From mboxrd@z Thu Jan 1 00:00:00 1970 From: Evan Green Subject: [PATCH RFC] mmc: sdhci: Don't get card-detect without preemption Date: Tue, 1 May 2018 16:47:21 -0700 Message-ID: <20180501234721.187135-1-evgreen@chromium.org> Return-path: Sender: linux-kernel-owner@vger.kernel.org To: Adrian Hunter , Ulf Hansson , linux-mmc@vger.kernel.org, linux-arm-msm@vger.kernel.org, linux-kernel@vger.kernel.org Cc: Evan Green List-Id: linux-arm-msm@vger.kernel.org For a controller with SDHCI_QUIRK_NO_CARD_NO_RESET, there are several conditions where sdhci_do_reset is called under a spinlock with interrupts disabled. The card detect is often a GPIO, which might sleep. Avoid asking for the card detect status if interrupts are disabled to prevent a warning like the following: [ 2.480199] Call trace: [ 2.480218] [] dump_backtrace+0x0/0x228 [ 2.480224] [] show_stack+0x20/0x28 [ 2.480235] [] dump_stack+0x90/0xb0 [ 2.480245] [] ___might_sleep+0x110/0x128 [ 2.480248] [] __might_sleep+0x78/0x88 [ 2.480260] [] gpiod_get_raw_value_cansleep+0x2c/0xc4 [ 2.480269] [] mmc_gpio_get_cd+0x3c/0x68 [ 2.480275] [] sdhci_get_cd+0x20/0x98 [ 2.480280] [] sdhci_do_reset+0x50/0x88 [ 2.480283] [] sdhci_tasklet_finish+0x1a8/0x270 [ 2.480292] [] tasklet_action+0x90/0xf4 [ 2.480296] [] __do_softirq+0x1c8/0x360 [ 2.480300] [] irq_exit+0x88/0xd0 --- I discovered this warning while trying to bring up SD, somewhat unsuccessfully, on a new device, and hit this error path. This change is not ideal as it only makes a best effort to adhere to the quirk, but may in some cases end up resetting the controller with no card present. Another option I considered was trying to call sdhci_do_reset within only sleepable contexts. I actually got as far as converting the tasklet to a work queue, and deferring the reset to the end of the function in sdhci_request_done. But 1) I'm not sure if that deferral was actually safe, and 2) we call sdhci_do_reset from under the lock in several other places, such as sdhci_card_event and sdhci_cqe_disable. Finally, I also considered calling the non-_maysleep gpio functions in mmc_gpio_get_cd. I assume this is not workable as someone somewhere has put a card detect off of a PMIC or GPIO expander. Any advice? --- drivers/mmc/host/sdhci.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/drivers/mmc/host/sdhci.c b/drivers/mmc/host/sdhci.c index 759278dd317d..c5273acf6987 100644 --- a/drivers/mmc/host/sdhci.c +++ b/drivers/mmc/host/sdhci.c @@ -210,7 +210,7 @@ static void sdhci_do_reset(struct sdhci_host *host, u8 mask) if (host->quirks & SDHCI_QUIRK_NO_CARD_NO_RESET) { struct mmc_host *mmc = host->mmc; - if (!mmc->ops->get_cd(mmc)) + if (preemptible() && !mmc->ops->get_cd(mmc)) return; } -- 2.13.5