Re: [PATCH 2/2] v3 more sampling fun 2

speck for mark gross <[email protected]> writes:
> diff --git a/Documentation/admin-guide/hw-vuln/special-register-buffer-data-sampling.rst b/Documentation/admin-guide/hw-vuln/special-register-buffer-data-sampling.rst
> new file mode 100644
> index 000000000000..5acc7748f8e9
> --- /dev/null
> +++ b/Documentation/admin-guide/hw-vuln/special-register-buffer-data-sampling.rst
> @@ -0,0 +1,145 @@
> +.. SPDX-License-Identifier: GPL-2.0
> +
> +SRBDS - Special Register Buffer Data Sampling
> +======================================

I doubt that this builds w/o warnings. And right:

Documentation/admin-guide/hw-vuln/special-register-buffer-data-sampling.rst:4: WARNING: Title underline too short.

SRBDS - Special Register Buffer Data Sampling
======================================^

Documentation/admin-guide/hw-vuln/special-register-buffer-data-sampling.rst:56: WARNING: Title underline too short.

Attack scenarios
---------------^

Documentation/admin-guide/hw-vuln/special-register-buffer-data-sampling.rst:76: WARNING: Unexpected indentation.
Documentation/admin-guide/hw-vuln/special-register-buffer-data-sampling.rst:77: WARNING: Block quote ends without a blank line; unexpected unindent.

Sigh...

> +SRBDS is a hardware vulnerability that allows MDS techniques to infer values

lacks a link to the MDS documentation

> +returned from special register accesses.  Special register accesses are
> +accesses to off core registers.  According to Intels evaluation, the special
> +register reads that have a security expectation of privacy are:
> +RDRAND, RDSEED and SGX EGETKEY.

And what 

> +
> +When RDRAND and RDSEED instructions are used the data is moved to the core
> +through the special register mechanism.

This sentence is not providing any new information and can be packed in
the above where you mention RDRAND, RDSEED already. EGETKEY is no instruction?

> +Affected processors
> +--------------------
> +Core models (desktop, mobile, Xeon-E3) that implement RDRAND and/or RDSEED and
> +are vulnerable to MFBDS (Micro architectural Fill Buffer Data
> Sampling) variant

s/to/to the/

> +of MDS (Micro architectural Data Sampling) or to TAA (TSX Asynchronous Abort)
> +when TSX is enabled,
> +
> +  =============  ============  ========
> +  common name    Family_Model  Stepping
> +  =============  ============  ========
> +  Ivybridge      06_3AH        All
> +
> +  Haswell        06_3CH        All
> +  Haswell_L      06_45H        All
> +  Haswell_G      06_46H        All
> +
> +  Broadwell_G    06_47H        All
> +  Broadwell      06_3DH        All
> +
> +  Skylake_L      06_4EH        All
> +  Skylake        06_5EH        All
> +
> +  Kabylake_L     06_8EH        <=A
> +  Kabylake_L     06_8EH        0xB only if TSX is enabled
> +  Kabylake_L     06_8EH        0xC only if TSX is enabled
> +
> +  Kabylake       06_9EH        <=B
> +  Kabylake       06_9EH        0xC only if TSX is enabled
> +  Kabylake       06_9EH        0xD only if TSX is enabled
> +  =============  ============  ========

> +  =============  ============  ===========================

ditto at the top of the table.

> +
> +Related CVEs
> +------------
> +
> +The following CVE entry is related to this SRBDS issue:
> +
> +    ==============  =====  ===================================================
> +    CVE-2020--0543
> +    ==============  =====  ===================================================

Is the void filled at some point?

> +Attack scenarios
> +---------------
> +An unprivileged user can extract returned values from RDRAND and RDSEED
> +executed on another core or sibling thread using MDS techniques.

Lacks EGETKEY again.

> +
> +
> +Mitigtion mechanism
> +-------------------
> +Intel will release microcode updates that modify the RDRAND, RDSEED, and
> +EGETKEY instructions to overwrite secret special register data in the shared
> +staging buffer before the secret data can be accessed by another logical
> +processor.
> +
> +During execution of the RDRAND, RDSEED, or EGETKEY instruction, off-core
> +accesses from other logical processors will be delayed until the special
> +register read is complete and the secret data in the shared staging buffer is
> +overwritten.
> +
> +This has three effects on performance:
> +1 RDRAND, RDSEED, or EGETKEY instruction have higher latency.
> +2 Executing RDRAND at the same time on multiple logical processors will be
> +  serialized, resulting in an overall reduction in the maximum RDRAND bandwidth.
> +3 Executing RDRAND, RDSEED or EGETKEY will delay memory accesses from other
> +  logical processors that miss their core caches, with an impact similar to
> +  legacy locked cache-line-split accesses.

I seriously doubt that you ever looked at the build output of this.

> +
> +Because of the performance impact of the mitigation, the microcode updates also
> +provide an opt-out mechanism (RNGDS_MITG_DIS) to disable the mitigation for
> +RDRAND and RDSEED instructions executed outside of Intel Software Guard
> +Extensions (Intel SGX) enclaves. On logical processors that disable the
> +mitigation using this opt-out mechanism, RDRAND and RDSEED do not take longer
> +to execute and do not impact performance of sibling logical processors memory
> +accesses. The opt-out mechanism does not affect Intel SGX enclaves (including
> +execution of RDRAND or RDSEED inside of an enclave, as well as EGETKEY
> +execution).
> +
> +IA32_MCU_OPT_CTRL MSR Definition
> +--------------------------------
> +Along with the mitigation for this issue, Intel added a new thread-scope
> +IA32_MCU_OPT_CTRL MSR, (address 0x123). The presence of this MSR and
> +RNGDS_MITG_DIS (bit 0) is enumerated by CPUID.(EAX=07H,ECX=0).EDX[SRBDS_CTRL =
> +9]==1. This MSR is introduced through the microcode update.
> +
> +Setting IA32_MCU_OPT_CTRL[0] (RNGDS_MITG_DIS) to 1 for a logical processor
> +disables the mitigation for RDRAND and RDSEED executed outside of an Intel SGX
> +enclave on that logical processor. Opting out of the mitigation for a
> +particular logical processor does not affect the RDRAND and RDSEED mitigations
> +for other logical processors.
> +
> +Note that inside of an Intel SGX enclave, the mitigation is applied regardless
> +of the value of RNGDS_MITG_DS.
> +
> +Mitigation control on the kernel command line
> +---------------------------------------------
> +The kernel command line allows for control over the SRBDS mitigation
> at boot

allows control (methinks)

> +time with the option "srbds=".  The option for this is:
> +
> +  ============= ============================================================
> +  off           This option disables SRBS mitigation for RDRAND and RDSEED on
> +                affected platforms.
> +  ============= ============================================================
> +
> +SRBS System Information
> +-----------------------
> +The Linux kernel provides vulnerability status information through sysfs.  For
> +SRBDS this can be accessed by the following sysfs file:
> +/sys/devices/system/cpu/vulnerabilities/special_register_data_sampling
> +
> +The possible values contained in this file are:
> +
> + ============================== ===========================================
> + Not affected                   Processor not vulnerable
> + Vulnerable                     Processor vulnerable and mitigation disabled
> + Vulnerable: no microcode       Processor vulnerable and microcode is missing
> +                                mitigation
> + Mitigated                      Processor is vulnerable and mitigation is in
> +                                effect.
> + Not affected (TSX disabled)    Processor is only vulnerable when TSX is
> +                                enabled while this system was booted with TSX
> +                                disabled.
> + Unknown                        Running on virtual guest processor that is
> +                                affected but with no way to know if host
> +                                processor is mitigated or vulnerable.
> + ============================== ===========================================
> +
> +Default mitigations
> +-------------------
> +This new microcode serializes processor access during execution of RDRAND,
> +RDSEED ensures that the shared buffer is overwritten before it is released for
> +reuse.

Errm. What has this to do with the default chosen by the kernel?

Thanks,

        tglx