net/mlx5: fix link state update
PATCH 1/1 net/mlx5: fix link state update
mlx5 PMD refuses to update link state if link speed is defined but status is down or if link speed is undefined but status is up, even if the ioctl() succeeded.
This prevents application to detect link up/down event, especially when the link speed is not correctly detected.
As link speed is nice to have whereas link status is mandatory for operations, always update link state regardless of link speed. The application can then check link speed if needs be.
---
Following the discussion on dpdk-users [1], I am submitting a tentative patch.
[1] http://mails.dpdk.org/archives/users/2020-March/thread.html#4744
drivers/net/mlx5/mlx5_ethdev.c | 10 ---------- 1 file changed, 10 deletions(-)
drivers/net/mlx5/mlx5_ethdev.c +0 -10
@@ -896,11 +896,6 @@ mlx5_link_update_unlocked_gset(struct rte_eth_dev *dev, ETH_LINK_HALF_DUPLEX : ETH_LINK_FULL_DUPLEX); dev_link.link_autoneg = !(dev->data->dev_conf.link_speeds Ð_LINK_SPEED_FIXED ); -if (((dev_link.link_speed && !dev_link.link_status) || - (!dev_link.link_speed && dev_link.link_status))) { - rte_errno =EAGAIN ; -return -rte_errno; - } *link = dev_link;return 0 ; }
@@ -1032,11 +1027,6 @@ mlx5_link_update_unlocked_gs(struct rte_eth_dev *dev, ETH_LINK_HALF_DUPLEX : ETH_LINK_FULL_DUPLEX); dev_link.link_autoneg = !(dev->data->dev_conf.link_speeds Ð_LINK_SPEED_FIXED ); -if (((dev_link.link_speed && !dev_link.link_status) || - (!dev_link.link_speed && dev_link.link_status))) { - rte_errno =EAGAIN ; -return -rte_errno; - } *link = dev_link;return 0 ; }
Signature · 1 line
2.17.1
Hi
From: Benoît Ganne
Quoted 5 lines
mlx5 PMD refuses to update link state if link speed is defined but status is down or if link speed is undefined but status is up, even if the
ioctl()succeeded.
This prevents application to detect link up/down event, especially when the link speed is not correctly detected.
Do you use the wait option? Or no wait?
Quoted 3 lines
As link speed is nice to have whereas link status is mandatory for operations, always update link state regardless of link speed. The application can then check link speed if needs be.
Is it documented well? I didn't find doc\description says link speed is best effort.
PMD cannot guess whether link speed is mandatory for the user or not....
I think, at least you should update ethdev relevant API descriptions and get agreement from more PMD maintainers...
Quoted 49 lines
Signed-off-by Benoît Ganne <[email protected]>---
Following the discussion on dpdk-users [1], I am submitting a tentative patch.
[1]
https://eur03.safelinks.protection.outlook.com/?url=http%3A%2F%2Fmails.
dpdk.org%2Farchives%2Fusers%2F2020- March%2Fthread.html%234744&data=02%7C01%7Cmatan%40mellanox. com%7C2cb881fc068c434e2f6b08d7d273c924%7Ca652971c7d2e4d9ba6a4d14 9256f461b%7C0%7C0%7C637209267055735109&sdata=WV%2Fsd%2B%2 BKssI3d8uxMf8cqackb%2FHrpqRIWOos2BWynU4%3D&reserved=0drivers/net/mlx5/mlx5_ethdev.c | 10 ---------- 1 file changed, 10 deletions(-)(quoted diff) +0
-0
b/drivers/net/mlx5/mlx5_ethdev.cindexd7d3bc73c..c15f4d62b100644drivers/net/mlx5/mlx5_ethdev.c +0
-5@@ -896,11 +896,6 @@ mlx5_link_update_unlocked_gset(structrte_eth_dev *dev, ETH_LINK_HALF_DUPLEX : ETH_LINK_FULL_DUPLEX); dev_link.link_autoneg = !(dev->data->dev_conf.link_speeds Ð_LINK_SPEED_FIXED ); -if (((dev_link.link_speed && !dev_link.link_status) || - (!dev_link.link_speed && dev_link.link_status))) { - rte_errno =EAGAIN ; -return -rte_errno; - } *link = dev_link;drivers/net/mlx5/mlx5_ethdev.c +0
-5return 0 ; }@@ -1032,11 +1027,6 @@ mlx5_link_update_unlocked_gs(structrte_eth_dev *dev, ETH_LINK_HALF_DUPLEX : ETH_LINK_FULL_DUPLEX); dev_link.link_autoneg = !(dev->data->dev_conf.link_speeds Ð_LINK_SPEED_FIXED ); -if (((dev_link.link_speed && !dev_link.link_status) || - (!dev_link.link_speed && dev_link.link_status))) { - rte_errno =EAGAIN ; -return -rte_errno; - } *link = dev_link;return 0;
}
--
2.17.1
Quoted 10 lines
From: Matan Azrad <[email protected]>
Quoted 6 lines
From: Benoît Ganne
mlx5 PMD refuses to update link state if link speed is defined but status is down or if link speed is undefined but status is up, even if theioctl()succeeded.
This prevents application to detect link up/down event, especially when the link speed is not correctly detected.Do you use the wait option? Or no wait?
We are using the no wait option.
Quoted 7 lines
Quoted 3 lines
As link speed is nice to have whereas link status is mandatory for operations, always update link state regardless of link speed. The application can then check link speed if needs be.
Is it documented well? I didn't find doc\description says link speed is best effort.
PMD cannot guess whether link speed is mandatory for the user or not....
I do not think it is documented and I suppose at the end of the day it depends from the HW and configurations...
My commit message is misleading, my apologizes. What I meant was to let the app decide whether it should retry or not, based on the data it gets.
Right now, the PMD *prevents* the app to get link state if the link speed is undefined even if the app does not care about link speed. It makes an assumption on behalf of the apps. Moreover, it seems to be one of the few PMDs to make this assumption: at least the mlx4 PMD and the i40e PMD does not seem to make this assumption. This is a problem for our app (fd.io VPP) but I was also told about other apps having the same issue too.
Quoted 2 lines
I think, at least you should update ethdev relevant API descriptions and get agreement from more PMD maintainers...
I hope to get feedback 😊 it is precisely the goal of this.
ben
Hi
From: Benoit Ganne (bganne) <[email protected]>
Quoted 9 lines
Quoted 8 lines
From: Matan Azrad <[email protected]>
Quoted 6 lines
From: Benoît Ganne
mlx5 PMD refuses to update link state if link speed is defined but status is down or if link speed is undefined but status is up, even if theioctl()succeeded.
This prevents application to detect link up/down event, especially when the link speed is not correctly detected.
Quoted 3 lines
Quoted 1 line
Do you use the wait option? Or no wait?
We are using the no wait option.
I suggest to call again if failed for N retries time.
Quoted 8 lines
Quoted 7 lines
Quoted 3 lines
As link speed is nice to have whereas link status is mandatory for operations, always update link state regardless of link speed. The application can then check link speed if needs be.
Is it documented well? I didn't find doc\description says link speed is best effort.
PMD cannot guess whether link speed is mandatory for the user or not....
Quoted 6 lines
I do not think it is documented and I suppose at the end of the day it depends from the HW and configurations...
My commit message is misleading, my apologizes. What I meant was to let the app decide whether it should retry or not, based on the data it gets. Right now, the PMD *prevents* the app to get link state if the link speed is undefined even if the app does not care about link speed.
In mlx5 this is not the case, we have no one updated and second not - there are going together:
You can see that we have 2 different system calls: 1 to get up\down and second to get link speed.
If link speed doesn't appropriate to the link state it may say that something was changed between the calls and the link status we got from the first call is not correct anymore.
In this case, we should call both calls again, that’s what we are doing in "nowait" option.
If the user doesn't want "nowait" option, (means PMD is not allowed to take more time for response) he should call again when the callback failed in the time and retries manner the user prefers.
Quoted 13 lines
It makes an
assumption on behalf of the apps. Moreover, it seems to be one of the few PMDs to make this assumption: at least the mlx4 PMD and the i40e PMD does not seem to make this assumption.
This is a problem for our app (fd.io VPP) but I was also told about other apps having the same issue too.Quoted 2 lines
I think, at least you should update ethdev relevant API descriptions and get agreement from more PMD maintainers...
I hope to get feedback 😊 it is precisely the goal of this.
ben
Hi Matan,
Quoted 8 lines
Quoted 7 lines
Quoted 6 lines
Quoted 5 lines
mlx5 PMD refuses to update link state if link speed is defined but status is down or if link speed is undefined but status is up, even if the
ioctl()succeeded.
This prevents application to detect link up/down event, especially when the link speed is not correctly detected.Do you use the wait option? Or no wait?
We are using the no wait option.
I suggest to call again if failed for N retries time.
Unfortunately it will not solve our problem: if link speed is undefined but the link is up then the test '!dev_link.link_speed && dev_link.link_status' at http://git.dpdk.org/dpdk/tree/drivers/net/mlx5/mlx5_ethdev.c#n899 will always be true and the function will always return EAGAIN. This actually happens in Azure with CX4-Lx VFs.
Quoted 5 lines
Quoted 4 lines
What I meant was to let the app decide whether it should retry or not, based on the data it gets.
Right now, the PMD *prevents* the app to get link state if the link speed is undefined even if the app does not care about link speed.
Quoted 12 lines
In mlx5 this is not the case, we have no one updated and second not - there are going together:
You can see that we have 2 different system calls: 1 to get up\down and second to get link speed.
If link speed doesn't appropriate to the link state it may say that something was changed between the calls and the link status we got from the first call is not correct anymore.
In this case, we should call both calls again, that’s what we are doing in "nowait" option.
If the user doesn't want "nowait" option, (means PMD is not allowed to take more time for response) he should call again when the callback failed in the time and retries manner the user prefers.
Ok, now I understand the logic behind the current behavior: the 2 syscalls being not atomics, you try to detect inconsistencies that way.
But if the link speed is undefined, then the state will never be correctly updated.
I still believe it is unnecessarily heavy-handed: in most networking application I have seen (and I have 2 examples of current shipping networking products), a missing link speed is not critical whereas link being reported as down means no traffic flowing.
Best
ben
Hi
From: Benoit Ganne
Quoted 44 lines
Sent: Monday, March 30, 2020 3:03 PM
To: Matan Azrad <[email protected]>; [email protected]
Subject: RE: [dpdk-dev] [PATCH] net/mlx5: fix link state updateHi Matan,
Quoted 8 lines
Quoted 7 lines
Quoted 6 lines
Quoted 5 lines
mlx5 PMD refuses to update link state if link speed is defined but status is down or if link speed is undefined but status is up, even if the
ioctl()succeeded.
This prevents application to detect link up/down event, especially when the link speed is not correctly detected.Do you use the wait option? Or no wait?
We are using the no wait option.
I suggest to call again if failed for N retries time.
Unfortunately it will not solve our problem: if link speed is undefined but the link is up then the test '!dev_link.link_speed && dev_link.link_status' at
https://eur03.safelinks.protection.outlook.com/?url=http%3A%2F%2Fgit.dp
dk.org%2Fdpdk%2Ftree%2Fdrivers%2Fnet%2Fmlx5%2Fmlx5_ethdev.c%23n8 99&data=02%7C01%7Cmatan%40mellanox.com%7C8a0fb9f9ba94422a4a 4208d7d4a2539c%7Ca652971c7d2e4d9ba6a4d149256f461b%7C0%7C0%7C637 211665962283891&sdata=%2FExOJyhm0Bi4kkp434P0U4kRpQs7U0knuTa kvsWFO5Q%3D&reserved=0 will always be true and the function will always return EAGAIN.
This actually happens in Azure with CX4-Lx VFs.Quoted 5 lines
Quoted 4 lines
What I meant was to let the app decide whether it should retry or not, based on the data it gets.
Right now, the PMD *prevents* the app to get link state if the link speed is undefined even if the app does not care about link speed.Quoted 12 lines
In mlx5 this is not the case, we have no one updated and second not - there are going together:
You can see that we have 2 different system calls: 1 to get up\down and second to get link speed.
If link speed doesn't appropriate to the link state it may say that something was changed between the calls and the link status we got from the first call is not correct anymore.
In this case, we should call both calls again, that’s what we are doing in "nowait" option.
If the user doesn't want "nowait" option, (means PMD is not allowed to take more time for response) he should call again when the callback failed in the time and retries manner the user prefers.
Quoted 4 lines
Ok, now I understand the logic behind the current behavior: the 2 syscalls being not atomics, you try to detect inconsistencies that way. But if the link speed is undefined, then the state will never be correctly updated.
Why link speed is undefined?
Old kernel?
Kernel mlx5 driver issue?
Do you know?
Quoted 7 lines
I still believe it is unnecessarily heavy-handed: in most networking application I have seen (and I have 2 examples of current shipping networking products), a missing link speed is not critical whereas link being reported as down means no traffic flowing.
Best
ben
Quoted 4 lines
Why link speed is undefined?
Old kernel?
Kernel mlx5 driver issue?
Do you know?
Seems to be a bug in kernel mlx5 driver: http://mails.dpdk.org/archives/users/2020-March/004758.html
I still think this test is too strict, and from what I can see mlx4 PMD do not implement it (even if it uses 2 syscalls too).
ben
Hi
From: Benoit Ganne (bganne)
Quoted 5 lines
Quoted 4 lines
Why link speed is undefined?
Old kernel?
Kernel mlx5 driver issue?
Do you know?
Quoted 9 lines
Seems to be a bug in kernel mlx5 driver:
https://eur03.safelinks.protection.outlook.com/?url=http%3A%2F%2Fmails.
dpdk.org%2Farchives%2Fusers%2F2020- March%2F004758.html&data=02%7C01%7Cmatan%40mellanox.com%7C 69ce3bfd96484267869908d7d4b1ca89%7Ca652971c7d2e4d9ba6a4d149256f46 1b%7C0%7C0%7C637211732388079395&sdata=%2BnfjST6fM4zoIYwJz7L Z2zL3rGmmOFXbpW0fK7Weps4%3D&reserved=0
I still think this test is too strict, and from what I can see mlx4 PMD do not implement it (even if it uses 2 syscalls too).
Yes, Doesn't mean this is not a bug in mlx4😊
Thanks Ben.
So we have 2 orthogonal efforts here:
1. To check with our driver team what is the issue to report a VF link speed : For me 😊
2. To change documentation of ethdev API to say that only link status is mandatory report. (all others "nice to have" or "best effort"):
This is for you - I suggest to send prior patch to ethdev for this and let the community to decide.
Are you agree?
Quoted 1 line
ben
Hi Matan,
Quoted 2 lines
Quoted 1 line
Seems to be a bug in kernel mlx5 driver:
1. To check with our driver team what is the issue to report a VF link
Note that even with the kernel patch, there still seem to be a valid case where it can return SPEED_UNKNOWN. If this is true, then I don't think we can rely on valid link speed to detect whether the info are correct.
ben
Hi
From: Benoit Ganne
Quoted 4 lines
Hi Matan,
Quoted 2 lines
Quoted 1 line
Seems to be a bug in kernel mlx5 driver:
1. To check with our driver team what is the issue to report a VF link
Yes, driver team said the 3 kernel patches should solve the issue.
Quoted 4 lines
Note that even with the kernel patch, there still seem to be a valid case where it can return
SPEED_UNKNOWN.
If this is true, then I don't think we can rely on valid link speed to detect whether the info are correct.
As I said I'm not against this approach, but need to define the API well.
Quoted 1 line
ben
30/03/2020 18:13, Matan Azrad:
Quoted 4 lines
2. To change documentation of ethdev API to say that only link status is mandatory report. (all others "nice to have" or "best effort"):
This is for you - I suggest to send prior patch to ethdev for this and let the community to decide.
Matan, Benoit, I guess nobody tried to update ethdev API doc yet? I can do it if you prefer.
Quoted 6 lines
Quoted 4 lines
2. To change documentation of ethdev API to say that only link status is mandatory report. (all others "nice to have" or "best effort"):
This is for you - I suggest to send prior patch to ethdev for this and let the community to decide.Matan, Benoit, I guess nobody tried to update ethdev API doc yet? I can do it if you prefer.
My understanding was that this [1] was basically saying 'returning UNKNOWN is ok'. But if this is not what you had in mind, I'll be glad you can take care of it 😊
[1] http://mails.dpdk.org/archives/dev/2020-April/thread.html#161538
Best
ben
+@Asaf Penso
From: Benoit Ganne
Quoted 19 lines
Quoted 6 lines
Quoted 4 lines
2. To change documentation of ethdev API to say that only link status is mandatory report. (all others "nice to have" or "best effort"):
This is for you - I suggest to send prior patch to ethdev for this and let the community to decide.Matan, Benoit, I guess nobody tried to update ethdev API doc yet? I can do it if you prefer.
My understanding was that this [1] was basically saying 'returning UNKNOWN is ok'. But if this is not what you had in mind, I'll be glad you can take care of it 😊
[1]
https://eur03.safelinks.protection.outlook.com/?url=http%3A%2F%2Fmails.dpdk
.org%2Farchives%2Fdev%2F2020- April%2Fthread.html%23161538&data=02%7C01%7Cmatan%40mellanox.c om%7Cee2c8eebc00d4ea3106b08d7daf97473%7Ca652971c7d2e4d9ba6a4d149 256f461b%7C0%7C0%7C637218637252717747&sdata=JSFaHkmSzE1tWB7 DqEaqVI5B1pIZuIdDrfw0SFyUE8s%3D&reserved=0
Quoted 2 lines
Best
ben
Hi
Now, when the API is updated that the speed value is best effort using ETH_SPEED_NUM_UNKNOWN new value,
You can go for your approach:
In case the link is up and speed is invalid by kernel we can return link up and ETH_SPEED_NUM_UNKNOWN.
In case applications wants a valid speed they should call the API again. Need to update it in mlx5 release notes section.
The original patch should be rebased and updated.
Matan
PATCH v2 1/1 net/mlx5: fix link state update
In reply to Benoît GanneFrom: Benoît Ganne <[email protected]>
mlx5 PMD refuses to update link state if link speed is defined but status is down or if link speed is undefined but status is up, even if the ioctl() succeeded.
This prevents application to detect link up/down event, especially when the link speed is not correctly detected.
As link speed is nice to have whereas link status is mandatory for operations, always update link state regardless of link speed. The application can then check link speed if needs be.
--- doc/guides/rel_notes/release_20_11.rst | 1 + drivers/net/mlx5/linux/mlx5_ethdev_os.c | 10 ---------- 2 files changed, 1 insertion(+), 10 deletions(-)
doc/guides/rel_notes/release_20_11.rst +1 -0
@@ -207,6 +207,7 @@ New Features by rte_flow API. * Added support of Age action query. * Added support of multi-ports hairpin. + * Allow unknown link speed. Updated Mellanox mlx5 vDPA driver:
drivers/net/mlx5/linux/mlx5_ethdev_os.c +0 -10
@@ -425,11 +425,6 @@ mlx5_link_update_unlocked_gset(struct rte_eth_dev *dev, ETH_LINK_HALF_DUPLEX : ETH_LINK_FULL_DUPLEX); dev_link.link_autoneg = !(dev->data->dev_conf.link_speeds Ð_LINK_SPEED_FIXED ); -if (((dev_link.link_speed && !dev_link.link_status) || - (!dev_link.link_speed && dev_link.link_status))) { - rte_errno =EAGAIN ; -return -rte_errno; - } *link = dev_link;return 0 ; }
@@ -571,11 +566,6 @@ mlx5_link_update_unlocked_gs(struct rte_eth_dev *dev, ETH_LINK_HALF_DUPLEX : ETH_LINK_FULL_DUPLEX); dev_link.link_autoneg = !(dev->data->dev_conf.link_speeds Ð_LINK_SPEED_FIXED ); -if (((dev_link.link_speed && !dev_link.link_status) || - (!dev_link.link_speed && dev_link.link_status))) { - rte_errno =EAGAIN ; -return -rte_errno; - } *link = dev_link;return 0 ; }
Signature · 1 line
2.17.1
Hi,
Quoted 20 lines
-----Original Message-----
From: dev <[email protected]> On Behalf Of Raslan Darawsheh
Sent: Thursday, November 19, 2020 4:20 PM
To: [email protected]
Cc: Matan Azrad <[email protected]>; Slava Ovsiienko <[email protected]>; NBU-Contact-Thomas Monjalon <[email protected]>; Benoît Ganne <[email protected]>
Subject: [dpdk-dev] [PATCH v2] net/mlx5: fix link state updateFrom: Benoît Ganne <[email protected]>
mlx5 PMD refuses to update link state if link speed is defined but status is down or if link speed is undefined but status is up, even if the
ioctl()succeeded.
This prevents application to detect link up/down event, especially when the link speed is not correctly detected.
As link speed is nice to have whereas link status is mandatory for operations, always update link state regardless of link speed. The application can then check link speed if needs be.
Quoted 2 lines
Signed-off-by Benoît Ganne <[email protected]>Signed-off-by Raslan Darawsheh <[email protected]>
Patch applied to next-net-mlx,
Kindest regards,
Raslan Darawsheh
Quoted 11 lines · Raslan Darawsheh
From: Benoît Ganne <[email protected]>
mlx5 PMD refuses to update link state if link speed is defined but status is down or if link speed is undefined but status is up, even if the
ioctl()succeeded.
This prevents application to detect link up/down event, especially when the link speed is not correctly detected.
As link speed is nice to have whereas link status is mandatory for operations, always update link state regardless of link speed. The application can then check link speed if needs be.
Hi Raslan, Matan,
Can you please provide the Fixes tag? Also should this fix backported?
Quoted 2 lines
Signed-off-by Benoît Ganne <[email protected]>Signed-off-by Raslan Darawsheh <[email protected]>
<...>
19/11/2020 18:48, Ferruh Yigit:
Quoted 13 lines · Raslan Darawsheh
Quoted 11 lines
From: Benoît Ganne <[email protected]>
mlx5 PMD refuses to update link state if link speed is defined but status is down or if link speed is undefined but status is up, even if the
ioctl()succeeded.
This prevents application to detect link up/down event, especially when the link speed is not correctly detected.
As link speed is nice to have whereas link status is mandatory for operations, always update link state regardless of link speed. The application can then check link speed if needs be.
Quoted 3 lines
Hi Raslan, Matan,
Can you please provide the Fixes tag? Also should this fix backported?
I think it should not be backported because it is a behaviour change relying on API doc change.
Quoted 14 lines · Thomas Monjalon
19/11/2020 18:48, Ferruh Yigit:
Quoted 13 lines · Raslan Darawsheh
Quoted 11 lines
From: Benoît Ganne <[email protected]>
mlx5 PMD refuses to update link state if link speed is defined but status is down or if link speed is undefined but status is up, even if the
ioctl()succeeded.
This prevents application to detect link up/down event, especially when the link speed is not correctly detected.
As link speed is nice to have whereas link status is mandatory for operations, always update link state regardless of link speed. The application can then check link speed if needs be.
Quoted 7 lines
Quoted 3 lines
Hi Raslan, Matan,
Can you please provide the Fixes tag? Also should this fix backported?
I think it should not be backported because it is a behaviour change relying on API doc change.
As far as I understand, API change you mention is making 'ETH_SPEED_NUM_UNKNOWN' speed value a valid value return value.
Commit log doesn't give the context that it relies on an ethdev behavior change,
it sounds like it is fixing behavior in the driver, it is not possible to figure out the relation without digging the mail list discussions.
It would be nice to update the commit log to give more details, I think problem is clear but can good to add why the check was there at first place and why it can be removed now, etc..
Btw with LTS, with the faulty kernel driver, PMD still won't able to get link speed which will prevent returning link status. Or is the kernel driver fixed and no need to worry about this anymore?
20/11/2020 11:51, Ferruh Yigit:
Quoted 26 lines · Thomas Monjalon
Quoted 14 lines
19/11/2020 18:48, Ferruh Yigit:
Quoted 13 lines · Raslan Darawsheh
Quoted 11 lines
From: Benoît Ganne <[email protected]>
mlx5 PMD refuses to update link state if link speed is defined but status is down or if link speed is undefined but status is up, even if the
ioctl()succeeded.
This prevents application to detect link up/down event, especially when the link speed is not correctly detected.
As link speed is nice to have whereas link status is mandatory for operations, always update link state regardless of link speed. The application can then check link speed if needs be.Quoted 7 lines
Quoted 3 lines
Hi Raslan, Matan,
Can you please provide the Fixes tag? Also should this fix backported?
I think it should not be backported because it is a behaviour change relying on API doc change.
As far as I understand, API change you mention is making '
ETH_SPEED_NUM_UNKNOWN' speed value a valid value return value.
Quoted 6 lines
Commit log doesn't give the context that it relies on an ethdev behavior change, it sounds like it is fixing behavior in the driver, it is not possible to figure out the relation without digging the mail list discussions.
It would be nice to update the commit log to give more details, I think problem is clear but can good to add why the check was there at first place and why it can be removed now, etc..
Yes I can help improving the explanations.
Quoted 2 lines
Btw with LTS, with the faulty kernel driver, PMD still won't able to get link speed which will prevent returning link status.
With this PMD change, the link status is returned even if link speed is not available.
Quoted 1 line
Or is the kernel driver fixed and no need to worry about this anymore?
Yes the kernel driver is fixed since a long time.
Quoted 34 lines · Thomas Monjalon
20/11/2020 11:51, Ferruh Yigit:
Quoted 26 lines · Thomas Monjalon
Quoted 14 lines
19/11/2020 18:48, Ferruh Yigit:
Quoted 13 lines · Raslan Darawsheh
Quoted 11 lines
From: Benoît Ganne <[email protected]>
mlx5 PMD refuses to update link state if link speed is defined but status is down or if link speed is undefined but status is up, even if the
ioctl()succeeded.
This prevents application to detect link up/down event, especially when the link speed is not correctly detected.
As link speed is nice to have whereas link status is mandatory for operations, always update link state regardless of link speed. The application can then check link speed if needs be.Quoted 7 lines
Quoted 3 lines
Hi Raslan, Matan,
Can you please provide the Fixes tag? Also should this fix backported?
I think it should not be backported because it is a behaviour change relying on API doc change.
As far as I understand, API change you mention is making '
ETH_SPEED_NUM_UNKNOWN' speed value a valid value return value.Quoted 6 lines
Commit log doesn't give the context that it relies on an ethdev behavior change, it sounds like it is fixing behavior in the driver, it is not possible to figure out the relation without digging the mail list discussions.
It would be nice to update the commit log to give more details, I think problem is clear but can good to add why the check was there at first place and why it can be removed now, etc..
Quoted 8 lines
Yes I can help improving the explanations.
Quoted 2 lines
Btw with LTS, with the faulty kernel driver, PMD still won't able to get link speed which will prevent returning link status.
With this PMD change, the link status is returned even if link speed is not available.
That is why I am asking if this patch should be backported :)
Quoted 6 lines
Quoted 1 line
Or is the kernel driver fixed and no need to worry about this anymore?
Yes the kernel driver is fixed since a long time.
Hi,
Quoted 51 lines
-----Original Message-----
From: Ferruh Yigit <[email protected]>
Sent: Friday, November 20, 2020 4:21 PM
To: NBU-Contact-Thomas Monjalon <[email protected]>
Cc: Raslan Darawsheh <[email protected]>; Matan Azrad <[email protected]>; Slava Ovsiienko <[email protected]>; Benoît Ganne <[email protected]>; [email protected]
Subject: Re: [dpdk-dev] [PATCH v2] net/mlx5: fix link state updateQuoted 10 lines · Thomas Monjalon
20/11/2020 11:51, Ferruh Yigit:
Quoted 9 lines · Thomas Monjalon
Quoted 8 lines
19/11/2020 18:48, Ferruh Yigit:
Quoted 7 lines · Raslan Darawsheh
Quoted 6 lines
From: Benoît Ganne <[email protected]>
mlx5 PMD refuses to update link state if link speed is defined but status is down or if link speed is undefined but status is up, even if the
ioctl()succeeded.
This prevents application to detect link up/down event, especiallywhen
Quoted 8 lines
Quoted 8 lines
Quoted 8 lines
Quoted 8 lines
Quoted 5 lines
the link speed is not correctly detected.
As link speed is nice to have whereas link status is mandatory for operations, always update link state regardless of link speed. The application can then check link speed if needs be.Hi Raslan, Matan,
Quoted 7 lines
Quoted 7 lines
Quoted 5 lines
Quoted 1 line
Can you please provide the Fixes tag? Also should this fix backported?
I think it should not be backported because it is a behaviour change relying on API doc change.
As far as I understand, API change you mention is making
'
ETH_SPEED_NUM_UNKNOWN'Quoted 3 lines
Quoted 3 lines
speed value a valid value return value.
Commit log doesn't give the context that it relies on an ethdev behavior
change,
Quoted 3 lines
Quoted 3 lines
it sounds like it is fixing behavior in the driver, it is not possible to figure out the relation without digging the mail list discussions.
It would be nice to update the commit log to give more details, I thinkproblem
Quoted 1 line
Quoted 1 line
is clear but can good to add why the check was there at first place and
why it
Quoted 4 lines
Quoted 1 line
can be removed now, etc..
Yes I can help improving the explanations.
Quoted 15 lines
Quoted 6 lines
Quoted 2 lines
Btw with LTS, with the faulty kernel driver, PMD still won't able to get link speed which will prevent returning link status.
With this PMD change, the link status is returned even if link speed is not available.
That is why I am asking if this patch should be backported :)
Quoted 6 lines
Quoted 1 line
Or is the kernel driver fixed and no need to worry about this anymore?
Yes the kernel driver is fixed since a long time.
Thanks for your comments and reviews,
I'll send a v3 changing the commit log accordingly.
Kindest regards,
Raslan Darawsheh
PATCH v3 1/1 net/mlx5: allow unknown link speed
In reply to Raslan DarawshehFrom: Benoît Ganne <[email protected]>
mlx5 PMD refuses to update link state if link speed is defined but status is down or if link speed is undefined but status is up, even if the ioctl() succeeded.
This prevents application to detect link up/down event, especially when the link speed is not correctly detected.
Commit [1] allowed returning unknown link speed, so now pmd allow the return of unknown link speed in the above case.
Due to some old kernel driver bug, link speed wasn't detected properly.
[1] http://git.dpdk.org/dpdk/commit/?id=810b17d116f03
---
v2: rebase the code and add doc update
v3: reword commit log, and return correct link speed
--- doc/guides/rel_notes/release_20_11.rst | 1 + drivers/net/mlx5/linux/mlx5_ethdev_os.c | 16 +++------------- 2 files changed, 4 insertions(+), 13 deletions(-)
doc/guides/rel_notes/release_20_11.rst +1 -0
@@ -207,6 +207,7 @@ New Features by rte_flow API. * Added support of Age action query. * Added support of multi-ports hairpin. + * Allow unknown link speed. Updated Mellanox mlx5 vDPA driver:
drivers/net/mlx5/linux/mlx5_ethdev_os.c +3 -13
@@ -405,7 +405,7 @@ mlx5_link_update_unlocked_gset(struct rte_eth_dev *dev, } link_speed =ethtool_cmd_speed (&edata);if (link_speed ==-1 ) - dev_link.link_speed =ETH_SPEED_NUM_NONE ; + dev_link.link_speed =ETH_SPEED_NUM_UNKNOWN ;else dev_link.link_speed = link_speed; priv->link_speed_capa =0 ;
@@ -425,11 +425,6 @@ mlx5_link_update_unlocked_gset(struct rte_eth_dev *dev, ETH_LINK_HALF_DUPLEX : ETH_LINK_FULL_DUPLEX); dev_link.link_autoneg = !(dev->data->dev_conf.link_speeds Ð_LINK_SPEED_FIXED ); -if (((dev_link.link_speed && !dev_link.link_status) || - (!dev_link.link_speed && dev_link.link_status))) { - rte_errno =EAGAIN ; -return -rte_errno; - } *link = dev_link;return 0 ; }
@@ -517,8 +512,8 @@ mlx5_link_update_unlocked_gs(struct rte_eth_dev *dev, dev->data->port_id, strerror(rte_errno));return ret; } - dev_link.link_speed = (ecmd->speed ==UINT32_MAX ) ?ETH_SPEED_NUM_NONE : - ecmd->speed; + dev_link.link_speed = (ecmd->speed ==UINT32_MAX ) ? +ETH_SPEED_NUM_UNKNOWN : ecmd->speed; sc = ecmd->link_mode_masks[0 ] | ((uint64_t )ecmd->link_mode_masks[1 ] <<32 ); priv->link_speed_capa =0 ;
@@ -571,11 +566,6 @@ mlx5_link_update_unlocked_gs(struct rte_eth_dev *dev, ETH_LINK_HALF_DUPLEX : ETH_LINK_FULL_DUPLEX); dev_link.link_autoneg = !(dev->data->dev_conf.link_speeds Ð_LINK_SPEED_FIXED ); -if (((dev_link.link_speed && !dev_link.link_status) || - (!dev_link.link_speed && dev_link.link_status))) { - rte_errno =EAGAIN ; -return -rte_errno; - } *link = dev_link;return 0 ; }
Signature · 1 line
2.17.1
Re: [dpdk-dev] [PATCH v3] net/mlx5: allow unknown link speed
Hi,
Quoted 35 lines
-----Original Message-----
From: dev <[email protected]> On Behalf Of Raslan Darawsheh
Sent: Sunday, November 22, 2020 12:04 PM
To: [email protected]
Cc: Matan Azrad <[email protected]>; Slava Ovsiienko <[email protected]>; NBU-Contact-Thomas Monjalon <[email protected]>; Benoît Ganne <[email protected]>; Benoît Ganne <[email protected]>
Subject: [dpdk-dev] [PATCH v3] net/mlx5: allow unknown link speedFrom: Benoît Ganne <[email protected]>
mlx5 PMD refuses to update link state if link speed is defined but status is down or if link speed is undefined but status is up, even if the
ioctl()succeeded.
This prevents application to detect link up/down event, especially when the link speed is not correctly detected.Commit [1] allowed returning unknown link speed, so now pmd allow the return of unknown link speed in the above case.
Due to some old kernel driver bug, link speed wasn't detected properly.
[1]
https://nam11.safelinks.protection.outlook.com/?url=http%3A%2F%2Fgit.dp
dk.org%2Fdpdk%2Fcommit%2F%3Fid%3D810b17d116f03&data=04%7C0 1%7Crasland%40nvidia.com%7C3965964e417641f146d008d88ece1678%7C430 83d15727340c1b7db39efd9ccc17a%7C0%7C0%7C637416363079871052%7CUn known%7CTWFpbGZsb3d8eyJWIjoiMC4wLjAwMDAiLCJQIjoiV2luMzIiLCJBTiI6 Ik1haWwiLCJXVCI6Mn0%3D%7C1000&sdata=t0Qj8hEjCBYi1bzy25lJAGTr bI0C4l6oImqAe174rNA%3D&reserved=0Signed-off-by Benoît Ganne <[email protected]>Signed-off-by Raslan Darawsheh <[email protected]>Acked-by Matan Azrad <[email protected]>
Quoted 7 lines
---
v2: rebase the code and add doc update
v3: reword commit log, and return correct link speed--- doc/guides/rel_notes/release_20_11.rst | 1 + drivers/net/mlx5/linux/mlx5_ethdev_os.c | 16 +++------------- 2 files changed, 4 insertions(+), 13 deletions(-)
Reverted V2 and applied v3 to next-net-mlx,
Kindest regards,
Raslan Darawsheh