mirror of https://lore.kernel.org/lkml/
 help / color / mirror / Atom feed
* [PATCH v3 0/3] Prevent races when doing read-modify-write of INTMASK
       [not found] <1382126905-14654-1-git-send-email-dianders@chromium.org>
@ 2013-10-18 21:55 ` Doug Anderson
  2013-10-18 21:56   ` [PATCH v3 1/3] mmc: core: Support the optional init_card() callback for MMC and SD Doug Anderson
                     ` (2 more replies)
  0 siblings, 3 replies; 4+ messages in thread
From: Doug Anderson @ 2013-10-18 21:55 UTC (permalink / raw)
  To: Jaehoon Chung, Seungwon Jeon, Chris Ball
  Cc: James Hogan, Grant Grundler, Alim Akhtar, Abhilash Kesavan,
	Tomasz Figa, Olof Johansson, Sonny Rao, Bing Zhao, Daniel Mack,
	Sascha Hauer, Doug Anderson, Namjae Jeon, linux-omap, Kevin Liu,
	linux-mmc, linux-kernel, Russell King, Johan Rudholm,
	Ulf Hansson, linux-arm-kernel, Tony Lindgren

Bing Zhao at Marvell discovered a race in the way that dw_mmc was
doing a read-modify-write of the INTMASK register.  This 2-patch
series attempts to fix the problem using a simple spinlock.  In order
to do so cleanly, we include a patch to tidy up the way that we
disable low power mode when using SDIO interrupts.

This patch series was not tested on ToT Linux other than basic
compiling and booting, since we don't have the whole Marvell SDIO
stack up and running in mainline yet.  This series is based on
mmc-next (e76b855 mmc: sdhci-esdhc-imx: set actual_clock in clock
setting) merged atop mainlinx Linux.

Changes in v3:
- Add fixup to pandora_wl1251_init_card().

Changes in v2:
- Core patch new for this version.
- Fixed "|" to "&".
- intmask_lock renamed to irq_lock

Doug Anderson (3):
  mmc: core: Support the optional init_card() callback for MMC and SD
  mmc: dw_mmc: Cleanup disable of low power mode w/ SDIO interrupts
  mmc: dw_mmc: Protect read-modify-write of INTMASK with a lock

 arch/arm/mach-omap2/board-omap3pandora.c | 14 +++---
 drivers/mmc/core/mmc.c                   |  6 +++
 drivers/mmc/core/sd.c                    |  7 ++-
 drivers/mmc/host/dw_mmc.c                | 79 +++++++++++++++++++-------------
 drivers/mmc/host/dw_mmc.h                |  1 +
 include/linux/mmc/dw_mmc.h               |  6 +++
 6 files changed, 74 insertions(+), 39 deletions(-)

-- 
1.8.4


^ permalink raw reply	[flat|nested] 4+ messages in thread

* [PATCH v3 1/3] mmc: core: Support the optional init_card() callback for MMC and SD
  2013-10-18 21:55 ` [PATCH v3 0/3] Prevent races when doing read-modify-write of INTMASK Doug Anderson
@ 2013-10-18 21:56   ` Doug Anderson
  2013-10-18 21:56   ` [PATCH v3 2/3] mmc: dw_mmc: Cleanup disable of low power mode w/ SDIO interrupts Doug Anderson
  2013-10-18 21:56   ` [PATCH v3 3/3] mmc: dw_mmc: Protect read-modify-write of INTMASK with a lock Doug Anderson
  2 siblings, 0 replies; 4+ messages in thread
From: Doug Anderson @ 2013-10-18 21:56 UTC (permalink / raw)
  To: Jaehoon Chung, Seungwon Jeon, Chris Ball
  Cc: James Hogan, Grant Grundler, Alim Akhtar, Abhilash Kesavan,
	Tomasz Figa, Olof Johansson, Sonny Rao, Bing Zhao, Daniel Mack,
	Sascha Hauer, Doug Anderson, Tony Lindgren, Russell King,
	Ulf Hansson, Johan Rudholm, Namjae Jeon, Kevin Liu,
	linux-arm-kernel, linux-omap, linux-kernel, linux-mmc

In (3fcb027 ARM: MXC: mxcmmc: work around a bug in the SDHC busy line
handling) the optional init_card() callback was added.  According to
the original change it was "for now only called from
mmc_sdio_init_card()".

This callback really ought to be called from the SD and MMC init
functions as well.  One current user of this callback
(mxcmci_init_card) will not work as expected if you insert an SDIO
card, then eject it and put a normal SD card in.  Specifically the
normal SD card will not get to run with 4-bit data.

I'd like to use the init_card() callback to handle a similar quirk on
dw_mmc when using SDIO Interrupts (the "low power" feature of the card
needs to be disabled), so that will add a second user of the function.

As part of this change fixup the one place that relied on the callback
only happening for SDIO cards.

Signed-off-by: Doug Anderson <dianders@chromium.org>
Reviewed-by: Grant Grundler <grundler@chromium.org>
---
Changes in v3:
- Add fixup to pandora_wl1251_init_card().

Changes in v2:
- New for this version.

 arch/arm/mach-omap2/board-omap3pandora.c | 14 ++++++++------
 drivers/mmc/core/mmc.c                   |  6 ++++++
 drivers/mmc/core/sd.c                    |  7 ++++++-
 3 files changed, 20 insertions(+), 7 deletions(-)

diff --git a/arch/arm/mach-omap2/board-omap3pandora.c b/arch/arm/mach-omap2/board-omap3pandora.c
index de1bc6b..6404595 100644
--- a/arch/arm/mach-omap2/board-omap3pandora.c
+++ b/arch/arm/mach-omap2/board-omap3pandora.c
@@ -254,12 +254,14 @@ static void pandora_wl1251_init_card(struct mmc_card *card)
 	 * We have TI wl1251 attached to MMC3. Pass this information to
 	 * SDIO core because it can't be probed by normal methods.
 	 */
-	card->quirks |= MMC_QUIRK_NONSTD_SDIO;
-	card->cccr.wide_bus = 1;
-	card->cis.vendor = 0x104c;
-	card->cis.device = 0x9066;
-	card->cis.blksize = 512;
-	card->cis.max_dtr = 20000000;
+	if (card->type == MMC_TYPE_SDIO || card->type == MMC_TYPE_SD_COMBO) {
+		card->quirks |= MMC_QUIRK_NONSTD_SDIO;
+		card->cccr.wide_bus = 1;
+		card->cis.vendor = 0x104c;
+		card->cis.device = 0x9066;
+		card->cis.blksize = 512;
+		card->cis.max_dtr = 20000000;
+	}
 }
 
 static struct omap2_hsmmc_info omap3pandora_mmc[] = {
diff --git a/drivers/mmc/core/mmc.c b/drivers/mmc/core/mmc.c
index 6d02012..7ad75c0 100644
--- a/drivers/mmc/core/mmc.c
+++ b/drivers/mmc/core/mmc.c
@@ -940,6 +940,12 @@ static int mmc_init_card(struct mmc_host *host, u32 ocr,
 	}
 
 	/*
+	 * Call the optional HC's init_card function to handle quirks.
+	 */
+	if (host->ops->init_card)
+		host->ops->init_card(host, card);
+
+	/*
 	 * For native busses:  set card RCA and quit open drain mode.
 	 */
 	if (!mmc_host_is_spi(host)) {
diff --git a/drivers/mmc/core/sd.c b/drivers/mmc/core/sd.c
index 5e8823d..59e2318 100644
--- a/drivers/mmc/core/sd.c
+++ b/drivers/mmc/core/sd.c
@@ -940,6 +940,12 @@ static int mmc_sd_init_card(struct mmc_host *host, u32 ocr,
 	}
 
 	/*
+	 * Call the optional HC's init_card function to handle quirks.
+	 */
+	if (host->ops->init_card)
+		host->ops->init_card(host, card);
+
+	/*
 	 * For native busses:  get card RCA and quit open drain mode.
 	 */
 	if (!mmc_host_is_spi(host)) {
@@ -1286,4 +1292,3 @@ err:
 
 	return err;
 }
-
-- 
1.8.4


^ permalink raw reply	[flat|nested] 4+ messages in thread

* [PATCH v3 2/3] mmc: dw_mmc: Cleanup disable of low power mode w/ SDIO interrupts
  2013-10-18 21:55 ` [PATCH v3 0/3] Prevent races when doing read-modify-write of INTMASK Doug Anderson
  2013-10-18 21:56   ` [PATCH v3 1/3] mmc: core: Support the optional init_card() callback for MMC and SD Doug Anderson
@ 2013-10-18 21:56   ` Doug Anderson
  2013-10-18 21:56   ` [PATCH v3 3/3] mmc: dw_mmc: Protect read-modify-write of INTMASK with a lock Doug Anderson
  2 siblings, 0 replies; 4+ messages in thread
From: Doug Anderson @ 2013-10-18 21:56 UTC (permalink / raw)
  To: Jaehoon Chung, Seungwon Jeon, Chris Ball
  Cc: James Hogan, Grant Grundler, Alim Akhtar, Abhilash Kesavan,
	Tomasz Figa, Olof Johansson, Sonny Rao, Bing Zhao, Daniel Mack,
	Sascha Hauer, Doug Anderson, linux-mmc, linux-kernel

In the patch (9623b5b mmc: dw_mmc: Disable low power mode if SDIO
interrupts are used) I added code that disabled the low power mode of
dw_mmc when SDIO interrupts are used.  That code worked but always
felt a little hacky because we ended up disabling low power as a side
effect of the first enable_sdio_irq() call.  That wouldn't be so bad
except that disabling low power involves a complicated process of
writing to the CMD/CMDARG registers and that extra process makes it
difficult to cleanly the read-modify-write race in
dw_mci_enable_sdio_irq() (see future patch in the series).

Change the code to take advantage of the init_card() callback of the
mmc core to know when an SDIO card has been inserted.  If we've got a
SDIO card and we're configured to use SDIO Interrupts then turn off
"low power mode" right away.

Signed-off-by: Doug Anderson <dianders@chromium.org>
---
Changes in v3: None
Changes in v2:
- Fixed "|" to "&".

 drivers/mmc/host/dw_mmc.c | 68 ++++++++++++++++++++++++-----------------------
 drivers/mmc/host/dw_mmc.h |  1 +
 2 files changed, 36 insertions(+), 33 deletions(-)

diff --git a/drivers/mmc/host/dw_mmc.c b/drivers/mmc/host/dw_mmc.c
index 0a6a512..6c8d6a9 100644
--- a/drivers/mmc/host/dw_mmc.c
+++ b/drivers/mmc/host/dw_mmc.c
@@ -27,6 +27,7 @@
 #include <linux/stat.h>
 #include <linux/delay.h>
 #include <linux/irq.h>
+#include <linux/mmc/card.h>
 #include <linux/mmc/host.h>
 #include <linux/mmc/mmc.h>
 #include <linux/mmc/sdio.h>
@@ -822,7 +823,7 @@ static void dw_mci_setup_bus(struct dw_mci_slot *slot, bool force_clkinit)
 
 		/* enable clock; only low power if no SDIO */
 		clk_en_a = SDMMC_CLKEN_ENABLE << slot->id;
-		if (!(mci_readl(host, INTMASK) & SDMMC_INT_SDIO(slot->id)))
+		if (!test_bit(DW_MMC_CARD_NO_LOW_PWR, &slot->flags))
 			clk_en_a |= SDMMC_CLKEN_LOW_PWR << slot->id;
 		mci_writel(host, CLKENA, clk_en_a);
 
@@ -1050,27 +1051,37 @@ static int dw_mci_get_cd(struct mmc_host *mmc)
 	return present;
 }
 
-/*
- * Disable lower power mode.
- *
- * Low power mode will stop the card clock when idle.  According to the
- * description of the CLKENA register we should disable low power mode
- * for SDIO cards if we need SDIO interrupts to work.
- *
- * This function is fast if low power mode is already disabled.
- */
-static void dw_mci_disable_low_power(struct dw_mci_slot *slot)
+static void dw_mci_init_card(struct mmc_host *mmc, struct mmc_card *card)
 {
+	struct dw_mci_slot *slot = mmc_priv(mmc);
 	struct dw_mci *host = slot->host;
-	u32 clk_en_a;
-	const u32 clken_low_pwr = SDMMC_CLKEN_LOW_PWR << slot->id;
 
-	clk_en_a = mci_readl(host, CLKENA);
+	/*
+	 * Low power mode will stop the card clock when idle.  According to the
+	 * description of the CLKENA register we should disable low power mode
+	 * for SDIO cards if we need SDIO interrupts to work.
+	 */
+	if (mmc->caps & MMC_CAP_SDIO_IRQ) {
+		const u32 clken_low_pwr = SDMMC_CLKEN_LOW_PWR << slot->id;
+		u32 clk_en_a_old;
+		u32 clk_en_a;
 
-	if (clk_en_a & clken_low_pwr) {
-		mci_writel(host, CLKENA, clk_en_a & ~clken_low_pwr);
-		mci_send_cmd(slot, SDMMC_CMD_UPD_CLK |
-			     SDMMC_CMD_PRV_DAT_WAIT, 0);
+		clk_en_a_old = mci_readl(host, CLKENA);
+
+		if (card->type == MMC_TYPE_SDIO ||
+		    card->type == MMC_TYPE_SD_COMBO) {
+			set_bit(DW_MMC_CARD_NO_LOW_PWR, &slot->flags);
+			clk_en_a = clk_en_a_old & ~clken_low_pwr;
+		} else {
+			clear_bit(DW_MMC_CARD_NO_LOW_PWR, &slot->flags);
+			clk_en_a = clk_en_a_old | clken_low_pwr;
+		}
+
+		if (clk_en_a != clk_en_a_old) {
+			mci_writel(host, CLKENA, clk_en_a);
+			mci_send_cmd(slot, SDMMC_CMD_UPD_CLK |
+				     SDMMC_CMD_PRV_DAT_WAIT, 0);
+		}
 	}
 }
 
@@ -1082,21 +1093,11 @@ static void dw_mci_enable_sdio_irq(struct mmc_host *mmc, int enb)
 
 	/* Enable/disable Slot Specific SDIO interrupt */
 	int_mask = mci_readl(host, INTMASK);
-	if (enb) {
-		/*
-		 * Turn off low power mode if it was enabled.  This is a bit of
-		 * a heavy operation and we disable / enable IRQs a lot, so
-		 * we'll leave low power mode disabled and it will get
-		 * re-enabled again in dw_mci_setup_bus().
-		 */
-		dw_mci_disable_low_power(slot);
-
-		mci_writel(host, INTMASK,
-			   (int_mask | SDMMC_INT_SDIO(slot->id)));
-	} else {
-		mci_writel(host, INTMASK,
-			   (int_mask & ~SDMMC_INT_SDIO(slot->id)));
-	}
+	if (enb)
+		int_mask |= SDMMC_INT_SDIO(slot->id);
+	else
+		int_mask &= ~SDMMC_INT_SDIO(slot->id);
+	mci_writel(host, INTMASK, int_mask);
 }
 
 static int dw_mci_execute_tuning(struct mmc_host *mmc, u32 opcode)
@@ -1140,6 +1141,7 @@ static const struct mmc_host_ops dw_mci_ops = {
 	.get_cd			= dw_mci_get_cd,
 	.enable_sdio_irq	= dw_mci_enable_sdio_irq,
 	.execute_tuning		= dw_mci_execute_tuning,
+	.init_card		= dw_mci_init_card,
 };
 
 static void dw_mci_request_end(struct dw_mci *host, struct mmc_request *mrq)
diff --git a/drivers/mmc/host/dw_mmc.h b/drivers/mmc/host/dw_mmc.h
index 6bf24ab..26d4657 100644
--- a/drivers/mmc/host/dw_mmc.h
+++ b/drivers/mmc/host/dw_mmc.h
@@ -227,6 +227,7 @@ struct dw_mci_slot {
 	unsigned long		flags;
 #define DW_MMC_CARD_PRESENT	0
 #define DW_MMC_CARD_NEED_INIT	1
+#define DW_MMC_CARD_NO_LOW_PWR	2
 	int			id;
 	int			last_detect_state;
 };
-- 
1.8.4


^ permalink raw reply	[flat|nested] 4+ messages in thread

* [PATCH v3 3/3] mmc: dw_mmc: Protect read-modify-write of INTMASK with a lock
  2013-10-18 21:55 ` [PATCH v3 0/3] Prevent races when doing read-modify-write of INTMASK Doug Anderson
  2013-10-18 21:56   ` [PATCH v3 1/3] mmc: core: Support the optional init_card() callback for MMC and SD Doug Anderson
  2013-10-18 21:56   ` [PATCH v3 2/3] mmc: dw_mmc: Cleanup disable of low power mode w/ SDIO interrupts Doug Anderson
@ 2013-10-18 21:56   ` Doug Anderson
  2 siblings, 0 replies; 4+ messages in thread
From: Doug Anderson @ 2013-10-18 21:56 UTC (permalink / raw)
  To: Jaehoon Chung, Seungwon Jeon, Chris Ball
  Cc: James Hogan, Grant Grundler, Alim Akhtar, Abhilash Kesavan,
	Tomasz Figa, Olof Johansson, Sonny Rao, Bing Zhao, Daniel Mack,
	Sascha Hauer, Doug Anderson, linux-mmc, linux-kernel

We're running into cases where our enabling of the SDIO interrupt in
dw_mmc doesn't actually take effect.  Specifically, adding patch like
this:

 +++ b/drivers/mmc/host/dw_mmc.c
 @@ -1076,6 +1076,9 @@ static void dw_mci_enable_sdio_irq(struct mmc_host *mmc, int enb)

      mci_writel(host, INTMASK,
           (int_mask | SDMMC_INT_SDIO(slot->id)));
 +    int_mask = mci_readl(host, INTMASK);
 +    if (!(int_mask & SDMMC_INT_SDIO(slot->id)))
 +      dev_err(&mmc->class_dev, "failed to enable sdio irq\n");
    } else {

...actually triggers the error message.  That's because the
dw_mci_enable_sdio_irq() unsafely does a read-modify-write of the
INTMASK register.

We can't just use the standard host->lock since that lock is not irq
safe and mmc_signal_sdio_irq() (called from interrupt context) calls
dw_mci_enable_sdio_irq().  Add a new irq-safe lock to protect INTMASK.

An alternate solution to this is to punt mmc_signal_sdio_irq() to the
tasklet and then protect INTMASK modifications by the standard host
lock.  This seemed like a bit more of a high-latency change.

Reported-by: Bing Zhao <bzhao@marvell.com>
Signed-off-by: Doug Anderson <dianders@chromium.org>
Reviewed-by: James Hogan <james.hogan@imgtec.com>
---
Changes in v3: None
Changes in v2:
- intmask_lock renamed to irq_lock

 drivers/mmc/host/dw_mmc.c  | 13 +++++++++++++
 include/linux/mmc/dw_mmc.h |  6 ++++++
 2 files changed, 19 insertions(+)

diff --git a/drivers/mmc/host/dw_mmc.c b/drivers/mmc/host/dw_mmc.c
index 6c8d6a9..3b910ec 100644
--- a/drivers/mmc/host/dw_mmc.c
+++ b/drivers/mmc/host/dw_mmc.c
@@ -657,6 +657,7 @@ disable:
 
 static int dw_mci_submit_data_dma(struct dw_mci *host, struct mmc_data *data)
 {
+	unsigned long irqflags;
 	int sg_len;
 	u32 temp;
 
@@ -693,9 +694,11 @@ static int dw_mci_submit_data_dma(struct dw_mci *host, struct mmc_data *data)
 	mci_writel(host, CTRL, temp);
 
 	/* Disable RX/TX IRQs, let DMA handle it */
+	spin_lock_irqsave(&host->irq_lock, irqflags);
 	temp = mci_readl(host, INTMASK);
 	temp  &= ~(SDMMC_INT_RXDR | SDMMC_INT_TXDR);
 	mci_writel(host, INTMASK, temp);
+	spin_unlock_irqrestore(&host->irq_lock, irqflags);
 
 	host->dma_ops->start(host, sg_len);
 
@@ -704,6 +707,7 @@ static int dw_mci_submit_data_dma(struct dw_mci *host, struct mmc_data *data)
 
 static void dw_mci_submit_data(struct dw_mci *host, struct mmc_data *data)
 {
+	unsigned long irqflags;
 	u32 temp;
 
 	data->error = -EINPROGRESS;
@@ -732,9 +736,12 @@ static void dw_mci_submit_data(struct dw_mci *host, struct mmc_data *data)
 		host->part_buf_count = 0;
 
 		mci_writel(host, RINTSTS, SDMMC_INT_TXDR | SDMMC_INT_RXDR);
+
+		spin_lock_irqsave(&host->irq_lock, irqflags);
 		temp = mci_readl(host, INTMASK);
 		temp |= SDMMC_INT_TXDR | SDMMC_INT_RXDR;
 		mci_writel(host, INTMASK, temp);
+		spin_unlock_irqrestore(&host->irq_lock, irqflags);
 
 		temp = mci_readl(host, CTRL);
 		temp &= ~SDMMC_CTRL_DMA_ENABLE;
@@ -1089,8 +1096,11 @@ static void dw_mci_enable_sdio_irq(struct mmc_host *mmc, int enb)
 {
 	struct dw_mci_slot *slot = mmc_priv(mmc);
 	struct dw_mci *host = slot->host;
+	unsigned long irqflags;
 	u32 int_mask;
 
+	spin_lock_irqsave(&host->irq_lock, irqflags);
+
 	/* Enable/disable Slot Specific SDIO interrupt */
 	int_mask = mci_readl(host, INTMASK);
 	if (enb)
@@ -1098,6 +1108,8 @@ static void dw_mci_enable_sdio_irq(struct mmc_host *mmc, int enb)
 	else
 		int_mask &= ~SDMMC_INT_SDIO(slot->id);
 	mci_writel(host, INTMASK, int_mask);
+
+	spin_unlock_irqrestore(&host->irq_lock, irqflags);
 }
 
 static int dw_mci_execute_tuning(struct mmc_host *mmc, u32 opcode)
@@ -2500,6 +2512,7 @@ int dw_mci_probe(struct dw_mci *host)
 	host->quirks = host->pdata->quirks;
 
 	spin_lock_init(&host->lock);
+	spin_lock_init(&host->irq_lock);
 	INIT_LIST_HEAD(&host->queue);
 
 	/*
diff --git a/include/linux/mmc/dw_mmc.h b/include/linux/mmc/dw_mmc.h
index 6ce7d2c..55e1d4e 100644
--- a/include/linux/mmc/dw_mmc.h
+++ b/include/linux/mmc/dw_mmc.h
@@ -102,6 +102,11 @@ struct mmc_data;
  * @cur_slot, @mrq and @state. These must always be updated
  * at the same time while holding @lock.
  *
+ * @irq_lock is an irq-safe spinlock protecting the INTMASK register
+ * to allow the interrupt handler to modify it directly.  Held for only long
+ * enough to read-modify-write INTMASK and no other locks are grabbed when
+ * holding this one.
+ *
  * The @mrq field of struct dw_mci_slot is also protected by @lock,
  * and must always be written at the same time as the slot is added to
  * @queue.
@@ -121,6 +126,7 @@ struct mmc_data;
  */
 struct dw_mci {
 	spinlock_t		lock;
+	spinlock_t		irq_lock;
 	void __iomem		*regs;
 
 	struct scatterlist	*sg;
-- 
1.8.4


^ permalink raw reply	[flat|nested] 4+ messages in thread

end of thread, other threads:[~2013-10-18 22:19 UTC | newest]

Thread overview: 4+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
     [not found] <1382126905-14654-1-git-send-email-dianders@chromium.org>
2013-10-18 21:55 ` [PATCH v3 0/3] Prevent races when doing read-modify-write of INTMASK Doug Anderson
2013-10-18 21:56   ` [PATCH v3 1/3] mmc: core: Support the optional init_card() callback for MMC and SD Doug Anderson
2013-10-18 21:56   ` [PATCH v3 2/3] mmc: dw_mmc: Cleanup disable of low power mode w/ SDIO interrupts Doug Anderson
2013-10-18 21:56   ` [PATCH v3 3/3] mmc: dw_mmc: Protect read-modify-write of INTMASK with a lock Doug Anderson

This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox

all inboxes | Powered by JetHome®