Re: 32bit x86 build broken (was: Re: [GIT PULL] Networking for 5.16-rc1)

[Linus Torvalds]

On Thu, Nov 11, 2021 at 5:46 PM Jakub Kicinski <kuba@xxxxxxxxxx> wrote:
>
> Rafael, Srinivas, we're getting 32 bit build failures after pulling back
> from Linus today.
>
> make[1]: *** [/home/nipa/net/Makefile:1850: drivers] Error 2
> make: *** [Makefile:219: __sub-make] Error 2
> ../drivers/thermal/intel/int340x_thermal/processor_thermal_mbox.c: In function ‘send_mbox_cmd’:
> ../drivers/thermal/intel/int340x_thermal/processor_thermal_mbox.c:79:37: error: implicit declaration of function ‘readq’; did you mean ‘readl’? [-Werror=implicit-function-declaration]
>    79 |                         *cmd_resp = readq((void __iomem *) (proc_priv->mmio_base + MBOX_OFFSET_DATA));
>       |                                     ^~~~~
>       |                                     readl

Gaah.

The trivial fix is *probably* just a simple

    #include <linux/io-64-nonatomic-lo-hi.h>

to say that a non-atomic readq() is ok done low word first. IOW, just a

  --- a/drivers/thermal/intel/int340x_thermal/processor_thermal_mbox.c
  +++ b/drivers/thermal/intel/int340x_thermal/processor_thermal_mbox.c
  @@ -7,6 +7,7 @@
   #include <linux/kernel.h>
   #include <linux/module.h>
   #include <linux/pci.h>
  +#include <linux/io-64-nonatomic-lo-hi.h>
   #include "processor_thermal_device.h"

   #define MBOX_CMD_WORKLOAD_TYPE_READ  0x0E

Of course, it depends on the hardware. It might not matter. Or maybe
it really wants the high word read first.

Looking at that driver, and how it didn't use to do 64-bit reads at
all, I suspect the answer is "doesn't matter".

                Linus