AMD Ryzen with AGESA microcode - FMA3 instruction sequence hang: A long series of FMA3 instructions hangs the system

Impact

A long series of FMA3 instructions hangs the system. FMA3 is the fused multiply-add path that every dense linear-algebra kernel hammers, so on an ML host this is not an exotic instruction sequence - it is the normal workload. An unprivileged tenant running a numerical benchmark can hang the node, deliberately or by accident.

Who can reach it

Local, unprivileged. Reachable from any container running numerical code.

What to do

Fixed by an AGESA microcode update from 2017 onward, delivered as an OEM BIOS package. Client Ryzen silicon rather than EPYC, so on a server fleet this is mostly a non-issue - verify your SKUs before spending a window on it.

References

Related entries

• Intel processors (speculative store bypass): Spectre v4: a load speculatively executes before an older storeCVE-2018-3639 · Intel processors (speculative store bypass)Medium
• Intel SGX Platform Software for Linux (AESM daemon): A local attacker can disable the AESM daemonCVE-2018-3689 · Intel SGX Platform Software for Linux (AESM daemon)Medium
• Intel SGX driver for Linux: Insufficient input validation in the out-of-tree SGX Linux driver lets a localCVE-2019-0157 · Intel SGX driver for LinuxMedium
• Intel PTT / fTPM (ECDSA and ECSchnorr timing): The firmware TPM's signing operation leaks nonce information throughCVE-2019-11090 · Intel PTT / fTPM (ECDSA and ECSchnorr timing)Medium
• Linux bnxt_re RoCE driver (bnxt_re_create_srq memory leak): A tenant can exhaust host memory by repeatedly triggeringCVE-2019-19077 · Linux bnxt_re RoCE driver (bnxt_re_create_srq memory leak)Medium
• Intel processors (vector register sampling): Stale values left in vector registers can be sampled by other contextsCVE-2020-0548 · Intel processors (vector register sampling)Medium