When accessing shared memory to check for the stat of submitted
descriptors, make sure to use READ_ONCE(). This will guarantee the
compiler treats these memory locations as volatile and doesn't apply
any caching.
While this doesn't fix any particular problem I ran into, it's best
practice to do it this way.
Note that this patch also removes the superflous extra condition check
in the do-while loop in reap_tx_dxes(), as the loop will break
instantly anyway in that case.
Signed-off-by: Daniel Mack <[email protected]>
---
drivers/net/wireless/ath/wcn36xx/dxe.c | 7 +++----
1 file changed, 3 insertions(+), 4 deletions(-)
diff --git a/drivers/net/wireless/ath/wcn36xx/dxe.c b/drivers/net/wireless/ath/wcn36xx/dxe.c
index 26e6c42f886a..c52f1c21ece9 100644
--- a/drivers/net/wireless/ath/wcn36xx/dxe.c
+++ b/drivers/net/wireless/ath/wcn36xx/dxe.c
@@ -363,7 +363,7 @@ static void reap_tx_dxes(struct wcn36xx *wcn, struct wcn36xx_dxe_ch *ch)
spin_lock_irqsave(&ch->lock, flags);
ctl = ch->tail_blk_ctl;
do {
- if (ctl->desc->ctrl & WCN36xx_DXE_CTRL_VLD)
+ if (READ_ONCE(ctl->desc->ctrl) & WCN36xx_DXE_CTRL_VLD)
break;
if (ctl->skb) {
dma_unmap_single(wcn->dev, ctl->desc->src_addr_l,
@@ -382,8 +382,7 @@ static void reap_tx_dxes(struct wcn36xx *wcn, struct wcn36xx_dxe_ch *ch)
ctl->skb = NULL;
}
ctl = ctl->next;
- } while (ctl != ch->head_blk_ctl &&
- !(ctl->desc->ctrl & WCN36xx_DXE_CTRL_VLD));
+ } while (ctl != ch->head_blk_ctl);
ch->tail_blk_ctl = ctl;
spin_unlock_irqrestore(&ch->lock, flags);
@@ -525,7 +524,7 @@ static int wcn36xx_rx_handle_packets(struct wcn36xx *wcn,
int_mask = WCN36XX_DXE_INT_CH3_MASK;
}
- while (!(dxe->ctrl & WCN36xx_DXE_CTRL_VLD)) {
+ while (!(READ_ONCE(dxe->ctrl) & WCN36xx_DXE_CTRL_VLD)) {
skb = ctl->skb;
dma_addr = dxe->dst_addr_l;
ret = wcn36xx_dxe_fill_skb(wcn->dev, ctl, GFP_ATOMIC);
--
2.14.3
On Tuesday, April 10, 2018 08:17 PM, Ramon Fried wrote:
> On 10 April 2018 at 20:35, Daniel Mack <[email protected]> wrote:
>> When accessing shared memory to check for the stat of submitted
>> descriptors, make sure to use READ_ONCE(). This will guarantee the
>> compiler treats these memory locations as volatile and doesn't apply
>> any caching.
>
> The structure that is tested is not shared memory. It's accessed only
> by the apps processor.
Hmm? ctl->desc is of type struct wcn36xx_dxe_desc, which is packed and
shared with the firmware. The WCN36xx_DXE_CTRL_VLD bit in ctrl bitfield
is set by the firmware when a frame is valid, and before asserting the
RX interrupt. So the host CPU must treat it as volatile and expect it to
change.
Am I reading this wrong?
Thanks,
Daniel
On 10 April 2018 at 20:35, Daniel Mack <[email protected]> wrote:
> When accessing shared memory to check for the stat of submitted
> descriptors, make sure to use READ_ONCE(). This will guarantee the
> compiler treats these memory locations as volatile and doesn't apply
> any caching.
The structure that is tested is not shared memory. It's accessed only
by the apps processor.
In this case, caching does make sense.
I'm not sure regarding READ_ONCE in that context I'll have to do some
reading regarding that.
But I don't think this is right here.
Thanks,
Ramon
>
> While this doesn't fix any particular problem I ran into, it's best
> practice to do it this way.
>
> Note that this patch also removes the superflous extra condition check
> in the do-while loop in reap_tx_dxes(), as the loop will break
> instantly anyway in that case.
>
> Signed-off-by: Daniel Mack <[email protected]>
> ---
> drivers/net/wireless/ath/wcn36xx/dxe.c | 7 +++----
> 1 file changed, 3 insertions(+), 4 deletions(-)
>
> diff --git a/drivers/net/wireless/ath/wcn36xx/dxe.c b/drivers/net/wireless/ath/wcn36xx/dxe.c
> index 26e6c42f886a..c52f1c21ece9 100644
> --- a/drivers/net/wireless/ath/wcn36xx/dxe.c
> +++ b/drivers/net/wireless/ath/wcn36xx/dxe.c
> @@ -363,7 +363,7 @@ static void reap_tx_dxes(struct wcn36xx *wcn, struct wcn36xx_dxe_ch *ch)
> spin_lock_irqsave(&ch->lock, flags);
> ctl = ch->tail_blk_ctl;
> do {
> - if (ctl->desc->ctrl & WCN36xx_DXE_CTRL_VLD)
> + if (READ_ONCE(ctl->desc->ctrl) & WCN36xx_DXE_CTRL_VLD)
> break;
> if (ctl->skb) {
> dma_unmap_single(wcn->dev, ctl->desc->src_addr_l,
> @@ -382,8 +382,7 @@ static void reap_tx_dxes(struct wcn36xx *wcn, struct wcn36xx_dxe_ch *ch)
> ctl->skb = NULL;
> }
> ctl = ctl->next;
> - } while (ctl != ch->head_blk_ctl &&
> - !(ctl->desc->ctrl & WCN36xx_DXE_CTRL_VLD));
> + } while (ctl != ch->head_blk_ctl);
>
> ch->tail_blk_ctl = ctl;
> spin_unlock_irqrestore(&ch->lock, flags);
> @@ -525,7 +524,7 @@ static int wcn36xx_rx_handle_packets(struct wcn36xx *wcn,
> int_mask = WCN36XX_DXE_INT_CH3_MASK;
> }
>
> - while (!(dxe->ctrl & WCN36xx_DXE_CTRL_VLD)) {
> + while (!(READ_ONCE(dxe->ctrl) & WCN36xx_DXE_CTRL_VLD)) {
> skb = ctl->skb;
> dma_addr = dxe->dst_addr_l;
> ret = wcn36xx_dxe_fill_skb(wcn->dev, ctl, GFP_ATOMIC);
> --
> 2.14.3
>
On 11 April 2018 at 09:58, Daniel Mack <[email protected]> wrote:
> On Tuesday, April 10, 2018 08:17 PM, Ramon Fried wrote:
>> On 10 April 2018 at 20:35, Daniel Mack <[email protected]> wrote:
>>> When accessing shared memory to check for the stat of submitted
>>> descriptors, make sure to use READ_ONCE(). This will guarantee the
>>> compiler treats these memory locations as volatile and doesn't apply
>>> any caching.
>>
>> The structure that is tested is not shared memory. It's accessed only
>> by the apps processor.
>
> Hmm? ctl->desc is of type struct wcn36xx_dxe_desc, which is packed and
> shared with the firmware. The WCN36xx_DXE_CTRL_VLD bit in ctrl bitfield
> is set by the firmware when a frame is valid, and before asserting the
> RX interrupt. So the host CPU must treat it as volatile and expect it to
> change.
>
> Am I reading this wrong?
>
No, you're right. I missed that.
>
> Thanks,
> Daniel
Daniel Mack <[email protected]> wrote:
> When accessing shared memory to check for the stat of submitted
> descriptors, make sure to use READ_ONCE(). This will guarantee the
> compiler treats these memory locations as volatile and doesn't apply
> any caching.
>
> While this doesn't fix any particular problem I ran into, it's best
> practice to do it this way.
>
> Note that this patch also removes the superflous extra condition check
> in the do-while loop in reap_tx_dxes(), as the loop will break
> instantly anyway in that case.
>
> Signed-off-by: Daniel Mack <[email protected]>
> Signed-off-by: Kalle Valo <[email protected]>
Patch applied to ath-next branch of ath.git, thanks.
9ff6fde86fd2 wcn36xx: use READ_ONCE() to access desc->ctrl
--
https://patchwork.kernel.org/patch/10333551/
https://wireless.wiki.kernel.org/en/developers/documentation/submittingpatches