136547dd7 · cryptotaxi247/netdata-1

@@ -646,29 +646,30 @@ cleanup:

646 static int read_clk_freq_file(procfile **p_ff, const char *const pathname, collected_number *num){

647 if(unlikely(!*p_ff)){

648 *p_ff = procfile_open(pathname, NULL, PROCFILE_FLAG_NO_ERROR_ON_FILE_IO);

649 - if(unlikely(!*p_ff)) return -2;

649 + if(unlikely(!*p_ff))

650 + return -1;

651 }

652

652 - if(unlikely(NULL == (*p_ff = procfile_readall(*p_ff)))) return -3;

653 + if(unlikely(NULL == (*p_ff = procfile_readall(*p_ff))))

654 + return -1;

655

656 for(size_t l = 0; l < procfile_lines(*p_ff) ; l++) {

657 char *str_with_units = NULL;

656 - if((*p_ff)->lines->lines[l].words >= 3 && !strcmp(procfile_lineword((*p_ff), l, 2), "*")) //format: X: collected_number *

658 +

659 + if(procfile_linewords(*p_ff, l) >= 3 && !strcmp(procfile_lineword((*p_ff), l, 2), "*")) //format: X: collected_number *

660 str_with_units = procfile_lineword((*p_ff), l, 1);

658 - else if ((*p_ff)->lines->lines[l].words == 2 && !strcmp(procfile_lineword((*p_ff), l, 1), "*")) //format: collected_number *

661 + else if (procfile_linewords(*p_ff, l) == 2 && !strcmp(procfile_lineword((*p_ff), l, 1), "*")) //format: collected_number *

662 str_with_units = procfile_lineword((*p_ff), l, 0);

663

664 if (str_with_units) {

662 - char *delim = strchr(str_with_units, 'M');

663 - char str_without_units[10];

664 - memcpy(str_without_units, str_with_units, delim - str_with_units);

665 - *num = str2ll(str_without_units, NULL);

665 + char *units = NULL;

666 + *num = str2ll(str_with_units, &units);

667 return 0;

668 }

669 }

670

670 - procfile_close((*p_ff));

671 - return -4;

671 + *num = 0; // the card is not active, so no speed reporting

672 + return 0;

673 }

674

675 static char *set_id(const char *const suf_1, const char *const suf_2, const char *const suf_3){

@@ -835,8 +836,8 @@ int do_sys_class_drm(int update_every, usec_t dt) {

836 int chart_prio = NETDATA_CHART_PRIO_DRM_AMDGPU;

837

838 if(unlikely(!drm_dir)) {

838 - char filename[FILENAME_MAX + 1];

839 - snprintfz(filename, FILENAME_MAX, "%s%s", netdata_configured_host_prefix, "/sys/class/drm");

839 + char filename[FILENAME_MAX];

840 + snprintfz(filename, sizeof(filename), "%s%s", netdata_configured_host_prefix, "/sys/class/drm");

841 const char *drm_dir_name = config_get(CONFIG_SECTION_PLUGIN_PROC_DRM, "directory to monitor", filename);

842 if(unlikely(NULL == (drm_dir = opendir(drm_dir_name)))){

843 collector_error("Cannot read directory '%s'", drm_dir_name);

@@ -849,23 +850,23 @@ int do_sys_class_drm(int update_every, usec_t dt) {

850 (de->d_name[0] == '.' && de->d_name[1] == '.' && de->d_name[2] == '\0'))) continue;

851

852 if(de->d_type == DT_LNK && !strncmp(de->d_name, "card", 4) && !strchr(de->d_name, '-')) {

852 - snprintfz(filename, FILENAME_MAX, "%s/%s/%s", drm_dir_name, de->d_name, "device/uevent");

853 + snprintfz(filename, sizeof(filename), "%s/%s/%s", drm_dir_name, de->d_name, "device/uevent");

854 if(check_card_is_amdgpu(filename)) continue;

855

856 /* Get static info */

857

858 struct card *const c = callocz(1, sizeof(struct card));

858 - snprintfz(filename, FILENAME_MAX, "%s/%s", drm_dir_name, de->d_name);

859 + snprintfz(filename, sizeof(filename), "%s/%s", drm_dir_name, de->d_name);

860 c->pathname = strdupz(filename);

861

861 - snprintfz(filename, FILENAME_MAX, "%s/%s", c->pathname, "device/device");

862 + snprintfz(filename, sizeof(filename), "%s/%s", c->pathname, "device/device");

863 if(read_single_base64_or_hex_number_file(filename, &c->id.asic_id)){

864 collector_error("Cannot read asic_id from '%s'", filename);

865 card_free(c);

866 continue;

867 }

868

868 - snprintfz(filename, FILENAME_MAX, "%s/%s", c->pathname, "device/revision");

869 + snprintfz(filename, sizeof(filename), "%s/%s", c->pathname, "device/revision");

870 if(read_single_base64_or_hex_number_file(filename, &c->id.pci_rev_id)){

871 collector_error("Cannot read pci_rev_id from '%s'", filename);

872 card_free(c);

@@ -883,10 +884,17 @@ int do_sys_class_drm(int update_every, usec_t dt) {

884

885

886 collected_number tmp_val;

886 - #define set_prop_pathname(prop_filename, prop_pathname, p_ff) do { \

887 - snprintfz(filename, FILENAME_MAX, "%s/%s", c->pathname, prop_filename); \

888 - if((p_ff && !read_clk_freq_file(p_ff, filename, &tmp_val)) || \

889 - !read_single_number_file(filename, (unsigned long long *) &tmp_val)) \

887 + #define set_prop_pathname_single_number_file(prop_filename, prop_pathname) do { \

888 + snprintfz(filename, sizeof(filename), "%s/%s", c->pathname, prop_filename); \

889 + if(!read_single_number_file(filename, (unsigned long long *) &tmp_val)) \

890 + prop_pathname = strdupz(filename); \

891 + else \

892 + collector_info("Cannot read file '%s'", filename); \

893 + } while(0)

894 +

895 + #define set_prop_pathname_clock_file(prop_filename, prop_pathname, p_ff) do { \

896 + snprintfz(filename, sizeof(filename), "%s/%s", c->pathname, prop_filename); \

897 + if(!read_clk_freq_file(p_ff, filename, &tmp_val)) \

898 prop_pathname = strdupz(filename); \

899 else \

900 collector_info("Cannot read file '%s'", filename); \

@@ -894,7 +902,7 @@ int do_sys_class_drm(int update_every, usec_t dt) {

902

903 /* Initialize GPU and VRAM utilization metrics */

904

897 - set_prop_pathname("device/gpu_busy_percent", c->pathname_util_gpu, NULL);

905 + set_prop_pathname_single_number_file("device/gpu_busy_percent", c->pathname_util_gpu);

906

907 if(c->pathname_util_gpu){

908 c->st_util_gpu = rrdset_create_localhost(

@@ -919,7 +927,7 @@ int do_sys_class_drm(int update_every, usec_t dt) {

927 add_do_rrd_x(c, do_rrd_util_gpu);

928 }

929

922 - set_prop_pathname("device/mem_busy_percent", c->pathname_util_mem, NULL);

930 + set_prop_pathname_single_number_file("device/mem_busy_percent", c->pathname_util_mem);

931

932 if(c->pathname_util_mem){

933 c->st_util_mem = rrdset_create_localhost(

@@ -947,7 +955,7 @@ int do_sys_class_drm(int update_every, usec_t dt) {

955

956 /* Initialize GPU and VRAM clock frequency metrics */

957

950 - set_prop_pathname("device/pp_dpm_sclk", c->pathname_clk_gpu, &c->ff_clk_gpu);

958 + set_prop_pathname_clock_file("device/pp_dpm_sclk", c->pathname_clk_gpu, &c->ff_clk_gpu);

959

960 if(c->pathname_clk_gpu){

961 c->st_clk_gpu = rrdset_create_localhost(

@@ -973,7 +981,7 @@ int do_sys_class_drm(int update_every, usec_t dt) {

981

982 }

983

976 - set_prop_pathname("device/pp_dpm_mclk", c->pathname_clk_mem, &c->ff_clk_mem);

984 + set_prop_pathname_clock_file("device/pp_dpm_mclk", c->pathname_clk_mem, &c->ff_clk_mem);

985

986 if(c->pathname_clk_mem){

987 c->st_clk_mem = rrdset_create_localhost(

@@ -1001,8 +1009,8 @@ int do_sys_class_drm(int update_every, usec_t dt) {

1009

1010 /* Initialize GPU memory usage metrics */

1011

1004 - set_prop_pathname("device/mem_info_vram_used", c->pathname_mem_used_vram, NULL);

1005 - set_prop_pathname("device/mem_info_vram_total", c->pathname_mem_total_vram, NULL);

1012 + set_prop_pathname_single_number_file("device/mem_info_vram_used", c->pathname_mem_used_vram);

1013 + set_prop_pathname_single_number_file("device/mem_info_vram_total", c->pathname_mem_total_vram);

1014 if(c->pathname_mem_total_vram) c->total_vram = tmp_val;

1015

1016 if(c->pathname_mem_used_vram && c->pathname_mem_total_vram){

@@ -1050,8 +1058,8 @@ int do_sys_class_drm(int update_every, usec_t dt) {

1058 add_do_rrd_x(c, do_rrd_vram);

1059 }

1060

1053 - set_prop_pathname("device/mem_info_vis_vram_used", c->pathname_mem_used_vis_vram, NULL);

1054 - set_prop_pathname("device/mem_info_vis_vram_total", c->pathname_mem_total_vis_vram, NULL);

1061 + set_prop_pathname_single_number_file("device/mem_info_vis_vram_used", c->pathname_mem_used_vis_vram);

1062 + set_prop_pathname_single_number_file("device/mem_info_vis_vram_total", c->pathname_mem_total_vis_vram);

1063 if(c->pathname_mem_total_vis_vram) c->total_vis_vram = tmp_val;

1064

1065 if(c->pathname_mem_used_vis_vram && c->pathname_mem_total_vis_vram){

@@ -1099,8 +1107,8 @@ int do_sys_class_drm(int update_every, usec_t dt) {

1107 add_do_rrd_x(c, do_rrd_vis_vram);

1108 }

1109

1102 - set_prop_pathname("device/mem_info_gtt_used", c->pathname_mem_used_gtt, NULL);

1103 - set_prop_pathname("device/mem_info_gtt_total", c->pathname_mem_total_gtt, NULL);

1110 + set_prop_pathname_single_number_file("device/mem_info_gtt_used", c->pathname_mem_used_gtt);

1111 + set_prop_pathname_single_number_file("device/mem_info_gtt_total", c->pathname_mem_total_gtt);

1112 if(c->pathname_mem_total_gtt) c->total_gtt = tmp_val;

1113

1114 if(c->pathname_mem_used_gtt && c->pathname_mem_total_gtt){

fix crash when the DRM file does not contain the right information (#19258)