[Letux-kernel] Unhandled EFAULT with letux kernels built with newer gcc compilers

[H. Nikolaus Schaller]

Hi William,

> Am 04.04.2021 um 23:56 schrieb William Cohen <wcohen at nc.rr.com>:
> 
> On 4/4/21 11:55 AM, H. Nikolaus Schaller wrote:
>> Hi Will,
>> I haven't seen the unhandled EFAULT you are describing but
>> there is something weird with futex on the jz4730. There
>> is no unhandled fault but a process can get stuck in
>> the futex system call.
>> 
>> So this may both be related. It may fault on the jz4780 with
>> newer compiler but work with older and get stuck on jz4730
>> with older...
> 
> Hi,
> 
> I have been building the kernels self-hosted on the Creator CI20 board.  I don't know if that might be causing issues.

It shouldn't but one never knows...

> 
> One attempt I tried was to turn off the CONFIG_FUTEX in the configure to work around the problem with the  futex_detect_cmpxchg causing a panic.  Turning that CONFIG_FUTEX off  caused:
> 
> The futex facility returned an unexpected error code.[    3.174558] Kernel panic - not syncing: Attempted to kill init! exitcode=0x00000005   
> [    3.186706] Rebooting in 10 seconds.. 
> 
> So for the time being I am running a kernel that has  CONFIG_HAVE_FUTEX_CMPXCHG patched to be set.
> 
> One other thought as to something that might be causing a problem is that address have the MSB set as in the backtrace below:

Ok, that could also be an influence.

> 
>                                                                                                                     
> [    0.183009] Call Trace:                                                                                                                    
> [    0.185486] [<801b97b4>] cmpxchg_futex_value_locked+0x30/0x6c                                                                              
> [    0.191318] [<80d8940c>] futex_init+0x80/0xe0                                                                                              
> [    0.195730] [<80100c68>] do_one_initcall+0x150/0x308                                                                                       
> [    0.113385] [<80d7db68>] do_initcall_level+0x128/0x160                                                                                     
> [    0.118590] [<80d7d9f8>] do_initcalls+0x60/0xa8                                                                                            
> [    0.123181] [<80d7d98c>] do_basic_setup+0x28/0x34                                                                                          
> [    0.127949] [<80d7d834>] kernel_init_freeable+0x6c/0xa8                                                                                    
> [    0.133249] [<80a0f4ac>] kernel_init+0x10/0x128                                                                                            
> [    0.137838] [<801030a0>] ret_from_kernel_thread+0x10/0x18 
> 
> Is there any place that the addresses are being treated as 64-bit quantities and/or getting signed extended with the newer compiler?  There have been problems in the past like:
> 
> https://github.com/u-boot/u-boot/commit/c190fbd010a00e16e9599575b43c5a7c7bc7ec09#diff-980273600cca5b17c252b4ce1edb11b4cfd826fe6b2bbcad20393aa392f3cdce

There are also more 32<->64 bit address conversion macros in the kernel to better support 64 bit processors.
Maybe one of these is compiler dependent and/or processor dependent.

I'll keep an eye on this issue.

BR,
Nikolaus


> 
> 
>> I need some time to upgrade my good old cross-toolchains
>> to something newer (yes, gcc 4.9 is quite old and arm64
>> already requires gcc 5.1)... I have to modify my own compiler
>> generator because e.g. cross-tool-ng or prebuilt toolchains
>> do not fit my requirements for a cross-compiling farm for
>> multiple architectures. Especially since a toolchain is
>> more than the C compiler.
>> 
>> BR and thanks,
>> Nikolaus
>> 
>> 
>> 
> Thanks,
> 
> -Will
> 
> -- 
> You received this message because you are subscribed to the Google Groups "MIPS Creator CI20 Development" group.
> To unsubscribe from this group and stop receiving emails from it, send an email to mips-creator-ci20-dev+unsubscribe at googlegroups.com.
> To view this discussion on the web, visit https://groups.google.com/d/msgid/mips-creator-ci20-dev/cd14c9d9-e3a0-b017-1141-cd7266fd3fa8%40nc.rr.com.