堆栈，缓存未命中和虚拟内存

Question

In general for what I remember the stack of a Program it is a special portion of the memory handled in a special way (by the means of a LIFO structure ie a 'stack').

I am working in Linux in C and C++ and I am not sure about the following points

1. being the stack a piece of general memory, does it mean that in a Linux process it is supposed to be in some page of the Virtual Memory of that process?

2. I am used to know that if a piece of memory (I always thought about heap only) resides in the L1 Cache will be quicker to retrieve than L3 Cache. Does it apply to the stack as well?

Now stack is usually faster than heap, but if point 2 is true, still some data from the stack might be in L3 line and introduce slowness in the system.

Am I right in reasoning in the following terms or am I am missing something?

Answer 1

It is processor specific: AMD and Intel are doing different things, and even within each brand it is model specific.

Some processors (I forgot which, perhaps older AMD) are relating stack machine instructions (ie PUSH , POP , RET and CALL etc...) to the L1 cache.

BTW, Andrew Appel wrote (in the previous century) garbage collection can be faster than stack allocation (for SML compiled using CPS techniques), but, IIRC, this is less true today because current processors have some behavior relating the call stack to the cache.

But I believe that hot pieces of the call stack are often in L1 cache (even without hardware to help that), because the data there (local variables, return addresses, ...) is often accessed.

Of course, the call stack is in virtual memory ; use proc(5) , eg try

 tail /proc/$$/maps

(you could use cat instead of tail ) to obtain perhaps:

7f6366db5000-7f6366dd5000 r-xp 00000000 08:11 2100860                    /lib/x86_64-linux-gnu/ld-2.19.so
7f6366fac000-7f6366fb0000 rw-p 00000000 00:00 0 
7f6366fcc000-7f6366fd3000 r--s 00000000 08:11 964796                     /usr/lib/x86_64-linux-gnu/gconv/gconv-modules.cache
7f6366fd3000-7f6366fd5000 rw-p 00000000 00:00 0 
7f6366fd5000-7f6366fd6000 r--p 00020000 08:11 2100860                    /lib/x86_64-linux-gnu/ld-2.19.so
7f6366fd6000-7f6366fd7000 rw-p 00021000 08:11 2100860                    /lib/x86_64-linux-gnu/ld-2.19.so
7f6366fd7000-7f6366fd8000 rw-p 00000000 00:00 0 
7fff59aa1000-7fff59ac2000 rw-p 00000000 00:00 0                          [stack]
7fff59bfe000-7fff59c00000 r-xp 00000000 00:00 0                          [vdso]
ffffffffff600000-ffffffffff601000 r-xp 00000000 00:00 0                  [vsyscall]

Notice the [stack] segment.

Read also about ASLR & vdso(7)

By definition (of CPU caches ), the L1 cache contain usually the most often accessed data. Cache misses are costly (an access to data in your RAM sticks can be 100x slower than to L1 cache).