On Wed, 15 Mar 2023 at 16:31, Jerome Brunet <jbrunet@xxxxxxxxxxxx> wrote: > > > On Wed 15 Mar 2023 at 11:20, Marc Gonzalez <marc.w.gonzalez@xxxxxxx> wrote: > > > On 14/03/2023 20:45, Heiner Kallweit wrote: > > > >> On 14.03.2023 18:24, Marc Gonzalez wrote: > >> > >>> With the default power-up delay, on small kernels, the host probes > >>> too soon, and mmc_send_io_op_cond() times out. > >> > >> Looking at mmc_power_up() and how power_delay_ms is used > >> I wonder what you mean with "host probes too soon". > > > > Hello Heiner, > > > > Thanks for your interest in my patch! :) > > > > I should have added a link to the thread that led to the patch. > > https://patchwork.kernel.org/project/linux-wireless/patch/c1a215cf-94be-871b-2a8a-3cc381588f83@xxxxxxx/ > > Start at "I have run into another issue." > > > > Basically, I have an S905X2-based board. > > I built a small kernel for it (with only a few drivers), that boots really fast. > > > > mmc2 (SDIO controller hooked to WiFi chip) would not probe at all, > > unless I added lots of printks. > > Basically, calling mmc_send_io_op_cond() too soon after the controller > > has been reset leads to the CMD5 request timing out. > > > > > > I tend to agree with Heiner here. > This patch is backing a contraint only reported on your design in the > driver of every AML SoC supported, for every MMC controller. > > I think you should look first in your vmmc and vqmmc regulators and > their setup times. > > "fixed-regulator" have properties which might be interesting to you, > like > * startup-delay-us > * off-on-delay-us If the problem is regulator specific, this would be the correct thing to do. Although, if the problem is pwrseq specific, like that we need a delay after enabling the clock and asserting the GPIO enable pin for the WiFi chip, then we have the "post-power-on-delay-ms" of the pwrseq node to play with instead. > > >> Are you sure that the additional delay is needed for the Amlogic MMC > >> block IP in general? Or could it be that your issue is caused by > >> a specific regulator and you need to add a delay there? > > > > The eternal question... > > > > I have only one type of board. (Actually, I have a reference design > > that is slightly different, so I should test on that one as well.) > > > > In vendor kernels, they add delays to the WiFi drivers. > > Maybe they have run into the issue, and they're just fixing the symptom? > > > > Default value for ios.power_delay_ms is 10 ms. > > msleep(ios.power_delay_ms) is called twice in mmc_power_up(). > > So raising the delay from 10 to 20 adds 20 ms > > to the latency of initializing SDIO/SD/MMC controllers. > > > > Would you be willing to test if the problem manifests on your board? > > > > Regards > Kind regards Uffe
