Linux USB
 help / color / mirror / Atom feed
* [PATCH v4] usb: uas: ignore UAS for ASMT135x bridges (174c:55aa)
@ 2026-11-14 17:49 Olivier Kaloudoff
  2026-11-16 12:00 ` Oliver Neukum
  0 siblings, 1 reply; 2+ messages in thread
From: Olivier Kaloudoff @ 2026-11-14 17:49 UTC (permalink / raw)
  To: linux-usb; +Cc: stern, gregkh, oneukum, usb-storage, stable, Olivier Kaloudoff

The ASMedia chip detection in uas_use_uas_driver() assumes that every
device with idVendor/idProduct 174c:5106 or 174c:55aa reporting a
config.MaxPower of 0 is an ASM1153 with working UAS.

ASMT135x bridges (iProduct "ASMT135x") share the same 174c:55aa id and
also report MaxPower == 0, so they take the "ASM1153, do nothing"
path — but their UAS implementation is broken:

- The first READ(16) issued by sd probing times out after 30 s;
  enumeration only completes after a uas_eh_abort_handler plus a full
  uas_eh_host_reset_handler cycle:

    scsi 3:0:0:0: Direct-Access  ASMT     ASMT135x        0  PQ: 0 ANSI: 6
    sd 3:0:0:0: [sdb] tag#26 uas_eh_abort_handler 0 uas-tag 1 inflight: CMD IN
    sd 3:0:0:0: [sdb] tag#26 CDB: Read(16) 88 00 00 00 00 00 00 00 00 00 00 00 00 01 00 00
    scsi host3: uas_eh_host_reset_handler start
    usb 6-2: reset SuperSpeed USB device number 2 using xhci_hcd
    scsi host3: uas_eh_host_reset_handler success

- Under concurrent load the bridge stalls: all inflight commands (20+
  UAS tags in the captured incident, mixed READ(16)/WRITE(16)) hit the
  30 s UAS timeout in the same event, forcing repeated abort/reset
  cycles and eventually a USB disconnect.  Writes dropped mid-flight
  during these resets returned zeros to the filesystem (observed as
  XFS metadata corruption on the mounted volume).

- Sequential-write throughput is ~45 MiB/s under UAS versus ~136 MiB/s
  after disabling UAS (usb-storage BOT mode), which is fully stable:
  several tens of GiB written without a single error.

- Bulk reads of healthy LBA ranges do work under UAS (~190 MiB/s, no
  errors); the failure pattern is command-timeout at enumeration and
  under queue depth / write load, not a broken transfer path.

An extended soak was also run at maintainer request: the device was
rebound to uas and pushed with ~640 GiB of mixed I/O over ~2h45
(randrw/seqwrite at queue depths 8-32, disk util ~100%).  Result: zero
uas_eh events; the queue-depth wedge did not reproduce on demand.  The
enumeration stall did reproduce once more at the first UAS probe and
self-recovered via host reset.  Per the maintainer, an inconclusive
load test still warrants the blacklist given the filesystem-loss risk.

Since 174c:55aa is shared across several ASMedia bridges and enclosure
rebrandings, distinguish the ASMT135x by its product string — the same
approach used for the HIKSEMI MD202 check a few lines below.  The
product strings of working bridges using the same id ("ASM1053E",
"ASM1153", "ASM1153E", "ASM1351", vendor rebrands such as "YZWY_TECH")
do not contain "ASMT135", so the match is safe.

Verified on an ORICO C10 enclosure (iManufacturer "ASMT", iProduct
"ASMT135x", bcdDevice 1.00, stock firmware) attached through an Alpine
Ridge TB3 dock: with the equivalent usb-storage quirks parameter
(174c:55aa:u) the device is completely stable.

Signed-off-by: Olivier Kaloudoff <olivier.kaloudoff@gmail.com>
---
v4: resend at maintainer request (Oliver Neukum will ack); documented
    the extended UAS soak results in the commit message.
v3: add braces around the if-body (per Greg KH).
v2: fix the comment block: added lines were missing the leading
    tab, breaking the *-alignment with the rest of the block.

 drivers/usb/storage/uas-detect.h | 8 +++++++-
 1 file changed, 7 insertions(+), 1 deletion(-)

diff --git a/drivers/usb/storage/uas-detect.h b/drivers/usb/storage/uas-detect.h
index 4d3b49e5b..523a17cf0 100644
--- a/drivers/usb/storage/uas-detect.h
+++ b/drivers/usb/storage/uas-detect.h
@@ -78,6 +78,8 @@ static int uas_use_uas_driver(struct usb_interface *intf,
         * ASM1051 - with broken (*) uas support
         * ASM1053 - with working uas support, but problems with large xfers
         * ASM1153 - with working uas support
+        * ASMT135x - broken uas support, shares the ASM1153 MaxPower == 0
+        *            signature, identify it by its product string
         *
         * Devices with these chips re-use a number of device-ids over the
         * entire line, so the device-id is useless to determine if we're
@@ -99,7 +101,11 @@ static int uas_use_uas_driver(struct usb_interface *intf,
                        (le16_to_cpu(udev->descriptor.idProduct) == 0x5106 ||
                         le16_to_cpu(udev->descriptor.idProduct) == 0x55aa)) {
                if (udev->actconfig->desc.bMaxPower == 0) {
-                       /* ASM1153, do nothing */
+                       if (udev->product && strstr(udev->product, "ASMT135")) {
+                               /* ASMT135x, broken uas */
+                               flags |= US_FL_IGNORE_UAS;
+                       }
+                       /* else ASM1153, do nothing */
                } else if (udev->speed < USB_SPEED_SUPER) {
                        /* No streams info, assume ASM1051 */
                        flags |= US_FL_IGNORE_UAS;
-- 
2.47.3

^ permalink raw reply related	[flat|nested] 2+ messages in thread

* Re: [PATCH v4] usb: uas: ignore UAS for ASMT135x bridges (174c:55aa)
  2026-11-14 17:49 [PATCH v4] usb: uas: ignore UAS for ASMT135x bridges (174c:55aa) Olivier Kaloudoff
@ 2026-11-16 12:00 ` Oliver Neukum
  0 siblings, 0 replies; 2+ messages in thread
From: Oliver Neukum @ 2026-11-16 12:00 UTC (permalink / raw)
  To: Olivier Kaloudoff, linux-usb; +Cc: stern, gregkh, usb-storage, stable



On 14.11.26 18:49, Olivier Kaloudoff wrote:
> The ASMedia chip detection in uas_use_uas_driver() assumes that every
> device with idVendor/idProduct 174c:5106 or 174c:55aa reporting a
> config.MaxPower of 0 is an ASM1153 with working UAS.
> 
> ASMT135x bridges (iProduct "ASMT135x") share the same 174c:55aa id and
> also report MaxPower == 0, so they take the "ASM1153, do nothing"
> path — but their UAS implementation is broken:
> 
> - The first READ(16) issued by sd probing times out after 30 s;
>    enumeration only completes after a uas_eh_abort_handler plus a full
>    uas_eh_host_reset_handler cycle:
> 
>      scsi 3:0:0:0: Direct-Access  ASMT     ASMT135x        0  PQ: 0 ANSI: 6
>      sd 3:0:0:0: [sdb] tag#26 uas_eh_abort_handler 0 uas-tag 1 inflight: CMD IN
>      sd 3:0:0:0: [sdb] tag#26 CDB: Read(16) 88 00 00 00 00 00 00 00 00 00 00 00 00 01 00 00
>      scsi host3: uas_eh_host_reset_handler start
>      usb 6-2: reset SuperSpeed USB device number 2 using xhci_hcd
>      scsi host3: uas_eh_host_reset_handler success
> 
> - Under concurrent load the bridge stalls: all inflight commands (20+
>    UAS tags in the captured incident, mixed READ(16)/WRITE(16)) hit the
>    30 s UAS timeout in the same event, forcing repeated abort/reset
>    cycles and eventually a USB disconnect.  Writes dropped mid-flight
>    during these resets returned zeros to the filesystem (observed as
>    XFS metadata corruption on the mounted volume).
> 
> - Sequential-write throughput is ~45 MiB/s under UAS versus ~136 MiB/s
>    after disabling UAS (usb-storage BOT mode), which is fully stable:
>    several tens of GiB written without a single error.
> 
> - Bulk reads of healthy LBA ranges do work under UAS (~190 MiB/s, no
>    errors); the failure pattern is command-timeout at enumeration and
>    under queue depth / write load, not a broken transfer path.
> 
> An extended soak was also run at maintainer request: the device was
> rebound to uas and pushed with ~640 GiB of mixed I/O over ~2h45
> (randrw/seqwrite at queue depths 8-32, disk util ~100%).  Result: zero
> uas_eh events; the queue-depth wedge did not reproduce on demand.  The
> enumeration stall did reproduce once more at the first UAS probe and
> self-recovered via host reset.  Per the maintainer, an inconclusive
> load test still warrants the blacklist given the filesystem-loss risk.
> 
> Since 174c:55aa is shared across several ASMedia bridges and enclosure
> rebrandings, distinguish the ASMT135x by its product string — the same
> approach used for the HIKSEMI MD202 check a few lines below.  The
> product strings of working bridges using the same id ("ASM1053E",
> "ASM1153", "ASM1153E", "ASM1351", vendor rebrands such as "YZWY_TECH")
> do not contain "ASMT135", so the match is safe.
> 
> Verified on an ORICO C10 enclosure (iManufacturer "ASMT", iProduct
> "ASMT135x", bcdDevice 1.00, stock firmware) attached through an Alpine
> Ridge TB3 dock: with the equivalent usb-storage quirks parameter
> (174c:55aa:u) the device is completely stable.
> 
> Signed-off-by: Olivier Kaloudoff <olivier.kaloudoff@gmail.com>Acked-by: Oliver Neukum <oneukum@suse.com>

^ permalink raw reply	[flat|nested] 2+ messages in thread

end of thread, other threads:[~2026-11-16 12:00 UTC | newest]

Thread overview: 2+ messages (download: mbox.gz follow: Atom feed
-- links below jump to the message on this page --
2026-11-14 17:49 [PATCH v4] usb: uas: ignore UAS for ASMT135x bridges (174c:55aa) Olivier Kaloudoff
2026-11-16 12:00 ` Oliver Neukum

This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox