Re: [PATCH 02/11] io_uring: get rid of remap_pfn_range() for mapping rings/sqes

[Jens Axboe <axboe@kernel.dk>]

On 3/29/24 9:50 PM, Gabriel Krisman Bertazi wrote:
> Jens Axboe <axboe@kernel.dk> writes:
> 
>> Rather than use remap_pfn_range() for this and manually free later,
>> switch to using vm_insert_pages() and have it Just Work.
>>
>> If possible, allocate a single compound page that covers the range that
>> is needed. If that works, then we can just use page_address() on that
>> page. If we fail to get a compound page, allocate single pages and use
>> vmap() to map them into the kernel virtual address space.
>>
>> This just covers the rings/sqes, the other remaining user of the mmap
>> remap_pfn_range() user will be converted separately. Once that is done,
>> we can kill the old alloc/free code.
>>
>> Signed-off-by: Jens Axboe <axboe@kernel.dk>
>> ---
>>  io_uring/io_uring.c | 136 +++++++++++++++++++++++++++++++++++++++++---
>>  io_uring/io_uring.h |   2 +
>>  2 files changed, 130 insertions(+), 8 deletions(-)
>>
>> diff --git a/io_uring/io_uring.c b/io_uring/io_uring.c
>> index 104899522bc5..982545ca23f9 100644
>> --- a/io_uring/io_uring.c
>> +++ b/io_uring/io_uring.c
>> @@ -2594,6 +2594,33 @@ static int io_cqring_wait(struct io_ring_ctx *ctx, int min_events,
>>  	return READ_ONCE(rings->cq.head) == READ_ONCE(rings->cq.tail) ? ret : 0;
>>  }
>>  
>> +static void io_pages_unmap(void *ptr, struct page ***pages,
>> +			   unsigned short *npages)
>> +{
>> +	bool do_vunmap = false;
>> +
>> +	if (*npages) {
>> +		struct page **to_free = *pages;
>> +		int i;
>> +
>> +		/*
>> +		 * Only did vmap for the non-compound multiple page case.
>> +		 * For the compound page, we just need to put the head.
>> +		 */
>> +		if (PageCompound(to_free[0]))
>> +			*npages = 1;
>> +		else if (*npages > 1)
>> +			do_vunmap = true;
>> +		for (i = 0; i < *npages; i++)
>> +			put_page(to_free[i]);
>> +	}
> 
> Hi Jens,
> 
> wouldn't it be simpler to handle the compound case separately as a
> folio?  Then you folio_put the compound page here and just handle the
> non-continuous case after.

I don't think it makes sense, as we're still dealing with pages for
insertion. Once there's some folio variant of inserting pages, then yeah
I think it'd make sense to unify it. If not, we're doing the page <->
folio transition in one spot anyway.

-- 
Jens Axboe