This reverts commit 452562abb5b7 ("base: arch_topology: fix section
mismatch build warnings"). It causes the notifier call hangs in some
use-cases.
In some cases with using maxcpus, some of cpus are booted first and
then the remaining cpus are booted. As an example, some users who want
to realize fast boot up often use the following procedure.
1) Define all CPUs on device tree (CA57x4 + CA53x4)
2) Add "maxcpus=4" in bootargs
3) Kernel boot up with CA57x4
4) After kernel boot up, CA53x4 is booted from user
When kernel init was finished, CPUFREQ_POLICY_NOTIFIER was not still
unregisterd. This means that "__init init_cpu_capacity_callback()"
will be called after kernel init sequence. To avoid this problem,
it needs to remove __init{,data} annotations by reverting this commit.
Also, this commit was needed to fix kernel compile issue below.
However, this issue was also fixed by another patch: commit 82d8ba717ccb
("arch_topology: Fix section miss match warning due to
free_raw_capacity()") in v4.15 as well.
Whereas commit 452562abb5b7 added all the missing __init annotations,
commit 82d8ba717ccb removed it from free_raw_capacity().
WARNING: vmlinux.o(.text+0x548f24): Section mismatch in reference
from the function init_cpu_capacity_callback() to the variable
.init.text:$x
The function init_cpu_capacity_callback() references
the variable __init $x.
This is often because init_cpu_capacity_callback lacks a __init
annotation or the annotation of $x is wrong.
Signed-off-by: Gaku Inami <[email protected]>
---
drivers/base/arch_topology.c | 12 ++++++------
1 file changed, 6 insertions(+), 6 deletions(-)
diff --git a/drivers/base/arch_topology.c b/drivers/base/arch_topology.c
index 4de87b0..8bfd498 100644
--- a/drivers/base/arch_topology.c
+++ b/drivers/base/arch_topology.c
@@ -175,11 +175,11 @@ bool __init topology_parse_cpu_capacity(struct device_node *cpu_node, int cpu)
}
#ifdef CONFIG_CPU_FREQ
-static cpumask_var_t cpus_to_visit __initdata;
-static void __init parsing_done_workfn(struct work_struct *work);
-static __initdata DECLARE_WORK(parsing_done_work, parsing_done_workfn);
+static cpumask_var_t cpus_to_visit;
+static void parsing_done_workfn(struct work_struct *work);
+static DECLARE_WORK(parsing_done_work, parsing_done_workfn);
-static int __init
+static int
init_cpu_capacity_callback(struct notifier_block *nb,
unsigned long val,
void *data)
@@ -215,7 +215,7 @@ init_cpu_capacity_callback(struct notifier_block *nb,
return 0;
}
-static struct notifier_block init_cpu_capacity_notifier __initdata = {
+static struct notifier_block init_cpu_capacity_notifier = {
.notifier_call = init_cpu_capacity_callback,
};
@@ -248,7 +248,7 @@ static int __init register_cpufreq_notifier(void)
}
core_initcall(register_cpufreq_notifier);
-static void __init parsing_done_workfn(struct work_struct *work)
+static void parsing_done_workfn(struct work_struct *work)
{
cpufreq_unregister_notifier(&init_cpu_capacity_notifier,
CPUFREQ_POLICY_NOTIFIER);
--
2.7.4
Hi Inami-san,
On 02/07/2018 09:32 AM, Gaku Inami wrote:
> This reverts commit 452562abb5b7 ("base: arch_topology: fix section
> mismatch build warnings"). It causes the notifier call hangs in some
> use-cases.
>
> In some cases with using maxcpus, some of cpus are booted first and
> then the remaining cpus are booted. As an example, some users who want
> to realize fast boot up often use the following procedure.
>
> 1) Define all CPUs on device tree (CA57x4 + CA53x4)
> 2) Add "maxcpus=4" in bootargs
> 3) Kernel boot up with CA57x4
> 4) After kernel boot up, CA53x4 is booted from user
>
> When kernel init was finished, CPUFREQ_POLICY_NOTIFIER was not still
> unregisterd. This means that "__init init_cpu_capacity_callback()"
> will be called after kernel init sequence. To avoid this problem,
> it needs to remove __init{,data} annotations by reverting this commit.
>
> Also, this commit was needed to fix kernel compile issue below.
> However, this issue was also fixed by another patch: commit 82d8ba717ccb
> ("arch_topology: Fix section miss match warning due to
> free_raw_capacity()") in v4.15 as well.
> Whereas commit 452562abb5b7 added all the missing __init annotations,
> commit 82d8ba717ccb removed it from free_raw_capacity().
>
> WARNING: vmlinux.o(.text+0x548f24): Section mismatch in reference
> from the function init_cpu_capacity_callback() to the variable
> .init.text:$x
> The function init_cpu_capacity_callback() references
> the variable __init $x.
> This is often because init_cpu_capacity_callback lacks a __init
> annotation or the annotation of $x is wrong.
>
> Signed-off-by: Gaku Inami <[email protected]>
Reviewed-by: Dietmar Eggemann <[email protected]>
Tested-by: Dietmar Eggemann <[email protected]>
This also helps with cpufreq driver module loading. W/o this revert I
get the following when trying to modprobe scpi-cpufreq on juno:
root@juno:/lib/modules# modprobe scpi-cpufreq
[ 171.994851] Unable to handle kernel paging request at virtual address
ffff000008f57708
[ 172.002708] Mem abort info:
[ 172.005471] ESR = 0x96000007
[ 172.008507] Exception class = DABT (current EL), IL = 32 bits
[ 172.014391] SET = 0, FnV = 0
[ 172.017410] EA = 0, S1PTW = 0
[ 172.020528] Data abort info:
[ 172.023384] ISV = 0, ISS = 0x00000007
[ 172.027190] CM = 0, WnR = 0
[ 172.030139] swapper pgtable: 4k pages, 48-bit VAs, pgd = 0000000015b245a1
[ 172.036867] [ffff000008f57708] *pgd=00000009ffffe003,
*pud=00000009ffffd003, *pmd=00000009ffffb003, *pte=0000000000000000
[ 172.047737] Internal error: Oops: 96000007 [#1] PREEMPT SMP
[ 172.053250] Modules linked in: scpi_cpufreq(+)
[ 172.057656] CPU: 1 PID: 2439 Comm: modprobe Not tainted
4.15.0-10759-ge237f98a9c13 #1
[ 172.065405] Hardware name: ARM Juno development board (r0) (DT)
[ 172.071265] pstate: a0000005 (NzCv daif -PAN -UAO)
[ 172.076012] pc : notifier_call_chain+0x34/0x90
[ 172.080411] lr : __blocking_notifier_call_chain+0x50/0x78
[ 172.085752] sp : ffff00000dc4b750
[ 172.089029] x29: ffff00000dc4b750 x28: ffff80097526b808
[ 172.094291] x27: 0000000000000040 x26: ffff0000090bbc30
[ 172.099551] x25: ffff00000913d000 x24: ffff00000913df98
[ 172.104811] x23: 0000000000000000 x22: ffff00000dc4b840
[ 172.110071] x21: ffff00000dc4b840 x20: 0000000000000000
[ 172.115330] x19: 00000000ffffffff x18: 0000000000000000
[ 172.120589] x17: 0000aaaaad4b02b0 x16: ffff00000814e9b8
[ 172.125848] x15: ffffffffffffffff x14: 0000000000000000
[ 172.131107] x13: ffff80097526b968 x12: ffff80097526b968
[ 172.136366] x11: 0000000000000000 x10: 0000000000000000
[ 172.141625] x9 : 0000000000000000 x8 : ffff8009753c1e00
[ 172.146884] x7 : 00000000000cf850 x6 : ffff000008f57708
[ 172.152143] x5 : 00000000ffffffff x4 : 0000000000000000
[ 172.157402] x3 : 00000000ffffffff x2 : ffff00000dc4b840
[ 172.162661] x1 : ffff000008f57708 x0 : ffff0000090bbcd8
[ 172.167922] Process modprobe (pid: 2439, stack limit =
0x000000005740a6c4)
[ 172.174724] Call trace:
[ 172.177143] notifier_call_chain+0x34/0x90
[ 172.181196] __blocking_notifier_call_chain+0x50/0x78
[ 172.186195] blocking_notifier_call_chain+0x14/0x20
[ 172.191025] cpufreq_set_policy+0x8c/0x1e0
[ 172.195077] cpufreq_init_policy+0x58/0x90
[ 172.199130] cpufreq_online+0x3d4/0x658
[ 172.202924] cpufreq_add_dev+0x78/0x88
[ 172.206635] subsys_interface_register+0x84/0xc8
[ 172.211203] cpufreq_register_driver+0x150/0x1d0
[ 172.215778] scpi_cpufreq_probe+0x28/0x80 [scpi_cpufreq]
[ 172.221036] platform_drv_probe+0x58/0xc0
[ 172.225003] driver_probe_device+0x224/0x308
[ 172.229227] __driver_attach+0xac/0xb0
[ 172.232935] bus_for_each_dev+0x54/0x98
[ 172.236729] driver_attach+0x20/0x28
[ 172.240265] bus_add_driver+0x108/0x228
[ 172.244060] driver_register+0x60/0xf8
[ 172.247769] __platform_driver_register+0x40/0x48
[ 172.252427] scpi_cpufreq_platdrv_init+0x18/0x1000 [scpi_cpufreq]
[ 172.258459] do_one_initcall+0x38/0x120
[ 172.262255] do_init_module+0x58/0x1b8
[ 172.265964] load_module+0x1ed8/0x21e8
[ 172.269672] SyS_finit_module+0xc0/0xd0
[ 172.273466] el0_svc_naked+0x20/0x24
[ 172.277004] Code: 54000320 2a0303f3 aa0203f6 aa0403f4 (f9400023)
[ 172.283035] ---[ end trace f70b068835918c13 ]---
Segmentation fault
[...]
On 07/02/18 08:32, Gaku Inami wrote:
> This reverts commit 452562abb5b7 ("base: arch_topology: fix section
> mismatch build warnings"). It causes the notifier call hangs in some
> use-cases.
>
> In some cases with using maxcpus, some of cpus are booted first and
> then the remaining cpus are booted. As an example, some users who want
> to realize fast boot up often use the following procedure.
>
> 1) Define all CPUs on device tree (CA57x4 + CA53x4)
> 2) Add "maxcpus=4" in bootargs
> 3) Kernel boot up with CA57x4
> 4) After kernel boot up, CA53x4 is booted from user
>
> When kernel init was finished, CPUFREQ_POLICY_NOTIFIER was not still
> unregisterd. This means that "__init init_cpu_capacity_callback()"
> will be called after kernel init sequence. To avoid this problem,
> it needs to remove __init{,data} annotations by reverting this commit.
>
> Also, this commit was needed to fix kernel compile issue below.
> However, this issue was also fixed by another patch: commit 82d8ba717ccb
> ("arch_topology: Fix section miss match warning due to
> free_raw_capacity()") in v4.15 as well.
> Whereas commit 452562abb5b7 added all the missing __init annotations,
> commit 82d8ba717ccb removed it from free_raw_capacity().
>
Ah I see commit 82d8ba717ccb made the commit 452562abb5b7 redundant
by removing the __init. Both attempt to solve the same problem in
different ways :)
Can you please add the the tag, so that it gets picked up accordingly ?
Fixes: 82d8ba717ccb ("arch_topology: Fix section miss match warning due
to free_raw_capacity()")
FWIW,
Acked-by: Sudeep Holla <[email protected]>
> WARNING: vmlinux.o(.text+0x548f24): Section mismatch in reference
> from the function init_cpu_capacity_callback() to the variable
> .init.text:$x
> The function init_cpu_capacity_callback() references
> the variable __init $x.
> This is often because init_cpu_capacity_callback lacks a __init
> annotation or the annotation of $x is wrong.
>
> Signed-off-by: Gaku Inami <[email protected]>
> ---
> drivers/base/arch_topology.c | 12 ++++++------
> 1 file changed, 6 insertions(+), 6 deletions(-)
>
> diff --git a/drivers/base/arch_topology.c b/drivers/base/arch_topology.c
> index 4de87b0..8bfd498 100644
> --- a/drivers/base/arch_topology.c
> +++ b/drivers/base/arch_topology.c
> @@ -175,11 +175,11 @@ bool __init topology_parse_cpu_capacity(struct device_node *cpu_node, int cpu)
> }
>
> #ifdef CONFIG_CPU_FREQ
> -static cpumask_var_t cpus_to_visit __initdata;
> -static void __init parsing_done_workfn(struct work_struct *work);
> -static __initdata DECLARE_WORK(parsing_done_work, parsing_done_workfn);
> +static cpumask_var_t cpus_to_visit;
> +static void parsing_done_workfn(struct work_struct *work);
> +static DECLARE_WORK(parsing_done_work, parsing_done_workfn);
>
> -static int __init
> +static int
> init_cpu_capacity_callback(struct notifier_block *nb,
> unsigned long val,
> void *data)
> @@ -215,7 +215,7 @@ init_cpu_capacity_callback(struct notifier_block *nb,
> return 0;
> }
>
> -static struct notifier_block init_cpu_capacity_notifier __initdata = {
> +static struct notifier_block init_cpu_capacity_notifier = {
> .notifier_call = init_cpu_capacity_callback,
> };
>
> @@ -248,7 +248,7 @@ static int __init register_cpufreq_notifier(void)
> }
> core_initcall(register_cpufreq_notifier);
>
> -static void __init parsing_done_workfn(struct work_struct *work)
> +static void parsing_done_workfn(struct work_struct *work)
> {
> cpufreq_unregister_notifier(&init_cpu_capacity_notifier,
> CPUFREQ_POLICY_NOTIFIER);
>
--
Regards,
Sudeep
> -----Original Message-----
> From: Sudeep Holla [mailto:[email protected]]
> Sent: Monday, February 12, 2018 9:29 PM
> To: Gaku Inami <[email protected]>; [email protected]; [email protected]
> Cc: Sudeep Holla <[email protected]>; [email protected]; [email protected];
> [email protected]
> Subject: Re: [PATCH] Revert "base: arch_topology: fix section mismatch build warnings"
<snip>
>
> Ah I see commit 82d8ba717ccb made the commit 452562abb5b7 redundant
> by removing the __init. Both attempt to solve the same problem in
> different ways :)
>
> Can you please add the the tag, so that it gets picked up accordingly ?
>
> Fixes: 82d8ba717ccb ("arch_topology: Fix section miss match warning due
> to free_raw_capacity()")
>
>
> FWIW,
>
> Acked-by: Sudeep Holla <[email protected]>
Thanks for your review and ack. I will send v2 patch soon.
Regards,
Inami