Re: Linux 2.6.25-rc4

[Bartlomiej Zolnierkiewicz <bzolnier@gmail.com>]

On Wednesday 19 March 2008, Linus Torvalds wrote:
> 
> On Wed, 19 Mar 2008, Bartlomiej Zolnierkiewicz wrote:
> > 
> > Which is with compliance with PIO-in protocol specification and years of
> > usage of the task_in_intr() code for fs code paths and HDIO_DRIVE_TASKFILE
> > ioctl has proven that real hardware also works this way (please note that
> > the changed code was used only for HDIO_DRIVE_CMD ioctl requests).
> 
> .. and years of drive_cmd_intr() which is much more widely used is *not* 
> in agreement.
> 
> > If INTRQ is asserted and "BSY is cleared to zero and DRQ is cleared, then
> > the device has completed the command with an error." (thus task_in_intr()
> > assumed ERR bit to be set), otherwise the value ERR may not be defined
> > (however task_in_intr() still assumed that it is OK to check ERR bit).
> 
> First off, the patch I sent out _works_.
> 
> Secondly, it's a hell of a lot more robust than yours is, exactly because 
> it doesn't get confused if the data direction or size bit disagrees with 
> the particular command in question.
> 
> > This is SMART command with SMART ENABLE ATTRIBUTE AUTOSAVE sub-command
> > (feat == 0xd2, nsect == 0xf1) but it should use no-data protocol instead of
> > PIO-in which brings us back to commit 18a056feccabdfa9764016a615121b194828bc72
> > ("ide: don't enable local IRQs for PIO-in in driver_cmd_intr() (take 2)"):
> 
> .. and what about all the magic special IDE commands that may be 
> drive-specific? What are we going to do about them?
> 
> In other words, we should not try to create an impossible-to-maintain 
> piece of shit code that does the wrong thing if you send a command to the 
> drive that the IDE layer doesn't understand (but the sending code 
> hopefully does).
> 
> We should make the core IDE code *robust*. 
> 
> Your "real fix" is nothing of the sort. It's just a workaround for the 
> fragility of the code that looks at the drive status. The real fix is to 
> be robust in the face of even unexpected drive status codes, *especially* 
> for the code that handles commands injected by the user!

Please take a closer look - my "real fix" _only_ affects WIN_SMART command
and _not_ vendor special ones (no, there are none vendor special commands
using the same opcode).

DRQ/ERR bits handling is a tangent issue (and yes it also needs fixing).

> In other words, you can talk about protocol specifications for things like 
> the regular filesystem READ/WRITE commands. But don't create total crap 
> like this that depends on the code knowing all possible outcomes of every 
> single possible command.
> 
> Your patch is utter crap.
> 
> You say (about commit 18a056feccabdfa9764016a615121b194828bc72):
> 
> > as can be seen before this patch protocol to use was decided basing on DRQ bit
> > being set - what would you call _that_ if taskfile code is horrible, he? ;-)
> 
> I would call that *correct*.
> 
> And my point is, we used to be better. You made the code buggy and fragile 
> with that crap commit, and with the others like it (ie the already 
> much-mentioned commit 4d977e43d8ae758434e603cf2455d955f71c77c4).
> 
> And that is and was *exactly* my point. The reason I called the taskfile 
> code horrible was exactly the fact that it only worked if it thought it 
> knew exactly what was going on.

This is actually how the ATA hardware operates.

You cannot just bang random commands and expect that later driver will be
able to 100% correctly deduce what the command wants from the drive status
itself (this may work more-or-less well for no-data and PIO-in by using DRQ
bit trick, but we also have PIO-out, DMA, PACKET etc).

IOW this won't fly as drive/controller alone doesn't provide use with enough
information about command being currently executed.

> Deciding what to do based on the DRQ bit (and the READY/BUSY/ERROR bits) 
> is the *right* thing to do. It's the intelligent approach - actually 
> tekign the response of the hardware into account, and being robust. The 
> *stupid* and horrible thing to do is to think that you know better than 
> what the hardware tells you, and think that you can look up every command 
> that the user sends in the spec and use *that* to figure out what to do.

There is some misunderstaning here - taskfile approach lets user specify
_both_ the command and what it really wants (protocol etc.), and no - we
don't do any command checking with the spec but give user the control...

Call taskfile crap all you want but the fact is that taskfile approach
handles _all_ protocols and _all_ commands while the beloved "robust"
and "intelligent" approach is able to handle only 28-bit commands using
no-data or PIO-in protocols (this is what HDIO_DRIVE_CMD is supporting
and there is no way it will ever support more than that with its design).

[ BTW libata also uses taskfile approach (including HDIO_DRIVE_CMD!) ]

> Trust the hardware, not the paper. Don't make the code only work when you 
> think you know what is going on. Make the code work _always_.

The problem here is that we cannot fully trust neither hardware nor the
paper (so we give the control to the user for the final decision).

> In short, there is no way in hell I'll apply a workaround patch for crap 
> code, when we already know what the robust solution is.

Please re-consider this decision given the above facts.

Thanks,
Bart