Hello, Debian ARM port team I have a suspicion regarding bug #914124 (https://bugs.debian.org/cgi-bin/bugreport.cgi?bug=914124) in bind9
It seems like between function calls register-passed parameter to qsort() n=4 have been changed into garbage. According to the coredump, dissassembly and stack unwind, parameter was passed in x25 register, then it was stored on the stack frame and read back as n=5 which caused qsort to call comparison function on 5th element of 4-element array. BIND9's assertions had stopped program abnormally. Currently I am running BIND9 on ARM64 (aarch64). The board I use is Orange Pi PC2, and it is based on Allwinner H5 SoC which is Cortex A53 (r0p4). As far as I know, it is susceptible to the CatA silicon erratas: 835769 and 843419. My question is: could the errata trigger such sequence? Could it be just a memory corruption due to the onboard DDR failure. P.S. Sometimes BIND9 crashes with SEGV on null pointer dereference. My other question is: is debian aarch64 port compiled with proper errata mitigation CFLAGS and LDFLAGS? -- Philippe Duke Network software engineer System-level developer NetAssist LLC Ukraine Khreshchatyk Street, 10B, office 8 AS29632 http://netassist.ua Our GitHub Repository: https://github.com/netassist-ua
