Date:   Mon, 25 Nov 2019 16:08:07 -0400
From:   Jason Gunthorpe <jgg@ziepe.ca>
To:     Andrew Boyer <aboyer@pensando.io>
Cc:     linux-rdma@vger.kernel.org, allenbh@pensando.io
Subject: Re: [PATCH] rdma-core: Recognize IBV_DEVICE_LOCAL_DMA_LKEY in
 ibv_devinfo
Message-ID: <20191125200807.GF11270@ziepe.ca>
References: <20191125152237.19084-1-aboyer@pensando.io>
 <20191125175603.GB11270@ziepe.ca>
 <77FA0F51-7A7D-42E3-AC2A-BF65550A4396@pensando.io>
MIME-Version: 1.0
Content-Type: text/plain; charset=utf-8
Content-Disposition: inline
Content-Transfer-Encoding: 8bit
In-Reply-To: <77FA0F51-7A7D-42E3-AC2A-BF65550A4396@pensando.io>
User-Agent: Mutt/1.9.4 (2018-02-28)
Sender: linux-rdma-owner@vger.kernel.org
Precedence: bulk

On Mon, Nov 25, 2019 at 02:32:31PM -0500, Andrew Boyer wrote:
> 
> > On Nov 25, 2019, at 12:56 PM, Jason Gunthorpe <jgg@ziepe.ca> wrote:
> > 
> > On Mon, Nov 25, 2019 at 07:22:37AM -0800, Andrew Boyer wrote:
> >> This bit is defined in the kernel but not displayed by ibv_devinfo.
> >> 
> >> Signed-off-by: Andrew Boyer <aboyer@pensando.io>
> >> libibverbs/examples/devinfo.c | 3 +++
> >> libibverbs/verbs.h            | 1 +
> >> 2 files changed, 4 insertions(+)
> >> 
> >> diff --git a/libibverbs/examples/devinfo.c b/libibverbs/examples/devinfo.c
> >> index bf53eac2..e3210f6e 100644
> >> +++ b/libibverbs/examples/devinfo.c
> >> @@ -220,6 +220,7 @@ static void print_device_cap_flags(uint32_t dev_cap_flags)
> >> 				   IBV_DEVICE_RC_RNR_NAK_GEN |
> >> 				   IBV_DEVICE_SRQ_RESIZE |
> >> 				   IBV_DEVICE_N_NOTIFY_CQ |
> >> +				   IBV_DEVICE_LOCAL_DMA_LKEY |
> >> 				   IBV_DEVICE_MEM_WINDOW |
> >> 				   IBV_DEVICE_UD_IP_CSUM |
> >> 				   IBV_DEVICE_XRC |
> >> @@ -260,6 +261,8 @@ static void print_device_cap_flags(uint32_t dev_cap_flags)
> >> 		printf("\t\t\t\t\tSRQ_RESIZE\n");
> >> 	if (dev_cap_flags & IBV_DEVICE_N_NOTIFY_CQ)
> >> 		printf("\t\t\t\t\tN_NOTIFY_CQ\n");
> >> +	if (dev_cap_flags & IBV_DEVICE_LOCAL_DMA_LKEY)
> >> +		printf("\t\t\t\t\tLOCAL_DMA_LKEY\n");
> >> 	if (dev_cap_flags & IBV_DEVICE_MEM_WINDOW)
> >> 		printf("\t\t\t\t\tMEM_WINDOW\n");
> >> 	if (dev_cap_flags & IBV_DEVICE_UD_IP_CSUM)
> >> diff --git a/libibverbs/verbs.h b/libibverbs/verbs.h
> >> index 7b8d4310..81e5812c 100644
> >> +++ b/libibverbs/verbs.h
> >> @@ -112,6 +112,7 @@ enum ibv_device_cap_flags {
> >> 	IBV_DEVICE_RC_RNR_NAK_GEN	= 1 << 12,
> >> 	IBV_DEVICE_SRQ_RESIZE		= 1 << 13,
> >> 	IBV_DEVICE_N_NOTIFY_CQ		= 1 << 14,
> >> +	IBV_DEVICE_LOCAL_DMA_LKEY	= 1 << 15,
> >> 	IBV_DEVICE_MEM_WINDOW           = 1 << 17,
> >> 	IBV_DEVICE_UD_IP_CSUM		= 1 << 18,
> >> 	IBV_DEVICE_XRC			= 1 << 20,
> > 
> > This flag really only has meaning for the kernel, it should come out
> > of the uapi at all.
> > 
> > It is a mistake that kernel internal bits have been mixed in with
> > userspace bits.
> > 
> > Jason
> 
> Isn’t there value in having the userspace tools tell the user about
> a device’s in-kernel capabilities?

Sure, but that is more a rdmatool thing, not uverbs

Jason