSPM specific semantics (for example, FIFO full/empty conditions for O_NONBLOCK or IO open/close conditions). Leaving such OSPM specific stuffs in the ACPICA debugger core blocks Linux debugger IO driver implementation. Several new OSL APIs are provided by this patch: 1. acpi_os_initialize_command_signals: initialize command handshake mechanism or any other OSPM specific stuffs. 2. acpi_os_terminate_command_signals: reversal of acpi_os_initialize_command_signals. 3. acpi_os_wait_command_ready: putting debugger task into wait state when a command is not ready. OSPMs can terminate command loop by returning AE_CTRL_TERMINATE from this API. Normally, wait_event() or wait_for_multiple_object() may be used to implement this API. 4. acpi_os_notify_command_complete: putting user task into running state when a command has been completed. OSPMs can terminate command loop by returning AE_CTRL_TERMINATE from this API. Normally, wake_up() or set_event() may be used to implement this API. This patch also converts current command signaling implementation into a generic debugger layer (osgendbg.c) to be used by the existing OSPMs or acpiexec, in return, Linux can have chance to implement its own command handshake mechanism. This patch also implements acpiexec batch mode in a multi-threading mode comaptible style as a demo (this can be confirmed by configuring acpiexec into DEBUGGER_MULTI_THREADED mode where the batch mode is still working). Lv Zheng. Note that the OSPM specific command handshake mechanism is required by Linux kernel because: 1. Linux kernel trends to use wait queue to synchronize two threads, using mutexes to achieve that will cause false "dead lock" warnings. 2. The command handshake mechanism implemented by ACPICA is implemented in this way because of a design issue in debugger IO streaming. Debugger IO outputs are simply cached using a giant buffer, this should be tuned by Linux in the future. Signed-off-by: Lv Zheng <lv.zheng@intel.com> Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
2015-12-03 02:42:46 +00:00
status = acpi_os_wait_command_ready();
if (ACPI_FAILURE(status)) {
goto error_exit;
}
status =
acpi_db_command_dispatch(acpi_gbl_db_line_buf, walk_state,
op);
}
/* acpi_ut_acquire_mutex (ACPI_MTX_NAMESPACE); */
ACPICA: Debugger: Convert some mechanisms to OSPM specific The following mechanisms are OSPM specific: 1. Redirect output destination to console: no file redirection will be needed by an in-kernel debugger, there is even no file can be accessed when the debugger is running in the kernel mode. 2. Output command prompts: programs other than acpiexec can have different prompt characters and the prompt characters may be implemented as a special character sequence to form a char device IO protocol. 3. Command ready/complete handshake: OSPM debugger may wait more conditions to implement OSPM specific semantics (for example, FIFO full/empty conditions for O_NONBLOCK or IO open/close conditions). Leaving such OSPM specific stuffs in the ACPICA debugger core blocks Linux debugger IO driver implementation. Several new OSL APIs are provided by this patch: 1. acpi_os_initialize_command_signals: initialize command handshake mechanism or any other OSPM specific stuffs. 2. acpi_os_terminate_command_signals: reversal of acpi_os_initialize_command_signals. 3. acpi_os_wait_command_ready: putting debugger task into wait state when a command is not ready. OSPMs can terminate command loop by returning AE_CTRL_TERMINATE from this API. Normally, wait_event() or wait_for_multiple_object() may be used to implement this API. 4. acpi_os_notify_command_complete: putting user task into running state when a command has been completed. OSPMs can terminate command loop by returning AE_CTRL_TERMINATE from this API. Normally, wake_up() or set_event() may be used to implement this API. This patch also converts current command signaling implementation into a generic debugger layer (osgendbg.c) to be used by the existing OSPMs or acpiexec, in return, Linux can have chance to implement its own command handshake mechanism. This patch also implements acpiexec batch mode in a multi-threading mode comaptible style as a demo (this can be confirmed by configuring acpiexec into DEBUGGER_MULTI_THREADED mode where the batch mode is still working). Lv Zheng. Note that the OSPM specific command handshake mechanism is required by Linux kernel because: 1. Linux kernel trends to use wait queue to synchronize two threads, using mutexes to achieve that will cause false "dead lock" warnings. 2. The command handshake mechanism implemented by ACPICA is implemented in this way because of a design issue in debugger IO streaming. Debugger IO outputs are simply cached using a giant buffer, this should be tuned by Linux in the future. Signed-off-by: Lv Zheng <lv.zheng@intel.com> Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
2015-12-03 02:42:46 +00:00
error_exit:
if (ACPI_FAILURE(status) && status != AE_CTRL_TERMINATE) {
ACPI_EXCEPTION((AE_INFO, status,
"While parsing/handling command line"));
}
return (status);
}
/*******************************************************************************
*
* FUNCTION: acpi_db_signal_break_point
*
* PARAMETERS: walk_state - Current walk
*
* RETURN: Status
*
* DESCRIPTION: Called for AML_BREAKPOINT_OP
*
******************************************************************************/
void acpi_db_signal_break_point(struct acpi_walk_state *walk_state)
{
#ifndef ACPI_APPLICATION
if (acpi_gbl_db_thread_id != acpi_os_get_thread_id()) {
return;
}
#endif
/*
* Set the single-step flag. This will cause the debugger (if present)
* to break to the console within the AML debugger at the start of the
* next AML instruction.
*/
acpi_gbl_cm_single_step = TRUE;
acpi_os_printf("**break** Executed AML BreakPoint opcode\n");
}
/*******************************************************************************
*
* FUNCTION: acpi_db_single_step
*
* PARAMETERS: walk_state - Current walk
* op - Current executing op (from aml interpreter)
* opcode_class - Class of the current AML Opcode
*
* RETURN: Status
*
* DESCRIPTION: Called just before execution of an AML opcode.
*
******************************************************************************/
acpi_status
acpi_db_single_step(struct acpi_walk_state *walk_state,
union acpi_parse_object *op, u32 opcode_class)
{
union acpi_parse_object *next;
acpi_status status = AE_OK;
u32 original_debug_level;
union acpi_parse_object *display_op;
union acpi_parse_object *parent_op;
u32 aml_offset;
ACPI_FUNCTION_ENTRY();
ACPICA: Debugger: Add thread ID support so that single step mode can only apply to the debugger thread When the debugger is running in the kernel mode, acpi_db_single_step() may also be invoked by the kernel runtime code path but the single stepping command prompt may be erronously logged as the kernel logs and runtime code path cannot proceed. This patch fixes this issue by adding acpi_gbl_db_thread_id for the debugger thread and preventing acpi_db_single_step() to be invoked from other threads. It is not suitable to add acpi_thread_id parameter for acpi_os_execute() as the function may be implemented as work queue on some hosts. So it is better to let the hosts invoke acpi_set_debugger_thread_id(). Currently acpiexec is not configured as DEBUGGER_MULTI_THREADED, but we can do this. When we do this, it is better to invoke acpi_set_debugger_thread_id() in acpi_os_execute() when the execution type is OSL_DEBUGGER_MAIN_THREAD. The support should look like: create_thread(&tid); if (type == OSL_DEBUGGER_MAIN_THREAD) acpi_set_debugger_thread_id(tid); resume_thread(tid); Similarly, semop() may be used for pthread implementation. But this patch simply skips debugger thread ID check for application instead of introducing such complications as there is no need to skip acpi_db_single_step() for an application debugger - acpiexec. Note that the debugger thread ID can also be used by acpi_os_printf() to filter out debugger output. Lv Zheng. Signed-off-by: Lv Zheng <lv.zheng@intel.com> Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
2015-10-19 02:25:50 +00:00
#ifndef ACPI_APPLICATION
if (acpi_gbl_db_thread_id != acpi_os_get_thread_id()) {
return (AE_OK);
}
#endif
/* Check the abort flag */
if (acpi_gbl_abort_method) {
acpi_gbl_abort_method = FALSE;
return (AE_ABORT_METHOD);
}
aml_offset = (u32)ACPI_PTR_DIFF(op->common.aml,
walk_state->parser_state.aml_start);
/* Check for single-step breakpoint */
if (walk_state->method_breakpoint &&
(walk_state->method_breakpoint <= aml_offset)) {
/* Check if the breakpoint has been reached or passed */
/* Hit the breakpoint, resume single step, reset breakpoint */
acpi_os_printf("***Break*** at AML offset %X\n", aml_offset);
acpi_gbl_cm_single_step = TRUE;
acpi_gbl_step_to_next_call = FALSE;
walk_state->method_breakpoint = 0;
}
/* Check for user breakpoint (Must be on exact Aml offset) */
else if (walk_state->user_breakpoint &&
(walk_state->user_breakpoint == aml_offset)) {
acpi_os_printf("***UserBreakpoint*** at AML offset %X\n",
aml_offset);
acpi_gbl_cm_single_step = TRUE;
acpi_gbl_step_to_next_call = FALSE;
walk_state->method_breakpoint = 0;
}
/*
* Check if this is an opcode that we are interested in --
* namely, opcodes that have arguments
*/
if (op->common.aml_opcode == AML_INT_NAMEDFIELD_OP) {
return (AE_OK);
}
switch (opcode_class) {
case AML_CLASS_UNKNOWN:
case AML_CLASS_ARGUMENT: /* constants, literals, etc. do nothing */
return (AE_OK);
default:
/* All other opcodes -- continue */
break;
}
/*
* Under certain debug conditions, display this opcode and its operands
*/
if ((acpi_gbl_db_output_to_file) ||
(acpi_gbl_cm_single_step) || (acpi_dbg_level & ACPI_LV_PARSE)) {
if ((acpi_gbl_db_output_to_file) ||
(acpi_dbg_level & ACPI_LV_PARSE)) {
acpi_os_printf
("\nAML Debug: Next AML Opcode to execute:\n");
}
/*
* Display this op (and only this op - zero out the NEXT field
* temporarily, and disable parser trace output for the duration of
* the display because we don't want the extraneous debug output)
*/
original_debug_level = acpi_dbg_level;
acpi_dbg_level &= ~(ACPI_LV_PARSE | ACPI_LV_FUNCTIONS);
next = op->common.next;
op->common.next = NULL;
display_op = op;
parent_op = op->common.parent;
if (parent_op) {
if ((walk_state->control_state) &&
(walk_state->control_state->common.state ==
ACPI_CONTROL_PREDICATE_EXECUTING)) {
/*
* We are executing the predicate of an IF or WHILE statement
* Search upwards for the containing IF or WHILE so that the
* entire predicate can be displayed.
*/
while (parent_op) {
if ((parent_op->common.aml_opcode ==
AML_IF_OP)
|| (parent_op->common.aml_opcode ==
AML_WHILE_OP)) {
display_op = parent_op;
break;
}
parent_op = parent_op->common.parent;
}
} else {
while (parent_op) {
if ((parent_op->common.aml_opcode ==
AML_IF_OP)
|| (parent_op->common.aml_opcode ==
AML_ELSE_OP)
|| (parent_op->common.aml_opcode ==
AML_SCOPE_OP)
|| (parent_op->common.aml_opcode ==
AML_METHOD_OP)
|| (parent_op->common.aml_opcode ==
AML_WHILE_OP)) {
break;
}
display_op = parent_op;
parent_op = parent_op->common.parent;
}
}
}
/* Now we can display it */
#ifdef ACPI_DISASSEMBLER
acpi_dm_disassemble(walk_state, display_op, ACPI_UINT32_MAX);
#endif
if ((op->common.aml_opcode == AML_IF_OP) ||
(op->common.aml_opcode == AML_WHILE_OP)) {
if (walk_state->control_state->common.value) {
acpi_os_printf
("Predicate = [True], IF block was executed\n");
} else {
acpi_os_printf
("Predicate = [False], Skipping IF block\n");
}
} else if (op->common.aml_opcode == AML_ELSE_OP) {
acpi_os_printf
("Predicate = [False], ELSE block was executed\n");
}
/* Restore everything */
op->common.next = next;
acpi_os_printf("\n");
if ((acpi_gbl_db_output_to_file) ||
(acpi_dbg_level & ACPI_LV_PARSE)) {
acpi_os_printf("\n");
}
acpi_dbg_level = original_debug_level;
}
/* If we are not single stepping, just continue executing the method */
if (!acpi_gbl_cm_single_step) {
return (AE_OK);
}
/*
* If we are executing a step-to-call command,
* Check if this is a method call.
*/
if (acpi_gbl_step_to_next_call) {
if (op->common.aml_opcode != AML_INT_METHODCALL_OP) {
/* Not a method call, just keep executing */
return (AE_OK);
}
/* Found a method call, stop executing */
acpi_gbl_step_to_next_call = FALSE;
}
/*
* If the next opcode is a method call, we will "step over" it
* by default.
*/
if (op->common.aml_opcode == AML_INT_METHODCALL_OP) {
/* Force no more single stepping while executing called method */
acpi_gbl_cm_single_step = FALSE;
/*
* Set the breakpoint on/before the call, it will stop execution
* as soon as we return
*/
walk_state->method_breakpoint = 1; /* Must be non-zero! */
}
acpi_ex_exit_interpreter();
status = acpi_db_start_command(walk_state, op);
acpi_ex_enter_interpreter();
/* User commands complete, continue execution of the interrupted method */
return (status);
}
/*******************************************************************************
*
* FUNCTION: acpi_initialize_debugger
*
* PARAMETERS: None
*
* RETURN: Status
*
* DESCRIPTION: Init and start debugger
*
******************************************************************************/
acpi_status acpi_initialize_debugger(void)
{
acpi_status status;
ACPI_FUNCTION_TRACE(acpi_initialize_debugger);
/* Init globals */
acpi_gbl_db_buffer = NULL;
acpi_gbl_db_filename = NULL;
acpi_gbl_db_output_to_file = FALSE;
acpi_gbl_db_debug_level = ACPI_LV_VERBOSITY2;
acpi_gbl_db_console_debug_level = ACPI_NORMAL_DEFAULT | ACPI_LV_TABLES;
acpi_gbl_db_output_flags = ACPI_DB_CONSOLE_OUTPUT;
acpi_gbl_db_opt_no_ini_methods = FALSE;
acpi_gbl_db_buffer = acpi_os_allocate(ACPI_DEBUG_BUFFER_SIZE);
if (!acpi_gbl_db_buffer) {
return_ACPI_STATUS(AE_NO_MEMORY);
}
memset(acpi_gbl_db_buffer, 0, ACPI_DEBUG_BUFFER_SIZE);
/* Initial scope is the root */
acpi_gbl_db_scope_buf[0] = AML_ROOT_PREFIX;
acpi_gbl_db_scope_buf[1] = 0;
acpi_gbl_db_scope_node = acpi_gbl_root_node;
/* Initialize user commands loop */
acpi_gbl_db_terminate_loop = FALSE;
/*
* If configured for multi-thread support, the debug executor runs in
* a separate thread so that the front end can be in another address
* space, environment, or even another machine.
*/
if (acpi_gbl_debugger_configuration & DEBUGGER_MULTI_THREADED) {
/* These were created with one unit, grab it */
status = acpi_os_initialize_debugger();
if (ACPI_FAILURE(status)) {
acpi_os_printf("Could not get debugger mutex\n");
return_ACPI_STATUS(status);
}
/* Create the debug execution thread to execute commands */
acpi_gbl_db_threads_terminated = FALSE;
ACPICA: Debugger: Add thread ID support so that single step mode can only apply to the debugger thread When the debugger is running in the kernel mode, acpi_db_single_step() may also be invoked by the kernel runtime code path but the single stepping command prompt may be erronously logged as the kernel logs and runtime code path cannot proceed. This patch fixes this issue by adding acpi_gbl_db_thread_id for the debugger thread and preventing acpi_db_single_step() to be invoked from other threads. It is not suitable to add acpi_thread_id parameter for acpi_os_execute() as the function may be implemented as work queue on some hosts. So it is better to let the hosts invoke acpi_set_debugger_thread_id(). Currently acpiexec is not configured as DEBUGGER_MULTI_THREADED, but we can do this. When we do this, it is better to invoke acpi_set_debugger_thread_id() in acpi_os_execute() when the execution type is OSL_DEBUGGER_MAIN_THREAD. The support should look like: create_thread(&tid); if (type == OSL_DEBUGGER_MAIN_THREAD) acpi_set_debugger_thread_id(tid); resume_thread(tid); Similarly, semop() may be used for pthread implementation. But this patch simply skips debugger thread ID check for application instead of introducing such complications as there is no need to skip acpi_db_single_step() for an application debugger - acpiexec. Note that the debugger thread ID can also be used by acpi_os_printf() to filter out debugger output. Lv Zheng. Signed-off-by: Lv Zheng <lv.zheng@intel.com> Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
2015-10-19 02:25:50 +00:00
status = acpi_os_execute(OSL_DEBUGGER_MAIN_THREAD,
acpi_db_execute_thread, NULL);
if (ACPI_FAILURE(status)) {
ACPI_EXCEPTION((AE_INFO, status,
"Could not start debugger thread"));
acpi_gbl_db_threads_terminated = TRUE;
return_ACPI_STATUS(status);
}
ACPICA: Debugger: Add thread ID support so that single step mode can only apply to the debugger thread When the debugger is running in the kernel mode, acpi_db_single_step() may also be invoked by the kernel runtime code path but the single stepping command prompt may be erronously logged as the kernel logs and runtime code path cannot proceed. This patch fixes this issue by adding acpi_gbl_db_thread_id for the debugger thread and preventing acpi_db_single_step() to be invoked from other threads. It is not suitable to add acpi_thread_id parameter for acpi_os_execute() as the function may be implemented as work queue on some hosts. So it is better to let the hosts invoke acpi_set_debugger_thread_id(). Currently acpiexec is not configured as DEBUGGER_MULTI_THREADED, but we can do this. When we do this, it is better to invoke acpi_set_debugger_thread_id() in acpi_os_execute() when the execution type is OSL_DEBUGGER_MAIN_THREAD. The support should look like: create_thread(&tid); if (type == OSL_DEBUGGER_MAIN_THREAD) acpi_set_debugger_thread_id(tid); resume_thread(tid); Similarly, semop() may be used for pthread implementation. But this patch simply skips debugger thread ID check for application instead of introducing such complications as there is no need to skip acpi_db_single_step() for an application debugger - acpiexec. Note that the debugger thread ID can also be used by acpi_os_printf() to filter out debugger output. Lv Zheng. Signed-off-by: Lv Zheng <lv.zheng@intel.com> Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
2015-10-19 02:25:50 +00:00
} else {
acpi_gbl_db_thread_id = acpi_os_get_thread_id();
}
return_ACPI_STATUS(AE_OK);
}
ACPI_EXPORT_SYMBOL(acpi_initialize_debugger)
/*******************************************************************************
*
* FUNCTION: acpi_terminate_debugger
*
* PARAMETERS: None
*
* RETURN: None
*
* DESCRIPTION: Stop debugger
*
******************************************************************************/
void acpi_terminate_debugger(void)
{
/* Terminate the AML Debugger */
acpi_gbl_db_terminate_loop = TRUE;
if (acpi_gbl_debugger_configuration & DEBUGGER_MULTI_THREADED) {
/* Wait the AML Debugger threads */
while (!acpi_gbl_db_threads_terminated) {
acpi_os_sleep(100);
}
ACPICA: Debugger: Convert some mechanisms to OSPM specific The following mechanisms are OSPM specific: 1. Redirect output destination to console: no file redirection will be needed by an in-kernel debugger, there is even no file can be accessed when the debugger is running in the kernel mode. 2. Output command prompts: programs other than acpiexec can have different prompt characters and the prompt characters may be implemented as a special character sequence to form a char device IO protocol. 3. Command ready/complete handshake: OSPM debugger may wait more conditions to implement OSPM specific semantics (for example, FIFO full/empty conditions for O_NONBLOCK or IO open/close conditions). Leaving such OSPM specific stuffs in the ACPICA debugger core blocks Linux debugger IO driver implementation. Several new OSL APIs are provided by this patch: 1. acpi_os_initialize_command_signals: initialize command handshake mechanism or any other OSPM specific stuffs. 2. acpi_os_terminate_command_signals: reversal of acpi_os_initialize_command_signals. 3. acpi_os_wait_command_ready: putting debugger task into wait state when a command is not ready. OSPMs can terminate command loop by returning AE_CTRL_TERMINATE from this API. Normally, wait_event() or wait_for_multiple_object() may be used to implement this API. 4. acpi_os_notify_command_complete: putting user task into running state when a command has been completed. OSPMs can terminate command loop by returning AE_CTRL_TERMINATE from this API. Normally, wake_up() or set_event() may be used to implement this API. This patch also converts current command signaling implementation into a generic debugger layer (osgendbg.c) to be used by the existing OSPMs or acpiexec, in return, Linux can have chance to implement its own command handshake mechanism. This patch also implements acpiexec batch mode in a multi-threading mode comaptible style as a demo (this can be confirmed by configuring acpiexec into DEBUGGER_MULTI_THREADED mode where the batch mode is still working). Lv Zheng. Note that the OSPM specific command handshake mechanism is required by Linux kernel because: 1. Linux kernel trends to use wait queue to synchronize two threads, using mutexes to achieve that will cause false "dead lock" warnings. 2. The command handshake mechanism implemented by ACPICA is implemented in this way because of a design issue in debugger IO streaming. Debugger IO outputs are simply cached using a giant buffer, this should be tuned by Linux in the future. Signed-off-by: Lv Zheng <lv.zheng@intel.com> Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
2015-12-03 02:42:46 +00:00
acpi_os_terminate_debugger();
}
if (acpi_gbl_db_buffer) {
acpi_os_free(acpi_gbl_db_buffer);
acpi_gbl_db_buffer = NULL;
}
/* Ensure that debug output is now disabled */
acpi_gbl_db_output_flags = ACPI_DB_DISABLE_OUTPUT;
}
ACPI_EXPORT_SYMBOL(acpi_terminate_debugger)
ACPICA: Debugger: Add thread ID support so that single step mode can only apply to the debugger thread When the debugger is running in the kernel mode, acpi_db_single_step() may also be invoked by the kernel runtime code path but the single stepping command prompt may be erronously logged as the kernel logs and runtime code path cannot proceed. This patch fixes this issue by adding acpi_gbl_db_thread_id for the debugger thread and preventing acpi_db_single_step() to be invoked from other threads. It is not suitable to add acpi_thread_id parameter for acpi_os_execute() as the function may be implemented as work queue on some hosts. So it is better to let the hosts invoke acpi_set_debugger_thread_id(). Currently acpiexec is not configured as DEBUGGER_MULTI_THREADED, but we can do this. When we do this, it is better to invoke acpi_set_debugger_thread_id() in acpi_os_execute() when the execution type is OSL_DEBUGGER_MAIN_THREAD. The support should look like: create_thread(&tid); if (type == OSL_DEBUGGER_MAIN_THREAD) acpi_set_debugger_thread_id(tid); resume_thread(tid); Similarly, semop() may be used for pthread implementation. But this patch simply skips debugger thread ID check for application instead of introducing such complications as there is no need to skip acpi_db_single_step() for an application debugger - acpiexec. Note that the debugger thread ID can also be used by acpi_os_printf() to filter out debugger output. Lv Zheng. Signed-off-by: Lv Zheng <lv.zheng@intel.com> Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
2015-10-19 02:25:50 +00:00
/*******************************************************************************
*
* FUNCTION: acpi_set_debugger_thread_id
*
* PARAMETERS: thread_id - Debugger thread ID
*
* RETURN: None
*
* DESCRIPTION: Set debugger thread ID
*
******************************************************************************/
void acpi_set_debugger_thread_id(acpi_thread_id thread_id)
{
acpi_gbl_db_thread_id = thread_id;
}
ACPI_EXPORT_SYMBOL(acpi_set_debugger_thread_id)