On Sun, Jun 9, 2019 at 10:40 PM Joe Perches <joe@xxxxxxxxxxx> wrote:
>
> On Sun, 2019-06-09 at 22:08 +0900, Masahiro Yamada wrote:
> > On Sun, Jun 9, 2019 at 8:57 PM Joe Perches <joe@xxxxxxxxxxx> wrote:
> > > On Sun, 2019-06-09 at 16:14 +0900, Masahiro Yamada wrote:
> > > > Hi Joe,
> > > >
> > > > On Thu, Jun 6, 2019 at 2:06 AM Joe Perches <joe@xxxxxxxxxxx> wrote:
> > > > > Perhaps a checkpatch change too:
> > > > >
> > > > > The first block updates unsigned only bitfields
> > > > > The second tests uapi definitions and suggests "__<kernel_types"
> > > >
> > > > Good.
> > > >
> > > > In addition,
> > > >
> > > > "warn if __u8, __u16, __u32, __u64 are used outside of uapi/"
> > > >
> > > > Lots of kernel-space headers use __u{8,16,32,64} instead of u{8,16,32,64}
> > > > just because developers often miss to understand when to use
> > > > the underscore-prefixed types.
> > >
> > > The problem there is that checkpatch can't know if the
> > > __<uapi_type> being used is for an actual uapi use or not.
> > >
> > > coccinelle could be much better at that.
> >
> > Why?
>
>
> Perhaps it's (somewhat) bad form to have a __uapi type in a
> structure, include that structure in a driver for something
> like a copy_to/from_user, and map the __<uapi_type> to a non
> underscore prefixed <kernel_type>
Linus Torvalds wrote 'sparse' to check this.
Any attempt to distinguish the address-space
by the presence of double-underscore-prefixes is pointless.
This is already checked by __kernel / __user.
It is absolutely correct to assign __u32 to u32, and vice versa.
If you think the following patch is wrong, please tell me why:
diff --git a/drivers/android/binder.c b/drivers/android/binder.c
index 748ac489ef7e..24c1b73d9fbd 100644
--- a/drivers/android/binder.c
+++ b/drivers/android/binder.c
@@ -1132,7 +1132,7 @@ static struct binder_node *binder_init_node_ilocked(
struct binder_node *node;
binder_uintptr_t ptr = fp ? fp->binder : 0;
binder_uintptr_t cookie = fp ? fp->cookie : 0;
- __u32 flags = fp ? fp->flags : 0;
+ u32 flags = fp ? fp->flags : 0;
assert_spin_locked(&proc->inner_lock);
@@ -4918,7 +4918,7 @@ static int
binder_ioctl_get_node_info_for_ref(struct binder_proc *proc,
{
struct binder_node *node;
struct binder_context *context = proc->context;
- __u32 handle = info->handle;
+ u32 handle = info->handle;
if (info->strong_count || info->weak_count || info->reserved1 ||
info->reserved2 || info->reserved3) {
>
> For instance
>
> struct flat_binder_object in drivers/android/binder.c
>
> How is checkpatch supposed to know that __u32 flags is
> inappropriate?
>
>
--
Best Regards
Masahiro Yamada
