Hi, we modified the block checker and run it again on linux 2.4.1. (The
block checker flags an error when blocking functions are called with
either interrupts disabled or a spin lock held. )
It gave us 4 warnings in kernel/module.c. Because we are unaware of the
contexts where these functions are called, we are not sure if these 4
warnings are real errors or false positives. Please help us to verify them
or show that they are false positives.
As usual, please CC us at [email protected]. Any help will be
appreciated.
---------------------------------------------------------
[UNKNOWN] get_mod_name->__get_free_page(GFP_KERNEL). This is in the
KERNEL. Definitely need to verify
/u2/acc/oses/linux/2.4.1/kernel/module.c:290:sys_create_module:
ERROR:BLOCK:289:290:calling blocking fn 'get_mod_name' w/ spin lock held
[type=GLOBAL]:289
Start --->
lock_kernel();
Error --->
if ((namelen = get_mod_name(name_user, &name)) < 0) {
error = namelen;
---------------------------------------------------------
[UNKNOWN] get_mod_name->__get_free_page(GFP_KERNEL) This is in the KERNEL.
Definitely need to verify
/u2/acc/oses/linux/2.4.1/kernel/module.c:599:sys_delete_module:
ERROR:BLOCK:597:599:calling blocking fn 'get_mod_name' w/ spin lock held
[type=GLOBAL]:597
Start --->
lock_kernel();
if (name_user) {
Error --->
if ((error = get_mod_name(name_user, &name)) < 0)
goto out;
---------------------------------------------------------
[UNKNOWN] need to verify. in the KERNEL!
/u2/acc/oses/linux/2.4.1/kernel/module.c:376:sys_init_module:
ERROR:BLOCK:342:376:calling blocking fn 'copy_from_user' w/ spin lock held
[type=LOCAL]:342
Start --->
lock_kernel();
Error --->
if ((namelen = get_mod_name(name_user, &name)) < 0) {
error = namelen;
goto err0;
}
... DELETED 26 lines ...
goto err1;
}
strcpy(name_tmp, mod->name);
Error --->
error = copy_from_user(mod, mod_user, mod_user_size);
if (error) {
---------------------------------------------------------
[UNKNOWN] need to verify. in the KERNEL!
/u2/acc/oses/linux/2.4.1/kernel/module.c:888:sys_query_module:
ERROR:BLOCK:881:888:calling blocking fn 'get_mod_name' w/ spin lock held
[type=GLOBAL]:881
Start --->
lock_kernel();
if (name_user == NULL)
mod = &kernel_module;
else {
long namelen;
char *name;
Error --->
if ((namelen = get_mod_name(name_user, &name)) < 0) {
err = namelen;
---------------------------------------------------------
A few questions:
1. Is it OK to call blocking functions in the functions like
/init/main.c:init and init/main.c:start_kernel with a spin lock held? It
seems OK because the system is booting when these functions are called.
2. Can functions like kmem_cache_create, kmem_cache_alloc, alloc_page
block?
> Hi, we modified the block checker and run it again on linux 2.4.1. (The
> block checker flags an error when blocking functions are called with
> either interrupts disabled or a spin lock held. )
lock_kernel() isnt a spinlock as such.
> 2. Can functions like kmem_cache_create, kmem_cache_alloc, alloc_page
> block?
They may block unless GFP_ATOMIC is specified in the arguments.
Alan
On Fri, 23 Mar 2001 02:41:40 -0800 (PST),
Junfeng Yang <[email protected]> wrote:
>Hi, we modified the block checker and run it again on linux 2.4.1. (The
>block checker flags an error when blocking functions are called with
>either interrupts disabled or a spin lock held. )
>
>It gave us 4 warnings in kernel/module.c. Because we are unaware of the
>contexts where these functions are called, we are not sure if these 4
>warnings are real errors or false positives. Please help us to verify them
>or show that they are false positives.
All false positives. The big kernel lock is a special case, you are
allowed to sleep while holding that lock. See release_kernel_lock()
and reacquire_kernel_lock() in sched().
On Fri, 23 Mar 2001, Alan Cox wrote:
> > Hi, we modified the block checker and run it again on linux 2.4.1. (The
> > block checker flags an error when blocking functions are called with
> > either interrupts disabled or a spin lock held. )
>
> lock_kernel() isnt a spinlock as such.
Thanks a lot. We just figured out that it is ok to block within
lock_kernel() unlock_kernel() scope. That will help us to eliminate
some false positives.
On Fri, 23 Mar 2001, Keith Owens wrote:
> On Fri, 23 Mar 2001 02:41:40 -0800 (PST),
> Junfeng Yang <[email protected]> wrote:
> >Hi, we modified the block checker and run it again on linux 2.4.1. (The
> >block checker flags an error when blocking functions are called with
> >either interrupts disabled or a spin lock held. )
> >
> >It gave us 4 warnings in kernel/module.c. Because we are unaware of the
> >contexts where these functions are called, we are not sure if these 4
> >warnings are real errors or false positives. Please help us to verify them
> >or show that they are false positives.
>
> All false positives. The big kernel lock is a special case, you are
> allowed to sleep while holding that lock. See release_kernel_lock()
> and reacquire_kernel_lock() in sched().
Thanks for pointing this out. We'll modify the checker again and remove
"lock_kernel" from the patterns.