aboutsummaryrefslogtreecommitdiff
path: root/sys/arm64
diff options
context:
space:
mode:
authorAndrew Turner <andrew@FreeBSD.org>2016-09-13 14:10:49 +0000
committerAndrew Turner <andrew@FreeBSD.org>2016-09-13 14:10:49 +0000
commit8b520540637e721fae1d447edc8312d9e9ebc623 (patch)
tree040da1457cde8de5d414731ec149f8c27f365ba4 /sys/arm64
parent5e61a9688275014158adccc72da159689e1dc3cb (diff)
downloadsrc-8b520540637e721fae1d447edc8312d9e9ebc623.tar.gz
src-8b520540637e721fae1d447edc8312d9e9ebc623.zip
Add a warning about a known erratum we have observed on ThunderX pass 1.1.
As this is evaluation hardware with only a few users, and there is a lack of information add a warning when booting on this hardware. Reported by: cognet Obtained from: ABT Systems Ltd MFC after: Instant Sponsored by: The FreeBSD Foundation
Notes
Notes: svn path=/head/; revision=305767
Diffstat (limited to 'sys/arm64')
-rw-r--r--sys/arm64/arm64/identcpu.c22
1 files changed, 22 insertions, 0 deletions
diff --git a/sys/arm64/arm64/identcpu.c b/sys/arm64/arm64/identcpu.c
index 70a2d84183f2..c047c5412db1 100644
--- a/sys/arm64/arm64/identcpu.c
+++ b/sys/arm64/arm64/identcpu.c
@@ -179,6 +179,28 @@ print_cpu_features(u_int cpu)
}
printf("\n");
+ /*
+ * There is a hardware errata where, if one CPU is performing a TLB
+ * invalidation while another is performing a store-exclusive the
+ * store-exclusive may return the wrong status. A workaround seems
+ * to be to use an IPI to invalidate on each CPU, however given the
+ * limited number of affected units (pass 1.1 is the evaluation
+ * hardware revision), and the lack of information from Cavium
+ * this has not been implemented.
+ *
+ * At the time of writing this the only information is from:
+ * https://lkml.org/lkml/2016/8/4/722
+ */
+ /*
+ * XXX: CPU_MATCH_ERRATA_CAVIUM_THUNDER_1_1 on it's own also
+ * triggers on pass 2.0+.
+ */
+ if (cpu == 0 && CPU_VAR(PCPU_GET(midr)) == 0 &&
+ CPU_MATCH_ERRATA_CAVIUM_THUNDER_1_1)
+ printf("WARNING: ThunderX Pass 1.1 detected.\nThis has known "
+ "hardware bugs that may cause the incorrect operation of "
+ "atomic operations.\n");
+
if (cpu != 0 && cpu_print_regs == 0)
return;