Re: [PATCH] barrier patch set

[Jeff Garzik]

Bartlomiej Zolnierkiewicz wrote:
> The fact that spec says "supported" not "enabled" in description of word86
> makes me wonder - can they be disabled? (FLUSH CACHE is mandatory for General
> feature set and FLUSH CACHE EXT is mandatory if 48-bit LBA is supported)

Yes, that's why there are separate 'supported' and 'enabled' bits for 
each feature.

Words 82-84 are 'supported' bits.  Words 85-87 are 'enabled' bits. 
These bits mirror each other, i.e. Word 83 and Word 86 have basically 
the same bits, except that Word 86 definitions change _slightly_ since 
the only bits that are relevant are the ones for features that can be 
disabled/enabled.

You use set-features command to enable and disable these features, and 
then the result shows up in subsequent identify-device command output.

If the driver is testing for a capability but does not enable it, then 
always use the 'enabled' set of bits, not the 'supported' set of bits.


> Jeff, please note that these bits were introduced by ATA-6 spec
> and take a look at ATA-5 spec:
>
> ...
> FLUSH CACHE
> General feature set
> - Mandatory for all devices
> ...
>
> and ATA-4 spec:
>
> ...
> FLUSH CACHE
> General feature set
> - Optional for all devices
> ...
>
> IMO to test if FLUSH CACHE works we should just issue it during disk setup
> and check result.  This way we can use FLUSH CACHE also on < ATA-6 devices
> (there is a lot of them).

I disagree.  "just issue it" is how those LG cdrom drives got cooked.

LG cdrom drives indicated in their identify-packet-device page that 
flush-cache was not supported...  and then re-used the flush-cache ATA 
opcode for their vendor-specific download-firmware command.  Combine 
that with a Linux patch that didn't properly check for flush-cache 
support.  Result: brick.

All drives that support flush-cache list the relevant bits in 
identify-device, even on pre-ATA-6 devices.  Whether the feature was 
optional or mandantory, we can check the feature bits.

	Jeff



-
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/