Return-path: Received: from mail-qa0-f51.google.com ([209.85.216.51]:34403 "EHLO mail-qa0-f51.google.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1753773AbaF0Ojr (ORCPT ); Fri, 27 Jun 2014 10:39:47 -0400 MIME-Version: 1.0 Date: Fri, 27 Jun 2014 16:39:42 +0200 Message-ID: (sfid-20140627_163953_850269_AAC37D9E) Subject: mwifiex card reset From: Andreas Fenkart To: Bing Zhao , Ulf Hansson , "linux-wireless@vger.kernel.org" , linux-mmc , devicetree@vger.kernel.org Cc: Tony Lindgren , Daniel Mack Content-Type: text/plain; charset=UTF-8 Sender: linux-wireless-owner@vger.kernel.org List-ID: I have an mwifiex module(sd8787) behind omap_hsmmc(am33xx-soc) The module is non-removable wired fix to soc. Now the wifi module needs 2 clocks; one is a sleep clock the other used when not sleeping. While the sleep clock is fixed, @32 kHz, the system clock can be one out of several values, but can be be derived automatically from the sleep clock. But this requires the clock to be present when the wifi module comes out of reset. Another problem is software reboot, at initial power on the wifi card will react to mmc discovery. After a software reset of the host it will not, because it has already been disvovered and didn't notice the host rebooted, unless we reset it as well. While this seems obvious, the problem is that the reset is needed before the card can be discovered. Which means we cannot move the reset logic into the mwifiex driver, since that driver has not yet been selected through vendor/product id. the reset logic: gpiod_set_value(card->gpiod_reset, 0); clk_enable(card->sleep_clock); udelay(1000); gpiod_set_value(card->gpiod_reset, 1); The idea so far was to extend the device-tree node of the mmc slot by some reset leafs; &mmc { ti,non-removable; .. peripheral-clocks = <&clk &clk2 ...>; peripheral-reset-gpios = <&gpio0 1 1 &gpio1 2 3 ...>; } in mmc_add_host, all clocks will be enabled and gpios be pulled low for 1 msec Is this a feasible solution? Another related issue, is the card reset in mwifiex driver: static void sdio_card_reset_worker(struct work_struct *work) { struct mmc_host *target = reset_host; pr_err("Resetting card...\n"); mmc_remove_host(target); /* 20ms delay is based on experiment with sdhci controller */ mdelay(20); mmc_add_host(target); } static DECLARE_WORK(card_reset_work, sdio_card_reset_worker); There are obviously a lot of problems with this, e.g. custom debugfs entries created in the driver will be lost. But sometimes this code might avert disaster in case the command handlers between driver and firmware lost synchronization How could this be done in a better way?