RE: [PATCH for-next v5 5/7] RDMA/rxe: Allow registering MRs for On-Demand Paging

[Daisuke Matsuda (Fujitsu)]

On Tue, June 13, 2023 1:19 AM Jason Gunthorpe wrote:

Sorry for my long silence.
I must spend most of my time for other works these days,
but I am still willing to complete this and subsequent works.

> 
> On Thu, May 18, 2023 at 05:21:50PM +0900, Daisuke Matsuda wrote:
> 
> > +static void rxe_mr_set_xarray(struct rxe_mr *mr, unsigned long start,
> > +			      unsigned long end, unsigned long *pfn_list)
> > +{
> > +	unsigned long lower, upper, idx;
> > +	struct page *page;
> > +
> > +	lower = rxe_mr_iova_to_index(mr, start);
> > +	upper = rxe_mr_iova_to_index(mr, end);
> > +
> > +	/* make pages visible in xarray. no sleep while taking the lock */
> > +	spin_lock(&mr->page_list.xa_lock);
> > +	for (idx = lower; idx <= upper; idx++) {
> > +		page = hmm_pfn_to_page(pfn_list[idx]);
> > +		__xa_store(&mr->page_list, idx, page, GFP_ATOMIC);
> 
> All of these loops can be performance improved a lot by using xas
> loops

Well, do you mean I should use 'xas_for_each()' here?
That is the same 'for-loop' after all, so performance may not be improved.
Additionally, the 'idx' value above must be counted separately in that case.

> 
> >  				    unsigned long cur_seq)
> > @@ -54,3 +72,105 @@ static bool rxe_ib_invalidate_range(struct mmu_interval_notifier *mni,
> >  const struct mmu_interval_notifier_ops rxe_mn_ops = {
> >  	.invalidate = rxe_ib_invalidate_range,
> >  };
> > +
> > +#define RXE_PAGEFAULT_RDONLY BIT(1)
> > +#define RXE_PAGEFAULT_SNAPSHOT BIT(2)
> > +static int rxe_odp_do_pagefault(struct rxe_mr *mr, u64 user_va, int bcnt, u32 flags)
> > +{
> > +	int np;
> > +	u64 access_mask;
> > +	bool fault = !(flags & RXE_PAGEFAULT_SNAPSHOT);
> > +	struct ib_umem_odp *umem_odp = to_ib_umem_odp(mr->umem);
> > +
> > +	access_mask = ODP_READ_ALLOWED_BIT;
> > +	if (umem_odp->umem.writable && !(flags & RXE_PAGEFAULT_RDONLY))
> > +		access_mask |= ODP_WRITE_ALLOWED_BIT;
> > +
> > +	/*
> > +	 * ib_umem_odp_map_dma_and_lock() locks umem_mutex on success.
> > +	 * Callers must release the lock later to let invalidation handler
> > +	 * do its work again.
> > +	 */
> > +	np = ib_umem_odp_map_dma_and_lock(umem_odp, user_va, bcnt,
> > +					  access_mask, fault);
> > +	if (np < 0)
> > +		return np;
> > +
> > +	/* umem_mutex is still locked here, so we can use hmm_pfn_to_page()
> > +	 * safely to fetch pages in the range.
> 
> All the comments should be in the style like the first one, not the
> second

I got it.

> 
> > +	 */
> > +	rxe_mr_set_xarray(mr, user_va, user_va + bcnt, umem_odp->pfn_list);
> > +
> > +	return np;
> > +}
> > +
> > +static int rxe_odp_init_pages(struct rxe_mr *mr)
> > +{
> > +	int ret;
> > +	struct ib_umem_odp *umem_odp = to_ib_umem_odp(mr->umem);
> > +
> > +	ret = rxe_odp_do_pagefault(mr, mr->umem->address, mr->umem->length,
> > +				   RXE_PAGEFAULT_SNAPSHOT);
> 
> Probably suffix this with "and_lock"

Agreed.

> 
> > +	mr->odp_enabled = true;
> > +	mr->umem = &umem_odp->umem;
> > +	mr->access = access_flags;
> > +	mr->ibmr.length = length;
> > +	mr->ibmr.iova = iova;
> > +	mr->page_offset = ib_umem_offset(&umem_odp->umem);
> > +
> > +	err = rxe_odp_init_pages(mr);
> > +	if (err) {
> > +		ib_umem_odp_release(umem_odp);
> > +		return err;
> > +	}
> > +
> > +	err = rxe_mr_fill_pages_from_sgt(mr, &umem_odp->umem.sgt_append.sgt);
> 
> Uh? What is this? The sgt is not used in the ODP mode?

This function fills the xarray in the non-ODP mode. 
You are correct the sgt is not used in ODP, but this works somehow.
I will fix this to fetch pages from umem_odp->pfn_list instead.

> 
> > diff --git a/drivers/infiniband/sw/rxe/rxe_verbs.h b/drivers/infiniband/sw/rxe/rxe_verbs.h
> > index b6fbd9b3d086..de5a982c7c7e 100644
> > --- a/drivers/infiniband/sw/rxe/rxe_verbs.h
> > +++ b/drivers/infiniband/sw/rxe/rxe_verbs.h
> > @@ -333,6 +333,8 @@ struct rxe_mr {
> >  	u32			nbuf;
> >
> >  	struct xarray		page_list;
> > +
> > +	bool			odp_enabled;
> 
> You can tell from the umem, don't need a flag

Certainly.

Daisuke

> 
> Jason