Set return value of atomic_get_and_set to prevent attempting to acquire spinlock twice

[arvindsraj]

The return value of atomic_get_and_set isn't set which causes the kernel to incorrectly assume that the spinlock has not been acquired and causes a panic(KDL) when trying to acquire it again.

[pulkomandy]

pdziepak proposed a C++ implementation of this, which could be used in a cross platform way:

int32_t atomic_get_and_set(int32_t* ptr, int32_t value)
{
    auto& obj = *reinterpret_cast<std::atomic<int32_t>*>(ptr);
    return obj.exchange(value);
}

This would make things work for all platforms, however it uses the C++11 std::atomic so we have to keep the current implementation at least for x86_gcc2. But it would be best to have all gcc4 platforms use this.

[arvindsraj]

In the newly uploaded patch, I've reimplemented atomic_get_and_set in C++.

[pulkomandy]

Applied in hrev47708.