Use own work queue
PATCH 1/1 Use own work queue
The MMC layer uses the standard work queue for doing card detection. As this queue is shared with other crucial subsystems, the effects of a long (and perhaps buggy) detection can cause the system to be unusable. E.g. the keyboard stops working while the detection routine is running.
The solution is to add a specific mmc work queue to run the detection code in. This is similar to how other subsystems handle detection (a full kernel thread is the most common theme).
---
drivers/mmc/mmc.c | 6 +++--- drivers/mmc/mmc.h | 4 ++++ drivers/mmc/mmc_sysfs.c | 35 ++++++++++++++++++++++++++++++++++- 3 files changed, 41 insertions(+), 4 deletions(-)
drivers/mmc/mmc.c +3 -3
@@ -1166,9 +1166,9 @@ static void mmc_setup(struct mmc_host *hvoid mmc_detect_change (struct mmc_host *host,unsigned long delay) {if (delay) -schedule_delayed_work (&host->detect, delay); +mmc_schedule_delayed_work (&host->detect, delay);else -schedule_work (&host->detect); +mmc_schedule_work (&host->detect); }EXPORT_SYMBOL (mmc_detect_change);
@@ -1311,7 +1311,7 @@ EXPORT_SYMBOL(mmc_remove_host);*/ void mmc_free_host (struct mmc_host *host) { -flush_scheduled_work (); +mmc_flush_scheduled_work ();mmc_free_host_sysfs (host); }
drivers/mmc/mmc.h +4 -0
@@ -18,4 +18,8 @@ struct mmc_host *mmc_alloc_host_sysfs(inint mmc_add_host_sysfs (struct mmc_host *host);void mmc_remove_host_sysfs (struct mmc_host *host);void mmc_free_host_sysfs (struct mmc_host *host); + +int mmc_schedule_work (struct work_struct *work); +int mmc_schedule_delayed_work (struct work_struct *work,unsigned long delay); +void mmc_flush_scheduled_work (void );#endif
drivers/mmc/mmc_sysfs.c +34 -1
@@ -13,6 +13,7 @@ #include <linux/module.h>#include <linux/init.h> #include <linux/device.h> #include <linux/idr.h> +#include <linux/workqueue.h> #include <linux/mmc/card.h> #include <linux/mmc/host.h>
@@ -317,10 +318,41 @@ void mmc_free_host_sysfs(struct mmc_hostclass_device_put (&host->class_dev); } +static struct workqueue_struct *workqueue; + +/* +* Internal function. Schedule work in the MMC work queue. +*/ +int mmc_schedule_work (struct work_struct *work) +{ +return queue_work (workqueue, work); +} + +/* +* Internal function. Schedule delayed work in the MMC work queue. +*/ +int mmc_schedule_delayed_work (struct work_struct *work,unsigned long delay) +{ +return queue_delayed_work (workqueue, work, delay); +} + +/* +* Internal function. Flush all scheduled work from the MMC work queue. +*/ +void mmc_flush_scheduled_work (void ) +{ +flush_workqueue (workqueue); +}static int __init mmc_init (void ) { -int ret =bus_register (&mmc_bus_type); +int ret; + + workqueue =create_singlethread_workqueue ("kmmcd" ); +if (!workqueue) +return -ENOMEM ; + + ret =bus_register (&mmc_bus_type);if (ret ==0 ) { ret =class_register (&mmc_host_class);if (ret)
@@ -333,6 +365,7 @@ static void __exit mmc_exit(void) {class_unregister (&mmc_host_class);bus_unregister (&mmc_bus_type); +destroy_workqueue (workqueue); }module_init (mmc_init);
Hi,
Quoted 10 lines · Pierre Ossman
The MMC layer uses the standard work queue for doing card detection. As this queue is shared with other crucial subsystems, the effects of a long (and perhaps buggy) detection can cause the system to be unusable. E.g. the keyboard stops working while the detection routine is running.
The solution is to add a specific mmc work queue to run the detection code in. This is similar to how other subsystems handle detection (a full kernel thread is the most common theme).
Signed-off-by Pierre Ossman <[email protected]>
This patch makes pxamci stop working for me on a HTC Magician (PXA272). Switching from 2.6.18 to 2.6.19-rc1 I got a kernel panic:
mmc0: clock 0Hz busmode 1 powermode 0 cs 0 Vdd 0 width 0
PXAMCI: clkrt = 0 cmdat = 0
VFS: Cannot open root device "mmcblk0p2" or unknown-block(0,0) Please append a correct "root=" boot option
Kernel panic - not syncing: VFS: Unable to mount root fs on unknown-block(0,0)
After removing this patch from 2.6.19-rc1, everything is working again. Are there any changes to pxamci.c needed to be compatible with it?
regards
Philipp
Quoted 8 lines · pHilipp Zabel
This patch makes pxamci stop working for me on a HTC Magician (PXA272). Switching from 2.6.18 to 2.6.19-rc1 I got a kernel panic:
mmc0: clock 0Hz busmode 1 powermode 0 cs 0 Vdd 0 width 0
PXAMCI: clkrt = 0 cmdat = 0
VFS: Cannot open root device "mmcblk0p2" or unknown-block(0,0) Please append a correct "root=" boot option
Quoted 6 lines
Kernel panic - not syncing: VFS: Unable to mount root fs onunknown-block(0,0)
After removing this patch from 2.6.19-rc1, everything is working again. Are there any changes to
pxamci.cneeded to be compatible with it?
No, the drivers shouldn't be affected. As this is a root device, my guess would be that you have a race in your bootup that is causing problem.
Are you using initrd for this device? And could you get a complete dmesg dump?
Rgds
Pierre
Quoted 12 lines · Pierre Ossman
Quoted 11 lines · pHilipp Zabel
This patch makes pxamci stop working for me on a HTC Magician (PXA272). Switching from 2.6.18 to 2.6.19-rc1 I got a kernel panic:
mmc0: clock 0Hz busmode 1 powermode 0 cs 0 Vdd 0 width 0
PXAMCI: clkrt = 0 cmdat = 0
VFS: Cannot open root device "mmcblk0p2" or unknown-block(0,0) Please append a correct "root=" boot optionKernel panic - not syncing: VFS: Unable to mount root fs onunknown-block(0,0)
Quoted 6 lines
Quoted 3 lines
After removing this patch from 2.6.19-rc1, everything is working again. Are there any changes to
pxamci.cneeded to be compatible with it?No, the drivers shouldn't be affected. As this is a root device, my guess would be that you have a race in your bootup that is causing problem.
The problem is likely that the boot is continuing in parallel with detecting the card, because the card detection is running in its own separate thread. Meanwhile, the init thread is trying to read from the as-yet missing root device and erroring out.
Signature · 3 lines
Russell King Linux kernel 2.6 ARM Linux - http://www.arm.linux.org.uk/ maintainer of: 2.6 Serial core
Quoted 17 lines · Russell King
Quoted 16 lines · Pierre Ossman
Quoted 14 lines · pHilipp Zabel
This patch makes pxamci stop working for me on a HTC Magician (PXA272). Switching from 2.6.18 to 2.6.19-rc1 I got a kernel panic:
mmc0: clock 0Hz busmode 1 powermode 0 cs 0 Vdd 0 width 0
PXAMCI: clkrt = 0 cmdat = 0
VFS: Cannot open root device "mmcblk0p2" or unknown-block(0,0) Please append a correct "root=" boot optionKernel panic - not syncing: VFS: Unable to mount root fs onunknown-block(0,0)
After removing this patch from 2.6.19-rc1, everything is working again. Are there any changes to
pxamci.cneeded to be compatible with it?
Quoted 7 lines
Quoted 2 lines
No, the drivers shouldn't be affected. As this is a root device, my guess would be that you have a race in your bootup that is causing problem.
The problem is likely that the boot is continuing in parallel with detecting the card, because the card detection is running in its own separate thread. Meanwhile, the init thread is trying to read from the as-yet missing root device and erroring out.
Thanks, I can work around this by using the rootdelay kernel parameter. So does that mean this is the expected behavior, or should I do anything in the bootup sequence to make the init process wait for mmc detection?
regards
Philipp
Quoted 9 lines · pHilipp Zabel
Quoted 4 lines · Russell King
The problem is likely that the boot is continuing in parallel with detecting the card, because the card detection is running in its own separate thread. Meanwhile, the init thread is trying to read from the as-yet missing root device and erroring out.
Thanks, I can work around this by using the rootdelay kernel parameter. So does that mean this is the expected behavior, or should I do anything in the bootup sequence to make the init process wait for mmc detection?
That's up to Pierre now. I originally ensured that the initial boot time card detection was synchronous with the boot to avoid unexpected problems like this.
Signature · 3 lines
Russell King Linux kernel 2.6 ARM Linux - http://www.arm.linux.org.uk/ maintainer of: 2.6 Serial core
Quoted 5 lines · pHilipp Zabel
Thanks, I can work around this by using the rootdelay kernel parameter. So does that mean this is the expected behavior, or should I do anything in the bootup sequence to make the init process wait for mmc detection?
USB, which has an almost identical problem, usually uses a "sleep" in initrd.
If you want to be a bit more fancy, you could try to listen to hotplug events and look for the kernel creating the relevant block device.
Rgds
Signature · 5 lines
-- Pierre Ossman Linux kernel, MMC maintainer http://www.kernel.org PulseAudio, core developer http://pulseaudio.org rdesktop, core developer http://www.rdesktop.org
Quoted 6 lines · Russell King
The problem is likely that the boot is continuing in parallel with detecting the card, because the card detection is running in its own separate thread. Meanwhile, the init thread is trying to read from the as-yet missing root device and erroring out.
That's what I suspect as well. I know using a root device on USB has these kinds of problems. And the solution I've mostly seen is adding a delay somewhere in initrd.
My experience with embedded systems is limited unfortunately, Perhaps Russell has some nice solution for Philipp? :)
Rgds
Signature · 6 lines
-- Pierre Ossman Linux kernel, MMC maintainer http://www.kernel.org PulseAudio, core developer http://pulseaudio.org rdesktop, core developer http://www.rdesktop.org OLPC, developer http://www.laptop.org