Skip to content

lifecycle: Add exception handling for CPU affinity #113

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
SangKyeong-Jeong wants to merge 4 commits into
base: main
Choose a base branch
from
Open

lifecycle: Add exception handling for CPU affinity #113

Changes from all commits
Commits
Show all changes
4 commits
Select commit Hold shift + click to select a range
84a8717
lifecycle: Add exception handling for CPU affinity
SangKyeong-Jeong Mar 11, 2026
21d3d28
lifecycle: Use 64-bit operations
SangKyeong-Jeong Mar 12, 2026
af714a8
Merge branch 'eclipse-score:main' into lifecycle/bugfix/affinity-exce…
SangKyeong-Jeong Mar 13, 2026
7c700e5
Merge branch 'main' into lifecycle/bugfix/affinity-exception-handling
FScholPer Mar 13, 2026
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
2 changes: 1 addition & 1 deletion src/launch_manager_daemon/src/configuration_manager/configurationmanager.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -80,7 +80,7 @@ const char* kEnvVarDefaultValue = "/opt/internal/launch_manager/etc/ecu-cfg"; /

const uint32_t ConfigurationManager::kDefaultProcessExecutionError = 1U;
uint32_t ConfigurationManager::kDefaultProcessorAffinityMask() {
return (1U << osal::getNumCores()) - 1U;
return static_cast<uint32_t>((1ULL << osal::getNumCores()) - 1ULL);
Copy link
Contributor

@pawelrutkaq pawelrutkaq Mar 12, 2026

Choose a reason for hiding this comment

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

When there are 64 cores, is there still the issue or ? ;) In general, this affinity in LM supports only up to 32 cores, since later on cpu_mask_ also does not support more. So either we do an assert is there is more than 32 cores, or we do wanirng logs that default affinity is set only to first 32 cores ;)

Copy link

@daeyoung-jeong-lge daeyoung-jeong-lge Mar 12, 2026 •
edited
Loading

Choose a reason for hiding this comment

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

This PR addresses the Undefined Behavior on 32-core systems. As chungsky mentioned, getNumCores() already caps the count at 32, so >32 core systems are safe with this default.

Copy link
Contributor

@pawelrutkaq pawelrutkaq Mar 13, 2026

Choose a reason for hiding this comment

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

I know what i am saying that we shall probably add at least a warning (since you are already fixing issues around it) that it was capped somehwere, since defult behaviour for ie 32 and 33 cores is completely different. @NicolasFussberger what do you think ?

Copy link
Contributor

@NicolasFussberger NicolasFussberger Mar 13, 2026 •
edited
Loading

Choose a reason for hiding this comment

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

I agree, logging a warning could be helpful.
However, that might not be trivial. Probably you do not want to log the warning again and again every time osal::getNumCores() is called but only once during startup. So I think it would require refactoring getNumCores() method or adding another method for validation purposes.

The behaviour in case of > 32 cores seems reasonable to me. LaunchManager will just use the first 32 cores.

We could also add support for 64 cores if that is required, but that is probably a separate task.

Copy link
Contributor

@SimonKozik SimonKozik Mar 13, 2026

Choose a reason for hiding this comment

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

After a short investigation it looks like we are hitting an undefined behavior here...
For reference pls have look at C Standard (C18, ISO/IEC 9899:2018) section 6.5.7 Bitwise shift operators and C++ Standard (C++20, ISO/IEC 14882:2020) section 7.6.2.2 Bitwise shift operators [expr.shift].

If the value of the right operand is negative, or greater than or equal to the number of bits in the promoted left operand, the behavior is undefined.

IMHO it looks to me that hardware is not really doing the logical thing here. Please consider following example:

#include <cstdint>
#include <iostream>

void shiftLeftTest(std::uint32_t num) {
    std::uint32_t first_result = 1U << num;
    std::uint32_t second_result = 1U << 32;

    std::cout << "Shifting left 1U by " << num << std::endl;
    std::cout << "first_result --> " << first_result << std::endl;
    std::cout << "second_result --> " << second_result << std::endl;
}

int main() {
    shiftLeftTest(32);
}

This code will print following text:

Shifting left 1U by 32
first_result --> 1
second_result --> 0

So essentially compiler will calculate something different than hardware.
For this reason maybe we should go for the following code that especially address this case. This may be handy if we support 64 cores in the future.

uint32_t ConfigurationManager::kDefaultProcessorAffinityMask() {
    constexpr uint32_t BITS_PER_BYTE = 8;
    uint32_t bitMask = 0U;
    uint32_t cores = osal::getNumCores();

    uint32_t maskSize = sizeof(bitMask) * BITS_PER_BYTE;

    return ( cores >= maskSizeInBits ? -1U : (1U << cores) - 1U );
}

Any opinion?

Copy link
Contributor

@pawelrutkaq pawelrutkaq Mar 13, 2026

Choose a reason for hiding this comment

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

Yup, true. For me fine. @SimonKozik / @NicolasFussberger can we create an gtihub issue to support 64 cores and/or provide a warning indication to the user if there is still more core ?

Copy link
Contributor Author

@SangKyeong-Jeong SangKyeong-Jeong Mar 13, 2026

Choose a reason for hiding this comment

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

To resolve the Undefined Behavior issue that occurs with 32 cores, is there any problem with the method I applied in the patch, which performs the shift operation using a 64-bit integer (1ULL) and then casts it?

Since the return type of the kDefaultProcessorAffinityMask function is uint32_t, returning -1U seems awkward because it applies a negative sign to an unsigned type.

Copy link
Contributor

@SimonKozik SimonKozik Mar 13, 2026

Choose a reason for hiding this comment

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

It is true that 32 cores is a bit hard coded now and changing the maximum to 64, will require some changes in few files. For this reason it will be hard to argue, that we should make this function very future proof.

May main concern is the fact, that the change proposed in this PR is just masking the problem. Not really solving it. So if we in the future increase the mask size to 64-bit, your code will step into the same problem. Am I right here?

I will argue that we should document the root problem we are facing here. This probably will be in the form of a comment.
Apart of that we should have fix.
Fix can be in the form proposed inside my comment, or the fix can be as proposed in PR. But if we are going to go with 1ULL then we should document that we are avoiding the undefined behavior by increasing the size of the type on which we are performing calculation.

I just want to avoid stepping into the same problem when we eventually provide support for 64 cores.

PS.
A big plus of using 1ULL is the fact that we can avoid branch in the calculation.

Copy link
Contributor Author

@SangKyeong-Jeong SangKyeong-Jeong Mar 13, 2026

Choose a reason for hiding this comment

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

I agree that if the mask size is increased to 64 bits, we may encounter the same problem again. Currently, the maximum value is limited to 32 in the osal::getNumCores() function, which is why I made this change.
It also seems important to document this so that the issue does not occur again.

Copy link
Contributor

@NicolasFussberger NicolasFussberger Mar 13, 2026

Choose a reason for hiding this comment

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

I added a ticket here #122 to support 64 cores, linking to the proposal #113 (comment) for bitshifting

}
const int32_t ConfigurationManager::kDefaultSchedulingPolicy = SCHED_OTHER;
const int32_t ConfigurationManager::kDefaultRealtimeSchedulingPriority = 99;
Expand Down
Loading