From mboxrd@z Thu Jan 1 00:00:00 1970 From: Adrian Hunter Subject: Re: [PATCH 08/10] mmc: card: Use R1 response for the stop cmd at recovery path Date: Mon, 27 Jan 2014 12:40:33 +0200 Message-ID: <52E637A1.8060602@intel.com> References: <1390402824-9850-1-git-send-email-ulf.hansson@linaro.org> <1390402824-9850-9-git-send-email-ulf.hansson@linaro.org> <52E0EA70.9010308@intel.com> <52E1274D.3050200@intel.com> Mime-Version: 1.0 Content-Type: text/plain; charset=ISO-8859-1 Content-Transfer-Encoding: 7bit Return-path: Received: from mga09.intel.com ([134.134.136.24]:2605 "EHLO mga09.intel.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1751827AbaA0Kiu (ORCPT ); Mon, 27 Jan 2014 05:38:50 -0500 In-Reply-To: Sender: linux-mmc-owner@vger.kernel.org List-Id: linux-mmc@vger.kernel.org To: Ulf Hansson Cc: linux-mmc , Chris Ball , Dong Aisheng , Stephen Warren , Vladimir Zapolskiy On 23/01/14 16:59, Ulf Hansson wrote: > On 23 January 2014 15:29, Adrian Hunter wrote: >> On 23/01/14 15:21, Ulf Hansson wrote: >>> On 23 January 2014 11:09, Adrian Hunter wrote: >>>> On 22/01/14 17:00, Ulf Hansson wrote: >>>>> Hosts supporting MMC_CAP_WAIT_WHILE_BUSY shall not be waiting for busy >>>>> detection completion in the recovery path, which were the case when >>>>> using R1B response. >>>>> >>>>> Start using R1 as response instead to align behavior, no matter if >>>>> MMC_CAP_WAIT_WHILE_BUSY is supported or not. >>>> >>>> This does not make sense to me. If you are sending a STOP command you >>>> should use the correct response type. R1B should be OK here because the >>>> card should release the busy signal in any case except failure. >>> >>> For those hosts not supporting MMC_CAP_WAIT_WHILE_BUSY a R1B is >>> assumed to be treated same as an R1, which means there are no busy >>> detection handled in the host. >> >> That is not entirely true. For hosts that do not set >> MMC_CAP_WAIT_WHILE_BUSY we don't know if they wait or not. I imagine most >> do because it is more efficient, but the kernel has always been programmed >> to poll the status anyway so you can't tell from the code. > > You are right, we can't know - unless we dive in into each host driver > and check. > > Surely there could be more than omap_hsmmc and sdhci that support > this. Still I think we need to conclude on how to go forward with > MMC_CAP_WAIT_WHILE_BUSY, since at the moment it seems a bit of a mess. > Obviously we need to be careful to not break anything. > >> >> MMC_CAP_WAIT_WHILE_BUSY was one of my inventions I am afraid. If I recall >> correctly it was mainly due to the SLEEP command because you can't poll in >> that case and you don't want to delay the system from sleeping - if you are >> certain that the controller has waited for busy to de-assert (i.e. >> MMC_CAP_WAIT_WHILE_BUSY) then you can exit immediately. > > I think MMC_CAP_WAIT_WHILE_BUSY was a needed feature, now we only have > to make it more mature. :-) > >> >>> >>> mmc_blk_cmd_recovery() is the only caller of the send_stop() function. >>> Additionally it does not care about to handle busy detection with >>> CDM13 polling. >>> >>> Now, since most hosts don't support MMC_CAP_WAIT_WHILE_BUSY which >>> means there no busy detection done, I wanted to align to this >>> behaviour - no matter if the host can do HW busy detection or not. >>> >>> I am not saying this is how it must be done, just trying to provide >>> you with some more reasons to why I wanted to change. >>> >>> If we instead decide keep the R1B for send_stop(), we should implement >>> CMD 13 polling to meet the same behaviour for hosts not supporting >>> MMC_CAP_WAIT_WHILE_BUSY. In this scenario, we need to set a select a >>> busy timeout, do you have any suggestion of what would be a reasonable >>> value for it? >> >> It is hard to tell if waiting is ever going to help more than hinder, so I >> would not change this. > > Fair enough, but certainly we should implement a CMD13 polling > mechanism - to align behaviour. Recovery probably isn't possible. The block driver heroically has a go at it. For some people it much more important to fail fast than to recover. Consequently, unless you has a specific use-case, I wouldn't add anything that would slow down that path. > > Are you then also indirectly suggesting that not specficing > "cmd.busy_timeout" should be interpreted by the host as "use whatever > timeout you want"? That is how it is now. The problem with trying to so something better is that sometimes the timeout really is undefined. > > Do note, there are another scenario, which also don't specify a busy > timeout, which is when we have used an open ended WRITE transmission > and using CMD12 to finalize it. > But, in this scenario we do polling with CMD13, also without a > timeout. So at least the behaviour are aligned here, but still no > timeout specified. I don't think that is right. The data timeout applies in that case too. > >> >>> >>> Kind regards >>> Ulf Hansson >>> >>>> >>>>> >>>>> Signed-off-by: Ulf Hansson >>>>> --- >>>>> drivers/mmc/card/block.c | 2 +- >>>>> 1 file changed, 1 insertion(+), 1 deletion(-) >>>>> >>>>> diff --git a/drivers/mmc/card/block.c b/drivers/mmc/card/block.c >>>>> index 87cd2b0..74169fa 100644 >>>>> --- a/drivers/mmc/card/block.c >>>>> +++ b/drivers/mmc/card/block.c >>>>> @@ -728,7 +728,7 @@ static int send_stop(struct mmc_card *card, u32 *status) >>>>> int err; >>>>> >>>>> cmd.opcode = MMC_STOP_TRANSMISSION; >>>>> - cmd.flags = MMC_RSP_SPI_R1B | MMC_RSP_R1B | MMC_CMD_AC; >>>>> + cmd.flags = MMC_RSP_SPI_R1 | MMC_RSP_R1 | MMC_CMD_AC; >>>>> err = mmc_wait_for_cmd(card->host, &cmd, 5); >>>>> if (err == 0) >>>>> *status = cmd.resp[0]; >>>>> >>>> >>> >>> >> > >