On 26.01.20 16:21, Hans de Goede wrote:
Hi,
On 25-01-2020 20:59, Tim Schumacher wrote:
On 25.01.20 19:37, Hans de Goede wrote:> Hi,
On 1/25/20 6:00 PM, Tim Schumacher wrote:
The JMicron JMS561U (notably used in the Sabrent SATA-to-USB
bridge) appears to have UAS-related issues when copying large
amounts of data, causing it to stall.
Disabling the advertised UAS (either through a command-line
quirk or through this patch) mitigates those issues.
Cc: stable@xxxxxxxxxxxxxxx
Signed-off-by: Tim Schumacher <timschumi@xxxxxx>
Hmm, this is a quiete popular usb2sata bridge and disabling uas
is quite bad for performance.
I haven't been able to notice any slowdown myself, averaging 350MB/s
while copying large files, before and after the patch. However, from
what I've been able to grasp, the actual advantage of UAS seems to
be located in even higher speeds, which I can't properly test with my
equipment.
The big difference is not so much linear throughput, as well as
iops / random access patterns. UAS allows sending multiple data
requests to the disk at once, which the old bulk mass storage
protocol does not allow.
Chances are that the times when you are seeing the hangs you
are also accessing the disk in some other way while copying a large
file. With the old mass storage protocol in this case the copy
will pause and then your other access will happen and then the copy
will resume. UAS allows both to happen "at once" and SSDs are very good
at this. As you can imagine making the SSD do multiple things at once
it is also something which really pushes the power-requirements up.
That makes sense, I haven't thought much about that.
Let's put this patch on hold then until I can figure out whether
it really is a power limit that I'm hitting.
It's a valid concern though, since SATA 3 theoretically goes way
higher than what I can reach.
I notice that there is no link to a bug report and AFAIK we have
no one else reporting this issue.
I haven't specifically looked on the kernel bug tracker yet, but I
found similiar UAS-related issues talking about the JMicron JMS567
and JMS579 on the Ubuntu kernel bug tracker [1], as well as the
Raspberry Pi bug tracker [2].
If it helps, I can make this a proper bug report first so that other
people can chime in, instead of burying the discussion in the mailing
list.
[1] https://bugs.launchpad.net/ubuntu/+source/linux/+bug/1789589
"When the device is under load and more than 1 hardrive is acessed,
the USB enclosure frequently restarts."
In other words, when I cause the power consumption to go through the
roof things crash. This smells very much like (yet another) power-supply
problem.
[2] https://github.com/raspberrypi/linux/issues/3070
Raspberry Pi's are notorious for having issues supplying enough power
to their USB ports due to the self-resetting electronic fuses they
are using, which increase there resistance when the current demand
peaks, which causes a voltage drop.
So this feels like a much too big hammer for the problem which you
are seeing.
When you say "stall" what exactly happens? Do you see any errors
in dmesg for example?
Basically, the transfer just freezes at one point (be it an actual
file transfer or just browsing directories quite fast), and a few
seconds later, UAS-related errors start appearing in dmesg.
At this point, the device either never recovers and requires a reconnect
to work correctly or it eventually recovers (after about 15 to 20
seconds) and continues the transfer as expected.
A dmesg of the device failing to recover can be found here: [3]
I can't reproduce a case where the device recovers right now, but
I found a StackExchange question with the same problem and an attached
dmesg. The general content of those error messages (maybe apart
from the hex output) is similiar to what I've been seeing: [4]
I'll try and see if I can hit a recoverable error myself in the next
few days.
[3] https://pastebin.com/raw/ny128rB4
Ok, so what we are seeing there is that the usb-sata bridge has basically
completely crashed. Normally it should always recover from a crash.
This really feels like a brownout (supply voltage too low) event has
happened, as that is typically the only thing which will hang the
bridge like this.
[4] https://pastebin.com/raw/i7KLzy6i
Also note that using UAS, since it has much better performance,
will often expose bugs which are not caused by it. One typical
example is bus-powered devices where the USB port does not deliver
enough power (typically the driver draws more then the port
guanrantees). Copying large amounts of data on a fast device is
a good way to make the current consumption go up and thus
trigger these kind of issues. Does the driver enclosure
you see this on use a separate power supply, or is it
bus-powered?
It is indeed a bus-powered enclosure/adapter, which I'm using with
an USB 3.0 port. The attached SSD is rated for 5V/0.7A. However,
(as mentioned above) I am reaching the same read speeds with and
without UAS, so I'm not quite sure whether it really is a power limit
caused by heavy load.
As I've tried to explain above, UAS allows more then one command
to be outstanding at once. Even if you are only copying a single file
then the Linux kernel will send more requests for blocks further ahead
in the file. This will make the SSD work harder to put the data in its
buffer, so even if the average through put stays the same the peak
energy consumption of the SSD may increase.
So far I've not really heard or seen anything which indicates that
there is a systematic problem with the JMS561U bridge. I'm actually
pretty sure you will find similar bug reports for Windows...
I set up my Windows installation again, so I'll check whether I get
similiar issues on there.
Have you tried using the drive in a different USB-3 port (on
a different machine perhaps) and/or with a different (shorter) USB3
cable?
I have tried this on both my PC (both front- and back-facing ports) and
my Laptop, the latter failing as well, but less often.
The cable is unfortunately joined to the enclosure itself, so I can't
try a different one.
Regards,
Hans
I'll try a few things in the next days to figure out whether I can confirm
that this is a power-related issue or if it happens on other platforms and
computers.
If I can't find anything related to those two causes, I'll probably report
back with more information.
Tim
---
v2: Fixed entry order. Also, CCing the correct people now.
---
drivers/usb/storage/unusual_uas.h | 7 +++++++
1 file changed, 7 insertions(+)
diff --git a/drivers/usb/storage/unusual_uas.h b/drivers/usb/storage/unusual_uas.h
index 1b23741036ee..a590f4a0d4b9 100644
--- a/drivers/usb/storage/unusual_uas.h
+++ b/drivers/usb/storage/unusual_uas.h
@@ -73,6 +73,13 @@ UNUSUAL_DEV(0x152d, 0x0578, 0x0000, 0x9999,
USB_SC_DEVICE, USB_PR_DEVICE, NULL,
US_FL_BROKEN_FUA),
+/* Reported-by: Tim Schumacher <timschumi@xxxxxx> */
+UNUSUAL_DEV(0x152d, 0x1561, 0x0000, 0x9999,
+ "JMicron",
+ "JMS561U",
+ USB_SC_DEVICE, USB_PR_DEVICE, NULL,
+ US_FL_IGNORE_UAS),
+
/* Reported-by: Hans de Goede <hdegoede@xxxxxxxxxx> */
UNUSUAL_DEV(0x2109, 0x0711, 0x0000, 0x9999,
"VIA",
--
2.25.0
