LOCK Prefix vs. MESI Protocol?

Question

What is the purpose of the x86 LOCK prefix if the MESI protocol prohibits other cores from writing "exclusive" data anyway?

Am I a little confused between what LOCK provides and what MESI provides?

I understand that the MESI protocol is related to the fact that all cores see a consistent state of memory, but, as I understand it, it also prevents the kernel from writing to memory to which another kernel is already writing.

Answer 1

The MESI protocol makes memory caches invisible. This means that multi-threaded programs do not have to worry about the kernel reading outdated data from them or two kernels, writing to different parts of the cache line, and receiving half of one record and half of the other sent to main memory.

However, this does not help with read-modify-write operations such as incrementing, comparing and swapping, etc. The MESI protocol does not stop two cores from each reading of the same memory fragment, each of which adds one to it, and then writes the same value back, turning two increments into one.

In modern CPUs, the LOCK prefix blocks the cache line, so the read-modify-write operation is logically atomic. They are simplified, but hopefully they will give you this idea.

Unlocked increment:

• Acquiring a cache line is well compatible. Read the meaning.
• Add a value to the read value.
• Acquire the cache line exclusively (if it is not already E or M) and block it.
• Enter the new value in the cache line.
• Change the cache line to change and unlock.

Blocked increment:

• Acquire the cache line exclusively (if it is not already E or M) and block it.
• Read the value.
• Add one to it.
• Enter the new value in the cache line.
• Change the cache line to change and unlock.

Pay attention to the difference? In an unlocked increment, the cache line is locked only during a write operation to memory, like all writes. In a locked increment, the cache line is saved throughout the entire command, from the read operation to the write operation and includes the increment itself.

In addition, some processors have things other than memory caches that can affect memory visibility. For example, some CPUs have a prefetcher reader or a hosted write buffer, which can cause memory operations to be out of order. Where necessary, the LOCK prefix (or equivalent functionality on other CPUs) will also do whatever it takes to handle memory ordering problems.

Answer 2

Yes, you mix two different things. The MESI protocol is a cache coherence protocol that ensures that each core / processor receives the latest data from the cache of other processors (or memory) upon request. If the cache line is in the "E" state, this tells the requesting processor that one (and only one) other processor has a copy of this line. That is all he does; State "E" in any case does not allow the requesting processor to access the data ; it just says that only one processor has a copy of the data (and this copy also matches what is in memory). So, if the main request for data in state "E", the kernel will receive a copy of it. Another copy that was in "E" will be modified depending on whether the kernel asks for a copy for "write" or "read". If requested for writing, the old copy will be invalidated ("I"), and if it is to read the old copy will be placed in the general state "S".