net: mana: Fix race on per-CQ variable napi work_done
Subsystems (5)
Haiyang Zhang
maintainerPATCH 1/1 net: mana: Fix race on per-CQ variable napi work_done
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.
--- drivers/net/ethernet/microsoft/mana/gdma.h | 9 ++++++++- drivers/net/ethernet/microsoft/mana/mana_en.c | 16 +++++++++++----- 2 files changed, 19 insertions(+), 6 deletions(-)
drivers/net/ethernet/microsoft/mana/gdma.h +8 -1
@@ -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
drivers/net/ethernet/microsoft/mana/mana_en.c +11 -5
@@ -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)
Signature · 1 line
2.25.1
This patch was applied to netdev/net.git (master) by Paolo Abeni <[email protected]>:
Hello:
This patch was applied to netdev/net.git (master)
by Paolo Abeni <[email protected]>:
Quoted 11 lines · On Fri, 2 Dec 2022 11:43:10 -0800 you wrote:
After calling
napi_complete_done(), theNAPIF_STATE_SCHEDbit may be cleared, and another CPU can start napi thread and access per-CQ variable,cq->work_done. If the other thread (for example, frombusy_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_donevariable in a local variable beforenapi_complete_done(), so it won't be corrupted by a possible concurrent thread afternapi_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!
Signature · 2 lines
Deet-doot-dot, I am a bot. https://korg.docs.kernel.org/patchwork/pwbot.html