[PATCH] Revert "OMAP/serial: Fix incorrect Rx FIFO threshold setting, LSR validation on Tx, and Tx FIFO IRQ generation"

[Paul Walmsley]

cc Kevin, Felipe

Hi

On Mon, 1 Apr 2013, Alexey Pelykh wrote:

> On Mon, Apr 1, 2013 at 12:13 PM, Paul Walmsley <paul at pwsan.com> wrote:
> > On Mon, 1 Apr 2013, Alexey Pelykh wrote:
> >
> >> Actually, I've tested my patch on DM3730,
> >
> > What board, bootloader, and test steps did you use?  Can you post a dmesg?
> 
> I've used LogicPD Torpedo Wireless SoM + DevKit board. Bootloader is
> stock from their BSP, u-boot 2011.06 + patches from LogicPD.
> I can provide dmesg of 3.8(inital kernel this patch was intented for)
> a bit later during the day, and a bit more later same patch rebased
> onto latest sources.
> Also, I should mention that LogicPD does not support 3.8, so most
> LogicPD-specific devices are not working.

A dmesg from v3.9-rc5 would be ideal, but sounds like it's a non-mainline 
kernel?

Are you using a serial console on that board, and if so, what UART is it 
on?

> And I'd like to point out that after patch gaps I've seen are not gone
> completely, ~1 packet out of 200 at 1mbaud is still corrupted, but
> only when PM is enabled in kernel,

Are you seeing corruption caused by the OMAP UART's receive path, or by 
the transmit path?

When you say "only when PM is enabled" do you mean by enabling CONFIG_PM, 
or do you mean after setting an autosuspend timeout on the serial drivers, 
or something else?

> what gives me a clue that it must be something in PM also. Maybe my
> patch just reveals PM-specific issue. Have you tried without PM to
> reproduce issues you're experiencing?

So I'm using omap2plus_defconfig, which has CONFIG_PM=y, but the issues 
appear before serial autosuspend is enabled.

> Indeed, TRMs appear to contain wrong information, but I believe that
> in this particular case, bug is somewhere around, but not in these
> lines.
> I believe if you take a look on related TRM pages, it seems quite
> logical that configuration of SCR (before my patch) is not quite
> proper:
>  - trigger interrupt only when it's entirely empty, not when threshold
> level is reached, what is incorrect, in my opinion.
>  - (less influencing) Granularity of counter indicates opposite to
> what is stated in comments
> 
> These issues may have never been experienced by others, since
> originally I've hit them on 1mbaud speed, what is not supported by
> original driver at all.

It wouldn't surprise me at all if something is wrong with the 
original driver's FIFO setup.  The challenge at this point is to keep 
the driver working while trying to fix what's wrong with it.


- Paul