X86-64 and long distance calls / jumps

Question

Quick summary: in x86-64 mode, are far transitions as slow as in x86-32 mode?

In the x86 processor, jumps are divided into three types:

• short, with PC offset +/- 127 bytes (2 byte instructions)
• next to the offset +/- 32k, which "collapses" the current segment (3 byte instruction)
• far that can jump anywhere (5 byte instruction)

short and close jumps take 1-2 cycles, and long jumps take 50-80 cycles, depending on the processor. This comes from my reading of the documentation because they "go beyond CS, the current code segment."

In x86-64 mode, code segments are not used. A segment is actually always 0..infinity. Ergo, there should be no penalty for going beyond the segment.

Thus, the question arises: does the number of clock cycles for the long jump change if the processor is in x86-64 mode?

A related issue with the bonus: most * nix-like operating systems running in 32-bit protected mode explicitly set the segment sizes to 0..infection and control the linear → physical translation completely through the page tables. Do they benefit from this in terms of call times (fewer clock cycles), or is it really the processor’s internal legacy of size segment registers since 8086?

Answer 1

CS is used not only for base and limitation, but also for permissions. CPL is encoded here, as well as other fields, such as:

• D-bit - 32-bit or 16-bit default segment size
• L-bit - selects compatibility or 64-bit mode for the segment (in this case, a significant base and limit)

Long jumps can also go through the goal gate, and long-distance calls can also go through the call gate. All of them should be processed regardless of the 64-bit mode.

To summarize, a down jump in 64-bit mode is not faster than in 32-bit mode. In fact, given that when 64-bit mode is enabled, segment descriptors are twice as large as when 64-bit mode is disabled, all accesses to descriptor tables are doubled, which can extend the transition time.