linux-kernel - Re: [RFC] x86, tsc: Add kcmdline args for skipping tsc calibration sequences

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Message-ID: <cb798dcd-8cd6-8a92-9f6c-50bec0afe204@supermicro.com>
Date:   Tue, 24 Jul 2018 12:45:22 -0700
From:   patrickg <patrickg@...ermicro.com>
To:     "Brown, Len" <len.brown@...el.com>,
        "linux-kernel@...r.kernel.org" <linux-kernel@...r.kernel.org>
CC:     "mingo@...nel.org" <mingo@...nel.org>,
        "Du, Alek" <alek.du@...el.com>,
        "arjan@...ux.intel.com" <arjan@...ux.intel.com>,
        "Tang, Feng" <feng.tang@...el.com>
Subject: Re: [RFC] x86, tsc: Add kcmdline args for skipping tsc calibration
 sequences

K, did significant poking.

native_calibrate_cpu is getting precidence no matter what because on SKL server, native_calibrate_tsc is always returning zero (Note that there is a caveat 2 lines down).

In native_calibrate_tsc, I'm seeing it always return zero after the `switch (boot_cpu_data.x86_model)`.  crystal_khz is zero so it rolls through that, never assigns it.

Now I'm apparently not testing on production CPU's.  I've requested that they drop em' in so I can ensure `CPUID 15H TSC/Crystal ratio` doesn't differ between the ES and Prod silicon, since that will effect the math 

Anyways, Then after calibrate is found to be zero, it's utilizing the CPUID calculation from native_calibrate_cpu for tsc_khz since that's returning a 'proper' value when it does the reassignment:

---
    if (tsc_khz == 0) /* Hits here and reassigns tsc_khz to the cpuid calculation. */
        tsc_khz = cpu_khz;
    else if (abs(cpu_khz - tsc_khz) * 10 > tsc_khz)
        cpu_khz = tsc_khz;
---

I'll post another update when I've checked with Prod CPU's, /probably/ tomorrow.