tty: serial: remove __init on pl011 console ops
PATCH 1/2 tty: serial: remove __init on pl011 console ops
If uart driver is probed defer, console_setup will be called later after __init && __initdata sections destroyed. And amba_console isn't defined in __init or __initdata section. So we needn't define pl011_console_setup() && pl011_console_get_options() in __init section.
--- drivers/tty/serial/amba-pl011.c | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-)
drivers/tty/serial/amba-pl011.c +2 -2
@@ -1810,7 +1810,7 @@ pl011_console_write(struct console *co, const char *s, unsigned int count)clk_disable (uap->clk); } -static void __init +static void pl011_console_get_options (struct uart_amba_port *uap,int *baud,int *parity,int *bits) {
@@ -1845,7 +1845,7 @@ pl011_console_get_options(struct uart_amba_port *uap, int *baud, } } -static int __init pl011_console_setup (struct console *co,char *options) +static int pl011_console_setup (struct console *co,char *options) {struct uart_amba_port *uap;int baud =38400 ;
Signature · 1 line
1.7.10.4
PATCH 2/2 tty: serial: use module_init on pl011_init
If amba serial driver is probed defer, amba serial driver may be probed after init process. So the error log shows in below.
[ 0.389403] Warning: unable to open an initial console. [ 0.390107] Freeing init memory: 2328K
It results in serial console not enabled. So replace arch_initcall by module_init on pl011_init(). The boot sequence is changed in below.
pinctrl driver --> amba serial driver --> init process
--- drivers/tty/serial/amba-pl011.c | 6 +----- 1 file changed, 1 insertion(+), 5 deletions(-)
drivers/tty/serial/amba-pl011.c +1 -5
@@ -2129,11 +2129,7 @@ static void __exit pl011_exit(void)uart_unregister_driver (&amba_reg); } -/* -* While this can be a module, if builtin it's most likely the console -* So let's leave module_exit but move module_init to an earlier place -*/ -arch_initcall (pl011_init); +module_init (pl011_init);module_exit (pl011_exit);MODULE_AUTHOR ("ARM Ltd/Deep Blue Solutions Ltd" );
Signature · 1 line
1.7.10.4
PATCH 2/2 tty: serial: use module_init on pl011_init
Please include Russell on patches to this driver.
Quoted 10 lines · Haojian Zhuang
If amba serial driver is probed defer, amba serial driver may be probed after init process. So the error log shows in below.
[ 0.389403] Warning: unable to open an initial console. [ 0.390107] Freeing init memory: 2328KIt results in serial console not enabled. So replace
arch_initcallbymodule_initonpl011_init(). The boot sequence is changed in below.pinctrl driver --> amba serial driver --> init process
ONLY on systems using pinctrl, nota bene.
Quoted 13 lines
Signed-off-by Haojian Zhuang <[email protected]>Cc Alan Cox <[email protected]>Cc Greg Kroah-Hartman <[email protected]>Cc linux-serial at vger.kernel.org--- drivers/tty/serial/amba-pl011.c | 6 +----- 1 file changed, 1 insertion(+), 5 deletions(-)drivers/tty/serial/amba-pl011.c +0
-0
drivers/tty/serial/amba-pl011.c static void __exit pl011_exit(void) · line 2132
Quoted 9 lines
drivers/tty/serial/amba-pl011.c +1
-5@@ -2129,11 +2129,7 @@ static void __exit pl011_exit(void)uart_unregister_driver (&amba_reg); } -/* -* While this can be a module, if builtin it's most likely the console -* So let's leave module_exit but move module_init to an earlier place -*/ -arch_initcall (pl011_init); +module_init (pl011_init);
Hm. PL011 is actually the only serial driver that gets called as arch_initcall().
Alessandro Rubini changed this in 2009.
Alessandro, can you explain? Is the sole purpose to get the console up early?
Yours,
Linus Walleij
PATCH 1/2 tty: serial: remove __init on pl011 console ops
In reply to Haojian ZhuangQuoted 4 lines · Haojian Zhuang
If uart driver is probed defer,
console_setupwill be called later after__init&&__initdatasections destroyed. Andamba_consoleisn't defined in__initor__initdatasection. So we needn't definepl011_console_setup()&&pl011_console_get_options()in__initsection.
It sounds like there's a deeper problem here - if this driver being deferred, why isn't it being retried after the pinctrl stuff gets its driver registered?
We've had bugs in this deferred probing before, I wouldn't be surprised if there's more... and we should not be "fixing" drivers because of bugs elsewhere.
PATCH 1/2 tty: serial: remove __init on pl011 console ops
Quoted 6 lines · Russell King - ARM Linux
Quoted 4 lines · Haojian Zhuang
If uart driver is probed defer,
console_setupwill be called later after__init&&__initdatasections destroyed. Andamba_consoleisn't defined in__initor__initdatasection. So we needn't definepl011_console_setup()&&pl011_console_get_options()in__initsection.
Quoted 7 lines
It sounds like there's a deeper problem here - if this driver being deferred, why isn't it being retried after the pinctrl stuff gets its driver registered?
We've had bugs in this deferred probing before, I wouldn't be surprised if there's more... and we should not be "fixing" drivers because of bugs elsewhere.
It retried if driver is being deferred. But those console setup functions are released
since they're declared in __init section.
rest_init()
--> creates kernel_init thread that could free __init section memory
deferred_probe_initicall() creates a workqueue that is used to retry deferred probe
function.
I think that these two threads are competing.
Best Regards
Haojian
PATCH 1/2 tty: serial: remove __init on pl011 console ops
On Wed, Feb 6, 2013 at 1:31 PM, Haojian Zhuang
<[email protected]> wrote:
CC Grant on this issue...
Quoted 20 lines
Quoted 6 lines · Russell King - ARM Linux
Quoted 4 lines · Haojian Zhuang
If uart driver is probed defer,
console_setupwill be called later after__init&&__initdatasections destroyed. Andamba_consoleisn't defined in__initor__initdatasection. So we needn't definepl011_console_setup()&&pl011_console_get_options()in__initsection.Quoted 7 lines
It sounds like there's a deeper problem here - if this driver being deferred, why isn't it being retried after the pinctrl stuff gets its driver registered?
We've had bugs in this deferred probing before, I wouldn't be surprised if there's more... and we should not be "fixing" drivers because of bugs elsewhere.
It retried if driver is being deferred. But those console setup functions are released
since they're declared in__initsection.
Quoted 8 lines
rest_init()
--> createskernel_initthread that could free__initsection memory
deferred_probe_initicall()creates a workqueue that is used to retry deferred probe
function.I think that these two threads are competing.
Now that is the *real* problem. The __init sections surely should not be discarded until *after* all deferred probes are complete and the deferral workqueue terminates.
How about writing a patch to fix that instead? (If possible...)
Yours,
Linus Walleij
PATCH 1/2 tty: serial: remove __init on pl011 console ops
Quoted 32 lines · Linus Walleij
On Wed, Feb 6, 2013 at 1:31 PM, Haojian Zhuang
<[email protected]> wrote:CC Grant on this issue...
Quoted 27 lines
Quoted 6 lines · Russell King - ARM Linux
Quoted 4 lines · Haojian Zhuang
If uart driver is probed defer,
console_setupwill be called later after__init&&__initdatasections destroyed. Andamba_consoleisn't defined in__initor__initdatasection. So we needn't definepl011_console_setup()&&pl011_console_get_options()in__initsection.Quoted 7 lines
It sounds like there's a deeper problem here - if this driver being deferred, why isn't it being retried after the pinctrl stuff gets its driver registered?
We've had bugs in this deferred probing before, I wouldn't be surprised if there's more... and we should not be "fixing" drivers because of bugs elsewhere.
It retried if driver is being deferred. But those console setup functions are released
since they're declared in__initsection.
rest_init()
--> createskernel_initthread that could free__initsection memory
deferred_probe_initicall()creates a workqueue that is used to retry deferred probe
function.
Quoted 7 lines
Quoted 1 line
I think that these two threads are competing.
Now that is the *real* problem. The
__initsections surely should not be discarded until *after* all deferred probes are complete and the deferral workqueue terminates.How about writing a patch to fix that instead? (If possible...)
Well, we have the facilities to flush workqueues, so it should be possible.
However, I'm just wondering if this shows that we _do_ need to get rid of a pile of __init marked functions as well as fixing this, thanks to the deferred probing we now have (maybe the whole __init thing becomes useless with deferred probing?) There's nothing really to guarantee that the pinctrl stuff will be available by the time the init sections are discarded (it could be in a loadable module?) or indeed any other resources that might be necessary.
I think in this regard, deferred probing has opened a similar can of worms which hotplug devices created (which eventually ended up with killing the __dev* marking).
PATCH 1/2 tty: serial: remove __init on pl011 console ops
Quoted 4 lines · Russell King - ARM Linux
However, I'm just wondering if this shows that we _do_ need to get rid of a pile of
__initmarked functions as well as fixing this, thanks to the deferred probing we now have (maybe the whole__initthing becomes useless with deferred probing?)
And all __initdata and __initconst as well.
Quoted 4 lines
There's nothing really to guarantee
that the pinctrl stuff will be available by the time the init sections are discarded (it could be in a loadable module?) or indeed any other resources that might be necessary.
I don't dare to guess how deferred probing is supposed to work with loadable modules. When I start to think of it my mind boggles. But I think maybe all drivers that support module loading are tagging things wrong. Shouldn't they all be using __init_or_module, __initdata_or_module, __initconst_or_module so they only get discarded unless compiled as modules?
As far as I can tell from <linux/init.h> that is the intention.
Talk about can of worms :-(
Quoted 3 lines
I think in this regard, deferred probing has opened a similar can of worms which hotplug devices created (which eventually ended up with killing the __dev* marking).
I'm afraid you're right.
Pinctrl core grabbing of pins seems to trigger the problem more often on platforms that are using it.
Yours,
Linus Walleij
PATCH 1/2 tty: serial: remove __init on pl011 console ops
In reply to Russell King - ARM LinuxQuoted 51 lines · Russell King - ARM Linux
Quoted 32 lines · Linus Walleij
On Wed, Feb 6, 2013 at 1:31 PM, Haojian Zhuang
<[email protected]> wrote:CC Grant on this issue...
Quoted 27 lines
Quoted 6 lines · Russell King - ARM Linux
Quoted 4 lines · Haojian Zhuang
If uart driver is probed defer,
console_setupwill be called later after__init&&__initdatasections destroyed. Andamba_consoleisn't defined in__initor__initdatasection. So we needn't definepl011_console_setup()&&pl011_console_get_options()in__initsection.Quoted 7 lines
It sounds like there's a deeper problem here - if this driver being deferred, why isn't it being retried after the pinctrl stuff gets its driver registered?
We've had bugs in this deferred probing before, I wouldn't be surprised if there's more... and we should not be "fixing" drivers because of bugs elsewhere.
It retried if driver is being deferred. But those console setup functions are released
since they're declared in__initsection.
rest_init()
--> createskernel_initthread that could free__initsection memory
deferred_probe_initicall()creates a workqueue that is used to retry deferred probe
function.Quoted 7 lines
Quoted 1 line
I think that these two threads are competing.
Now that is the *real* problem. The
__initsections surely should not be discarded until *after* all deferred probes are complete and the deferral workqueue terminates.How about writing a patch to fix that instead? (If possible...)
Well, we have the facilities to flush workqueues, so it should be possible.
However, I'm just wondering if this shows that we _do_ need to get rid of a pile of
__initmarked functions as well as fixing this, thanks to the deferred probing we now have (maybe the whole__initthing becomes useless with deferred probing?) There's nothing really to guarantee that the pinctrl stuff will be available by the time the init sections are discarded (it could be in a loadable module?) or indeed any other resources that might be necessary.
Quoted 3 lines
I think in this regard, deferred probing has opened a similar can of worms which hotplug devices created (which eventually ended up with killing the __dev* marking).
At first, I considered to remove those __init functions. But it may seem unreasonable.
Now I'm trying to reuse wait_for_device_probe() after do_initcalls(). Please help to review.
Thanks
Haojian
PATCH 1/2 tty: serial: remove __init on pl011 console ops
Quoted 16 lines · Haojian Zhuang
Quoted 14 lines · Russell King - ARM Linux
Quoted 13 lines · Linus Walleij
On Wed, Feb 6, 2013 at 1:31 PM, Haojian Zhuang
<[email protected]> wrote:CC Grant on this issue...
Quoted 8 lines
On 6 February 2013 19:52, Russell King - ARM Linux
rest_init()
--> createskernel_initthread that could free__initsection memory
deferred_probe_initicall()creates a workqueue that is used to retry deferred probe
function.
Quoted 7 lines
Quoted 7 lines
Quoted 7 lines
Quoted 1 line
I think that these two threads are competing.
Now that is the *real* problem. The
__initsections surely should not be discarded until *after* all deferred probes are complete and the deferral workqueue terminates.How about writing a patch to fix that instead? (If possible...)
That just masks the issue. See below.
Quoted 10 lines
Quoted 10 lines
Well, we have the facilities to flush workqueues, so it should be possible.
However, I'm just wondering if this shows that we _do_ need to get rid of a pile of
__initmarked functions as well as fixing this, thanks to the deferred probing we now have (maybe the whole__initthing becomes useless with deferred probing?) There's nothing really to guarantee that the pinctrl stuff will be available by the time the init sections are discarded (it could be in a loadable module?) or indeed any other resources that might be necessary.
If it is a probe function, then yes the __init annotations need to be removed or the driver needs to use platform_driver_probe(). This has been the model as long as I can remember. As long as a driver has a .probe hook, the driver core can call it at any time. Deferred probe has only exposed the issue.
If you want to discard the .probe() hook, then the probe() pointer needs to be cleared from the driver structure before discarding the sections. Otherwise the driver core can still call it which is what happens in deferred probe and can happen if a device gets unbound from a driver and then re-attached.
Quoted 9 lines
Quoted 3 lines
I think in this regard, deferred probing has opened a similar can of worms which hotplug devices created (which eventually ended up with killing the __dev* marking).
At first, I considered to remove those
__initfunctions. But it may seem unreasonable.Now I'm trying to reuse
wait_for_device_probe()afterdo_initcalls(). Please help to review.
That patch doesn't actually fix the problem, it just masks it.
g.
Signature · 2 lines
Grant Likely, B.Sc., P.Eng. Secret Lab Technologies Ltd.
PATCH 1/2 tty: serial: remove __init on pl011 console ops
Quoted 11 lines · Grant Likely
If it is a probe function, then yes the
__initannotations need to be removed or the driver needs to useplatform_driver_probe(). This has been the model as long as I can remember. As long as a driver has a .probe hook, the driver core can call it at any time. Deferred probe has only exposed the issue.If you want to discard the .probe() hook, then the
probe()pointer needs to be cleared from the driver structure before discarding the sections. Otherwise the driver core can still call it which is what happens in deferred probe and can happen if a device gets unbound from a driver and then re-attached.
You're talking about something completely different on the assumption that what is being talked about is the probe hook. It isn't. It's a path for the console initialization which has _always_ been __init marked since the dawn of time, because modules are not supposed to be calling that path.
What has been exposed is that console drivers which have resources which are not immediately available no longer have the same guarantees that they used to have (that is, they will be called before the __init section is given away.)
Remember: console drivers used not to support removal of themselves. Many still do not support that - in fact, these do not on the basis that they call register_console() but not unregister_console():
drivers/net/ethernet/sgi/ioc3-eth.cdrivers/s390/char/con3215.cdrivers/s390/char/con3270.cdrivers/s390/char/sclp_con.cdrivers/s390/char/sclp_vt220.cdrivers/tty/amiserial.cdrivers/tty/bfin_jtag_comm.cdrivers/tty/ehv_bytechan.cdrivers/tty/hvc/hvsi.cdrivers/tty/serial/21285.cdrivers/tty/serial/68328serial.cdrivers/tty/serial/8250/8250.cdrivers/tty/serial/8250/8250_early.cdrivers/tty/serial/altera_jtaguart.cdrivers/tty/serial/altera_uart.cdrivers/tty/serial/apbuart.cdrivers/tty/serial/arc_uart.cdrivers/tty/serial/atmel_serial.cdrivers/tty/serial/bcm63xx_uart.cdrivers/tty/serial/bfin_sport_uart.cdrivers/tty/serial/bfin_uart.cdrivers/tty/serial/cpm_uart/cpm_uart_core.cdrivers/tty/serial/dz.cdrivers/tty/serial/lantiq.cdrivers/tty/serial/lpc32xx_hs.cdrivers/tty/serial/m32r_sio.cdrivers/tty/serial/mcf.cdrivers/tty/serial/mpc52xx_uart.cdrivers/tty/serial/mpsc.cdrivers/tty/serial/netx-serial.cdrivers/tty/serial/nwpserial.cdrivers/tty/serial/pmac_zilog.cdrivers/tty/serial/pnx8xxx_uart.cdrivers/tty/serial/sa1100.cdrivers/tty/serial/samsung.cdrivers/tty/serial/sb1250-duart.cdrivers/tty/serial/serial_core.cdrivers/tty/serial/serial_ks8695.cdrivers/tty/serial/serial_txx9.cdrivers/tty/serial/sh-sci.cdrivers/tty/serial/sirfsoc_uart.cdrivers/tty/serial/uartlite.cdrivers/tty/serial/vr41xx_siu.cdrivers/tty/serial/xilinx_uartps.cdrivers/tty/serial/zs.cdrivers/tty/vt/vt.c
PATCH 1/2 tty: serial: remove __init on pl011 console ops
Quoted 66 lines · Russell King - ARM Linux
Quoted 11 lines · Grant Likely
If it is a probe function, then yes the
__initannotations need to be removed or the driver needs to useplatform_driver_probe(). This has been the model as long as I can remember. As long as a driver has a .probe hook, the driver core can call it at any time. Deferred probe has only exposed the issue.If you want to discard the .probe() hook, then the
probe()pointer needs to be cleared from the driver structure before discarding the sections. Otherwise the driver core can still call it which is what happens in deferred probe and can happen if a device gets unbound from a driver and then re-attached.You're talking about something completely different on the assumption that what is being talked about is the probe hook. It isn't. It's a path for the console initialization which has _always_ been
__initmarked since the dawn of time, because modules are not supposed to be calling that path.What has been exposed is that console drivers which have resources which are not immediately available no longer have the same guarantees that they used to have (that is, they will be called before the
__initsection is given away.)Remember: console drivers used not to support removal of themselves. Many still do not support that - in fact, these do not on the basis that they call
register_console()but notunregister_console():
drivers/net/ethernet/sgi/ioc3-eth.cdrivers/s390/char/con3215.cdrivers/s390/char/con3270.cdrivers/s390/char/sclp_con.cdrivers/s390/char/sclp_vt220.cdrivers/tty/amiserial.cdrivers/tty/bfin_jtag_comm.cdrivers/tty/ehv_bytechan.cdrivers/tty/hvc/hvsi.cdrivers/tty/serial/21285.cdrivers/tty/serial/68328serial.cdrivers/tty/serial/8250/8250.cdrivers/tty/serial/8250/8250_early.cdrivers/tty/serial/altera_jtaguart.cdrivers/tty/serial/altera_uart.cdrivers/tty/serial/apbuart.cdrivers/tty/serial/arc_uart.cdrivers/tty/serial/atmel_serial.cdrivers/tty/serial/bcm63xx_uart.cdrivers/tty/serial/bfin_sport_uart.cdrivers/tty/serial/bfin_uart.cdrivers/tty/serial/cpm_uart/cpm_uart_core.cdrivers/tty/serial/dz.cdrivers/tty/serial/lantiq.cdrivers/tty/serial/lpc32xx_hs.cdrivers/tty/serial/m32r_sio.cdrivers/tty/serial/mcf.cdrivers/tty/serial/mpc52xx_uart.cdrivers/tty/serial/mpsc.cdrivers/tty/serial/netx-serial.cdrivers/tty/serial/nwpserial.cdrivers/tty/serial/pmac_zilog.cdrivers/tty/serial/pnx8xxx_uart.cdrivers/tty/serial/sa1100.cdrivers/tty/serial/samsung.cdrivers/tty/serial/sb1250-duart.cdrivers/tty/serial/serial_core.cdrivers/tty/serial/serial_ks8695.c
Quoted 8 lines
drivers/tty/serial/serial_txx9.cdrivers/tty/serial/sh-sci.cdrivers/tty/serial/sirfsoc_uart.cdrivers/tty/serial/uartlite.cdrivers/tty/serial/vr41xx_siu.cdrivers/tty/serial/xilinx_uartps.cdrivers/tty/serial/zs.cdrivers/tty/vt/vt.c
I agree on Russell. And console isn't the only one device. In kernel_init_freeable(), we can find that we have an assumption that all initial calls are already finished, and we need to discard them & start the user-mode stuff. But at that time, deferred probe may still not run since they're in another kernel thread. It's very dangerous that user mode stuff start running but device driver isn't ready yet.
Then let's move to my error case. Serial driver fails to bind pinctrl, and it's moved into deferred probe list. And kernel_init_freeable() in kernel_init thread keeps working on CPU, the deferred probe doesn't have chances to be scheduled yet. So two issues happen.
1. /dev/console is accessed before serial driver is ready. Since it fails, user can't input anything on console any more.
2. __init memory section is released before serial driver is ready. It results in system hang while deferred probe runs.
Regards
Haojian
PATCH 1/2 tty: serial: remove __init on pl011 console ops
Quoted 8 lines · Haojian Zhuang
I agree on Russell. And console isn't the only one device. In
kernel_init_freeable(), we can find that we have an assumption that all initial calls are already finished, and we need to discard them & start the user-mode stuff. But at that time, deferred probe may still not run since they're in another kernel thread. It's very dangerous that user mode stuff start running but device driver isn't ready yet.
Linux has been moving in that direction for quite a while now. Userspace can start before a lot of devices are ready. Mass storage for instance. An initramfs can start userspace, but if the real rootfs is on rotating storage, then it will have to wait for it to become ready before mounting the real root. Distributions ran into this problem quite a while ago.
That said, console is pretty important, and it isn't good that the fallout of deferred probe is that console gets initialized even later.
Linus, we should get some engineers around a whiteboard at connect and map out the initialization order issues. I've been pushing for moving as much as possible to device_initcall() time, which in general I think is the right thing to do, but this is the downside.
Quoted 8 lines
Then let's move to my error case. Serial driver fails to bind pinctrl, and it's moved into deferred probe list. And
kernel_init_freeable()inkernel_initthread keeps working on CPU, the deferred probe doesn't have chances to be scheduled yet. So two issues happen.
1./dev/consoleis accessed before serial driver is ready. Since it fails, user can't input anything on console any more.
This is a fair point and is a better argument for me to having a pass over the deferred drivers before exiting the device initcalls.
Quoted 3 lines
2.
__initmemory section is released before serial driver is ready. It results in system hang while deferred probe runs.
The __init section in a .probe() path is a problem in and of itself. I do think that the __init annotations either need to be removed or the hooks into them need to be cleared when the init sections are discarded.
g.
PATCH 1/2 tty: serial: remove __init on pl011 console ops
In reply to Russell King - ARM LinuxQuoted 13 lines · Russell King - ARM Linux
Quoted 11 lines · Grant Likely
If it is a probe function, then yes the
__initannotations need to be removed or the driver needs to useplatform_driver_probe(). This has been the model as long as I can remember. As long as a driver has a .probe hook, the driver core can call it at any time. Deferred probe has only exposed the issue.If you want to discard the .probe() hook, then the
probe()pointer needs to be cleared from the driver structure before discarding the sections. Otherwise the driver core can still call it which is what happens in deferred probe and can happen if a device gets unbound from a driver and then re-attached.
Quoted 5 lines
You're talking about something completely different on the assumption that what is being talked about is the probe hook. It isn't. It's a path for the console initialization which has _always_ been
__initmarked since the dawn of time, because modules are not supposed to be calling that path.
Sorry, you're right. Hitting reply too quickly I guess.
However, the point still stands that anything that can be called from a .probe() path cannot be in an __init section with the current driver model. A driver can even be unbound from a device and reattached at runtime. That would also cause problems.
Quoted 4 lines
What has been exposed is that console drivers which have resources which are not immediately available no longer have the same guarantees that they used to have (that is, they will be called before the
__initsection is given away.)
Okay, I'll reply to Haojian's proposed solution patch on this point.
g.
PATCH 1/2 tty: serial: remove __init on pl011 console ops
Quoted 20 lines · Grant Likely
Quoted 13 lines · Russell King - ARM Linux
Quoted 11 lines · Grant Likely
If it is a probe function, then yes the
__initannotations need to be removed or the driver needs to useplatform_driver_probe(). This has been the model as long as I can remember. As long as a driver has a .probe hook, the driver core can call it at any time. Deferred probe has only exposed the issue.If you want to discard the .probe() hook, then the
probe()pointer needs to be cleared from the driver structure before discarding the sections. Otherwise the driver core can still call it which is what happens in deferred probe and can happen if a device gets unbound from a driver and then re-attached.Quoted 5 lines
You're talking about something completely different on the assumption that what is being talked about is the probe hook. It isn't. It's a path for the console initialization which has _always_ been
__initmarked since the dawn of time, because modules are not supposed to be calling that path.
Quoted 6 lines
Sorry, you're right. Hitting reply too quickly I guess.
However, the point still stands that anything that can be called from a .probe() path cannot be in an
__initsection with the current driver model. A driver can even be unbound from a device and reattached at runtime. That would also cause problems.
However, it reveals much bigger problems - that is, if you defer the probe of the console, what happens when we need to open the console...
Although you can say that, there's _much_ bigger problems here if you delay console initialization - you'll effectively end up calling init with no system console in place, which means you'll see no output from init until the deferred probe sorts itself out.
Yes, it's an unintended side effect of deferred probing, but nevertheless one which needs to be resolved such that the console _is_ initialized before it's required, which happens _before_ the init section is discarded.
PATCH 1/2 tty: serial: remove __init on pl011 console ops
Quoted 36 lines · Russell King - ARM Linux
Quoted 20 lines · Grant Likely
Quoted 13 lines · Russell King - ARM Linux
Quoted 11 lines · Grant Likely
If it is a probe function, then yes the
__initannotations need to be removed or the driver needs to useplatform_driver_probe(). This has been the model as long as I can remember. As long as a driver has a .probe hook, the driver core can call it at any time. Deferred probe has only exposed the issue.If you want to discard the .probe() hook, then the
probe()pointer needs to be cleared from the driver structure before discarding the sections. Otherwise the driver core can still call it which is what happens in deferred probe and can happen if a device gets unbound from a driver and then re-attached.Quoted 5 lines
You're talking about something completely different on the assumption that what is being talked about is the probe hook. It isn't. It's a path for the console initialization which has _always_ been
__initmarked since the dawn of time, because modules are not supposed to be calling that path.Quoted 6 lines
Sorry, you're right. Hitting reply too quickly I guess.
However, the point still stands that anything that can be called from a .probe() path cannot be in an
__initsection with the current driver model. A driver can even be unbound from a device and reattached at runtime. That would also cause problems.However, it reveals much bigger problems - that is, if you defer the probe of the console, what happens when we need to open the console...
Although you can say that, there's _much_ bigger problems here if you delay console initialization - you'll effectively end up calling init with no system console in place, which means you'll see no output from init until the deferred probe sorts itself out.
Quoted 4 lines
Yes, it's an unintended side effect of deferred probing, but nevertheless one which needs to be resolved such that the console _is_ initialized before it's required, which happens _before_ the init section is discarded.
I've replied with a suggested solution to Haojian's patch to driver core. If he can test it and post a proper patch, then I'll ack it.
g.
Signature · 2 lines
Grant Likely, B.Sc., P.Eng. Secret Lab Technologies Ltd.