Skip to content

If backends fail it is possible to try using a null pointer #3313

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
wants to merge 3 commits into
base: master
Choose a base branch
from
Open

If backends fail it is possible to try using a null pointer #3313

Changes from 2 commits
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
925695f
Check for null pointers in backends before use
peardox Jul 9, 2025
377c749
Check for null pointers in backends before use
peardox Jul 9, 2025
1a64dd4
if nit-picked * 4
peardox Jul 21, 2025
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
45 changes: 33 additions & 12 deletions src/whisper.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1326,6 +1326,10 @@ static ggml_backend_t whisper_backend_init_gpu(const whisper_context_params & pa
if (params.use_gpu) {
for (size_t i = 0; i < ggml_backend_dev_count(); ++i) {
ggml_backend_dev_t dev_cur = ggml_backend_dev_get(i);
// Prevent null pointer being used in following code
if(dev_cur == nullptr) {
continue;
}
if (ggml_backend_dev_type(dev_cur) == GGML_BACKEND_DEVICE_TYPE_GPU) {
if (cnt == 0 || cnt == params.gpu_device) {
dev = dev_cur;
Expand Down Expand Up @@ -1364,6 +1368,10 @@ static std::vector<ggml_backend_t> whisper_backend_init(const whisper_context_pa
// ACCEL backends
for (size_t i = 0; i < ggml_backend_dev_count(); ++i) {
ggml_backend_dev_t dev = ggml_backend_dev_get(i);
// Prevent null pointer being used in following code
if(dev == nullptr) {
continue;
}
if (ggml_backend_dev_type(dev) == GGML_BACKEND_DEVICE_TYPE_ACCEL) {
WHISPER_LOG_INFO("%s: using %s backend\n", __func__, ggml_backend_dev_name(dev));
ggml_backend_t backend = ggml_backend_dev_init(dev, nullptr);
Expand Down Expand Up @@ -1395,6 +1403,10 @@ static buft_list_t make_buft_list(whisper_context_params & params) {
int cnt = 0;
for (size_t i = 0; i < ggml_backend_dev_count(); ++i) {
ggml_backend_dev_t dev = ggml_backend_dev_get(i);
// Prevent null pointer being used in following code
if(dev == nullptr) {
continue;
}
if (ggml_backend_dev_type(dev) == GGML_BACKEND_DEVICE_TYPE_GPU) {
if (cnt == 0 || cnt == params.gpu_device) {
auto * buft = ggml_backend_dev_buffer_type(dev);
Expand All @@ -1412,20 +1424,23 @@ static buft_list_t make_buft_list(whisper_context_params & params) {

// CPU Extra
auto * cpu_dev = ggml_backend_dev_by_type(GGML_BACKEND_DEVICE_TYPE_CPU);
auto * cpu_reg = ggml_backend_dev_backend_reg(cpu_dev);
auto get_extra_bufts_fn = (ggml_backend_dev_get_extra_bufts_t)
ggml_backend_reg_get_proc_address(cpu_reg, "ggml_backend_dev_get_extra_bufts");
if (get_extra_bufts_fn) {
ggml_backend_buffer_type_t * extra_bufts = get_extra_bufts_fn(cpu_dev);
while (extra_bufts && *extra_bufts) {
buft_list.emplace_back(cpu_dev, *extra_bufts);
++extra_bufts;
// Prevent null pointer being used in following code
if(cpu_dev != nullptr) {
Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Nit: Can you add a space after the if keyword.

Copy link
Contributor Author

@peardox peardox Jul 21, 2025 •
edited
Loading

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Added space after if *4 to fix nit

The most easiest way to trigger the state this avoids is by commenting out the ggml_backend_load_all() in whisper-bench and build shared all_cpu_variants. OK that error would be pretty stupid but at least with this it'll tell you there are no backends. The alternative is a core-dump which tells you less than zero

I know most perople will build static for the build-box but if you want to pass something around as a release shared all_cpu_variants is a must

Also if you pass a static release around you run into the issue of it not loading the right CPU on older kit (probably same result - not tested - yet, only just though of it)

auto * cpu_reg = ggml_backend_dev_backend_reg(cpu_dev);
auto get_extra_bufts_fn = (ggml_backend_dev_get_extra_bufts_t)
ggml_backend_reg_get_proc_address(cpu_reg, "ggml_backend_dev_get_extra_bufts");
if (get_extra_bufts_fn) {
ggml_backend_buffer_type_t * extra_bufts = get_extra_bufts_fn(cpu_dev);
while (extra_bufts && *extra_bufts) {
buft_list.emplace_back(cpu_dev, *extra_bufts);
++extra_bufts;
}
}
}

// CPU
buft_list.emplace_back(cpu_dev, ggml_backend_cpu_buffer_type());

// CPU
buft_list.emplace_back(cpu_dev, ggml_backend_cpu_buffer_type());
}

return buft_list;
}

Expand Down Expand Up @@ -1728,6 +1743,12 @@ static bool whisper_model_load(struct whisper_model_loader * loader, whisper_con

// Create a list of available bufts, in priority order
buft_list_t buft_list = make_buft_list(wctx.params);
// Under rare circumstances it may not be possible to build a buft_list
// This can occur, for example, is all backends fail
if (buft_list.empty()) {
WHISPER_LOG_ERROR("%s: Failed to find a create a list of available bufts - possibly all backends failed\n", __func__);
return false;
}

auto create_tensor = [&](asr_tensor type, asr_system system, ggml_tensor * meta, int layer = 0) -> ggml_tensor * {
ggml_op op = ASR_TENSOR_INFO.at(type);
Expand Down