Spectre Mitigation Added To GCC 8, Seeking Backport To GCC 7

[phoronix]

Phoronix: Spectre Mitigation Added To GCC 8, Seeking Backport To GCC 7

Hitting the GCC 8 compiler Git/SVN code this Sunday morning are the changes needed compiler-side for CVE-2017-5715 / Spectre mitigation...

http://www.phoronix.com/scan.php?page=news_item&px=GCC-8-Spectre-Mitigation-Lands

[entropy]

What about GCC 4.x?

Just think of RHEL 6,7 which will still have a quite long period of paid support contracts.

[Emmanuel Deloget]

What about GCC 4.x?

Just think of RHEL 6,7 which will still have a quite long period of paid support contracts.

The GCC team will probably never provide backport for versions before GCC7 (maybe 6 but I highly doubt that) so RH customer will have to rely on RH.

Best regards,

-- Emmanuel Deloget

[slacka]

Is there any performance impact? If so, is there any way to disabled it for code that could not be attacked?

[pipe13]

The GCC team will probably never provide backport for versions before GCC7 (maybe 6 but I highly doubt that) so RH customer will have to rely on RH.

Best regards,

-- Emmanuel Deloget

RedHat Devtoolsets are your friend

https://access.redhat.com/documentation/en-us/red_hat_developer_toolset/

[carewolf]

Is there any performance impact? If so, is there any way to disabled it for code that could not be attacked?

You don't need to disable it. Just don't enable it.

[eva2000]

The GCC team will probably never provide backport for versions before GCC7 (maybe 6 but I highly doubt that) so RH customer will have to rely on RH.

Best regards,

-- Emmanuel Deloget

Redhat has devtoolset to give access to GCC 5.3.1, 6.3.1, 7.2.1 and Clang 4.0.1. Works well on my CentOS 6/7 based servers. Example Nginx compiled with GCC 4.8.5, 5.3.1, 6.3.1, 7.2.1 and 8.0 and Clang 3.4/4 https://community.centminmod.com/thr...nstalls.13729/

But yeah, for native system GCC, Redhat is behind usually. Maybe they'll make an exception for security reasons and backport

[caligula]

So how does the compiler fix the issue?

[arjan_intel]

So how does the compiler fix the issue?

it fixes the issue by not emitting the instruction that gets to use the indirect branch prediction.

[numacross]

it fixes the issue by not emitting the instruction that gets to use the indirect branch prediction.

Does it rely on special-sauce instructions introduced in latest microcode for the newer generations like the Windows solution?