* [PATCH net] gve: fix dma sync bug where not all pages synced
@ 2019-11-18 22:36 Adi Suresh
2019-11-18 23:19 ` Jakub Kicinski
0 siblings, 1 reply; 3+ messages in thread
From: Adi Suresh @ 2019-11-18 22:36 UTC (permalink / raw)
To: netdev; +Cc: Adi Suresh, Catherine Sullivan
The previous commit "Fixes DMA synchronization" had a bug where the
last page in the memory range could not be synced. This change fixes
the behavior so that all the required pages are synced.
Signed-off-by: Adi Suresh <adisuresh@google.com>
Reviewed-by: Catherine Sullivan <csully@google.com>
---
drivers/net/ethernet/google/gve/gve_tx.c | 10 +++++-----
1 file changed, 5 insertions(+), 5 deletions(-)
diff --git a/drivers/net/ethernet/google/gve/gve_tx.c b/drivers/net/ethernet/google/gve/gve_tx.c
index 0a9a7ee2a866..89271380bbfd 100644
--- a/drivers/net/ethernet/google/gve/gve_tx.c
+++ b/drivers/net/ethernet/google/gve/gve_tx.c
@@ -393,12 +393,12 @@ static void gve_tx_fill_seg_desc(union gve_tx_desc *seg_desc,
static void gve_dma_sync_for_device(struct device *dev, dma_addr_t *page_buses,
u64 iov_offset, u64 iov_len)
{
- dma_addr_t dma;
- u64 addr;
+ u64 last_page = (iov_offset + iov_len - 1) / PAGE_SIZE;
+ u64 first_page = iov_offset / PAGE_SIZE;
+ u64 page;
- for (addr = iov_offset; addr < iov_offset + iov_len;
- addr += PAGE_SIZE) {
- dma = page_buses[addr / PAGE_SIZE];
+ for (page = first_page; page <= last_page; page++) {
+ dma_addr_t dma = page_buses[page];
dma_sync_single_for_device(dev, dma, PAGE_SIZE, DMA_TO_DEVICE);
}
}
--
2.24.0.432.g9d3f5f5b63-goog
^ permalink raw reply related [flat|nested] 3+ messages in thread
* Re: [PATCH net] gve: fix dma sync bug where not all pages synced
2019-11-18 22:36 [PATCH net] gve: fix dma sync bug where not all pages synced Adi Suresh
@ 2019-11-18 23:19 ` Jakub Kicinski
2019-11-18 23:43 ` Adi Suresh
0 siblings, 1 reply; 3+ messages in thread
From: Jakub Kicinski @ 2019-11-18 23:19 UTC (permalink / raw)
To: Adi Suresh; +Cc: netdev, Catherine Sullivan
On Mon, 18 Nov 2019 14:36:30 -0800, Adi Suresh wrote:
> The previous commit "Fixes DMA synchronization" had a bug where the
Please use the standard way of quoting commits.
> last page in the memory range could not be synced. This change fixes
> the behavior so that all the required pages are synced.
>
Please add a Fixes tag.
> Signed-off-by: Adi Suresh <adisuresh@google.com>
> Reviewed-by: Catherine Sullivan <csully@google.com>
> ---
> drivers/net/ethernet/google/gve/gve_tx.c | 10 +++++-----
> 1 file changed, 5 insertions(+), 5 deletions(-)
>
> diff --git a/drivers/net/ethernet/google/gve/gve_tx.c b/drivers/net/ethernet/google/gve/gve_tx.c
> index 0a9a7ee2a866..89271380bbfd 100644
> --- a/drivers/net/ethernet/google/gve/gve_tx.c
> +++ b/drivers/net/ethernet/google/gve/gve_tx.c
> @@ -393,12 +393,12 @@ static void gve_tx_fill_seg_desc(union gve_tx_desc *seg_desc,
> static void gve_dma_sync_for_device(struct device *dev, dma_addr_t *page_buses,
> u64 iov_offset, u64 iov_len)
> {
> - dma_addr_t dma;
> - u64 addr;
> + u64 last_page = (iov_offset + iov_len - 1) / PAGE_SIZE;
> + u64 first_page = iov_offset / PAGE_SIZE;
> + u64 page;
>
> - for (addr = iov_offset; addr < iov_offset + iov_len;
> - addr += PAGE_SIZE) {
> - dma = page_buses[addr / PAGE_SIZE];
> + for (page = first_page; page <= last_page; page++) {
> + dma_addr_t dma = page_buses[page];
Empty line after variable declaration. Perhaps keep the dma declaration
outside the loop, since this is a fix the smaller the better.
> dma_sync_single_for_device(dev, dma, PAGE_SIZE, DMA_TO_DEVICE);
> }
> }
^ permalink raw reply [flat|nested] 3+ messages in thread
* Re: [PATCH net] gve: fix dma sync bug where not all pages synced
2019-11-18 23:19 ` Jakub Kicinski
@ 2019-11-18 23:43 ` Adi Suresh
0 siblings, 0 replies; 3+ messages in thread
From: Adi Suresh @ 2019-11-18 23:43 UTC (permalink / raw)
To: Jakub Kicinski; +Cc: netdev, Catherine Sullivan
On Mon, Nov 18, 2019 at 3:19 PM Jakub Kicinski
<jakub.kicinski@netronome.com> wrote:
>
> On Mon, 18 Nov 2019 14:36:30 -0800, Adi Suresh wrote:
> > The previous commit "Fixes DMA synchronization" had a bug where the
>
> Please use the standard way of quoting commits.
>
> > last page in the memory range could not be synced. This change fixes
> > the behavior so that all the required pages are synced.
> >
>
> Please add a Fixes tag.
>
> > Signed-off-by: Adi Suresh <adisuresh@google.com>
> > Reviewed-by: Catherine Sullivan <csully@google.com>
> > ---
> > drivers/net/ethernet/google/gve/gve_tx.c | 10 +++++-----
> > 1 file changed, 5 insertions(+), 5 deletions(-)
> >
> > diff --git a/drivers/net/ethernet/google/gve/gve_tx.c b/drivers/net/ethernet/google/gve/gve_tx.c
> > index 0a9a7ee2a866..89271380bbfd 100644
> > --- a/drivers/net/ethernet/google/gve/gve_tx.c
> > +++ b/drivers/net/ethernet/google/gve/gve_tx.c
> > @@ -393,12 +393,12 @@ static void gve_tx_fill_seg_desc(union gve_tx_desc *seg_desc,
> > static void gve_dma_sync_for_device(struct device *dev, dma_addr_t *page_buses,
> > u64 iov_offset, u64 iov_len)
> > {
> > - dma_addr_t dma;
> > - u64 addr;
> > + u64 last_page = (iov_offset + iov_len - 1) / PAGE_SIZE;
> > + u64 first_page = iov_offset / PAGE_SIZE;
> > + u64 page;
> >
> > - for (addr = iov_offset; addr < iov_offset + iov_len;
> > - addr += PAGE_SIZE) {
> > - dma = page_buses[addr / PAGE_SIZE];
> > + for (page = first_page; page <= last_page; page++) {
> > + dma_addr_t dma = page_buses[page];
>
> Empty line after variable declaration. Perhaps keep the dma declaration
> outside the loop, since this is a fix the smaller the better.
>
> > dma_sync_single_for_device(dev, dma, PAGE_SIZE, DMA_TO_DEVICE);
> > }
> > }
>
Comments Addressed in v2
^ permalink raw reply [flat|nested] 3+ messages in thread
end of thread, other threads:[~2019-11-18 23:43 UTC | newest]
Thread overview: 3+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2019-11-18 22:36 [PATCH net] gve: fix dma sync bug where not all pages synced Adi Suresh
2019-11-18 23:19 ` Jakub Kicinski
2019-11-18 23:43 ` Adi Suresh
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).