[dpdk-dev] [PATCH v3 0/4] fix the issue that DPDK takes over virtio device blindly
On Fri, Feb 26, 2016 at 7:09 AM, Xie, Huawei wrote: > On 2/24/2016 8:45 PM, Thomas Monjalon wrote: >>> Huawei Xie (4): >>> eal: make the comment more accurate >>> eal: set kdrv to RTE_KDRV_NONE if kernel driver isn't manipulating the >>> device. >>> virtio: return 1 to tell the kernel we don't take over this device >>> virtio: check if kernel driver is manipulating the virtio device >> The virtio PCI code has been refactored. >> Please Huawei, would it be possible to rebase on master? > > OK. Since IO port map is moved to EAL layer, it is not straightforward > like before for virtio PMD to distinguish the reason why port map fails. > We have two choices. Return 1 to the upper layer to say that we don't > take over the device for all the map failures or we check the driver > type, return -1 for UIO/VFIO driver error, return 1 for kernel driver, > which is a bit overelaborate. The important thing is to have eal report "none" driver first (your 2nd patch) , then in ioport_map, "none" driver will trigger the x86 special case (see other discussion [1]). For "uio" drivers, code (when fixed for uio_pci_generic) already does the right stuff. "vfio" is handled. Anything else should fail once we have the "none" driver correctly reported. What did I miss ? [1] http://dpdk.org/ml/archives/dev/2016-February/034035.html -- David Marchand
[dpdk-dev] [PATCH v3 0/4] fix the issue that DPDK takes over virtio device blindly
On 2/26/2016 4:41 PM, David Marchand wrote:
> On Fri, Feb 26, 2016 at 7:09 AM, Xie, Huawei wrote:
>> On 2/24/2016 8:45 PM, Thomas Monjalon wrote:
Huawei Xie (4):
eal: make the comment more accurate
eal: set kdrv to RTE_KDRV_NONE if kernel driver isn't manipulating the
device.
virtio: return 1 to tell the kernel we don't take over this device
virtio: check if kernel driver is manipulating the virtio device
>>> The virtio PCI code has been refactored.
>>> Please Huawei, would it be possible to rebase on master?
>> OK. Since IO port map is moved to EAL layer, it is not straightforward
>> like before for virtio PMD to distinguish the reason why port map fails.
>> We have two choices. Return 1 to the upper layer to say that we don't
>> take over the device for all the map failures or we check the driver
>> type, return -1 for UIO/VFIO driver error, return 1 for kernel driver,
>> which is a bit overelaborate.
> The important thing is to have eal report "none" driver first (your
> 2nd patch) , then in ioport_map, "none" driver will trigger the x86
> special case (see other discussion [1]).
> For "uio" drivers, code (when fixed for uio_pci_generic) already does
> the right stuff.
> "vfio" is handled.
> Anything else should fail once we have the "none" driver correctly reported.
>
in rte_eal_pci_map_device:
case RTE_KDRV_NONE:
#if defined(RTE_ARCH_X86)
ret = pci_ioport_map(dev, bar, p);
#endif
break;
}
in vtpci_init:
if (legacy_virtio_resource_init(dev, hw) < 0) {
if (dev->kdrv == RTE_KDRV_UNKNOWN) {
PMD_INIT_LOG(INFO,
"skip kernel managed virtio device.");
return 1;
}
return -1;
}
in dev_init
ret = vt_pci_init
if (ret)
return ret
> What did I miss ?
>
>
> [1] http://dpdk.org/ml/archives/dev/2016-February/034035.html
>
[dpdk-dev] [PATCH v3 0/4] fix the issue that DPDK takes over virtio device blindly
On 2/24/2016 8:45 PM, Thomas Monjalon wrote: >> Huawei Xie (4): >> eal: make the comment more accurate >> eal: set kdrv to RTE_KDRV_NONE if kernel driver isn't manipulating the >> device. >> virtio: return 1 to tell the kernel we don't take over this device >> virtio: check if kernel driver is manipulating the virtio device > The virtio PCI code has been refactored. > Please Huawei, would it be possible to rebase on master? OK. Since IO port map is moved to EAL layer, it is not straightforward like before for virtio PMD to distinguish the reason why port map fails. We have two choices. Return 1 to the upper layer to say that we don't take over the device for all the map failures or we check the driver type, return -1 for UIO/VFIO driver error, return 1 for kernel driver, which is a bit overelaborate. >
[dpdk-dev] [PATCH v3 0/4] fix the issue that DPDK takes over virtio device blindly
> Huawei Xie (4): > eal: make the comment more accurate > eal: set kdrv to RTE_KDRV_NONE if kernel driver isn't manipulating the > device. > virtio: return 1 to tell the kernel we don't take over this device > virtio: check if kernel driver is manipulating the virtio device The virtio PCI code has been refactored. Please Huawei, would it be possible to rebase on master?
[dpdk-dev] [PATCH v3 0/4] fix the issue that DPDK takes over virtio device blindly
On Wed, Jan 27, 2016 at 11:21:18PM +0800, Huawei Xie wrote: > v3 changes: > change log message to tell user that the virtio device is skipped > due to it is managed by kernel driver, instead of asking user to > unbind it from kernel driver. > > v2 changes: > Remove unnecessary assignment of NULL to dev->data->mac_addrs > Ajust one comment's position > change LOG level from ERR to INFO > > virtio PMD doesn't set RTE_PCI_DRV_NEED_MAPPING in drv_flags of its > eth_driver. It will try igb_uio and PORT IO in turn to configure > virtio device. Even user in guest VM doesn't want to use virtio for > DPDK, virtio PMD will take over the device blindly. > > The more serious problem is kernel driver is still manipulating the > device, which causes driver conflict. > > This patch checks if there is any kernel driver manipulating the > virtio device before virtio PMD uses port IO to configure the device. Series acked: Acked-by: Yuanhan Liu --yliu
[dpdk-dev] [PATCH v3 0/4] fix the issue that DPDK takes over virtio device blindly
v3 changes: change log message to tell user that the virtio device is skipped due to it is managed by kernel driver, instead of asking user to unbind it from kernel driver. v2 changes: Remove unnecessary assignment of NULL to dev->data->mac_addrs Ajust one comment's position change LOG level from ERR to INFO virtio PMD doesn't set RTE_PCI_DRV_NEED_MAPPING in drv_flags of its eth_driver. It will try igb_uio and PORT IO in turn to configure virtio device. Even user in guest VM doesn't want to use virtio for DPDK, virtio PMD will take over the device blindly. The more serious problem is kernel driver is still manipulating the device, which causes driver conflict. This patch checks if there is any kernel driver manipulating the virtio device before virtio PMD uses port IO to configure the device. Huawei Xie (4): eal: make the comment more accurate eal: set kdrv to RTE_KDRV_NONE if kernel driver isn't manipulating the device. virtio: return 1 to tell the kernel we don't take over this device virtio: check if kernel driver is manipulating the virtio device drivers/net/virtio/virtio_ethdev.c | 14 -- lib/librte_eal/common/eal_common_pci.c | 8 lib/librte_eal/linuxapp/eal/eal_pci.c | 2 +- 3 files changed, 17 insertions(+), 7 deletions(-) -- 1.8.1.4
