4 ms·

macOS 10.13.3 fixes the Intel KPTI issue

101 points by misterdata 9y ago

qubex 9y ago
Any noticeable performance penalty? Any benchmarks?
- epmaybe 9y ago
  Yes, what kind of real world performance will users notice?
  - gbrown_ 9y ago
    This is the real question, too many people seem to be blindly throwing around "Regression of X%". Phoronix have put out some benchmarks for Linux [1][2]. As one may expect I/O seems to be hit noticeably given it's handled by the kernel. [1] https://www.phoronix.com/scan.php?page=article&item=linux-415-x86pti&num=1 https://www.phoronix.com/scan.php?page=article&item=linux-41... [2] https://www.phoronix.com/scan.php?page=article&item=linux-more-x86pti&num=1 https://www.phoronix.com/scan.php?page=article&item=linux-mo...
    - pier25 9y ago
      IO performance is actually more worrying than CPU performance IMO and the hit is considerable for PCIe SSDs. So many applications depend on it. From databases to pro video, pro audio, etc.
      - jorvi 9y ago
        The question becomes: does Apple need to compensate customers? If Ford sold a car with a certain amount of torque advertised, and then an ECU patch (due to emission problems) slashed that torque by 30%, you bet they’d have to compensate either monetarily or with a buyback. MacBooks go for as much as secondhand cars too.
        CodeWriter23 9y ago
        That’s what Volkswagen and Bosch had to do to make amends for Dieselgate.
        pier25 9y ago
        I imagine Apple, and all PC manufacturers.
- faitswulff 9y ago
  From the Twitter thread: > The performance drop on a system with PCID is minimal. Most Macs have PCID. https://twitter.com/aionescu/status/948613035861553152 https://twitter.com/aionescu/status/948613035861553152
  - Matt3o12_ 9y ago
    Could you eleblrate what PCID is and which models (CPUs) Support that?
    - tadfisher 9y ago
      Process context identifiers, which are basically tags for regions of cache/TLB, allowing for fast context switching on multi-CPU systems. Introduced in Haswell I believe.
      - jorvi 9y ago
        Reading the Twitter responses PCID was introduced with Westmere and INVPCID with Haswell.
mnm1 9y ago
What about 10.12 and earlier versions?
- hivacruz 9y ago
  I hope so. As far as I know, the three latest OS X versions get Security Updates. So it should be 10.11+
- lunixbochs 9y ago
  10.12.6 + 2017-002 does not appear to patch this. I couldn't find any code related to double map, and sysenter did not change like in 10.13.2: https://twitter.com/lunixbochs/status/948763821933506560 https://twitter.com/lunixbochs/status/948763821933506560
faitswulff 9y ago
It's a nice surprise to read some good news about Apple's software engineering, for once.
- gumby 9y ago
  “For a change.” They’ve got some deserved bad press lately, but “for once” is absurd.
  - acdha 9y ago
    I think this says more about the online culture which draws attention to problems real, exaggerated, and imaginary in equal levels.
    - woolvalley 9y ago
      With a memory set of about 1 month.
acdha 9y ago
According to that tweet 10.13.2, the latest public release, fixes it and 10.13.3 will have some improvements.
rconti 9y ago
10.13.2 actually. >Say hello to the "Double Map" since 10.13.2 -- and with some surprises in 10.13.3 > Question: Is that right that there is need for 10.13.3 for full bug mitigation? > No, sorry, did not mean to imply that. There are some changes in 10.13.3 but they shouldn't affect the quality of the mitigation _as far as I can see_
zimbatm 9y ago
Does it restore the passwordless login feature? :p

101 points by misterdata 9y ago

qubex 9y ago

Any noticeable performance penalty? Any benchmarks?

epmaybe 9y ago
Yes, what kind of real world performance will users notice?
- gbrown_ 9y ago
  This is the real question, too many people seem to be blindly throwing around "Regression of X%". Phoronix have put out some benchmarks for Linux [1][2]. As one may expect I/O seems to be hit noticeably given it's handled by the kernel. [1] https://www.phoronix.com/scan.php?page=article&item=linux-415-x86pti&num=1 https://www.phoronix.com/scan.php?page=article&item=linux-41... [2] https://www.phoronix.com/scan.php?page=article&item=linux-more-x86pti&num=1 https://www.phoronix.com/scan.php?page=article&item=linux-mo...
  - pier25 9y ago
    IO performance is actually more worrying than CPU performance IMO and the hit is considerable for PCIe SSDs. So many applications depend on it. From databases to pro video, pro audio, etc.
    - jorvi 9y ago
      The question becomes: does Apple need to compensate customers? If Ford sold a car with a certain amount of torque advertised, and then an ECU patch (due to emission problems) slashed that torque by 30%, you bet they’d have to compensate either monetarily or with a buyback. MacBooks go for as much as secondhand cars too.
      - CodeWriter23 9y ago
        That’s what Volkswagen and Bosch had to do to make amends for Dieselgate.
      - pier25 9y ago
        I imagine Apple, and all PC manufacturers.
faitswulff 9y ago
From the Twitter thread: > The performance drop on a system with PCID is minimal. Most Macs have PCID. https://twitter.com/aionescu/status/948613035861553152 https://twitter.com/aionescu/status/948613035861553152
- Matt3o12_ 9y ago
  Could you eleblrate what PCID is and which models (CPUs) Support that?
  - tadfisher 9y ago
    Process context identifiers, which are basically tags for regions of cache/TLB, allowing for fast context switching on multi-CPU systems. Introduced in Haswell I believe.
    - jorvi 9y ago
      Reading the Twitter responses PCID was introduced with Westmere and INVPCID with Haswell.

mnm1 9y ago

What about 10.12 and earlier versions?

hivacruz 9y ago
I hope so. As far as I know, the three latest OS X versions get Security Updates. So it should be 10.11+
lunixbochs 9y ago
10.12.6 + 2017-002 does not appear to patch this. I couldn't find any code related to double map, and sysenter did not change like in 10.13.2: https://twitter.com/lunixbochs/status/948763821933506560 https://twitter.com/lunixbochs/status/948763821933506560

faitswulff 9y ago

It's a nice surprise to read some good news about Apple's software engineering, for once.

gumby 9y ago
“For a change.” They’ve got some deserved bad press lately, but “for once” is absurd.
- acdha 9y ago
  I think this says more about the online culture which draws attention to problems real, exaggerated, and imaginary in equal levels.
  - woolvalley 9y ago
    With a memory set of about 1 month.

acdha 9y ago

According to that tweet 10.13.2, the latest public release, fixes it and 10.13.3 will have some improvements.

rconti 9y ago

10.13.2 actually. >Say hello to the "Double Map" since 10.13.2 -- and with some surprises in 10.13.3 > Question: Is that right that there is need for 10.13.3 for full bug mitigation? > No, sorry, did not mean to imply that. There are some changes in 10.13.3 but they shouldn't affect the quality of the mitigation _as far as I can see_

zimbatm 9y ago

Does it restore the passwordless login feature? :p