In this example, T1 executes some code in the atomic block, where it sets ptr to A and then does some other operations. If the two atomic blocks were joined together, since there is no code in between, we would be guaranteed that when we access ptr, it would point to A because all the operations in the braces would be atomic - no conflicting memory operations are allowed to occur anywhere in between the start and end of the atomic block.
However, the programmer here has accidentally split the two atomic blocks, so that conflicting operations can actually occur (the transaction has been committed too early, so other threads are now allowed to commit their transactions, and thus modify memory, before the next transaction in T1 begins). So, if we have the interleaving T1 atomic1 -> T2 atomic 1 -> T1 atomic 2, then ptr will be set to NULL before we access ptr again, causing a NULL pointer access.
In this example, T1 executes some code in the atomic block, where it sets ptr to A and then does some other operations. If the two atomic blocks were joined together, since there is no code in between, we would be guaranteed that when we access ptr, it would point to A because all the operations in the braces would be atomic - no conflicting memory operations are allowed to occur anywhere in between the start and end of the atomic block.
However, the programmer here has accidentally split the two atomic blocks, so that conflicting operations can actually occur (the transaction has been committed too early, so other threads are now allowed to commit their transactions, and thus modify memory, before the next transaction in T1 begins). So, if we have the interleaving T1 atomic1 -> T2 atomic 1 -> T1 atomic 2, then ptr will be set to NULL before we access ptr again, causing a NULL pointer access.
This comment was marked helpful 0 times.