Re: SSD trim crash
- Reply: Chris Stephan : "Re: SSD trim crash"
- In reply to: Lee MATTHEWS via freebsd-drivers : "SSD trim crash"
- Go to: [ bottom of page ] [ top of archives ] [ this month ]
Date: Mon, 03 Jan 2022 16:15:59 UTC
You can set the trim method to nothing w/o that patch.
kern.cam.ada.X.delete_method=DISABLE
There's also a mechanism through quirks that can disable known
troublemakers, though to date we've only disable those drives that cause
data corruption with TRIMs.
FreeBSD makes heavy use of TRIMs at times to improve the info the drive's
firmware has on what's actually in use.
Some drives love this and their write amplification is reduced (so
performance and longevity increased). Other drives,
as you've discovered, implement TRIMs in a sub-optimal manner. I've seen
too many TRIMs cause drives to hang,
but it's never been as severe as you are seeing (I have seen it on reboot
once or twice, but my usual heartache
with TRIMs has been just really bad performance). Older versions of FreeBSD
didn't have TRIM collapsing in any
real sense, so would send lots and lots of TRIMs to drives which often
times would lead to poor performance. UFS
and ada have both been enhanced to reduce that in newer versions of FreeBSD.
FreeBSD 11.3 is out of support. You may have slightly better luck with a
newer FreeBSD in general (there's
more TRIM management in newer FreeBSDs, especially 13.0 and newer). There's
a chance that the improvements
will help, but if the drives firmware is bad enough, disabling trim on the
models of drives where there are problems
may be your only option.
Warner
On Mon, Jan 3, 2022 at 9:01 AM Lee MATTHEWS via freebsd-drivers <
freebsd-drivers@freebsd.org> wrote:
> Hello,
>
> I have been tasked with working on an intermittent problem that has been
> affecting some of our products. During in-house testing, we observe that
> occasionally our system crashes and reboots. After reboot, the system gets
> blocked in the BIOS and it does not advance any further because the BIOS
> does not detect the SSD disk. The crash appears to happen randomly during
> testing, as it can take anything from several hours of continuous testing
> to a week of running without stop for the crash to occur. The only way to
> get the system back up and running again is to power down and then up the
> machine, at which point the disk once again becomes visible to the BIOS.
>
> Our products are based on FreeBSD version 11.3.
>
> The crash appears to be linked to the type of disk that is being used. We
> have one brand of disk that has never had a problem (even after a month of
> continuous testing) and others that, after testing, fail and after reboot
> get blocked in the BIOS.
>
>
> Logging via the serial console, I have observed the following crash :
>
> --
> [2021-12-10 20:56:10] ahcich1: Timeout on slot 5 port 0
> [2021-12-10 20:56:10] ahcich1: is 00000000 cs 00000020 ss 00000000 rs
> 00000020 tfd c0 serr 00000000 cmd 0000c517
> [2021-12-10 20:56:11] ahcich1: (ada0:ahcich1:0:0:0): DSM TRIM. ACB: 06 01
> 00 00 00 40 00 00 00 00 01 00
> [2021-12-10 20:56:11] AHCI reset...
> [2021-12-10 20:56:11] (ada0:ahcich1:0:0:0): CAM status: Command timeout
> [2021-12-10 20:56:11] ahcich1: (ada0:ahcich1:0:SATA connect time=100us
> status=00000133
> [2021-12-10 20:56:11] 0:0): Retrying command
> [2021-12-10 20:56:11] ahcich1: AHCI reset: device found
> [2021-12-10 20:56:42] ahcich1: AHCI reset: device not ready after 31000ms
> (tfd = 00000080)
> [2021-12-10 20:57:15] ahcich1: Timeout on slot 6 port 0
> [2021-12-10 20:57:15] ahcich1: is 00000000 cs 00000040 ss 00000000 rs
> 00000040 tfd 80 serr 00000000 cmd 0000c617
> [2021-12-10 20:57:15] ahcich1: (aprobe0:ahcich1:0:0:0): ATA_IDENTIFY. ACB:
> ec 00 00 00 00 40 00 00 00 00 00 00
> [2021-12-10 20:57:16] AHCI reset...
> [2021-12-10 20:57:16] (aprobe0:ahcich1:0:0:0): CAM status: Command timeout
> [2021-12-10 20:57:16] ahcich1: (aprobe0:SATA connect time=100us
> status=00000133
> [2021-12-10 20:57:16] ahcich1:0:ahcich1: AHCI reset: device found
> [2021-12-10 20:57:16] 0:0): Retrying command
> [2021-12-10 20:57:47] ahcich1: AHCI reset: device not ready after 31000ms
> (tfd = 00000080)
> [2021-12-10 20:58:20] ahcich1: Timeout on slot 7 port 0
> [2021-12-10 20:58:20] ahcich1: is 00000000 cs 00000080 ss 00000000 rs
> 00000080 tfd 80 serr 00000000 cmd 0000c717
> [2021-12-10 20:58:20] ahcich1: (aprobe0:ahcich1:0:0:0): ATA_IDENTIFY. ACB:
> ec 00 00 00 00 40 00 00 00 00 00 00
> [2021-12-10 20:58:21] AHCI reset...
> [2021-12-10 20:58:21] (aprobe0:ahcich1:0:0:0): CAM status: Command timeout
> [2021-12-10 20:58:21] ahcich1: (aprobe0:ahcich1:0:SATA connect time=100us
> status=00000133
> [2021-12-10 20:58:21] 0:0): ahcich1: AHCI reset: device found
> [2021-12-10 20:58:21] Error 5, Retries exhausted
> [2021-12-10 20:58:52] ahcich1: AHCI reset: device not ready after 31000ms
> (tfd = 00000080)
> [2021-12-10 20:59:25] ahcich1: Timeout on slot 8 port 0
> [2021-12-10 20:59:25] ahcich1: is 00000000 cs 00000100 ss 00000000 rs
> 00000100 tfd 80 serr 00000000 cmd 0000c817
> [2021-12-10 20:59:25] ahcich1: (aprobe0:ahcich1:0:0:0): ATA_IDENTIFY. ACB:
> ec 00 00 00 00 40 00 00 00 00 00 00
> [2021-12-10 20:59:26] AHCI reset...
> [2021-12-10 20:59:26] (aprobe0:ahcich1:0:0:0): CAM status: Command timeout
> [2021-12-10 20:59:26] ahcich1: (aprobe0:SATA connect time=100us
> status=00000133
> [2021-12-10 20:59:26] ahcich1:0:ahcich1: 0:AHCI reset: device found
> [2021-12-10 20:59:26] 0): Error 5, Retry was blocked
> [2021-12-10 20:59:26] ada0 at ahcich1 bus 0 scbus0 target 0 lun 0
> [2021-12-10 20:59:26] ada0: <WDC PC SA530 SDASB8Y256G 40103000> s/n
> 200914802684 detached
> [2021-12-10 20:59:26] g_vfs_done():ufs/log[WRITE(offset=5309202432,
> length=32768)]error = 6
> [2021-12-10 20:59:26] pass0 at ahcich1 bus 0 scbus0 target 0 lun 0
> [2021-12-10 20:59:26] g_vfs_done():pass0: <WDC PC SA530 SDASB8Y256G
> 40103000>ufs/log[WRITE(offset=65027883008, length=16384)] s/n
> 200914802684error = 6
> [2021-12-10 20:59:26] detached
> [2021-12-10 20:59:26]
> g_vfs_done():(pass0:ahcich1:0:ufs/log[WRITE(offset=65124368384,
> length=32768)]0:error = 6
> [2021-12-10 20:59:26] 0): g_vfs_done():Periph destroyed
> [2021-12-10 20:59:26] ufs/log[WRITE(offset=65124368384,
> length=32768)]error = 6
> [2021-12-10 20:59:26] g_vfs_done():ufs/log[WRITE(offset=65126465536,
> length=32768)]error = 6
> [2021-12-10 20:59:26] g_vfs_done():ufs/log[WRITE(offset=65126465536,
> length=32768)]error = 6
> [2021-12-10 20:59:26]
> g_vfs_done():g_vfs_done():ufs/log[WRITE(offset=24702976,
> length=512)]ufs/log[WRITE(offset=65126662144, length=16384)]error = 6
> [2021-12-10 20:59:27] panic: cannot reassign paging buffer
> [2021-12-10 20:59:27] cpuid = 0
> [2021-12-10 20:59:27] __HardenedBSD_version = 1100056 __FreeBSD_version =
> 1103500
> [2021-12-10 20:59:27] version = NS-BSD 4.2.0.beta--HBSD #0 : Wed Dec 1
> 14:00:48 CET 2021
> [2021-12-10 20:59:27] build@BuildFreeBSD-11.3-hardened
> :/home/build/build/kernel/work-OPTIM/sys/amd64/compile/NETASQ.XL.SMP.HW.RELEASE
> [2021-12-10 20:59:27] Uptime: 2d0h46m1s
> [2021-12-10 20:59:27] Dumping 2004 out of 8064
> MB:..1%..11%..21%..31%..41%..51%..61%..71%..81%..91%
> [2021-12-10 21:00:29] Dump complete
> [2021-12-10 21:00:31] þVersion 2.15.1236. Copyright (C) 2012 American
> Megatrends, Inc.
> --
>
> It appears that after sending a SATA TRIM command, that the disk stops
> responding. I've also observed another crash that happens just after a TRIM
> command.
>
> According to camcontrol, the disk supports the TRIM command :
>
> --
> camcontrol identify /dev/ada0
> pass0: <WDC PC SA530 SDASB8Y256G 40103000> ACS-4 ATA SATA 3.x device
> pass0: 600.000MB/s transfers (SATA 3.x, UDMA6, PIO 512bytes)
>
> protocol ATA/ATAPI-11 SATA 3.x
> device model WDC PC SA530 SDASB8Y256G
> firmware revision 40103000
> serial number 200914802684
> WWN 5001b448b13c08ca
> cylinders 16383
> heads 16
> sectors/track 63
> sector size logical 512, physical 512, offset 0
> LBA supported 268435455 sectors
> LBA48 supported 500118192 sectors
> PIO supported PIO4
> DMA supported WDMA2 UDMA6
> media RPM non-rotating
> Zoned-Device Commands no
>
> Feature Support Enabled Value Vendor
> read ahead yes yes
> write cache yes yes
> flush cache yes yes
> overlap no
> Tagged Command Queuing (TCQ) no no
> Native Command Queuing (NCQ) yes 32 tags
> NCQ Queue Management no
> NCQ Streaming no
> Receive & Send FPDMA Queued no
> SMART yes yes
> microcode download yes yes
> security yes no
> power management yes yes
> advanced power management yes yes 128/0x80
> automatic acoustic management no no
> media status notification no no
> power-up in Standby no no
> write-read-verify no no
> unload no no
> general purpose logging yes yes
> free-fall no no
> Data Set Management (DSM/TRIM) yes
> DSM - max 512byte blocks yes 8
> DSM - deterministic read yes zeroed
> Host Protected Area (HPA) no
> --
>
>
> I found this patch concerning a similar problem :
> https://bugs.freebsd.org/bugzilla/show_bug.cgi?id=222802
>
> I've modified sys/cam/ata/ata_da.c to deactivate the trim functionality.
>
> --
> +diff --git sys/cam/ata/ata_da.c sys/cam/ata/ata_da.c
> +index cba52c5458be..ad256da0a495 100644
> +--- sys/cam/ata/ata_da.c
> ++++ sys/cam/ata/ata_da.c
> +@@ -1798,7 +1798,7 @@ adaregister(struct cam_periph *periph, void *arg)
> + softc->disk->d_flags = DISKFLAG_DIRECT_COMPLETION |
> DISKFLAG_CANZONE;
> + if (softc->flags & ADA_FLAG_CAN_FLUSHCACHE)
> + softc->disk->d_flags |= DISKFLAG_CANFLUSHCACHE;
> +- if (softc->flags & ADA_FLAG_CAN_TRIM) {
> ++ if (0 /*softc->flags & ADA_FLAG_CAN_TRIM */) {
> + softc->disk->d_flags |= DISKFLAG_CANDELETE;
> + softc->disk->d_delmaxsize = softc->params.secsize *
> + ATA_DSM_RANGE_MAX *
> --
>
> I've been running a set of tests on one of our products for over two weeks
> with the trim deactivated (using the above patch) and there have been no
> crashes.
>
> Could this issue be an internal disk firmware problem?
>
> Is it possible that camcontrol reports that the disk supports TRIM yet in
> reality it isn't supported, or isn't supported fully?
>
> I've observed that this crash doesn't happen on the first TRIM command, is
> it possible that a set of SATA commands destabilize the disk firmware and
> cause it to crash?
>
> Is it possible that the timeout isn't long enough for the TRIM command?
>
> Thanks in advance for any help.
>
> Best wishes,
> Lee Matthews
>
>
>
>
>
>
>
>
>
>
>