[0/2] serial: qcom_geni_serial: Use the FIFOs properly for console

Message

Doug Anderson June 26, 2020, 8 p.m. UTC

This series of two patches gets rid of some ugly hacks that were in
the qcom_geni_serial driver around dealing with a port that was used
for console output and dealing with a port that was being used for
kgdb.

While the character reading/writing code is now slightly more complex,
it's better to be consistently configuring the serial port the same
way and doing so avoids some corner cases where the old hacks weren't
always catching properly.

This change is slightly larger than it needs to be because I was
trying not to use global variables in the read/write functions.
Unfortunately the functions were sometimes called earlycon which
didn't have any "private_data" pointer set.  I've tried to do the
minimal change here to have some shared "private_data" that's always
present, but longer term it wouldn't hurt to see if we could unify
more.

Greg / Andy / Bjorn:

This series of patches is atop the current Qualcomm tree to avoid
conflicts.  Assuming it looks OK, presumably the best way for it to
land would be to get an Ack from Greg and then Bjorn or Andy could
land it.


Douglas Anderson (2):
  serial: qcom_geni_serial: Make kgdb work even if UART isn't console
  serial: qcom_geni_serial: Always use 4 bytes per TX FIFO word

 drivers/tty/serial/qcom_geni_serial.c | 129 ++++++++++++++++++--------
 1 file changed, 88 insertions(+), 41 deletions(-)

Comments

Greg KH June 27, 2020, 1:59 p.m. UTC | #1

On Fri, Jun 26, 2020 at 01:00:31PM -0700, Douglas Anderson wrote:
> This series of two patches gets rid of some ugly hacks that were in
> the qcom_geni_serial driver around dealing with a port that was used
> for console output and dealing with a port that was being used for
> kgdb.
> 
> While the character reading/writing code is now slightly more complex,
> it's better to be consistently configuring the serial port the same
> way and doing so avoids some corner cases where the old hacks weren't
> always catching properly.
> 
> This change is slightly larger than it needs to be because I was
> trying not to use global variables in the read/write functions.
> Unfortunately the functions were sometimes called earlycon which
> didn't have any "private_data" pointer set.  I've tried to do the
> minimal change here to have some shared "private_data" that's always
> present, but longer term it wouldn't hurt to see if we could unify
> more.
> 
> Greg / Andy / Bjorn:
> 
> This series of patches is atop the current Qualcomm tree to avoid
> conflicts.  Assuming it looks OK, presumably the best way for it to
> land would be to get an Ack from Greg and then Bjorn or Andy could
> land it.

Acked-by: Greg Kroah-Hartman <[e-mail]>