2019-12-10 13:09:01

by Mehta, Sanju

[permalink] [raw]
Subject: [PATCH] ntb_perf: pass correct struct device to dma_alloc_coherent

From: Sanjay R Mehta <[email protected]>

Currently, ntb->dev is passed to dma_alloc_coherent
and dma_free_coherent calls. The returned dma_addr_t
is the CPU physical address. This works fine as long
as IOMMU is disabled. But when IOMMU is enabled, we
need to make sure that IOVA is returned for dma_addr_t.
So the correct way to achieve this is by changing the
first parameter of dma_alloc_coherent() as ntb->pdev->dev
instead.

Fixes: 5648e56 ("NTB: ntb_perf: Add full multi-port NTB API support")
Signed-off-by: Sanjay R Mehta <[email protected]>
---
drivers/ntb/test/ntb_perf.c | 7 ++++---
1 file changed, 4 insertions(+), 3 deletions(-)

diff --git a/drivers/ntb/test/ntb_perf.c b/drivers/ntb/test/ntb_perf.c
index f5df33a..8729838 100644
--- a/drivers/ntb/test/ntb_perf.c
+++ b/drivers/ntb/test/ntb_perf.c
@@ -559,7 +559,7 @@ static void perf_free_inbuf(struct perf_peer *peer)
return;

(void)ntb_mw_clear_trans(peer->perf->ntb, peer->pidx, peer->gidx);
- dma_free_coherent(&peer->perf->ntb->dev, peer->inbuf_size,
+ dma_free_coherent(&peer->perf->ntb->pdev->dev, peer->inbuf_size,
peer->inbuf, peer->inbuf_xlat);
peer->inbuf = NULL;
}
@@ -588,8 +588,9 @@ static int perf_setup_inbuf(struct perf_peer *peer)

perf_free_inbuf(peer);

- peer->inbuf = dma_alloc_coherent(&perf->ntb->dev, peer->inbuf_size,
- &peer->inbuf_xlat, GFP_KERNEL);
+ peer->inbuf = dma_alloc_coherent(&perf->ntb->pdev->dev,
+ peer->inbuf_size, &peer->inbuf_xlat,
+ GFP_KERNEL);
if (!peer->inbuf) {
dev_err(&perf->ntb->dev, "Failed to alloc inbuf of %pa\n",
&peer->inbuf_size);
--
2.7.4


2019-12-10 17:32:58

by Logan Gunthorpe

[permalink] [raw]
Subject: Re: [PATCH] ntb_perf: pass correct struct device to dma_alloc_coherent


On 2019-12-10 6:07 a.m., Sanjay R Mehta wrote:
> From: Sanjay R Mehta <[email protected]>
>
> Currently, ntb->dev is passed to dma_alloc_coherent
> and dma_free_coherent calls. The returned dma_addr_t
> is the CPU physical address. This works fine as long
> as IOMMU is disabled. But when IOMMU is enabled, we
> need to make sure that IOVA is returned for dma_addr_t.
> So the correct way to achieve this is by changing the
> first parameter of dma_alloc_coherent() as ntb->pdev->dev
> instead.
>
> Fixes: 5648e56 ("NTB: ntb_perf: Add full multi-port NTB API support")
> Signed-off-by: Sanjay R Mehta <[email protected]>

Yes, I did the same thing as one of the patches in my fix-up series that
never got merged. See [1].

Hopefully you can make better progress than I did.

While you're at it I think it's worth doing the same thing in ntb_tool
as well as removing the dma_coerce_mask_and_coherent() calls that are in
the NTB drivers which are meaningless once we get back to using the
correct DMA device.

Thanks,

Logan

[1] https://lore.kernel.org/lkml/[email protected]/

> ---
> drivers/ntb/test/ntb_perf.c | 7 ++++---
> 1 file changed, 4 insertions(+), 3 deletions(-)
>
> diff --git a/drivers/ntb/test/ntb_perf.c b/drivers/ntb/test/ntb_perf.c
> index f5df33a..8729838 100644
> --- a/drivers/ntb/test/ntb_perf.c
> +++ b/drivers/ntb/test/ntb_perf.c
> @@ -559,7 +559,7 @@ static void perf_free_inbuf(struct perf_peer *peer)
> return;
>
> (void)ntb_mw_clear_trans(peer->perf->ntb, peer->pidx, peer->gidx);
> - dma_free_coherent(&peer->perf->ntb->dev, peer->inbuf_size,
> + dma_free_coherent(&peer->perf->ntb->pdev->dev, peer->inbuf_size,
> peer->inbuf, peer->inbuf_xlat);
> peer->inbuf = NULL;
> }
> @@ -588,8 +588,9 @@ static int perf_setup_inbuf(struct perf_peer *peer)
>
> perf_free_inbuf(peer);
>
> - peer->inbuf = dma_alloc_coherent(&perf->ntb->dev, peer->inbuf_size,
> - &peer->inbuf_xlat, GFP_KERNEL);
> + peer->inbuf = dma_alloc_coherent(&perf->ntb->pdev->dev,
> + peer->inbuf_size, &peer->inbuf_xlat,
> + GFP_KERNEL);
> if (!peer->inbuf) {
> dev_err(&perf->ntb->dev, "Failed to alloc inbuf of %pa\n",
> &peer->inbuf_size);
>

2019-12-11 12:14:18

by Sanjay R Mehta

[permalink] [raw]
Subject: Re: [PATCH] ntb_perf: pass correct struct device to dma_alloc_coherent


On 12/10/2019 11:01 PM, Logan Gunthorpe wrote:
> [CAUTION: External Email]
>
> On 2019-12-10 6:07 a.m., Sanjay R Mehta wrote:
>> From: Sanjay R Mehta <[email protected]>
>>
>> Currently, ntb->dev is passed to dma_alloc_coherent
>> and dma_free_coherent calls. The returned dma_addr_t
>> is the CPU physical address. This works fine as long
>> as IOMMU is disabled. But when IOMMU is enabled, we
>> need to make sure that IOVA is returned for dma_addr_t.
>> So the correct way to achieve this is by changing the
>> first parameter of dma_alloc_coherent() as ntb->pdev->dev
>> instead.
>>
>> Fixes: 5648e56 ("NTB: ntb_perf: Add full multi-port NTB API support")
>> Signed-off-by: Sanjay R Mehta <[email protected]>
> Yes, I did the same thing as one of the patches in my fix-up series that
> never got merged. See [1].
>
> Hopefully you can make better progress than I did.
>
> While you're at it I think it's worth doing the same thing in ntb_tool
> as well as removing the dma_coerce_mask_and_coherent() calls that are in
> the NTB drivers which are meaningless once we get back to using the
> correct DMA device.
Yes, What you said make sense. I will address your comment in the next version of patch, but I will break them down into two separate patches.
>
> Thanks,
>
> Logan
>
> [1] https://nam11.safelinks.protection.outlook.com/?url=https%3A%2F%2Flore.kernel.org%2Flkml%2F20190109192233.5752-3-logang%40deltatee.com%2F&amp;data=02%7C01%7CSanju.Mehta%40amd.com%7C1f043932c1c1425a836908d77d96e265%7C3dd8961fe4884e608e11a82d994e183d%7C0%7C0%7C637115959326469198&amp;sdata=BoDyfqTML39JhIvO7%2FvYGnKNRKe2rSKgZxJ0uT6dtsQ%3D&amp;reserved=0
>
>> ---
>> drivers/ntb/test/ntb_perf.c | 7 ++++---
>> 1 file changed, 4 insertions(+), 3 deletions(-)
>>
>> diff --git a/drivers/ntb/test/ntb_perf.c b/drivers/ntb/test/ntb_perf.c
>> index f5df33a..8729838 100644
>> --- a/drivers/ntb/test/ntb_perf.c
>> +++ b/drivers/ntb/test/ntb_perf.c
>> @@ -559,7 +559,7 @@ static void perf_free_inbuf(struct perf_peer *peer)
>> return;
>>
>> (void)ntb_mw_clear_trans(peer->perf->ntb, peer->pidx, peer->gidx);
>> - dma_free_coherent(&peer->perf->ntb->dev, peer->inbuf_size,
>> + dma_free_coherent(&peer->perf->ntb->pdev->dev, peer->inbuf_size,
>> peer->inbuf, peer->inbuf_xlat);
>> peer->inbuf = NULL;
>> }
>> @@ -588,8 +588,9 @@ static int perf_setup_inbuf(struct perf_peer *peer)
>>
>> perf_free_inbuf(peer);
>>
>> - peer->inbuf = dma_alloc_coherent(&perf->ntb->dev, peer->inbuf_size,
>> - &peer->inbuf_xlat, GFP_KERNEL);
>> + peer->inbuf = dma_alloc_coherent(&perf->ntb->pdev->dev,
>> + peer->inbuf_size, &peer->inbuf_xlat,
>> + GFP_KERNEL);
>> if (!peer->inbuf) {
>> dev_err(&perf->ntb->dev, "Failed to alloc inbuf of %pa\n",
>> &peer->inbuf_size);
>>