[U-Boot] [PATCH v2] net/eth.c: fix eth_write_hwaddr() to use dev->enetaddr as fall back
From: Eric Miao eric.miao@linaro.org
Ignore the return value of eth_getenv_enetaddr_by_index(), and if it fails, fall back to use dev->enetaddr, which could be filled up by the ethernet device driver:
With the current code, introduced with below commit, eth_write_hwaddr() will fail immediately if there is no eth<n>addr in the environment variables.
However, e.g. for an overo based product that uses the SMSC911x ethernet chip (with the MAC address set via EEPROM connected to the SMSC911x chip), the MAC address is still OK.
On mx28 boards that are depending on the OCOTP bits to set the MAC address (like the Denx m28 board), the OCOTP bits should be used instead of failing on the environment variables.
Actually, this was the original behavior, and was later changed by commit 7616e7850804c7c69e0a22c179dfcba9e8f3f587.
Signed-off-by: Eric Miao eric.miao@linaro.org Acked-by: Simon Glass sjg@chromium.org Acked-by: Dirk Behme dirk.behme@de.bosch.com CC: Stefan Roese sr@denx.de CC: Eric Miao eric.miao@linaro.org CC: Wolfgang Denk wd@denx.de CC: Philip Balister philip@balister.org CC: Zach Sadecki zach@itwatchdogs.com --- v2: Correct the referenced commit ID and update the commit message. No functional change at the code itself.
Note: This resend is based on my understanding from
http://lists.denx.de/pipermail/u-boot/2012-January/116118.html
Please let Eric and me know if I missed anything there.
net/eth.c | 3 +-- 1 files changed, 1 insertions(+), 2 deletions(-)
diff --git a/net/eth.c b/net/eth.c index b4b9b43..451568f 100644 --- a/net/eth.c +++ b/net/eth.c @@ -175,8 +175,7 @@ int eth_write_hwaddr(struct eth_device *dev, const char *base_name, unsigned char env_enetaddr[6]; int ret = 0;
- if (!eth_getenv_enetaddr_by_index(base_name, eth_number, env_enetaddr)) - return -1; + eth_getenv_enetaddr_by_index(base_name, eth_number, env_enetaddr);
if (memcmp(env_enetaddr, "\0\0\0\0\0\0", 6)) { if (memcmp(dev->enetaddr, "\0\0\0\0\0\0", 6) &&
Hi,
On Thu, Jan 19, 2012 at 12:56 AM, Dirk Behme dirk.behme@de.bosch.com wrote:
From: Eric Miao eric.miao@linaro.org
Ignore the return value of eth_getenv_enetaddr_by_index(), and if it fails, fall back to use dev->enetaddr, which could be filled up by the ethernet device driver:
With the current code, introduced with below commit, eth_write_hwaddr() will fail immediately if there is no eth<n>addr in the environment variables.
However, e.g. for an overo based product that uses the SMSC911x ethernet chip (with the MAC address set via EEPROM connected to the SMSC911x chip), the MAC address is still OK.
On mx28 boards that are depending on the OCOTP bits to set the MAC address (like the Denx m28 board), the OCOTP bits should be used instead of failing on the environment variables.
Actually, this was the original behavior, and was later changed by commit 7616e7850804c7c69e0a22c179dfcba9e8f3f587.
Signed-off-by: Eric Miao eric.miao@linaro.org Acked-by: Simon Glass sjg@chromium.org Acked-by: Dirk Behme dirk.behme@de.bosch.com CC: Stefan Roese sr@denx.de CC: Eric Miao eric.miao@linaro.org CC: Wolfgang Denk wd@denx.de CC: Philip Balister philip@balister.org CC: Zach Sadecki zach@itwatchdogs.com
v2: Correct the referenced commit ID and update the commit message. No functional change at the code itself.
Note: This resend is based on my understanding from
http://lists.denx.de/pipermail/u-boot/2012-January/116118.html
Please let Eric and me know if I missed anything there.
I don't think you have missed anything and I have already acked this. But I want to start a related discussion.
The code structure does bug me a bit - I think it is too confusing. eth_getenv_enetaddr() returns an error if there is no environment variable set or if the address it gets from the environment variable is invalid. We should probably not conflate those two. The first is ok here, but the second isn't, I think.
What if the driver has no write_hwaddr method? Do we silently ignore the environment variable value?
Why use memcmp() against env_enetaddr when the function we just called returns an error that tells us whether it is supposed to be valid (the error return your patch squashes)?
We set the hwaddr by writing directly into the dev->enet_addr field and then calling write_hwaddr() if it exists. Maybe that is ok - is the lack of write_hwaddr() an indication that the driver does MAC address handling on the fly, or just that it can't set the MAC address at all?
Overall I feel that eth_write_hwaddr() should return success or failure, confident in its determination that there is either a valid MAC address or there is not. The message you are seeing is I suppose an indication that it thinks there is a problem, when in fact none exists in this case. At the moment it feels fragile.
I wonder whether a little refactor here would be best?
That said, your patch restores the original behaviour, hiding the problem which isn't actually a problem in this case, and which we don't want to report. So it is better than the status quo. Perhaps a little comment as to why the return value doesn't matter?
Regards, Simon
net/eth.c | 3 +-- 1 files changed, 1 insertions(+), 2 deletions(-)
diff --git a/net/eth.c b/net/eth.c index b4b9b43..451568f 100644 --- a/net/eth.c +++ b/net/eth.c @@ -175,8 +175,7 @@ int eth_write_hwaddr(struct eth_device *dev, const char *base_name, unsigned char env_enetaddr[6]; int ret = 0;
- if (!eth_getenv_enetaddr_by_index(base_name, eth_number, env_enetaddr))
- return -1;
- eth_getenv_enetaddr_by_index(base_name, eth_number, env_enetaddr);
if (memcmp(env_enetaddr, "\0\0\0\0\0\0", 6)) { if (memcmp(dev->enetaddr, "\0\0\0\0\0\0", 6) && -- 1.7.0.4
U-Boot mailing list U-Boot@lists.denx.de http://lists.denx.de/mailman/listinfo/u-boot
On 23.01.2012 08:31, Simon Glass wrote:
Hi,
On Thu, Jan 19, 2012 at 12:56 AM, Dirk Behme dirk.behme@de.bosch.com wrote:
From: Eric Miao eric.miao@linaro.org
Ignore the return value of eth_getenv_enetaddr_by_index(), and if it fails, fall back to use dev->enetaddr, which could be filled up by the ethernet device driver:
With the current code, introduced with below commit, eth_write_hwaddr() will fail immediately if there is no eth<n>addr in the environment variables.
However, e.g. for an overo based product that uses the SMSC911x ethernet chip (with the MAC address set via EEPROM connected to the SMSC911x chip), the MAC address is still OK.
On mx28 boards that are depending on the OCOTP bits to set the MAC address (like the Denx m28 board), the OCOTP bits should be used instead of failing on the environment variables.
Actually, this was the original behavior, and was later changed by commit 7616e7850804c7c69e0a22c179dfcba9e8f3f587.
Signed-off-by: Eric Miao eric.miao@linaro.org Acked-by: Simon Glass sjg@chromium.org Acked-by: Dirk Behme dirk.behme@de.bosch.com CC: Stefan Roese sr@denx.de CC: Eric Miao eric.miao@linaro.org CC: Wolfgang Denk wd@denx.de CC: Philip Balister philip@balister.org CC: Zach Sadecki zach@itwatchdogs.com
v2: Correct the referenced commit ID and update the commit message. No functional change at the code itself.
Note: This resend is based on my understanding from
http://lists.denx.de/pipermail/u-boot/2012-January/116118.html Please let Eric and me know if I missed anything there.I don't think you have missed anything and I have already acked this. But I want to start a related discussion.
The code structure does bug me a bit - I think it is too confusing. eth_getenv_enetaddr() returns an error if there is no environment variable set or if the address it gets from the environment variable is invalid. We should probably not conflate those two. The first is ok here, but the second isn't, I think.
What if the driver has no write_hwaddr method? Do we silently ignore the environment variable value?
Why use memcmp() against env_enetaddr when the function we just called returns an error that tells us whether it is supposed to be valid (the error return your patch squashes)?
We set the hwaddr by writing directly into the dev->enet_addr field and then calling write_hwaddr() if it exists. Maybe that is ok - is the lack of write_hwaddr() an indication that the driver does MAC address handling on the fly, or just that it can't set the MAC address at all?
Overall I feel that eth_write_hwaddr() should return success or failure, confident in its determination that there is either a valid MAC address or there is not. The message you are seeing is I suppose an indication that it thinks there is a problem, when in fact none exists in this case. At the moment it feels fragile.
I wonder whether a little refactor here would be best?
That said, your patch restores the original behaviour, hiding the problem which isn't actually a problem in this case, and which we don't want to report. So it is better than the status quo.
Ok, thanks.
I'm not an expert for this code, nor is the patch from me. It's from Eric ;) I just try to help to mainline all the stuff we have collected for i.MX6.
Therefore I wonder if it would be possible to split this into two steps:
a) Improve the status quo by applying this patch b) In parallel discuss how to refactor and improve this code as you describe above
It's my feeling that with (a) we still have a chance to improve v2012.03. But I doubt that (b) would make it into v2012.03.
Best regards
Dirk
Hi Dirk,
On Jan 23, 2012 12:30 AM, "Dirk Behme" dirk.behme@de.bosch.com wrote:
On 23.01.2012 08:31, Simon Glass wrote:
Hi,
On Thu, Jan 19, 2012 at 12:56 AM, Dirk Behme dirk.behme@de.bosch.com
wrote:
From: Eric Miao eric.miao@linaro.org
Ignore the return value of eth_getenv_enetaddr_by_index(), and if it fails, fall back to use dev->enetaddr, which could be filled up by the ethernet device driver:
With the current code, introduced with below commit, eth_write_hwaddr() will fail immediately if there is no eth<n>addr in the environment
variables.
However, e.g. for an overo based product that uses the SMSC911x ethernet chip (with the MAC address set via EEPROM connected to the SMSC911x
chip),
the MAC address is still OK.
On mx28 boards that are depending on the OCOTP bits to set the MAC
address
(like the Denx m28 board), the OCOTP bits should be used instead of failing on the environment variables.
Actually, this was the original behavior, and was later changed by commit 7616e7850804c7c69e0a22c179dfcba9e8f3f587.
Signed-off-by: Eric Miao eric.miao@linaro.org Acked-by: Simon Glass sjg@chromium.org Acked-by: Dirk Behme dirk.behme@de.bosch.com CC: Stefan Roese sr@denx.de CC: Eric Miao eric.miao@linaro.org CC: Wolfgang Denk wd@denx.de CC: Philip Balister philip@balister.org CC: Zach Sadecki zach@itwatchdogs.com
v2: Correct the referenced commit ID and update the commit message. No functional change at the code itself.
Note: This resend is based on my understanding from
http://lists.denx.de/pipermail/u-boot/2012-January/116118.html Please let Eric and me know if I missed anything there.I don't think you have missed anything and I have already acked this. But I want to start a related discussion.
The code structure does bug me a bit - I think it is too confusing. eth_getenv_enetaddr() returns an error if there is no environment variable set or if the address it gets from the environment variable is invalid. We should probably not conflate those two. The first is ok here, but the second isn't, I think.
What if the driver has no write_hwaddr method? Do we silently ignore the environment variable value?
Why use memcmp() against env_enetaddr when the function we just called returns an error that tells us whether it is supposed to be valid (the error return your patch squashes)?
We set the hwaddr by writing directly into the dev->enet_addr field and then calling write_hwaddr() if it exists. Maybe that is ok - is the lack of write_hwaddr() an indication that the driver does MAC address handling on the fly, or just that it can't set the MAC address at all?
Overall I feel that eth_write_hwaddr() should return success or failure, confident in its determination that there is either a valid MAC address or there is not. The message you are seeing is I suppose an indication that it thinks there is a problem, when in fact none exists in this case. At the moment it feels fragile.
I wonder whether a little refactor here would be best?
That said, your patch restores the original behaviour, hiding the problem which isn't actually a problem in this case, and which we don't want to report. So it is better than the status quo.
Ok, thanks.
I'm not an expert for this code, nor is the patch from me. It's from Eric
;) I just try to help to mainline all the stuff we have collected for i.MX6.
Therefore I wonder if it would be possible to split this into two steps:
a) Improve the status quo by applying this patch b) In parallel discuss how to refactor and improve this code as you
describe above
It's my feeling that with (a) we still have a chance to improve v2012.03.
But I doubt that (b) would make it into v2012.03.
Yes agreed, it is a separate discussion. I added Wolfgang on cc to see what he thinks.
Regards, Simon
Best regards
Dirk
On 23.01.2012 17:17, Simon Glass wrote:
Hi Dirk,
On Jan 23, 2012 12:30 AM, "Dirk Behme" <dirk.behme@de.bosch.com mailto:dirk.behme@de.bosch.com> wrote:
On 23.01.2012 08:31, Simon Glass wrote:
Hi,
On Thu, Jan 19, 2012 at 12:56 AM, Dirk Behme
<dirk.behme@de.bosch.com mailto:dirk.behme@de.bosch.com> wrote:
From: Eric Miao <eric.miao@linaro.org mailto:eric.miao@linaro.org>
Ignore the return value of eth_getenv_enetaddr_by_index(), and if it fails, fall back to use dev->enetaddr, which could be filled up by the ethernet device driver:
With the current code, introduced with below commit, eth_write_hwaddr() will fail immediately if there is no eth<n>addr in the environment
variables.
However, e.g. for an overo based product that uses the SMSC911x
ethernet
chip (with the MAC address set via EEPROM connected to the SMSC911x
chip),
the MAC address is still OK.
On mx28 boards that are depending on the OCOTP bits to set the MAC
address
(like the Denx m28 board), the OCOTP bits should be used instead of failing on the environment variables.
Actually, this was the original behavior, and was later changed by commit 7616e7850804c7c69e0a22c179dfcba9e8f3f587.
Signed-off-by: Eric Miao <eric.miao@linaro.org
Acked-by: Simon Glass <sjg@chromium.org mailto:sjg@chromium.org> Acked-by: Dirk Behme <dirk.behme@de.bosch.com
mailto:dirk.behme@de.bosch.com>
CC: Stefan Roese <sr@denx.de mailto:sr@denx.de> CC: Eric Miao <eric.miao@linaro.org mailto:eric.miao@linaro.org> CC: Wolfgang Denk <wd@denx.de mailto:wd@denx.de> CC: Philip Balister <philip@balister.org mailto:philip@balister.org> CC: Zach Sadecki <zach@itwatchdogs.com mailto:zach@itwatchdogs.com>
v2: Correct the referenced commit ID and update the commit message. No functional change at the code itself.
Note: This resend is based on my understanding from
http://lists.denx.de/pipermail/u-boot/2012-January/116118.html Please let Eric and me know if I missed anything there.I don't think you have missed anything and I have already acked this. But I want to start a related discussion.
The code structure does bug me a bit - I think it is too confusing. eth_getenv_enetaddr() returns an error if there is no environment variable set or if the address it gets from the environment variable is invalid. We should probably not conflate those two. The first is ok here, but the second isn't, I think.
What if the driver has no write_hwaddr method? Do we silently ignore the environment variable value?
Why use memcmp() against env_enetaddr when the function we just called returns an error that tells us whether it is supposed to be valid (the error return your patch squashes)?
We set the hwaddr by writing directly into the dev->enet_addr field and then calling write_hwaddr() if it exists. Maybe that is ok - is the lack of write_hwaddr() an indication that the driver does MAC address handling on the fly, or just that it can't set the MAC address at all?
Overall I feel that eth_write_hwaddr() should return success or failure, confident in its determination that there is either a valid MAC address or there is not. The message you are seeing is I suppose an indication that it thinks there is a problem, when in fact none exists in this case. At the moment it feels fragile.
I wonder whether a little refactor here would be best?
That said, your patch restores the original behaviour, hiding the problem which isn't actually a problem in this case, and which we don't want to report. So it is better than the status quo.
Ok, thanks.
I'm not an expert for this code, nor is the patch from me. It's from
Eric ;) I just try to help to mainline all the stuff we have collected for i.MX6.
Therefore I wonder if it would be possible to split this into two steps:
a) Improve the status quo by applying this patch b) In parallel discuss how to refactor and improve this code as you
describe above
It's my feeling that with (a) we still have a chance to improve
v2012.03. But I doubt that (b) would make it into v2012.03.
Yes agreed, it is a separate discussion. I added Wolfgang on cc to see what he thinks.
Any news to this?
Many thanks
Dirk
Hi Dirk,
On Tue, Feb 7, 2012 at 11:13 PM, Dirk Behme dirk.behme@de.bosch.com wrote:
On 23.01.2012 17:17, Simon Glass wrote:
Hi Dirk,
On Jan 23, 2012 12:30 AM, "Dirk Behme" <dirk.behme@de.bosch.com mailto:dirk.behme@de.bosch.com> wrote: > > On 23.01.2012 08:31, Simon Glass wrote: >> >> Hi, >> >> On Thu, Jan 19, 2012 at 12:56 AM, Dirk Behme <dirk.behme@de.bosch.com mailto:dirk.behme@de.bosch.com> wrote: >>> >>> From: Eric Miao <eric.miao@linaro.org mailto:eric.miao@linaro.org>
>>> >>> Ignore the return value of eth_getenv_enetaddr_by_index(), and if it >>> fails, fall back to use dev->enetaddr, which could be filled up by >>> the ethernet device driver: >>> >>> With the current code, introduced with below commit, eth_write_hwaddr() >>> will fail immediately if there is no eth<n>addr in the environment variables. >>> >>> However, e.g. for an overo based product that uses the SMSC911x ethernet >>> chip (with the MAC address set via EEPROM connected to the SMSC911x chip), >>> the MAC address is still OK. >>> >>> On mx28 boards that are depending on the OCOTP bits to set the MAC address >>> (like the Denx m28 board), the OCOTP bits should be used instead of >>> failing on the environment variables. >>> >>> Actually, this was the original behavior, and was later changed by >>> commit 7616e7850804c7c69e0a22c179dfcba9e8f3f587. >>> >>> Signed-off-by: Eric Miao <eric.miao@linaro.org mailto:eric.miao@linaro.org> >>> Acked-by: Simon Glass <sjg@chromium.org mailto:sjg@chromium.org> >>> Acked-by: Dirk Behme <dirk.behme@de.bosch.com mailto:dirk.behme@de.bosch.com> >>> CC: Stefan Roese <sr@denx.de mailto:sr@denx.de> >>> CC: Eric Miao <eric.miao@linaro.org mailto:eric.miao@linaro.org> >>> CC: Wolfgang Denk <wd@denx.de mailto:wd@denx.de> >>> CC: Philip Balister <philip@balister.org mailto:philip@balister.org> >>> CC: Zach Sadecki <zach@itwatchdogs.com mailto:zach@itwatchdogs.com>
>>> --- >>> v2: Correct the referenced commit ID and update the commit message. >>> No functional change at the code itself. >>> >>> Note: This resend is based on my understanding from >>> >>> http://lists.denx.de/pipermail/u-boot/2012-January/116118.html >>> >>> Please let Eric and me know if I missed anything there. >> >> >> I don't think you have missed anything and I have already acked this. >> But I want to start a related discussion. >> >> The code structure does bug me a bit - I think it is too confusing. >> eth_getenv_enetaddr() returns an error if there is no environment >> variable set or if the address it gets from the environment variable >> is invalid. We should probably not conflate those two. The first is ok >> here, but the second isn't, I think. >> >> What if the driver has no write_hwaddr method? Do we silently ignore >> the environment variable value? >> >> Why use memcmp() against env_enetaddr when the function we just called >> returns an error that tells us whether it is supposed to be valid (the >> error return your patch squashes)? >> >> We set the hwaddr by writing directly into the dev->enet_addr field >> and then calling write_hwaddr() if it exists. Maybe that is ok - is >> the lack of write_hwaddr() an indication that the driver does MAC >> address handling on the fly, or just that it can't set the MAC address >> at all? >> >> Overall I feel that eth_write_hwaddr() should return success or >> failure, confident in its determination that there is either a valid >> MAC address or there is not. The message you are seeing is I suppose >> an indication that it thinks there is a problem, when in fact none >> exists in this case. At the moment it feels fragile. >> >> I wonder whether a little refactor here would be best? >> >> That said, your patch restores the original behaviour, hiding the >> problem which isn't actually a problem in this case, and which we >> don't want to report. So it is better than the status quo. > > > Ok, thanks. > > I'm not an expert for this code, nor is the patch from me. It's from Eric ;) I just try to help to mainline all the stuff we have collected for i.MX6. > > Therefore I wonder if it would be possible to split this into two steps: > > a) Improve the status quo by applying this patch > b) In parallel discuss how to refactor and improve this code as you describe above > > It's my feeling that with (a) we still have a chance to improve v2012.03. But I doubt that (b) would make it into v2012.03.
Yes agreed, it is a separate discussion. I added Wolfgang on cc to see what he thinks.
Any news to this?
Already acked by me. Are you going to start a separate discussion on how to clean up this code? If so please cc me.
Regards, Simon
Many thanks
Dirk
On 23.01.2012 08:31, Simon Glass wrote: ...
Note: This resend is based on my understanding from
http://lists.denx.de/pipermail/u-boot/2012-January/116118.html Please let Eric and me know if I missed anything there.I don't think you have missed anything and I have already acked this. But I want to start a related discussion.
The code structure does bug me a bit - I think it is too confusing. eth_getenv_enetaddr() returns an error if there is no environment variable set or if the address it gets from the environment variable is invalid. We should probably not conflate those two. The first is ok here, but the second isn't, I think.
What if the driver has no write_hwaddr method? Do we silently ignore the environment variable value?
Why use memcmp() against env_enetaddr when the function we just called returns an error that tells us whether it is supposed to be valid (the error return your patch squashes)?
We set the hwaddr by writing directly into the dev->enet_addr field and then calling write_hwaddr() if it exists. Maybe that is ok - is the lack of write_hwaddr() an indication that the driver does MAC address handling on the fly, or just that it can't set the MAC address at all?
Overall I feel that eth_write_hwaddr() should return success or failure, confident in its determination that there is either a valid MAC address or there is not. The message you are seeing is I suppose an indication that it thinks there is a problem, when in fact none exists in this case. At the moment it feels fragile.
I wonder whether a little refactor here would be best?
While discussing about [1] we found that this is only a short term fix (which should go into 2012.03 [2]) and that we should discuss about a more general clean up of eth_write_hwaddr() and friends:
http://git.denx.de/cgi-bin/gitweb.cgi?p=u-boot.git;a=blob;f=net/eth.c;h=b4b9...
See Simon's ideas above.
Comments? Opinions?
Best regards
Dirk
[1] http://lists.denx.de/pipermail/u-boot/2012-January/116224.html
[2] http://lists.denx.de/pipermail/u-boot/2012-January/116436.html
participants (2)
-
Dirk Behme -
Simon Glass