After calling napi_complete_done(), the NAPIF_STATE_SCHED bit may be cleared, and another CPU can start napi thread and access per-CQ variable, cq->work_done. If the other thread (for example, from busy_poll) sets it to a value >= budget, this thread will continue to run when it should stop, and cause memory corruption and panic.
To fix this issue, save the per-CQ work_done variable in a local variable before napi_complete_done(), so it won't be corrupted by a possible concurrent thread after napi_complete_done().
Also, add a flag bit to advertise to the NIC firmware: the NAPI work_done variable race is fixed, so the driver is able to reliably support features like busy_poll.
Cc: stable@vger.kernel.org Fixes: e1b5683ff62e ("net: mana: Move NAPI from EQ to CQ") Signed-off-by: Haiyang Zhang haiyangz@microsoft.com --- drivers/net/ethernet/microsoft/mana/gdma.h | 9 ++++++++- drivers/net/ethernet/microsoft/mana/mana_en.c | 16 +++++++++++----- 2 files changed, 19 insertions(+), 6 deletions(-)
diff --git a/drivers/net/ethernet/microsoft/mana/gdma.h b/drivers/net/ethernet/microsoft/mana/gdma.h index 4a6efe6ada08..65c24ee49efd 100644 --- a/drivers/net/ethernet/microsoft/mana/gdma.h +++ b/drivers/net/ethernet/microsoft/mana/gdma.h @@ -498,7 +498,14 @@ enum {
#define GDMA_DRV_CAP_FLAG_1_EQ_SHARING_MULTI_VPORT BIT(0)
-#define GDMA_DRV_CAP_FLAGS1 GDMA_DRV_CAP_FLAG_1_EQ_SHARING_MULTI_VPORT +/* Advertise to the NIC firmware: the NAPI work_done variable race is fixed, + * so the driver is able to reliably support features like busy_poll. + */ +#define GDMA_DRV_CAP_FLAG_1_NAPI_WKDONE_FIX BIT(2) + +#define GDMA_DRV_CAP_FLAGS1 \ + (GDMA_DRV_CAP_FLAG_1_EQ_SHARING_MULTI_VPORT | \ + GDMA_DRV_CAP_FLAG_1_NAPI_WKDONE_FIX)
#define GDMA_DRV_CAP_FLAGS2 0
diff --git a/drivers/net/ethernet/microsoft/mana/mana_en.c b/drivers/net/ethernet/microsoft/mana/mana_en.c index 9259a74eca40..27a0f3af8aab 100644 --- a/drivers/net/ethernet/microsoft/mana/mana_en.c +++ b/drivers/net/ethernet/microsoft/mana/mana_en.c @@ -1303,10 +1303,11 @@ static void mana_poll_rx_cq(struct mana_cq *cq) xdp_do_flush(); }
-static void mana_cq_handler(void *context, struct gdma_queue *gdma_queue) +static int mana_cq_handler(void *context, struct gdma_queue *gdma_queue) { struct mana_cq *cq = context; u8 arm_bit; + int w;
WARN_ON_ONCE(cq->gdma_cq != gdma_queue);
@@ -1315,26 +1316,31 @@ static void mana_cq_handler(void *context, struct gdma_queue *gdma_queue) else mana_poll_tx_cq(cq);
- if (cq->work_done < cq->budget && - napi_complete_done(&cq->napi, cq->work_done)) { + w = cq->work_done; + + if (w < cq->budget && + napi_complete_done(&cq->napi, w)) { arm_bit = SET_ARM_BIT; } else { arm_bit = 0; }
mana_gd_ring_cq(gdma_queue, arm_bit); + + return w; }
static int mana_poll(struct napi_struct *napi, int budget) { struct mana_cq *cq = container_of(napi, struct mana_cq, napi); + int w;
cq->work_done = 0; cq->budget = budget;
- mana_cq_handler(cq, cq->gdma_cq); + w = mana_cq_handler(cq, cq->gdma_cq);
- return min(cq->work_done, budget); + return min(w, budget); }
static void mana_schedule_napi(void *context, struct gdma_queue *gdma_queue)
Hello:
This patch was applied to netdev/net.git (master) by Paolo Abeni pabeni@redhat.com:
On Fri, 2 Dec 2022 11:43:10 -0800 you wrote:
After calling napi_complete_done(), the NAPIF_STATE_SCHED bit may be cleared, and another CPU can start napi thread and access per-CQ variable, cq->work_done. If the other thread (for example, from busy_poll) sets it to a value >= budget, this thread will continue to run when it should stop, and cause memory corruption and panic.
To fix this issue, save the per-CQ work_done variable in a local variable before napi_complete_done(), so it won't be corrupted by a possible concurrent thread after napi_complete_done().
[...]
Here is the summary with links: - [net] net: mana: Fix race on per-CQ variable napi work_done https://git.kernel.org/netdev/net/c/18010ff776fa
You are awesome, thank you!
linux-stable-mirror@lists.linaro.org