On Thu, Aug 22, 2019 at 03:29:40AM +0300, Jarkko Sakkinen wrote:
> On Wed, Aug 21, 2019 at 10:24:00AM -0700, Sean Christopherson wrote:
> > How does the current code not align with this approach? The core subsystem
> > should tear itself down if loading the driver fails, which includes failing
> > because it doesn't exist.
>
> I get now the real issue that you are trying to point out (did not
> earlier). Still, I think your approach to fix it needs some
> reconsideration.
>
> Something that *does not exist* can never fail. That should be dead
> obvious.
>
> If the SGX driver does not exit and KVM does not have SGX support
> compiled in, then the only logical conclusion that you can end up with
> is that neither the SGX core should exist in vmlinux in the first place.
>
> This all summarizes to that I have to remove the INTEL_SGX_DRIVER
> kconfig flag. Its existence can only be considered when there >= 2
> in-kernel users for SGX.
The subdirectory for the driver has clearly turned into legacy with only
two source files and one header file in it.
Even if/when we want to add INTEL_SGX_DRIVER back it is easiest use it
in the main makefile.
I prepared snippets to execute these changes.
This is for driver.h, ioctl.c, main.c (rename):
git filter-branch -f --index-filter 'git ls-files -s | \
sed "s/sgx\/driver\/main\.c/sgx\/driver.c/" | \
GIT_INDEX_FILE=$GIT_INDEX_FILE.new git update-index --index-info && \
mv "$GIT_INDEX_FILE.new" "$GIT_INDEX_FILE"' mainline/master..master
This for for Makefile (remove):
git filter-branch -f --index-filter 'git ls-files -s | \
sed "/sgx\/driver\/Makefile/d" | \
GIT_INDEX_FILE=$GIT_INDEX_FILE.new git update-index --index-info && \
mv "$GIT_INDEX_FILE.new" "$GIT_INDEX_FILE"' mainline/master..master
/Jarkko
