From mboxrd@z Thu Jan 1 00:00:00 1970 From: Y.b. Lu Date: Thu, 24 Aug 2017 03:54:42 +0000 Subject: [U-Boot] [PATCH] blk: Remove various places that do flush cache after read In-Reply-To: References: <1503373600-20847-1-git-send-email-bmeng.cn@gmail.com> Message-ID: List-Id: MIME-Version: 1.0 Content-Type: text/plain; charset="us-ascii" Content-Transfer-Encoding: 7bit To: u-boot@lists.denx.de > -----Original Message----- > From: Bin Meng [mailto:bmeng.cn at gmail.com] > Sent: Wednesday, August 23, 2017 1:29 PM > To: Y.b. Lu > Cc: Qiang Zhao; York Sun; Grygorii Strashko; Michal Simek; Angelo > Dureghello; Kever Yang; Jaehoon Chung; Albert Aribaud; Marcel Ziswiler; > Markus Niebel; Z.q. Hou; Sumit Garg; Joe Hershberger; Stefan Roese; > Wenbin Song; Simon Glass; Max Krummenacher; Masahiro Yamada; Tomas Melin; > Q.y. Gong; Tom Rini; U-Boot Mailing List > Subject: Re: [PATCH] blk: Remove various places that do flush cache after > read > > Hi Yangbo, > > On Wed, Aug 23, 2017 at 11:29 AM, Y.b. Lu wrote: > > cmd/mmc.c > > > > Maybe SD controller on some vendor platforms doesn't support SNOOP. > > flush_cache() is required to sync between DDR and cache. > > Should be careful to remove this. > > > > Please do not top-posting. > > What you mentioned if snoop is not supported, flush_cache() is required, > is true, however that's not the issue I am trying to fix here. The > flush_cache() here after a DMA read will cause issues because stale data > in the cache will overwrite the correct data in the RAM that is DMAed by > hardware (in this case, MMC controller). > [Lu Yangbo-B47093] Sorry for my misunderstanding. It looks ok to me now. flush_cache is indeed redundant after DMA reading. > Regards, > Bin