[dm-devel] [PATCH] block: set the bounce_pfn to the actual DMA limit rather than to max memory

[Jens Axboe]

On 2010-10-01 04:30, Malahal Naineni wrote:
> Jens Axboe [jaxboe at fusionio.com] wrote:
>> On 2010-09-24 21:20, Malahal Naineni wrote:
>>> Jens Axboe [jaxboe at fusionio.com] wrote:
>>>>> This patch also _fixed_ our problem. So we are fine with either patch,
>>>>> but this patch is preferred as it enables more request merges. Also,
>>>>> both patches maybe needed for some configurations.
>>>>
>>>> Plus it doesn't needlessly bounce, that's the real problem you want to
>>>> fix. I have applied this thread patch to for-2.6.37/core, thanks.
>>>
>>> There is a shortcut check in blk_queue_bounce() that uses blk_max_pfn to
>>> return without doing anything.  blk_max_pfn is not updated when we do
>>> hot-plug memory add, so bounce buffers are NOT really used in our case
>>> (thankfully)! 
>>
>> Any reason we can't just add a hot mem add notifier and update the block
>> copies when we need to?
> 
> Actually it is quite simple, but power arch guys decided not to update
> max_pfn and max_low_pfn when someone adds or removes memory. They do
> seem to indicate that it is not an over sight! On the other hand, x86
> arch does update those.
> 
> Another approach would be defining max_possible_pfn (maximum allowable
> memory) which would be constant (based on architecture) and hopefully
> adapters that we care still take the short-cut.
> 
> Also, this patch broke some USB hosts. Can you please back out this
> patch and apply the one below instead:
> 
> http://thread.gmane.org/gmane.linux.kernel.next/14158

Done, thanks.

-- 
Jens Axboe