lkml.org 
[lkml]   [2011]   [Jun]   [17]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
Date
SubjectRe: [PATCH] ring_buffer: Ensure that buffer page data is aligned.
From
On Thu, Jun 16, 2011 at 6:24 PM, Steven Rostedt <rostedt@goodmis.org> wrote:
> On Thu, 2011-06-16 at 14:32 +0100, Will Newton wrote:
>> On Thu, Jun 16, 2011 at 2:17 PM, Steven Rostedt <rostedt@goodmis.org> wrote:
>> > On Thu, 2011-06-16 at 14:03 +0100, Will Newton wrote:
>> >> Explicitly align the start of the buffer page data array to the
>> >> required arch alignment. This is required for architectures that
>> >> require 8 byte alignment but do not have a 8 byte local_t.
>> >
>> > What arch does that? A 32bit arch that forces 8 byte alignment?
>>
>> Yes, it's likely for all 64bit arches will be aligned correctly.
>
> What arch is this exactly. Is it because 64bit variables must be aligned
> on 8byte boundaries?

Yes, it's a 32bit system with 64bit alignment required for 64bit
variables. I'm not sure whether there are other architectures in the
tree that have this characteristic but it's relatively common in cores
with DSP heritage.

http://imgtec.com/meta/meta-technology.asp

>>
>> >>
>> >> Signed-off-by: Will Newton <will.newton@imgtec.com>
>> >> ---
>> >>  kernel/trace/ring_buffer.c |    4 +++-
>> >>  1 files changed, 3 insertions(+), 1 deletions(-)
>> >>
>> >> I don't believe that any currently in-tree architecture is affected by
>> >> this, but it can be an issue on 32bit architectures that require an
>> >> 8 byte aligment but only have a 32bit local_t. I think it's potentially
>> >> cleaner to make the alignment explicit anyway.
>> >>
>> >> diff --git a/kernel/trace/ring_buffer.c b/kernel/trace/ring_buffer.c
>> >> index 0ef7b4b..36d5699 100644
>> >> --- a/kernel/trace/ring_buffer.c
>> >> +++ b/kernel/trace/ring_buffer.c
>> >> @@ -215,6 +215,8 @@ EXPORT_SYMBOL_GPL(tracing_is_on);
>> >>  # define RB_ARCH_ALIGNMENT           8U
>> >>  #endif
>> >>
>> >> +#define RB_ALIGN_DATA                __aligned(RB_ARCH_ALIGNMENT)
>> >> +
>> >
>> > Note, the code above this is:
>> >
>> > #define RB_EVNT_HDR_SIZE (offsetof(struct ring_buffer_event, array))
>> > #define RB_ALIGNMENT            4U
>> > #define RB_MAX_SMALL_DATA       (RB_ALIGNMENT * RINGBUF_TYPE_DATA_TYPE_LEN_MAX)
>> > #define RB_EVNT_MIN_SIZE        8U      /* two 32bit words */
>> >
>> > #if !defined(CONFIG_64BIT) || defined(CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS)
>> > # define RB_FORCE_8BYTE_ALIGNMENT       0
>> > # define RB_ARCH_ALIGNMENT              RB_ALIGNMENT
>> > #else
>> > # define RB_FORCE_8BYTE_ALIGNMENT       1
>> > # define RB_ARCH_ALIGNMENT              8U
>> > #endif
>> >
>> >
>> > Which means that when CONFIG_64BIT is not set, RB_ARCH_ALIGNMENT is 4.
>> > This means that this patch is really a nop and doesn't do anything for
>> > your arch.
>>
>> I have modified that conditional in my local tree so the patch is not
>> a nop there. Obviously it doesn't make any sense to merge that change
>> yet. I thought this part of the patch was a reasonable cleanup to make
>> the code a bit more self documenting though.
>
> Unfortunately, if anything it confuses the code more. If I were
> reviewing it and saw that we are aligning something that is already
> aligned with the same value, I would question that code.

It does however make the conditional more flexible as you could change
it to #if defined (CONFIG_WHATEVER) and it will still work, rather
than the 64bit assumption leaking throughout the code.

> Now if/when your arch is merged, we can add this with the other changes
> you have, which will make reviewing the code not as confusing.

Sure, no problem.
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/

\
 
 \ /
  Last update: 2011-06-17 12:07    [W:0.075 / U:0.624 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site