Compiling for RISC-V

Tue Mar 10 20:31:29 UTC 2020

Hi Richard,

You're right, in fact, some newer releases of Debian even seem to ship with Configuration files having entries for riscv platforms explicitly!

I've realized that the problems I was seeing were more nuanced than I originally thought.  The RISC-V toolchain actually has two flavors: one that is newlib-based and one that is linux-gnu-based.  I am using the newlib toolchain, its an embedded environment, where in fact I don't expect to have any OS support at all.  The errors I was seeing were the result of some extra flags needed to allow newlib to play nicely with the OpenSSL build, such as nostdlib.  However, of course there are still errors associated with the syscalls that aren't implemented in this case, but I believe that is now the only remaining issue.

This brings me to another point, which may be outside the scope of this thread and please let me know if I should open another one to address this: for constrained environments without an OS, is there a best practice for compiling OpenSSL? I actually don't even need *all* of OpenSSL, just the crypto, even more specifically, the RSA encrypt/decrypt.  For syscall functionality, do folks typically pull them from open-source kernels as an easy way to get up-and-running?  Any kind of insights on this would be great.

Thanks,
Kristin

On 3/9/20, 9:30 PM, "openssl-users on behalf of Richard Levitte" <openssl-users-bounces at openssl.org on behalf of levitte at openssl.org> wrote:

    Hi,

    when you mentioned cross compiling, that got me a bit more curious, so
    I went looking, and noticed that Debian [sid] (which is what I run on
    my laptop) has all the cross compiling tools I needed (see
    https://wiki.debian.org/RISC-V#Cross_compilation), so I installed
    them, and then tried this in my checkout of OpenSSL's master branch:

        ./Configure linux-generic64 no-asm no-threads \
                    --cross-compile-prefix=riscv64-linux-gnu-

    Running 'make' was a breeze, it went through flawlessly.

    I haven't done much further tests, though.

    Cheers,
    Richard

    On Mon, 09 Mar 2020 20:23:09 +0100,
    Kristin Barber wrote:
    > 
    > 
    > I did also try configuring for "no-asm", but there still seemed to be architecture-specific issues
    > based on which files the errors were coming from.  I should probably also mention that I am
    > attempting to cross-compile for RV64 from an x86 machine.
    > 
    > On Mon, Mar 9, 2020 at 3:12 PM Scott Neugroschl <scott_n at xypro.com> wrote:
    > 
    >      
    >    
    >     Is the “no-asm” configuration option still supported?
    >    
    >      
    >    
    >     From: openssl-users <openssl-users-bounces at openssl.org> On Behalf Of Kristin Barber
    >     Sent: Monday, March 9, 2020 12:03 PM
    >     To: Richard Levitte <levitte at openssl.org>
    >     Cc: openssl-users at openssl.org
    >     Subject: Re: Compiling for RISC-V
    >    
    >      
    >    
    >     Hi Richard, thanks for the reply. It was helpful.
    >    
    >      
    >    
    >     You are correct, I was able to find a configuration that worked by passing the RISC-V compiler
    >     via "make variable" assignment, along with some relevant options.  Things start compiling, but
    >     the build fails on what seems to be architecture-specific assembly files which are selected
    >     based on which "platform" has been configured.  It did not seem to me that there were RISC-V
    >     assembly-specific files as an option here, and based on your reply, I think that is indeed the
    >     issue.  Am I understanding this correctly?
    >    
    >      
    >    
    >     Thanks,
    >    
    >      
    >    
    >     Kristin
    >    
    >      
    >    
    >     On Mon, Mar 9, 2020 at 3:03 AM Richard Levitte <levitte at openssl.org> wrote:
    >    
    >         On Mon, 09 Mar 2020 05:18:17 +0100,
    >         Kristin Barber wrote:
    >         > I've looked at the INSTALL docs, and it doesn't seem that RISC-V processors are
    >         supported
    >         > currently as a platform. Is this correct?
    >        
    >         That is correct.  No one has implemented that support yet.
    >        
    >         > Is there a branch which enables configuring for a RISC-V machine that hasn't yet made it
    >         into a
    >         > stable release?  
    >        
    >         Not that I know of.  Although, this same question has also been raised
    >         on github (I forget the issue number).
    >        
    >         > Any advice on where to look for information or changes to the build process in order to
    >         compile
    >         > for RISC-V?
    >        
    >         The first thing to attempt is a generic build with no assembler.
    >         There are some really simply config targets that could be a first
    >         step, one of:
    >        
    >             ./Configure cc
    >        
    >             ./Configure gcc
    >        
    >         A (pretty big) step up from that, at least if Linux is your target,
    >         would be one of these:
    >        
    >             ./Configure linux-generic32
    >        
    >             ./Configure linux-generic64
    >        
    >         Note that in either case, you may have to add C flags and ld flags,
    >         which you can do in one of two ways:
    >        
    >         1)  directly on the configuration command line, like this (Configure
    >             makes an educated guess on what flags go where):
    >        
    >             ./Configure linux-generic64 -m64 -DWHATEVER=value -Wl,-something
    >        
    >         2)  via "make variable" assignment:
    >        
    >             ./Configure linux-generic64 \
    >                         CPPFLAGS='-DWHATEVER=value' \
    >                         CFLAGS='-m64' \
    >                         LDFLAGS='-Wl,-something'
    >        
    >         At some point, you might find a combination that works for you.  We
    >         would definitely like to know what you figure out, and it may be that
    >         the result makes it into our database of config targets (which, if
    >         you're curious, are the files Configurations/*.conf).
    >        
    >         Now, configuration is the easy bit when it comes to new CPUs,
    >         relatively speaking.  I assume that part of your question is whether
    >         there is assembler support.  This is the hard part in terms of
    >         effort.  We currently have no such thing at all for RISC-V, and I
    >         haven't seen any attempts to start such an effort...  PRs would
    >         certainly be welcome, but anyone who tries this will have to be
    >         prepared for it to take a while to get into the main source.
    >        
    >         Cheers,
    >         Richard
    >        
    >         --
    >         Richard Levitte         levitte at openssl.org
    >         OpenSSL Project         http://www.openssl.org/~levitte/
    > 
    > 
    -- 
    Richard Levitte         levitte at openssl.org
    OpenSSL Project         http://www.openssl.org/~levitte/