From: Michael Williamson <michael.williamson@criticallink•com>
To: cyril@ti•com
Cc: Kevin Hilman <khilman@deeprootsystems•com>,
"tony@atomide•com" <tony@atomide•com>,
"netdev@vger•kernel.org" <netdev@vger•kernel.org>,
"davinci-linux-open-source@linux•davincidsp.com"
<davinci-linux-open-source@linux•davincidsp.com>,
"linux-omap@vger•kernel.org" <linux-omap@vger•kernel.org>,
"davem@davemloft•net" <davem@davemloft•net>
Subject: Re: [PATCH v3 00/10] split out emac cpdma and mdio for reuse
Date: Sat, 11 Sep 2010 09:14:14 -0400 [thread overview]
Message-ID: <4C8B80A6.5060609@criticallink.com> (raw)
In-Reply-To: <4C8AB85A.8060909@ti.com>
Hi Cyril,
On 09/10/2010 06:59 PM, Cyril Chemparathy wrote:
> Hi Mike,
>
> I have merged your latest two emails and responded to both here.
>
[...]
> One of the patches posted on my repo [1] replaces the dumb mdelay() with
> a bit more logic that calculates the worst-case access time. This
> mechanism may work a lot better for you. Would you mind trying it out?
>
You patch from [1] is working much more reliably now (well, 6 for 6 boot cycles as well as
several ifup/ifdown cycles). I do get the "resetting idled controlled" console message
every cycle, it seems that will be expected now.
[...]
>>
>> Also, your while(1) loops with the continue conditions on the second wait_for_user_access()
>> in the read and writes might need some consideration, i.e.:
>>
>> while (1) {
>> ret = wait_for_user_access(data);
>> if (ret == -EAGAIN)
>> continue;
>> if (ret < 0)
>> break;
>>
>> __raw_writel(reg, &data->regs->user[0].access);
>>
>> ret = wait_for_user_access(data);
>> if (ret == -EAGAIN)
>> continue;
>> ^^^^^^^^^ <--- this will re-issue the request.... what you want?
>
> Yes, the wait_for_user_access() would have reset the controller,
> throwing the current transaction out. The intent here is to restart the
> current transaction with a newly initialized controller.
>
OK. Makes sense. Looking at it felt like there was a chance for end endless spin, but that
seems unlikely given how that condition might fire.
[...]
>> Also, on the shutdown, I get a major kernel trace. Here is the dump, as much
>> as I could catch of it.... (I need a better terminal program)
>
> [...]
>> WARNING: at drivers/net/davinci_emac.c:1025 __cpdma_chan_free+0xac/0xb0()
>
> The current code spits out a huge volume of stuff as a result of a
> WARN_ON in the rx handler. The gitweb on [1] has a patch that fixes this.
>
Yes, these messages are no longer issued with the patches from [1]. Thanks.
> [...]
>>>> The MDIO module upgrade (rev 1.4 -> 1.5) could have something to do with
>>>> this behavior. Even so, I can't explain why this issue wasn't seen on
>>>> da8xx prior to this series. The original code should (at least in
>>>> theory) have sporadically locked up on emac open.
>>>>
>> I think, if I understand it correctly, that in the previous version of
>> this code, the emac was reset *prior* to enabling, scanning, and assigning
>> the associated phy on the MDIO bus. The new implementation sets up and scans
>> the MDIO bus first, then comes back around to the EMAC second... hits a reset,
>> and doesn't re-ENABLE the MDIO.
>
> AFAICS, that isn't entirely accurate. In the previous version, the mdio
> bus was being brought up at probe time in davinci_emac_probe(). The
> soft-reset was happening later on when the device is opened, in
> emac_hw_enable().
>
> The difference, however, is that the original code forced an
> emac_mii_reset() immediately after the emac_hw_enable(). This is not
> being done with the separated mdio, and that is the problem. In terms
> of behavior, with the current work around, the new and old versions
> should be close to identical. More below...
>
>> Also, maybe hitting the EMAC reset while the MDIO state machine is up is *bad*, I
>> seem to recall some text in the user's guide about waiting for the state
>> machine to stop before disabling it. I wonder if that also applies to reset?
>
> You are correct. EMAC soft-reset stops the MDIO mid-transaction, quite
> unlike disabling the module via the control register. Therefore, there
> is a risk that a badly designed phy could be left hanging in an
> arbitrary state. However, all earlier versions of the emac code have
> been exposed to this very same vulnerability (i.e. arbitrary emac
> soft-reset regardless of mdio state) all along.
>
Thanks for straightening me out on this, Cryil. Your patch series in [1] seems to
have resolved the issues I've been able to see on the da850 based board I'm using
here. I appreciate your patience and quick response. I may try to beat on it a
bit more with some network performance tests (even though it's not at all related
to the immediate problems you've fixed) -- we've had that on our list of todos
anyway for this module.
I think it would be good to float your patches over to davinci-next, if possible, before the
37 merge window opens... Speaking, of course, from a very partial perspective.
> [1]
> http://arago-project.org/git/people/?p=cyril/linux-tnetv107x.git;a=shortlog;h=refs/heads/emac-cpdma-mdio-fixes
-Mike
prev parent reply other threads:[~2010-09-11 13:14 UTC|newest]
Thread overview: 29+ messages / expand[flat|nested] mbox.gz Atom feed top
2010-09-07 20:25 [PATCH v3 00/10] split out emac cpdma and mdio for reuse Cyril Chemparathy
2010-09-07 20:25 ` [PATCH v3 01/10] net: davinci_emac: separate out davinci mdio Cyril Chemparathy
2010-09-07 20:25 ` [PATCH v3 02/10] davinci: add mdio platform devices Cyril Chemparathy
2010-09-07 20:25 ` [PATCH v3 03/10] omap: " Cyril Chemparathy
2010-09-08 1:00 ` Tony Lindgren
2010-09-07 20:25 ` [PATCH v3 04/10] net: davinci_emac: switch to new mdio Cyril Chemparathy
2010-09-07 20:25 ` [PATCH v3 05/10] davinci: cleanup unused davinci mdio arch code Cyril Chemparathy
2010-09-07 20:25 ` [PATCH v3 06/10] omap: " Cyril Chemparathy
2010-09-08 1:00 ` Tony Lindgren
2010-09-07 20:25 ` [PATCH v3 07/10] net: davinci_emac: cleanup unused mdio emac code Cyril Chemparathy
2010-09-07 20:25 ` [PATCH v3 08/10] net: davinci_emac: separate out cpdma code Cyril Chemparathy
2010-09-07 20:25 ` [PATCH v3 09/10] net: davinci_emac: switch to new cpdma layer Cyril Chemparathy
2010-09-07 20:25 ` [PATCH v3 10/10] net: davinci_emac: cleanup unused cpdma code Cyril Chemparathy
2010-09-08 1:18 ` [PATCH v3 00/10] split out emac cpdma and mdio for reuse Kevin Hilman
2010-09-08 2:22 ` Michael Williamson
2010-09-08 21:59 ` Cyril Chemparathy
2010-09-09 0:47 ` Michael Williamson
2010-09-09 18:43 ` Michael Williamson
2010-09-09 19:51 ` Cyril Chemparathy
2010-09-09 21:24 ` Cyril Chemparathy
2010-09-09 21:45 ` Michael Williamson
2010-09-09 21:25 ` Michael Williamson
2010-09-10 15:23 ` Caglar Akyuz
2010-09-11 8:54 ` Caglar Akyuz
2010-09-13 14:09 ` Cyril Chemparathy
2010-09-13 15:46 ` Cyril Chemparathy
2010-09-13 17:51 ` Caglar Akyuz
2010-09-10 22:59 ` Cyril Chemparathy
2010-09-11 13:14 ` Michael Williamson [this message]
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=4C8B80A6.5060609@criticallink.com \
--to=michael.williamson@criticallink$(echo .)com \
--cc=cyril@ti$(echo .)com \
--cc=davem@davemloft$(echo .)net \
--cc=davinci-linux-open-source@linux$(echo .)davincidsp.com \
--cc=khilman@deeprootsystems$(echo .)com \
--cc=linux-omap@vger$(echo .)kernel.org \
--cc=netdev@vger$(echo .)kernel.org \
--cc=tony@atomide$(echo .)com \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox