Re: [PATCH] Performance Improvement in CRC16 Calculations.

[Ard Biesheuvel]

On 24 August 2018 at 17:29, Martin K. Petersen
<martin.petersen@xxxxxxxxxx> wrote:
>
> Ard,
>
>> Would it be possible to allocate the crypto transform upon first use
>> instead of from an initcall? If crc_t10dif() is mostly called from
>> non-process context, that would not really work, but otherwise, we
>> could simply defer it (and occasional calls from non-process context
>> that do occur would use the generic code until the point where another
>> call from process context allocates the transform)
>
> The function is always called from user context. However, postponing the
> crypto transform registration doesn't solve the common scenario of the
> user booting off of a Fibre Channel/SAS/NVMe device with the desired
> crct10dif-pclmul.ko module located on the boot drive.
>
> If there is no good way to teach crypto to update existing registrations
> when a higher priority transformation becomes available, then we
> probably need to explore tweaking dracut to unconditionally load
> crct10dif-pclmul (and your ARM equivalent). Looks like there are already
> hacks in place in dracut to preload crc32c for btrfs and XFS.
>

I'd prefer to handle this without help from userland.

It shouldn't be too difficult to register a module notifier that only
sets a flag (or the static key, even?), and to free and re-allocate
the crc_t10dif transform if the flag is set.


> Anyway. Just seems like the kernel is violating the principle of least
> surprise here. The kernel should always pick the best available tool for
> the job...
>
> --
> Martin K. Petersen      Oracle Linux Engineering