Hi Some more info: the actual regression was introduced with the commit: ath10k: enable ASPM - reverting this change only is enough to have card operating as expected.
Still, this patch would apply, but in slightly different manner. On OpenWrt router, masked value would always read 0x27 with two cards tested QCA9862 and QCA9880 v1. But on HP laptop, QCA9862 masked value reading was 0x0f, and writing 0x27 to this register still seems fine, as long as the faulty ASPM change is kept reverted. I did also notice about 10°C temperature drop in ath10k_hwmon and no more crashes when resuming from sleep, but that might be related to the fact these tests were now done with backports-5.1 instead of the in-tree 4.9 driver.
