[PATCH] bus/pci: fix secondary process PCI uio resource map problem

[PATCH] bus/pci: fix secondary process PCI uio resource map problem

[Chaoyong He]

From: Zerun Fu <zerun.fu@corigine.com>

For the primary process, the logic loops all BARs and will skip
the map of BAR with an invalid physical address (0), also will
assign 'uio_res->nb_maps' with the real mapped BARs number. But
for the secondary process, instead of loops all BARs, the logic
using the 'uio_res->nb_map' as index. If the device uses continuous
BARs there will be no problem, whereas if it uses discrete BARs,
it will lead to mapping errors.

Fix this problem by also loops all BARs and skip the map of BAR
with an invalid physical address in secondary process.

Fixes: 9b957f378abf ("pci: merge uio functions for linux and bsd")
Cc: mukawa@igel.co.jp
Cc: stable@dpdk.org

Signed-off-by: Zerun Fu <zerun.fu@corigine.com>
Reviewed-by: Chaoyong He <chaoyong.he@corigine.com>
Reviewed-by: Long Wu <long.wu@corigine.com>
Reviewed-by: Peng Zhang <peng.zhang@corigine.com>
---
 drivers/bus/pci/pci_common_uio.c | 94 ++++++++++++++++++++------------
 1 file changed, 60 insertions(+), 34 deletions(-)

diff --git a/drivers/bus/pci/pci_common_uio.c b/drivers/bus/pci/pci_common_uio.c
index 76c661f054..fcd8a49daf 100644
--- a/drivers/bus/pci/pci_common_uio.c
+++ b/drivers/bus/pci/pci_common_uio.c
@@ -23,10 +23,57 @@ static struct rte_tailq_elem rte_uio_tailq = {
 };
 EAL_REGISTER_TAILQ(rte_uio_tailq)
 
+static int
+pci_uio_map_secondary_resource_by_index(struct rte_pci_device *dev,
+		int res_idx, struct mapped_pci_resource *uio_res, int map_idx)
+{
+	int fd, i;
+
+	if (map_idx >= uio_res->nb_maps)
+		return -1;
+
+	/*
+	 * open devname, to mmap it
+	 */
+	fd = open(uio_res->maps[map_idx].path, O_RDWR);
+	if (fd < 0) {
+		RTE_LOG(ERR, EAL, "Cannot open %s: %s\n",
+			uio_res->maps[map_idx].path, strerror(errno));
+		return -1;
+	}
+
+	void *mapaddr = pci_map_resource(uio_res->maps[map_idx].addr,
+			fd, (off_t)uio_res->maps[map_idx].offset,
+			(size_t)uio_res->maps[map_idx].size, 0);
+
+	/* fd is not needed in secondary process, close it */
+	close(fd);
+	if (mapaddr != uio_res->maps[map_idx].addr) {
+		RTE_LOG(ERR, EAL,
+			"Cannot mmap device resource file %s to address: %p\n",
+			uio_res->maps[map_idx].path,
+			uio_res->maps[map_idx].addr);
+		if (mapaddr != NULL) {
+			/* unmap addrs correctly mapped */
+			for (i = 0; i < map_idx; i++)
+				pci_unmap_resource(
+					uio_res->maps[i].addr,
+					(size_t)uio_res->maps[i].size);
+			/* unmap addr wrongly mapped */
+			pci_unmap_resource(mapaddr,
+				(size_t)uio_res->maps[map_idx].size);
+		}
+		return -1;
+	}
+	dev->mem_resource[res_idx].addr = mapaddr;
+
+	return 0;
+}
+
 static int
 pci_uio_map_secondary(struct rte_pci_device *dev)
 {
-	int fd, i, j;
+	int map_idx = 0, res_idx, ret;
 	struct mapped_pci_resource *uio_res;
 	struct mapped_pci_res_list *uio_res_list =
 			RTE_TAILQ_CAST(rte_uio_tailq.head, mapped_pci_res_list);
@@ -37,41 +84,20 @@ pci_uio_map_secondary(struct rte_pci_device *dev)
 		if (rte_pci_addr_cmp(&uio_res->pci_addr, &dev->addr))
 			continue;
 
-		for (i = 0; i != uio_res->nb_maps; i++) {
-			/*
-			 * open devname, to mmap it
-			 */
-			fd = open(uio_res->maps[i].path, O_RDWR);
-			if (fd < 0) {
-				RTE_LOG(ERR, EAL, "Cannot open %s: %s\n",
-					uio_res->maps[i].path, strerror(errno));
-				return -1;
+		/* Map all BARs */
+		for (res_idx = 0; res_idx != PCI_MAX_RESOURCE; res_idx++) {
+			 /* skip empty BAR */
+			if (dev->mem_resource[res_idx].phys_addr == 0)
+				continue;
+
+			ret = pci_uio_map_secondary_resource_by_index(dev, res_idx,
+					uio_res, map_idx);
+			if (ret < 0) {
+				RTE_LOG(ERR, EAL, "Failed to map resources\n");
+				return ret;
 			}
 
-			void *mapaddr = pci_map_resource(uio_res->maps[i].addr,
-					fd, (off_t)uio_res->maps[i].offset,
-					(size_t)uio_res->maps[i].size, 0);
-
-			/* fd is not needed in secondary process, close it */
-			close(fd);
-			if (mapaddr != uio_res->maps[i].addr) {
-				RTE_LOG(ERR, EAL,
-					"Cannot mmap device resource file %s to address: %p\n",
-					uio_res->maps[i].path,
-					uio_res->maps[i].addr);
-				if (mapaddr != NULL) {
-					/* unmap addrs correctly mapped */
-					for (j = 0; j < i; j++)
-						pci_unmap_resource(
-							uio_res->maps[j].addr,
-							(size_t)uio_res->maps[j].size);
-					/* unmap addr wrongly mapped */
-					pci_unmap_resource(mapaddr,
-						(size_t)uio_res->maps[i].size);
-				}
-				return -1;
-			}
-			dev->mem_resource[i].addr = mapaddr;
+			map_idx++;
 		}
 		return 0;
 	}
-- 
2.39.1

[PATCH v2 1/2] bus/pci: fix secondary process PCI uio resource map problem

[Chaoyong He]

From: Zerun Fu <zerun.fu@corigine.com>

For the primary process, the logic loops all BARs and will skip
the map of BAR with an invalid physical address (0), also will
assign 'uio_res->nb_maps' with the real mapped BARs number. But
for the secondary process, instead of loops all BARs, the logic
using the 'uio_res->nb_map' as index. If the device uses continuous
BARs there will be no problem, whereas if it uses discrete BARs,
it will lead to mapping errors.

Fix this problem by also loops all BARs and skip the map of BAR
with an invalid physical address in secondary process.

Fixes: 9b957f378abf ("pci: merge uio functions for linux and bsd")
Cc: mukawa@igel.co.jp
Cc: stable@dpdk.org

Signed-off-by: Zerun Fu <zerun.fu@corigine.com>
Reviewed-by: Chaoyong He <chaoyong.he@corigine.com>
Reviewed-by: Long Wu <long.wu@corigine.com>
Reviewed-by: Peng Zhang <peng.zhang@corigine.com>
---
 drivers/bus/pci/pci_common_uio.c | 94 ++++++++++++++++++++------------
 1 file changed, 60 insertions(+), 34 deletions(-)

diff --git a/drivers/bus/pci/pci_common_uio.c b/drivers/bus/pci/pci_common_uio.c
index 76c661f054..fcd8a49daf 100644
--- a/drivers/bus/pci/pci_common_uio.c
+++ b/drivers/bus/pci/pci_common_uio.c
@@ -23,10 +23,57 @@ static struct rte_tailq_elem rte_uio_tailq = {
 };
 EAL_REGISTER_TAILQ(rte_uio_tailq)
 
+static int
+pci_uio_map_secondary_resource_by_index(struct rte_pci_device *dev,
+		int res_idx, struct mapped_pci_resource *uio_res, int map_idx)
+{
+	int fd, i;
+
+	if (map_idx >= uio_res->nb_maps)
+		return -1;
+
+	/*
+	 * open devname, to mmap it
+	 */
+	fd = open(uio_res->maps[map_idx].path, O_RDWR);
+	if (fd < 0) {
+		RTE_LOG(ERR, EAL, "Cannot open %s: %s\n",
+			uio_res->maps[map_idx].path, strerror(errno));
+		return -1;
+	}
+
+	void *mapaddr = pci_map_resource(uio_res->maps[map_idx].addr,
+			fd, (off_t)uio_res->maps[map_idx].offset,
+			(size_t)uio_res->maps[map_idx].size, 0);
+
+	/* fd is not needed in secondary process, close it */
+	close(fd);
+	if (mapaddr != uio_res->maps[map_idx].addr) {
+		RTE_LOG(ERR, EAL,
+			"Cannot mmap device resource file %s to address: %p\n",
+			uio_res->maps[map_idx].path,
+			uio_res->maps[map_idx].addr);
+		if (mapaddr != NULL) {
+			/* unmap addrs correctly mapped */
+			for (i = 0; i < map_idx; i++)
+				pci_unmap_resource(
+					uio_res->maps[i].addr,
+					(size_t)uio_res->maps[i].size);
+			/* unmap addr wrongly mapped */
+			pci_unmap_resource(mapaddr,
+				(size_t)uio_res->maps[map_idx].size);
+		}
+		return -1;
+	}
+	dev->mem_resource[res_idx].addr = mapaddr;
+
+	return 0;
+}
+
 static int
 pci_uio_map_secondary(struct rte_pci_device *dev)
 {
-	int fd, i, j;
+	int map_idx = 0, res_idx, ret;
 	struct mapped_pci_resource *uio_res;
 	struct mapped_pci_res_list *uio_res_list =
 			RTE_TAILQ_CAST(rte_uio_tailq.head, mapped_pci_res_list);
@@ -37,41 +84,20 @@ pci_uio_map_secondary(struct rte_pci_device *dev)
 		if (rte_pci_addr_cmp(&uio_res->pci_addr, &dev->addr))
 			continue;
 
-		for (i = 0; i != uio_res->nb_maps; i++) {
-			/*
-			 * open devname, to mmap it
-			 */
-			fd = open(uio_res->maps[i].path, O_RDWR);
-			if (fd < 0) {
-				RTE_LOG(ERR, EAL, "Cannot open %s: %s\n",
-					uio_res->maps[i].path, strerror(errno));
-				return -1;
+		/* Map all BARs */
+		for (res_idx = 0; res_idx != PCI_MAX_RESOURCE; res_idx++) {
+			 /* skip empty BAR */
+			if (dev->mem_resource[res_idx].phys_addr == 0)
+				continue;
+
+			ret = pci_uio_map_secondary_resource_by_index(dev, res_idx,
+					uio_res, map_idx);
+			if (ret < 0) {
+				RTE_LOG(ERR, EAL, "Failed to map resources\n");
+				return ret;
 			}
 
-			void *mapaddr = pci_map_resource(uio_res->maps[i].addr,
-					fd, (off_t)uio_res->maps[i].offset,
-					(size_t)uio_res->maps[i].size, 0);
-
-			/* fd is not needed in secondary process, close it */
-			close(fd);
-			if (mapaddr != uio_res->maps[i].addr) {
-				RTE_LOG(ERR, EAL,
-					"Cannot mmap device resource file %s to address: %p\n",
-					uio_res->maps[i].path,
-					uio_res->maps[i].addr);
-				if (mapaddr != NULL) {
-					/* unmap addrs correctly mapped */
-					for (j = 0; j < i; j++)
-						pci_unmap_resource(
-							uio_res->maps[j].addr,
-							(size_t)uio_res->maps[j].size);
-					/* unmap addr wrongly mapped */
-					pci_unmap_resource(mapaddr,
-						(size_t)uio_res->maps[i].size);
-				}
-				return -1;
-			}
-			dev->mem_resource[i].addr = mapaddr;
+			map_idx++;
 		}
 		return 0;
 	}
-- 
2.39.1

Re: [PATCH v2 1/2] bus/pci: fix secondary process PCI uio resource map problem

[fengchengwen]

Hi,

Nice to see this fix.
ps: our team also founds this problem, and the bugfix is still being reviewed internally.

As for this patch, I prefer not extract subfunction, because the father is simple and just
need do some minor adjustment.

		map_idx = 0;
		for (i = 0; i != PCI_MAX_RESOURCE; i++) {
			/* skip empty BAR */
			if (dev->mem_resource[i].phys_addr == 0)
				continue;
			... // use map_idx instead i
			map_idx++;
			dev->mem_resource[i].addr = mapaddr;
		}

Also suggest use pci_uio_find_resource() to refactor pci_uio_map_secondary() function.

Thanks

On 2024/1/29 17:22, Chaoyong He wrote:
> From: Zerun Fu <zerun.fu@corigine.com>
> 
> For the primary process, the logic loops all BARs and will skip
> the map of BAR with an invalid physical address (0), also will
> assign 'uio_res->nb_maps' with the real mapped BARs number. But
> for the secondary process, instead of loops all BARs, the logic
> using the 'uio_res->nb_map' as index. If the device uses continuous
> BARs there will be no problem, whereas if it uses discrete BARs,
> it will lead to mapping errors.
> 
> Fix this problem by also loops all BARs and skip the map of BAR
> with an invalid physical address in secondary process.
> 
> Fixes: 9b957f378abf ("pci: merge uio functions for linux and bsd")
> Cc: mukawa@igel.co.jp
> Cc: stable@dpdk.org
> 
> Signed-off-by: Zerun Fu <zerun.fu@corigine.com>
> Reviewed-by: Chaoyong He <chaoyong.he@corigine.com>
> Reviewed-by: Long Wu <long.wu@corigine.com>
> Reviewed-by: Peng Zhang <peng.zhang@corigine.com>
> ---
>  drivers/bus/pci/pci_common_uio.c | 94 ++++++++++++++++++++------------
>  1 file changed, 60 insertions(+), 34 deletions(-)
> 
> diff --git a/drivers/bus/pci/pci_common_uio.c b/drivers/bus/pci/pci_common_uio.c
> index 76c661f054..fcd8a49daf 100644
> --- a/drivers/bus/pci/pci_common_uio.c
> +++ b/drivers/bus/pci/pci_common_uio.c
> @@ -23,10 +23,57 @@ static struct rte_tailq_elem rte_uio_tailq = {
>  };
>  EAL_REGISTER_TAILQ(rte_uio_tailq)
>  
> +static int
> +pci_uio_map_secondary_resource_by_index(struct rte_pci_device *dev,
> +		int res_idx, struct mapped_pci_resource *uio_res, int map_idx)
> +{
> +	int fd, i;
> +
> +	if (map_idx >= uio_res->nb_maps)
> +		return -1;
> +
> +	/*
> +	 * open devname, to mmap it
> +	 */
> +	fd = open(uio_res->maps[map_idx].path, O_RDWR);
> +	if (fd < 0) {
> +		RTE_LOG(ERR, EAL, "Cannot open %s: %s\n",
> +			uio_res->maps[map_idx].path, strerror(errno));
> +		return -1;
> +	}
> +
> +	void *mapaddr = pci_map_resource(uio_res->maps[map_idx].addr,
> +			fd, (off_t)uio_res->maps[map_idx].offset,
> +			(size_t)uio_res->maps[map_idx].size, 0);
> +
> +	/* fd is not needed in secondary process, close it */
> +	close(fd);
> +	if (mapaddr != uio_res->maps[map_idx].addr) {
> +		RTE_LOG(ERR, EAL,
> +			"Cannot mmap device resource file %s to address: %p\n",
> +			uio_res->maps[map_idx].path,
> +			uio_res->maps[map_idx].addr);
> +		if (mapaddr != NULL) {
> +			/* unmap addrs correctly mapped */
> +			for (i = 0; i < map_idx; i++)
> +				pci_unmap_resource(
> +					uio_res->maps[i].addr,
> +					(size_t)uio_res->maps[i].size);
> +			/* unmap addr wrongly mapped */
> +			pci_unmap_resource(mapaddr,
> +				(size_t)uio_res->maps[map_idx].size);
> +		}
> +		return -1;
> +	}
> +	dev->mem_resource[res_idx].addr = mapaddr;
> +
> +	return 0;
> +}
> +
>  static int
>  pci_uio_map_secondary(struct rte_pci_device *dev)
>  {
> -	int fd, i, j;
> +	int map_idx = 0, res_idx, ret;
>  	struct mapped_pci_resource *uio_res;
>  	struct mapped_pci_res_list *uio_res_list =
>  			RTE_TAILQ_CAST(rte_uio_tailq.head, mapped_pci_res_list);
> @@ -37,41 +84,20 @@ pci_uio_map_secondary(struct rte_pci_device *dev)
>  		if (rte_pci_addr_cmp(&uio_res->pci_addr, &dev->addr))
>  			continue;
>  
> -		for (i = 0; i != uio_res->nb_maps; i++) {
> -			/*
> -			 * open devname, to mmap it
> -			 */
> -			fd = open(uio_res->maps[i].path, O_RDWR);
> -			if (fd < 0) {
> -				RTE_LOG(ERR, EAL, "Cannot open %s: %s\n",
> -					uio_res->maps[i].path, strerror(errno));
> -				return -1;
> +		/* Map all BARs */
> +		for (res_idx = 0; res_idx != PCI_MAX_RESOURCE; res_idx++) {
> +			 /* skip empty BAR */
> +			if (dev->mem_resource[res_idx].phys_addr == 0)
> +				continue;
> +
> +			ret = pci_uio_map_secondary_resource_by_index(dev, res_idx,
> +					uio_res, map_idx);
> +			if (ret < 0) {
> +				RTE_LOG(ERR, EAL, "Failed to map resources\n");
> +				return ret;
>  			}
>  
> -			void *mapaddr = pci_map_resource(uio_res->maps[i].addr,
> -					fd, (off_t)uio_res->maps[i].offset,
> -					(size_t)uio_res->maps[i].size, 0);
> -
> -			/* fd is not needed in secondary process, close it */
> -			close(fd);
> -			if (mapaddr != uio_res->maps[i].addr) {
> -				RTE_LOG(ERR, EAL,
> -					"Cannot mmap device resource file %s to address: %p\n",
> -					uio_res->maps[i].path,
> -					uio_res->maps[i].addr);
> -				if (mapaddr != NULL) {
> -					/* unmap addrs correctly mapped */
> -					for (j = 0; j < i; j++)
> -						pci_unmap_resource(
> -							uio_res->maps[j].addr,
> -							(size_t)uio_res->maps[j].size);
> -					/* unmap addr wrongly mapped */
> -					pci_unmap_resource(mapaddr,
> -						(size_t)uio_res->maps[i].size);
> -				}
> -				return -1;
> -			}
> -			dev->mem_resource[i].addr = mapaddr;
> +			map_idx++;
>  		}
>  		return 0;
>  	}
> 

Re: [PATCH v2 1/2] bus/pci: fix secondary process PCI uio resource map problem

[huangdengdui]

On 2024/1/29 17:22, Chaoyong He wrote:
> From: Zerun Fu <zerun.fu@corigine.com>
> 
> For the primary process, the logic loops all BARs and will skip
> the map of BAR with an invalid physical address (0), also will
> assign 'uio_res->nb_maps' with the real mapped BARs number. But
> for the secondary process, instead of loops all BARs, the logic
> using the 'uio_res->nb_map' as index. If the device uses continuous
> BARs there will be no problem, whereas if it uses discrete BARs,
> it will lead to mapping errors.
> 
> Fix this problem by also loops all BARs and skip the map of BAR
> with an invalid physical address in secondary process.
> 
> Fixes: 9b957f378abf ("pci: merge uio functions for linux and bsd")Hi, Chaoyong
This patch is just a code refactoring, this bug is caused by the following patch:

Fixes: e6cf7bee1c77 ("bus/pci: fix UIO resource access from secondary process")

> Cc: mukawa@igel.co.jp
> Cc: stable@dpdk.org
> 
> Signed-off-by: Zerun Fu <zerun.fu@corigine.com>
> Reviewed-by: Chaoyong He <chaoyong.he@corigine.com>
> Reviewed-by: Long Wu <long.wu@corigine.com>
> Reviewed-by: Peng Zhang <peng.zhang@corigine.com>

test by the igb_uio driver on the ARM.
Tested-by: Dengdui Huang <huangdengdui@huawei.com>

> ---
>  drivers/bus/pci/pci_common_uio.c | 94 ++++++++++++++++++++------------
>  1 file changed, 60 insertions(+), 34 deletions(-)
> 
> diff --git a/drivers/bus/pci/pci_common_uio.c b/drivers/bus/pci/pci_common_uio.c
> index 76c661f054..fcd8a49daf 100644
> --- a/drivers/bus/pci/pci_common_uio.c
> +++ b/drivers/bus/pci/pci_common_uio.c
> @@ -23,10 +23,57 @@ static struct rte_tailq_elem rte_uio_tailq = {
>  };
>  EAL_REGISTER_TAILQ(rte_uio_tailq)
>  
> +static int
> +pci_uio_map_secondary_resource_by_index(struct rte_pci_device *dev,
> +		int res_idx, struct mapped_pci_resource *uio_res, int map_idx)
> +{
> +	int fd, i;
> +
> +	if (map_idx >= uio_res->nb_maps)
> +		return -1;
> +
> +	/*
> +	 * open devname, to mmap it
> +	 */
> +	fd = open(uio_res->maps[map_idx].path, O_RDWR);
> +	if (fd < 0) {
> +		RTE_LOG(ERR, EAL, "Cannot open %s: %s\n",
> +			uio_res->maps[map_idx].path, strerror(errno));
> +		return -1;
> +	}
> +
> +	void *mapaddr = pci_map_resource(uio_res->maps[map_idx].addr,
> +			fd, (off_t)uio_res->maps[map_idx].offset,
> +			(size_t)uio_res->maps[map_idx].size, 0);
> +
> +	/* fd is not needed in secondary process, close it */
> +	close(fd);
> +	if (mapaddr != uio_res->maps[map_idx].addr) {
> +		RTE_LOG(ERR, EAL,
> +			"Cannot mmap device resource file %s to address: %p\n",
> +			uio_res->maps[map_idx].path,
> +			uio_res->maps[map_idx].addr);
> +		if (mapaddr != NULL) {
> +			/* unmap addrs correctly mapped */
> +			for (i = 0; i < map_idx; i++)
> +				pci_unmap_resource(
> +					uio_res->maps[i].addr,
> +					(size_t)uio_res->maps[i].size);
> +			/* unmap addr wrongly mapped */
> +			pci_unmap_resource(mapaddr,
> +				(size_t)uio_res->maps[map_idx].size);
> +		}
When mapaddr == NULL, the addrs correctly mapped should also be unmaped
It may be better to make the following changes

for (i = 0; i < map_idx; i++)
	pci_unmap_resource(uio_res->maps[i].addr, (size_t)uio_res->maps[i].size);
if (mapaddr != NULL)
	pci_unmap_resource(mapaddr, (size_t)uio_res->maps[map_idx].size);


> +		return -1;
> +	}
> +	dev->mem_resource[res_idx].addr = mapaddr;
> +
> +	return 0;
> +}
> +
>  static int
>  pci_uio_map_secondary(struct rte_pci_device *dev)
>  {
> -	int fd, i, j;
> +	int map_idx = 0, res_idx, ret;
>  	struct mapped_pci_resource *uio_res;
>  	struct mapped_pci_res_list *uio_res_list =
>  			RTE_TAILQ_CAST(rte_uio_tailq.head, mapped_pci_res_list);
> @@ -37,41 +84,20 @@ pci_uio_map_secondary(struct rte_pci_device *dev)
>  		if (rte_pci_addr_cmp(&uio_res->pci_addr, &dev->addr))
>  			continue;
>  
> -		for (i = 0; i != uio_res->nb_maps; i++) {
> -			/*
> -			 * open devname, to mmap it
> -			 */
> -			fd = open(uio_res->maps[i].path, O_RDWR);
> -			if (fd < 0) {
> -				RTE_LOG(ERR, EAL, "Cannot open %s: %s\n",
> -					uio_res->maps[i].path, strerror(errno));
> -				return -1;
> +		/* Map all BARs */
> +		for (res_idx = 0; res_idx != PCI_MAX_RESOURCE; res_idx++) {
> +			 /* skip empty BAR */
> +			if (dev->mem_resource[res_idx].phys_addr == 0)
> +				continue;
> +
> +			ret = pci_uio_map_secondary_resource_by_index(dev, res_idx,
> +					uio_res, map_idx);
> +			if (ret < 0) {
> +				RTE_LOG(ERR, EAL, "Failed to map resources\n");
> +				return ret;
>  			}
>  
> -			void *mapaddr = pci_map_resource(uio_res->maps[i].addr,
> -					fd, (off_t)uio_res->maps[i].offset,
> -					(size_t)uio_res->maps[i].size, 0);
> -
> -			/* fd is not needed in secondary process, close it */
> -			close(fd);
> -			if (mapaddr != uio_res->maps[i].addr) {
> -				RTE_LOG(ERR, EAL,
> -					"Cannot mmap device resource file %s to address: %p\n",
> -					uio_res->maps[i].path,
> -					uio_res->maps[i].addr);
> -				if (mapaddr != NULL) {
> -					/* unmap addrs correctly mapped */
> -					for (j = 0; j < i; j++)
> -						pci_unmap_resource(
> -							uio_res->maps[j].addr,
> -							(size_t)uio_res->maps[j].size);
> -					/* unmap addr wrongly mapped */
> -					pci_unmap_resource(mapaddr,
> -						(size_t)uio_res->maps[i].size);
> -				}
> -				return -1;
> -			}
> -			dev->mem_resource[i].addr = mapaddr;
> +			map_idx++;
>  		}
>  		return 0;
>  	}

Re: [PATCH v2 1/2] bus/pci: fix secondary process PCI uio resource map problem

[Burakov, Anatoly]

On 1/29/2024 10:22 AM, Chaoyong He wrote:
> From: Zerun Fu <zerun.fu@corigine.com>
> 
> For the primary process, the logic loops all BARs and will skip
> the map of BAR with an invalid physical address (0), also will
> assign 'uio_res->nb_maps' with the real mapped BARs number. But
> for the secondary process, instead of loops all BARs, the logic
> using the 'uio_res->nb_map' as index. If the device uses continuous
> BARs there will be no problem, whereas if it uses discrete BARs,
> it will lead to mapping errors.
> 
> Fix this problem by also loops all BARs and skip the map of BAR
> with an invalid physical address in secondary process.
> 
> Fixes: 9b957f378abf ("pci: merge uio functions for linux and bsd")
> Cc: mukawa@igel.co.jp
> Cc: stable@dpdk.org
> 
> Signed-off-by: Zerun Fu <zerun.fu@corigine.com>
> Reviewed-by: Chaoyong He <chaoyong.he@corigine.com>
> Reviewed-by: Long Wu <long.wu@corigine.com>
> Reviewed-by: Peng Zhang <peng.zhang@corigine.com>
> ---
>   drivers/bus/pci/pci_common_uio.c | 94 ++++++++++++++++++++------------
>   1 file changed, 60 insertions(+), 34 deletions(-)
> 
> diff --git a/drivers/bus/pci/pci_common_uio.c b/drivers/bus/pci/pci_common_uio.c
> index 76c661f054..fcd8a49daf 100644
> --- a/drivers/bus/pci/pci_common_uio.c
> +++ b/drivers/bus/pci/pci_common_uio.c
> @@ -23,10 +23,57 @@ static struct rte_tailq_elem rte_uio_tailq = {
>   };
>   EAL_REGISTER_TAILQ(rte_uio_tailq)
>   
> +static int
> +pci_uio_map_secondary_resource_by_index(struct rte_pci_device *dev,
> +		int res_idx, struct mapped_pci_resource *uio_res, int map_idx)
> +{
> +	int fd, i;
> +
> +	if (map_idx >= uio_res->nb_maps)
> +		return -1;
> +
> +	/*
> +	 * open devname, to mmap it
> +	 */
> +	fd = open(uio_res->maps[map_idx].path, O_RDWR);
> +	if (fd < 0) {
> +		RTE_LOG(ERR, EAL, "Cannot open %s: %s\n",
> +			uio_res->maps[map_idx].path, strerror(errno));
> +		return -1;
> +	}
> +
> +	void *mapaddr = pci_map_resource(uio_res->maps[map_idx].addr,
> +			fd, (off_t)uio_res->maps[map_idx].offset,
> +			(size_t)uio_res->maps[map_idx].size, 0);
> +
> +	/* fd is not needed in secondary process, close it */
> +	close(fd);
> +	if (mapaddr != uio_res->maps[map_idx].addr) {
> +		RTE_LOG(ERR, EAL,
> +			"Cannot mmap device resource file %s to address: %p\n",
> +			uio_res->maps[map_idx].path,
> +			uio_res->maps[map_idx].addr);
> +		if (mapaddr != NULL) {
> +			/* unmap addrs correctly mapped */
> +			for (i = 0; i < map_idx; i++)
> +				pci_unmap_resource(
> +					uio_res->maps[i].addr,
> +					(size_t)uio_res->maps[i].size);
> +			/* unmap addr wrongly mapped */
> +			pci_unmap_resource(mapaddr,
> +				(size_t)uio_res->maps[map_idx].size);
> +		}
> +		return -1;
> +	}
> +	dev->mem_resource[res_idx].addr = mapaddr;
> +
> +	return 0;
> +}
> +
>   static int
>   pci_uio_map_secondary(struct rte_pci_device *dev)
>   {
> -	int fd, i, j;
> +	int map_idx = 0, res_idx, ret;
>   	struct mapped_pci_resource *uio_res;
>   	struct mapped_pci_res_list *uio_res_list =
>   			RTE_TAILQ_CAST(rte_uio_tailq.head, mapped_pci_res_list);
> @@ -37,41 +84,20 @@ pci_uio_map_secondary(struct rte_pci_device *dev)
>   		if (rte_pci_addr_cmp(&uio_res->pci_addr, &dev->addr))
>   			continue;
>   
> -		for (i = 0; i != uio_res->nb_maps; i++) {
> -			/*
> -			 * open devname, to mmap it
> -			 */
> -			fd = open(uio_res->maps[i].path, O_RDWR);
> -			if (fd < 0) {
> -				RTE_LOG(ERR, EAL, "Cannot open %s: %s\n",
> -					uio_res->maps[i].path, strerror(errno));
> -				return -1;
> +		/* Map all BARs */
> +		for (res_idx = 0; res_idx != PCI_MAX_RESOURCE; res_idx++) {
> +			 /* skip empty BAR */
> +			if (dev->mem_resource[res_idx].phys_addr == 0)
> +				continue;
> +
> +			ret = pci_uio_map_secondary_resource_by_index(dev, res_idx,
> +					uio_res, map_idx);
> +			if (ret < 0) {
> +				RTE_LOG(ERR, EAL, "Failed to map resources\n");
> +				return ret;
>   			}
>   
> -			void *mapaddr = pci_map_resource(uio_res->maps[i].addr,
> -					fd, (off_t)uio_res->maps[i].offset,
> -					(size_t)uio_res->maps[i].size, 0);
> -
> -			/* fd is not needed in secondary process, close it */
> -			close(fd);
> -			if (mapaddr != uio_res->maps[i].addr) {
> -				RTE_LOG(ERR, EAL,
> -					"Cannot mmap device resource file %s to address: %p\n",
> -					uio_res->maps[i].path,
> -					uio_res->maps[i].addr);
> -				if (mapaddr != NULL) {
> -					/* unmap addrs correctly mapped */
> -					for (j = 0; j < i; j++)
> -						pci_unmap_resource(
> -							uio_res->maps[j].addr,
> -							(size_t)uio_res->maps[j].size);
> -					/* unmap addr wrongly mapped */
> -					pci_unmap_resource(mapaddr,
> -						(size_t)uio_res->maps[i].size);

Nitpicking, but I feel like this would've been better done from the 
caller, not here. This is how it's done in primary process.

> -				}
> -				return -1;
> -			}
> -			dev->mem_resource[i].addr = mapaddr;
> +			map_idx++;
>   		}
>   		return 0;
>   	}

With fix above,

Acked-by: Anatoly Burakov <anatoly.burakov@intel.com>

-- 
Thanks,
Anatoly

[PATCH v2 2/2] bus/pci: fix secondary process save 'FD' problem

[Chaoyong He]

From: Zerun Fu <zerun.fu@corigine.com>

In the previous logic the 'fd' was only saved in the primary process,
but for some devices this value is also used in the secondary logic.

For example, the call of 'rte_pci_find_ext_capability()' will fail in
the secondary process.

Fix this problem by getting and saving the value of 'fd' also in the
secondary process logic.

Fixes: 9b957f378abf ("pci: merge uio functions for linux and bsd")
Cc: mukawa@igel.co.jp
Cc: stable@dpdk.org

Signed-off-by: Zerun Fu <zerun.fu@corigine.com>
Reviewed-by: Chaoyong He <chaoyong.he@corigine.com>
Reviewed-by: Long Wu <long.wu@corigine.com>
Reviewed-by: Peng Zhang <peng.zhang@corigine.com>
---
 drivers/bus/pci/linux/pci_uio.c  | 5 ++++-
 drivers/bus/pci/pci_common_uio.c | 8 ++++----
 2 files changed, 8 insertions(+), 5 deletions(-)

diff --git a/drivers/bus/pci/linux/pci_uio.c b/drivers/bus/pci/linux/pci_uio.c
index 97d740dfe5..6680e42efb 100644
--- a/drivers/bus/pci/linux/pci_uio.c
+++ b/drivers/bus/pci/linux/pci_uio.c
@@ -237,7 +237,7 @@ pci_uio_alloc_resource(struct rte_pci_device *dev,
 	}
 	snprintf(devname, sizeof(devname), "/dev/uio%u", uio_num);
 
-	/* save fd if in primary process */
+	/* save fd */
 	fd = open(devname, O_RDWR);
 	if (fd < 0) {
 		RTE_LOG(ERR, EAL, "Cannot open %s: %s\n",
@@ -261,6 +261,9 @@ pci_uio_alloc_resource(struct rte_pci_device *dev,
 	if (rte_intr_dev_fd_set(dev->intr_handle, uio_cfg_fd))
 		goto error;
 
+	if (rte_eal_process_type() != RTE_PROC_PRIMARY)
+		return 0;
+
 	if (dev->kdrv == RTE_PCI_KDRV_IGB_UIO) {
 		if (rte_intr_type_set(dev->intr_handle, RTE_INTR_HANDLE_UIO))
 			goto error;
diff --git a/drivers/bus/pci/pci_common_uio.c b/drivers/bus/pci/pci_common_uio.c
index fcd8a49daf..b6f79b067d 100644
--- a/drivers/bus/pci/pci_common_uio.c
+++ b/drivers/bus/pci/pci_common_uio.c
@@ -122,15 +122,15 @@ pci_uio_map_resource(struct rte_pci_device *dev)
 	if (rte_intr_dev_fd_set(dev->intr_handle, -1))
 		return -1;
 
-	/* secondary processes - use already recorded details */
-	if (rte_eal_process_type() != RTE_PROC_PRIMARY)
-		return pci_uio_map_secondary(dev);
-
 	/* allocate uio resource */
 	ret = pci_uio_alloc_resource(dev, &uio_res);
 	if (ret)
 		return ret;
 
+	/* secondary processes - use already recorded details */
+	if (rte_eal_process_type() != RTE_PROC_PRIMARY)
+		return pci_uio_map_secondary(dev);
+
 	/* Map all BARs */
 	for (i = 0; i != PCI_MAX_RESOURCE; i++) {
 		/* skip empty BAR */
-- 
2.39.1

Re: [PATCH v2 2/2] bus/pci: fix secondary process save 'FD' problem

[fengchengwen]

Hi,

The pci_uio_alloc_resource in drivers/bus/pci/bsd should also modified.

How about add subfunction: pci_uio_init_intr() which includes access/rte_intr_fd_set/rte_intr_fd_get/rte_intr_type_set?

Thanks

On 2024/1/29 17:22, Chaoyong He wrote:
> From: Zerun Fu <zerun.fu@corigine.com>
> 
> In the previous logic the 'fd' was only saved in the primary process,
> but for some devices this value is also used in the secondary logic.
> 
> For example, the call of 'rte_pci_find_ext_capability()' will fail in
> the secondary process.
> 
> Fix this problem by getting and saving the value of 'fd' also in the
> secondary process logic.
> 
> Fixes: 9b957f378abf ("pci: merge uio functions for linux and bsd")
> Cc: mukawa@igel.co.jp
> Cc: stable@dpdk.org
> 
> Signed-off-by: Zerun Fu <zerun.fu@corigine.com>
> Reviewed-by: Chaoyong He <chaoyong.he@corigine.com>
> Reviewed-by: Long Wu <long.wu@corigine.com>
> Reviewed-by: Peng Zhang <peng.zhang@corigine.com>
> ---
>  drivers/bus/pci/linux/pci_uio.c  | 5 ++++-
>  drivers/bus/pci/pci_common_uio.c | 8 ++++----
>  2 files changed, 8 insertions(+), 5 deletions(-)
> 
> diff --git a/drivers/bus/pci/linux/pci_uio.c b/drivers/bus/pci/linux/pci_uio.c
> index 97d740dfe5..6680e42efb 100644
> --- a/drivers/bus/pci/linux/pci_uio.c
> +++ b/drivers/bus/pci/linux/pci_uio.c
> @@ -237,7 +237,7 @@ pci_uio_alloc_resource(struct rte_pci_device *dev,
>  	}
>  	snprintf(devname, sizeof(devname), "/dev/uio%u", uio_num);
>  
> -	/* save fd if in primary process */
> +	/* save fd */
>  	fd = open(devname, O_RDWR);
>  	if (fd < 0) {
>  		RTE_LOG(ERR, EAL, "Cannot open %s: %s\n",
> @@ -261,6 +261,9 @@ pci_uio_alloc_resource(struct rte_pci_device *dev,
>  	if (rte_intr_dev_fd_set(dev->intr_handle, uio_cfg_fd))
>  		goto error;
>  
> +	if (rte_eal_process_type() != RTE_PROC_PRIMARY)
> +		return 0;
> +
>  	if (dev->kdrv == RTE_PCI_KDRV_IGB_UIO) {
>  		if (rte_intr_type_set(dev->intr_handle, RTE_INTR_HANDLE_UIO))
>  			goto error;
> diff --git a/drivers/bus/pci/pci_common_uio.c b/drivers/bus/pci/pci_common_uio.c
> index fcd8a49daf..b6f79b067d 100644
> --- a/drivers/bus/pci/pci_common_uio.c
> +++ b/drivers/bus/pci/pci_common_uio.c
> @@ -122,15 +122,15 @@ pci_uio_map_resource(struct rte_pci_device *dev)
>  	if (rte_intr_dev_fd_set(dev->intr_handle, -1))
>  		return -1;
>  
> -	/* secondary processes - use already recorded details */
> -	if (rte_eal_process_type() != RTE_PROC_PRIMARY)
> -		return pci_uio_map_secondary(dev);
> -
>  	/* allocate uio resource */
>  	ret = pci_uio_alloc_resource(dev, &uio_res);
>  	if (ret)
>  		return ret;
>  
> +	/* secondary processes - use already recorded details */
> +	if (rte_eal_process_type() != RTE_PROC_PRIMARY)
> +		return pci_uio_map_secondary(dev);
> +
>  	/* Map all BARs */
>  	for (i = 0; i != PCI_MAX_RESOURCE; i++) {
>  		/* skip empty BAR */
> 

Re: [PATCH v2 2/2] bus/pci: fix secondary process save 'FD' problem

[Burakov, Anatoly]

On 1/29/2024 10:22 AM, Chaoyong He wrote:
> From: Zerun Fu <zerun.fu@corigine.com>
> 
> In the previous logic the 'fd' was only saved in the primary process,
> but for some devices this value is also used in the secondary logic.
> 
> For example, the call of 'rte_pci_find_ext_capability()' will fail in
> the secondary process.
> 
> Fix this problem by getting and saving the value of 'fd' also in the
> secondary process logic.
> 
> Fixes: 9b957f378abf ("pci: merge uio functions for linux and bsd")
> Cc: mukawa@igel.co.jp
> Cc: stable@dpdk.org
> 
> Signed-off-by: Zerun Fu <zerun.fu@corigine.com>
> Reviewed-by: Chaoyong He <chaoyong.he@corigine.com>
> Reviewed-by: Long Wu <long.wu@corigine.com>
> Reviewed-by: Peng Zhang <peng.zhang@corigine.com>
> ---
Acked-by: Anatoly Burakov <anatoly.burakov@intel.com>

-- 
Thanks,
Anatoly

[PATCH v3 1/2] bus/pci: fix secondary process PCI uio resource map problem

[Chaoyong He]

From: Zerun Fu <zerun.fu@corigine.com>

For the primary process, the logic loops all BARs and will skip
the map of BAR with an invalid physical address (0), also will
assign 'uio_res->nb_maps' with the real mapped BARs number. But
for the secondary process, instead of loops all BARs, the logic
using the 'uio_res->nb_map' as index. If the device uses continuous
BARs there will be no problem, whereas if it uses discrete BARs,
it will lead to mapping errors.

Fix this problem by also loops all BARs and skip the map of BAR
with an invalid physical address in secondary process.

Fixes: 9b957f378abf ("pci: merge uio functions for linux and bsd")
Cc: mukawa@igel.co.jp
Cc: stable@dpdk.org

Signed-off-by: Zerun Fu <zerun.fu@corigine.com>
Reviewed-by: Chaoyong He <chaoyong.he@corigine.com>
Reviewed-by: Long Wu <long.wu@corigine.com>
Reviewed-by: Peng Zhang <peng.zhang@corigine.com>
---
 drivers/bus/pci/pci_common_uio.c | 40 ++++++++++++++++++++------------
 1 file changed, 25 insertions(+), 15 deletions(-)

diff --git a/drivers/bus/pci/pci_common_uio.c b/drivers/bus/pci/pci_common_uio.c
index 76c661f054..432d68b12c 100644
--- a/drivers/bus/pci/pci_common_uio.c
+++ b/drivers/bus/pci/pci_common_uio.c
@@ -26,7 +26,7 @@ EAL_REGISTER_TAILQ(rte_uio_tailq)
 static int
 pci_uio_map_secondary(struct rte_pci_device *dev)
 {
-	int fd, i, j;
+	int fd, i, map_idx = 0, res_idx;
 	struct mapped_pci_resource *uio_res;
 	struct mapped_pci_res_list *uio_res_list =
 			RTE_TAILQ_CAST(rte_uio_tailq.head, mapped_pci_res_list);
@@ -37,41 +37,51 @@ pci_uio_map_secondary(struct rte_pci_device *dev)
 		if (rte_pci_addr_cmp(&uio_res->pci_addr, &dev->addr))
 			continue;
 
-		for (i = 0; i != uio_res->nb_maps; i++) {
+		/* Map all BARs */
+		for (res_idx = 0; res_idx != PCI_MAX_RESOURCE; res_idx++) {
+			 /* skip empty BAR */
+			if (dev->mem_resource[res_idx].phys_addr == 0)
+				continue;
+
+			if (map_idx >= uio_res->nb_maps)
+				return -1;
+
 			/*
 			 * open devname, to mmap it
 			 */
-			fd = open(uio_res->maps[i].path, O_RDWR);
+			fd = open(uio_res->maps[map_idx].path, O_RDWR);
 			if (fd < 0) {
 				RTE_LOG(ERR, EAL, "Cannot open %s: %s\n",
-					uio_res->maps[i].path, strerror(errno));
+					uio_res->maps[map_idx].path, strerror(errno));
 				return -1;
 			}
 
-			void *mapaddr = pci_map_resource(uio_res->maps[i].addr,
-					fd, (off_t)uio_res->maps[i].offset,
-					(size_t)uio_res->maps[i].size, 0);
+			void *mapaddr = pci_map_resource(uio_res->maps[map_idx].addr,
+					fd, (off_t)uio_res->maps[map_idx].offset,
+					(size_t)uio_res->maps[map_idx].size, 0);
 
 			/* fd is not needed in secondary process, close it */
 			close(fd);
-			if (mapaddr != uio_res->maps[i].addr) {
+			if (mapaddr != uio_res->maps[map_idx].addr) {
 				RTE_LOG(ERR, EAL,
 					"Cannot mmap device resource file %s to address: %p\n",
-					uio_res->maps[i].path,
-					uio_res->maps[i].addr);
+					uio_res->maps[map_idx].path,
+					uio_res->maps[map_idx].addr);
 				if (mapaddr != NULL) {
 					/* unmap addrs correctly mapped */
-					for (j = 0; j < i; j++)
+					for (i = 0; i < map_idx; i++)
 						pci_unmap_resource(
-							uio_res->maps[j].addr,
-							(size_t)uio_res->maps[j].size);
+							uio_res->maps[i].addr,
+							(size_t)uio_res->maps[i].size);
 					/* unmap addr wrongly mapped */
 					pci_unmap_resource(mapaddr,
-						(size_t)uio_res->maps[i].size);
+						(size_t)uio_res->maps[map_idx].size);
 				}
 				return -1;
 			}
-			dev->mem_resource[i].addr = mapaddr;
+			dev->mem_resource[res_idx].addr = mapaddr;
+
+			map_idx++;
 		}
 		return 0;
 	}
-- 
2.39.1

Re: [PATCH v3 1/2] bus/pci: fix secondary process PCI uio resource map problem

[Thomas Monjalon]

19/04/2024 05:26, Chaoyong He:
> From: Zerun Fu <zerun.fu@corigine.com>
> 
> For the primary process, the logic loops all BARs and will skip
> the map of BAR with an invalid physical address (0), also will
> assign 'uio_res->nb_maps' with the real mapped BARs number. But
> for the secondary process, instead of loops all BARs, the logic
> using the 'uio_res->nb_map' as index. If the device uses continuous
> BARs there will be no problem, whereas if it uses discrete BARs,
> it will lead to mapping errors.
> 
> Fix this problem by also loops all BARs and skip the map of BAR
> with an invalid physical address in secondary process.
> 
> Fixes: 9b957f378abf ("pci: merge uio functions for linux and bsd")
> Cc: mukawa@igel.co.jp
> Cc: stable@dpdk.org
> 
> Signed-off-by: Zerun Fu <zerun.fu@corigine.com>
> Reviewed-by: Chaoyong He <chaoyong.he@corigine.com>
> Reviewed-by: Long Wu <long.wu@corigine.com>
> Reviewed-by: Peng Zhang <peng.zhang@corigine.com>

You should have kept the ack from Anatoly here.

>  drivers/bus/pci/pci_common_uio.c | 40 ++++++++++++++++++++------------
>  1 file changed, 25 insertions(+), 15 deletions(-)

There are too many changes in this sensitive code.
Please could you introduce a first patch for the renaming of the variable "i"?
It should make this patch simpler to read.

Thank you

RE: [PATCH v3 1/2] bus/pci: fix secondary process PCI uio resource map problem

[Chaoyong He]

> 19/04/2024 05:26, Chaoyong He:
> > From: Zerun Fu <zerun.fu@corigine.com>
> >
> > For the primary process, the logic loops all BARs and will skip the
> > map of BAR with an invalid physical address (0), also will assign
> > 'uio_res->nb_maps' with the real mapped BARs number. But for the
> > secondary process, instead of loops all BARs, the logic using the
> > 'uio_res->nb_map' as index. If the device uses continuous BARs there
> > will be no problem, whereas if it uses discrete BARs, it will lead to
> > mapping errors.
> >
> > Fix this problem by also loops all BARs and skip the map of BAR with
> > an invalid physical address in secondary process.
> >
> > Fixes: 9b957f378abf ("pci: merge uio functions for linux and bsd")
> > Cc: mukawa@igel.co.jp
> > Cc: stable@dpdk.org
> >
> > Signed-off-by: Zerun Fu <zerun.fu@corigine.com>
> > Reviewed-by: Chaoyong He <chaoyong.he@corigine.com>
> > Reviewed-by: Long Wu <long.wu@corigine.com>
> > Reviewed-by: Peng Zhang <peng.zhang@corigine.com>
> 
> You should have kept the ack from Anatoly here.
> 
> >  drivers/bus/pci/pci_common_uio.c | 40
> > ++++++++++++++++++++------------
> >  1 file changed, 25 insertions(+), 15 deletions(-)
> 
> There are too many changes in this sensitive code.
> Please could you introduce a first patch for the renaming of the variable "i"?
> It should make this patch simpler to read.
> 
> Thank you
> 

Okay, will do it in the next version patch.

Thanks for your review.

[PATCH v3 2/2] bus/pci: fix secondary process save 'FD' problem

[Chaoyong He]

From: Zerun Fu <zerun.fu@corigine.com>

In the previous logic the 'fd' was only saved in the primary process,
but for some devices this value is also used in the secondary logic.

For example, the call of 'rte_pci_find_ext_capability()' will fail in
the secondary process.

Fix this problem by getting and saving the value of 'fd' also in the
secondary process logic.

Fixes: 9b957f378abf ("pci: merge uio functions for linux and bsd")
Cc: mukawa@igel.co.jp
Cc: stable@dpdk.org

Signed-off-by: Zerun Fu <zerun.fu@corigine.com>
Reviewed-by: Chaoyong He <chaoyong.he@corigine.com>
Reviewed-by: Long Wu <long.wu@corigine.com>
Reviewed-by: Peng Zhang <peng.zhang@corigine.com>
---
 drivers/bus/pci/linux/pci_uio.c  | 5 ++++-
 drivers/bus/pci/pci_common_uio.c | 8 ++++----
 2 files changed, 8 insertions(+), 5 deletions(-)

diff --git a/drivers/bus/pci/linux/pci_uio.c b/drivers/bus/pci/linux/pci_uio.c
index 97d740dfe5..6680e42efb 100644
--- a/drivers/bus/pci/linux/pci_uio.c
+++ b/drivers/bus/pci/linux/pci_uio.c
@@ -237,7 +237,7 @@ pci_uio_alloc_resource(struct rte_pci_device *dev,
 	}
 	snprintf(devname, sizeof(devname), "/dev/uio%u", uio_num);
 
-	/* save fd if in primary process */
+	/* save fd */
 	fd = open(devname, O_RDWR);
 	if (fd < 0) {
 		RTE_LOG(ERR, EAL, "Cannot open %s: %s\n",
@@ -261,6 +261,9 @@ pci_uio_alloc_resource(struct rte_pci_device *dev,
 	if (rte_intr_dev_fd_set(dev->intr_handle, uio_cfg_fd))
 		goto error;
 
+	if (rte_eal_process_type() != RTE_PROC_PRIMARY)
+		return 0;
+
 	if (dev->kdrv == RTE_PCI_KDRV_IGB_UIO) {
 		if (rte_intr_type_set(dev->intr_handle, RTE_INTR_HANDLE_UIO))
 			goto error;
diff --git a/drivers/bus/pci/pci_common_uio.c b/drivers/bus/pci/pci_common_uio.c
index 432d68b12c..8688862738 100644
--- a/drivers/bus/pci/pci_common_uio.c
+++ b/drivers/bus/pci/pci_common_uio.c
@@ -106,15 +106,15 @@ pci_uio_map_resource(struct rte_pci_device *dev)
 	if (rte_intr_dev_fd_set(dev->intr_handle, -1))
 		return -1;
 
-	/* secondary processes - use already recorded details */
-	if (rte_eal_process_type() != RTE_PROC_PRIMARY)
-		return pci_uio_map_secondary(dev);
-
 	/* allocate uio resource */
 	ret = pci_uio_alloc_resource(dev, &uio_res);
 	if (ret)
 		return ret;
 
+	/* secondary processes - use already recorded details */
+	if (rte_eal_process_type() != RTE_PROC_PRIMARY)
+		return pci_uio_map_secondary(dev);
+
 	/* Map all BARs */
 	for (i = 0; i != PCI_MAX_RESOURCE; i++) {
 		/* skip empty BAR */
-- 
2.39.1

[PATCH v4 2/3] bus/pci: fix secondary process PCI uio resource map problem

[Chaoyong He]

From: Zerun Fu <zerun.fu@corigine.com>

For the primary process, the logic loops all BARs and will skip
the map of BAR with an invalid physical address (0), also will
assign 'uio_res->nb_maps' with the real mapped BARs number. But
for the secondary process, instead of loops all BARs, the logic
using the 'uio_res->nb_map' as index. If the device uses continuous
BARs there will be no problem, whereas if it uses discrete BARs,
it will lead to mapping errors.

Fix this problem by also loops all BARs and skip the map of BAR
with an invalid physical address in secondary process.

Fixes: 9b957f378abf ("pci: merge uio functions for linux and bsd")
Cc: mukawa@igel.co.jp
Cc: stable@dpdk.org

Signed-off-by: Zerun Fu <zerun.fu@corigine.com>
Reviewed-by: Chaoyong He <chaoyong.he@corigine.com>
Reviewed-by: Long Wu <long.wu@corigine.com>
Reviewed-by: Peng Zhang <peng.zhang@corigine.com>
---
 drivers/bus/pci/pci_common_uio.c | 16 +++++++++++++---
 1 file changed, 13 insertions(+), 3 deletions(-)

diff --git a/drivers/bus/pci/pci_common_uio.c b/drivers/bus/pci/pci_common_uio.c
index caacbb69d9..432d68b12c 100644
--- a/drivers/bus/pci/pci_common_uio.c
+++ b/drivers/bus/pci/pci_common_uio.c
@@ -26,7 +26,7 @@ EAL_REGISTER_TAILQ(rte_uio_tailq)
 static int
 pci_uio_map_secondary(struct rte_pci_device *dev)
 {
-	int fd, i, map_idx;
+	int fd, i, map_idx = 0, res_idx;
 	struct mapped_pci_resource *uio_res;
 	struct mapped_pci_res_list *uio_res_list =
 			RTE_TAILQ_CAST(rte_uio_tailq.head, mapped_pci_res_list);
@@ -37,7 +37,15 @@ pci_uio_map_secondary(struct rte_pci_device *dev)
 		if (rte_pci_addr_cmp(&uio_res->pci_addr, &dev->addr))
 			continue;
 
-		for (map_idx = 0; map_idx != uio_res->nb_maps; map_idx++) {
+		/* Map all BARs */
+		for (res_idx = 0; res_idx != PCI_MAX_RESOURCE; res_idx++) {
+			 /* skip empty BAR */
+			if (dev->mem_resource[res_idx].phys_addr == 0)
+				continue;
+
+			if (map_idx >= uio_res->nb_maps)
+				return -1;
+
 			/*
 			 * open devname, to mmap it
 			 */
@@ -71,7 +79,9 @@ pci_uio_map_secondary(struct rte_pci_device *dev)
 				}
 				return -1;
 			}
-			dev->mem_resource[map_idx].addr = mapaddr;
+			dev->mem_resource[res_idx].addr = mapaddr;
+
+			map_idx++;
 		}
 		return 0;
 	}
-- 
2.39.1

[PATCH v4 3/3] bus/pci: fix secondary process save 'FD' problem

[Chaoyong He]

From: Zerun Fu <zerun.fu@corigine.com>

In the previous logic the 'fd' was only saved in the primary process,
but for some devices this value is also used in the secondary logic.

For example, the call of 'rte_pci_find_ext_capability()' will fail in
the secondary process.

Fix this problem by getting and saving the value of 'fd' also in the
secondary process logic.

Fixes: 9b957f378abf ("pci: merge uio functions for linux and bsd")
Cc: mukawa@igel.co.jp
Cc: stable@dpdk.org

Signed-off-by: Zerun Fu <zerun.fu@corigine.com>
Reviewed-by: Chaoyong He <chaoyong.he@corigine.com>
Reviewed-by: Long Wu <long.wu@corigine.com>
Reviewed-by: Peng Zhang <peng.zhang@corigine.com>
---
 drivers/bus/pci/linux/pci_uio.c  | 5 ++++-
 drivers/bus/pci/pci_common_uio.c | 8 ++++----
 2 files changed, 8 insertions(+), 5 deletions(-)

diff --git a/drivers/bus/pci/linux/pci_uio.c b/drivers/bus/pci/linux/pci_uio.c
index 97d740dfe5..6680e42efb 100644
--- a/drivers/bus/pci/linux/pci_uio.c
+++ b/drivers/bus/pci/linux/pci_uio.c
@@ -237,7 +237,7 @@ pci_uio_alloc_resource(struct rte_pci_device *dev,
 	}
 	snprintf(devname, sizeof(devname), "/dev/uio%u", uio_num);
 
-	/* save fd if in primary process */
+	/* save fd */
 	fd = open(devname, O_RDWR);
 	if (fd < 0) {
 		RTE_LOG(ERR, EAL, "Cannot open %s: %s\n",
@@ -261,6 +261,9 @@ pci_uio_alloc_resource(struct rte_pci_device *dev,
 	if (rte_intr_dev_fd_set(dev->intr_handle, uio_cfg_fd))
 		goto error;
 
+	if (rte_eal_process_type() != RTE_PROC_PRIMARY)
+		return 0;
+
 	if (dev->kdrv == RTE_PCI_KDRV_IGB_UIO) {
 		if (rte_intr_type_set(dev->intr_handle, RTE_INTR_HANDLE_UIO))
 			goto error;
diff --git a/drivers/bus/pci/pci_common_uio.c b/drivers/bus/pci/pci_common_uio.c
index 432d68b12c..8688862738 100644
--- a/drivers/bus/pci/pci_common_uio.c
+++ b/drivers/bus/pci/pci_common_uio.c
@@ -106,15 +106,15 @@ pci_uio_map_resource(struct rte_pci_device *dev)
 	if (rte_intr_dev_fd_set(dev->intr_handle, -1))
 		return -1;
 
-	/* secondary processes - use already recorded details */
-	if (rte_eal_process_type() != RTE_PROC_PRIMARY)
-		return pci_uio_map_secondary(dev);
-
 	/* allocate uio resource */
 	ret = pci_uio_alloc_resource(dev, &uio_res);
 	if (ret)
 		return ret;
 
+	/* secondary processes - use already recorded details */
+	if (rte_eal_process_type() != RTE_PROC_PRIMARY)
+		return pci_uio_map_secondary(dev);
+
 	/* Map all BARs */
 	for (i = 0; i != PCI_MAX_RESOURCE; i++) {
 		/* skip empty BAR */
-- 
2.39.1

Re: [PATCH v4 3/3] bus/pci: fix secondary process save 'FD' problem

[David Marchand]

On Fri, Jun 28, 2024 at 9:37 AM Chaoyong He <chaoyong.he@corigine.com> wrote:
>
> From: Zerun Fu <zerun.fu@corigine.com>
>
> In the previous logic the 'fd' was only saved in the primary process,
> but for some devices this value is also used in the secondary logic.
>
> For example, the call of 'rte_pci_find_ext_capability()' will fail in
> the secondary process.
>
> Fix this problem by getting and saving the value of 'fd' also in the
> secondary process logic.
>
> Fixes: 9b957f378abf ("pci: merge uio functions for linux and bsd")
> Cc: mukawa@igel.co.jp
> Cc: stable@dpdk.org
>
> Signed-off-by: Zerun Fu <zerun.fu@corigine.com>
> Reviewed-by: Chaoyong He <chaoyong.he@corigine.com>
> Reviewed-by: Long Wu <long.wu@corigine.com>
> Reviewed-by: Peng Zhang <peng.zhang@corigine.com>

I did not enter in depth, but I wonder if this change leaks fd in a
secondary process when unmapping UIO resources.


-- 
David Marchand

RE: [PATCH v4 3/3] bus/pci: fix secondary process save 'FD' problem

[Chaoyong He]

> On Fri, Jun 28, 2024 at 9:37 AM Chaoyong He <chaoyong.he@corigine.com>
> wrote:
> >
> > From: Zerun Fu <zerun.fu@corigine.com>
> >
> > In the previous logic the 'fd' was only saved in the primary process,
> > but for some devices this value is also used in the secondary logic.
> >
> > For example, the call of 'rte_pci_find_ext_capability()' will fail in
> > the secondary process.
> >
> > Fix this problem by getting and saving the value of 'fd' also in the
> > secondary process logic.
> >
> > Fixes: 9b957f378abf ("pci: merge uio functions for linux and bsd")
> > Cc: mukawa@igel.co.jp
> > Cc: stable@dpdk.org
> >
> > Signed-off-by: Zerun Fu <zerun.fu@corigine.com>
> > Reviewed-by: Chaoyong He <chaoyong.he@corigine.com>
> > Reviewed-by: Long Wu <long.wu@corigine.com>
> > Reviewed-by: Peng Zhang <peng.zhang@corigine.com>
> 
> I did not enter in depth, but I wonder if this change leaks fd in a secondary
> process when unmapping UIO resources.
> 

Yeah, it does have problem when unmapping UIO resources, and we will send a new version patch to fix it soon.
Many thanks for your carefully review.

> 
> --
> David Marchand

[PATCH v5 2/3] bus/pci: fix secondary process PCI uio resource map problem

[Chaoyong He]

From: Zerun Fu <zerun.fu@corigine.com>

For the primary process, the logic loops all BARs and will skip
the map of BAR with an invalid physical address (0), also will
assign 'uio_res->nb_maps' with the real mapped BARs number. But
for the secondary process, instead of loops all BARs, the logic
using the 'uio_res->nb_map' as index. If the device uses continuous
BARs there will be no problem, whereas if it uses discrete BARs,
it will lead to mapping errors.

Fix this problem by also loops all BARs and skip the map of BAR
with an invalid physical address in secondary process.

Fixes: 9b957f378abf ("pci: merge uio functions for linux and bsd")
Cc: mukawa@igel.co.jp
Cc: stable@dpdk.org

Signed-off-by: Zerun Fu <zerun.fu@corigine.com>
Reviewed-by: Chaoyong He <chaoyong.he@corigine.com>
Reviewed-by: Long Wu <long.wu@corigine.com>
Reviewed-by: Peng Zhang <peng.zhang@corigine.com>
---
 drivers/bus/pci/pci_common_uio.c | 16 +++++++++++++---
 1 file changed, 13 insertions(+), 3 deletions(-)

diff --git a/drivers/bus/pci/pci_common_uio.c b/drivers/bus/pci/pci_common_uio.c
index caacbb69d9..432d68b12c 100644
--- a/drivers/bus/pci/pci_common_uio.c
+++ b/drivers/bus/pci/pci_common_uio.c
@@ -26,7 +26,7 @@ EAL_REGISTER_TAILQ(rte_uio_tailq)
 static int
 pci_uio_map_secondary(struct rte_pci_device *dev)
 {
-	int fd, i, map_idx;
+	int fd, i, map_idx = 0, res_idx;
 	struct mapped_pci_resource *uio_res;
 	struct mapped_pci_res_list *uio_res_list =
 			RTE_TAILQ_CAST(rte_uio_tailq.head, mapped_pci_res_list);
@@ -37,7 +37,15 @@ pci_uio_map_secondary(struct rte_pci_device *dev)
 		if (rte_pci_addr_cmp(&uio_res->pci_addr, &dev->addr))
 			continue;
 
-		for (map_idx = 0; map_idx != uio_res->nb_maps; map_idx++) {
+		/* Map all BARs */
+		for (res_idx = 0; res_idx != PCI_MAX_RESOURCE; res_idx++) {
+			 /* skip empty BAR */
+			if (dev->mem_resource[res_idx].phys_addr == 0)
+				continue;
+
+			if (map_idx >= uio_res->nb_maps)
+				return -1;
+
 			/*
 			 * open devname, to mmap it
 			 */
@@ -71,7 +79,9 @@ pci_uio_map_secondary(struct rte_pci_device *dev)
 				}
 				return -1;
 			}
-			dev->mem_resource[map_idx].addr = mapaddr;
+			dev->mem_resource[res_idx].addr = mapaddr;
+
+			map_idx++;
 		}
 		return 0;
 	}
-- 
2.39.1

[PATCH v5 3/3] bus/pci: fix secondary process save 'FD' problem

[Chaoyong He]

From: Zerun Fu <zerun.fu@corigine.com>

In the previous logic the 'fd' was only saved in the primary process,
but for some devices this value is also used in the secondary logic.

For example, the call of 'rte_pci_find_ext_capability()' will fail in
the secondary process.

Fix this problem by getting and saving the value of 'fd' also in the
secondary process logic.

Fixes: 9b957f378abf ("pci: merge uio functions for linux and bsd")
Cc: mukawa@igel.co.jp
Cc: stable@dpdk.org

Signed-off-by: Zerun Fu <zerun.fu@corigine.com>
Reviewed-by: Chaoyong He <chaoyong.he@corigine.com>
Reviewed-by: Long Wu <long.wu@corigine.com>
Reviewed-by: Peng Zhang <peng.zhang@corigine.com>
---
 drivers/bus/pci/linux/pci_uio.c  |  5 ++++-
 drivers/bus/pci/pci_common_uio.c | 32 ++++++++++++++++----------------
 2 files changed, 20 insertions(+), 17 deletions(-)

diff --git a/drivers/bus/pci/linux/pci_uio.c b/drivers/bus/pci/linux/pci_uio.c
index 97d740dfe5..4afda97858 100644
--- a/drivers/bus/pci/linux/pci_uio.c
+++ b/drivers/bus/pci/linux/pci_uio.c
@@ -237,7 +237,7 @@ pci_uio_alloc_resource(struct rte_pci_device *dev,
 	}
 	snprintf(devname, sizeof(devname), "/dev/uio%u", uio_num);
 
-	/* save fd if in primary process */
+	/* save fd */
 	fd = open(devname, O_RDWR);
 	if (fd < 0) {
 		RTE_LOG(ERR, EAL, "Cannot open %s: %s\n",
@@ -275,6 +275,9 @@ pci_uio_alloc_resource(struct rte_pci_device *dev,
 		}
 	}
 
+	if (rte_eal_process_type() != RTE_PROC_PRIMARY)
+		return 0;
+
 	/* allocate the mapping details for secondary processes*/
 	*uio_res = rte_zmalloc("UIO_RES", sizeof(**uio_res), 0);
 	if (*uio_res == NULL) {
diff --git a/drivers/bus/pci/pci_common_uio.c b/drivers/bus/pci/pci_common_uio.c
index 432d68b12c..09f30bee6c 100644
--- a/drivers/bus/pci/pci_common_uio.c
+++ b/drivers/bus/pci/pci_common_uio.c
@@ -106,15 +106,15 @@ pci_uio_map_resource(struct rte_pci_device *dev)
 	if (rte_intr_dev_fd_set(dev->intr_handle, -1))
 		return -1;
 
-	/* secondary processes - use already recorded details */
-	if (rte_eal_process_type() != RTE_PROC_PRIMARY)
-		return pci_uio_map_secondary(dev);
-
 	/* allocate uio resource */
 	ret = pci_uio_alloc_resource(dev, &uio_res);
 	if (ret)
 		return ret;
 
+	/* secondary processes - use already recorded details */
+	if (rte_eal_process_type() != RTE_PROC_PRIMARY)
+		return pci_uio_map_secondary(dev);
+
 	/* Map all BARs */
 	for (i = 0; i != PCI_MAX_RESOURCE; i++) {
 		/* skip empty BAR */
@@ -230,6 +230,18 @@ pci_uio_unmap_resource(struct rte_pci_device *dev)
 	if (uio_res == NULL)
 		return;
 
+	/* close fd */
+	if (rte_intr_fd_get(dev->intr_handle) >= 0)
+		close(rte_intr_fd_get(dev->intr_handle));
+	uio_cfg_fd = rte_intr_dev_fd_get(dev->intr_handle);
+	if (uio_cfg_fd >= 0) {
+		close(uio_cfg_fd);
+		rte_intr_dev_fd_set(dev->intr_handle, -1);
+	}
+
+	rte_intr_fd_set(dev->intr_handle, -1);
+	rte_intr_type_set(dev->intr_handle, RTE_INTR_HANDLE_UNKNOWN);
+
 	/* secondary processes - just free maps */
 	if (rte_eal_process_type() != RTE_PROC_PRIMARY)
 		return pci_uio_unmap(uio_res);
@@ -241,16 +253,4 @@ pci_uio_unmap_resource(struct rte_pci_device *dev)
 
 	/* free uio resource */
 	rte_free(uio_res);
-
-	/* close fd if in primary process */
-	if (rte_intr_fd_get(dev->intr_handle) >= 0)
-		close(rte_intr_fd_get(dev->intr_handle));
-	uio_cfg_fd = rte_intr_dev_fd_get(dev->intr_handle);
-	if (uio_cfg_fd >= 0) {
-		close(uio_cfg_fd);
-		rte_intr_dev_fd_set(dev->intr_handle, -1);
-	}
-
-	rte_intr_fd_set(dev->intr_handle, -1);
-	rte_intr_type_set(dev->intr_handle, RTE_INTR_HANDLE_UNKNOWN);
 }
-- 
2.39.1

[PATCH v6 1/2] bus/pci: fix secondary process PCI uio resource map problem

[Chaoyong He]

From: Zerun Fu <zerun.fu@corigine.com>

For the primary process, the logic loops all BARs and will skip
the map of BAR with an invalid physical address (0), also will
assign 'uio_res->nb_maps' with the real mapped BARs number. But
for the secondary process, instead of loops all BARs, the logic
using the 'uio_res->nb_map' as index. If the device uses continuous
BARs there will be no problem, whereas if it uses discrete BARs,
it will lead to mapping errors.

Fix this problem by also loops all BARs and skip the map of BAR
with an invalid physical address in secondary process.

Fixes: 9b957f378abf ("pci: merge uio functions for linux and bsd")
Cc: mukawa@igel.co.jp
Cc: stable@dpdk.org

Signed-off-by: Zerun Fu <zerun.fu@corigine.com>
Reviewed-by: Chaoyong He <chaoyong.he@corigine.com>
Reviewed-by: Long Wu <long.wu@corigine.com>
Reviewed-by: Peng Zhang <peng.zhang@corigine.com>
Acked-by: Anatoly Burakov <anatoly.burakov@intel.com>
---
 drivers/bus/pci/pci_common_uio.c | 16 +++++++++++++---
 1 file changed, 13 insertions(+), 3 deletions(-)

diff --git a/drivers/bus/pci/pci_common_uio.c b/drivers/bus/pci/pci_common_uio.c
index 76c661f054..f44ccdf27c 100644
--- a/drivers/bus/pci/pci_common_uio.c
+++ b/drivers/bus/pci/pci_common_uio.c
@@ -26,7 +26,7 @@ EAL_REGISTER_TAILQ(rte_uio_tailq)
 static int
 pci_uio_map_secondary(struct rte_pci_device *dev)
 {
-	int fd, i, j;
+	int fd, i = 0, j, res_idx;
 	struct mapped_pci_resource *uio_res;
 	struct mapped_pci_res_list *uio_res_list =
 			RTE_TAILQ_CAST(rte_uio_tailq.head, mapped_pci_res_list);
@@ -37,7 +37,15 @@ pci_uio_map_secondary(struct rte_pci_device *dev)
 		if (rte_pci_addr_cmp(&uio_res->pci_addr, &dev->addr))
 			continue;
 
-		for (i = 0; i != uio_res->nb_maps; i++) {
+		/* Map all BARs */
+		for (res_idx = 0; res_idx != PCI_MAX_RESOURCE; res_idx++) {
+			/* skip empty BAR */
+			if (dev->mem_resource[res_idx].phys_addr == 0)
+				continue;
+
+			if (i >= uio_res->nb_maps)
+				return -1;
+
 			/*
 			 * open devname, to mmap it
 			 */
@@ -71,7 +79,9 @@ pci_uio_map_secondary(struct rte_pci_device *dev)
 				}
 				return -1;
 			}
-			dev->mem_resource[i].addr = mapaddr;
+			dev->mem_resource[res_idx].addr = mapaddr;
+
+			i++;
 		}
 		return 0;
 	}
-- 
2.39.1

Re: [PATCH v6 1/2] bus/pci: fix secondary process PCI uio resource map problem

[Chenbo Xia]

> On Jul 2, 2024, at 15:40, Chaoyong He <chaoyong.he@corigine.com> wrote:
> 
> External email: Use caution opening links or attachments
> 
> 
> From: Zerun Fu <zerun.fu@corigine.com>
> 
> For the primary process, the logic loops all BARs and will skip
> the map of BAR with an invalid physical address (0), also will
> assign 'uio_res->nb_maps' with the real mapped BARs number. But
> for the secondary process, instead of loops all BARs, the logic
> using the 'uio_res->nb_map' as index. If the device uses continuous
> BARs there will be no problem, whereas if it uses discrete BARs,
> it will lead to mapping errors.
> 
> Fix this problem by also loops all BARs and skip the map of BAR
> with an invalid physical address in secondary process.
> 
> Fixes: 9b957f378abf ("pci: merge uio functions for linux and bsd")
> Cc: mukawa@igel.co.jp
> Cc: stable@dpdk.org
> 
> Signed-off-by: Zerun Fu <zerun.fu@corigine.com>
> Reviewed-by: Chaoyong He <chaoyong.he@corigine.com>
> Reviewed-by: Long Wu <long.wu@corigine.com>
> Reviewed-by: Peng Zhang <peng.zhang@corigine.com>
> Acked-by: Anatoly Burakov <anatoly.burakov@intel.com>
> ---
> drivers/bus/pci/pci_common_uio.c | 16 +++++++++++++---
> 1 file changed, 13 insertions(+), 3 deletions(-)
> 
> diff --git a/drivers/bus/pci/pci_common_uio.c b/drivers/bus/pci/pci_common_uio.c
> index 76c661f054..f44ccdf27c 100644
> --- a/drivers/bus/pci/pci_common_uio.c
> +++ b/drivers/bus/pci/pci_common_uio.c
> @@ -26,7 +26,7 @@ EAL_REGISTER_TAILQ(rte_uio_tailq)
> static int
> pci_uio_map_secondary(struct rte_pci_device *dev)
> {
> -       int fd, i, j;
> +       int fd, i = 0, j, res_idx;
>        struct mapped_pci_resource *uio_res;
>        struct mapped_pci_res_list *uio_res_list =
>                        RTE_TAILQ_CAST(rte_uio_tailq.head, mapped_pci_res_list);
> @@ -37,7 +37,15 @@ pci_uio_map_secondary(struct rte_pci_device *dev)
>                if (rte_pci_addr_cmp(&uio_res->pci_addr, &dev->addr))
>                        continue;
> 
> -               for (i = 0; i != uio_res->nb_maps; i++) {
> +               /* Map all BARs */
> +               for (res_idx = 0; res_idx != PCI_MAX_RESOURCE; res_idx++) {
> +                       /* skip empty BAR */
> +                       if (dev->mem_resource[res_idx].phys_addr == 0)
> +                               continue;
> +
> +                       if (i >= uio_res->nb_maps)
> +                               return -1;
> +
>                        /*
>                         * open devname, to mmap it
>                         */
> @@ -71,7 +79,9 @@ pci_uio_map_secondary(struct rte_pci_device *dev)
>                                }
>                                return -1;
>                        }
> -                       dev->mem_resource[i].addr = mapaddr;
> +                       dev->mem_resource[res_idx].addr = mapaddr;
> +
> +                       i++;
>                }
>                return 0;
>        }
> —
> 2.39.1
> 

Reviewed-by: Chenbo Xia <chenbox@nvidia.com>

[PATCH v6 2/2] bus/pci: fix secondary process save 'FD' problem

[Chaoyong He]

From: Zerun Fu <zerun.fu@corigine.com>

In the previous logic the 'fd' was only saved in the primary process,
but for some devices this value is also used in the secondary logic.

For example, the call of 'rte_pci_find_ext_capability()' will fail in
the secondary process.

Fix this problem by getting and saving the value of 'fd' also in the
secondary process logic.

Fixes: 9b957f378abf ("pci: merge uio functions for linux and bsd")
Cc: mukawa@igel.co.jp
Cc: stable@dpdk.org

Signed-off-by: Zerun Fu <zerun.fu@corigine.com>
Reviewed-by: Chaoyong He <chaoyong.he@corigine.com>
Reviewed-by: Long Wu <long.wu@corigine.com>
Reviewed-by: Peng Zhang <peng.zhang@corigine.com>
Acked-by: Anatoly Burakov <anatoly.burakov@intel.com>
---
 drivers/bus/pci/linux/pci_uio.c  |  5 ++++-
 drivers/bus/pci/pci_common_uio.c | 32 ++++++++++++++++----------------
 2 files changed, 20 insertions(+), 17 deletions(-)

diff --git a/drivers/bus/pci/linux/pci_uio.c b/drivers/bus/pci/linux/pci_uio.c
index 97d740dfe5..4afda97858 100644
--- a/drivers/bus/pci/linux/pci_uio.c
+++ b/drivers/bus/pci/linux/pci_uio.c
@@ -237,7 +237,7 @@ pci_uio_alloc_resource(struct rte_pci_device *dev,
 	}
 	snprintf(devname, sizeof(devname), "/dev/uio%u", uio_num);
 
-	/* save fd if in primary process */
+	/* save fd */
 	fd = open(devname, O_RDWR);
 	if (fd < 0) {
 		RTE_LOG(ERR, EAL, "Cannot open %s: %s\n",
@@ -275,6 +275,9 @@ pci_uio_alloc_resource(struct rte_pci_device *dev,
 		}
 	}
 
+	if (rte_eal_process_type() != RTE_PROC_PRIMARY)
+		return 0;
+
 	/* allocate the mapping details for secondary processes*/
 	*uio_res = rte_zmalloc("UIO_RES", sizeof(**uio_res), 0);
 	if (*uio_res == NULL) {
diff --git a/drivers/bus/pci/pci_common_uio.c b/drivers/bus/pci/pci_common_uio.c
index f44ccdf27c..a06378b239 100644
--- a/drivers/bus/pci/pci_common_uio.c
+++ b/drivers/bus/pci/pci_common_uio.c
@@ -106,15 +106,15 @@ pci_uio_map_resource(struct rte_pci_device *dev)
 	if (rte_intr_dev_fd_set(dev->intr_handle, -1))
 		return -1;
 
-	/* secondary processes - use already recorded details */
-	if (rte_eal_process_type() != RTE_PROC_PRIMARY)
-		return pci_uio_map_secondary(dev);
-
 	/* allocate uio resource */
 	ret = pci_uio_alloc_resource(dev, &uio_res);
 	if (ret)
 		return ret;
 
+	/* secondary processes - use already recorded details */
+	if (rte_eal_process_type() != RTE_PROC_PRIMARY)
+		return pci_uio_map_secondary(dev);
+
 	/* Map all BARs */
 	for (i = 0; i != PCI_MAX_RESOURCE; i++) {
 		/* skip empty BAR */
@@ -230,6 +230,18 @@ pci_uio_unmap_resource(struct rte_pci_device *dev)
 	if (uio_res == NULL)
 		return;
 
+	/* close fd */
+	if (rte_intr_fd_get(dev->intr_handle) >= 0)
+		close(rte_intr_fd_get(dev->intr_handle));
+	uio_cfg_fd = rte_intr_dev_fd_get(dev->intr_handle);
+	if (uio_cfg_fd >= 0) {
+		close(uio_cfg_fd);
+		rte_intr_dev_fd_set(dev->intr_handle, -1);
+	}
+
+	rte_intr_fd_set(dev->intr_handle, -1);
+	rte_intr_type_set(dev->intr_handle, RTE_INTR_HANDLE_UNKNOWN);
+
 	/* secondary processes - just free maps */
 	if (rte_eal_process_type() != RTE_PROC_PRIMARY)
 		return pci_uio_unmap(uio_res);
@@ -241,16 +253,4 @@ pci_uio_unmap_resource(struct rte_pci_device *dev)
 
 	/* free uio resource */
 	rte_free(uio_res);
-
-	/* close fd if in primary process */
-	if (rte_intr_fd_get(dev->intr_handle) >= 0)
-		close(rte_intr_fd_get(dev->intr_handle));
-	uio_cfg_fd = rte_intr_dev_fd_get(dev->intr_handle);
-	if (uio_cfg_fd >= 0) {
-		close(uio_cfg_fd);
-		rte_intr_dev_fd_set(dev->intr_handle, -1);
-	}
-
-	rte_intr_fd_set(dev->intr_handle, -1);
-	rte_intr_type_set(dev->intr_handle, RTE_INTR_HANDLE_UNKNOWN);
 }
-- 
2.39.1

Re: [PATCH v6 2/2] bus/pci: fix secondary process save 'FD' problem

[David Marchand]

Hello,

On Tue, Jul 2, 2024 at 9:40 AM Chaoyong He <chaoyong.he@corigine.com> wrote:
>
> From: Zerun Fu <zerun.fu@corigine.com>
>
> In the previous logic the 'fd' was only saved in the primary process,
> but for some devices this value is also used in the secondary logic.
>
> For example, the call of 'rte_pci_find_ext_capability()' will fail in
> the secondary process.
>
> Fix this problem by getting and saving the value of 'fd' also in the
> secondary process logic.
>
> Fixes: 9b957f378abf ("pci: merge uio functions for linux and bsd")
> Cc: mukawa@igel.co.jp
> Cc: stable@dpdk.org
>
> Signed-off-by: Zerun Fu <zerun.fu@corigine.com>
> Reviewed-by: Chaoyong He <chaoyong.he@corigine.com>
> Reviewed-by: Long Wu <long.wu@corigine.com>
> Reviewed-by: Peng Zhang <peng.zhang@corigine.com>
> Acked-by: Anatoly Burakov <anatoly.burakov@intel.com>

Chenbo,
Are you ok with this fix?

Thanks.

-- 
David Marchand

Re: [PATCH v6 2/2] bus/pci: fix secondary process save 'FD' problem

[Chenbo Xia]

Hi David,

> On Jul 13, 2024, at 01:30, David Marchand <david.marchand@redhat.com> wrote:
> 
> External email: Use caution opening links or attachments
> 
> 
> Hello,
> 
> On Tue, Jul 2, 2024 at 9:40 AM Chaoyong He <chaoyong.he@corigine.com> wrote:
>> 
>> From: Zerun Fu <zerun.fu@corigine.com>
>> 
>> In the previous logic the 'fd' was only saved in the primary process,
>> but for some devices this value is also used in the secondary logic.
>> 
>> For example, the call of 'rte_pci_find_ext_capability()' will fail in
>> the secondary process.
>> 
>> Fix this problem by getting and saving the value of 'fd' also in the
>> secondary process logic.
>> 
>> Fixes: 9b957f378abf ("pci: merge uio functions for linux and bsd")
>> Cc: mukawa@igel.co.jp
>> Cc: stable@dpdk.org
>> 
>> Signed-off-by: Zerun Fu <zerun.fu@corigine.com>
>> Reviewed-by: Chaoyong He <chaoyong.he@corigine.com>
>> Reviewed-by: Long Wu <long.wu@corigine.com>
>> Reviewed-by: Peng Zhang <peng.zhang@corigine.com>
>> Acked-by: Anatoly Burakov <anatoly.burakov@intel.com>
> 
> Chenbo,
> Are you ok with this fix?

Sorry that I was interrupted when I was reviewing this and later I forgot..

For this patch:

Reviewed-by: Chenbo Xia <chenbox@nvidia.com>

> 
> Thanks.
> 
> --
> David Marchand