DPDK patches and discussions
 help / color / mirror / Atom feed
From: Bruce Richardson <bruce.richardson@intel.com>
To: Neil Horman <nhorman@tuxdriver.com>
Cc: pbhagavatula@marvell.com, jerinj@marvell.com, dev@dpdk.org,
	stable@dpdk.org
Subject: Re: [dpdk-dev] [PATCH] buildtools: fix pmdinfogen compilation
Date: Wed, 31 Jul 2019 15:21:46 +0100	[thread overview]
Message-ID: <20190731142145.GC1705@bricha3-MOBL.ger.corp.intel.com> (raw)
In-Reply-To: <20190731113503.GB9823@hmswarspite.think-freely.org>

On Wed, Jul 31, 2019 at 07:35:03AM -0400, Neil Horman wrote:
> On Wed, Jul 31, 2019 at 11:57:05AM +0530, pbhagavatula@marvell.com wrote:
> > From: Pavan Nikhilesh <pbhagavatula@marvell.com>
> > 
> > Pmdinfogen is always compiled with host gcc.
> > If host gcc version is lessthan 7 and target gcc is greaterthan 7
> > pmdinfogen fails to compile due to unsupported cflags.
> > This patch removes unsupported host cflags when the above condition is
> > met.
> > 
> > Fixes: 98b0fdb0ffc6 ("pmdinfogen: add buildtools and pmdinfogen utility")
> > Cc: stable@dpdk.org
> > 
> > Signed-off-by: Pavan Nikhilesh <pbhagavatula@marvell.com>
> > ---
> >  buildtools/pmdinfogen/Makefile | 8 ++++++++
> >  1 file changed, 8 insertions(+)
> > 
> > diff --git a/buildtools/pmdinfogen/Makefile b/buildtools/pmdinfogen/Makefile
> > index a97a7648f..86f883e05 100644
> > --- a/buildtools/pmdinfogen/Makefile
> > +++ b/buildtools/pmdinfogen/Makefile
> > @@ -9,6 +9,14 @@ include $(RTE_SDK)/mk/rte.vars.mk
> >  #
> >  HOSTAPP = dpdk-pmdinfogen
> >  
> > +HOST_GCC_MAJOR = $(shell echo __GNUC__ | $(HOSTCC) -E -x c - | tail -n 1)
> > +HOST_GCC_MINOR = $(shell echo __GNUC_MINOR__ | $(HOSTCC) -E -x c - | tail -n 1)
> > +HOST_GCC_VERSION = $(HOST_GCC_MAJOR)$(HOST_GCC_MINOR)
> > +
> > +ifeq ($(shell test $(HOST_GCC_VERSION) -gt 70 && echo 1), 1)
> > +HOST_WERROR_FLAGS = $(filter-out -Wimplicit-fallthrough=2, $(WERROR_FLAGS))
> > +endif
> > +
> 
> A few things:
> 
> 1) HOST_GCC_MAJOR and HOST_GCC_MINOR seem to already be computed in
> rte.toolchain-compat.mk and so I don't think you need to recompute them here
> 
> 2) This seems limited in its function.  That is to say, ostensibly there are
> simmilar incompatibilities with icc and clang that may need addressing for which
> there are different environment variables (CLANG_MAJOR_VERSION, etc)
> 
> 3) This may also need to be reflected into the meson build environment

Initially I though that meson cross-compilation support would make this a
non-issue, but looking into it further it could theoretically be a problem
with meson too. Where the issue would arise is where we use
"add_project_arguments()" in "config/meson.build" for the cflags. When
adding those flags we only check the standard compiler, which would be the
cross-compiler in the cross compilation case. Unfortunately, I don't
believe that meson supports setting project options only for native or
non-native compiles - it assumes that per-project arguments are really
global to that whole project, and expect you to specify them per-target if
not.

To fix this possible issue for cross-compilation, what we really need to do
in to get the native compiler too (using "meson.get_compiler('c', native:
'true')") and check that the cflags are valid for it also. In case of a
flag that is supported by one compiler but not the other, the flag would be
omitted, which means that instead of a compiler error we should instead get
an error with reduced warning flags. (I'm assuming that it's only the
warnings need adjusting here - any compiler that doesn't support
"-D<define>" syntax just won't work anyway, I suspect :-))

I'll see if I can do up a quick patch to fix this for meson.


> 
> 4) I'm not sure how this is a problem at all.  In your description, you indicate
> that:
> 	a) pmdinfogen is always compiled with host gcc
> 	b) this fails when the target gcc uses a compiler version older than the 
> 	   host gcc version, leading to corresponding flag incompatibilities
> If (a) is true, why does (b) matter?  If we are using the host gcc, then the
> host gcc flags should work.  If we're mixing and matching host gcc and target
> gcc flags, that seems like a bug that should be fixed in the target rte.vars.mk
> 
Yes, this is the issue - the mixing of host and target gcc flags.

/Bruce

  reply	other threads:[~2019-07-31 14:22 UTC|newest]

Thread overview: 6+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2019-07-31  6:27 pbhagavatula
2019-07-31 11:35 ` Neil Horman
2019-07-31 14:21   ` Bruce Richardson [this message]
2019-07-31 14:30     ` Bruce Richardson
2019-07-31 14:31 ` Bruce Richardson
2019-08-02  3:35   ` [dpdk-dev] [EXT] " Pavan Nikhilesh Bhagavatula

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=20190731142145.GC1705@bricha3-MOBL.ger.corp.intel.com \
    --to=bruce.richardson@intel.com \
    --cc=dev@dpdk.org \
    --cc=jerinj@marvell.com \
    --cc=nhorman@tuxdriver.com \
    --cc=pbhagavatula@marvell.com \
    --cc=stable@dpdk.org \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).