RTM/HLE abort on stack pointer mod

[Oliver_K_]

Hi,

to be confirmed: modification on stack pointer register (RSP/ESP) will not abort HLE/RTM?!

ty,

Oliver

[James_C_Intel2]

 modification on stack pointer register (RSP/ESP) will not abort HLE/RTM?!

Of course it won't abort.  It's just a register like any other from the TSX hardware's point of view.

(If you couldn't modify it you'd be unable to use the call or ret instructions inside a speculative region!)

[Oliver_K_]

I assume the same is true for exchanging the complete stack (e.g. assigning address from a complete different address range than the current one points to).

[James_C_Intel2]

I assume the same is true for exchanging the complete stack (e.g. assigning address from a complete different address range than the current one points to)

As before : " It's just a register like any other from the TSX hardware's point of view." So you can put whatever you like into it, and the hardware doesn't care. 

[jimdempseyatthecove]

James,

While changing the content of the stack pointer would not terminate the transaction, using the stack could when the access causes a page fault. This may be relatively rare after the first use of the specific page(s) of stack address space.

Jim Dempsey

[James_C_Intel2]

While changing the content of the stack pointer would not terminate the transaction, using the stack could when the access causes a page fault. 

Of course, but so could any memory access, so I see that as a different concern.