On Sun, Nov 27, 2016 at 3:25 PM, Alex Lemberg <Alex.Lemberg@xxxxxxxxxxx> wrote: > I understand your concerns about the current packed commands > support and the problem of transition to blk_mq. It may be possible to re-implement the feature later, but then only if we get some queue inspection features in blk-mq. Do you agree with my assessment in the thread that this feature can only really make an improvement on random writes, i.e. it will not improve reads (even random reads) and it will not improve sequential writes? > But as I mentioned in previous email thread, the packed commands > feature is still in use on eMMC4.5 or eMMC5.1 devices on hosts > without SW/HW CMDQ support. True. In the market place of devices, sure. But noone is maintaining or testing the code in the Linux kernel. Noone bothered to enable it for their platform so we can test it. > Taking this feature out meaning stop supporting one of important > performance features of eMMC devices. > Looking forward this feature is replaced with a CMDQ, but should we > kill it for those who doesn’t have a CMDQ? What we need from memory card vendors like you (SanDisk) is a target test rig. That is: a board with an SoC that we developers can get hold of (for reasonable cost) and where both the host controller and the eMMC card has the desired features so we can test it. Also that this board has support in the upstream kernel. There are too many vendor trees with too much forked code out there, I know this is common, but we cannot develop important core features in the kernel that have no in-kernel user and no way for us to test it if we need to re-write the code, it is just unmaintainable. Yours, Linus Walleij -- To unsubscribe from this list: send the line "unsubscribe linux-mmc" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
