On Mon, 23 Jun 2008, Valentin QUEQUET wrote: > Package: linux-image-2.6.25-2-amd64 > Version: 2.6.25-5 > Severity: normal > > > Hello, all. > > While scanning (with SANE) a document from my 'Canon PIXMA MP150' > scanner/printer combo, I stumbled upon a grave problem. > > I discovered that my configuration - out-of-the-box up-to-date lenny - was > using module 'ehci_hcd' (possibly amongst others) to dialog with my scanner. > > And I discovered that there was some activity on IRQ_7, which was > 'listened-to' by module 'ehci_hcd', when I queried scanner list typing: > scanimage -L > and when I put my scanner ON or OFF. > > I was able to query this list 3 times without a problem ; this step did no > hurt. > > But a grave problem suddenly arose when I typed: > scanimage -T > which is a scanner/communication test which consist in scanning a full line - > equivalent to effectively scanning a small part of the document. > > Not only the test failed, but 3 more very bad things happened: > - I got alarming messages on console (and in 'dmesg' below) : "ehci_hdc: > ... HC died", and kernel said it would forget about IRQ_7 (which 'ehci_hcd' > was 'listening-to' earlier). > - I became unable to query the list of scanners anymore : the scanner > ceased to respond. > - The kernel no longer notified (eg on console) when I put my scanner OFF, > ON, and OFF again. > > Fortunately, I discovered that in this situation, I had just to unload module > 'ehci_hcd' to get my scanner+SANE functional (including scanner power ON/OFF > notifications). > > But a few points draw my attention: > > I retried this scenario, implying scanner OFF, PC reboot, delay, scanner ON. > > A few times the scenario showed to be exactly like I described above. > > While the other times, symptoms of the bug (alarming messages + unability to > communicate with the scanner) happened merely whenever I chose to put my > scanner OFF (alarm) and ON (no comm.) instead of doing the scanner test (via > scanimage -T). > > Again, unloading module 'ehci_hcd' got my scanner+SANE functional. > > I found very strange that IRQ_7 would just be used for some handchecking and > not for the transfer phase. > But it's obvious: 'ehci_hcd', which was the sole user of IRQ_7, succeeded at > handchecking and broke on scanning/transfer attempts. > > So, when I unloaded 'ehci_hcd', the transfer phase likely falled back to some > non-IRQ (polling) method. > > I was sorry that my new Debian pre-packaged linux-image-2.6.25-2-amd64 (ver > 2.6.25-5) made my scanning experience so tricky. > > And I asked myself what it would have been if module 'echi_hcd' didn't broke. > Would 'ehci_hcd' be usefull after all ? > Would IRQ_7 be used ? With many interruptions in scanning/transfer phase ? > And would transfers get faster ? > > So, I decided to repeat the whole thing with Debian pre-packaged > linux-image-2.6.24-1-amd64 (ver 2.6.24-7). > You can't believe it ! Scanning (scanner+SANE) functioned right out of the > box, without having root to unload module 'ehci_hcd' (it was effectively > loaded). > > And I was witnessing an intensive use of IRQ_7 by module 'ehci_hcd' ; > transfers were similar than with 2.6.25 : certainly no far from perfect. > > And whenever I put my scanner ON or OFF, I got the matching notifications on > console and in dmesg. > > And whenever I wanted to scan some paper, all was good and perfect. > > To help you understand what happened, and in the case the piece of dmesg > below would not be enough, I give you a pointer to my full dmesg log: > http://pagesperso-orange.fr/mandolosse/logs/2008-06-23__dmesg__ehci_hcd__died.txt > > I also captured 'scanimage', 'lsmod' output and snapshots of > '/proc/interrupts' at different times. I plan to scrutinize all this data, to > comment the relevant parts and to post them later. > > In hope my report will prove useful. > > Sincerely, > Valentin QUEQUET
urrgs indeed. but please can you test out latest upstream linux images 2.6.26-rcX they install just fine in testing/unstable. see trunk apt lines -> http://wiki.debian.org/DebianKernel kind regards -- maks -- To UNSUBSCRIBE, email to [EMAIL PROTECTED] with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]
