On Thu, Jan 26, 2023 at 11:12:36AM -0800, Dave Hansen wrote: > On 1/26/23 09:52, Jann Horn wrote: > >> Maybe I'm totally missing something, but I thought the scope here was > >> the "non-data operand independent timing behavior for the listed > >> instructions" referenced here: > >> > >>> https://www.intel.com/content/www/us/en/developer/articles/technical/software-security-guidance/best-practices/data-operand-independent-timing-isa-guidance.html > >> where the "listed instructions" is this list: > >> > >>> https://www.intel.com/content/www/us/en/developer/articles/technical/software-security-guidance/resources/data-operand-independent-timing-instructions.html > >> For example, that includes XOR with the 0x31 and 0x81 opcodes which > >> there are plenty of in the kernel. > > That list says at the top: "The table below lists instructions that > > have data-independent timing." > > So, first of all, apologies for the documentation. It needs some work, > and I see where the confusion is coming from. > > But, I did just confirm with the folks that wrote it. The "listed > instructions" *ARE* within the scope of being affected by the DOITM=0/1 > setting. > > Instead of saying: > > The table below lists instructions that have data-independent > timing. > > I think it should probably say something like: > > The table below lists instructions that have data-independent > timing when DOITM is enabled. > > (Modulo the MXCSR interactions for now) > > Somebody from Intel please thwack me over the head if I'm managing to > get this wrong (wouldn't be the first time). That's my understanding too, based on the part of https://www.intel.com/content/www/us/en/developer/articles/technical/software-security-guidance/best-practices/data-operand-independent-timing-isa-guidance.html that describes DOITM. The page https://www.intel.com/content/www/us/en/developer/articles/technical/software-security-guidance/resources/data-operand-independent-timing-instructions.html is actively misleading, so yes please get that fixed. I think the following is also extremely important: For Intel® Core™ family processors based on microarchitectures before Ice Lake and Intel Atom® family processors based on microarchitectures before Gracemont that do not enumerate IA32_UARCH_MISC_CTL, developers may assume that the instructions listed here operate as if DOITM is enabled. The end result is that on older CPUs, Intel explicitly guarantees that the instructions in https://www.intel.com/content/www/us/en/developer/articles/technical/software-security-guidance/resources/data-operand-independent-timing-instructions.html have data operand independent timing. But on newer CPUs, Intel has explicitly removed that guarantee, and enabling DOITM is needed to get it back. By the way, surely the importance of using DOITM on a particular CPU correlates strongly with its performance overhead? So I'm not sure that benchmarks of DOITM would even be very interesting, as we couldn't necessarily decide on something like "don't use DOITM if the overhead is more than X percent", since that would exclude exactly the CPUs where it's the most important to use... I think the real takeaway here is that the optimizations that Intel is apparently trying to introduce are a bad idea and not safe at all. To the extent that they exist at all, they should be an opt-in thing, not out-opt. The CPU gets that wrong, but Linux can flip that and do it right. - Eric