[PATCH 6.1 v2] gve: Fixes for napi_poll when budget is 0

[PATCH 6.1 v2] gve: Fixes for napi_poll when budget is 0

[Ziwei Xiao]

Netpoll will explicitly pass the polling call with a budget of 0 to
indicate it's clearing the Tx path only. For the gve_rx_poll and
gve_xdp_poll, they were mistakenly taking the 0 budget as the indication
to do all the work. Add check to avoid the rx path and xdp path being
called when budget is 0. And also avoid napi_complete_done being called
when budget is 0 for netpoll.

The original fix was merged here:
https://lore.kernel.org/r/20231114004144.2022268-1-ziweixiao@google.com
Resend it since the original one was not cleanly applied to 6.1 kernel.

commit 278a370c1766 ("gve: Fixes for napi_poll when budget is 0")

Fixes: f5cedc84a30d ("gve: Add transmit and receive support")
Signed-off-by: Ziwei Xiao <ziweixiao@google.com>
Reviewed-by: Praveen Kaligineedi <pkaligineedi@google.com>
Signed-off-by: Praveen Kaligineedi <pkaligineedi@google.com>
---
 Changes in v2:
 * Add the original git commit id
---
 drivers/net/ethernet/google/gve/gve_main.c | 7 +++++++
 drivers/net/ethernet/google/gve/gve_rx.c   | 4 ----
 drivers/net/ethernet/google/gve/gve_tx.c   | 4 ----
 3 files changed, 7 insertions(+), 8 deletions(-)

diff --git a/drivers/net/ethernet/google/gve/gve_main.c b/drivers/net/ethernet/google/gve/gve_main.c
index d3f6ad586ba1..8771ccfc69b4 100644
--- a/drivers/net/ethernet/google/gve/gve_main.c
+++ b/drivers/net/ethernet/google/gve/gve_main.c
@@ -202,6 +202,10 @@ static int gve_napi_poll(struct napi_struct *napi, int budget)
 
 	if (block->tx)
 		reschedule |= gve_tx_poll(block, budget);
+
+	if (!budget)
+		return 0;
+
 	if (block->rx) {
 		work_done = gve_rx_poll(block, budget);
 		reschedule |= work_done == budget;
@@ -242,6 +246,9 @@ static int gve_napi_poll_dqo(struct napi_struct *napi, int budget)
 	if (block->tx)
 		reschedule |= gve_tx_poll_dqo(block, /*do_clean=*/true);
 
+	if (!budget)
+		return 0;
+
 	if (block->rx) {
 		work_done = gve_rx_poll_dqo(block, budget);
 		reschedule |= work_done == budget;
diff --git a/drivers/net/ethernet/google/gve/gve_rx.c b/drivers/net/ethernet/google/gve/gve_rx.c
index 021bbf308d68..639eb6848c7d 100644
--- a/drivers/net/ethernet/google/gve/gve_rx.c
+++ b/drivers/net/ethernet/google/gve/gve_rx.c
@@ -778,10 +778,6 @@ int gve_rx_poll(struct gve_notify_block *block, int budget)
 
 	feat = block->napi.dev->features;
 
-	/* If budget is 0, do all the work */
-	if (budget == 0)
-		budget = INT_MAX;
-
 	if (budget > 0)
 		work_done = gve_clean_rx_done(rx, budget, feat);
 
diff --git a/drivers/net/ethernet/google/gve/gve_tx.c b/drivers/net/ethernet/google/gve/gve_tx.c
index 5e11b8236754..bf1ac0d1dc6f 100644
--- a/drivers/net/ethernet/google/gve/gve_tx.c
+++ b/drivers/net/ethernet/google/gve/gve_tx.c
@@ -725,10 +725,6 @@ bool gve_tx_poll(struct gve_notify_block *block, int budget)
 	u32 nic_done;
 	u32 to_do;
 
-	/* If budget is 0, do all the work */
-	if (budget == 0)
-		budget = INT_MAX;
-
 	/* In TX path, it may try to clean completed pkts in order to xmit,
 	 * to avoid cleaning conflict, use spin_lock(), it yields better
 	 * concurrency between xmit/clean than netif's lock.
-- 
2.47.0.338.g60cca15819-goog

Re: [PATCH 6.1 v2] gve: Fixes for napi_poll when budget is 0

[Sasha Levin]

[ Sasha's backport helper bot ]

Hi,

Found matching upstream commit: 278a370c1766060d2144d6cf0b06c101e1043b6d


Status in newer kernel trees:
6.12.y | Present (exact SHA1)
6.6.y | Present (different SHA1: ff33be9cecee)
6.1.y | Not found

Note: The patch differs from the upstream commit:
---
1:  278a370c17660 ! 1:  68463e6a71027 gve: Fixes for napi_poll when budget is 0
    @@ Metadata
      ## Commit message ##
         gve: Fixes for napi_poll when budget is 0
     
    -    Netpoll will explicilty pass the polling call with a budget of 0 to
    +    Netpoll will explicitly pass the polling call with a budget of 0 to
         indicate it's clearing the Tx path only. For the gve_rx_poll and
         gve_xdp_poll, they were mistakenly taking the 0 budget as the indication
         to do all the work. Add check to avoid the rx path and xdp path being
         called when budget is 0. And also avoid napi_complete_done being called
         when budget is 0 for netpoll.
     
    +    The original fix was merged here:
    +    https://lore.kernel.org/r/20231114004144.2022268-1-ziweixiao@google.com
    +    Resend it since the original one was not cleanly applied to 6.1 kernel.
    +
    +    commit 278a370c1766 ("gve: Fixes for napi_poll when budget is 0")
    +
         Fixes: f5cedc84a30d ("gve: Add transmit and receive support")
         Signed-off-by: Ziwei Xiao <ziweixiao@google.com>
    -    Link: https://lore.kernel.org/r/20231114004144.2022268-1-ziweixiao@google.com
    -    Signed-off-by: Jakub Kicinski <kuba@kernel.org>
    +    Reviewed-by: Praveen Kaligineedi <pkaligineedi@google.com>
    +    Signed-off-by: Praveen Kaligineedi <pkaligineedi@google.com>
     
      ## drivers/net/ethernet/google/gve/gve_main.c ##
     @@ drivers/net/ethernet/google/gve/gve_main.c: static int gve_napi_poll(struct napi_struct *napi, int budget)
    - 	if (block->tx) {
    - 		if (block->tx->q_num < priv->tx_cfg.num_queues)
    - 			reschedule |= gve_tx_poll(block, budget);
    --		else
    -+		else if (budget)
    - 			reschedule |= gve_xdp_poll(block, budget);
    - 	}
      
    + 	if (block->tx)
    + 		reschedule |= gve_tx_poll(block, budget);
    ++
     +	if (!budget)
     +		return 0;
     +
    @@ drivers/net/ethernet/google/gve/gve_rx.c: int gve_rx_poll(struct gve_notify_bloc
      
     
      ## drivers/net/ethernet/google/gve/gve_tx.c ##
    -@@ drivers/net/ethernet/google/gve/gve_tx.c: bool gve_xdp_poll(struct gve_notify_block *block, int budget)
    - 	bool repoll;
    +@@ drivers/net/ethernet/google/gve/gve_tx.c: bool gve_tx_poll(struct gve_notify_block *block, int budget)
    + 	u32 nic_done;
      	u32 to_do;
      
     -	/* If budget is 0, do all the work */
     -	if (budget == 0)
     -		budget = INT_MAX;
     -
    - 	/* Find out how much work there is to be done */
    - 	nic_done = gve_tx_load_event_counter(priv, tx);
    - 	to_do = min_t(u32, (nic_done - tx->done), budget);
    + 	/* In TX path, it may try to clean completed pkts in order to xmit,
    + 	 * to avoid cleaning conflict, use spin_lock(), it yields better
    + 	 * concurrency between xmit/clean than netif's lock.
---

Results of testing on various branches:

| Branch                    | Patch Apply | Build Test |
|---------------------------|-------------|------------|
| stable/linux-6.1.y        |  Success    |  Success   |