Skip to content

Commit

Permalink
AMDGPU max temps and PCIe gen/width and usage
Browse files Browse the repository at this point in the history
  • Loading branch information
@Syllo
Syllo committed Apr 2, 2022
1 parent 52140a1 commit a37f12b
Showing 1 changed file with 131 additions and 4 deletions.
135 changes: 131 additions & 4 deletions src/extract_gpuinfo_amdgpu.c
View file
Original file line number Diff line number Diff line change
Expand Up @@ -30,6 +30,7 @@
#include <drm/amdgpu_drm.h>
#include <errno.h>
#include <fcntl.h>
#include <inttypes.h>
#include <linux/kcmp.h>
#include <math.h>
#include <stdarg.h>
Expand Down Expand Up @@ -189,7 +190,10 @@ struct gpu_info_amdgpu {
int hwmonFD; // file descriptor for the /sys/bus/pci/devices/<this gpu>/hwmon/hwmon[0-9]+ folder

// We poll the fan frequently enough and want to avoid the open/close overhead of the sysfs file
FILE *fanSpeedFILE; // file descriptor for the current fan speed
FILE *fanSpeedFILE; // FILE* for this device current fan speed
FILE *PCIeLinkSpeed; // FILE* for this device PCIe link speed
FILE *PCIeLinkWidth; // FILE* for this device PCIe link width
FILE *PCIeBW; // FILE* for this device PCIe bandwidth over one second
// Used to compute the actual fan speed
unsigned maxFanValue;
};
Expand Down Expand Up @@ -534,6 +538,34 @@ static int readValueFromFileAt(int folderFD, const char *fileName, const char *f
return nread;
}

// Converts the link speed in MT/s and laneWidth to a PCIe generation
static unsigned pcieGenFromLinkSpeedAndWidth(unsigned linkSpeed, unsigned laneWidth) {
// Multiply by 100 to distinguish PCIe v1 and v2
unsigned laneSpeed = linkSpeed / laneWidth;
unsigned pcieGen = 0;
switch (laneSpeed) {
case 250:
pcieGen = 1;
break;
case 500:
pcieGen = 2;
break;
case 1000:
pcieGen = 3;
break;
case 2000:
pcieGen = 4;
break;
case 4000:
pcieGen = 5;
break;
case 8000:
pcieGen = 6;
break;
}
return pcieGen;
}

static void gpuinfo_amdgpu_populate_static_info(struct gpu_info *_gpu_info) {
struct gpu_info_amdgpu *gpu_info =
container_of(_gpu_info, struct gpu_info_amdgpu, base);
Expand Down Expand Up @@ -623,9 +655,9 @@ static void gpuinfo_amdgpu_populate_static_info(struct gpu_info *_gpu_info) {
char *fanSensorFile = usePWMSensor ? "pwm1" : "fan1_input";
unsigned maxSpeedVal;
NreadPatterns = readValueFromFileAt(gpu_info->hwmonFD, maxFanSpeedFile, "%u", &maxSpeedVal);
if (NreadPatterns) {
if (NreadPatterns == 1) {
gpu_info->maxFanValue = maxSpeedVal;
// Open the fan file
// Open the fan file for dynamic info gathering
int fanSpeedFD = openat(gpu_info->hwmonFD, fanSensorFile, O_RDONLY);
if (fanSpeedFD >= 0) {
gpu_info->fanSpeedFILE = fdopen(fanSpeedFD, "r");
Expand All @@ -634,7 +666,63 @@ static void gpuinfo_amdgpu_populate_static_info(struct gpu_info *_gpu_info) {
}
}
}
// TODO: temperature crit/emergency and PCIE max link/width

// Critical temparature
// temp1_* files should always be the GPU die in millidegrees Celsius
unsigned criticalTemp;
NreadPatterns = readValueFromFileAt(gpu_info->hwmonFD, "temp1_crit", "%u", &criticalTemp);
if (NreadPatterns == 1) {
static_info->temperature_slowdown_threshold = criticalTemp;
SET_VALID(gpuinfo_temperature_slowdown_valid, static_info->valid);
}

// Emergency/shutdown temparature
unsigned emergemcyTemp;
NreadPatterns = readValueFromFileAt(gpu_info->hwmonFD, "temp1_emergency", "%u", &emergemcyTemp);
if (NreadPatterns == 1) {
static_info->temperature_shutdown_threshold = emergemcyTemp;
SET_VALID(gpuinfo_temperature_shutdown_valid, static_info->valid);
}

// PCIe max link width
unsigned maxLinkWidth;
NreadPatterns = readValueFromFileAt(gpu_info->sysfsFD, "max_link_width", "%u", &maxLinkWidth);
if (NreadPatterns == 1) {
static_info->max_pcie_link_width = maxLinkWidth;
SET_VALID(gpuinfo_max_link_width_valid, static_info->valid);
}
// Open current link width for dynamic info gathering
gpu_info->PCIeLinkWidth = NULL;
int linkWidthFD = openat(gpu_info->sysfsFD, "current_link_width", O_RDONLY);
if (linkWidthFD) {
gpu_info->PCIeLinkWidth = fdopen(linkWidthFD, "r");
}

// PCIe max link speed
// [max|current]_link_speed export the value as "x.y GT/s PCIe" where x.y is a float value.
float maxLinkSpeedf;
NreadPatterns = readValueFromFileAt(gpu_info->sysfsFD, "max_link_speed", "%f GT/s PCIe", &maxLinkSpeedf);
if (NreadPatterns == 1 && IS_VALID(gpuinfo_max_link_width_valid, static_info->valid)) {
unsigned maxLinkSpeed = (unsigned)(maxLinkSpeedf * 1000.f);
unsigned pcieGen = pcieGenFromLinkSpeedAndWidth(maxLinkSpeed, static_info->max_pcie_link_width);
if (pcieGen) {
static_info->max_pcie_gen = pcieGen;
SET_VALID(gpuinfo_max_pcie_gen_valid, static_info->valid);
}
}
// Open current link speed
gpu_info->PCIeLinkSpeed = NULL;
int linkSpeedFD = openat(gpu_info->sysfsFD, "current_link_speed", O_RDONLY);
if (linkSpeedFD) {
gpu_info->PCIeLinkSpeed = fdopen(linkSpeedFD, "r");
}

// Open the PCIe bandwidth file for dynamic info gathering
gpu_info->PCIeBW = NULL;
int pcieBWFD = openat(gpu_info->sysfsFD, "pcie_bw", O_RDONLY);
if (pcieBWFD) {
gpu_info->PCIeBW = fdopen(pcieBWFD, "r");
}
}

static void gpuinfo_amdgpu_refresh_dynamic_info(struct gpu_info *_gpu_info) {
Expand Down Expand Up @@ -765,6 +853,45 @@ static void gpuinfo_amdgpu_refresh_dynamic_info(struct gpu_info *_gpu_info) {
SET_VALID(gpuinfo_power_draw_valid, dynamic_info->valid);
} else
RESET_VALID(gpuinfo_power_draw_valid, dynamic_info->valid);

// Current PCIe link used
if (gpu_info->PCIeLinkWidth) {
unsigned currentLinkWidth = 0;
int NreadPatterns = rewindAndReadPattern(gpu_info->PCIeLinkWidth, "%u", &currentLinkWidth);
if (NreadPatterns == 1) {
dynamic_info->curr_pcie_link_width = currentLinkWidth;
SET_VALID(gpuinfo_pcie_link_width_valid, dynamic_info->valid);
if (gpu_info->PCIeLinkSpeed) {
float currentLinkSpeedf;
NreadPatterns = rewindAndReadPattern(gpu_info->PCIeLinkSpeed, "%f GT/s PCIe", &currentLinkSpeedf);
if (NreadPatterns == 1) {
unsigned currentLinkSpeed = (unsigned)(currentLinkSpeedf * 1000.f);
unsigned pcieGen = pcieGenFromLinkSpeedAndWidth(currentLinkSpeed, currentLinkWidth);
if (pcieGen) {
dynamic_info->curr_pcie_link_gen = pcieGen;
SET_VALID(gpuinfo_pcie_link_gen_valid, dynamic_info->valid);
}
}
}
}
}
// PCIe bandwidth
if (gpu_info->PCIeBW) {
// According to https://github.com/torvalds/linux/blob/master/drivers/gpu/drm/amd/pm/amdgpu_pm.c, under the pcie_bw
// section, we should be able to read the number of packets received and sent by the GPU and get the maximum payload
// size during the last second. This is untested but should work when the file is populated by the driver.
uint64_t received, transmitted;
int maxPayloadSize;
int NreadPatterns = rewindAndReadPattern(gpu_info->PCIeBW, "%" SCNu64 " %" SCNu64 " %i", &received, &transmitted, &maxPayloadSize);
if (NreadPatterns == 3) {
received *= maxPayloadSize;
transmitted *= maxPayloadSize;
dynamic_info->pcie_rx = received;
dynamic_info->pcie_tx = transmitted;
SET_VALID(gpuinfo_pcie_rx_valid, dynamic_info->valid);
SET_VALID(gpuinfo_pcie_tx_valid, dynamic_info->valid);
}
}
}

static bool is_drm_fd(int fd_dir_fd, const char *name) {
Expand Down

0 comments on commit a37f12b

Please sign in to comment.