Hi Brian, list,
Thanks for bringing this up on the mailing list.
On Wed, Apr 24, 2024 at 9:16 PM Brian Norris briannorris@chromium.org wrote:
Background: https://review.coreboot.org/c/flashrom/+/80807
A long time ago, in the pre-git times [1], flashrom added a 1 second delay to all verification, and claimed that some chips "need some time to calm down." The commit message claims it "fixes a few reports where verify directly after erase had unpleasant side effects like corrupting flash or at least getting incorrect verify results." It provides no details of what systems, chips, or programmers were involved.
Back then, SPI flash chips weren't as ubiquitous as they currently are. This workaround is most likely for Parallel/LPC/FWH flash chips, which can actually be quite weird.
This delay remains in the --verify path today, and IMO, it's a big waste of time. If there are truly chips that are, say, deasserting the BUSY line before they're truly finished programming ... well, we should track those chips down and add targeted quirk flags for them. We shouldn't penalize all flashrom users in all cases for all time.
I agree that this delay shouldn't be unconditional.
Personally, I highly doubt that this delay is still relevant today -- there may have been a bug in some programmer that has since been fixed; there may have been some malfunctioning system that is no longer in use; ... or it's possible this is still hiding a real buggy chip somewhere out there.
If we don't know for sure whether this delay is no longer needed, I wouldn't risk re-introducing issues in flashrom.
I propose: we still remove the delay. There's plenty of description in the above Gerrit link about why, and how we can handle regressions. (For one, it's relatively simple to split up a --verify operation into its constituent --write/sleep/--read operations, for debugging.)
The request:
- Tests: I've tested a few Chromebooks, but imagine this had to have
been some more esoteric system. Extra testing is welcome.
Do Chromebooks use non-SPI flash chips? They probably don't.
- Thoughts: does my proposal make sense? Am I missing something obvious?
Did you conside making it so that only SPI programmers / flash chips skip the delay? The SPI bus' strict timing leaves no room for this problem to occur, so it should be safe to skip this delay. And this would keep non-SPI unaffected (which is most likely what needs this extra delay).
- Awareness: if nothing else, this email may serve to highlight in
case we land the above patch and later hear back that there are some sketchy --verify reports from users.
I agree that awareness is important, but bear in mind that external programmers for non-SPI flash chips are increasingly rare, so recovering from a brick can be quite convoluted (e.g. having to desolder a TSOP32 flash chip to reflash it somehow).
Thanks, Brian
[1] The svn->git commit in question:
commit 8ab49e72af8465d4527de2ec37b22cd44f7a1169 Date: Wed Aug 19 13:55:34 2009 +0000 Disallow erase/write for known bad chips so people won't try without a clear understanding _______________________________________________ flashrom mailing list -- flashrom@flashrom.org To unsubscribe send an email to flashrom-leave@flashrom.org
Best regards, Angel