Re: Fwd: [PATCH v7 1/2] x86: Add Isolated Memory Regions for Quark X1000

["Bryan O'Donoghue" <pure.logic@nexus-software.ie>]

On 30/01/15 12:55, Bryan O'Donoghue wrote:
> Oops.
>
> Hit reply not reply-all

On 30/01/15 12:03, Andy Shevchenko wrote:
 >> +               return -ENOMEM;
 >
 > When CONFIG_DEBUGFS=n, you will get error pointer here, which is not 
NULL.
 >
 > So, the proper check is
 > if (IS_ERR())
 >   return PTR_ERR();
 > if (!file)
 >   return -ENOMEM;

Yeah I saw that. Also saw that most other code doesn't bother trapping
those return values - so skipped it.
No issue adding.

 >
 >> +               } else {
 >> +                       reg = i;
 >
 > Do we go always through all IMRs and choose the last one?
 > If no, break is missed here.

Yep - we always choose the last one.


 >> +               }
 >> +       }
 >> +
 >> +       /* Error out if we have no free IMR entries. */
 >> +       if (reg == -1) {
 >> +               ret = -ENODEV;
 >
 > -ENOMEM ? Like you said there is no *free* IMR.

OK

 >> + * imr_remove_range(0, size, base); delete IMR at index 0 base/size 
ignored.
 >> + * imr_remove_range(-1, base, size); delete IMR from base to base+size.
 >
 > (size, base) or (base, size) ?

base, size that's a documentation typo :)

 >> +
 >> +       ret = imr_check_params(base, size);
 >> +       if (ret == -EINVAL || (ret == -ENOMEM && reg == -1))
 >
 > reg base size (0 correct, 1 wrong):
 >
 > 0 0 0 — which should be used? what is the priority?
 > 0 x 1 — index
 > 0 1 x — index
 > 1 0 0 — address
 > 1 0 1 — an error
 > 1 1 1 — an error
 >
 > Thus, could it be simpler? Like
 > if (reg < 0 && ret) ?

ret will be EINVAL for unaligned base or size
ret will be ENOMEM when reg == -1 and size == 0

I could probably write it like this to make it clearer
(ret == -EINVAL || (reg == -1 && size == 0)
	return -EINVAL;

traps unaligned input - for address range tear-down
traps zero sized - for address range tear-down

Allows index based teardown i.e. reg >= 0

 > if (ret)
 >
 >> +               pr_warn("debugfs register failed!\n");
 >
 > Do we actually need this? Or move it to debug level.

It was your suggestion @ a previous review ....

 > Here is the mix of kernel levels. What about to align them?
 >
 > For example I doubt we need to distinguish messages by level:
 >
 > pr_info();
 > vprintk(KERN_INFO fmt, …);

OK fine.