As if AMD wasn't already having a tough time in the CPU market, as the firm struggles to promote its Bulldozer-based FX CPUs in light of Intel's latest offerings, news recently surfaced that a bug has been discovered in some of the firm's families of processors.
Discovered by Matt Dillion, member of the DragonFly BSD team, the bug was reproduced by creating a test-case that, "through a very specific sequence of consecutive back-to-back pops and (near) return instructions, can create a condition where the process incorrectly updates the stack pointer," leading to a segmentation fault.
Whilst it's not openly known how many processor families are affected, AMD is expected to update documentation errata and so all shall be revealed in due course, however the bug has been reproduced on a 4-way Opteron 6168 and a Phenom II x4 820. Matt himself stated that the bug was so difficult to reproduce that he had to send a test-case USB image over to AMD and so it seems unlikely that this bug would have any significant impact on the stability of a production system, that and, if it did, the bug would likely have been spotted much sooner.
In DragonFly, however, the bug manifested as random segmentation faults under heavy load in a single program 'cc1', that before creation of a test case that was able to reproduce the bug in under 60 seconds, took anywhere up to two days to trigger with a 48-core system under heavy load.
At the very least, the bug results in a crash as opposed to the CPU presenting incorrect data to a program, ensuring the integrity of the data processed. For AMD, this news is just another round of bad PR.
