Re: [PATCH v2 1/1] x86/early_printk: Use __iomem address space for IO

[Andy Shevchenko <andriy.shevchenko@linux.intel.com>]

On Sun, 2015-10-11 at 21:18 +0200, Thomas Gleixner wrote:
> Andy,
> 
> On Fri, 9 Oct 2015, Andy Shevchenko wrote:
> >  #ifdef CONFIG_PCI
> >  static void mem32_serial_out(unsigned long addr, int offset, int 
> value)
> >  {
> > -     u32 *vaddr = (u32 *)addr;
> > +     void __iomem *vaddr = (void __iomem *)addr;
> >       /* shift implied by pointer type */
> >       writel(value, vaddr + offset);
> 
> This is broken. Assume vaddr = 0x1000 and offset = 1

Yes, I noticed in reply to myself  
http://www.spinics.net/lists/kernel/msg2094143.html

Tests were okay because platform I have is using 8-bit I/O.

Does it make sense to use the following approach?

 void __iomem *vaddr = (void __iomem *)addr;
 writel(value, vaddr + offset << 2);
 

> 
> ==>  u32 *vaddr = 0x1000;
> ==>  vaddr + offset = 0x1004
> 
> with your change
> 
> ==>  void __iomem *vaddr = 0x1000;
> 
> ==>  vaddr + offset = 0x1001
> 
> This comment is there for a reason:
> >       /* shift implied by pointer type */
> 
> Thanks,
> 
>         tglx

-- 
Andy Shevchenko <andriy.shevchenko@linux.intel.com>
Intel Finland Oy