On Tue, May 24, 2022 at 11:48 AM Kefeng Wang <wangkefeng.wang@xxxxxxxxxx> wrote:Right, I found this too, and using ___force in local, will update, thank.
I think this need an extra __force to actually suppress the sparse1) Will add generic "arch_ioremap/arch_iounmap define"mm/ioremap.c:59:16: sparse: expected void const *addrmm/ioremap.c:59:16: sparse: sparse: incorrect type in argument 1 (different address spaces) @@ expected void const *addr @@ got void [noderef] __iomem *vaddr @@
mm/ioremap.c:59:16: sparse: got void [noderef] __iomem *vaddr
vim +59 mm/ioremap.c
51
52 void iounmap(volatile void __iomem *addr)
53 {
54 void __iomem *vaddr = (void __iomem *)((unsigned long)addr & PAGE_MASK);
55
56 if (arch_iounmap(vaddr))
57 return;
58
> 59 vunmap(vaddr);
2) and change this to vunmap((void *)vaddr);
warning, as in
vunmap((void __force *)vaddr);
Using __force is usually wrong, this is one of the exceptions, so
maybe add a comment
as well.
Arnd
.