-----"Jason Gunthorpe" <jgg@xxxxxxxx> wrote: -----
>To: "Bernard Metzler" <BMT@xxxxxxxxxxxxxx>
>From: "Jason Gunthorpe" <jgg@xxxxxxxx>
>Date: 06/11/2020 04:24PM
>Cc: "Tom Seewald" <tseewald@xxxxxxxxx>, linux-rdma@xxxxxxxxxxxxxxx,
>"Doug Ledford" <dledford@xxxxxxxxxx>
>Subject: [EXTERNAL] Re: Re: [PATCH next] siw: Fix pointer-to-int-cast
>warning in siw_rx_pbl()
>
>On Thu, Jun 11, 2020 at 02:11:51PM +0000, Bernard Metzler wrote:
>>
>> >To: "Bernard Metzler" <BMT@xxxxxxxxxxxxxx>
>> >From: "Jason Gunthorpe" <jgg@xxxxxxxx>
>> >Date: 06/11/2020 01:35PM
>> >Cc: "Tom Seewald" <tseewald@xxxxxxxxx>,
>linux-rdma@xxxxxxxxxxxxxxx,
>> >"Doug Ledford" <dledford@xxxxxxxxxx>
>> >Subject: [EXTERNAL] Re: Re: [PATCH next] siw: Fix
>pointer-to-int-cast
>> >warning in siw_rx_pbl()
>> >
>> >On Thu, Jun 11, 2020 at 10:21:49AM +0000, Bernard Metzler wrote:
>> >>
>> >> >To: "Tom Seewald" <tseewald@xxxxxxxxx>
>> >> >From: "Jason Gunthorpe" <jgg@xxxxxxxx>
>> >> >Date: 06/10/2020 07:50PM
>> >> >Cc: linux-rdma@xxxxxxxxxxxxxxx, "Bernard Metzler"
>> >> ><bmt@xxxxxxxxxxxxxx>, "Doug Ledford" <dledford@xxxxxxxxxx>
>> >> >Subject: [EXTERNAL] Re: [PATCH next] siw: Fix
>pointer-to-int-cast
>> >> >warning in siw_rx_pbl()
>> >> >
>> >> >On Wed, Jun 10, 2020 at 12:47:17PM -0500, Tom Seewald wrote:
>> >> >> The variable buf_addr is type dma_addr_t, which may not be
>the
>> >same
>> >> >size
>> >> >> as a pointer. To ensure it is the correct size, cast to a
>> >> >uintptr_t.
>> >> >>
>> >> >> Signed-off-by: Tom Seewald <tseewald@xxxxxxxxx>
>> >> >> drivers/infiniband/sw/siw/siw_qp_rx.c | 3 ++-
>> >> >> 1 file changed, 2 insertions(+), 1 deletion(-)
>> >> >>
>> >> >> diff --git a/drivers/infiniband/sw/siw/siw_qp_rx.c
>> >> >b/drivers/infiniband/sw/siw/siw_qp_rx.c
>> >> >> index 650520244ed0..7271d705f4b0 100644
>> >> >> +++ b/drivers/infiniband/sw/siw/siw_qp_rx.c
>> >> >> @@ -139,7 +139,8 @@ static int siw_rx_pbl(struct
>siw_rx_stream
>> >> >*srx, int *pbl_idx,
>> >> >> break;
>> >> >>
>> >> >> bytes = min(bytes, len);
>> >> >> - if (siw_rx_kva(srx, (void *)buf_addr, bytes) == bytes) {
>> >> >> + if (siw_rx_kva(srx, (void *)(uintptr_t)buf_addr, bytes) ==
>> >> >> + bytes) {
>> >> >
>> >> >How is a dma_addr_t being cast to a void *? That can't be
>right?
>> >> >Bernard??
>> >> >
>> >> >Jason
>> >> >
>> >> Hi Tom, Hi Jason,
>> >>
>> >> Thanks for looking into that.
>> >>
>> >> siw_rx_kva() calls skb_copy_bits() to move data to its
>> >> kernel clients destination. It expects a void * target
>> >> address. This is why I chose it for siw_rx_kva() as well.
>> >> One could say siw_rx_kva() should better get an uintptr_t
>> >> as target argument, which would probably make it look
>> >> more clean. And we rename it to siw_rx_kbuf(), and we
>> >> cast from uintptr_t to (void *) just for
>> >> skb_copy_bits(skb *, off, (void *)dest, len)
>> >>
>> >> This would avoid all those nasty (void *) casting at all (!)
>> >> the places we are calling siw_rx_kva().
>> >
>> >But where did the dma_addr_t come from?
>> >
>> It initially comes from the scatterlist provided by the
>> kernel user via drivers .map_mr_sg() method. There we get a
>> dma_addr_t describing the users buffer.
>
>For the SW dma maps you have to convert the dma_addr_t to a kva using
>kmap, it cannot just be casted.
>
True for a real dma addr. But here the user initially came
with an address it got from dma_virt_ops.dma_virt_map_page(),
which provides the virtual address of the page referenced,
casted to dma_addr_t. So we mimic some great dma_addr_t stuff
we do not need for a SW driver and in the end we even have
to call heavy kmap_atomic() to just get the very same addr again.
This I don't want. It's why I just casted it back to void *.
hmmm.
Thanks,
Bernard.
