The 'tail' pointer is also free-running count, so it needs to be masked
as 'adminq_prod_cnt' does, to become an index value of AdminQ buffer.
Fixes: 5cdad90de62c ("gve: Batch AQ commands for creating and destroying queues.")
Signed-off-by: Haiyue Wang <[email protected]>
---
drivers/net/ethernet/google/gve/gve_adminq.c | 6 ++++--
1 file changed, 4 insertions(+), 2 deletions(-)
diff --git a/drivers/net/ethernet/google/gve/gve_adminq.c b/drivers/net/ethernet/google/gve/gve_adminq.c
index 5bb56b454541..f089d33dd48e 100644
--- a/drivers/net/ethernet/google/gve/gve_adminq.c
+++ b/drivers/net/ethernet/google/gve/gve_adminq.c
@@ -322,7 +322,8 @@ static int gve_adminq_issue_cmd(struct gve_priv *priv,
tail = ioread32be(&priv->reg_bar0->adminq_event_counter);
// Check if next command will overflow the buffer.
- if (((priv->adminq_prod_cnt + 1) & priv->adminq_mask) == tail) {
+ if (((priv->adminq_prod_cnt + 1) & priv->adminq_mask) ==
+ (tail & priv->adminq_mask)) {
int err;
// Flush existing commands to make room.
@@ -332,7 +333,8 @@ static int gve_adminq_issue_cmd(struct gve_priv *priv,
// Retry.
tail = ioread32be(&priv->reg_bar0->adminq_event_counter);
- if (((priv->adminq_prod_cnt + 1) & priv->adminq_mask) == tail) {
+ if (((priv->adminq_prod_cnt + 1) & priv->adminq_mask) ==
+ (tail & priv->adminq_mask)) {
// This should never happen. We just flushed the
// command queue so there should be enough space.
return -ENOMEM;
--
2.32.0
On Wed, Jul 14, 2021 at 12:58 AM Haiyue Wang <[email protected]> wrote:
>
> The 'tail' pointer is also free-running count, so it needs to be masked
> as 'adminq_prod_cnt' does, to become an index value of AdminQ buffer.
>
> Fixes: 5cdad90de62c ("gve: Batch AQ commands for creating and destroying queues.")
> Signed-off-by: Haiyue Wang <[email protected]>
Reviewed-by: Catherine Sullivan <[email protected]>
> ---
> drivers/net/ethernet/google/gve/gve_adminq.c | 6 ++++--
> 1 file changed, 4 insertions(+), 2 deletions(-)
>
> diff --git a/drivers/net/ethernet/google/gve/gve_adminq.c b/drivers/net/ethernet/google/gve/gve_adminq.c
> index 5bb56b454541..f089d33dd48e 100644
> --- a/drivers/net/ethernet/google/gve/gve_adminq.c
> +++ b/drivers/net/ethernet/google/gve/gve_adminq.c
> @@ -322,7 +322,8 @@ static int gve_adminq_issue_cmd(struct gve_priv *priv,
> tail = ioread32be(&priv->reg_bar0->adminq_event_counter);
>
> // Check if next command will overflow the buffer.
> - if (((priv->adminq_prod_cnt + 1) & priv->adminq_mask) == tail) {
> + if (((priv->adminq_prod_cnt + 1) & priv->adminq_mask) ==
> + (tail & priv->adminq_mask)) {
> int err;
>
> // Flush existing commands to make room.
> @@ -332,7 +333,8 @@ static int gve_adminq_issue_cmd(struct gve_priv *priv,
>
> // Retry.
> tail = ioread32be(&priv->reg_bar0->adminq_event_counter);
> - if (((priv->adminq_prod_cnt + 1) & priv->adminq_mask) == tail) {
> + if (((priv->adminq_prod_cnt + 1) & priv->adminq_mask) ==
> + (tail & priv->adminq_mask)) {
> // This should never happen. We just flushed the
> // command queue so there should be enough space.
> return -ENOMEM;
> --
> 2.32.0
>
Thanks!
Hello:
This patch was applied to netdev/net-next.git (refs/heads/master):
On Wed, 14 Jul 2021 15:34:59 +0800 you wrote:
> The 'tail' pointer is also free-running count, so it needs to be masked
> as 'adminq_prod_cnt' does, to become an index value of AdminQ buffer.
>
> Fixes: 5cdad90de62c ("gve: Batch AQ commands for creating and destroying queues.")
> Signed-off-by: Haiyue Wang <[email protected]>
> ---
> drivers/net/ethernet/google/gve/gve_adminq.c | 6 ++++--
> 1 file changed, 4 insertions(+), 2 deletions(-)
Here is the summary with links:
- [v1] gve: fix the wrong AdminQ buffer overflow check
https://git.kernel.org/netdev/net-next/c/63a9192b8fa1
You are awesome, thank you!
--
Deet-doot-dot, I am a bot.
https://korg.docs.kernel.org/patchwork/pwbot.html