Kernel 5.4 Patch Request

[rifsxd]

Is it possible for bore to work on 5.4? Since 4.19 and 5.15 backports doesnt boot but builds fine after a bit of manual patching

[firelzrd]

Can I see your "backported" patch, please?

[rifsxd]

Can I see your "backported" patch, please?

Bummer I deleted that backport branch, which patch you want me to apply to show you the changes? 5.15 backport would a better approach imo

[firelzrd]

It seems that 4.19 is more similar to 5.4. For example, the usage of min() macro and also the variable types in sysctl should be more like 4.19's.

[rifsxd]

linux5.4.y-bore5.1.0.patch.txt

CC kernel/sched/fair.o ../kernel/sched/fair.c:152:26: error: implicit declaration of function 'task_of' [-Werror,-Wimplicit-function-declaration] struct task_struct p = task_of(se); ^ ../kernel/sched/fair.c:152:22: error: incompatible integer to pointer conversion initializing 'struct task_struct ' with an expression of type 'int' [-Werror,-Wint-conversion] struct task_struct p = task_of(se); ^ ~~~ ../kernel/sched/fair.c:152:22: error: ISO C90 forbids mixing declarations and code [-Werror,-Wdeclaration-after-statement] ../kernel/sched/fair.c:347:35: error: static declaration of 'task_of' follows non-static declaration static inline struct task_struct task_of(struct sched_entity se) ^ ../kernel/sched/fair.c:152:26: note: previous implicit declaration is here struct task_struct p = task_of(se); ^ 4 errors generated. make[3]: [../scripts/Makefile.build:288: kernel/sched/fair.o] Error 1 make[2]: [../scripts/Makefile.build:556: kernel/sched] Error 2

[rifsxd]

https://github.com/rifsxd/android_kernel_realme_RMX3511/commit/5bcf5f8aae26f7fcd4fa95efaa621bbb02463d8c

[firelzrd]

For implicit declaration of function 'task_of', It seems that in fair.c static inline struct task_struct *task_of() is used before declared. Please try declaring it before the use. Insert the below line before the BORE function block. (anywhere before #150 in patched fair.c)

static inline struct task_struct *task_of(struct sched_entity *se);

For ISO C90 forbids mixing declarations and code, you have 2 options:

• Move all local variable declarations to the top of function. Here's why
• Turn off -Werror or -Wdeclaration-after-statement

[rifsxd]

static void update_burst_score(struct sched_entity *se) {
    struct task_struct *p;
    u8 prio;
    u8 prev_prio;
    u8 new_prio;

    // Check if sched_entity corresponds to a task
    if (!entity_is_task(se))
        return;

    // Resolve task_struct from sched_entity
    p = container_of(se, struct task_struct, se);
    prio = p->static_prio - MAX_RT_PRIO;

    // Calculate previous priority
    prev_prio = min(39, prio + se->burst_score);

    // Update the burst score
    se->burst_score = se->burst_penalty >> 2;

    // Calculate new priority
    new_prio = min(39, prio + se->burst_score);
    if (new_prio != prev_prio) {
        reweight_task(p, new_prio); // Adjust task weight if priority changes
    }
}

I did this and it built without errors but as always whatever I do bore makes my kernel panic and doesn't boot

[firelzrd]

I see. Is there any kernel panic log?

[rifsxd]

If I could get out logs from this, then I wouldn't even need to hit you up lol the device gets stuck at boot logo (not boot animation) so no way to get logs, I know this is a kernel panic that's all

[firelzrd]

Okay, got it. I'll build my own 5.4 kernel when I next find free time.

[firelzrd]

https://github.com/firelzrd/bore-scheduler/blob/main/patches/legacy/linux-5.4-bore/0001-linux5.4.y-bore5.1.0.patch

confirmed bootable

[rifsxd]

https://github.com/rifsxd/android_kernel_realme_RMX3511/commits/bore/

still not booting and kernel panics but built succesfully after a little bit of code changes for my kernel source

0001-linux5.4.y-bore5.1.0.patch.txt 0002-constgran-vanilla-max.patch.txt

these are the new patches

and also now i can give you the kernel panic logs, could you inspect?

console-ramoops-0.log dmesg-ramoops-0.log

[firelzrd]

Thank you for providing those files.

<14>[    0.831215]c6 [    T1] init: [libfs_mgr]ReadFstabFromDt(): failed to read fstab from dt
<14>[    0.832569]c6 [    T1] init: Using Android DT directory /proc/device-tree/firmware/android/
<14>[    0.885179]c6 [    T1] init: bool android::init::BlockDevInitializer::InitDevices(std::set<std::string>): partition(s) not found in /sys, waiting for their uevent(s): boot_b, dtbo_b, l_agdsp_b, l_gdsp_b, l_ldsp_b, l_modem_b, metadata, pm_sys_b, super, vbmeta_b, vbmeta_product_b, vbmeta_system_b, vbmeta_system_ext_b, vbmeta_vendor_b, vendor_boot_b
<14>[   10.895536]c6 [    T1] init: Wait for partitions returned after 10010ms
<11>[   10.895585]c6 [    T1] init: bool android::init::BlockDevInitializer::InitDevices(std::set<std::string>): partition(s) not found after polling timeout: boot_b, dtbo_b, l_agdsp_b, l_gdsp_b, l_ldsp_b, l_modem_b, metadata, pm_sys_b, super, vbmeta_b, vbmeta_product_b, vbmeta_system_b, vbmeta_system_ext_b, vbmeta_vendor_b, vendor_boot_b
<10>[   10.895709]c6 [    T1] init: Failed to mount required partitions early ...
<0>[   10.895821]c6 [    T1] Kernel panic - not syncing: Attempted to kill the init task!

It looks to be a partition detection error to me. Can you try without BORE and see if it still fails?

[rifsxd]

Thank you for providing those files.

<14>[    0.831215]c6 [    T1] init: [libfs_mgr]ReadFstabFromDt(): failed to read fstab from dt
<14>[    0.832569]c6 [    T1] init: Using Android DT directory /proc/device-tree/firmware/android/
<14>[    0.885179]c6 [    T1] init: bool android::init::BlockDevInitializer::InitDevices(std::set<std::string>): partition(s) not found in /sys, waiting for their uevent(s): boot_b, dtbo_b, l_agdsp_b, l_gdsp_b, l_ldsp_b, l_modem_b, metadata, pm_sys_b, super, vbmeta_b, vbmeta_product_b, vbmeta_system_b, vbmeta_system_ext_b, vbmeta_vendor_b, vendor_boot_b
<14>[   10.895536]c6 [    T1] init: Wait for partitions returned after 10010ms
<11>[   10.895585]c6 [    T1] init: bool android::init::BlockDevInitializer::InitDevices(std::set<std::string>): partition(s) not found after polling timeout: boot_b, dtbo_b, l_agdsp_b, l_gdsp_b, l_ldsp_b, l_modem_b, metadata, pm_sys_b, super, vbmeta_b, vbmeta_product_b, vbmeta_system_b, vbmeta_system_ext_b, vbmeta_vendor_b, vendor_boot_b
<10>[   10.895709]c6 [    T1] init: Failed to mount required partitions early ...
<0>[   10.895821]c6 [    T1] Kernel panic - not syncing: Attempted to kill the init task!

It looks to be a partition detection error to me. Can you try without BORE and see if it still fails?

Exactly when I went through the logs I was so confused like why it doesn't get the partitions correctly loaded? What does it have to do with a scheduler? It doesn't occur when I do CONFIG_SCHED_BORE=n

[rifsxd]

Could it be cuz of sched_init_bore?

[firelzrd]

void sched_post_fork(struct task_struct *p)
{
#ifdef CONFIG_SCHED_BORE
    sched_post_fork_bore(p);
#endif // CONFIG_SCHED_BORE
    uclamp_post_fork(p);
}

I couldn't find the definition of uclamp_post_fork in Linux 5.4 (at least in 5.4.0 and 5.4.286), but in Linux 5.10 there is.

    p->last_sleep_ts        = 0;
#ifdef CONFIG_SCHED_BORE
    sched_fork_bore(p);
#endif // CONFIG_SCHED_BORE
    INIT_LIST_HEAD(&p->se.group_node);

And I couldn't find anything like last_sleep_ts in core.c of Linux 5.4 or 5.10.

Are you using a modified Linux source code?

[rifsxd]

This is my device kernel source

https://github.com/rifsxd/android_kernel_realme_RMX3511/tree/bore

Which closely follows this android_common_kernel branch android12-5.4 which I upstream from

https://github.com/aosp-mirror/kernel_common/tree/0c2544add9fc25c0e54a2167d6a2cfd2e696cf58

I upstreamed UpTo 5.4.256 and those functions exists

I tried with my stock oem kernel source patch 254 , my own 255, 256 and all fails to boot with bore on

EDIT: Wait p->last_sleep_ts doesn't have exist on android_common_kernel!

[firelzrd]

Thank you. I'll investigate it.

[rifsxd]

Thank you. I'll investigate it.

Edited above

[rifsxd]

Thank you. I'll investigate it.

And thanks ^^

[firelzrd]

0001-linux5.4.256-android-bore5.1.0.patch.txt I don't see any fault in the modifications. A little (but not much) difference between your patch and the one I made. Please recheck Kconfigs and filesystem-related config files.

[firelzrd]

[ 0.830188]c6 [ T1] sprd_wdt_fiq: disagrees about version of symbol module_layout

This may be indicating that there's a version mismatch between the kernel and the sprd_wdt_fiq module.

[rifsxd]

[ 0.830188]c6 [ T1] sprd_wdt_fiq: disagrees about version of symbol module_layout

This may be indicating that there's a version mismatch between the kernel and the sprd_wdt_fiq module.

You are not getting it, the kernel boots if I turn off CONFIG_SCHED_BORE in defconfig but doesn't if I turn it on and bore is included in the kernel

[firelzrd]

Alright. I've got to go for 2-day Tokyo trip. Please let me think over.

[rifsxd]

Alright. I've got to go for 2-day Tokyo trip. Please let me think over.

Sure thing!

[backslashxx]

@rifsxd can you try to patch out modversions check for now? atleast for testing do something like

https://github.com/rifsxd/android_kernel_realme_RMX3511/blob/bore/kernel/module.c#L1349 return 1 on this

like https://github.com/backslashxx/mojito_krenol/commit/0ad68a503f03ad788c92133dd84c5252fa882868

[rifsxd]

@rifsxd can you try to patch out modversions check for now? atleast for testing do something like

https://github.com/rifsxd/android_kernel_realme_RMX3511/blob/bore/kernel/module.c#L1349 return 1 on this

like backslashxx/mojito_krenol@0ad68a5

Sure, I will try it tomorrow.. I'm going out for something.