Date: Sun, 12 Jan 2014 14:28:09 -0800 (PST) From: Alex Goncharov <alex_goncharov_usa@yahoo.com> To: freebsd-usb@FreeBSD.org, Hans Petter Selasky <hps@bitfrost.no> Subject: Re: usb/185628: usbd_req_re_enumerate set address failed USB_ERR_STALLED for Seagate USB drives between r259425 and r260321 Message-ID: <1389565689.83194.YahooMailBasic@web162101.mail.bf1.yahoo.com> In-Reply-To: <52D30966.4090002@bitfrost.no>
next in thread | previous in thread | raw e-mail | index | archive | help
,-- From: Alex Goncharov <alex_goncharov_usa@yahoo.com>
> Date: Sunday, January 12, 2014, 5:01 PM
> I just noticed your recent
>
> ----
> r260575 | hselasky | 2014-01-12
>
> and am beginning a full rebuild; the results will be known in about
> three hours.
Hans,
While I am doing the rebuild, may I return to the topic I touched
slightly in my original PR submission?
A sporadic USB HDD device loss, sometimes with a system crash:
I had this with a WD drive, when "da0" could disappear at any moment,
a file system vnode could not be found for reading or writing and bad
things would happen. Now the same story with the Sony USB drive.
My observations of many USB HDD's led me to conclude that some are
smarter than the others -- the smarter ones may be slower to react to
just about anything but they don't get lost. My Seagates may have a
huge operation queues for either writing or reading, but I've never
lost those drives' devices ("da0"s) when using them. 500G Buffalo
never has a long queue, and good for it, but I am fine with a longer
queue of the 1T Seagates, as long as their "da0"s don't go down. 1.5T
Toshiba is another story: it seems like it often needs a significant
wake-up period after sitting idle, but 'da0' never goes away, either.
What WD and Sony exhibit on FreeBSD is plain horrible. It doesn't
make sense to quickly write the first 10G of 100G of data if the
system goes down after those 10G. And losing "da0" on reading or
after idling (the WD's behavior) is just as bad.
As I mentioned, I didn't observe the Sony issue when using it on Linux
(I didn't with WD -- just sent it back.)
Can something be done about it along the Linux's lines, which you
briefly mentioned and seemed to be critical about? As a data user, I
strongly disagree that Linux's approach here is inferior to the one
FreeBSD took, if I understand both correctly.
Thank you,
-- Alex
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?1389565689.83194.YahooMailBasic>