Skip to contentSkip to conversation

net/virtio: Add virtio support for Windows.

RFC PATCH In review William Tu 3 participants 3 messages Aug 15, 2021 – Aug 17, 2021

Subsystems (2)
lore
Share & tools
lore

William Tu

author
#0 lore raw

RFC PATCH 1/1 net/virtio: Add virtio support for Windows.

I looked at Dmitry's patch [1] a while ago and found that now it's not much work to enable the virtio support. Pass build, but I haven't tested it yet. I plan to try it on GCP.

[1] http://inbox.dpdk.org/dev/[email protected]/

---
 drivers/net/virtio/meson.build                   | 6 ------
 drivers/net/virtio/virtio_ethdev.c               | 4 ++--
 drivers/net/virtio/virtio_pci_ethdev.c           | 2 ++
 drivers/net/virtio/virtio_user/virtio_user_dev.h | 4 ++++
 4 files changed, 8 insertions(+), 8 deletions(-)
drivers/net/virtio/meson.build +0 -6
index 01a333ada2..14c02ee671 100644
@@ -1,12 +1,6 @@
 # SPDX-License-Identifier: BSD-3-Clause
 # Copyright(c) 2018 Intel Corporation
 
-if is_windows
-    build = false
-    reason = 'not supported on Windows'
-    subdir_done()
-endif
-
 sources += files(
         'virtio.c',
         'virtio_ethdev.c',
drivers/net/virtio/virtio_ethdev.c +2 -2
index e58085a2c9..7dbf6d0ea7 100644
@@ -198,7 +198,7 @@ virtio_send_command_packed(struct virtnet_ctl *cvq,
 	 * desc_is_used has a load-acquire or rte_io_rmb inside
 	 */
 	while (!desc_is_used(&desc[head], vq))
-		usleep(100);
+		rte_delay_us_sleep(100);
 
 	/* now get used descriptors */
 	vq->vq_free_cnt += nb_descs;
@@ -274,7 +274,7 @@ virtio_send_command_split(struct virtnet_ctl *cvq,
 	virtqueue_notify(vq);
 
 	while (virtqueue_nused(vq) == 0)
-		usleep(100);
+		rte_delay_us_sleep(100);
 
 	while (virtqueue_nused(vq)) {
 		uint32_t idx, desc_idx, used_idx;
drivers/net/virtio/virtio_pci_ethdev.c +2 -0
index 4083853c48..03e9dfab71 100644
@@ -220,7 +220,9 @@ static struct rte_pci_driver rte_virtio_net_pci_pmd = {
 
 RTE_INIT(rte_virtio_net_pci_pmd_init)
 {
+#if !(defined(_WIN32) || defined(_WIN64))
 	rte_eal_iopl_init();
+#endif
 	rte_pci_register(&rte_virtio_net_pci_pmd);
 }
 
drivers/net/virtio/virtio_user/virtio_user_dev.h +4 -0
index 819f6463ba..01cb962770 100644
@@ -11,6 +11,10 @@
 #include "../virtio.h"
 #include "../virtio_ring.h"
 
+#if defined(_WIN32) || defined(_WIN64)
+#define PATH_MAX MAX_PATH
+#endif
+
 enum virtio_user_backend_type {
 	VIRTIO_USER_BACKEND_UNKNOWN,
 	VIRTIO_USER_BACKEND_VHOST_USER,
Signature · 1 line
2.30.2
#1 lore raw

2021-08-15 14:56 (UTC+0000), William Tu:

Quoted 3 lines

I looked at Dmitry's patch [1] a while ago and found that now it's not much work to enable the virtio support. Pass build, but I haven't tested it yet. I plan to try it on GCP.

1. Build fails, the reason is __rte_weak: there are no weak symbols on Windows. The best solution to this I came up with so far is to create a file with stubs for weak functions. The -fno-asynchronous-unwind-tables flag will also be required for MinGW to link SSE files.

2. I can say in advance it won't work without an ability to read capabilities form PCI configuration space (rte_pci_read_config). This requires adding support in bus/pci.

Quoted 55 lines

[1] http://inbox.dpdk.org/dev/[email protected]/

---
 drivers/net/virtio/meson.build                   | 6 ------
 drivers/net/virtio/virtio_ethdev.c               | 4 ++--
 drivers/net/virtio/virtio_pci_ethdev.c           | 2 ++
 drivers/net/virtio/virtio_user/virtio_user_dev.h | 4 ++++
 4 files changed, 8 insertions(+), 8 deletions(-)
drivers/net/virtio/meson.build +0 -6
index 01a333ada2..14c02ee671 100644
@@ -1,12 +1,6 @@
 # SPDX-License-Identifier: BSD-3-Clause
 # Copyright(c) 2018 Intel Corporation
 
-if is_windows
-    build = false
-    reason = 'not supported on Windows'
-    subdir_done()
-endif
-
 sources += files(
         'virtio.c',
         'virtio_ethdev.c',
drivers/net/virtio/virtio_ethdev.c +2 -2
index e58085a2c9..7dbf6d0ea7 100644
@@ -198,7 +198,7 @@ virtio_send_command_packed(struct virtnet_ctl *cvq,
 	 * desc_is_used has a load-acquire or rte_io_rmb inside
 	 */
 	while (!desc_is_used(&desc[head], vq))
-		usleep(100);
+		rte_delay_us_sleep(100);
 
 	/* now get used descriptors */
 	vq->vq_free_cnt += nb_descs;
@@ -274,7 +274,7 @@ virtio_send_command_split(struct virtnet_ctl *cvq,
 	virtqueue_notify(vq);
 
 	while (virtqueue_nused(vq) == 0)
-		usleep(100);
+		rte_delay_us_sleep(100);
 
 	while (virtqueue_nused(vq)) {
 		uint32_t idx, desc_idx, used_idx;
drivers/net/virtio/virtio_pci_ethdev.c +0 -0
index 4083853c48..03e9dfab71 100644

drivers/net/virtio/virtio_pci_ethdev.c static struct rte_pci_driver rte_virtio_net_pci_pmd = { · line 225

Quoted 6 lines
drivers/net/virtio/virtio_pci_ethdev.c +2 -0
@@ -220,7 +220,9 @@ static struct rte_pci_driver rte_virtio_net_pci_pmd = {
 
 RTE_INIT(rte_virtio_net_pci_pmd_init)
 {
+#if !(defined(_WIN32) || defined(_WIN64))
 	rte_eal_iopl_init();
+#endif

I think rte_eal_iopl_init() should not be called from this PMD at all, because it is allready called by PCI bus driver before probing.

Also, DPDK code should use #ifdef RTE_EXEC_ENV_WINDOWS, etc. to test if it is building for a specific execution environment.

Quoted 8 lines
drivers/net/virtio/virtio_pci_ethdev.c +0 -0
 	rte_pci_register(&rte_virtio_net_pci_pmd);
 }
 
drivers/net/virtio/virtio_user/virtio_user_dev.h +0 -0
index 819f6463ba..01cb962770 100644

drivers/net/virtio/virtio_user/virtio_user_dev.h line 17

Quoted 7 lines
drivers/net/virtio/virtio_user/virtio_user_dev.h +4 -0
@@ -11,6 +11,10 @@
 #include "../virtio.h"
 #include "../virtio_ring.h"
 
+#if defined(_WIN32) || defined(_WIN64)
+#define PATH_MAX MAX_PATH
+#endif
+

There's rte_os_shim.h for that.

Quoted 3 lines
drivers/net/virtio/virtio_user/virtio_user_dev.h +0 -0
 enum virtio_user_backend_type {
 	VIRTIO_USER_BACKEND_UNKNOWN,
 	VIRTIO_USER_BACKEND_VHOST_USER,
#2 lore raw
Quoted 6 lines · Dmitry Kozlyuk
Quoted 6 lines
drivers/net/virtio/virtio_pci_ethdev.c +0 -0
index 4083853c48..03e9dfab71 100644
@@ -220,7 +220,9 @@ static struct rte_pci_driver rte_virtio_net_pci_pmd = {
 
Quoted 8 lines

drivers/net/virtio/virtio_pci_ethdev.c static struct rte_pci_driver rte_virtio_net_pci_pmd = { · line 225

Quoted 5 lines
drivers/net/virtio/virtio_pci_ethdev.c +2 -0
 RTE_INIT(rte_virtio_net_pci_pmd_init)
 {
+#if !(defined(_WIN32) || defined(_WIN64))
      rte_eal_iopl_init();
+#endif

I think rte_eal_iopl_init() should not be called from this PMD at all, because it is allready called by PCI bus driver before probing.

This call to iopl() must be kept in a pmd constructor (at least on Linux) so that, for example, the interrupt thread inherits the IO priviledge level.

Quoted 7 lines

Also, DPDK code should use #ifdef RTE_EXEC_ENV_WINDOWS, etc. to test if it is building for a specific execution environment.

Quoted 3 lines
drivers/net/virtio/virtio_pci_ethdev.c +0 -0
      rte_pci_register(&rte_virtio_net_pci_pmd);
 }
 
Signature · 1 line
David Marchand