backport protection against userspace-userspace spectreRSB
authorThomas Lamprecht <t.lamprecht@proxmox.com>
Thu, 30 Aug 2018 11:02:23 +0000 (13:02 +0200)
committerThomas Lamprecht <t.lamprecht@proxmox.com>
Thu, 30 Aug 2018 11:05:06 +0000 (13:05 +0200)
Signed-off-by: Thomas Lamprecht <t.lamprecht@proxmox.com>
patches/kernel/0009-x86-speculation-Protect-against-userspace-userspace-.patch [new file with mode: 0644]

diff --git a/patches/kernel/0009-x86-speculation-Protect-against-userspace-userspace-.patch b/patches/kernel/0009-x86-speculation-Protect-against-userspace-userspace-.patch
new file mode 100644 (file)
index 0000000..4202dd6
--- /dev/null
@@ -0,0 +1,95 @@
+From 0000000000000000000000000000000000000000 Mon Sep 17 00:00:00 2001
+From: Jiri Kosina <jkosina@suse.cz>
+Date: Wed, 29 Aug 2018 19:17:40 +0000
+Subject: [PATCH] x86/speculation: Protect against userspace-userspace
+ spectreRSB
+
+The article "Spectre Returns! Speculation Attacks using the Return Stack
+Buffer" [1] describes two new (sub-)variants of spectrev2-like attacks,
+making use solely of the RSB contents even on CPUs that don't fallback to
+BTB on RSB underflow (Skylake+).
+
+Mitigate userspace-userspace attacks by always unconditionally filling RSB on
+context switch when the generic spectrev2 mitigation has been enabled.
+
+[1] https://arxiv.org/pdf/1807.07940.pdf
+
+Signed-off-by: Jiri Kosina <jkosina@suse.cz>
+Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
+Reviewed-by: Josh Poimboeuf <jpoimboe@redhat.com>
+Acked-by: Tim Chen <tim.c.chen@linux.intel.com>
+Cc: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
+Cc: Borislav Petkov <bp@suse.de>
+Cc: David Woodhouse <dwmw@amazon.co.uk>
+Cc: Peter Zijlstra <peterz@infradead.org>
+Cc: Linus Torvalds <torvalds@linux-foundation.org>
+Cc: stable@vger.kernel.org
+Link: https://lkml.kernel.org/r/nycvar.YFH.7.76.1807261308190.997@cbobk.fhfr.pm
+
+CVE-2017-5715 (SpectreRSB sub-variant)
+
+(cherry picked from commit fdf82a7856b32d905c39afc85e34364491e46346)
+Signed-off-by: Tyler Hicks <tyhicks@canonical.com>
+Signed-off-by: Thomas Lamprecht <t.lamprecht@proxmox.com>
+---
+ arch/x86/kernel/cpu/bugs.c | 38 +++++++-------------------------------
+ 1 file changed, 7 insertions(+), 31 deletions(-)
+
+diff --git a/arch/x86/kernel/cpu/bugs.c b/arch/x86/kernel/cpu/bugs.c
+index 700b4c0a93a2..edfc64a8a154 100644
+--- a/arch/x86/kernel/cpu/bugs.c
++++ b/arch/x86/kernel/cpu/bugs.c
+@@ -322,23 +322,6 @@ static enum spectre_v2_mitigation_cmd __init spectre_v2_parse_cmdline(void)
+       return cmd;
+ }
+-/* Check for Skylake-like CPUs (for RSB handling) */
+-static bool __init is_skylake_era(void)
+-{
+-      if (boot_cpu_data.x86_vendor == X86_VENDOR_INTEL &&
+-          boot_cpu_data.x86 == 6) {
+-              switch (boot_cpu_data.x86_model) {
+-              case INTEL_FAM6_SKYLAKE_MOBILE:
+-              case INTEL_FAM6_SKYLAKE_DESKTOP:
+-              case INTEL_FAM6_SKYLAKE_X:
+-              case INTEL_FAM6_KABYLAKE_MOBILE:
+-              case INTEL_FAM6_KABYLAKE_DESKTOP:
+-                      return true;
+-              }
+-      }
+-      return false;
+-}
+-
+ static void __init spectre_v2_select_mitigation(void)
+ {
+       enum spectre_v2_mitigation_cmd cmd = spectre_v2_parse_cmdline();
+@@ -399,22 +382,15 @@ static void __init spectre_v2_select_mitigation(void)
+       pr_info("%s\n", spectre_v2_strings[mode]);
+       /*
+-       * If neither SMEP nor PTI are available, there is a risk of
+-       * hitting userspace addresses in the RSB after a context switch
+-       * from a shallow call stack to a deeper one. To prevent this fill
+-       * the entire RSB, even when using IBRS.
++       * If spectre v2 protection has been enabled, unconditionally fill
++       * RSB during a context switch; this protects against two independent
++       * issues:
+        *
+-       * Skylake era CPUs have a separate issue with *underflow* of the
+-       * RSB, when they will predict 'ret' targets from the generic BTB.
+-       * The proper mitigation for this is IBRS. If IBRS is not supported
+-       * or deactivated in favour of retpolines the RSB fill on context
+-       * switch is required.
++       *      - RSB underflow (and switch to BTB) on Skylake+
++       *      - SpectreRSB variant of spectre v2 on X86_BUG_SPECTRE_V2 CPUs
+        */
+-      if ((!boot_cpu_has(X86_FEATURE_PTI) &&
+-           !boot_cpu_has(X86_FEATURE_SMEP)) || is_skylake_era()) {
+-              setup_force_cpu_cap(X86_FEATURE_RSB_CTXSW);
+-              pr_info("Spectre v2 mitigation: Filling RSB on context switch\n");
+-      }
++      setup_force_cpu_cap(X86_FEATURE_RSB_CTXSW);
++      pr_info("Spectre v2 / SpectreRSB mitigation: Filling RSB on context switch\n");
+       /* Initialize Indirect Branch Prediction Barrier if supported */
+       if (boot_cpu_has(X86_FEATURE_IBPB)) {