Re: [PATCH] gdb/riscv: Add target description support

[Andrew Burgess <andrew.burgess@embecosm.com>]

* Jim Wilson <jimw@sifive.com> [2019-02-22 11:24:15 -0800]:

> On Fri, Feb 22, 2019 at 9:42 AM Tom Tromey <tom@tromey.com> wrote:
> > With an older gdb (8.2), with remote debugging enabled:
> >
> >     (gdb) p $fflags
> >     Sending packet: $p42#d6...Ack
> >     Packet received: 0000000000000000
> >     $1 = 0
> >...
> > However with a newer gdb, with an older qemu, I get a failure:
> >
> >     Sending packet: $p41#d5...Ack
> >     Packet received: E14
> >     Could not fetch register "fflags"; remote failure reply 'E14'
> >
> > Here you can see gdb is sending 0x41.
> 
> The renumbering is unfortunate, but there are a lot of historical
> problems here, and compatibility with old tools may not be worth the
> trouble.  We need to make this work right first, before we can worry
> about backward compatibility.
> 
> fflags is a CSR not an F register.  But gdb is treating it as both an
> F register and a CSR which may lead to some confusion.

I think we should be OK.  The existing CSR feature file should
possibly be deleted, it's not actually used right now.

When targets supply a target description then the fflags, frm, and
fcsr should only be in one feature otherwise (I think, but haven't
tested) GDB will get upset.  However, I don't think we should prevent
bundling the FP status registers in the FP feature set, personally, I
think that the argument for having them there is almost as strong as
for having them in with the CSRs.

> 
> Also, fflags is not a real register.  It is an alias for a field
> inside the fcsr register.  If gdb is talking to something like
> openocd/qemu that can translate fflags to fcsr that is OK, but for a
> native riscv-linux gdb trying to read fflags won't do anything useful,
> gdb needs to translate fflags into the fcsr field itself or this isn't
> going to work.  Though maybe we can do something inside one of the
> riscv-linux files to handle this.

I started working on, and do plan to finish soon(ish) a change where
if GDB finds an $fcsr register, but no $fflags or $frm then a
pseudo-register would be setup (for each) that just reads/writes the
bits for $fcsr.

This would work for all targets, so the native linux case should be
covered.

> 
> There are a lot of problems with the old qemu gdbstub support.  I have
> a patch set submitted for qemu to add the gdb xml files and proper
> gdbstub support to use them.  gdb+qemu works much better with these
> patches.  This patch set has been approved, and is queued for commit
> behind one other patch in progress I think.  With this patch,
> accessing fflags works with both a system qemu and a user qemu using
> the current development gdb (8.3 pre-release).

That's great news.  Thanks for all your effort on this.

Thanks,
Andrew


> 
> (gdb) info registers fflags
> fflags         Sending packet: $p4c#07...Ack
> Packet received: 0000000000000000
> Packet p (fetch-register) is supported
> 0x0 RD:0 NV:0 DZ:0 OF:0 UF:0 NX:0
> (gdb)
> 
> Jim