Bhyve with UEFI edk2-stable201903 and CSM support - debugging the crash
Rodney W. Grimes
freebsd-rwg at gndrsh.dnsmgr.net
Thu Apr 4 06:27:08 UTC 2019
> I've spent some time trying to figure out where and why the UEFI 2.70
> firmware is crashing when built with CSM support under Bhyve.
>
> The first thing I noticed is that the code that builds the E820 table
> doesn't appear to be finding the location of the EBDA at 0x40E: the
> first entry should be something like 0x0 to 0x4cf00, but is 0-0 instead.
Since your building a bios there should be a table some
place in some header or c file that builds the EBDA and
puts a pointer to it at 0x40E. This should be someplace
in the CSM legacy support stuff since the EBSA is a legacy
bios thingy.
You may want to look at the prior port of UEFI/CSM and see
how this was done in that code, it may be a hard coded hack.
> E820[ 0]: 0x?????????????? 0 ---- 0x?????????????? 0, Type = 0x1
> E820[ 1]: 0x?????????????? 0 ---- 0x?????????? A0000, Type = 0x2
> E820[ 2]: 0x?????????? E0000 ---- 0x????????? 100000, Type = 0x2
> E820[ 3]: 0x????????? 100000 ---- 0x????????? 406000, Type = 0x1
>
> ...
>
>
> As an aside, I'm working on a patch to submit upstream to fix the output
> so it pads with zeros instead of spaces.
>
>
> I've also found that as was probably pretty obvious the system is
> crashing when trying to call into the CSM to initialize it. It gets as
> far as calling AsmThunk16 (AsmThunk.nasm), but never appears to reach
> CsmMain in BhyveCsm16.c. I'll see if I can work on it some more to
> narrow down the error.
Well I would think if the legacy memory region table is bogus
that would be a good start point to fix.
>
> Another thing: the GNUmakefile in OvmfPkg/Csm/BhyveCsm16 has "CC ?=
> gcc", but it appears the build system already sets CC to "cc", which
> uses the system's clang compiler, which can't compile the assembly code.
>
> I don't know if we might want to either hard-code CC to "gcc" for now,
> or have users create a 'cc' symlinks in BaseTools/Bin/FreeBSD-amd64 ?
I would firmcode it, if CC=cc then CC=gcc, this would allow a person
to override CC with something else and still have all the other stuff
work out, except for this specific point that does infact need gcc.
> Rebecca Cran
--
Rod Grimes rgrimes at freebsd.org
More information about the freebsd-virtualization
mailing list