On 22-11-01, Marek Vasut wrote: > On 11/1/22 17:51, Marco Felsch wrote: > > On 22-11-01, Marek Vasut wrote: > > > On 11/1/22 17:06, Marco Felsch wrote: > > > > > > Hi, > > > > > > > > > Also I understood the thresholds in such a way, that the FIFO > > > > > > watermark > > > > > > must be higher but there is no place left when it is set to 3/3. In > > > > > > such > > > > > > case this means that the PANIC will never left once it was entered. > > > > > > > > > > I think this part is wrong. > > > > > > > > > > Consider that the FIFO fill drops below 2/3 so PANIC signal asserts. > > > > > > > > ? I thought the PANIC is triggered if the FIFO drops below the 1/3 > > > > threshold and is active till the 2/3 threshold. > > > > > > Yes, although I think the ASSERT/DEASSERT are one-way switches. > > > > > > > > After a bit of time, the FIFO fill reaches full 3/3 (maybe during > > > > > blanking period, where the data can be read in quickly without being > > > > > scanned out again), and the PANIC signal de-asserts. > > > > > > > > > > So the LCDIF won't be in constant PANIC asserted, but it will be > > > > > there for > > > > > quite a bit longer. > > > > > > > > > > > > > It also seems to me that tuning these thresholds might be > > > > > > > > related to > > > > > > > > some special use-case of the SoC, and it is most likely not > > > > > > > > just the > > > > > > > > LCDIF thresholds which have been adjusted in such case, I would > > > > > > > > expect > > > > > > > > the NOC and GPV NIC priorities to be adjusted at that point too. > > > > > > > > > > > > As far as I understood, the PANIC signal triggers the NOC to use the > > > > > > PANIC signal priorities instead of the normal ones. I found a patch > > > > > > laying in our downstream [1] repo which configures the threshold. > > > > > > This > > > > > > raises the question which PANIC prio do you use? Do you have a > > > > > > patch for > > > > > > this? If I remember correctly some TF-A's like the NXP downstream > > > > > > one > > > > > > configure this but the upstream TF-A don't. Which TF-A do you use? > > > > > > > > > > Upstream 2.6 or 2.7 , so this tuning does not apply. > > > > > > > > So your panic priority is what? > > > > > > If you tell me which register (physical address) to read, I will do that > > > on > > > this board right now. > > > > According our patch it is: > > > > #define HDMI_TX_CONTROL0 0x200 > > #define HDMI_LCDIF_NOC_HURRY_PRIO_MASK GENMASK(14, 12) > > #define HDMI_LCDIF_NOC_HURRY_PRIO(p) (((p) << 12) & > > HDMI_LCDIF_NOC_HURRY_PRIO_MASK) > > > > and we set it within imx8mp_hdmi_blk_ctrl_power_on() by: > > > > regmap_set_bits(bc->regmap, HDMI_TX_CONTROL0, > > HDMI_LCDIF_NOC_HURRY_PRIO(7)); > > > > imx8mp_hdmi_blk_ctrl_power_off(): > > > > regmap_clear_bits(bc->regmap, HDMI_TX_CONTROL0, > > HDMI_LCDIF_NOC_HURRY_PRIO(7)); > > Hmmm, bc->regmap does not exist in either upstream or downstream ATF > versions per git grep, maybe you can give me the exact physical address of > the register you would like me to read?
Arg.. sorry if I confused you. This is the code we are using within Linux. We don't wanna use the TF-A to configure these bits. Regards, Marco