update spectre v2 microcodes blacklist
Hello,
Intel released an updated microcode revision guidance [1] (8 Feb). The link is available in press-release [2], but it's quite easy to miss.
Some microcode revisions that were blacklisted before are now tested and pronounced production-ready:
Quoted 1 line
Lines with “***” were previously recommended to discontinue use. Subsequent testing by Intel has determined that these were unaffected by the stability issues and have been re-released without modification.
I didn't fully match the updated revision guidance and spectre_bad_microcodes[] in arch/x86/kernel/cpu/intel.c, but at least microcode 0xc2 for intel skylakes should not be blacklisted anymore.
The update is also revelant to a recent blacklist backport to 4.9-stable:
https://patchwork.kernel.org/patch/10209021/
[1] https://newsroom.intel.com/wp-content/uploads/sites/11/2018/02/microcode-update-guidance.pdf
[2] https://newsroom.intel.com/news/security-issue-update-progress-continues-firmware-updates/
Quoted 1 line
I didn't fully match the updated revision guidance and
spectre_bad_microcodes
I compared these lists and it seems that the only difference is about skylakes. Everything else is covered by less-or-equal criteria on revision version.
Both desktop and mobile skylakes are stated to be unaffected for 0xC2:
{ INTEL_FAM6_SKYLAKE_DESKTOP, 0x03, 0xc2 }, // signature 0x000506E3
{ INTEL_FAM6_SKYLAKE_MOBILE, 0x03, 0xc2 }, // signature 0x000406E3
I reformated the Intel bulletin into json format for ease of automation and batch processing -- it is attached.
Raw diff between the mainline blacklist and the bulletin looks like:
@@ -1,5 +1,6 @@
{ INTEL_FAM6_BROADWELL_CORE, 0x04, 0x28 },
{ INTEL_FAM6_BROADWELL_GT3E, 0x01, 0x1b },
+{ INTEL_FAM6_BROADWELL_X, 0x01, 0x0b000023 },
{ INTEL_FAM6_BROADWELL_X, 0x01, 0x0b000025 },
{ INTEL_FAM6_BROADWELL_XEON_D, 0x02, 0x14 },
{ INTEL_FAM6_BROADWELL_XEON_D, 0x03, 0x07000011 },
@@ -9,12 +10,10 @@
{ INTEL_FAM6_HASWELL_X, 0x02, 0x3b },
{ INTEL_FAM6_HASWELL_X, 0x04, 0x10 },
{ INTEL_FAM6_IVYBRIDGE_X, 0x04, 0x42a },
-{ INTEL_FAM6_KABYLAKE_DESKTOP, 0x09, 0x84 },
-{ INTEL_FAM6_KABYLAKE_DESKTOP, 0x0a, 0x84 },
-{ INTEL_FAM6_KABYLAKE_DESKTOP, 0x0b, 0x84 },
-{ INTEL_FAM6_KABYLAKE_MOBILE, 0x09, 0x84 },
-{ INTEL_FAM6_KABYLAKE_MOBILE, 0x0a, 0x84 },
-{ INTEL_FAM6_SKYLAKE_DESKTOP, 0x03, 0xc2 },
-{ INTEL_FAM6_SKYLAKE_MOBILE, 0x03, 0xc2 },
-{ INTEL_FAM6_SKYLAKE_X, 0x03, 0x0100013e },
+{ INTEL_FAM6_KABYLAKE_DESKTOP, 0x09, 0x80 },
+{ INTEL_FAM6_KABYLAKE_DESKTOP, 0x0a, 0x80 },
+{ INTEL_FAM6_KABYLAKE_DESKTOP, 0x0b, 0x80 },
+{ INTEL_FAM6_KABYLAKE_MOBILE, 0x09, 0x80 },
+{ INTEL_FAM6_KABYLAKE_MOBILE, 0x0a, 0x80 },
+{ INTEL_FAM6_SKYLAKE_X, 0x04, 0x0200003a },{ INTEL_FAM6_SKYLAKE_X, 0x04, 0x0200003c },
Note: Gemini Lake and Sandy Bridge are not considered (observed in wild).
Quoted 15 lines · Alexander Sergeyev
Quoted 2 lines
I didn't fully match the updated revision guidance and
spectre_bad_microcodesI compared these lists and it seems that the only difference is about skylakes. Everything else is covered by less-or-equal criteria on revision version.
Both desktop and mobile skylakes are stated to be unaffected for
0xC2:{ INTEL_FAM6_SKYLAKE_DESKTOP, 0x03,
0xc2}, // signature0x000506E3{ INTEL_FAM6_SKYLAKE_MOBILE, 0x03,0xc2}, // signature0x000406E3I reformated the Intel bulletin into json format for ease of automation and batch processing -- it is attached.
Raw diff between the mainline blacklist and the bulletin looks like:
Review line 4
Quoted 4 lines
(quoted diff) +1
-0@@ -1,5 +1,6 @@ { INTEL_FAM6_BROADWELL_CORE, 0x04, 0x28 }, { INTEL_FAM6_BROADWELL_GT3E, 0x01, 0x1b }, +{ INTEL_FAM6_BROADWELL_X, 0x01, 0x0b000023 }, { INTEL_FAM6_BROADWELL_X, 0x01, 0x0b000025 },
That's redundant since blacklisting 0x0b000025 will also stop us from using 0x0b000023 anyway.
Review
Quoted 5 lines
(quoted diff) +0
-5-{ INTEL_FAM6_KABYLAKE_DESKTOP, 0x09, 0x84 }, -{ INTEL_FAM6_KABYLAKE_DESKTOP, 0x0a, 0x84 }, -{ INTEL_FAM6_KABYLAKE_DESKTOP, 0x0b, 0x84 }, -{ INTEL_FAM6_KABYLAKE_MOBILE, 0x09, 0x84 }, -{ INTEL_FAM6_KABYLAKE_MOBILE, 0x0a, 0x84 },
No, let's not assume 0x84 will be safe on those just yet. Intel mentioned 0x84 as bad in one of the revisions of the doc, and it's safer to leave it blacklisted until they explicitly say otherwise.
Review
Quoted 2 lines
(quoted diff) +0
-2-{ INTEL_FAM6_SKYLAKE_DESKTOP, 0x03, 0xc2 }, -{ INTEL_FAM6_SKYLAKE_MOBILE, 0x03, 0xc2 },
OK. With 0xc2 being exonerated, I believe there is *no* publicly released microcode for these which needs to be blacklisted, so they can be removed entirely.
Quoted 1 line
-{ INTEL_FAM6_SKYLAKE_X, 0x03,
0x0100013e},
Again, that appeared in an earlier version of the document and hasn't been explicitly cleared.
Quoted 2 lines
+{ INTEL_FAM6_SKYLAKE_X, 0x04,
0x0200003a}, { INTEL_FAM6_SKYLAKE_X, 0x04,0x0200003c},
Redundant again.
Quoted 1 line
Note: Gemini Lake and Sandy Bridge are not considered (observed in wild).
Ah, but we have more information now from the doc.
The Gemini Lake 0x22 was observed in the wild, sure, but is also explicitly mentioned as OK in the latest doc. So I removed it from the blacklist (again, with no prior Gemini Lake bad microcode, so the line can be removed entirely).
The Sandy Bridge ones observed in the wild are, however, *newer* than the "Pre-Mitigation Production MCU" listed for the corresponding CPU in the table, and thus we should assume they should still be blacklisted.
Quoted 9 lines
Review line 4
Quoted 6 lines
Raw diff between the mainline blacklist and the bulletin looks like:
(quoted diff) +1
-0@@ -1,5 +1,6 @@ { INTEL_FAM6_BROADWELL_CORE, 0x04, 0x28 }, { INTEL_FAM6_BROADWELL_GT3E, 0x01, 0x1b }, +{ INTEL_FAM6_BROADWELL_X, 0x01, 0x0b000023 }, { INTEL_FAM6_BROADWELL_X, 0x01, 0x0b000025 },That's redundant since blacklisting
0x0b000025will also stop us from using0x0b000023anyway.
and 23 is the pre security version so known good.
Quoted 5 lines
Quoted 3 lines
Quoted 1 line
+{ INTEL_FAM6_BROADWELL_X, 0x01,
0x0b000023},That's redundant since blacklisting
0x0b000025will also stop us from using0x0b000023anyway.and 23 is the pre security version so known good.
But 0x0b000023 is listed under "stop deploying these MCU revs" (page 4, row 4). Has something changed?