Development mbed library for MAX32630FTHR

Dependents:   blinky_max32630fthr

Revision:
0:5c4d7b2438d3
--- /dev/null	Thu Jan 01 00:00:00 1970 +0000
+++ b/tools/memap.py	Fri Nov 11 20:59:50 2016 +0000
@@ -0,0 +1,674 @@
+#!/usr/bin/env python
+
+"""Memory Map File Analyser for ARM mbed"""
+
+import sys
+import os
+import re
+import csv
+import json
+import argparse
+from prettytable import PrettyTable
+
+from utils import argparse_filestring_type, \
+    argparse_lowercase_hyphen_type, argparse_uppercase_type
+
+DEBUG = False
+
+RE_ARMCC = re.compile(
+    r'^\s+0x(\w{8})\s+0x(\w{8})\s+(\w+)\s+(\w+)\s+(\d+)\s+[*]?.+\s+(.+)$')
+RE_IAR = re.compile(
+    r'^\s+(.+)\s+(zero|const|ro code|inited|uninit)\s'
+    r'+0x(\w{8})\s+0x(\w+)\s+(.+)\s.+$')
+
+class MemapParser(object):
+    """An object that represents parsed results, parses the memory map files,
+    and writes out different file types of memory results
+    """
+
+    print_sections = ('.text', '.data', '.bss')
+
+    misc_flash_sections = ('.interrupts', '.flash_config')
+
+    other_sections = ('.interrupts_ram', '.init', '.ARM.extab',
+                      '.ARM.exidx', '.ARM.attributes', '.eh_frame',
+                      '.init_array', '.fini_array', '.jcr', '.stab',
+                      '.stabstr', '.ARM.exidx', '.ARM')
+
+    # sections to print info (generic for all toolchains)
+    sections = ('.text', '.data', '.bss', '.heap', '.stack')
+
+    def __init__(self, detailed_misc=False):
+        """ General initialization
+        """
+        # 
+        self.detailed_misc = detailed_misc
+        
+        # list of all modules and their sections
+        self.modules = dict()
+
+        # sections must be defined in this order to take irrelevant out
+        self.all_sections = self.sections + self.other_sections + \
+                            self.misc_flash_sections + ('unknown', 'OUTPUT')
+
+        # list of all object files and mappting to module names
+        self.object_to_module = dict()
+
+        # Memory report (sections + summary)
+        self.mem_report = []
+
+        # Just the memory summary section
+        self.mem_summary = dict()
+
+        self.subtotal = dict()
+
+    def module_add(self, module_name, size, section):
+        """ Adds a module / section to the list
+
+        Positional arguments:
+        module_name - name of the module to add
+        size - the size of the module being added
+        section - the section the module contributes to
+        """
+
+        if module_name in self.modules:
+            self.modules[module_name][section] += size
+        else:
+            temp_dic = dict()
+            for section_idx in self.all_sections:
+                temp_dic[section_idx] = 0
+            temp_dic[section] = size
+            self.modules[module_name] = temp_dic
+
+    def check_new_section_gcc(self, line):
+        """ Check whether a new section in a map file has been detected (only
+        applies to gcc)
+
+        Positional arguments:
+        line - the line to check for a new section
+        """
+
+        for i in self.all_sections:
+            if line.startswith(i):
+                # should name of the section (assuming it's a known one)
+                return i
+
+        if line.startswith('.'):
+            return 'unknown'     # all others are classified are unknown
+        else:
+            return False         # everything else, means no change in section
+
+    
+    def path_object_to_module_name(self, txt):
+        """ Parse a path to object file to extract it's module and object data
+
+        Positional arguments:
+        txt - the path to parse the object and module name from
+        """
+
+        txt = txt.replace('\\', '/')
+        rex_mbed_os_name = r'^.+mbed-os\/(.+)\/(.+\.o)$'
+        test_rex_mbed_os_name = re.match(rex_mbed_os_name, txt)
+
+        if test_rex_mbed_os_name:
+
+            object_name = test_rex_mbed_os_name.group(2)
+            data = test_rex_mbed_os_name.group(1).split('/')
+            ndata = len(data)
+
+            if ndata == 1:
+                module_name = data[0]
+            else:
+                module_name = data[0] + '/' + data[1]
+
+            return [module_name, object_name]
+            
+        elif self.detailed_misc:           
+            rex_obj_name = r'^.+\/(.+\.o\)*)$'
+            test_rex_obj_name = re.match(rex_obj_name, txt)
+            if test_rex_obj_name:
+                object_name = test_rex_obj_name.group(1)
+                return ['Misc/' + object_name, ""]        
+                
+            return ['Misc', ""]
+        else: 
+            return ['Misc', ""]
+
+    def parse_section_gcc(self, line):
+        """ Parse data from a section of gcc map file
+
+        examples:
+                        0x00004308       0x7c ./.build/K64F/GCC_ARM/mbed-os/hal/targets/hal/TARGET_Freescale/TARGET_KPSDK_MCUS/spi_api.o
+         .text          0x00000608      0x198 ./.build/K64F/GCC_ARM/mbed-os/core/mbed-rtos/rtx/TARGET_CORTEX_M/TARGET_RTOS_M4_M7/TOOLCHAIN_GCC/HAL_CM4.o
+
+        Positional arguments:
+        line - the line to parse a section from
+        """
+        rex_address_len_name = re.compile(
+            r'^\s+.*0x(\w{8,16})\s+0x(\w+)\s(.+)$')
+
+        test_address_len_name = re.match(rex_address_len_name, line)
+
+        if test_address_len_name:
+
+            if int(test_address_len_name.group(2), 16) == 0: # size == 0
+                return ["", 0] # no valid entry
+            else:
+                m_name, _ = self.path_object_to_module_name(
+                    test_address_len_name.group(3))
+                m_size = int(test_address_len_name.group(2), 16)
+                return [m_name, m_size]
+
+        else: # special corner case for *fill* sections
+            #  example
+            # *fill*         0x0000abe4        0x4
+            rex_address_len = r'^\s+\*fill\*\s+0x(\w{8,16})\s+0x(\w+).*$'
+            test_address_len = re.match(rex_address_len, line)
+
+            if test_address_len:
+                if int(test_address_len.group(2), 16) == 0: # size == 0
+                    return ["", 0] # no valid entry
+                else:
+                    m_name = 'Fill'
+                    m_size = int(test_address_len.group(2), 16)
+                    return [m_name, m_size]
+            else:
+                return ["", 0] # no valid entry
+
+    def parse_map_file_gcc(self, file_desc):
+        """ Main logic to decode gcc map files
+
+        Positional arguments:
+        file_desc - a stream object to parse as a gcc map file
+        """
+
+        current_section = 'unknown'
+
+        with file_desc as infile:
+
+            # Search area to parse
+            for line in infile:
+                if line.startswith('Linker script and memory map'):
+                    current_section = "unknown"
+                    break
+
+            # Start decoding the map file
+            for line in infile:
+
+                change_section = self.check_new_section_gcc(line)
+
+                if change_section == "OUTPUT": # finish parsing file: exit
+                    break
+                elif change_section != False:
+                    current_section = change_section
+
+                [module_name, module_size] = self.parse_section_gcc(line)
+
+                if module_size == 0 or module_name == "":
+                    pass
+                else:
+                    self.module_add(module_name, module_size, current_section)
+
+                if DEBUG:
+                    print "Line: %s" % line,
+                    print "Module: %s\tSection: %s\tSize: %s" % \
+                        (module_name, current_section, module_size)
+                    raw_input("----------")
+
+    def parse_section_armcc(self, line):
+        """ Parse data from an armcc map file
+
+        Examples of armcc map file:
+            Base_Addr    Size         Type   Attr      Idx    E Section Name        Object
+            0x00000000   0x00000400   Data   RO        11222    RESET               startup_MK64F12.o
+            0x00000410   0x00000008   Code   RO        49364  * !!!main             c_w.l(__main.o)
+
+        Positional arguments:
+        line - the line to parse the section data from
+        """
+
+        test_rex_armcc = re.match(RE_ARMCC, line)
+
+        if test_rex_armcc:
+
+            size = int(test_rex_armcc.group(2), 16)
+
+            if test_rex_armcc.group(4) == 'RO':
+                section = '.text'
+            else:
+                if test_rex_armcc.group(3) == 'Data':
+                    section = '.data'
+                elif test_rex_armcc.group(3) == 'Zero':
+                    section = '.bss'
+                else:
+                    print "BUG armcc map parser"
+                    raw_input()
+
+            # lookup object in dictionary and return module name
+            object_name = test_rex_armcc.group(6)
+            if object_name in self.object_to_module:
+                module_name = self.object_to_module[object_name]
+            else:
+                module_name = 'Misc'
+
+            return [module_name, size, section]
+
+        else:
+            return ["", 0, ""] # no valid entry
+
+    def parse_section_iar(self, line):
+        """ Parse data from an IAR map file
+
+        Examples of IAR map file:
+         Section             Kind        Address     Size  Object
+         .intvec             ro code  0x00000000    0x198  startup_MK64F12.o [15]
+         .rodata             const    0x00000198      0x0  zero_init3.o [133]
+         .iar.init_table     const    0x00008384     0x2c  - Linker created -
+         Initializer bytes   const    0x00000198     0xb2  <for P3 s0>
+         .data               inited   0x20000000     0xd4  driverAtmelRFInterface.o [70]
+         .bss                zero     0x20000598    0x318  RTX_Conf_CM.o [4]
+         .iar.dynexit        uninit   0x20001448    0x204  <Block tail>
+           HEAP              uninit   0x20001650  0x10000  <Block tail>
+
+        Positional_arguments:
+        line - the line to parse section data from
+        """
+
+        test_rex_iar = re.match(RE_IAR, line)
+
+        if test_rex_iar:
+
+            size = int(test_rex_iar.group(4), 16)
+
+            if test_rex_iar.group(2) == 'const' or \
+               test_rex_iar.group(2) == 'ro code':
+                section = '.text'
+            elif test_rex_iar.group(2) == 'zero' or \
+            test_rex_iar.group(2) == 'uninit':
+                if test_rex_iar.group(1)[0:4] == 'HEAP':
+                    section = '.heap'
+                elif test_rex_iar.group(1)[0:6] == 'CSTACK':
+                    section = '.stack'
+                else:
+                    section = '.bss' #  default section
+
+            elif test_rex_iar.group(2) == 'inited':
+                section = '.data'
+            else:
+                print "BUG IAR map parser"
+                raw_input()
+
+            # lookup object in dictionary and return module name
+            object_name = test_rex_iar.group(5)
+            if object_name in self.object_to_module:
+                module_name = self.object_to_module[object_name]
+            else:
+                module_name = 'Misc'
+
+            return [module_name, size, section]
+
+        else:
+            return ["", 0, ""] # no valid entry
+
+    def parse_map_file_armcc(self, file_desc):
+        """ Main logic to decode armc5 map files
+
+        Positional arguments:
+        file_desc - a file like object to parse as an armc5 map file
+        """
+
+        with file_desc as infile:
+
+            # Search area to parse
+            for line in infile:
+                if line.startswith('    Base Addr    Size'):
+                    break
+
+            # Start decoding the map file
+            for line in infile:
+
+                [name, size, section] = self.parse_section_armcc(line)
+
+                if size == 0 or name == "" or section == "":
+                    pass
+                else:
+                    self.module_add(name, size, section)
+
+    def parse_map_file_iar(self, file_desc):
+        """ Main logic to decode IAR map files
+
+        Positional arguments:
+        file_desc - a file like object to parse as an IAR map file
+        """
+
+        with file_desc as infile:
+
+            # Search area to parse
+            for line in infile:
+                if line.startswith('  Section  '):
+                    break
+
+            # Start decoding the map file
+            for line in infile:
+
+                [name, size, section] = self.parse_section_iar(line)
+
+                if size == 0 or name == "" or section == "":
+                    pass
+                else:
+                    self.module_add(name, size, section)
+
+    def search_objects(self, path):
+        """ Searches for object files and creates mapping: object --> module
+
+        Positional arguments:
+        path - the path to an object file
+        """
+
+        path = path.replace('\\', '/')
+
+        # check location of map file
+        rex = r'^(.+)' + r'\/(.+\.map)$'
+        test_rex = re.match(rex, path)
+
+        if test_rex:
+            search_path = test_rex.group(1) + '/mbed-os/'
+        else:
+            print "Warning: this doesn't look like an mbed project"
+            return
+
+        for root, _, obj_files in os.walk(search_path):
+            for obj_file in obj_files:
+                if obj_file.endswith(".o"):
+                    module_name, object_name = self.path_object_to_module_name(
+                        os.path.join(root, obj_file))
+
+                    if object_name in self.object_to_module:
+                        if DEBUG:
+                            print "WARNING: multiple usages of object file: %s"\
+                                % object_name
+                            print "    Current: %s" % \
+                                self.object_to_module[object_name]
+                            print "    New:     %s" % module_name
+                            print " "
+                    else:
+                        self.object_to_module.update({object_name:module_name})
+
+    export_formats = ["json", "csv-ci", "table"]
+
+    def generate_output(self, export_format, file_output=None):
+        """ Generates summary of memory map data
+
+        Positional arguments:
+        export_format - the format to dump
+
+        Keyword arguments:
+        file_desc - descriptor (either stdout or file)
+
+        Returns: generated string for the 'table' format, otherwise None
+        """
+
+        try:
+            if file_output:
+                file_desc = open(file_output, 'wb')
+            else:
+                file_desc = sys.stdout
+        except IOError as error:
+            print "I/O error({0}): {1}".format(error.errno, error.strerror)
+            return False
+
+        to_call = {'json': self.generate_json,
+                   'csv-ci': self.generate_csv,
+                   'table': self.generate_table}[export_format]
+        output = to_call(file_desc)
+
+        if file_desc is not sys.stdout:
+            file_desc.close()
+
+        return output
+
+    def generate_json(self, file_desc):
+        """Generate a json file from a memory map
+
+        Positional arguments:
+        file_desc - the file to write out the final report to
+        """
+        file_desc.write(json.dumps(self.mem_report, indent=4))
+        file_desc.write('\n')
+
+        return None
+
+    def generate_csv(self, file_desc):
+        """Generate a CSV file from a memoy map
+
+        Positional arguments:
+        file_desc - the file to write out the final report to
+        """
+        csv_writer = csv.writer(file_desc, delimiter=',',
+                                quoting=csv.QUOTE_MINIMAL)
+
+        csv_module_section = []
+        csv_sizes = []
+        for i in sorted(self.modules):
+            for k in self.print_sections:
+                csv_module_section += [i+k]
+                csv_sizes += [self.modules[i][k]]
+
+        csv_module_section += ['static_ram']
+        csv_sizes += [self.mem_summary['static_ram']]
+
+        csv_module_section += ['heap']
+        if self.mem_summary['heap'] == 0:
+            csv_sizes += ['unknown']
+        else:
+            csv_sizes += [self.mem_summary['heap']]
+
+        csv_module_section += ['stack']
+        if self.mem_summary['stack'] == 0:
+            csv_sizes += ['unknown']
+        else:
+            csv_sizes += [self.mem_summary['stack']]
+
+        csv_module_section += ['total_ram']
+        csv_sizes += [self.mem_summary['total_ram']]
+
+        csv_module_section += ['total_flash']
+        csv_sizes += [self.mem_summary['total_flash']]
+
+        csv_writer.writerow(csv_module_section)
+        csv_writer.writerow(csv_sizes)
+
+        return None
+
+    def generate_table(self, file_desc):
+        """Generate a table from a memoy map
+
+        Positional arguments:
+        file_desc - the file to write out the final report to
+
+        Returns: string of the generated table
+        """
+        # Create table
+        columns = ['Module']
+        columns.extend(self.print_sections)
+
+        table = PrettyTable(columns)
+        table.align["Module"] = "l"
+        for col in self.print_sections:
+            table.align[col] = 'r'
+
+        for i in list(self.print_sections):
+            table.align[i] = 'r'
+
+        for i in sorted(self.modules):
+            row = [i]
+
+            for k in self.print_sections:
+                row.append(self.modules[i][k])
+
+            table.add_row(row)
+
+        subtotal_row = ['Subtotals']
+        for k in self.print_sections:
+            subtotal_row.append(self.subtotal[k])
+
+        table.add_row(subtotal_row)
+
+        output = table.get_string()
+        output += '\n'
+
+        if self.mem_summary['heap'] == 0:
+            output += "Allocated Heap: unknown\n"
+        else:
+            output += "Allocated Heap: %s bytes\n" % \
+                        str(self.mem_summary['heap'])
+
+        if self.mem_summary['stack'] == 0:
+            output += "Allocated Stack: unknown\n"
+        else:
+            output += "Allocated Stack: %s bytes\n" % \
+                        str(self.mem_summary['stack'])
+
+        output += "Total Static RAM memory (data + bss): %s bytes\n" % \
+                        str(self.mem_summary['static_ram'])
+        output += "Total RAM memory (data + bss + heap + stack): %s bytes\n" % \
+                        str(self.mem_summary['total_ram'])
+        output += "Total Flash memory (text + data + misc): %s bytes\n" % \
+                        str(self.mem_summary['total_flash'])
+
+        return output
+
+    toolchains = ["ARM", "ARM_STD", "ARM_MICRO", "GCC_ARM", "IAR"]
+
+    def compute_report(self):
+        for k in self.sections:
+            self.subtotal[k] = 0
+
+        for i in sorted(self.modules):
+            for k in self.sections:
+                self.subtotal[k] += self.modules[i][k]
+
+        # Calculate misc flash sections
+        self.misc_flash_mem = 0
+        for i in self.modules:
+            for k in self.misc_flash_sections:
+                if self.modules[i][k]:
+                    self.misc_flash_mem += self.modules[i][k]
+
+        self.mem_summary = {
+            'static_ram': (self.subtotal['.data'] + self.subtotal['.bss']),
+            'heap': (self.subtotal['.heap']),
+            'stack': (self.subtotal['.stack']),
+            'total_ram': (self.subtotal['.data'] + self.subtotal['.bss'] +
+                          self.subtotal['.heap']+self.subtotal['.stack']),
+            'total_flash': (self.subtotal['.text'] + self.subtotal['.data'] +
+                            self.misc_flash_mem),
+        }
+
+        self.mem_report = []
+        for i in sorted(self.modules):
+            self.mem_report.append({
+                "module":i,
+                "size":{
+                    k:self.modules[i][k] for k in self.print_sections
+                }
+            })
+
+        self.mem_report.append({
+            'summary': self.mem_summary
+        })
+
+    def parse(self, mapfile, toolchain):
+        """ Parse and decode map file depending on the toolchain
+
+        Positional arguments:
+        mapfile - the file name of the memory map file
+        toolchain - the toolchain used to create the file
+        """
+
+        result = True
+        try:
+            with open(mapfile, 'r') as file_input:
+                if toolchain == "ARM" or toolchain == "ARM_STD" or\
+                   toolchain == "ARM_MICRO":
+                    self.search_objects(os.path.abspath(mapfile))
+                    self.parse_map_file_armcc(file_input)
+                elif toolchain == "GCC_ARM":
+                    self.parse_map_file_gcc(file_input)
+                elif toolchain == "IAR":
+                    self.search_objects(os.path.abspath(mapfile))
+                    self.parse_map_file_iar(file_input)
+                else:
+                    result = False
+            
+            self.compute_report()
+        
+        except IOError as error:
+            print "I/O error({0}): {1}".format(error.errno, error.strerror)
+            result = False
+        return result
+
+def main():
+    """Entry Point"""
+
+    version = '0.3.12'
+
+    # Parser handling
+    parser = argparse.ArgumentParser(
+        description="Memory Map File Analyser for ARM mbed\nversion %s" %
+        version)
+
+    parser.add_argument(
+        'file', type=argparse_filestring_type, help='memory map file')
+
+    parser.add_argument(
+        '-t', '--toolchain', dest='toolchain',
+        help='select a toolchain used to build the memory map file (%s)' %
+        ", ".join(MemapParser.toolchains),
+        required=True,
+        type=argparse_uppercase_type(MemapParser.toolchains, "toolchain"))
+
+    parser.add_argument(
+        '-o', '--output', help='output file name', required=False)
+
+    parser.add_argument(
+        '-e', '--export', dest='export', required=False, default='table',
+        type=argparse_lowercase_hyphen_type(MemapParser.export_formats,
+                                            'export format'),
+        help="export format (examples: %s: default)" %
+        ", ".join(MemapParser.export_formats))
+
+    parser.add_argument('-v', '--version', action='version', version=version)
+    
+    parser.add_argument('-d', '--detailed', action='store_true', help='Displays the elements in "Misc" in a detailed fashion', required=False)
+
+    # Parse/run command
+    if len(sys.argv) <= 1:
+        parser.print_help()
+        sys.exit(1)
+
+
+    args = parser.parse_args()
+
+    # Create memap object
+    memap = MemapParser(detailed_misc=args.detailed)
+
+    # Parse and decode a map file
+    if args.file and args.toolchain:
+        if memap.parse(args.file, args.toolchain) is False:
+            sys.exit(0)
+
+    returned_string = None
+    # Write output in file
+    if args.output != None:
+        returned_string = memap.generate_output(args.export, args.output)
+    else: # Write output in screen
+        returned_string = memap.generate_output(args.export)
+
+    if args.export == 'table' and returned_string:
+        print returned_string
+
+    sys.exit(0)
+
+if __name__ == "__main__":
+    main()