[PATCH v3 16/29] serial: Add semihosting driver

[Sean Anderson]

On 3/28/22 2:35 AM, Simon Glass wrote:
> Hi Sean,
> 
> On Tue, 22 Mar 2022 at 15:00, Sean Anderson <sean.anderson at seco.com> wrote:
>>
>> This adds a serial driver which uses semihosting calls to read and write
>> to the host's console. For convenience, if CONFIG_DM_SERIAL is enabled,
>> we will instantiate a serial driver. This allows users to enable this
>> driver (which has no physical device) without modifying their device
>> trees or board files. We also implement a non-DM driver for SPL, or for
>> much faster output in U-Boot proper.
>>
>> There are three ways to print to the console:
>>
>> Method              Baud
>> ================== =====
>> smh_putc in a loop   170
>> smh_puts            1600
>> smh_write with :tt 20000
>> ================== =====
>>
>> These speeds were measured using a 175 character message with a J-Link
>> adapter. For reference, U-Boot typically prints around 2700 characters
>> during boot on this board. There are two major factors affecting the
>> speed of these functions. First, each breakpoint incurs a delay. Second,
>> each debugger memory transaction incurs a delay. smh_putc has a
>> breakpoint and memory transaction for every character. smh_puts has one
>> breakpoint, but still has to use a transaction for every character. This
>> is because we don't know the length up front, so OpenOCD has to check if
>> each character is nul. smh_write has only one breakpoint and one memory
>> transfer.
>>
>> DM serial drivers can only implement a putc interface, so we are stuck
>> with the slowest API. Non-DM drivers can implement puts, which is vastly
>> more efficient. When the driver starts up, we try to open :tt. Since
>> this is an extension, this may fail. If it does, we fall back to
>> smh_puts. We don't check :semihosting-features, since there are
>> nonconforming implementations (OpenOCD) which don't implement it (but
>> *do* implement :tt).
>>
>> Some semihosting implementations (QEMU) don't handle READC properly. To
>> work around this, we try to use open/read (much like for stdin) if
>> possible.
>>
>> There is no non-blocking I/O available, so we don't implement pending.
>> This will cause __serial_tstc to always return true. If
>> CONFIG_SERIAL_RX_BUFFER is enabled, _serial_tstc will try and read
>> characters forever. To avoid this, we depend on this config being
>> disabled.
>>
>> Signed-off-by: Sean Anderson <sean.anderson at seco.com>
>> ---
>>
>> (no changes since v2)
>>
>> Changes in v2:
>> - Fix baud numbers being off by 10
>> - Fix typos in commit message
>> - Rename non-DM driver struct to match format of other drivers
>>
>>  drivers/serial/Kconfig              |  22 +++++
>>  drivers/serial/Makefile             |   1 +
>>  drivers/serial/serial.c             |   2 +
>>  drivers/serial/serial_semihosting.c | 147 ++++++++++++++++++++++++++++
>>  include/serial.h                    |   1 +
>>  5 files changed, 173 insertions(+)
>>  create mode 100644 drivers/serial/serial_semihosting.c
>>
> 
> Reviewed-by: Simon Glass <sjg at chromium.org>
> 
> But please can we drop the non-DM support?

Unfortunately, Layerscape does not support DM serial. I tried converting
it, but I ran into some unusual aborts. At the moment, I don't have time
to debug things further. And I thought that non-DM serial was ok for
SPL?

--Sean