Skip to contentSkip to conversation

hns3: work around stack size warning

PATCH In review Arnd Bergmann 5 participants 14 messages Jun 10, 2025 – Jun 23, 2025

Subsystems (4)
lore
Share & tools
lore
#0 lore raw

PATCH 1/1 hns3: work around stack size warning

From: Arnd Bergmann <[email protected]>

The hns3 debugfs functions all use an extra on-stack buffer to store temporary text output before copying that to the debugfs file.

In some configurations with clang, this can trigger the warning limit for the total stack size:

drivers/net/ethernet/hisilicon/hns3/hns3_debugfs.c:788:12: error: stack frame size (1456) exceeds limit (1280) in 'hns3_dbg_tx_queue_info' [-Werror,-Wframe-larger-than]

The problem here is that both hns3_dbg_tx_spare_info() and hns3_dbg_tx_queue_info() have a large on-stack buffer, and clang decides to inline them into a single function.

Annotate hns3_dbg_tx_spare_info() as noinline_for_stack to force the behavior that gcc has, regardless of the compiler.

Ideally all the functions in here would be changed to avoid on-stack output buffers.

---
 drivers/net/ethernet/hisilicon/hns3/hns3_debugfs.c | 5 +++--
 1 file changed, 3 insertions(+), 2 deletions(-)
drivers/net/ethernet/hisilicon/hns3/hns3_debugfs.c +3 -2
index 4e5d8bc39a1b..97dc47eeb44c 100644
@@ -580,8 +580,9 @@ static const struct hns3_dbg_item tx_spare_info_items[] = {
 	{ "DMA", 17 },
 };
 
-static void hns3_dbg_tx_spare_info(struct hns3_enet_ring *ring, char *buf,
-				   int len, u32 ring_num, int *pos)
+static noinline_for_stack void
+hns3_dbg_tx_spare_info(struct hns3_enet_ring *ring, char *buf,
+			int len, u32 ring_num, int *pos)
 {
 	char data_str[ARRAY_SIZE(tx_spare_info_items)][HNS3_DBG_DATA_STR_LEN];
 	struct hns3_tx_spare *tx_spare = ring->tx_spare;
Signature · 1 line
2.39.5
#1 lore raw
Quoted 8 lines · Arnd Bergmann

From: Arnd Bergmann <[email protected]>

The hns3 debugfs functions all use an extra on-stack buffer to store temporary text output before copying that to the debugfs file.

In some configurations with clang, this can trigger the warning limit for the total stack size:

Quoted 8 lines

drivers/net/ethernet/hisilicon/hns3/hns3_debugfs.c:788:12: error: stack frame size (1456) exceeds limit (1280) in 'hns3_dbg_tx_queue_info' [-Werror,-Wframe-larger-than]

The problem here is that both hns3_dbg_tx_spare_info() and hns3_dbg_tx_queue_info() have a large on-stack buffer, and clang decides to inline them into a single function.

Annotate hns3_dbg_tx_spare_info() as noinline_for_stack to force the behavior that gcc has, regardless of the compiler.

This warning is about the potential to exceed the stack space. That potential still exists because of the tail call from hns3_dbg_tx_queue_info() to hns3_dbg_tx_spare_info(), preventing the compile from inlining does nothing against that.

Quoted 2 lines

Ideally all the functions in here would be changed to avoid on-stack output buffers.

That would be my preference as well. Lets give Huawei a bit of time to rewrite this code.

Andrew

maintainer
#2 lore raw
In reply to Arnd Bergmann
Quoted 5 lines · Arnd Bergmann

From: Arnd Bergmann <[email protected]>

The hns3 debugfs functions all use an extra on-stack buffer to store temporary text output before copying that to the debugfs file.

Quoted 8 lines

In some configurations with clang, this can trigger the warning limit for the total stack size:

drivers/net/ethernet/hisilicon/hns3/hns3_debugfs.c:788:12: error: stack frame size (1456) exceeds limit (1280) in 'hns3_dbg_tx_queue_info' [-Werror,-Wframe-larger-than]

The problem here is that both hns3_dbg_tx_spare_info() and hns3_dbg_tx_queue_info() have a large on-stack buffer, and clang decides to inline them into a single function.

Hi Arnd:

Thank you for your report.

Quoted 6 lines

Annotate hns3_dbg_tx_spare_info() as noinline_for_stack to force the behavior that gcc has, regardless of the compiler.

Ideally all the functions in here would be changed to avoid on-stack output buffers.

Would you please help test whether the following changes have solved your problem, And I'm not sure if this patch should be sent to net or net-next...

From d8d1ec419d45411762dd1c8ba24510e5a40bad08 Mon Sep 17 00:00:00 2001
From: Jian Shen <[email protected]>
Date: Tue, 10 Jun 2025 19:35:19 +0800
Subject: [PATCH] net: hns3: clean up compile warning in debugfs

Arnd reported that there are two build warning for on-stasck buffer oversize[1]. So use kmalloc instead of on-stack buffer.

1: https://lore.kernel.org/all/[email protected]/

---
 .../ethernet/hisilicon/hns3/hns3_debugfs.c    | 37 ++++++++++++-------
 1 file changed, 24 insertions(+), 13 deletions(-)
drivers/net/ethernet/hisilicon/hns3/hns3_debugfs.c +24 -13
index dd86a3f66040..0246d9ef26ab 100644
@@ -592,10 +592,11 @@ static const struct hns3_dbg_item tx_spare_info_items[] = {
 static void hns3_dbg_tx_spare_info(struct hns3_enet_ring *ring, char *buf,
 				   int len, u32 ring_num, int *pos)
 {
-	char data_str[ARRAY_SIZE(tx_spare_info_items)][HNS3_DBG_DATA_STR_LEN];
+	size_t item_num = ARRAY_SIZE(tx_spare_info_items);
 	struct hns3_tx_spare *tx_spare = ring->tx_spare;
 	char *result[ARRAY_SIZE(tx_spare_info_items)];
 	char content[HNS3_DBG_INFO_LEN];
+	char *data_str;
 	u32 i, j;
 
 	if (!tx_spare) {
@@ -604,12 +605,16 @@ static void hns3_dbg_tx_spare_info(struct hns3_enet_ring *ring, char *buf,
 		return;
 	}
 
-	for (i = 0; i < ARRAY_SIZE(tx_spare_info_items); i++)
-		result[i] = &data_str[i][0];
+	data_str = kzalloc(item_num * HNS3_DBG_DATA_STR_LEN, GFP_KERNEL);
+	if (!data_str)
+		return;
+
+	for (i = 0; i < item_num; i++)
+		result[i] = &data_str[i * HNS3_DBG_DATA_STR_LEN];
 
 	*pos += scnprintf(buf + *pos, len - *pos, "tx spare buffer info\n");
 	hns3_dbg_fill_content(content, sizeof(content), tx_spare_info_items,
-			      NULL, ARRAY_SIZE(tx_spare_info_items));
+			      NULL, item_num);
 	*pos += scnprintf(buf + *pos, len - *pos, "%s", content);
 
 	for (i = 0; i < ring_num; i++) {
@@ -623,10 +628,11 @@ static void hns3_dbg_tx_spare_info(struct hns3_enet_ring *ring, char *buf,
 		sprintf(result[j++], "%pad", &tx_spare->dma);
 		hns3_dbg_fill_content(content, sizeof(content),
 				      tx_spare_info_items,
-				      (const char **)result,
-				      ARRAY_SIZE(tx_spare_info_items));
+				      (const char **)result, item_num);
 		*pos += scnprintf(buf + *pos, len - *pos, "%s", content);
 	}
+
+	kfree(data_str);
 }
 
 static const struct hns3_dbg_item rx_queue_info_items[] = {
@@ -793,12 +799,13 @@ static void hns3_dump_tx_queue_info(struct hns3_enet_ring *ring,
 static int hns3_dbg_tx_queue_info(struct hnae3_handle *h,
 				  char *buf, int len)
 {
-	char data_str[ARRAY_SIZE(tx_queue_info_items)][HNS3_DBG_DATA_STR_LEN];
+	size_t item_num = ARRAY_SIZE(tx_queue_info_items);
 	struct hnae3_ae_dev *ae_dev = hns3_get_ae_dev(h);
 	char *result[ARRAY_SIZE(tx_queue_info_items)];
 	struct hns3_nic_priv *priv = h->priv;
 	char content[HNS3_DBG_INFO_LEN];
 	struct hns3_enet_ring *ring;
+	char *data_str;
 	int pos = 0;
 	u32 i;
 
@@ -807,11 +814,15 @@ static int hns3_dbg_tx_queue_info(struct hnae3_handle *h,
 		return -EFAULT;
 	}
 
-	for (i = 0; i < ARRAY_SIZE(tx_queue_info_items); i++)
-		result[i] = &data_str[i][0];
+	data_str = kzalloc(item_num * HNS3_DBG_DATA_STR_LEN, GFP_KERNEL);
+	if (!data_str)
+		return -ENOMEM;
+
+	for (i = 0; i < item_num; i++)
+		result[i] = &data_str[i * HNS3_DBG_DATA_STR_LEN];
 
 	hns3_dbg_fill_content(content, sizeof(content), tx_queue_info_items,
-			      NULL, ARRAY_SIZE(tx_queue_info_items));
+			      NULL, item_num);
 	pos += scnprintf(buf + pos, len - pos, "%s", content);
 
 	for (i = 0; i < h->kinfo.num_tqps; i++) {
@@ -827,13 +838,13 @@ static int hns3_dbg_tx_queue_info(struct hnae3_handle *h,
 		hns3_dump_tx_queue_info(ring, ae_dev, result, i);
 		hns3_dbg_fill_content(content, sizeof(content),
 				      tx_queue_info_items,
-				      (const char **)result,
-				      ARRAY_SIZE(tx_queue_info_items));
+				      (const char **)result, item_num);
 		pos += scnprintf(buf + pos, len - pos, "%s", content);
 	}
 
-	hns3_dbg_tx_spare_info(ring, buf, len, h->kinfo.num_tqps, &pos);
+	kfree(data_str);
 
+	hns3_dbg_tx_spare_info(ring, buf, len, h->kinfo.num_tqps, &pos);
 	return 0;
 }
 
Signature · 4 lines
2.33.0


Thanks very much!
Quoted 12 lines
---
  drivers/net/ethernet/hisilicon/hns3/hns3_debugfs.c | 5 +++--
  1 file changed, 3 insertions(+), 2 deletions(-)
drivers/net/ethernet/hisilicon/hns3/hns3_debugfs.c +0 -0
index 4e5d8bc39a1b..97dc47eeb44c 100644
@@ -580,8 +580,9 @@ static const struct hns3_dbg_item tx_spare_info_items[] = {
  	{ "DMA", 17 },

drivers/net/ethernet/hisilicon/hns3/hns3_debugfs.c static const struct hns3_dbg_item tx_spare_info_items[] = { · line 588

Quoted 10 lines
drivers/net/ethernet/hisilicon/hns3/hns3_debugfs.c +3 -2
  };
 
-static void hns3_dbg_tx_spare_info(struct hns3_enet_ring *ring, char *buf,
-				   int len, u32 ring_num, int *pos)
+static noinline_for_stack void
+hns3_dbg_tx_spare_info(struct hns3_enet_ring *ring, char *buf,
+			int len, u32 ring_num, int *pos)
  {
  	char data_str[ARRAY_SIZE(tx_spare_info_items)][HNS3_DBG_DATA_STR_LEN];
  	struct hns3_tx_spare *tx_spare = ring->tx_spare;

From d8d1ec419d45411762dd1c8ba24510e5a40bad08 Mon Sep 17 00:00:00 2001
From: Jian Shen <[email protected]>
Date: Tue, 10 Jun 2025 19:35:19 +0800
Subject: [PATCH] net: hns3: clean up compile warning in debugfs

Arnd reported that there are two build warning for on-stasck buffer oversize[1]. So use kmalloc instead of on-stack buffer.

1: https://lore.kernel.org/all/[email protected]/

---
 .../ethernet/hisilicon/hns3/hns3_debugfs.c    | 37 ++++++++++++-------
 1 file changed, 24 insertions(+), 13 deletions(-)
drivers/net/ethernet/hisilicon/hns3/hns3_debugfs.c +24 -13
index dd86a3f66040..0246d9ef26ab 100644
@@ -592,10 +592,11 @@ static const struct hns3_dbg_item tx_spare_info_items[] = {
 static void hns3_dbg_tx_spare_info(struct hns3_enet_ring *ring, char *buf,
 				   int len, u32 ring_num, int *pos)
 {
-	char data_str[ARRAY_SIZE(tx_spare_info_items)][HNS3_DBG_DATA_STR_LEN];
+	size_t item_num = ARRAY_SIZE(tx_spare_info_items);
 	struct hns3_tx_spare *tx_spare = ring->tx_spare;
 	char *result[ARRAY_SIZE(tx_spare_info_items)];
 	char content[HNS3_DBG_INFO_LEN];
+	char *data_str;
 	u32 i, j;
 
 	if (!tx_spare) {
@@ -604,12 +605,16 @@ static void hns3_dbg_tx_spare_info(struct hns3_enet_ring *ring, char *buf,
 		return;
 	}
 
-	for (i = 0; i < ARRAY_SIZE(tx_spare_info_items); i++)
-		result[i] = &data_str[i][0];
+	data_str = kzalloc(item_num * HNS3_DBG_DATA_STR_LEN, GFP_KERNEL);
+	if (!data_str)
+		return;
+
+	for (i = 0; i < item_num; i++)
+		result[i] = &data_str[i * HNS3_DBG_DATA_STR_LEN];
 
 	*pos += scnprintf(buf + *pos, len - *pos, "tx spare buffer info\n");
 	hns3_dbg_fill_content(content, sizeof(content), tx_spare_info_items,
-			      NULL, ARRAY_SIZE(tx_spare_info_items));
+			      NULL, item_num);
 	*pos += scnprintf(buf + *pos, len - *pos, "%s", content);
 
 	for (i = 0; i < ring_num; i++) {
@@ -623,10 +628,11 @@ static void hns3_dbg_tx_spare_info(struct hns3_enet_ring *ring, char *buf,
 		sprintf(result[j++], "%pad", &tx_spare->dma);
 		hns3_dbg_fill_content(content, sizeof(content),
 				      tx_spare_info_items,
-				      (const char **)result,
-				      ARRAY_SIZE(tx_spare_info_items));
+				      (const char **)result, item_num);
 		*pos += scnprintf(buf + *pos, len - *pos, "%s", content);
 	}
+
+	kfree(data_str);
 }
 
 static const struct hns3_dbg_item rx_queue_info_items[] = {
@@ -793,12 +799,13 @@ static void hns3_dump_tx_queue_info(struct hns3_enet_ring *ring,
 static int hns3_dbg_tx_queue_info(struct hnae3_handle *h,
 				  char *buf, int len)
 {
-	char data_str[ARRAY_SIZE(tx_queue_info_items)][HNS3_DBG_DATA_STR_LEN];
+	size_t item_num = ARRAY_SIZE(tx_queue_info_items);
 	struct hnae3_ae_dev *ae_dev = hns3_get_ae_dev(h);
 	char *result[ARRAY_SIZE(tx_queue_info_items)];
 	struct hns3_nic_priv *priv = h->priv;
 	char content[HNS3_DBG_INFO_LEN];
 	struct hns3_enet_ring *ring;
+	char *data_str;
 	int pos = 0;
 	u32 i;
 
@@ -807,11 +814,15 @@ static int hns3_dbg_tx_queue_info(struct hnae3_handle *h,
 		return -EFAULT;
 	}
 
-	for (i = 0; i < ARRAY_SIZE(tx_queue_info_items); i++)
-		result[i] = &data_str[i][0];
+	data_str = kzalloc(item_num * HNS3_DBG_DATA_STR_LEN, GFP_KERNEL);
+	if (!data_str)
+		return -ENOMEM;
+
+	for (i = 0; i < item_num; i++)
+		result[i] = &data_str[i * HNS3_DBG_DATA_STR_LEN];
 
 	hns3_dbg_fill_content(content, sizeof(content), tx_queue_info_items,
-			      NULL, ARRAY_SIZE(tx_queue_info_items));
+			      NULL, item_num);
 	pos += scnprintf(buf + pos, len - pos, "%s", content);
 
 	for (i = 0; i < h->kinfo.num_tqps; i++) {
@@ -827,13 +838,13 @@ static int hns3_dbg_tx_queue_info(struct hnae3_handle *h,
 		hns3_dump_tx_queue_info(ring, ae_dev, result, i);
 		hns3_dbg_fill_content(content, sizeof(content),
 				      tx_queue_info_items,
-				      (const char **)result,
-				      ARRAY_SIZE(tx_queue_info_items));
+				      (const char **)result, item_num);
 		pos += scnprintf(buf + pos, len - pos, "%s", content);
 	}
 
-	hns3_dbg_tx_spare_info(ring, buf, len, h->kinfo.num_tqps, &pos);
+	kfree(data_str);
 
+	hns3_dbg_tx_spare_info(ring, buf, len, h->kinfo.num_tqps, &pos);
 	return 0;
 }
 
Signature · 1 line
2.33.0
#3 lore raw
Quoted 6 lines · Jijie Shao
Quoted 5 lines · Arnd Bergmann

From: Arnd Bergmann <[email protected]>

Quoted 1 line
Quoted 1 line

Annotate hns3_dbg_tx_spare_info() as noinline_for_stack to force the behavior that gcc has, regardless of the compiler.

Quoted 6 lines
Quoted 2 lines

Ideally all the functions in here would be changed to avoid on-stack output buffers.

Would you please help test whether the following changes have solved your problem,
And I'm not sure if this patch should be sent to net or net-next...

Your patch arrived with whitespace corruption here, so I could not try it, but I'm sure it would help avoid the warning.

However, this is not what meant with my suggestion: you already allocate a temporary buffer in hns3_dbg_open() and I would expect it to be possible to read into that buffer directly without a second temporary buffer (on stack or kmalloc).

The normal way of doing this would be to use the infrastructure from seq_file and then seq_printf() and not have any extra buffers on top of that.

Arnd

maintainer
#4 lore raw
Quoted 19 lines · Arnd Bergmann
Quoted 10 lines · Jijie Shao
Quoted 6 lines · Arnd Bergmann

From: Arnd Bergmann <[email protected]>
Annotate hns3_dbg_tx_spare_info() as noinline_for_stack to force the behavior that gcc has, regardless of the compiler.

Ideally all the functions in here would be changed to avoid on-stack output buffers.

Would you please help test whether the following changes have solved your problem,
And I'm not sure if this patch should be sent to net or net-next...

Your patch arrived with whitespace corruption here, so I could not try it, but I'm sure it would help avoid the warning.

However, this is not what meant with my suggestion: you already allocate a temporary buffer in hns3_dbg_open() and I would expect it to be possible to read into that buffer directly without a second temporary buffer (on stack or kmalloc).

Quoted 5 lines

The normal way of doing this would be to use the infrastructure from seq_file and then seq_printf() and not have any extra buffers on top of that.

Arnd

Hi,

seq_file is good. But the change is quite big.
I need to discuss it internally, and it may not be completed so quickly. I will also need consider the maintainer's suggestion.

Thanks,
Jijie Shao

maintainer
#5 lore raw
Quoted 11 lines · On Thu, 12 Jun 2025 21:09:40 +0800 Jijie Shao wrote:
Quoted 11 lines
Quoted 3 lines

Would you please help test whether the following changes have solved your problem,
And I'm not sure if this patch should be sent to net or net-next...

Your patch arrived with whitespace corruption here, so I could not try it, but I'm sure it would help avoid the warning.

However, this is not what meant with my suggestion: you already allocate a temporary buffer in hns3_dbg_open() and I would expect it to be possible to read into that buffer directly without a second temporary buffer (on stack or kmalloc).

Quoted 7 lines
Quoted 3 lines

The normal way of doing this would be to use the infrastructure from seq_file and then seq_printf() and not have any extra buffers on top of that.

seq_file is good. But the change is quite big.
I need to discuss it internally, and it may not be completed so quickly. I will also need consider the maintainer's suggestion.

Arnd, do you mind waiting? We can apply your patch as is if the warning is a nuisance.

#6 lore raw
Quoted 11 lines · Jakub Kicinski
Quoted 7 lines · On Thu, 12 Jun 2025 21:09:40 +0800 Jijie Shao wrote:
Quoted 3 lines

The normal way of doing this would be to use the infrastructure from seq_file and then seq_printf() and not have any extra buffers on top of that.

seq_file is good. But the change is quite big.
I need to discuss it internally, and it may not be completed so quickly. I will also need consider the maintainer's suggestion.

Arnd, do you mind waiting? We can apply your patch as is if the warning is a nuisance.

It's not urgent, as the current warning limit is still 2048 bytes for 64-bit targets. I am hoping to have all patches in place for the next merge window to be able to reduce the default limit to 1280 bytes and I can just carry my local fix for my own testing until then.

Arnd

maintainer
#7 lore raw
In reply to Jijie Shao
Quoted 3 lines · On Thu, 12 Jun 2025 21:09:40 +0800 Jijie Shao wrote:

seq_file is good. But the change is quite big.
I need to discuss it internally, and it may not be completed so quickly. I will also need consider the maintainer's suggestion.

Please work on the seq_file conversion, given that the merge window just closed you have around 6 weeks to get it done, so hopefully plenty of time.

maintainer
#8 lore raw
Quoted 7 lines · Jakub Kicinski
Quoted 3 lines · On Thu, 12 Jun 2025 21:09:40 +0800 Jijie Shao wrote:

seq_file is good. But the change is quite big.
I need to discuss it internally, and it may not be completed so quickly. I will also need consider the maintainer's suggestion.

Please work on the seq_file conversion, given that the merge window just closed you have around 6 weeks to get it done, so hopefully plenty of time.

Ok

I will try to send patch as soon as possible to complete this conversion

Thanks
Jijie Shao

maintainer
#9 lore raw
Quoted 10 lines · Jijie Shao
Quoted 7 lines · Jakub Kicinski
Quoted 3 lines · On Thu, 12 Jun 2025 21:09:40 +0800 Jijie Shao wrote:

seq_file is good. But the change is quite big.
I need to discuss it internally, and it may not be completed so quickly. I will also need consider the maintainer's suggestion.

Please work on the seq_file conversion, given that the merge window just closed you have around 6 weeks to get it done, so hopefully plenty of time.

Quoted 9 lines

Ok

I will try to send patch as soon as possible to complete this conversion

Thanks
Jijie Shao

We tried modifying the relevant debugfs files,
and both compilation and tests passed.
Please help us check if the solution is OK.
If it is OK, we will modify all the debugfs files in this way.

====================

From c62568f3e91eb5725211fde8e63d44f68452b4e3 Mon Sep 17 00:00:00 2001
From: Jian Shen <[email protected]>
Date: Thu, 19 Jun 2025 16:21:17 +0800
Subject: [PATCH net-next] net: hns3: clean up the build warning in debugfs by
use seq file

Arnd reported that there are two build warning for on-stasck buffer oversize[1]. As Arnd's suggestion, using seq file way to avoid the stack buffer or kmalloc buffer allocating.

---
 drivers/net/ethernet/hisilicon/hns3/hnae3.h   |   3 +
 .../ethernet/hisilicon/hns3/hns3_debugfs.c    | 160 ++++++++----------
 .../net/ethernet/hisilicon/hns3/hns3_enet.c   |   2 +
 3 files changed, 79 insertions(+), 86 deletions(-)
drivers/net/ethernet/hisilicon/hns3/hnae3.h +3 -0
index 8dc7d6fae224..58a63d2eb69b 100644
@@ -434,8 +434,11 @@ struct hnae3_ae_dev {
 	u32 dev_version;
 	DECLARE_BITMAP(caps, HNAE3_DEV_CAPS_MAX_NUM);
 	void *priv;
+	struct hnae3_handle *handle;
 };
 
+typedef int (*read_func)(struct seq_file *s, void *data);
+
 /* This struct defines the operation on the handle.
  *
  * init_ae_dev(): (mandatory)
drivers/net/ethernet/hisilicon/hns3/hns3_debugfs.c +74 -86

241 lines not on this page · raw message

drivers/net/ethernet/hisilicon/hns3/hns3_enet.c +2 -0
index 5c8c62ea6ac0..f01c7e45e674 100644
@@ -5299,6 +5299,8 @@ static int hns3_client_init(struct hnae3_handle *handle)
 	struct net_device *netdev;
 	int ret;
 
+	ae_dev->handle = handle;
+
 	handle->ae_algo->ops->get_tqps_and_rss_info(handle, &alloc_tqps,
 						    &max_rss_size);
 	netdev = alloc_etherdev_mq(sizeof(struct hns3_nic_priv), alloc_tqps);
Signature · 1 line
2.33.0
maintainer
#10 lore raw
In reply to Jijie Shao
Quoted 10 lines · Jijie Shao
Quoted 7 lines · Jakub Kicinski
Quoted 3 lines · On Thu, 12 Jun 2025 21:09:40 +0800 Jijie Shao wrote:

seq_file is good. But the change is quite big.
I need to discuss it internally, and it may not be completed so quickly. I will also need consider the maintainer's suggestion.

Please work on the seq_file conversion, given that the merge window just closed you have around 6 weeks to get it done, so hopefully plenty of time.

Quoted 9 lines

Ok

I will try to send patch as soon as possible to complete this conversion

Thanks
Jijie Shao

*Hi Jakub, Arnd We have changed the impleament as your suggestion. Would you please help check it ? If it's OK, we will rewrite the rest parts of our debugfs code. Thanks! *

==========

From c62568f3e91eb5725211fde8e63d44f68452b4e3 Mon Sep 17 00:00:00 2001
From: Jian Shen <[email protected]>
Date: Thu, 19 Jun 2025 16:21:17 +0800
Subject: [PATCH  net-next] net: hns3: clean up the build warning in debugfs by use seq file

Arnd reported that there are two build warning for on-stasck buffer oversize[1]. As Arnd's suggestion, using seq file way to avoid the stack buffer or kmalloc buffer allocating.

---
 drivers/net/ethernet/hisilicon/hns3/hnae3.h   |   3 + .../ethernet/hisilicon/hns3/hns3_debugfs.c    | 160 ++++++++---------- .../net/ethernet/hisilicon/hns3/hns3_enet.c   |   2 +

 3 files changed, 79 insertions(+), 86 deletions(-)
drivers/net/ethernet/hisilicon/hns3/hnae3.h +0 -0
index 8dc7d6fae224..58a63d2eb69b 100644
@@ -434,8 +434,11 @@ struct hnae3_ae_dev {

     u32 dev_version;
     DECLARE_BITMAP(caps, HNAE3_DEV_CAPS_MAX_NUM); void *priv;
+    struct hnae3_handle *handle;
 };

+typedef int (*read_func)(struct seq_file *s, void *data); +
 /* This struct defines the operation on the handle. *
  * init_ae_dev(): (mandatory)

drivers/net/ethernet/hisilicon/hns3/hns3_debugfs.c +0 -0
index 4e5d8bc39a1b..458a2944ee3c 100644
@@ -3,6 +3,7 @@
 

 #include <linux/debugfs.h>
 #include <linux/device.h>
+#include <linux/seq_file.h>
 #include <linux/string_choices.h>

 #include "hnae3.h"
@@ -41,6 +42,7 @@ static struct hns3_dbg_dentry_info hns3_dbg_dentry[] = {

 static int hns3_dbg_bd_file_init(struct hnae3_handle *handle, u32 cmd); static int hns3_dbg_common_file_init(struct hnae3_handle *handle, u32 cmd); +static int hns3_dbg_common_init_t1(struct hnae3_handle *handle, u32 cmd);

 static struct hns3_dbg_cmd_info hns3_dbg_cmd[] = {
     {
@@ -300,7 +302,7 @@ static struct hns3_dbg_cmd_info hns3_dbg_cmd[] = { .cmd = HNAE3_DBG_CMD_TX_QUEUE_INFO,
         .dentry = HNS3_DBG_DENTRY_QUEUE,
         .buf_len = HNS3_DBG_READ_LEN_1MB,
-        .init = hns3_dbg_common_file_init,
+        .init = hns3_dbg_common_init_t1,
     },
     {
         .name = "fd_tcam",
@@ -580,44 +582,29 @@ static const struct hns3_dbg_item tx_spare_info_items[] = { { "DMA", 17 },
 };

-static void hns3_dbg_tx_spare_info(struct hns3_enet_ring *ring, char *buf, -                   int len, u32 ring_num, int *pos)
+static void hns3_dbg_tx_spare_info(struct hns3_enet_ring *ring, +                   struct seq_file *s, u32 ring_num)
 {
-    char data_str[ARRAY_SIZE(tx_spare_info_items)][HNS3_DBG_DATA_STR_LEN]; struct hns3_tx_spare *tx_spare = ring->tx_spare;
-    char *result[ARRAY_SIZE(tx_spare_info_items)];
-    char content[HNS3_DBG_INFO_LEN];
-    u32 i, j;
+    u32 i;

     if (!tx_spare) {
-        *pos += scnprintf(buf + *pos, len - *pos, -                  "tx spare buffer is not enabled\n"); +        seq_puts(s, "tx spare buffer is not enabled\n"); return;
     }

-    for (i = 0; i < ARRAY_SIZE(tx_spare_info_items); i++)
-        result[i] = &data_str[i][0];
-
-    *pos += scnprintf(buf + *pos, len - *pos, "tx spare buffer info\n"); -    hns3_dbg_fill_content(content, sizeof(content), tx_spare_info_items, -                  NULL, ARRAY_SIZE(tx_spare_info_items));
-    *pos += scnprintf(buf + *pos, len - *pos, "%s", content); +    seq_puts(s, "tx spare buffer info\n");

-    for (i = 0; i < ring_num; i++) {
-        j = 0;
-        sprintf(result[j++], "%u", i);
-        sprintf(result[j++], "%u", ring->tx_copybreak);
-        sprintf(result[j++], "%u", tx_spare->len);
-        sprintf(result[j++], "%u", tx_spare->next_to_use);
-        sprintf(result[j++], "%u", tx_spare->next_to_clean); -        sprintf(result[j++], "%u", tx_spare->last_to_clean); -        sprintf(result[j++], "%pad", &tx_spare->dma);
-        hns3_dbg_fill_content(content, sizeof(content),
-                      tx_spare_info_items,
-                      (const char **)result,
-                      ARRAY_SIZE(tx_spare_info_items));
-        *pos += scnprintf(buf + *pos, len - *pos, "%s", content); -    }
+    for (i = 0; i < ARRAY_SIZE(tx_spare_info_items); i++)
+        seq_printf(s, "%s%*s", tx_spare_info_items[i].name,
+               tx_spare_info_items[i].interval, " ");
+    seq_puts(s, "\n");
+
+    for (i = 0; i < ring_num; i++)
+        seq_printf(s, "%-4u%6s%-5u%6s%-8u%2s%-5u%2s%-5u%2s%-5u%2s%pad\n", +               i, " ", ring->tx_copybreak, " ", tx_spare->len, " ", +               tx_spare->next_to_use, " ", tx_spare->next_to_clean, +               " ", tx_spare->last_to_clean, " ", &tx_spare->dma); }

 static const struct hns3_dbg_item rx_queue_info_items[] = {
@@ -739,62 +726,52 @@ static const struct hns3_dbg_item tx_queue_info_items[] = { };

 static void hns3_dump_tx_queue_info(struct hns3_enet_ring *ring, -                    struct hnae3_ae_dev *ae_dev, char **result, -                    u32 index)
+                    struct hnae3_ae_dev *ae_dev,
+                    struct seq_file *s, u32 index)
 {
+    void __iomem *base = ring->tqp->io_base;
     u32 base_add_l, base_add_h;
-    u32 j = 0;
-
-    sprintf(result[j++], "%u", index);
-    sprintf(result[j++], "%u", readl_relaxed(ring->tqp->io_base + -        HNS3_RING_TX_RING_BD_NUM_REG));
-
-    sprintf(result[j++], "%u", readl_relaxed(ring->tqp->io_base + -        HNS3_RING_TX_RING_TC_REG));
-
-    sprintf(result[j++], "%u", readl_relaxed(ring->tqp->io_base + -        HNS3_RING_TX_RING_TAIL_REG));
-
-    sprintf(result[j++], "%u", readl_relaxed(ring->tqp->io_base + -        HNS3_RING_TX_RING_HEAD_REG));
-
-    sprintf(result[j++], "%u", readl_relaxed(ring->tqp->io_base + -        HNS3_RING_TX_RING_FBDNUM_REG));
-
-    sprintf(result[j++], "%u", readl_relaxed(ring->tqp->io_base + -        HNS3_RING_TX_RING_OFFSET_REG));
-
-    sprintf(result[j++], "%u", readl_relaxed(ring->tqp->io_base + -        HNS3_RING_TX_RING_PKTNUM_RECORD_REG));

-    sprintf(result[j++], "%s",
-        str_on_off(readl_relaxed(ring->tqp->io_base +
-                     HNS3_RING_EN_REG)));
+    seq_printf(s, "%-4u%6s", index, " ");
+    seq_printf(s, "%-5u%3s",
+           readl_relaxed(base + HNS3_RING_TX_RING_BD_NUM_REG), " "); +    seq_printf(s, "%u%3s",
+           readl_relaxed(base + HNS3_RING_TX_RING_TC_REG), " "); +    seq_printf(s, "%-4u%2s",
+           readl_relaxed(base + HNS3_RING_TX_RING_TAIL_REG), " "); +    seq_printf(s, "%-4u%2s",
+           readl_relaxed(base + HNS3_RING_TX_RING_HEAD_REG), " "); +    seq_printf(s, "%-4u%4s",
+           readl_relaxed(base + HNS3_RING_TX_RING_FBDNUM_REG), " "); +    seq_printf(s, "%-4u%4s",
+           readl_relaxed(base + HNS3_RING_TX_RING_OFFSET_REG), " "); +    seq_printf(s, "%-9u%2s",
+           readl_relaxed(base + HNS3_RING_TX_RING_PKTNUM_RECORD_REG), +           " ");
+    seq_printf(s, "%-3s%6s",
+           str_on_off(readl_relaxed(base + HNS3_RING_EN_REG)), " ");

     if (hnae3_ae_dev_tqp_txrx_indep_supported(ae_dev)) -        sprintf(result[j++], "%s",
-            str_on_off(readl_relaxed(ring->tqp->io_base + -                         HNS3_RING_TX_EN_REG))); +        seq_printf(s, "%-3s%9s",
+               str_on_off(readl_relaxed(base +
+                            HNS3_RING_TX_EN_REG)), +               " ");
     else
-        sprintf(result[j++], "%s", "NA");
+        seq_printf(s, "%-3s%9s", "NA", " ");

     base_add_h = readl_relaxed(ring->tqp->io_base + HNS3_RING_TX_RING_BASEADDR_H_REG); base_add_l = readl_relaxed(ring->tqp->io_base + HNS3_RING_TX_RING_BASEADDR_L_REG); -    sprintf(result[j++], "0x%08x%08x", base_add_h, base_add_l); +    seq_printf(s, "0x%08x%08x\n", base_add_h, base_add_l); }

-static int hns3_dbg_tx_queue_info(struct hnae3_handle *h,
-                  char *buf, int len)
+static int hns3_dbg_tx_queue_info(struct seq_file *s, void *data) {
-    char data_str[ARRAY_SIZE(tx_queue_info_items)][HNS3_DBG_DATA_STR_LEN]; -    struct hnae3_ae_dev *ae_dev = pci_get_drvdata(h->pdev);
-    char *result[ARRAY_SIZE(tx_queue_info_items)];
+    struct hnae3_ae_dev *ae_dev = dev_get_drvdata(s->private); +    struct hnae3_handle *h = ae_dev->handle;
     struct hns3_nic_priv *priv = h->priv;
-    char content[HNS3_DBG_INFO_LEN];
     struct hns3_enet_ring *ring;
-    int pos = 0;
     u32 i;

     if (!priv->ring) {
@@ -803,11 +780,10 @@ static int hns3_dbg_tx_queue_info(struct hnae3_handle *h, }

     for (i = 0; i < ARRAY_SIZE(tx_queue_info_items); i++) -        result[i] = &data_str[i][0];
+        seq_printf(s, "%s%*s", tx_queue_info_items[i].name, +               tx_queue_info_items[i].interval, " ");

-    hns3_dbg_fill_content(content, sizeof(content), tx_queue_info_items, -                  NULL, ARRAY_SIZE(tx_queue_info_items));
-    pos += scnprintf(buf + pos, len - pos, "%s", content);
+    seq_puts(s, "\n");

     for (i = 0; i < h->kinfo.num_tqps; i++) {
         /* Each cycle needs to determine whether the instance is reset, @@ -819,15 +795,10 @@ static int hns3_dbg_tx_queue_info(struct hnae3_handle *h, return -EPERM;

         ring = &priv->ring[i];
-        hns3_dump_tx_queue_info(ring, ae_dev, result, i); -        hns3_dbg_fill_content(content, sizeof(content), -                      tx_queue_info_items,
-                      (const char **)result,
-                      ARRAY_SIZE(tx_queue_info_items)); -        pos += scnprintf(buf + pos, len - pos, "%s", content); +        hns3_dump_tx_queue_info(ring, ae_dev, s, i); }

-    hns3_dbg_tx_spare_info(ring, buf, len, h->kinfo.num_tqps, &pos); +    hns3_dbg_tx_spare_info(ring, s, h->kinfo.num_tqps);

     return 0;
 }
@@ -1222,10 +1193,6 @@ static const struct hns3_dbg_func hns3_dbg_cmd_func[] = { .cmd = HNAE3_DBG_CMD_RX_QUEUE_INFO,
         .dbg_dump = hns3_dbg_rx_queue_info,
     },
-    {
-        .cmd = HNAE3_DBG_CMD_TX_QUEUE_INFO,
-        .dbg_dump = hns3_dbg_tx_queue_info,
-    },
     {
         .cmd = HNAE3_DBG_CMD_PAGE_POOL_INFO,
         .dbg_dump = hns3_dbg_page_pool_info,
@@ -1362,6 +1329,27 @@ hns3_dbg_common_file_init(struct hnae3_handle *handle, u32 cmd) return 0;
 }

+static int hns3_dbg_common_init_t1(struct hnae3_handle *handle, u32 cmd) +{
+    struct device *dev = &handle->pdev->dev;
+    struct dentry *entry_dir;
+    read_func func = NULL;
+
+    switch (hns3_dbg_cmd[cmd].cmd) {
+    case HNAE3_DBG_CMD_TX_QUEUE_INFO:
+        func = hns3_dbg_tx_queue_info;
+        break;
+    default:
+        return -EINVAL;
+    }
+
+    entry_dir = hns3_dbg_dentry[hns3_dbg_cmd[cmd].dentry].dentry; +    debugfs_create_devm_seqfile(dev, hns3_dbg_cmd[cmd].name, entry_dir, +                    func);
+
+    return 0;
+}
+
 int hns3_dbg_init(struct hnae3_handle *handle)
 {
     struct hnae3_ae_dev *ae_dev = pci_get_drvdata(handle->pdev);

drivers/net/ethernet/hisilicon/hns3/hns3_enet.c +0 -0
index 5c8c62ea6ac0..f01c7e45e674 100644
@@ -5299,6 +5299,8 @@ static int hns3_client_init(struct hnae3_handle *handle)

     struct net_device *netdev;
     int ret;

+    ae_dev->handle = handle;
+
     handle->ae_algo->ops->get_tqps_and_rss_info(handle, &alloc_tqps, &max_rss_size);
     netdev = alloc_etherdev_mq(sizeof(struct hns3_nic_priv), alloc_tqps);

Signature · 1 line
2.33.0
#11 lore raw
Quoted 6 lines · Jijie Shao
Quoted 2 lines

on 2025/6/12 23:33, Jakub Kicinski wrote:

*Hi Jakub, Arnd We have changed the impleament as your suggestion. Would you please help check it ? If it's OK, we will rewrite the rest parts of our debugfs code. Thanks! *

The conversion to seq_file looks good to me, this does address the stack usage problems I was observing.
Thanks for cleaning this up!

Review

Quoted 6 lines
(quoted diff) +3 -3
-    sprintf(result[j++], "%u", index);
-    sprintf(result[j++], "%u", readl_relaxed(ring->tqp->io_base +
-        HNS3_RING_TX_RING_BD_NUM_REG));
+    seq_printf(s, "%-4u%6s", index, " ");
+    seq_printf(s, "%-5u%3s",
+           readl_relaxed(base + HNS3_RING_TX_RING_BD_NUM_REG), " ");

I'm not sure I understand the format string changes here, I did not think they were necessary.

Are you doing this to keep the output the same as before, or are you reformatting the contents for readability?

Quoted 10 lines

+static int hns3_dbg_common_init_t1(struct hnae3_handle *handle, u32 cmd)

(quoted diff) +8 -0
+{
+    struct device *dev = &handle->pdev->dev;
+    struct dentry *entry_dir;
+    read_func func = NULL;
+
+    switch (hns3_dbg_cmd[cmd].cmd) {
+    case HNAE3_DBG_CMD_TX_QUEUE_INFO:
+        func = hns3_dbg_tx_queue_info;

Review

Quoted 10 lines
(quoted diff) +10 -0
+        break;
+    default:
+        return -EINVAL;
+    }
+
+    entry_dir = hns3_dbg_dentry[hns3_dbg_cmd[cmd].dentry].dentry;
+    debugfs_create_devm_seqfile(dev, hns3_dbg_cmd[cmd].name, entry_dir,
+                    func);
+
+    return 0;

This will work fine as well, but I think you can do slightly better by having your own file_operations with a read function based on single_open() and your current hns3_dbg_read_cmd().

I don't think you gain anything from using debugfs_create_devm_seqfile() since you use debugfs_remove_recursive() for cleaning it up anyway.

Arnd

maintainer
#12 lore raw
Quoted 16 lines · Arnd Bergmann
Quoted 5 lines · Jijie Shao
Quoted 2 lines

on 2025/6/12 23:33, Jakub Kicinski wrote:

*Hi Jakub, Arnd We have changed the impleament as your suggestion. Would you please help check it ? If it's OK, we will rewrite the rest parts of our debugfs code. Thanks! *

The conversion to seq_file looks good to me, this does address the stack usage problems I was observing.
Thanks for cleaning this up!

Review

Quoted 6 lines
(quoted diff) +3 -3
-    sprintf(result[j++], "%u", index);
-    sprintf(result[j++], "%u", readl_relaxed(ring->tqp->io_base +
-        HNS3_RING_TX_RING_BD_NUM_REG));
+    seq_printf(s, "%-4u%6s", index, " ");
+    seq_printf(s, "%-5u%3s",
+           readl_relaxed(base + HNS3_RING_TX_RING_BD_NUM_REG), " ");
Quoted 5 lines

I'm not sure I understand the format string changes here, I did not think they were necessary.

Are you doing this to keep the output the same as before, or are you reformatting the contents for readability?

yeah, just to keep the output the same as before

Quoted 25 lines
Quoted 11 lines

+static int hns3_dbg_common_init_t1(struct hnae3_handle *handle, u32 cmd)

(quoted diff) +8 -0
+{
+    struct device *dev = &handle->pdev->dev;
+    struct dentry *entry_dir;
+    read_func func = NULL;
+
+    switch (hns3_dbg_cmd[cmd].cmd) {
+    case HNAE3_DBG_CMD_TX_QUEUE_INFO:
+        func = hns3_dbg_tx_queue_info;

Review

Quoted 10 lines
(quoted diff) +10 -0
+        break;
+    default:
+        return -EINVAL;
+    }
+
+    entry_dir = hns3_dbg_dentry[hns3_dbg_cmd[cmd].dentry].dentry;
+    debugfs_create_devm_seqfile(dev, hns3_dbg_cmd[cmd].name, entry_dir,
+                    func);
+
+    return 0;

This will work fine as well, but I think you can do slightly better by having your own file_operations with a read function based on single_open() and your current hns3_dbg_read_cmd().

Quoted 4 lines

I don't think you gain anything from using debugfs_create_devm_seqfile() since you use debugfs_remove_recursive() for cleaning it up anyway.

Arnd

Using debugfs_create_devm_seqfile() is just to simplify the code. We only need to focus on the implementation of .read() function.

Thanks
Jijie Shao

#13 lore raw
Quoted 8 lines · Jijie Shao
Quoted 7 lines · Arnd Bergmann
Quoted 7 lines
(quoted diff) +3 -3
-    sprintf(result[j++], "%u", index);
-    sprintf(result[j++], "%u", readl_relaxed(ring->tqp->io_base +
-        HNS3_RING_TX_RING_BD_NUM_REG));
+    seq_printf(s, "%-4u%6s", index, " ");
+    seq_printf(s, "%-5u%3s",
+           readl_relaxed(base + HNS3_RING_TX_RING_BD_NUM_REG), " ");
Quoted 7 lines
Quoted 5 lines

I'm not sure I understand the format string changes here, I did not think they were necessary.

Are you doing this to keep the output the same as before, or are you reformatting the contents for readability?

yeah, just to keep the output the same as before

Ok.

Quoted 24 lines
Quoted 24 lines
Quoted 10 lines

+static int hns3_dbg_common_init_t1(struct hnae3_handle *handle, u32 cmd)

(quoted diff) +8 -0
+{
+    struct device *dev = &handle->pdev->dev;
+    struct dentry *entry_dir;
+    read_func func = NULL;
+
+    switch (hns3_dbg_cmd[cmd].cmd) {
+    case HNAE3_DBG_CMD_TX_QUEUE_INFO:
+        func = hns3_dbg_tx_queue_info;

Review

Quoted 10 lines
(quoted diff) +10 -0
+        break;
+    default:
+        return -EINVAL;
+    }
+
+    entry_dir = hns3_dbg_dentry[hns3_dbg_cmd[cmd].dentry].dentry;
+    debugfs_create_devm_seqfile(dev, hns3_dbg_cmd[cmd].name, entry_dir,
+                    func);
+
+    return 0;

This will work fine as well, but I think you can do slightly better by having your own file_operations with a read function based on single_open() and your current hns3_dbg_read_cmd().

Quoted 5 lines
Quoted 2 lines

I don't think you gain anything from using debugfs_create_devm_seqfile() since you use debugfs_remove_recursive() for cleaning it up anyway.

Using debugfs_create_devm_seqfile() is just to simplify the code. We only need to focus on the implementation of .read() function.

What I meant is that it doesn't seem simpler to me, as it adds one level of indirection to both the file creation and the read() function compared to having a single_open() helpe with that switch()/case or the corresponding hns3_dbg_cmd_func[] array.

Either way, I'm not worried about it, there is no actual problem that I see with your version.

Arnd