Groups | Search | Server Info | Keyboard shortcuts | Login | Register [http] [https] [nntp] [nntps]

Groups > linux.kernel > #1722961 > unrolled thread

[PATCH v3 0/2] Avoid system abort by move pm domain's detach after devres_release_all

Back to article view | Back to linux.kernel

Contents

  [PATCH v3 0/2] Avoid system abort by move pm domain's detach after devres_release_all Shawn Lin <shawn.lin@rock-chips.com> - 2017-08-30 03:40 +0200
    [PATCH v3 2/2] mmc: dw_mmc: fix potential system abort if activating CONFIG_DEBUG_SHIRQ Shawn Lin <shawn.lin@rock-chips.com> - 2017-08-30 03:40 +0200

#1722961 — [PATCH v3 0/2] Avoid system abort by move pm domain's detach after devres_release_all

CONFIG_DEBUG_SHIRQ will fire extra irq action to call the registered
irq callback after driver is removed or failed to probe. In general,
the irq callback provided by driver should read its internal registers
to see who fires the irq. So this leads a situation that we access the
registers with a powered-off pm domain that the system abort. This is
a system-wide issue may break lots of drivers, IMHO.

dwmmc driver is one of them suffered from this, please see the commit
message of patch 2 for how that happened.

I haven't find a proper way to freeze the genpd_power_off_work_fn
and fire it again when finish devres_release_all indeed. And it
seems to me that device's pm domain detach is always called when
failing to probe or removing the driver. So I have to cook patch 1
to see if folks think this's the best way to fix that, otherwise we may
need to fix it everywhere for other drivers.


Changes in v3:
- fix the code path for consolidating the attach for both of driver
  and bus driver, and then move detach to the error path
- rework the changelog
- include a driver core change to fix the genpd issue.

Shawn Lin (2):
  driver core: detach device's pm_domain after devres_release_all
  mmc: dw_mmc: fix potential system abort if activating
    CONFIG_DEBUG_SHIRQ

 drivers/base/dd.c         | 16 ++++++++++++++++
 drivers/base/platform.c   | 18 ++----------------
 drivers/mmc/host/dw_mmc.c | 48 +++++++++++++++++++++++------------------------
 3 files changed, 41 insertions(+), 41 deletions(-)

-- 
1.9.1

[toc] | [next] | [standalone]

#1722962 — [PATCH v3 2/2] mmc: dw_mmc: fix potential system abort if activating CONFIG_DEBUG_SHIRQ

With CONFIG_DEBUG_SHIRQ enabled, the irq tear down routine
would still access the irq handler registed as a shard irq.
Per the comment within the function of __free_irq, it says
"It's a shared IRQ -- the driver ought to be prepared for
an IRQ event to happen even now it's being freed". However
when failing to probe the driver, dw_mmc disables the clock
and asserts the reset pin, even power off its genpd for accessing
the registers and the following check for shared irq state would call
the irq handler which accesses the register w/o all necessary resource
prepared. That will hang the system forever.

With adding some dump_stack we could see how that happened.

Synopsys Designware Multimedia Card Interface Driver
dwmmc_rockchip fe320000.dwmmc: IDMAC supports 32-bit address mode.
dwmmc_rockchip fe320000.dwmmc: Using internal DMA controller.
dwmmc_rockchip fe320000.dwmmc: Version ID is 270a
CPU: 0 PID: 1 Comm: swapper/0 Not tainted
4.13.0-rc3-next-20170807-00004-g93d3644-dirty #5
Hardware name: Firefly-RK3399 Board (DT)
Call trace:
[<ffff20000808b5a0>] dump_backtrace+0x0/0x300
[<ffff20000808ba1c>] show_stack+0x14/0x20
[<ffff200008dc480c>] dump_stack+0xa4/0xc8
[<ffff200008b9691c>] dw_mci_interrupt+0x3c/0x6a8
[<ffff200008157450>] __free_irq+0x308/0x410
[<ffff20000815760c>] free_irq+0x54/0xb0
[<ffff20000815d630>] devm_irq_release+0x30/0x40
[<ffff2000087f0174>] release_nodes+0x1e4/0x390
[<ffff2000087f04c0>] devres_release_all+0x50/0x78
[<ffff2000087e9bc0>] driver_probe_device+0x128/0x3b8
[<ffff2000087e9f34>] __driver_attach+0xe4/0xe8
[<ffff2000087e7048>] bus_for_each_dev+0xe0/0x138
[<ffff2000087e93b8>] driver_attach+0x30/0x40
[<ffff2000087e8c00>] bus_add_driver+0x1d0/0x328
[<ffff2000087ead0c>] driver_register+0xb4/0x198
[<ffff2000087ec98c>] __platform_driver_register+0x7c/0x88
[<ffff2000095bc564>] dw_mci_rockchip_pltfm_driver_init+0x18/0x20
[<ffff200008083a8c>] do_one_initcall+0x14c/0x1b8
[<ffff200009560ff8>] kernel_init_freeable+0x238/0x2d8
[<ffff200008dde500>] kernel_init+0x10/0x110
[<ffff2000080836c0>] ret_from_fork+0x10/0x50
Synchronous External Abort: synchronous external abort (0x96000010) at
0xffff20000aaa4040
Internal error: : 96000010 [#1] PREEMPT SMP
Modules linked in:
CPU: 0 PID: 1 Comm: swapper/0 Not tainted
task: ffff80006ba28080 task.stack: ffff80006ba24000
PC is at dw_mci_interrupt+0x4c/0x6a8
LR is at dw_mci_interrupt+0x44/0x6a8
pc : [<ffff200008b9692c>] lr : [<ffff200008b96924>] pstate: 600001c5

...

Kernel panic - not syncing: Attempted to kill init! exitcode=0x0000000b

SMP: stopping secondary CPUs
Kernel Offset: disabled
CPU features: 0x00200c
Memory Limit: none
---[ end Kernel panic - not syncing: Attempted to kill init!
exitcode=0x0000000b

In order to fix this, we remove all the clock-disabling from
the error handle path and driver's remove function. And replying
on the devm_add_action_or_reset to fire the clock-disabling and reset
signal at the appropriate time.

Signed-off-by: Shawn Lin <shawn.lin@rock-chips.com>

---

Changes in v3:
- include a driver core change to fix the genpd issue.

 drivers/mmc/host/dw_mmc.c | 48 +++++++++++++++++++++++------------------------
 1 file changed, 23 insertions(+), 25 deletions(-)

diff --git a/drivers/mmc/host/dw_mmc.c b/drivers/mmc/host/dw_mmc.c
index 860313b..610613a 100644
--- a/drivers/mmc/host/dw_mmc.c
+++ b/drivers/mmc/host/dw_mmc.c
@@ -3057,6 +3057,18 @@ static void dw_mci_enable_cd(struct dw_mci *host)
 	}
 }
 
+static void dw_mci_post_cleanup(void *data)
+{
+	struct dw_mci *host = data;
+
+	clk_disable_unprepare(host->ciu_clk);
+	clk_disable_unprepare(host->biu_clk);
+
+	if (!IS_ERR(host->pdata->rstc))
+		reset_control_assert(host->pdata->rstc);
+
+}
+
 int dw_mci_probe(struct dw_mci *host)
 {
 	const struct dw_mci_drv_data *drv_data = host->drv_data;
@@ -3092,7 +3104,7 @@ int dw_mci_probe(struct dw_mci *host)
 		ret = clk_prepare_enable(host->ciu_clk);
 		if (ret) {
 			dev_err(host->dev, "failed to enable ciu clock\n");
-			goto err_clk_biu;
+			return ret;
 		}
 
 		if (host->pdata->bus_hz) {
@@ -3105,11 +3117,16 @@ int dw_mci_probe(struct dw_mci *host)
 		host->bus_hz = clk_get_rate(host->ciu_clk);
 	}
 
+	ret = devm_add_action_or_reset(host->dev, dw_mci_post_cleanup, host);
+	if (ret) {
+		dev_err(host->dev, "unable to add action or reset\n");
+		return ret;
+	}
+
 	if (!host->bus_hz) {
 		dev_err(host->dev,
 			"Platform data must supply bus speed\n");
-		ret = -ENODEV;
-		goto err_clk_ciu;
+		return -ENODEV;
 	}
 
 	if (!IS_ERR(host->pdata->rstc)) {
@@ -3123,7 +3140,7 @@ int dw_mci_probe(struct dw_mci *host)
 		if (ret) {
 			dev_err(host->dev,
 				"implementation specific init failed\n");
-			goto err_clk_ciu;
+			return ret;
 		}
 	}
 
@@ -3167,10 +3184,8 @@ int dw_mci_probe(struct dw_mci *host)
 	}
 
 	/* Reset all blocks */
-	if (!dw_mci_ctrl_reset(host, SDMMC_CTRL_ALL_RESET_FLAGS)) {
-		ret = -ENODEV;
-		goto err_clk_ciu;
-	}
+	if (!dw_mci_ctrl_reset(host, SDMMC_CTRL_ALL_RESET_FLAGS))
+		return -ENODEV;
 
 	host->dma_ops = host->pdata->dma_ops;
 	dw_mci_init_dma(host);
@@ -3257,15 +3272,6 @@ int dw_mci_probe(struct dw_mci *host)
 	if (host->use_dma && host->dma_ops->exit)
 		host->dma_ops->exit(host);
 
-	if (!IS_ERR(host->pdata->rstc))
-		reset_control_assert(host->pdata->rstc);
-
-err_clk_ciu:
-	clk_disable_unprepare(host->ciu_clk);
-
-err_clk_biu:
-	clk_disable_unprepare(host->biu_clk);
-
 	return ret;
 }
 EXPORT_SYMBOL(dw_mci_probe);
@@ -3285,17 +3291,9 @@ void dw_mci_remove(struct dw_mci *host)
 
 	if (host->use_dma && host->dma_ops->exit)
 		host->dma_ops->exit(host);
-
-	if (!IS_ERR(host->pdata->rstc))
-		reset_control_assert(host->pdata->rstc);
-
-	clk_disable_unprepare(host->ciu_clk);
-	clk_disable_unprepare(host->biu_clk);
 }
 EXPORT_SYMBOL(dw_mci_remove);
 
-
-
 #ifdef CONFIG_PM
 int dw_mci_runtime_suspend(struct device *dev)
 {
-- 
1.9.1

[toc] | [prev] | [standalone]

Back to top | Article view | linux.kernel

csiph-web