Dear Linux folks,
Trying to decrease the resume time of Linux 5.3-rc3 on the Dell OptiPlex
5040 with the device below
$ lspci -nn -s 00:1f.6
00:1f.6 Ethernet controller [0200]: Intel Corporation Ethernet Connection (2) I219-V [8086:15b8] (rev 31)
pm-graph’s script `sleepgraph.py` shows, that the driver *e1000e* takes
around 400 ms, which is quite a lot. The call graph trace shows that
`e1000e_read_phy_reg_mdic()` is responsible for a lot of those. From
`drivers/net/ethernet/intel/e1000e/phy.c` [1]:
for (i = 0; i < (E1000_GEN_POLL_TIMEOUT * 3); i++) {
udelay(50);
mdic = er32(MDIC);
if (mdic & E1000_MDIC_READY)
break;
}
if (!(mdic & E1000_MDIC_READY)) {
e_dbg("MDI Read did not complete\n");
return -E1000_ERR_PHY;
}
if (mdic & E1000_MDIC_ERROR) {
e_dbg("MDI Error\n");
return -E1000_ERR_PHY;
}
Unfortunately, errors are not logged if dynamic debug is disabled,
so rebuilding the Linux kernel with `CONFIG_DYNAMIC_DEBUG`, and
echo "file drivers/net/ethernet/* +p" | sudo tee /sys/kernel/debug/dynamic_debug/control
I got the messages below.
[ 4159.204192] e1000e 0000:00:1f.6 net00: MDI Error
[ 4160.267950] e1000e 0000:00:1f.6 net00: MDI Write did not complete
[ 4160.359855] e1000e 0000:00:1f.6 net00: MDI Error
Can you please shed a little more light into these errors? Please
find the full log attached.
Kind regards,
Paul
[1]: https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/tree/drivers/net/ethernet/intel/e1000e/phy.c#n206
> -----Original Message-----
> From: Paul Menzel <[email protected]>
> Sent: Tuesday, August 6, 2019 10:36 AM
> To: Jeff Kirsher
> Cc: [email protected]; Linux Kernel Mailing List; Limonciello, Mario
> Subject: MDI errors during resume from ACPI S3 (suspend to ram)
>
> Dear Linux folks,
>
>
> Trying to decrease the resume time of Linux 5.3-rc3 on the Dell OptiPlex
> 5040 with the device below
>
> $ lspci -nn -s 00:1f.6
> 00:1f.6 Ethernet controller [0200]: Intel Corporation Ethernet Connection (2)
> I219-V [8086:15b8] (rev 31)
>
> pm-graph’s script `sleepgraph.py` shows, that the driver *e1000e* takes
> around 400 ms, which is quite a lot. The call graph trace shows that
> `e1000e_read_phy_reg_mdic()` is responsible for a lot of those. From
> `drivers/net/ethernet/intel/e1000e/phy.c` [1]:
>
> for (i = 0; i < (E1000_GEN_POLL_TIMEOUT * 3); i++) {
> udelay(50);
> mdic = er32(MDIC);
> if (mdic & E1000_MDIC_READY)
> break;
> }
> if (!(mdic & E1000_MDIC_READY)) {
> e_dbg("MDI Read did not complete\n");
> return -E1000_ERR_PHY;
> }
> if (mdic & E1000_MDIC_ERROR) {
> e_dbg("MDI Error\n");
> return -E1000_ERR_PHY;
> }
>
> Unfortunately, errors are not logged if dynamic debug is disabled,
> so rebuilding the Linux kernel with `CONFIG_DYNAMIC_DEBUG`, and
>
> echo "file drivers/net/ethernet/* +p" | sudo tee
> /sys/kernel/debug/dynamic_debug/control
>
> I got the messages below.
>
> [ 4159.204192] e1000e 0000:00:1f.6 net00: MDI Error
> [ 4160.267950] e1000e 0000:00:1f.6 net00: MDI Write did not complete
> [ 4160.359855] e1000e 0000:00:1f.6 net00: MDI Error
>
> Can you please shed a little more light into these errors? Please
> find the full log attached.
>
>
> Kind regards,
>
> Paul
>
>
> [1]:
> https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/tree/drivers/n
> et/ethernet/intel/e1000e/phy.c#n206
Strictly as a reference point you may consider trying the out-of-tree driver to see if these
behaviors persist.
https://sourceforge.net/projects/e1000/
On 8/6/2019 18:53, [email protected] wrote:
>> -----Original Message-----
>> From: Paul Menzel <[email protected]>
>> Sent: Tuesday, August 6, 2019 10:36 AM
>> To: Jeff Kirsher
>> Cc: [email protected]; Linux Kernel Mailing List; Limonciello, Mario
>> Subject: MDI errors during resume from ACPI S3 (suspend to ram)
>>
>> Dear Linux folks,
>>
>>
>> Trying to decrease the resume time of Linux 5.3-rc3 on the Dell OptiPlex
>> 5040 with the device below
>>
>> $ lspci -nn -s 00:1f.6
>> 00:1f.6 Ethernet controller [0200]: Intel Corporation Ethernet Connection (2)
>> I219-V [8086:15b8] (rev 31)
>>
>> pm-graph’s script `sleepgraph.py` shows, that the driver *e1000e* takes
>> around 400 ms, which is quite a lot. The call graph trace shows that
>> `e1000e_read_phy_reg_mdic()` is responsible for a lot of those. From
>> `drivers/net/ethernet/intel/e1000e/phy.c` [1]:
>>
>> for (i = 0; i < (E1000_GEN_POLL_TIMEOUT * 3); i++) {
>> udelay(50);
>> mdic = er32(MDIC);
>> if (mdic & E1000_MDIC_READY)
>> break;
>> }
>> if (!(mdic & E1000_MDIC_READY)) {
>> e_dbg("MDI Read did not complete\n");
>> return -E1000_ERR_PHY;
>> }
>> if (mdic & E1000_MDIC_ERROR) {
>> e_dbg("MDI Error\n");
>> return -E1000_ERR_PHY;
>> }
>>
>> Unfortunately, errors are not logged if dynamic debug is disabled,
>> so rebuilding the Linux kernel with `CONFIG_DYNAMIC_DEBUG`, and
>>
>> echo "file drivers/net/ethernet/* +p" | sudo tee
>> /sys/kernel/debug/dynamic_debug/control
>>
>> I got the messages below.
>>
>> [ 4159.204192] e1000e 0000:00:1f.6 net00: MDI Error
>> [ 4160.267950] e1000e 0000:00:1f.6 net00: MDI Write did not complete
>> [ 4160.359855] e1000e 0000:00:1f.6 net00: MDI Error
>>
>> Can you please shed a little more light into these errors? Please
>> find the full log attached.
>>
>>
>> Kind regards,
>>
>> Paul
>>
>>
>> [1]:
>> https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/tree/drivers/n
>> et/ethernet/intel/e1000e/phy.c#n206
>
> Strictly as a reference point you may consider trying the out-of-tree driver to see if these
> behaviors persist.
>
> https://sourceforge.net/projects/e1000/
>
> _______________________________________________
> Intel-wired-lan mailing list
> [email protected]
> https://lists.osuosl.org/mailman/listinfo/intel-wired-lan
>
We are using external PHY. Required ~200 ms to complete MDIC transaction
(depended on the project). You need to take to consider this time before
access to the PHY. I do not recommend decrease timer in a
'e1000e_read_phy_reg_mdic()' method. We could hit on wrong MDI access.
Dear Sasha,
On 07.08.19 09:23, Neftin, Sasha wrote:
> On 8/6/2019 18:53, [email protected] wrote:
>>> -----Original Message-----
>>> From: Paul Menzel <[email protected]>
>>> Sent: Tuesday, August 6, 2019 10:36 AM
>>> To: Jeff Kirsher
>>> Cc: [email protected]; Linux Kernel Mailing List; Limonciello, Mario
>>> Subject: MDI errors during resume from ACPI S3 (suspend to ram)
>>>
>>> Dear Linux folks,
>>>
>>>
>>> Trying to decrease the resume time of Linux 5.3-rc3 on the Dell OptiPlex
>>> 5040 with the device below
>>>
>>> $ lspci -nn -s 00:1f.6
>>> 00:1f.6 Ethernet controller [0200]: Intel Corporation Ethernet Connection (2)
>>> I219-V [8086:15b8] (rev 31)
>>>
>>> pm-graph’s script `sleepgraph.py` shows, that the driver *e1000e* takes
>>> around 400 ms, which is quite a lot. The call graph trace shows that
>>> `e1000e_read_phy_reg_mdic()` is responsible for a lot of those. From
>>> `drivers/net/ethernet/intel/e1000e/phy.c` [1]:
>>>
>>> for (i = 0; i < (E1000_GEN_POLL_TIMEOUT * 3); i++) {
>>> udelay(50);
>>> mdic = er32(MDIC);
>>> if (mdic & E1000_MDIC_READY)
>>> break;
>>> }
>>> if (!(mdic & E1000_MDIC_READY)) {
>>> e_dbg("MDI Read did not complete\n");
>>> return -E1000_ERR_PHY;
>>> }
>>> if (mdic & E1000_MDIC_ERROR) {
>>> e_dbg("MDI Error\n");
>>> return -E1000_ERR_PHY;
>>> }
>>>
>>> Unfortunately, errors are not logged if dynamic debug is disabled,
>>> so rebuilding the Linux kernel with `CONFIG_DYNAMIC_DEBUG`, and
>>>
>>> echo "file drivers/net/ethernet/* +p" | sudo tee
>>> /sys/kernel/debug/dynamic_debug/control
>>>
>>> I got the messages below.
>>>
>>> [ 4159.204192] e1000e 0000:00:1f.6 net00: MDI Error
>>> [ 4160.267950] e1000e 0000:00:1f.6 net00: MDI Write did not complete
>>> [ 4160.359855] e1000e 0000:00:1f.6 net00: MDI Error
>>>
>>> Can you please shed a little more light into these errors? Please
>>> find the full log attached.
>>> [1]: https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/tree/drivers/net/ethernet/intel/e1000e/phy.c#n206
>>
>> Strictly as a reference point you may consider trying the out-of-tree driver to see if these
>> behaviors persist.
>>
>> https://sourceforge.net/projects/e1000/
I can try that in the next days.
> We are using external PHY. Required ~200 ms to complete MDIC
> transaction (depended on the project).
Are you referring to the out-of-tree driver?
> You need to take to consider this time before access to the PHY. I do
> not recommend decrease timer in a 'e1000e_read_phy_reg_mdic()'
> method. We could hit on wrong MDI access.
My point was more, if you know that more time is needed, before the MDI
setting(?) will succeed, why try it anyway and go into the error paths?
Isn’t there some polling possible to find out, when MDI can be set up?
Kind regards,
Paul
On 8/7/2019 17:55, Paul Menzel wrote:
>
> Dear Sasha,
>
>
> On 07.08.19 09:23, Neftin, Sasha wrote:
>> On 8/6/2019 18:53, [email protected] wrote:
>>>> -----Original Message-----
>>>> From: Paul Menzel <[email protected]>
>>>> Sent: Tuesday, August 6, 2019 10:36 AM
>>>> To: Jeff Kirsher
>>>> Cc: [email protected]; Linux Kernel Mailing List; Limonciello, Mario
>>>> Subject: MDI errors during resume from ACPI S3 (suspend to ram)
>>>>
>>>> Dear Linux folks,
>>>>
>>>>
>>>> Trying to decrease the resume time of Linux 5.3-rc3 on the Dell OptiPlex
>>>> 5040 with the device below
>>>>
>>>> $ lspci -nn -s 00:1f.6
>>>> 00:1f.6 Ethernet controller [0200]: Intel Corporation Ethernet Connection (2)
>>>> I219-V [8086:15b8] (rev 31)
>>>>
>>>> pm-graph’s script `sleepgraph.py` shows, that the driver *e1000e* takes
>>>> around 400 ms, which is quite a lot. The call graph trace shows that
>>>> `e1000e_read_phy_reg_mdic()` is responsible for a lot of those. From
>>>> `drivers/net/ethernet/intel/e1000e/phy.c` [1]:
>>>>
>>>> for (i = 0; i < (E1000_GEN_POLL_TIMEOUT * 3); i++) {
>>>> udelay(50);
>>>> mdic = er32(MDIC);
>>>> if (mdic & E1000_MDIC_READY)
>>>> break;
>>>> }
>>>> if (!(mdic & E1000_MDIC_READY)) {
>>>> e_dbg("MDI Read did not complete\n");
>>>> return -E1000_ERR_PHY;
>>>> }
>>>> if (mdic & E1000_MDIC_ERROR) {
>>>> e_dbg("MDI Error\n");
>>>> return -E1000_ERR_PHY;
>>>> }
>>>>
>>>> Unfortunately, errors are not logged if dynamic debug is disabled,
>>>> so rebuilding the Linux kernel with `CONFIG_DYNAMIC_DEBUG`, and
>>>>
>>>> echo "file drivers/net/ethernet/* +p" | sudo tee
>>>> /sys/kernel/debug/dynamic_debug/control
>>>>
>>>> I got the messages below.
>>>>
>>>> [ 4159.204192] e1000e 0000:00:1f.6 net00: MDI Error
>>>> [ 4160.267950] e1000e 0000:00:1f.6 net00: MDI Write did not complete
>>>> [ 4160.359855] e1000e 0000:00:1f.6 net00: MDI Error
>>>>
>>>> Can you please shed a little more light into these errors? Please
>>>> find the full log attached.
>
>>>> [1]: https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/tree/drivers/net/ethernet/intel/e1000e/phy.c#n206
>>>
>>> Strictly as a reference point you may consider trying the out-of-tree driver to see if these
>>> behaviors persist.
>>>
>>> https://sourceforge.net/projects/e1000/
>
> I can try that in the next days.
>
>> We are using external PHY. Required ~200 ms to complete MDIC
>> transaction (depended on the project).
>
> Are you referring to the out-of-tree driver?
>
I believe the out of tree driver have a same approach to MDIC access.
>> You need to take to consider this time before access to the PHY. I do
>> not recommend decrease timer in a 'e1000e_read_phy_reg_mdic()'
>> method. We could hit on wrong MDI access.
> My point was more, if you know that more time is needed, before the MDI
> setting(?) will succeed, why try it anyway and go into the error paths?
> Isn’t there some polling possible to find out, when MDI can be set up?
>
e1000e is very old driver and serve pretty lot of 1G clients. Each 1Gbe
MAC/PHY controller have a different configuration depend platform.
>
> Kind regards,
>
> Paul
>
Hello Paul,
Let me back later with more information specific your device. I will try
find out more details with design team.