Skip to contentSkip to conversation

ixgbe: xsk: resolve the negative overflow of budget in ixgbe_xmit_zc

PATCH In review Jason Xing 6 participants 12 messages Jul 26, 2025 – Aug 19, 2025 1 message not in the archive

Subsystems (3)
lore
Share & tools
lore

Jason Xing

author
#0 lore raw

PATCH v2 1/1 ixgbe: xsk: resolve the negative overflow of budget in ixgbe_xmit_zc

From: Jason Xing <[email protected]>

Resolve the budget negative overflow which leads to returning true in ixgbe_xmit_zc even when the budget of descs are thoroughly consumed.

Before this patch, when the budget is decreased to zero and finishes sending the last allowed desc in ixgbe_xmit_zc, it will always turn back and enter into the while() statement to see if it should keep processing packets, but in the meantime it unexpectedly decreases the value again to 'unsigned int (0--)', namely, UINT_MAX. Finally, the ixgbe_xmit_zc returns true, showing 'we complete cleaning the budget'. That also means 'clean_complete = true' in ixgbe_poll.

The true theory behind this is if that budget number of descs are consumed, it implies that we might have more descs to be done. So we should return false in ixgbe_xmit_zc to tell napi poll to find another chance to start polling to handle the rest of descs. On the contrary, returning true here means job done and we know we finish all the possible descs this time and we don't intend to start a new napi poll.

It is apparently against our expectations. Please also see how ixgbe_clean_tx_irq() handles the problem: it uses do..while() statement to make sure the budget can be decreased to zero at most and the negative overflow never happens.

The patch adds 'likely' because we rarely would not hit the loop codition since the standard budget is 256.

Fixes 8221c5eba8c1 ("ixgbe: add AF_XDP zero-copy Tx support")

---

1. use 'negative overflow' instead of 'underflow' (Willem)
2. add reviewed-by tag (Larysa)
3. target iwl-net branch (Larysa)
4. add the reason why the patch adds likely() (Larysa)

---
 drivers/net/ethernet/intel/ixgbe/ixgbe_xsk.c | 4 +++-
 1 file changed, 3 insertions(+), 1 deletion(-)
drivers/net/ethernet/intel/ixgbe/ixgbe_xsk.c +3 -1
index ac58964b2f08..7b941505a9d0 100644
@@ -398,7 +398,7 @@ static bool ixgbe_xmit_zc(struct ixgbe_ring *xdp_ring, unsigned int budget)
 	dma_addr_t dma;
 	u32 cmd_type;
 
-	while (budget-- > 0) {
+	while (likely(budget)) {
 		if (unlikely(!ixgbe_desc_unused(xdp_ring))) {
 			work_done = false;
 			break;
@@ -433,6 +433,8 @@ static bool ixgbe_xmit_zc(struct ixgbe_ring *xdp_ring, unsigned int budget)
 		xdp_ring->next_to_use++;
 		if (xdp_ring->next_to_use == xdp_ring->count)
 			xdp_ring->next_to_use = 0;
+
+		budget--;
 	}
 
 	if (tx_desc) {
Signature · 1 line
2.41.3
Reviewed-by on v2 1/1
#1 lore raw

Dear Jason,

Thank you for the improved version.

Am 26.07.25 um 09:03 schrieb Jason Xing:

Quoted 55 lines

From: Jason Xing <[email protected]>

Resolve the budget negative overflow which leads to returning true in ixgbe_xmit_zc even when the budget of descs are thoroughly consumed.

Before this patch, when the budget is decreased to zero and finishes sending the last allowed desc in ixgbe_xmit_zc, it will always turn back and enter into the while() statement to see if it should keep processing packets, but in the meantime it unexpectedly decreases the value again to 'unsigned int (0--)', namely, UINT_MAX. Finally, the ixgbe_xmit_zc returns true, showing 'we complete cleaning the budget'. That also means 'clean_complete = true' in ixgbe_poll.

The true theory behind this is if that budget number of descs are consumed, it implies that we might have more descs to be done. So we should return false in ixgbe_xmit_zc to tell napi poll to find another chance to start polling to handle the rest of descs. On the contrary, returning true here means job done and we know we finish all the possible descs this time and we don't intend to start a new napi poll.

It is apparently against our expectations. Please also see how ixgbe_clean_tx_irq() handles the problem: it uses do..while() statement to make sure the budget can be decreased to zero at most and the negative overflow never happens.

The patch adds 'likely' because we rarely would not hit the loop codition since the standard budget is 256.

Fixes 8221c5eba8c1 ("ixgbe: add AF_XDP zero-copy Tx support")

---

1. use 'negative overflow' instead of 'underflow' (Willem)
2. add reviewed-by tag (Larysa)
3. target iwl-net branch (Larysa)
4. add the reason why the patch adds likely() (Larysa)

---
  drivers/net/ethernet/intel/ixgbe/ixgbe_xsk.c | 4 +++-
  1 file changed, 3 insertions(+), 1 deletion(-)
drivers/net/ethernet/intel/ixgbe/ixgbe_xsk.c +1 -1
index ac58964b2f08..7b941505a9d0 100644
@@ -398,7 +398,7 @@ static bool ixgbe_xmit_zc(struct ixgbe_ring *xdp_ring, unsigned int budget)
  	dma_addr_t dma;
  	u32 cmd_type;
 
-	while (budget-- > 0) {
+	while (likely(budget)) {
  		if (unlikely(!ixgbe_desc_unused(xdp_ring))) {
  			work_done = false;
  			break;

drivers/net/ethernet/intel/ixgbe/ixgbe_xsk.c static bool ixgbe_xmit_zc(struct ixgbe_ring *xdp_ring, unsigned int budget) · line 440

Quoted 8 lines
drivers/net/ethernet/intel/ixgbe/ixgbe_xsk.c +2 -0
@@ -433,6 +433,8 @@ static bool ixgbe_xmit_zc(struct ixgbe_ring *xdp_ring, unsigned int budget)
  		xdp_ring->next_to_use++;
  		if (xdp_ring->next_to_use == xdp_ring->count)
  			xdp_ring->next_to_use = 0;
+
+		budget--;
  	}
 
  	if (tx_desc) {

Is this just the smallest fix, and the rewrite to the more idiomatic for loop going to be done in a follow-up?

Kind regards,

Paul

author
#2 lore raw

Hi Paul,

Quoted 69 lines · Paul Menzel

Dear Jason,

Thank you for the improved version.

Am 26.07.25 um 09:03 schrieb Jason Xing:

Quoted 62 lines

From: Jason Xing <[email protected]>

Resolve the budget negative overflow which leads to returning true in ixgbe_xmit_zc even when the budget of descs are thoroughly consumed.

Before this patch, when the budget is decreased to zero and finishes sending the last allowed desc in ixgbe_xmit_zc, it will always turn back and enter into the while() statement to see if it should keep processing packets, but in the meantime it unexpectedly decreases the value again to 'unsigned int (0--)', namely, UINT_MAX. Finally, the ixgbe_xmit_zc returns true, showing 'we complete cleaning the budget'. That also means 'clean_complete = true' in ixgbe_poll.

The true theory behind this is if that budget number of descs are consumed, it implies that we might have more descs to be done. So we should return false in ixgbe_xmit_zc to tell napi poll to find another chance to start polling to handle the rest of descs. On the contrary, returning true here means job done and we know we finish all the possible descs this time and we don't intend to start a new napi poll.

It is apparently against our expectations. Please also see how ixgbe_clean_tx_irq() handles the problem: it uses do..while() statement to make sure the budget can be decreased to zero at most and the negative overflow never happens.

The patch adds 'likely' because we rarely would not hit the loop codition since the standard budget is 256.

Fixes 8221c5eba8c1 ("ixgbe: add AF_XDP zero-copy Tx support")

---

1. use 'negative overflow' instead of 'underflow' (Willem)
2. add reviewed-by tag (Larysa)
3. target iwl-net branch (Larysa)
4. add the reason why the patch adds likely() (Larysa)

---
  drivers/net/ethernet/intel/ixgbe/ixgbe_xsk.c | 4 +++-
  1 file changed, 3 insertions(+), 1 deletion(-)
drivers/net/ethernet/intel/ixgbe/ixgbe_xsk.c +3 -1
index ac58964b2f08..7b941505a9d0 100644
@@ -398,7 +398,7 @@ static bool ixgbe_xmit_zc(struct ixgbe_ring *xdp_ring, unsigned int budget)
      dma_addr_t dma;
      u32 cmd_type;
 
-     while (budget-- > 0) {
+     while (likely(budget)) {
              if (unlikely(!ixgbe_desc_unused(xdp_ring))) {
                      work_done = false;
                      break;
@@ -433,6 +433,8 @@ static bool ixgbe_xmit_zc(struct ixgbe_ring *xdp_ring, unsigned int budget)
              xdp_ring->next_to_use++;
              if (xdp_ring->next_to_use == xdp_ring->count)
                      xdp_ring->next_to_use = 0;
+
+             budget--;
      }
 
Quoted 6 lines

drivers/net/ethernet/intel/ixgbe/ixgbe_xsk.c static bool ixgbe_xmit_zc(struct ixgbe_ring *xdp_ring, unsigned int budget) · line 440

Quoted 1 line
drivers/net/ethernet/intel/ixgbe/ixgbe_xsk.c +0 -0
      if (tx_desc) {

Is this just the smallest fix, and the rewrite to the more idiomatic for loop going to be done in a follow-up?

Thanks for the review. But I'm not that sure if it's worth a follow-up patch. Or if anyone else also expects to see a 'for loop' version, I can send a V3 patch then. I have no strong opinion either way.

Thanks,
Jason

reviewer
#3 lore raw
Quoted 76 lines · Jason Xing

Hi Paul,

Quoted 73 lines · Paul Menzel

Dear Jason,

Thank you for the improved version.

Am 26.07.25 um 09:03 schrieb Jason Xing:

Quoted 55 lines

From: Jason Xing <[email protected]>

Resolve the budget negative overflow which leads to returning true in ixgbe_xmit_zc even when the budget of descs are thoroughly consumed.

Before this patch, when the budget is decreased to zero and finishes sending the last allowed desc in ixgbe_xmit_zc, it will always turn back and enter into the while() statement to see if it should keep processing packets, but in the meantime it unexpectedly decreases the value again to 'unsigned int (0--)', namely, UINT_MAX. Finally, the ixgbe_xmit_zc returns true, showing 'we complete cleaning the budget'. That also means 'clean_complete = true' in ixgbe_poll.

The true theory behind this is if that budget number of descs are consumed, it implies that we might have more descs to be done. So we should return false in ixgbe_xmit_zc to tell napi poll to find another chance to start polling to handle the rest of descs. On the contrary, returning true here means job done and we know we finish all the possible descs this time and we don't intend to start a new napi poll.

It is apparently against our expectations. Please also see how ixgbe_clean_tx_irq() handles the problem: it uses do..while() statement to make sure the budget can be decreased to zero at most and the negative overflow never happens.

The patch adds 'likely' because we rarely would not hit the loop codition since the standard budget is 256.

Fixes 8221c5eba8c1 ("ixgbe: add AF_XDP zero-copy Tx support")

---

1. use 'negative overflow' instead of 'underflow' (Willem)
2. add reviewed-by tag (Larysa)
3. target iwl-net branch (Larysa)
4. add the reason why the patch adds likely() (Larysa)

---
  drivers/net/ethernet/intel/ixgbe/ixgbe_xsk.c | 4 +++-
  1 file changed, 3 insertions(+), 1 deletion(-)
drivers/net/ethernet/intel/ixgbe/ixgbe_xsk.c +1 -1
index ac58964b2f08..7b941505a9d0 100644
@@ -398,7 +398,7 @@ static bool ixgbe_xmit_zc(struct ixgbe_ring *xdp_ring, unsigned int budget)
      dma_addr_t dma;
      u32 cmd_type;
 
-     while (budget-- > 0) {
+     while (likely(budget)) {
              if (unlikely(!ixgbe_desc_unused(xdp_ring))) {
                      work_done = false;
                      break;

drivers/net/ethernet/intel/ixgbe/ixgbe_xsk.c static bool ixgbe_xmit_zc(struct ixgbe_ring *xdp_ring, unsigned int budget) · line 440

Quoted 8 lines
drivers/net/ethernet/intel/ixgbe/ixgbe_xsk.c +2 -0
@@ -433,6 +433,8 @@ static bool ixgbe_xmit_zc(struct ixgbe_ring *xdp_ring, unsigned int budget)
              xdp_ring->next_to_use++;
              if (xdp_ring->next_to_use == xdp_ring->count)
                      xdp_ring->next_to_use = 0;
+
+             budget--;
      }
 
      if (tx_desc) {
Quoted 6 lines
Quoted 2 lines

Is this just the smallest fix, and the rewrite to the more idiomatic for loop going to be done in a follow-up?

Thanks for the review. But I'm not that sure if it's worth a follow-up patch. Or if anyone else also expects to see a 'for loop' version, I can send a V3 patch then. I have no strong opinion either way.

I think we have iterated over this enough (pun intended). If this patch is correct then lets stick with this approach.

author
#4 lore raw

Hi Simon,

Quoted 78 lines · Simon Horman
Quoted 76 lines · Jason Xing

Hi Paul,

Quoted 73 lines · Paul Menzel

Dear Jason,

Thank you for the improved version.

Am 26.07.25 um 09:03 schrieb Jason Xing:

Quoted 55 lines

From: Jason Xing <[email protected]>

Resolve the budget negative overflow which leads to returning true in ixgbe_xmit_zc even when the budget of descs are thoroughly consumed.

Before this patch, when the budget is decreased to zero and finishes sending the last allowed desc in ixgbe_xmit_zc, it will always turn back and enter into the while() statement to see if it should keep processing packets, but in the meantime it unexpectedly decreases the value again to 'unsigned int (0--)', namely, UINT_MAX. Finally, the ixgbe_xmit_zc returns true, showing 'we complete cleaning the budget'. That also means 'clean_complete = true' in ixgbe_poll.

The true theory behind this is if that budget number of descs are consumed, it implies that we might have more descs to be done. So we should return false in ixgbe_xmit_zc to tell napi poll to find another chance to start polling to handle the rest of descs. On the contrary, returning true here means job done and we know we finish all the possible descs this time and we don't intend to start a new napi poll.

It is apparently against our expectations. Please also see how ixgbe_clean_tx_irq() handles the problem: it uses do..while() statement to make sure the budget can be decreased to zero at most and the negative overflow never happens.

The patch adds 'likely' because we rarely would not hit the loop codition since the standard budget is 256.

Fixes 8221c5eba8c1 ("ixgbe: add AF_XDP zero-copy Tx support")

---

1. use 'negative overflow' instead of 'underflow' (Willem)
2. add reviewed-by tag (Larysa)
3. target iwl-net branch (Larysa)
4. add the reason why the patch adds likely() (Larysa)

---
  drivers/net/ethernet/intel/ixgbe/ixgbe_xsk.c | 4 +++-
  1 file changed, 3 insertions(+), 1 deletion(-)
drivers/net/ethernet/intel/ixgbe/ixgbe_xsk.c +1 -1
index ac58964b2f08..7b941505a9d0 100644
@@ -398,7 +398,7 @@ static bool ixgbe_xmit_zc(struct ixgbe_ring *xdp_ring, unsigned int budget)
      dma_addr_t dma;
      u32 cmd_type;
 
-     while (budget-- > 0) {
+     while (likely(budget)) {
              if (unlikely(!ixgbe_desc_unused(xdp_ring))) {
                      work_done = false;
                      break;

drivers/net/ethernet/intel/ixgbe/ixgbe_xsk.c static bool ixgbe_xmit_zc(struct ixgbe_ring *xdp_ring, unsigned int budget) · line 440

Quoted 8 lines
drivers/net/ethernet/intel/ixgbe/ixgbe_xsk.c +2 -0
@@ -433,6 +433,8 @@ static bool ixgbe_xmit_zc(struct ixgbe_ring *xdp_ring, unsigned int budget)
              xdp_ring->next_to_use++;
              if (xdp_ring->next_to_use == xdp_ring->count)
                      xdp_ring->next_to_use = 0;
+
+             budget--;
      }
 
      if (tx_desc) {
Quoted 8 lines
Quoted 6 lines
Quoted 2 lines

Is this just the smallest fix, and the rewrite to the more idiomatic for loop going to be done in a follow-up?

Thanks for the review. But I'm not that sure if it's worth a follow-up patch. Or if anyone else also expects to see a 'for loop' version, I can send a V3 patch then. I have no strong opinion either way.

I think we have iterated over this enough (pun intended).

Hhh, interesting.

Quoted 1 line

If this patch is correct then lets stick with this approach.

No problem. Tomorrow I will send the 'for loop' version :)

Thanks,
Jason

reviewer
#5 lore raw
Quoted 88 lines · Jason Xing

Hi Simon,

Quoted 85 lines · Simon Horman
Quoted 82 lines · Jason Xing

Hi Paul,

Quoted 69 lines · Paul Menzel

Dear Jason,

Thank you for the improved version.

Am 26.07.25 um 09:03 schrieb Jason Xing:

Quoted 62 lines

From: Jason Xing <[email protected]>

Resolve the budget negative overflow which leads to returning true in ixgbe_xmit_zc even when the budget of descs are thoroughly consumed.

Before this patch, when the budget is decreased to zero and finishes sending the last allowed desc in ixgbe_xmit_zc, it will always turn back and enter into the while() statement to see if it should keep processing packets, but in the meantime it unexpectedly decreases the value again to 'unsigned int (0--)', namely, UINT_MAX. Finally, the ixgbe_xmit_zc returns true, showing 'we complete cleaning the budget'. That also means 'clean_complete = true' in ixgbe_poll.

The true theory behind this is if that budget number of descs are consumed, it implies that we might have more descs to be done. So we should return false in ixgbe_xmit_zc to tell napi poll to find another chance to start polling to handle the rest of descs. On the contrary, returning true here means job done and we know we finish all the possible descs this time and we don't intend to start a new napi poll.

It is apparently against our expectations. Please also see how ixgbe_clean_tx_irq() handles the problem: it uses do..while() statement to make sure the budget can be decreased to zero at most and the negative overflow never happens.

The patch adds 'likely' because we rarely would not hit the loop codition since the standard budget is 256.

Fixes 8221c5eba8c1 ("ixgbe: add AF_XDP zero-copy Tx support")

---

1. use 'negative overflow' instead of 'underflow' (Willem)
2. add reviewed-by tag (Larysa)
3. target iwl-net branch (Larysa)
4. add the reason why the patch adds likely() (Larysa)

---
  drivers/net/ethernet/intel/ixgbe/ixgbe_xsk.c | 4 +++-
  1 file changed, 3 insertions(+), 1 deletion(-)
drivers/net/ethernet/intel/ixgbe/ixgbe_xsk.c +3 -1
index ac58964b2f08..7b941505a9d0 100644
@@ -398,7 +398,7 @@ static bool ixgbe_xmit_zc(struct ixgbe_ring *xdp_ring, unsigned int budget)
      dma_addr_t dma;
      u32 cmd_type;
 
-     while (budget-- > 0) {
+     while (likely(budget)) {
              if (unlikely(!ixgbe_desc_unused(xdp_ring))) {
                      work_done = false;
                      break;
@@ -433,6 +433,8 @@ static bool ixgbe_xmit_zc(struct ixgbe_ring *xdp_ring, unsigned int budget)
              xdp_ring->next_to_use++;
              if (xdp_ring->next_to_use == xdp_ring->count)
                      xdp_ring->next_to_use = 0;
+
+             budget--;
      }
 
Quoted 6 lines

drivers/net/ethernet/intel/ixgbe/ixgbe_xsk.c static bool ixgbe_xmit_zc(struct ixgbe_ring *xdp_ring, unsigned int budget) · line 440

Quoted 1 line
drivers/net/ethernet/intel/ixgbe/ixgbe_xsk.c +0 -0
      if (tx_desc) {

Is this just the smallest fix, and the rewrite to the more idiomatic for loop going to be done in a follow-up?

Thanks for the review. But I'm not that sure if it's worth a follow-up patch. Or if anyone else also expects to see a 'for loop' version, I can send a V3 patch then. I have no strong opinion either way.

Quoted 7 lines
Quoted 1 line

I think we have iterated over this enough (pun intended).

Hhh, interesting.

Quoted 1 line

If this patch is correct then lets stick with this approach.

No problem. Tomorrow I will send the 'for loop' version :)

I meant, I think the while loop is fine.
But anything that is correct is fine by me :)

author
#6 lore raw
Quoted 92 lines · Simon Horman
Quoted 90 lines · Jason Xing

Hi Simon,

Quoted 86 lines · Simon Horman
Quoted 76 lines · Jason Xing

Hi Paul,

Quoted 73 lines · Paul Menzel

Dear Jason,

Thank you for the improved version.

Am 26.07.25 um 09:03 schrieb Jason Xing:

Quoted 55 lines

From: Jason Xing <[email protected]>

Resolve the budget negative overflow which leads to returning true in ixgbe_xmit_zc even when the budget of descs are thoroughly consumed.

Before this patch, when the budget is decreased to zero and finishes sending the last allowed desc in ixgbe_xmit_zc, it will always turn back and enter into the while() statement to see if it should keep processing packets, but in the meantime it unexpectedly decreases the value again to 'unsigned int (0--)', namely, UINT_MAX. Finally, the ixgbe_xmit_zc returns true, showing 'we complete cleaning the budget'. That also means 'clean_complete = true' in ixgbe_poll.

The true theory behind this is if that budget number of descs are consumed, it implies that we might have more descs to be done. So we should return false in ixgbe_xmit_zc to tell napi poll to find another chance to start polling to handle the rest of descs. On the contrary, returning true here means job done and we know we finish all the possible descs this time and we don't intend to start a new napi poll.

It is apparently against our expectations. Please also see how ixgbe_clean_tx_irq() handles the problem: it uses do..while() statement to make sure the budget can be decreased to zero at most and the negative overflow never happens.

The patch adds 'likely' because we rarely would not hit the loop codition since the standard budget is 256.

Fixes 8221c5eba8c1 ("ixgbe: add AF_XDP zero-copy Tx support")

---

1. use 'negative overflow' instead of 'underflow' (Willem)
2. add reviewed-by tag (Larysa)
3. target iwl-net branch (Larysa)
4. add the reason why the patch adds likely() (Larysa)

---
  drivers/net/ethernet/intel/ixgbe/ixgbe_xsk.c | 4 +++-
  1 file changed, 3 insertions(+), 1 deletion(-)
drivers/net/ethernet/intel/ixgbe/ixgbe_xsk.c +1 -1
index ac58964b2f08..7b941505a9d0 100644
@@ -398,7 +398,7 @@ static bool ixgbe_xmit_zc(struct ixgbe_ring *xdp_ring, unsigned int budget)
      dma_addr_t dma;
      u32 cmd_type;
 
-     while (budget-- > 0) {
+     while (likely(budget)) {
              if (unlikely(!ixgbe_desc_unused(xdp_ring))) {
                      work_done = false;
                      break;

drivers/net/ethernet/intel/ixgbe/ixgbe_xsk.c static bool ixgbe_xmit_zc(struct ixgbe_ring *xdp_ring, unsigned int budget) · line 440

Quoted 8 lines
drivers/net/ethernet/intel/ixgbe/ixgbe_xsk.c +2 -0
@@ -433,6 +433,8 @@ static bool ixgbe_xmit_zc(struct ixgbe_ring *xdp_ring, unsigned int budget)
              xdp_ring->next_to_use++;
              if (xdp_ring->next_to_use == xdp_ring->count)
                      xdp_ring->next_to_use = 0;
+
+             budget--;
      }
 
      if (tx_desc) {
Quoted 6 lines
Quoted 2 lines

Is this just the smallest fix, and the rewrite to the more idiomatic for loop going to be done in a follow-up?

Thanks for the review. But I'm not that sure if it's worth a follow-up patch. Or if anyone else also expects to see a 'for loop' version, I can send a V3 patch then. I have no strong opinion either way.

I think we have iterated over this enough (pun intended).

Quoted 8 lines
Quoted 5 lines

Hhh, interesting.

Quoted 1 line

If this patch is correct then lets stick with this approach.

No problem. Tomorrow I will send the 'for loop' version :)

I meant, I think the while loop is fine.
But anything that is correct is fine by me :)

Okay, great, then no more updates will be made unless other reviewers have opinions :)

Thanks,
Jason

Reviewed-by on v2 1/1
#7 lore raw
In reply to Jason Xing
Quoted 42 lines

-----Original Message-----
From: Intel-wired-lan <[email protected]> On Behalf Of Jason Xing
Sent: Saturday, July 26, 2025 9:04 AM
To: Nguyen, Anthony L <[email protected]>; Kitszel, Przemyslaw <[email protected]>; Zaremba, Larysa <[email protected]>; [email protected]; [email protected]; [email protected]; [email protected]; [email protected]; [email protected]; Fijalkowski, Maciej <[email protected]>
Cc: [email protected]; [email protected]; Jason Xing <[email protected]>
Subject: [Intel-wired-lan] [PATCH v2 iwl-net] ixgbe: xsk: resolve the negative overflow of budget in ixgbe_xmit_zc

From: Jason Xing <[email protected]>

Resolve the budget negative overflow which leads to returning true in ixgbe_xmit_zc even when the budget of descs are thoroughly consumed.

Before this patch, when the budget is decreased to zero and finishes sending the last allowed desc in ixgbe_xmit_zc, it will always turn back and enter into the while() statement to see if it should keep processing packets, but in the meantime it unexpectedly decreases the value again to 'unsigned int (0--)', namely, UINT_MAX. Finally, the ixgbe_xmit_zc returns true, showing 'we complete cleaning the budget'. That also means 'clean_complete = true' in ixgbe_poll.

The true theory behind this is if that budget number of descs are consumed, it implies that we might have more descs to be done. So we should return false in ixgbe_xmit_zc to tell napi poll to find another chance to start polling to handle the rest of descs. On the contrary, returning true here means job done and we know we finish all the possible descs this time and we don't intend to start a new napi poll.

It is apparently against our expectations. Please also see how ixgbe_clean_tx_irq() handles the problem: it uses do..while() statement to make sure the budget can be decreased to zero at most and the negative overflow never happens.

Quoted 6 lines

The patch adds 'likely' because we rarely would not hit the loop codition since the standard budget is 256.

Fixes 8221c5eba8c1 ("ixgbe: add AF_XDP zero-copy Tx support")
Quoted 37 lines

---

[email protected]/
1. use 'negative overflow' instead of 'underflow' (Willem) 2. add reviewed-by tag (Larysa) 3. target iwl-net branch (Larysa) 4. add the reason why the patch adds likely() (Larysa)

---
 drivers/net/ethernet/intel/ixgbe/ixgbe_xsk.c | 4 +++-
 1 file changed, 3 insertions(+), 1 deletion(-)
drivers/net/ethernet/intel/ixgbe/ixgbe_xsk.c +1 -1
index ac58964b2f08..7b941505a9d0 100644
@@ -398,7 +398,7 @@ static bool ixgbe_xmit_zc(struct ixgbe_ring
 *xdp_ring, unsigned int budget)
 	dma_addr_t dma;
 	u32 cmd_type;
 
-	while (budget-- > 0) {
+	while (likely(budget)) {
 		if (unlikely(!ixgbe_desc_unused(xdp_ring))) {
 			work_done = false;
drivers/net/ethernet/intel/ixgbe/ixgbe_xsk.c +2 -0
 			break;
@@ -433,6 +433,8 @@ static bool ixgbe_xmit_zc(struct ixgbe_ring
 *xdp_ring, unsigned int budget)
 		xdp_ring->next_to_use++;
 		if (xdp_ring->next_to_use == xdp_ring->count)
 			xdp_ring->next_to_use = 0;
+
+		budget--;
 	}
 

if (tx_desc) {
--
2.41.3

author
#8 lore raw
In reply to Jason Xing
Quoted 26 lines · Jason Xing

From: Jason Xing <[email protected]>

Resolve the budget negative overflow which leads to returning true in ixgbe_xmit_zc even when the budget of descs are thoroughly consumed.

Before this patch, when the budget is decreased to zero and finishes sending the last allowed desc in ixgbe_xmit_zc, it will always turn back and enter into the while() statement to see if it should keep processing packets, but in the meantime it unexpectedly decreases the value again to 'unsigned int (0--)', namely, UINT_MAX. Finally, the ixgbe_xmit_zc returns true, showing 'we complete cleaning the budget'. That also means 'clean_complete = true' in ixgbe_poll.

The true theory behind this is if that budget number of descs are consumed, it implies that we might have more descs to be done. So we should return false in ixgbe_xmit_zc to tell napi poll to find another chance to start polling to handle the rest of descs. On the contrary, returning true here means job done and we know we finish all the possible descs this time and we don't intend to start a new napi poll.

It is apparently against our expectations. Please also see how ixgbe_clean_tx_irq() handles the problem: it uses do..while() statement to make sure the budget can be decreased to zero at most and the negative overflow never happens.

Quoted 6 lines

The patch adds 'likely' because we rarely would not hit the loop codition since the standard budget is 256.

Fixes 8221c5eba8c1 ("ixgbe: add AF_XDP zero-copy Tx support")

Hi Tony,

Any update here? Thanks! I'm asking because I'm ready to send an afxdp patch series based on the patch :)

Thanks,
jason

maintainer
#9 lore raw
Quoted 30 lines · Jason Xing
Quoted 29 lines · Jason Xing

From: Jason Xing <[email protected]>

Resolve the budget negative overflow which leads to returning true in ixgbe_xmit_zc even when the budget of descs are thoroughly consumed.

Before this patch, when the budget is decreased to zero and finishes sending the last allowed desc in ixgbe_xmit_zc, it will always turn back and enter into the while() statement to see if it should keep processing packets, but in the meantime it unexpectedly decreases the value again to 'unsigned int (0--)', namely, UINT_MAX. Finally, the ixgbe_xmit_zc returns true, showing 'we complete cleaning the budget'. That also means 'clean_complete = true' in ixgbe_poll.

The true theory behind this is if that budget number of descs are consumed, it implies that we might have more descs to be done. So we should return false in ixgbe_xmit_zc to tell napi poll to find another chance to start polling to handle the rest of descs. On the contrary, returning true here means job done and we know we finish all the possible descs this time and we don't intend to start a new napi poll.

It is apparently against our expectations. Please also see how ixgbe_clean_tx_irq() handles the problem: it uses do..while() statement to make sure the budget can be decreased to zero at most and the negative overflow never happens.

The patch adds 'likely' because we rarely would not hit the loop codition since the standard budget is 256.

Quoted 8 lines
Quoted 3 lines
Fixes 8221c5eba8c1 ("ixgbe: add AF_XDP zero-copy Tx support")

Hi Tony,

Any update here? Thanks! I'm asking because I'm ready to send an afxdp patch series based on the patch :)

Hi Jason,

I have this slated to be part of my next net series. I do already have this patch applied/staged on the Intel tree [1] (dev-queue branch). You can base it on that and send the changes now; I'll ensure that this patch is merged before sending the other series.

Thanks,
Tony

[1] https://git.kernel.org/pub/scm/linux/kernel/git/tnguy/next-queue.git/

Quoted 2 lines

Thanks,
jason

author
#10 lore raw
Quoted 50 lines · Tony Nguyen
Quoted 30 lines · Jason Xing
Quoted 29 lines · Jason Xing

From: Jason Xing <[email protected]>

Resolve the budget negative overflow which leads to returning true in ixgbe_xmit_zc even when the budget of descs are thoroughly consumed.

Before this patch, when the budget is decreased to zero and finishes sending the last allowed desc in ixgbe_xmit_zc, it will always turn back and enter into the while() statement to see if it should keep processing packets, but in the meantime it unexpectedly decreases the value again to 'unsigned int (0--)', namely, UINT_MAX. Finally, the ixgbe_xmit_zc returns true, showing 'we complete cleaning the budget'. That also means 'clean_complete = true' in ixgbe_poll.

The true theory behind this is if that budget number of descs are consumed, it implies that we might have more descs to be done. So we should return false in ixgbe_xmit_zc to tell napi poll to find another chance to start polling to handle the rest of descs. On the contrary, returning true here means job done and we know we finish all the possible descs this time and we don't intend to start a new napi poll.

It is apparently against our expectations. Please also see how ixgbe_clean_tx_irq() handles the problem: it uses do..while() statement to make sure the budget can be decreased to zero at most and the negative overflow never happens.

The patch adds 'likely' because we rarely would not hit the loop codition since the standard budget is 256.

Quoted 8 lines
Quoted 3 lines
Fixes 8221c5eba8c1 ("ixgbe: add AF_XDP zero-copy Tx support")

Hi Tony,

Any update here? Thanks! I'm asking because I'm ready to send an afxdp patch series based on the patch :)

Hi Jason,

I have this slated to be part of my next net series. I do already have this patch applied/staged on the Intel tree [1] (dev-queue branch). You can base it on that and send the changes now; I'll ensure that this patch is merged before sending the other series.

Quoted 4 lines

Thanks,
Tony

[1] https://git.kernel.org/pub/scm/linux/kernel/git/tnguy/next-queue.git/

Great, thanks for the pointer!

Thanks,
Jason

A message that is not in the archive (sent off-list or not yet received): PH0PR11MB50135E015152E30636D2AC37962AA@PH0PR11MB5013.namprd11.prod.outlook.com

Tested-by
#12 lore raw
Quoted 80 lines

-----Original Message-----
From: Intel-wired-lan <[email protected]> On Behalf Of Jason Xing
Sent: Saturday, July 26, 2025 12:34 PM
To: Nguyen, Anthony L <[email protected]>; Kitszel, Przemyslaw <[email protected]>; Zaremba, Larysa <[email protected]>; [email protected]; [email protected]; [email protected]; [email protected]; [email protected]; [email protected]; Fijalkowski, Maciej <[email protected]>
Cc: [email protected]; [email protected]; Jason Xing <[email protected]>
Subject: [Intel-wired-lan] [PATCH v2 iwl-net] ixgbe: xsk: resolve the negative overflow of budget in ixgbe_xmit_zc

From: Jason Xing <[email protected]>

Resolve the budget negative overflow which leads to returning true in ixgbe_xmit_zc even when the budget of descs are thoroughly consumed.

Before this patch, when the budget is decreased to zero and finishes sending the last allowed desc in ixgbe_xmit_zc, it will always turn back and enter into the while() statement to see if it should keep processing packets, but in the meantime it unexpectedly decreases the value again to 'unsigned int (0--)', namely, UINT_MAX. Finally, the ixgbe_xmit_zc returns true, showing 'we complete cleaning the budget'. That also means 'clean_complete = true' in ixgbe_poll.

The true theory behind this is if that budget number of descs are consumed, it implies that we might have more descs to be done. So we should return false in ixgbe_xmit_zc to tell napi poll to find another chance to start polling to handle the rest of descs. On the contrary, returning true here means job done and we know we finish all the possible descs this time and we don't intend to start a new napi poll.

It is apparently against our expectations. Please also see how ixgbe_clean_tx_irq() handles the problem: it uses do..while() statement to make sure the budget can be decreased to zero at most and the negative overflow never happens.

The patch adds 'likely' because we rarely would not hit the loop codition since the standard budget is 256.

Fixes 8221c5eba8c1 ("ixgbe: add AF_XDP zero-copy Tx support")

---

[email protected]/
1. use 'negative overflow' instead of 'underflow' (Willem) 2. add reviewed-by tag (Larysa) 3. target iwl-net branch (Larysa) 4. add the reason why the patch adds likely() (Larysa)

---
 drivers/net/ethernet/intel/ixgbe/ixgbe_xsk.c | 4 +++-
 1 file changed, 3 insertions(+), 1 deletion(-)
drivers/net/ethernet/intel/ixgbe/ixgbe_xsk.c +1 -1
index ac58964b2f08..7b941505a9d0 100644
@@ -398,7 +398,7 @@ static bool ixgbe_xmit_zc(struct ixgbe_ring
 *xdp_ring, unsigned int budget)
 	dma_addr_t dma;
 	u32 cmd_type;
 
-	while (budget-- > 0) {
+	while (likely(budget)) {
 		if (unlikely(!ixgbe_desc_unused(xdp_ring))) {
 			work_done = false;
drivers/net/ethernet/intel/ixgbe/ixgbe_xsk.c +2 -0
 			break;
@@ -433,6 +433,8 @@ static bool ixgbe_xmit_zc(struct ixgbe_ring
 *xdp_ring, unsigned int budget)
 		xdp_ring->next_to_use++;
 		if (xdp_ring->next_to_use == xdp_ring->count)
 			xdp_ring->next_to_use = 0;
+
+		budget--;
 	}
 
Quoted 3 lines

if (tx_desc) {
--
2.41.3