Re: [PATCH 04/23] dmaengine: sdxi: Feature discovery and initial configuration

["Lynch, Nathan" <nathan.lynch@amd.com>]

On 4/20/2026 1:51 AM, Frank Li wrote:
> On Fri, Apr 10, 2026 at 08:07:14AM -0500, Nathan Lynch wrote:
>> After bus-specific initialization, force the SDXI function to stopped
>> state. This is the expected state from reset, but kexec or driver bugs
>> can leave a function in other states from which the initialization
>> code must be able to recover.
> 
> Move this paragraph to last because this major funciton is
> "Discover via the capability register ..."
> 

OK.


>> +/* Get the device to the GSV_STOP state. */
>> +static int sdxi_dev_stop(struct sdxi_dev *sdxi)
>> +{
>> +     unsigned long deadline = jiffies + msecs_to_jiffies(1000);
>> +     bool reset_issued = false;
>> +
>> +     do {
>> +             enum sdxi_fn_gsv status = sdxi_dev_gsv(sdxi);
>> +
>> +             sdxi_dbg(sdxi, "%s: function state: %s\n", __func__, gsv_str(status));
>> +
>> +             switch (status) {
>> +             case SDXI_GSV_ACTIVE:
>> +                     sdxi_write_fn_gsr(sdxi, SDXI_GSRV_STOP_SF);
>> +                     break;
>> +             case SDXI_GSV_ERROR:
>> +                     if (!reset_issued) {
>> +                             sdxi_info(sdxi,
>> +                                       "function in error state, issuing reset\n");
>> +                             sdxi_write_fn_gsr(sdxi, SDXI_GSRV_RESET);
>> +                             reset_issued = true;
>> +                     } else {
>> +                             fsleep(1000);
>> +                     }
>> +                     break;
>> +             case SDXI_GSV_STOP:
>> +                     return 0;
>> +             case SDXI_GSV_INIT:
>> +             case SDXI_GSV_STOPG_SF:
>> +             case SDXI_GSV_STOPG_HD:
>> +                     /* transitional states, wait */
>> +                     sdxi_dbg(sdxi, "waiting for stop (gsv = %u)\n",
>> +                              status);
>> +                     fsleep(1000);
>> +                     break;
>> +             default:
>> +                     sdxi_err(sdxi, "unknown gsv %u, giving up\n", status);
>> +                     return -EIO;
>> +             }
>> +     } while (time_before(jiffies, deadline));
> 
> does read_poll_timeout() work?

Maybe so. Thanks for the pointer.


>> +#define sdxi_dbg(s, fmt, ...) dev_dbg(sdxi_to_dev(s), fmt, ## __VA_ARGS__)
>> +#define sdxi_info(s, fmt, ...) dev_info(sdxi_to_dev(s), fmt, ## __VA_ARGS__)
>> +#define sdxi_err(s, fmt, ...) dev_err(sdxi_to_dev(s), fmt, ## __VA_ARGS__)
> 
> suggest direct use dev_*

OK. I guess these would be more justified if the driver were adding
SDXI-specific information such as group ID or sfunc to the messages, but
it's not as of now.