[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <aa5a91defcca4a4cb8e2317e26385010@AcuMS.aculab.com>
Date: Wed, 24 Mar 2021 17:12:38 +0000
From: David Laight <David.Laight@...LAB.COM>
To: 'Alex Elder' <elder@...e.org>, 'Alex Elder' <elder@...aro.org>,
"davem@...emloft.net" <davem@...emloft.net>,
"kuba@...nel.org" <kuba@...nel.org>
CC: "rdunlap@...radead.org" <rdunlap@...radead.org>,
"bjorn.andersson@...aro.org" <bjorn.andersson@...aro.org>,
"evgreen@...omium.org" <evgreen@...omium.org>,
"cpratapa@...eaurora.org" <cpratapa@...eaurora.org>,
"subashab@...eaurora.org" <subashab@...eaurora.org>,
"elder@...nel.org" <elder@...nel.org>,
"netdev@...r.kernel.org" <netdev@...r.kernel.org>,
"linux-kernel@...r.kernel.org" <linux-kernel@...r.kernel.org>
Subject: RE: [PATCH net-next] net: ipa: avoid 64-bit modulus
From: Alex Elder
> Sent: 24 March 2021 17:07
>
> On 3/24/21 11:27 AM, David Laight wrote:
> > From: Alex Elder
> >> Sent: 23 March 2021 01:05
> >> It is possible for a 32 bit x86 build to use a 64 bit DMA address.
> >>
> >> There are two remaining spots where the IPA driver does a modulo
> >> operation to check alignment of a DMA address, and under certain
> >> conditions this can lead to a build error on i386 (at least).
> >>
> >> The alignment checks we're doing are for power-of-2 values, and this
> >> means the lower 32 bits of the DMA address can be used. This ensures
> >> both operands to the modulo operator are 32 bits wide.
> >>
> >> Reported-by: Randy Dunlap <rdunlap@...radead.org>
> >> Signed-off-by: Alex Elder <elder@...aro.org>
> >> ---
> >> drivers/net/ipa/gsi.c | 11 +++++++----
> >> drivers/net/ipa/ipa_table.c | 9 ++++++---
> >> 2 files changed, 13 insertions(+), 7 deletions(-)
> >>
> >> diff --git a/drivers/net/ipa/gsi.c b/drivers/net/ipa/gsi.c
> >> index 7f3e338ca7a72..b6355827bf900 100644
> >> --- a/drivers/net/ipa/gsi.c
> >> +++ b/drivers/net/ipa/gsi.c
> >> @@ -1436,15 +1436,18 @@ static void gsi_evt_ring_rx_update(struct gsi_evt_ring *evt_ring, u32
> index)
> >> /* Initialize a ring, including allocating DMA memory for its entries */
> >> static int gsi_ring_alloc(struct gsi *gsi, struct gsi_ring *ring, u32 count)
> >> {
> >> - size_t size = count * GSI_RING_ELEMENT_SIZE;
> >> + u32 size = count * GSI_RING_ELEMENT_SIZE;
> >> struct device *dev = gsi->dev;
> >> dma_addr_t addr;
> >>
> >> - /* Hardware requires a 2^n ring size, with alignment equal to size */
> >> + /* Hardware requires a 2^n ring size, with alignment equal to size.
> >> + * The size is a power of 2, so we can check alignment using just
> >> + * the bottom 32 bits for a DMA address of any size.
> >> + */
> >> ring->virt = dma_alloc_coherent(dev, size, &addr, GFP_KERNEL);
> >
> > Doesn't dma_alloc_coherent() guarantee that alignment?
> > I doubt anywhere else checks?
>
> I normally wouldn't check something like this if it
> weren't guaranteed. I'm not sure why I did it here.
>
> I see it's "guaranteed to be aligned to the smallest
> PAGE_SIZE order which is greater than or equal to
> the requested size." So I think the answer to your
> question is "yes, it does guarantee that."
>
> I'll make a note to remove this check in a future
> patch, and will credit you with the suggestion.
I think 'count' is also required to be a power of 2.
so you could have checked 'addr & (size - 1)'.
David
-
Registered Address Lakeside, Bramley Road, Mount Farm, Milton Keynes, MK1 1PT, UK
Registration No: 1397386 (Wales)
Powered by blists - more mailing lists