[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

[xen stable-4.16] x86/tsx: Cope with TSX deprecation on WHL-R/CFL-R



commit 0c46d108b74bd38e0887cdd02cb82dd1084586cb
Author:     Andrew Cooper <andrew.cooper3@xxxxxxxxxx>
AuthorDate: Wed Sep 16 16:15:52 2020 +0100
Commit:     Andrew Cooper <andrew.cooper3@xxxxxxxxxx>
CommitDate: Tue Feb 8 18:01:32 2022 +0000

    x86/tsx: Cope with TSX deprecation on WHL-R/CFL-R
    
    The February 2022 microcode is formally de-featuring TSX on the TAA-impacted
    client CPUs.  The backup TAA mitigation (VERW regaining its flushing side
    effect) is being dropped, meaning that `smt=0 spec-ctrl=md-clear` no longer
    protects against TAA on these parts.
    
    The new functionality enumerates itself via the RTM_ALWAYS_ABORT CPUID
    bit (the same as June 2021), but has its control in MSR_MCU_OPT_CTRL as
    opposed to MSR_TSX_FORCE_ABORT.
    
    TSX now defaults to being disabled on ucode load.  Furthermore, if SGX is
    enabled in the BIOS, TSX is locked and cannot be re-enabled.  In this case,
    override opt_tsx to 0, so the RTM/HLE CPUID bits get hidden by default.
    
    While updating the command line documentation, take the opportunity to add a
    paragraph explaining what TSX being disabled actually means, and how 
migration
    compatibility works.
    
    Signed-off-by: Andrew Cooper <andrew.cooper3@xxxxxxxxxx>
    Reviewed-by: Jan Beulich <jbeulich@xxxxxxxx>
    (cherry picked from commit ad9f7c3b2e0df38ad6d54f4769d4dccf765fbcee)
---
 docs/misc/xen-command-line.pandoc | 25 ++++++++++---
 xen/arch/x86/spec_ctrl.c          |  7 ++--
 xen/arch/x86/tsx.c                | 76 +++++++++++++++++++++++++++++++++++++++
 xen/include/asm-x86/msr-index.h   |  2 ++
 4 files changed, 103 insertions(+), 7 deletions(-)

diff --git a/docs/misc/xen-command-line.pandoc 
b/docs/misc/xen-command-line.pandoc
index f7797ea233..995197f4b2 100644
--- a/docs/misc/xen-command-line.pandoc
+++ b/docs/misc/xen-command-line.pandoc
@@ -2371,7 +2371,9 @@ Several microcode updates are relevant:
    Introduced MSR_TSX_CTRL on all TSX-enabled MDS_NO parts to date,
    CLX/WHL-R/CFL-R, with the controls becoming architectural moving forward
    and formally retiring HLE from the architecture.  The user can disable TSX
-   to mitigate TAA, and elect to hide the HLE/RTM CPUID bits.
+   to mitigate TAA, and elect to hide the HLE/RTM CPUID bits.  Also causes
+   VERW to once-again flush the microarchiectural buffers in case a TAA
+   mitigation is wanted along with TSX being enabled.
 
  * June 2021, removing the workaround for March 2019 on client CPUs and
    formally de-featured TSX on SKL/KBL/WHL/CFL (Note: SKX still retains the
@@ -2379,19 +2381,32 @@ Several microcode updates are relevant:
    PCR3 works fine, and TSX is disabled by default, but the user can re-enable
    TSX at their own risk, accepting that the memory order erratum is unfixed.
 
+ * February 2022, removing the VERW flushing workaround from November 2019 on
+   client CPUs and formally de-featuring TSX on WHL-R/CFL-R (Note: CLX still
+   retains the VERW flushing workaround).  TSX defaults to disabled, and is
+   locked off when SGX is enabled in the BIOS.  When SGX is not enabled, TSX
+   can be re-enabled at the users own risk, as it reintroduces the TSX Async
+   Abort speculative vulnerability.
+
 On systems with the ability to configure TSX, this boolean offers system wide
 control of whether TSX is enabled or disabled.
 
+When TSX is disabled, transactions unconditionally abort.  This is compatible
+with the TSX spec, which requires software to have a non-transactional path as
+a fallback.  The RTM and HLE CPUID bits are hidden from VMs by default, but
+can be re-enabled if required.  This allows VMs which previously saw RTM/HLE
+to be migrated in, although any TSX-enabled software will run with reduced
+performance.
+
+ * When TSX is locked off by firmware, `tsx=` is ignored and treated as
+   `false`.
+
  * An explicit `tsx=` choice is honoured, even if it is `true` and would
    result in a vulnerable system.
 
  * When no explicit `tsx=` choice is given, parts vulnerable to TAA will be
    mitigated by disabling TSX, as this is the lowest overhead option.
 
-   If the use of TSX is important, the more expensive TAA mitigations can be
-   opted in to with `smt=0 spec-ctrl=md-clear`, at which point TSX will remain
-   active by default.
-
  * When no explicit `tsx=` option is given, parts susceptible to the memory
    ordering errata default to `true` to enable working TSX.  Alternatively,
    selecting `tsx=0` will disable TSX and restore PCR3 to a working state.
diff --git a/xen/arch/x86/spec_ctrl.c b/xen/arch/x86/spec_ctrl.c
index 3628b4b415..2b93468d39 100644
--- a/xen/arch/x86/spec_ctrl.c
+++ b/xen/arch/x86/spec_ctrl.c
@@ -1233,11 +1233,14 @@ void __init init_speculation_mitigations(void)
      * the MDS mitigation of disabling HT and using VERW flushing.
      *
      * On CPUs which advertise MDS_NO, VERW has no flushing side effect until
-     * the TSX_CTRL microcode is loaded, despite the MD_CLEAR CPUID bit being
+     * the TSX_CTRL microcode (Nov 2019), despite the MD_CLEAR CPUID bit being
      * advertised, and there isn't a MD_CLEAR_2 flag to use...
      *
+     * Furthermore, the VERW flushing side effect is removed again on client
+     * parts with the Feb 2022 microcode.
+     *
      * If we're on affected hardware, able to do something about it (which
-     * implies that VERW now works), no explicit TSX choice and traditional
+     * implies that VERW might work), no explicit TSX choice and traditional
      * MDS mitigations (no-SMT, VERW) not obviosuly in use (someone might
      * plausibly value TSX higher than Hyperthreading...), disable TSX to
      * mitigate TAA.
diff --git a/xen/arch/x86/tsx.c b/xen/arch/x86/tsx.c
index c3b8a7ec00..be89741a2f 100644
--- a/xen/arch/x86/tsx.c
+++ b/xen/arch/x86/tsx.c
@@ -14,6 +14,9 @@
  * This is arranged such that the bottom bit encodes whether TSX is actually
  * disabled, while identifying various explicit (>=0) and implicit (<0)
  * conditions.
+ *
+ * This option only has any effect on systems presenting a mechanism of
+ * controlling TSX behaviour, and where TSX isn't force-disabled by firmware.
  */
 int8_t __read_mostly opt_tsx = -1;
 int8_t __read_mostly cpu_has_tsx_ctrl = -1;
@@ -54,6 +57,66 @@ void tsx_init(void)
         cpu_has_tsx_ctrl = !!(caps & ARCH_CAPS_TSX_CTRL);
         has_rtm_always_abort = cpu_has_rtm_always_abort;
 
+        if ( cpu_has_tsx_ctrl && cpu_has_srbds_ctrl )
+        {
+            /*
+             * On a TAA-vulnerable or later part with at least the May 2020
+             * microcode mitigating SRBDS.
+             */
+            uint64_t val;
+
+            rdmsrl(MSR_MCU_OPT_CTRL, val);
+
+            /*
+             * Probe for the February 2022 microcode which de-features TSX on
+             * TAA-vulnerable client parts - WHL-R/CFL-R.
+             *
+             * RTM_ALWAYS_ABORT (read above) enumerates the new functionality,
+             * but is read as zero if MCU_OPT_CTRL.RTM_ALLOW has been set
+             * before we run.  Undo this.
+             */
+            if ( val & MCU_OPT_CTRL_RTM_ALLOW )
+                has_rtm_always_abort = true;
+
+            if ( has_rtm_always_abort )
+            {
+                if ( val & MCU_OPT_CTRL_RTM_LOCKED )
+                {
+                    /*
+                     * If RTM_LOCKED is set, TSX is disabled because SGX is
+                     * enabled, and there is nothing we can do.  Override with
+                     * tsx=0 so all other logic takes sensible actions.
+                     */
+                    printk(XENLOG_WARNING "TSX locked by firmware - 
disabling\n");
+                    opt_tsx = 0;
+                }
+                else
+                {
+                    /*
+                     * Otherwise, set RTM_ALLOW.  Not because we necessarily
+                     * intend to enable RTM, but it prevents
+                     * MSR_TSX_CTRL.RTM_DISABLE from being ignored, thus
+                     * allowing the rest of the TSX selection logic to work as
+                     * before.
+                     */
+                    val |= MCU_OPT_CTRL_RTM_ALLOW;
+                }
+
+                set_in_mcu_opt_ctrl(
+                    MCU_OPT_CTRL_RTM_LOCKED | MCU_OPT_CTRL_RTM_ALLOW, val);
+
+                /*
+                 * If no explicit tsx= option is provided, pick a default.
+                 *
+                 * With RTM_ALWAYS_ABORT, the default ucode behaviour is to
+                 * disable, so match that.  This does not override explicit 
user
+                 * choices, or implicit choices as a side effect of 
spec-ctrl=0.
+                 */
+                if ( opt_tsx == -1 )
+                    opt_tsx = 0;
+            }
+        }
+
         if ( cpu_has_tsx_force_abort )
         {
             /*
@@ -142,6 +205,19 @@ void tsx_init(void)
      */
     if ( cpu_has_tsx_ctrl )
     {
+        /*
+         * On a TAA-vulnerable part with at least the November 2019 microcode,
+         * or newer part with TAA fixed.
+         *
+         * Notes:
+         *  - With the February 2022 microcode, if SGX has caused TSX to be
+         *    locked off, opt_tsx is overridden to 0.  TSX_CTRL.RTM_DISABLE is
+         *    an ignored bit, but we write it such that it matches the
+         *    behaviour enforced by microcode.
+         *  - Otherwise, if SGX isn't enabled and TSX is available to be
+         *    controlled, we have or will set MSR_MCU_OPT_CTRL.RTM_ALLOW to
+         *    let TSX_CTRL.RTM_DISABLE be usable.
+         */
         uint32_t hi, lo;
 
         rdmsr(MSR_TSX_CTRL, lo, hi);
diff --git a/xen/include/asm-x86/msr-index.h b/xen/include/asm-x86/msr-index.h
index ab68ef2681..9df1959fe5 100644
--- a/xen/include/asm-x86/msr-index.h
+++ b/xen/include/asm-x86/msr-index.h
@@ -78,6 +78,8 @@
 
 #define MSR_MCU_OPT_CTRL                    0x00000123
 #define  MCU_OPT_CTRL_RNGDS_MITG_DIS        (_AC(1, ULL) <<  0)
+#define  MCU_OPT_CTRL_RTM_ALLOW             (_AC(1, ULL) <<  1)
+#define  MCU_OPT_CTRL_RTM_LOCKED            (_AC(1, ULL) <<  2)
 
 #define MSR_RTIT_OUTPUT_BASE                0x00000560
 #define MSR_RTIT_OUTPUT_MASK                0x00000561
--
generated by git-patchbot for /home/xen/git/xen.git#stable-4.16



 


Rackspace

Lists.xenproject.org is hosted with RackSpace, monitoring our
servers 24x7x365 and backed by RackSpace's Fanatical Support®.