Skip to contentSkip to conversation

igb_uio: remove out-of-date comment

PATCH Superseded Xiaolong Ye 4 participants 12 messages Jan 15, 2019 – Jan 31, 2019

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 igb_uio: remove out-of-date comment

The comment for igbuio_pci_irqhandler is out of date as the code evolves, remove it to avoid misleading.

---
 kernel/linux/igb_uio/igb_uio.c | 5 -----
 1 file changed, 5 deletions(-)
kernel/linux/igb_uio/igb_uio.c +0 -5
index 3cf394bdf..7e69b4498 100644
@@ -185,11 +185,6 @@ igbuio_pci_irqcontrol(struct uio_info *info, s32 irq_state)
 	return 0;
 }
 
-/**
- * This is interrupt handler which will check if the interrupt is for the right device.
- * If yes, disable it here and will be enable later.
- */
-static irqreturn_t
 igbuio_pci_irqhandler(int irq, void *dev_id)
 {
 	struct rte_uio_pci_dev *udev = (struct rte_uio_pci_dev *)dev_id;
Signature · 1 line
2.17.1
#1 lore raw
Quoted 15 lines · Xiaolong Ye

The comment for igbuio_pci_irqhandler is out of date as the code evolves, remove it to avoid misleading.

---
  kernel/linux/igb_uio/igb_uio.c | 5 -----
  1 file changed, 5 deletions(-)
kernel/linux/igb_uio/igb_uio.c +0 -0
index 3cf394bdf..7e69b4498 100644
@@ -185,11 +185,6 @@ igbuio_pci_irqcontrol(struct uio_info *info, s32 irq_state)
  	return 0;

kernel/linux/igb_uio/igb_uio.c igbuio_pci_irqcontrol(struct uio_info *info, s32 irq_state) · line 190

Quoted 10 lines
kernel/linux/igb_uio/igb_uio.c +0 -5
  }
 
-/**
- * This is interrupt handler which will check if the interrupt is for the right device.
- * If yes, disable it here and will be enable later.
- */
-static irqreturn_t
  igbuio_pci_irqhandler(int irq, void *dev_id)
  {
  	struct rte_uio_pci_dev *udev = (struct rte_uio_pci_dev *)dev_id;

I think you removed one extra line there...

Signature · 2 lines
Thanks,
Anatoly
#2 lore raw
Quoted 19 lines · Anatoly
Quoted 18 lines · Xiaolong Ye

The comment for igbuio_pci_irqhandler is out of date as the code evolves, remove it to avoid misleading.

---
  kernel/linux/igb_uio/igb_uio.c | 5 -----
  1 file changed, 5 deletions(-)
kernel/linux/igb_uio/igb_uio.c +0 -3
index 3cf394bdf..7e69b4498 100644
@@ -185,11 +185,6 @@ igbuio_pci_irqcontrol(struct uio_info *info, s32 irq_state)
  	return 0;
  }
-/**
- * This is interrupt handler which will check if the interrupt is for the right device.
- * If yes, disable it here and will be enable later.
Quoted 8 lines

kernel/linux/igb_uio/igb_uio.c igbuio_pci_irqcontrol(struct uio_info *info, s32 irq_state)

Quoted 6 lines
kernel/linux/igb_uio/igb_uio.c +0 -2
- */
-static irqreturn_t
  igbuio_pci_irqhandler(int irq, void *dev_id)
  {
  	struct rte_uio_pci_dev *udev = (struct rte_uio_pci_dev *)dev_id;
 

I think you removed one extra line there...

oops, thanks for the catch, I'll send a v2.

Thanks,
Xiaolong

Quoted 4 lines
Signature · 3 lines
-- 
Thanks,
Anatoly
#3 lore raw

PATCH v2 1/1 igb_uio: remove out-of-date comment

In reply to Xiaolong Ye

The comment for igbuio_pci_irqhandler is out of date as the code evolves, remove it to avoid misleading.

---
 kernel/linux/igb_uio/igb_uio.c | 4 ----
 1 file changed, 4 deletions(-)
kernel/linux/igb_uio/igb_uio.c +0 -4
index 3cf394bdf..d6ac51e79 100644
@@ -185,10 +185,6 @@ igbuio_pci_irqcontrol(struct uio_info *info, s32 irq_state)
 	return 0;
 }
 
-/**
- * This is interrupt handler which will check if the interrupt is for the right device.
- * If yes, disable it here and will be enable later.
- */
 static irqreturn_t
 igbuio_pci_irqhandler(int irq, void *dev_id)
 {
Signature · 1 line
2.17.1
#4 lore raw
Quoted 13 lines · On Wed, 16 Jan 2019 08:34:52 +0800 Xiaolong Ye <[email protected]> wrote:

The comment for igbuio_pci_irqhandler is out of date as the code evolves, remove it to avoid misleading.

---
 kernel/linux/igb_uio/igb_uio.c | 4 ----
 1 file changed, 4 deletions(-)
kernel/linux/igb_uio/igb_uio.c +0 -0
index 3cf394bdf..d6ac51e79 100644

kernel/linux/igb_uio/igb_uio.c igbuio_pci_irqcontrol(struct uio_info *info, s32 irq_state) · line 190

Quoted 10 lines
kernel/linux/igb_uio/igb_uio.c +0 -4
@@ -185,10 +185,6 @@ igbuio_pci_irqcontrol(struct uio_info *info, s32 irq_state)
 	return 0;
 }
 
-/**
- * This is interrupt handler which will check if the interrupt is for the right device.
- * If yes, disable it here and will be enable later.
- */
 static irqreturn_t
 igbuio_pci_irqhandler(int irq, void *dev_id)
 {

The comment is partially correct; if you look at the legacy case.

Maybe better to move the comment to pci_check_and_mask_intx in compat.h? I see there is another incorrect comment there.

#5 lore raw
Quoted 29 lines · Stephen Hemminger
Quoted 13 lines · On Wed, 16 Jan 2019 08:34:52 +0800 Xiaolong Ye <[email protected]> wrote:

The comment for igbuio_pci_irqhandler is out of date as the code evolves, remove it to avoid misleading.

---
 kernel/linux/igb_uio/igb_uio.c | 4 ----
 1 file changed, 4 deletions(-)
kernel/linux/igb_uio/igb_uio.c +0 -0
index 3cf394bdf..d6ac51e79 100644

kernel/linux/igb_uio/igb_uio.c igbuio_pci_irqcontrol(struct uio_info *info, s32 irq_state) · line 190

Quoted 10 lines
kernel/linux/igb_uio/igb_uio.c +0 -4
@@ -185,10 +185,6 @@ igbuio_pci_irqcontrol(struct uio_info *info, s32 irq_state)
 	return 0;
 }
 
-/**
- * This is interrupt handler which will check if the interrupt is for the right device.
- * If yes, disable it here and will be enable later.
- */
 static irqreturn_t
 igbuio_pci_irqhandler(int irq, void *dev_id)
 {

The comment is partially correct; if you look at the legacy case.

Quoted 3 lines

Maybe better to move the comment to pci_check_and_mask_intx in compat.h? I see there is another incorrect comment there.

Hmm, you're right, this comment is correct for the legacy case, I'll move the comment to pci_check_and_mask_intx as you suggested.

Thanks,
Xiaolong

#6 lore raw
In reply to Stephen Hemminger

Hi, Stephen

Quoted 29 lines · Stephen Hemminger
Quoted 13 lines · On Wed, 16 Jan 2019 08:34:52 +0800 Xiaolong Ye <[email protected]> wrote:

The comment for igbuio_pci_irqhandler is out of date as the code evolves, remove it to avoid misleading.

---
 kernel/linux/igb_uio/igb_uio.c | 4 ----
 1 file changed, 4 deletions(-)
kernel/linux/igb_uio/igb_uio.c +0 -0
index 3cf394bdf..d6ac51e79 100644

kernel/linux/igb_uio/igb_uio.c igbuio_pci_irqcontrol(struct uio_info *info, s32 irq_state) · line 190

Quoted 10 lines
kernel/linux/igb_uio/igb_uio.c +0 -4
@@ -185,10 +185,6 @@ igbuio_pci_irqcontrol(struct uio_info *info, s32 irq_state)
 	return 0;
 }
 
-/**
- * This is interrupt handler which will check if the interrupt is for the right device.
- * If yes, disable it here and will be enable later.
- */
 static irqreturn_t
 igbuio_pci_irqhandler(int irq, void *dev_id)
 {

The comment is partially correct; if you look at the legacy case.

Quoted 3 lines

Maybe better to move the comment to pci_check_and_mask_intx in compat.h? I see there is another incorrect comment there.

As I tried to understand pci_check_and_mask_intx behavior, I noticed that it was introduced by commit 399a3f0d ("igb_uio: fix IRQ mode handling"), and it was derived from igbuio_set_interrupt_mask, however there were two parameters in igbuio_set_interrupt_mask as below:

static int
igbuio_set_interrupt_mask(struct rte_uio_pci_dev *udev, int32_t state)

while the pci_check_and_mask_intx only has 1 parameter,

static bool pci_check_and_mask_intx(struct pci_dev *pdev)

but in its function body it still use the "state" accorrding to commit 399a3f0d

+static bool pci_check_and_mask_intx(struct pci_dev *pdev)
{
- /* Some function names changes between 3.2.0 and 3.3.0... */
-#if LINUX_VERSION_CODE < KERNEL_VERSION(3, 3, 0)
- pci_unblock_user_cfg_access(pdev);
-#else
- pci_cfg_access_unlock(pdev);
-#endif
+ bool pending;
+ uint32_t status;
+
+ pci_block_user_cfg_access(dev);
+ pci_read_config_dword(pdev, PCI_COMMAND, &status);
+

+       /* interrupt is not ours, goes to out */
+       pending = (((status >> 16) & PCI_STATUS_INTERRUPT) != 0);
+       if (pending) {

+ uint16_t old, new;
+
+ old = status;
+ if (state != 0) <=========================== state still in use
+ new = old & (~PCI_COMMAND_INTX_DISABLE);
+ else
+ new = old | PCI_COMMAND_INTX_DISABLE;
+
+ if (old != new)
+ pci_write_config_word(pdev, PCI_COMMAND, new);
+ }
+ pci_unblock_user_cfg_access(dev);
+
+ return pending;

and later this was fixed as a typo in commit 5b2f8137 ("igb_uio: fix typos for kernel older than 3.3") which seems not correct.

                old = status;

- if (state != 0)
+ if (status != 0)

I feel like that pci_check_and_mask_intx still needs two parameters from code point of view, please correct me if I was wrong or miss anything, or you think it's sensible, I'll cook a patch for it.

Thanks,
Xiaolong

#7 lore raw
Quoted 89 lines · On Wed, 16 Jan 2019 15:48:36 +0800 Ye Xiaolong <[email protected]> wrote:

Hi, Stephen

Quoted 29 lines · Stephen Hemminger
Quoted 13 lines · On Wed, 16 Jan 2019 08:34:52 +0800 Xiaolong Ye <[email protected]> wrote:

The comment for igbuio_pci_irqhandler is out of date as the code evolves, remove it to avoid misleading.

---
 kernel/linux/igb_uio/igb_uio.c | 4 ----
 1 file changed, 4 deletions(-)
kernel/linux/igb_uio/igb_uio.c +0 -0
index 3cf394bdf..d6ac51e79 100644

kernel/linux/igb_uio/igb_uio.c igbuio_pci_irqcontrol(struct uio_info *info, s32 irq_state) · line 190

Quoted 10 lines
kernel/linux/igb_uio/igb_uio.c +0 -4
@@ -185,10 +185,6 @@ igbuio_pci_irqcontrol(struct uio_info *info, s32 irq_state)
 	return 0;
 }
 
-/**
- * This is interrupt handler which will check if the interrupt is for the right device.
- * If yes, disable it here and will be enable later.
- */
 static irqreturn_t
 igbuio_pci_irqhandler(int irq, void *dev_id)
 {  

The comment is partially correct; if you look at the legacy case.

Quoted 3 lines

Maybe better to move the comment to pci_check_and_mask_intx in compat.h? I see there is another incorrect comment there.

As I tried to understand pci_check_and_mask_intx behavior, I noticed that it was introduced by commit 399a3f0d ("igb_uio: fix IRQ mode handling"), and it was derived from igbuio_set_interrupt_mask, however there were two parameters in igbuio_set_interrupt_mask as below:

static int
igbuio_set_interrupt_mask(struct rte_uio_pci_dev *udev, int32_t state)

while the pci_check_and_mask_intx only has 1 parameter,

static bool pci_check_and_mask_intx(struct pci_dev *pdev)

but in its function body it still use the "state" accorrding to commit 399a3f0d

(quoted diff) +24 -6
+static bool pci_check_and_mask_intx(struct pci_dev *pdev)
 {
-       /* Some function names changes between 3.2.0 and 3.3.0... */
-#if LINUX_VERSION_CODE < KERNEL_VERSION(3, 3, 0)
-       pci_unblock_user_cfg_access(pdev);
-#else
-       pci_cfg_access_unlock(pdev);
-#endif
+       bool pending;
+       uint32_t status;
+
+       pci_block_user_cfg_access(dev);
+       pci_read_config_dword(pdev, PCI_COMMAND, &status);
+
+       /* interrupt is not ours, goes to out */
+       pending = (((status >> 16) & PCI_STATUS_INTERRUPT) != 0);
+       if (pending) {
+               uint16_t old, new;
+
+               old = status;
+               if (state != 0) <=========================== state still in use
+                       new = old & (~PCI_COMMAND_INTX_DISABLE);
+               else
+                       new = old | PCI_COMMAND_INTX_DISABLE;
+
+               if (old != new)
+                       pci_write_config_word(pdev, PCI_COMMAND, new);
+       }
+       pci_unblock_user_cfg_access(dev);
+
+       return pending;

and later this was fixed as a typo in commit 5b2f8137 ("igb_uio: fix typos for kernel older than 3.3") which seems not correct.

                old = status;
(quoted diff) +1 -1
-               if (state != 0)
+               if (status != 0)
Quoted 6 lines

I feel like that pci_check_and_mask_intx still needs two parameters from code point of view, please correct me if I was wrong or miss anything, or you think it's sensible, I'll cook a patch for it.

Thanks,
Xiaolong

This code is copy of upstream kernel function, you should look at that.

#8 lore raw
Quoted 92 lines · Stephen Hemminger
Quoted 89 lines · On Wed, 16 Jan 2019 15:48:36 +0800 Ye Xiaolong <[email protected]> wrote:

Hi, Stephen

Quoted 29 lines · Stephen Hemminger
Quoted 13 lines · On Wed, 16 Jan 2019 08:34:52 +0800 Xiaolong Ye <[email protected]> wrote:

The comment for igbuio_pci_irqhandler is out of date as the code evolves, remove it to avoid misleading.

---
 kernel/linux/igb_uio/igb_uio.c | 4 ----
 1 file changed, 4 deletions(-)
kernel/linux/igb_uio/igb_uio.c +0 -0
index 3cf394bdf..d6ac51e79 100644

kernel/linux/igb_uio/igb_uio.c igbuio_pci_irqcontrol(struct uio_info *info, s32 irq_state) · line 190

Quoted 10 lines
kernel/linux/igb_uio/igb_uio.c +0 -4
@@ -185,10 +185,6 @@ igbuio_pci_irqcontrol(struct uio_info *info, s32 irq_state)
 	return 0;
 }
 
-/**
- * This is interrupt handler which will check if the interrupt is for the right device.
- * If yes, disable it here and will be enable later.
- */
 static irqreturn_t
 igbuio_pci_irqhandler(int irq, void *dev_id)
 {  

The comment is partially correct; if you look at the legacy case.

Quoted 3 lines

Maybe better to move the comment to pci_check_and_mask_intx in compat.h? I see there is another incorrect comment there.

As I tried to understand pci_check_and_mask_intx behavior, I noticed that it was introduced by commit 399a3f0d ("igb_uio: fix IRQ mode handling"), and it was derived from igbuio_set_interrupt_mask, however there were two parameters in igbuio_set_interrupt_mask as below:

static int
igbuio_set_interrupt_mask(struct rte_uio_pci_dev *udev, int32_t state)

while the pci_check_and_mask_intx only has 1 parameter,

static bool pci_check_and_mask_intx(struct pci_dev *pdev)

but in its function body it still use the "state" accorrding to commit 399a3f0d

(quoted diff) +24 -6
+static bool pci_check_and_mask_intx(struct pci_dev *pdev)
 {
-       /* Some function names changes between 3.2.0 and 3.3.0... */
-#if LINUX_VERSION_CODE < KERNEL_VERSION(3, 3, 0)
-       pci_unblock_user_cfg_access(pdev);
-#else
-       pci_cfg_access_unlock(pdev);
-#endif
+       bool pending;
+       uint32_t status;
+
+       pci_block_user_cfg_access(dev);
+       pci_read_config_dword(pdev, PCI_COMMAND, &status);
+
+       /* interrupt is not ours, goes to out */
+       pending = (((status >> 16) & PCI_STATUS_INTERRUPT) != 0);
+       if (pending) {
+               uint16_t old, new;
+
+               old = status;
+               if (state != 0) <=========================== state still in use
+                       new = old & (~PCI_COMMAND_INTX_DISABLE);
+               else
+                       new = old | PCI_COMMAND_INTX_DISABLE;
+
+               if (old != new)
+                       pci_write_config_word(pdev, PCI_COMMAND, new);
+       }
+       pci_unblock_user_cfg_access(dev);
+
+       return pending;

and later this was fixed as a typo in commit 5b2f8137 ("igb_uio: fix typos for kernel older than 3.3") which seems not correct.

                old = status;
(quoted diff) +1 -1
-               if (state != 0)
+               if (status != 0)
Quoted 8 lines
Quoted 6 lines

I feel like that pci_check_and_mask_intx still needs two parameters from code point of view, please correct me if I was wrong or miss anything, or you think it's sensible, I'll cook a patch for it.

Thanks,
Xiaolong

This code is copy of upstream kernel function, you should look at that.

Thanks, just checked the kernel version of pci_check_and_mask_intx, and it is a wrap of pci_check_and_set_intx_mask(pdev, true).

	bool pci_check_and_mask_intx(struct pci_dev *dev)
	{
		return pci_check_and_set_intx_mask(dev, true);
	}


	static bool pci_check_and_set_intx_mask(struct pci_dev *dev, bool mask)
	{
		struct pci_bus *bus = dev->bus;
		...
		origcmd = cmd_status_dword;
		newcmd = origcmd & ~PCI_COMMAND_INTX_DISABLE;
		if (mask)
			newcmd |= PCI_COMMAND_INTX_DISABLE;
		if (newcmd != origcmd)
			bus->ops->write(bus, dev->devfn, PCI_COMMAND, 2, newcmd);

	done:
		raw_spin_unlock_irqrestore(&pci_lock, flags);

		return mask_updated;
	}

if mask is set to ture, PCI_COMMAND_INTX_DISABLE will be set in newcmd.

And current dpdk version doesn't align with it:

	static bool pci_check_and_mask_intx(struct pci_dev *pdev)
	{
		bool pending;
		uint32_t status;

		pci_block_user_cfg_access(pdev);
		pci_read_config_dword(pdev, PCI_COMMAND, &status);

		/* interrupt is not ours, goes to out */
		pending = (((status >> 16) & PCI_STATUS_INTERRUPT) != 0);
		if (pending) {
			uint16_t old, new;

			old = status;
			if (status != 0)
				new = old & (~PCI_COMMAND_INTX_DISABLE);
			else
				new = old | PCI_COMMAND_INTX_DISABLE;

			if (old != new)
				pci_write_config_word(pdev, PCI_COMMAND, new);
		}
		pci_unblock_user_cfg_access(pdev);

		return pending;
	}

In this code, if the interrupt is triggered by this device, both pending and status will not be 0, so the PCI_COMMAND_INTX_DISABLE will be cleared for new.

To align the behavior with upstream kernel, we may need change like:

kernel/linux/igb_uio/compat.h +1 -4
index 8dbb896ae..a78c97f91 100644
@@ -107,10 +107,7 @@ static bool pci_check_and_mask_intx(struct pci_dev *pdev)
                uint16_t old, new;
 
                old = status;
-               if (status != 0)
-                       new = old & (~PCI_COMMAND_INTX_DISABLE);
-               else
-                       new = old | PCI_COMMAND_INTX_DISABLE;
+               new = old | PCI_COMMAND_INTX_DISABLE;
 
                if (old != new)
                        pci_write_config_word(pdev, PCI_COMMAND, new);

Please correct me if I was wrong.

Thanks,
Xiaolong

#9 lore raw

Hi, Stephen

Quoted 183 lines · Ye Xiaolong
Quoted 92 lines · Stephen Hemminger
Quoted 89 lines · On Wed, 16 Jan 2019 15:48:36 +0800 Ye Xiaolong <[email protected]> wrote:

Hi, Stephen

Quoted 29 lines · Stephen Hemminger
Quoted 13 lines · On Wed, 16 Jan 2019 08:34:52 +0800 Xiaolong Ye <[email protected]> wrote:

The comment for igbuio_pci_irqhandler is out of date as the code evolves, remove it to avoid misleading.

---
 kernel/linux/igb_uio/igb_uio.c | 4 ----
 1 file changed, 4 deletions(-)
kernel/linux/igb_uio/igb_uio.c +0 -0
index 3cf394bdf..d6ac51e79 100644

kernel/linux/igb_uio/igb_uio.c igbuio_pci_irqcontrol(struct uio_info *info, s32 irq_state) · line 190

Quoted 10 lines
kernel/linux/igb_uio/igb_uio.c +0 -4
@@ -185,10 +185,6 @@ igbuio_pci_irqcontrol(struct uio_info *info, s32 irq_state)
 	return 0;
 }
 
-/**
- * This is interrupt handler which will check if the interrupt is for the right device.
- * If yes, disable it here and will be enable later.
- */
 static irqreturn_t
 igbuio_pci_irqhandler(int irq, void *dev_id)
 {  

The comment is partially correct; if you look at the legacy case.

Quoted 3 lines

Maybe better to move the comment to pci_check_and_mask_intx in compat.h? I see there is another incorrect comment there.

As I tried to understand pci_check_and_mask_intx behavior, I noticed that it was introduced by commit 399a3f0d ("igb_uio: fix IRQ mode handling"), and it was derived from igbuio_set_interrupt_mask, however there were two parameters in igbuio_set_interrupt_mask as below:

static int
igbuio_set_interrupt_mask(struct rte_uio_pci_dev *udev, int32_t state)

while the pci_check_and_mask_intx only has 1 parameter,

static bool pci_check_and_mask_intx(struct pci_dev *pdev)

but in its function body it still use the "state" accorrding to commit 399a3f0d

(quoted diff) +24 -6
+static bool pci_check_and_mask_intx(struct pci_dev *pdev)
 {
-       /* Some function names changes between 3.2.0 and 3.3.0... */
-#if LINUX_VERSION_CODE < KERNEL_VERSION(3, 3, 0)
-       pci_unblock_user_cfg_access(pdev);
-#else
-       pci_cfg_access_unlock(pdev);
-#endif
+       bool pending;
+       uint32_t status;
+
+       pci_block_user_cfg_access(dev);
+       pci_read_config_dword(pdev, PCI_COMMAND, &status);
+
+       /* interrupt is not ours, goes to out */
+       pending = (((status >> 16) & PCI_STATUS_INTERRUPT) != 0);
+       if (pending) {
+               uint16_t old, new;
+
+               old = status;
+               if (state != 0) <=========================== state still in use
+                       new = old & (~PCI_COMMAND_INTX_DISABLE);
+               else
+                       new = old | PCI_COMMAND_INTX_DISABLE;
+
+               if (old != new)
+                       pci_write_config_word(pdev, PCI_COMMAND, new);
+       }
+       pci_unblock_user_cfg_access(dev);
+
+       return pending;

and later this was fixed as a typo in commit 5b2f8137 ("igb_uio: fix typos for kernel older than 3.3") which seems not correct.

                old = status;
(quoted diff) +1 -1
-               if (state != 0)
+               if (status != 0)
Quoted 8 lines
Quoted 6 lines

I feel like that pci_check_and_mask_intx still needs two parameters from code point of view, please correct me if I was wrong or miss anything, or you think it's sensible, I'll cook a patch for it.

Thanks,
Xiaolong

This code is copy of upstream kernel function, you should look at that.

Thanks, just checked the kernel version of pci_check_and_mask_intx, and it is a wrap of pci_check_and_set_intx_mask(pdev, true).

	bool pci_check_and_mask_intx(struct pci_dev *dev)
	{
		return pci_check_and_set_intx_mask(dev, true);
	}


	static bool pci_check_and_set_intx_mask(struct pci_dev *dev, bool mask)
	{
		struct pci_bus *bus = dev->bus;
		...
		origcmd = cmd_status_dword;
		newcmd = origcmd & ~PCI_COMMAND_INTX_DISABLE;
		if (mask)
			newcmd |= PCI_COMMAND_INTX_DISABLE;
		if (newcmd != origcmd)
			bus->ops->write(bus, dev->devfn, PCI_COMMAND, 2, newcmd);

	done:
		raw_spin_unlock_irqrestore(&pci_lock, flags);

		return mask_updated;
	}

if mask is set to ture, PCI_COMMAND_INTX_DISABLE will be set in newcmd.

And current dpdk version doesn't align with it:

	static bool pci_check_and_mask_intx(struct pci_dev *pdev)
	{
		bool pending;
		uint32_t status;

		pci_block_user_cfg_access(pdev);
		pci_read_config_dword(pdev, PCI_COMMAND, &status);

		/* interrupt is not ours, goes to out */
		pending = (((status >> 16) & PCI_STATUS_INTERRUPT) != 0);
		if (pending) {
			uint16_t old, new;

			old = status;
			if (status != 0)
				new = old & (~PCI_COMMAND_INTX_DISABLE);
			else
				new = old | PCI_COMMAND_INTX_DISABLE;

			if (old != new)
				pci_write_config_word(pdev, PCI_COMMAND, new);
		}
		pci_unblock_user_cfg_access(pdev);

		return pending;
	}

In this code, if the interrupt is triggered by this device, both pending and status will not be 0, so the PCI_COMMAND_INTX_DISABLE will be cleared for new.

To align the behavior with upstream kernel, we may need change like:

kernel/linux/igb_uio/compat.h +1 -4
index 8dbb896ae..a78c97f91 100644
@@ -107,10 +107,7 @@ static bool pci_check_and_mask_intx(struct pci_dev *pdev)
               uint16_t old, new;
 
               old = status;
-               if (status != 0)
-                       new = old & (~PCI_COMMAND_INTX_DISABLE);
-               else
-                       new = old | PCI_COMMAND_INTX_DISABLE;
+               new = old | PCI_COMMAND_INTX_DISABLE;
 
               if (old != new)
                       pci_write_config_word(pdev, PCI_COMMAND, new);
Quoted 1 line

Please correct me if I was wrong.

Does my above fix make any sense?

Thanks,
Xiaolong

Quoted 3 lines

Thanks,
Xiaolong

#10 lore raw
In reply to Ye Xiaolong
Quoted 183 lines · Ye Xiaolong
Quoted 92 lines · Stephen Hemminger
Quoted 89 lines · On Wed, 16 Jan 2019 15:48:36 +0800 Ye Xiaolong <[email protected]> wrote:

Hi, Stephen

Quoted 29 lines · Stephen Hemminger
Quoted 13 lines · On Wed, 16 Jan 2019 08:34:52 +0800 Xiaolong Ye <[email protected]> wrote:

The comment for igbuio_pci_irqhandler is out of date as the code evolves, remove it to avoid misleading.

---
 kernel/linux/igb_uio/igb_uio.c | 4 ----
 1 file changed, 4 deletions(-)
kernel/linux/igb_uio/igb_uio.c +0 -0
index 3cf394bdf..d6ac51e79 100644

kernel/linux/igb_uio/igb_uio.c igbuio_pci_irqcontrol(struct uio_info *info, s32 irq_state) · line 190

Quoted 10 lines
kernel/linux/igb_uio/igb_uio.c +0 -4
@@ -185,10 +185,6 @@ igbuio_pci_irqcontrol(struct uio_info *info, s32 irq_state)
 	return 0;
 }
 
-/**
- * This is interrupt handler which will check if the interrupt is for the right device.
- * If yes, disable it here and will be enable later.
- */
 static irqreturn_t
 igbuio_pci_irqhandler(int irq, void *dev_id)
 {  

The comment is partially correct; if you look at the legacy case.

Quoted 3 lines

Maybe better to move the comment to pci_check_and_mask_intx in compat.h? I see there is another incorrect comment there.

As I tried to understand pci_check_and_mask_intx behavior, I noticed that it was introduced by commit 399a3f0d ("igb_uio: fix IRQ mode handling"), and it was derived from igbuio_set_interrupt_mask, however there were two parameters in igbuio_set_interrupt_mask as below:

static int
igbuio_set_interrupt_mask(struct rte_uio_pci_dev *udev, int32_t state)

while the pci_check_and_mask_intx only has 1 parameter,

static bool pci_check_and_mask_intx(struct pci_dev *pdev)

but in its function body it still use the "state" accorrding to commit 399a3f0d

(quoted diff) +24 -6
+static bool pci_check_and_mask_intx(struct pci_dev *pdev)
 {
-       /* Some function names changes between 3.2.0 and 3.3.0... */
-#if LINUX_VERSION_CODE < KERNEL_VERSION(3, 3, 0)
-       pci_unblock_user_cfg_access(pdev);
-#else
-       pci_cfg_access_unlock(pdev);
-#endif
+       bool pending;
+       uint32_t status;
+
+       pci_block_user_cfg_access(dev);
+       pci_read_config_dword(pdev, PCI_COMMAND, &status);
+
+       /* interrupt is not ours, goes to out */
+       pending = (((status >> 16) & PCI_STATUS_INTERRUPT) != 0);
+       if (pending) {
+               uint16_t old, new;
+
+               old = status;
+               if (state != 0) <=========================== state still in use
+                       new = old & (~PCI_COMMAND_INTX_DISABLE);
+               else
+                       new = old | PCI_COMMAND_INTX_DISABLE;
+
+               if (old != new)
+                       pci_write_config_word(pdev, PCI_COMMAND, new);
+       }
+       pci_unblock_user_cfg_access(dev);
+
+       return pending;

and later this was fixed as a typo in commit 5b2f8137 ("igb_uio: fix typos for kernel older than 3.3") which seems not correct.

                old = status;
(quoted diff) +1 -1
-               if (state != 0)
+               if (status != 0)
Quoted 8 lines
Quoted 6 lines

I feel like that pci_check_and_mask_intx still needs two parameters from code point of view, please correct me if I was wrong or miss anything, or you think it's sensible, I'll cook a patch for it.

Thanks,
Xiaolong

This code is copy of upstream kernel function, you should look at that.

Thanks, just checked the kernel version of pci_check_and_mask_intx, and it is a wrap of pci_check_and_set_intx_mask(pdev, true).

	bool pci_check_and_mask_intx(struct pci_dev *dev)
	{
		return pci_check_and_set_intx_mask(dev, true);
	}


	static bool pci_check_and_set_intx_mask(struct pci_dev *dev, bool mask)
	{
		struct pci_bus *bus = dev->bus;
		...
		origcmd = cmd_status_dword;
		newcmd = origcmd & ~PCI_COMMAND_INTX_DISABLE;
		if (mask)
			newcmd |= PCI_COMMAND_INTX_DISABLE;
		if (newcmd != origcmd)
			bus->ops->write(bus, dev->devfn, PCI_COMMAND, 2, newcmd);

	done:
		raw_spin_unlock_irqrestore(&pci_lock, flags);

		return mask_updated;
	}

if mask is set to ture, PCI_COMMAND_INTX_DISABLE will be set in newcmd.

And current dpdk version doesn't align with it:

	static bool pci_check_and_mask_intx(struct pci_dev *pdev)
	{
		bool pending;
		uint32_t status;

		pci_block_user_cfg_access(pdev);
		pci_read_config_dword(pdev, PCI_COMMAND, &status);

		/* interrupt is not ours, goes to out */
		pending = (((status >> 16) & PCI_STATUS_INTERRUPT) != 0);
		if (pending) {
			uint16_t old, new;

			old = status;
			if (status != 0)
				new = old & (~PCI_COMMAND_INTX_DISABLE);
			else
				new = old | PCI_COMMAND_INTX_DISABLE;

			if (old != new)
				pci_write_config_word(pdev, PCI_COMMAND, new);
		}
		pci_unblock_user_cfg_access(pdev);

		return pending;
	}

In this code, if the interrupt is triggered by this device, both pending and status will not be 0, so the PCI_COMMAND_INTX_DISABLE will be cleared for new.

To align the behavior with upstream kernel, we may need change like:

kernel/linux/igb_uio/compat.h +1 -4
index 8dbb896ae..a78c97f91 100644
@@ -107,10 +107,7 @@ static bool pci_check_and_mask_intx(struct pci_dev *pdev)
                uint16_t old, new;
 
                old = status;
-               if (status != 0)
-                       new = old & (~PCI_COMMAND_INTX_DISABLE);
-               else
-                       new = old | PCI_COMMAND_INTX_DISABLE;
+               new = old | PCI_COMMAND_INTX_DISABLE;
 
                if (old != new)
                        pci_write_config_word(pdev, PCI_COMMAND, new);
Quoted 1 line

Please correct me if I was wrong.

Hi Xiaolong,

Your analysis looks correct, thanks for history provided, it seems renaming "state" to "status" was not correct [1].

The intention of the 'pci_check_and_mask_intx()' is mask the interrupt if it is from this device, so your suggestion looks correct to me.

Only concern is how we can test the change, most probably existing code is also failing but able to find its way since it is not tested fully. To be sure we are also not missing anything, I think better to test the code before getting it. Are you aware of any way to test updated code?

And related to the initial patch, removing the comment, I think the comment is informative within the context of 'igbuio_pci_irqhandler()', so I suggest moving if from begging of the function into it just above legacy interrupt change.

[1]
5b2f8137 ("igb_uio: fix typos for kernel older than 3.3")

Thanks,
ferruh

#11 lore raw

Hi, Ferruh

On 01/29, Ferruh Yigit wrote:
[snip]

Quoted 6 lines

Hi Xiaolong,

Your analysis looks correct, thanks for history provided, it seems renaming "state" to "status" was not correct [1].

Quoted 8 lines

The intention of the 'pci_check_and_mask_intx()' is mask the interrupt if it is from this device, so your suggestion looks correct to me.

Only concern is how we can test the change, most probably existing code is also failing but able to find its way since it is not tested fully. To be sure we are also not missing anything, I think better to test the code before getting it. Are you aware of any way to test updated code?

Yes, I am also thinking about how to verify the change, will update once I find a proper method.

Quoted 5 lines

And related to the initial patch, removing the comment, I think the comment is informative within the context of 'igbuio_pci_irqhandler()', so I suggest moving if from begging of the function into it just above legacy interrupt change.

Sure, will do.

Thanks,
Xiaolong

Quoted 6 lines

[1]
5b2f8137 ("igb_uio: fix typos for kernel older than 3.3")

Thanks,
ferruh