By: Linus Torvalds (torvalds.delete@this.linux-foundation.org), March 31, 2021 12:37 pm
Room: Moderated Discussions
Mark Roulo (nothanks.delete@this.xxx.com) on March 31, 2021 12:15 pm wrote:
>
> I can imagine that the concept works fine, that IBM got it right
I admittedly haven't followed it. I've seen more of the ppc side, and that was so complicated (with system-visible "transactional state" and transaction suspend/restart etc) that I refuse to believe that it's possibly even remotely the right way to do things.
IBM has had bugs on the Power side too, and afaik TM is disabled on power9: power9 TM CPU bugs
Maybe it works on zSeries. Maybe it's even widely (for some admittedly very odd definition of "widely") used. But honestly, from what I've seen on x86 and ppc, I'd be very surprised, and I can state from the power9 fiasco that no, it being IBM does not mean that they got it right.
Linus
>
> I can imagine that the concept works fine, that IBM got it right
I admittedly haven't followed it. I've seen more of the ppc side, and that was so complicated (with system-visible "transactional state" and transaction suspend/restart etc) that I refuse to believe that it's possibly even remotely the right way to do things.
IBM has had bugs on the Power side too, and afaik TM is disabled on power9: power9 TM CPU bugs
Maybe it works on zSeries. Maybe it's even widely (for some admittedly very odd definition of "widely") used. But honestly, from what I've seen on x86 and ppc, I'd be very surprised, and I can state from the power9 fiasco that no, it being IBM does not mean that they got it right.
Linus