Re: [PATCH] IDE: Fix HDIO_DRIVE_RESET handling

Alan Cox <alan@xxxxxxxxxxxxxxxxxxx> wrote:
in various ways. Most importantly, it is treated as an out of band
request in an illegal way which may very likely lead to system lock ups.
Use the drive's request queue to avoid this problem (and fix a locking
issue for free along the way).

It was always designed to be, and used out of band. One of the important
uses of the ioctl is to abort a running command when an interface has
jammed up. If you end up queueing it behind that command you've lost most
of the reason for the ioctl anyway (and you might as well just remove it
really given SG_IO exists).

Well, I can see your point. In fact, there really doesn't seem to be an
alternative to the out of band approach for the purposes you described.
Now, I even think that I could perhaps fix the request aborting properly
and restore the original behaviour. Moreover, I may very likely live to
regret having removed ide_abort() and friends when implementing disk
shock protection in the IDE layer. Maybe I should try to send an
alternative patch for discussion. On the other hand I don't see the
equivalent for HDIO_DRIVE_RESET in libata which makes me wonder whether
this ioctl has actually been used in real life for the purposes you
described.


Other than the command aborting bit, it looks a good idea - that code
has
always been racy and raced against timer handlers, irq handlers and if
neither of them got it then a speed changedown raced the lot 8(

My idea to solve this would be roughly this: Change ide_set_handler to
leave the ->handler and ->expiry members alone if they have been set on
entry. If a request is being processed by the time a HDIO_DRIVE_RESET
ioctl is received, these callbacks will be changed so the reset sequence
will be started on the next interrupt, timeout, or when the ->busy flag
is cleared. I'm not quite sure yet whether things will work out the way
I want them to and I don't know whether HDIO_DRIVE_RESET actually
justifies the effort since I don't knowof an equivalent in libata
anyway. But as I said, it might come in handy for other purposes.

Comments?

Elias
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/

Relevant Pages

  • [RFC] Re: [PATCH 4/4] autofs4 - add miscelaneous device for ioctls
    ... We seem to be passing some string into a misc-device ioctl and getting some ... I'm not aware of what the netlink interface may be ... completion of the original ioctl expire request. ...
    (Linux-Kernel)
  • Re: Driver verifier cause SYSTEM_SERVICE_EXCEPTION
    ... a request that has a cancel routine assigned to it (e.g. you called ... An exception happened while executing a system service routine. ... The problem is that when the driver receives its first IOCTL and it is ... to complete or even forward this request a BSOD is reported. ...
    (microsoft.public.development.device.drivers)
  • Re: Get service from a user mode application/ service
    ... The typical problem here is that you need more than one ... speed increase by good development of the IOCTL approach. ... Windows 2k/XP/2k3 Filesystem and Driver Consulting ... From user mode I send an IOCTRL request to driver. ...
    (microsoft.public.development.device.drivers)
  • Re: How user wait for kernel?
    ... My kernel IOCTL function call ... than my IOCTL function will return to user. ... have your IOCTL's put into a cancel safe queue (if this is WDF the ... user, until I completed request, isn't it? ...
    (microsoft.public.development.device.drivers)
  • Re: Licensing - The biter- bit?
    ... is the problem with the request? ... about 'purposes which include the purpose'. ... You can't write directly to the Parliamentary Ombudsman, the complaint has to be submitted via an/your MP, and in any case I'm not sure that this level of complaint would get very far - I'm afraid it looks like pretty small beer compared to most of those of which I had experience. ...
    (uk.music.folk)
Loading