Date: Fri, 23 Jan 2009 15:21:48 -0800 From: Sam Leffler <sam@freebsd.org> To: John Baldwin <jhb@freebsd.org> Cc: freebsd-hackers@freebsd.org, Mehul Chadha <mehulc87@gmail.com> Subject: Re: doubts regarding System Initialization working (SYSINIT) Message-ID: <497A510C.3030106@freebsd.org> In-Reply-To: <200901231755.15548.jhb@freebsd.org> References: <251d650c0901230755n769de861u1fe2e76dd28bbde8@mail.gmail.com> <200901231755.15548.jhb@freebsd.org>
next in thread | previous in thread | raw e-mail | index | archive | help
John Baldwin wrote:
> On Friday 23 January 2009 10:55:32 am Mehul Chadha wrote:
>
>> Hello all,
>> I have been browsing through the FreeBSD kernel's
>> source code trying to understand its working .
>>
>> In the mi_startup() in /sys/kern/init_main.c all the SYSINIT objects
>> are sorted using bubble sort and then they are executed in order.
>>
>> My doubt is that we have declared the pointer to the struct sysinit as
>> const pointer to a const in the macro definition of SYSINIT ie when
>> the macro
>>
>> SYSINIT(kmem, SI_SUB_KMEM, SI_ORDER_FIRST, kmeminit, NULL) is
>> expanded completely we get the following
>>
>> static struct sysinit kmem_sys_init = { SI_SUB_KMEM, SI_ORDER_FIRST,
>> (sysinit_cfunc_t)(sysinit_
>> nfunc_t)kmeminit, ((void *)(((void *)0))) }; static void const * const
>> __set_sysinit_set_sym_kmem_sys_init __attribute__((__section__("set_"
>> "sysinit_set"))) __attribute__((__used__)) = &kmem_sys_init;
>>
>> Here we see that the pointer is of type const and to a const but when we
>>
> sort
>
>> and swap using
>> *sipp=*xipp;
>>
>> We are trying to change the address of const pointer to a new address
>> in which case it should segfault but it works fine.
>>
>> Why does it not segfault it seems I have not understood the concept
>> behind using const *const... I will be very thankful if you can help
>> me with it.
>>
>
> I'm guessing the startup code doesn't map the SYSINIT pages read only because
> it is not smart enough to honor that request perhaps. That is, I wouldn't be
> surprised if all of .rodata in the kernel was mapped as R/W instead of R/O.
>
>
I think I have an ancient patch from someone to fix the code to not do
this; let me dig for it.
Sam
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?497A510C.3030106>