Skip to content

GPUOP-907: fix gpuagent SIGSEGV on MI350P from clock freq OOB read #75

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
bhatturu wants to merge 1 commit into
base: main
Choose a base branch
from
Open

GPUOP-907: fix gpuagent SIGSEGV on MI350P from clock freq OOB read #75

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
19 changes: 13 additions & 6 deletions sw/nic/gpuagent/api/smi/amdsmi/smi_api.cc
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -529,6 +529,7 @@ smi_fill_clock_status_ (aga_gpu_handle_t gpu_handle,
amdsmi_gpu_metrics_t *metrics_info)
{
uint32_t clk_cnt = 0;
uint64_t cur_freq;
amdsmi_status_t amdsmi_ret;
uint32_t low_freq, high_freq;
amdsmi_frequencies_t freq = {};
Expand Down Expand Up @@ -642,54 +643,60 @@ smi_fill_clock_status_ (aga_gpu_handle_t gpu_handle,
}
// data fabric clock
amdsmi_ret = amdsmi_get_clk_freq(gpu_handle, AMDSMI_CLK_TYPE_DF, &freq);
cur_freq = current_frequency_hz(&freq);
if (unlikely(amdsmi_ret != AMDSMI_STATUS_SUCCESS)) {
AGA_TRACE_ERR("Failed to get data fabric clock frequencies for GPU {}, "
"err {}", gpu_handle, amdsmi_ret);
} else {
} else if (cur_freq != AMDSMI_INVALID_UINT32) {
// skip clock types whose current frequency is reported as NA
low_freq = high_freq = 0;
clock_status = &status->clock_status[clk_cnt];
// min and max frequencies are per clock type
find_low_high_frequency(&freq,
&clock_status->low_frequency,
&clock_status->high_frequency);
clock_status->clock_type = AGA_GPU_CLOCK_TYPE_FABRIC;
clock_status->frequency = freq.frequency[freq.current]/1000000;
clock_status->frequency = cur_freq / 1000000;
clock_status->deep_sleep =
(clock_status->frequency < clock_status->low_frequency);
clk_cnt++;
}
// DCE clock
amdsmi_ret = amdsmi_get_clk_freq(gpu_handle, AMDSMI_CLK_TYPE_DCEF, &freq);
cur_freq = current_frequency_hz(&freq);
if (unlikely(amdsmi_ret != AMDSMI_STATUS_SUCCESS)) {
AGA_TRACE_ERR("Failed to get DCE clock frequencies for GPU {}, err {}",
gpu_handle, amdsmi_ret);
} else {
} else if (cur_freq != AMDSMI_INVALID_UINT32) {
// skip clock types whose current frequency is reported as NA
low_freq = high_freq = 0;
clock_status = &status->clock_status[clk_cnt];
// min and max frequencies are per clock type
find_low_high_frequency(&freq,
&clock_status->low_frequency,
&clock_status->high_frequency);
clock_status->clock_type = AGA_GPU_CLOCK_TYPE_DCE;
clock_status->frequency = freq.frequency[freq.current]/1000000;
clock_status->frequency = cur_freq / 1000000;
clock_status->deep_sleep =
(clock_status->frequency < clock_status->low_frequency);
clk_cnt++;
}
// PCIe clock
amdsmi_ret = amdsmi_get_clk_freq(gpu_handle, AMDSMI_CLK_TYPE_PCIE, &freq);
cur_freq = current_frequency_hz(&freq);
if (unlikely(amdsmi_ret != AMDSMI_STATUS_SUCCESS)) {
AGA_TRACE_ERR("Failed to get PCIe clock frequencies for GPU {}, err {}",
gpu_handle, amdsmi_ret);
} else {
} else if (cur_freq != AMDSMI_INVALID_UINT32) {
// skip clock types whose current frequency is reported as NA
low_freq = high_freq = 0;
clock_status = &status->clock_status[clk_cnt];
// min and max frequencies are per clock type
find_low_high_frequency(&freq,
&clock_status->low_frequency,
&clock_status->high_frequency);
clock_status->clock_type = AGA_GPU_CLOCK_TYPE_PCIE;
clock_status->frequency = freq.frequency[freq.current]/1000000;
clock_status->frequency = cur_freq / 1000000;
clock_status->deep_sleep =
(clock_status->frequency < clock_status->low_frequency);
clk_cnt++;
Expand Down
23 changes: 20 additions & 3 deletions sw/nic/gpuagent/api/smi/amdsmi/smi_utils.hpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -46,9 +46,10 @@ static inline void
find_low_high_frequency (amdsmi_frequencies_t *freq,
uint32_t *min, uint32_t *max)
{
// create a vector of the valid frequencies
std::vector<uint64_t> f(freq->frequency,
freq->frequency + freq->num_supported);
// clamp num_supported to the array bound before building the vector
uint32_t n = (freq->num_supported < AMDSMI_MAX_NUM_FREQUENCIES) ?
freq->num_supported : AMDSMI_MAX_NUM_FREQUENCIES;
std::vector<uint64_t> f(freq->frequency, freq->frequency + n);

// sort vector
std::sort(f.begin(), f.end());
Expand Down Expand Up @@ -76,6 +77,22 @@ find_low_high_frequency (amdsmi_frequencies_t *freq,
return;
}

/// \brief return current raw frequency in Hz, clamping an out-of-bounds index
/// \param[in] freq frequencies struct from amdsmi
/// \return current frequency in Hz; 0 when no supported frequencies,
/// else the first entry if current is out of range
static inline uint64_t
current_frequency_hz (amdsmi_frequencies_t *freq)
{
if (freq->num_supported == 0) {
return 0;
}
uint32_t n = (freq->num_supported < AMDSMI_MAX_NUM_FREQUENCIES) ?
freq->num_supported : AMDSMI_MAX_NUM_FREQUENCIES;
uint32_t idx = (freq->current < n) ? freq->current : 0;
return freq->frequency[idx];
}

/// \brief convert amdsmi virtualization mode to aga vritualization mode
/// \param[in] virt_mode amdsmi virtualization mode
/// \return aga virtualization mode
Expand Down