Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 2620:137:e000::1:20 as permitted sender) client-ip=2620:137:e000::1:20;
Date:   Sat, 9 Apr 2022 15:38:51 +0300
From:   Ivan Bornyakov <i.bornyakov@metrotek.ru>
To:     Xu Yilun <yilun.xu@intel.com>
Cc:     mdf@kernel.org, hao.wu@intel.com, trix@redhat.com,
        conor.dooley@microchip.com, robh+dt@kernel.org,
        krzysztof.kozlowski+dt@linaro.org, system@metrotek.ru,
        linux-kernel@vger.kernel.org, linux-fpga@vger.kernel.org,
        devicetree@vger.kernel.org
Subject: Re: [PATCH v9 1/3] fpga: fpga-mgr: support bitstream offset in image
 buffer
Message-ID: <20220409123851.tcjflctnuurag2yb@x260>
References: <20220407133658.15699-1-i.bornyakov@metrotek.ru>
 <20220407133658.15699-2-i.bornyakov@metrotek.ru>
 <20220409050423.GA265355@yilunxu-OptiPlex-7050>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <20220409050423.GA265355@yilunxu-OptiPlex-7050>
Precedence: bulk

On Sat, Apr 09, 2022 at 01:04:23PM +0800, Xu Yilun wrote:
> On Thu, Apr 07, 2022 at 04:36:56PM +0300, Ivan Bornyakov wrote:
> > 
> > ... snip ...
> > 
> > @@ -148,12 +149,10 @@ static int fpga_mgr_write_init_buf(struct fpga_manager *mgr,
> >  	int ret;
> >  
> >  	mgr->state = FPGA_MGR_STATE_WRITE_INIT;
> > -	if (!mgr->mops->initial_header_size)
> > -		ret = fpga_mgr_write_init(mgr, info, NULL, 0);
> > -	else
> > -		ret = fpga_mgr_write_init(
> > -		    mgr, info, buf, min(mgr->mops->initial_header_size, count));
> > +	if (mgr->mops->initial_header_size)
> > +		count = min(mgr->mops->initial_header_size, count);
> >  
> > +	ret = fpga_mgr_write_init(mgr, info, buf, count);
> 
> Here we pass the whole buffer for write_init(). Maybe it works for mapped buf,
> but still doesn't work for sg buf.
> 
> It is also inefficient if we change to map and copy all sg buffers just for
> write_init().
> 
> We could discuss on the solution.
> 
> My quick mind is we introduce an optional fpga_manager_ops.parse_header()
> callback, and a header_size (dynamic header size) field in
> fpga_image_info. FPGA core starts with mapping a buf of initial_header_size
> for parse_header(), let the drivers decide the dynamic header_size.
> 
> The parse_header() could be called several times with updated dynamic
> header_size, if drivers doesn't get enough buffer for final decision and
> return -EAGAIN.
> 
> Then write_init() be called with the final dynamic header size.
> 
> For mapped buffer, just passing the whole buffer for write_init() is
> fine.
> 

Ok, that's sounds reasonable. Should we pass PAGE_SIZE of a buffer to
the parse_header() if initial_header_size is not set? The other option
would be to make initial_header_size to be mandatory for usage of
parse_header().

> > 
> > ... snip ...
> > 
> > @@ -98,6 +101,8 @@ struct fpga_image_info {
> >  	struct sg_table *sgt;
> >  	const char *buf;
> >  	size_t count;
> > +	size_t bitstream_start;
> 
> How about we name it header_size?
> 
> > +	size_t bitstream_size;
> 
> And how about data_size?
> 

Sure, I don't mind.