Re:Re: [f2fs-dev] [PATCH v5 00/16] f2fs: support & optimize large folios for writable files

[Nanzhe Zhao]

Hi Jaeguk,
  
  I am trying to reproduce the failure in your log: sd 0:0:0:4: Retrying unaligned write for zone 2639 → critical target error → sde:
  zone 2639: Aborting plugged BIOs → f2fs_stop_checkpoint (kernel 6.12.81-android16-6) and got two two questions.
  
  1. Kernel baseline. Could you confirm that the kernel you tested has Bart Van Assche's zoned write pipelining, i.e. multiple in-flight
  sequential writes per zone, with the order kept by BLK_FEAT_ORDERED_HWQ and the per-zone from_cpu in block/blk-zoned.c? And that it
  does not include these three follow-up fixes:
  
  - "ANDROID: block: Preserve the submission order of zoned writes"
  - "ANDROID: block: Preserve the order of zoned REQ_FUA writes"
  - "ANDROID: block: Restore the write pointer offset check"
  
  In AOSP they are 050de570f, c616f7c1a, d50601d0f on android16-6.12, and 6068ed9f0, d786f8609, f08430f80 on android17-6.18. The
  simplest check is whether REQ_ZONE_SERIALIZED and work_busy(&zwplug->bio_work) appear in your block/blk-zoned.c. The output of git log
  --oneline -- block/blk-zoned.c would also do.
  
  2. Test workload. If that is the case, could you share the test workload that triggers it?  So far I have not been able to reproduce it using null_blk to simulate 
zone devices with qemu and android17-6.18 gki kernel. Any suggestions would be much appreciated.
Thanks,