From mboxrd@z Thu Jan 1 00:00:00 1970 From: Christoph Hellwig Subject: Re: REQ_META performance impact on eMMC Date: Sun, 21 Jun 2015 07:02:33 -0700 Message-ID: <20150621140233.GA28075@infradead.org> References: <26E7A31274623843B0E8CF86148BFE32011C36E354@NTXBOIMBX08.micron.com> Mime-Version: 1.0 Content-Type: text/plain; charset=us-ascii Cc: "linux-fsdevel@vger.kernel.org" , Alex Lemberg , "ulf.hansson@linaro.org" , "tlinder@codeaurora.org" To: "Luca Porzio (lporzio)" Return-path: Received: from bombadil.infradead.org ([198.137.202.9]:34877 "EHLO bombadil.infradead.org" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1752559AbbFUOCf (ORCPT ); Sun, 21 Jun 2015 10:02:35 -0400 Content-Disposition: inline In-Reply-To: <26E7A31274623843B0E8CF86148BFE32011C36E354@NTXBOIMBX08.micron.com> Sender: linux-fsdevel-owner@vger.kernel.org List-ID: On Fri, Jun 19, 2015 at 04:06:49PM +0000, Luca Porzio (lporzio) wrote: > This performance impact comes from the existing MMC SW driver solution where each IO request marked with REQ_META is handled as a "Reliable Write" operation. > Pretty much as a as a FUA write access, a "Reliable write" means that the data will be written to the non-volatile memory. > Due to the fact that every REQ_META is immediately bypassing the internal cache of eMMC device, write performance is affected significantly. > > Please advise how critical is to send REQ_META as "Reliable Write"? Not at all. > Can REQ_META be sent as a regular Write operation? Yes, and it should. The MMC driver is doing something completely dumb here. -- To unsubscribe from this list: send the line "unsubscribe linux-fsdevel" in