microSD card "partition not aligned" warning

I

Ignacio

I have 2 identical arm devices running FreeBSD 11.4 from a 64 GB microSD card that is mounted read-only most of the time, whenever I mount it read-write and then back to read-only with these commands:
Code: 
mount -u -w /
mount -u -r /
It will take between 10 and 40 seconds to mount it read-only again, freezing disk access during that time, and in dmesg it appears these messages:
Code: 
Mar 27 07:56:00 bl12-247-99 kernel: GEOM_PART: partition 1 on (diskid/DISK-E89ADB97, MBR) is not aligned on 4194304 bytes
Mar 27 07:56:00 bl12-247-99 kernel: GEOM_PART: partition 2 on (diskid/DISK-E89ADB97, MBR) is not aligned on 4194304 bytes
Mar 27 07:56:00 bl12-247-99 kernel: GEOM_PART: partition 1 on (ufsid/5e37468114d819f1, BSD) is not aligned on 4194304 bytes
Mar 27 07:56:00 bl12-247-99 kernel: GEOM_PART: partition 1 on (diskid/DISK-E89ADB97s2, BSD) is not aligned on 4194304 bytes
The partitions of both cards look like this:
Code: 
# gpart show mmcsd0
=>       63  124735425  mmcsd0  MBR  (59G)
         63       2016          - free -  (1.0M)
       2079      65520       1  !12  [active]  (32M)
      67599       1009          - free -  (505K)
      68608  121533440       2  freebsd  (58G)
  121602048    3133440          - free -  (1.5G)
Is this a driver related problem? or maybe I should partition the cards in a different way to keep partitions aligned somewhat?
 
I made a custom image with crochet, then copied it with dd to the microSD card and in the first boot it made a growfs to use all the space available, maybe there is some crochet config that I overlooked.

Anyway, how can you tell that those numbers are not 4K boundaries? I understand that they should be a multiple of 8 as we are talking about 512 bytes sectors, isn't it? so 2079 wouldn't do but 2080 would do? but then 68608 is a multiple of 8.
 
Ignacio said:
I understand that they should be a multiple of 8 as we are talking about 512 bytes sectors, isn't it? so 2079 wouldn't do but 2080 would do?
Click to expand...
Correct.
Ignacio said:
but then 68608 is a multiple of 8.
Click to expand...
It's a slice, there are partitions inside that slice. The partitions inside the slice aren't aligned.

If I recall correctly gpart(8) already uses 4K boundaries by default but you can always add -a 4K to make sure it does.
 
VladiBG said:
What is the sector size of your SD card?
Click to expand...
I understand it's 512 bytes:
Code: 
# diskinfo -c mmcsd0
mmcsd0
        512             # sectorsize
        63864569856     # mediasize in bytes (59G)
        124735488       # mediasize in sectors
        4194304         # stripesize
        0               # stripeoffset
        SDHC SB64G 8.0 SN E89ADB97 MFG 05/2019 by 3 SD  # Disk descr.
        E89ADB97        # Disk ident.
        Yes             # TRIM/UNMAP support
        0               # Rotation rate in RPM

I/O command overhead:
        time to read 10MB block      1.737557 sec       =    0.085 msec/sector
        time to read 20480 sectors   7.293083 sec       =    0.356 msec/sector
        calculated command overhead                     =    0.271 msec/sector
Almost 2 seconds to read 10MB looks very poor, it may be caused by the partitions misalignment, isn't it? it could also make the card fail much sooner?
 
Your SD card is with Stripesize of 4K and your partition is not aligned.
68608 / 4096 = 16.75

The read speed is against the entire card, so it may be worst when you read from not aligned partition.
 
You must log in or register to reply here.
Back
Top