The card is clearly buggy. CMD49 (write extension register) is explicitly allowed if the card is in CQ mode, and Linux waits until the CQE is idle before doing the write, then polls for cache flush completion. The card reports back that the cache is flushed, then ignores any CQ commands thereafter.
I would wager that the vast majority of cards claiming A2 support have had a very limited set of real-world testing. The main use-case drive for application-class performance is apps on removable mobile storage (i.e. SD cards in phones), but Android doesn't use CQ in the base kernel. Vendor BSPs may have patch sets that enable it, but I've not come across any that publish sources to that effect.
So far, the only flash vendor with a clean bill of health is Sandisk.
I have accumulated a bigger set of patches following more testing here - being more pessimistic about CQ support, more debug, etc. This should hit rpi-update probably by the end of the week.
I would wager that the vast majority of cards claiming A2 support have had a very limited set of real-world testing. The main use-case drive for application-class performance is apps on removable mobile storage (i.e. SD cards in phones), but Android doesn't use CQ in the base kernel. Vendor BSPs may have patch sets that enable it, but I've not come across any that publish sources to that effect.
So far, the only flash vendor with a clean bill of health is Sandisk.
I have accumulated a bigger set of patches following more testing here - being more pessimistic about CQ support, more debug, etc. This should hit rpi-update probably by the end of the week.
Statistics: Posted by jdb — Tue Mar 26, 2024 6:59 pm