-----Original Message----- From: Richard Weinberger <richard@xxxxxx> Sent: Tuesday, May 9, 2023 1:28 PM To: Christian Loehle <CLoehle@xxxxxxxxxxxxxx> Cc: Biju Das <biju.das.jz@xxxxxxxxxxxxxx>; linux-mmc <linux-mmc@xxxxxxxxxxxxxxx>; linux-renesas-soc <linux-renesas-soc@xxxxxxxxxxxxxxx>; wsa+renesas <wsa+renesas@xxxxxxxxxxxxxxxxxxxx>; ulf hansson <ulf.hansson@xxxxxxxxxx> Subject: Re: Poor write performance to boot area using rcar-gen3-sdhi > So I went down the rabbit hole. > After analyzing the operation traces from Linux and u-boot in more detail I had another chat with the vendor. It turned out their boot area *is* different, it operates on a much larger internal block size. > > While u-boot does basically all IO with one op code, Linux's writeback via page cache writes 4k wise which leads to catastrophic performance. > Using O_DIRECT mitigates the issue, though. Interesting, without having tried it out I would expect the 4k writebacks to be merged to a bigger IO, too? Given that they are filled sequentially, which is the case. But good that you got it working. If I find the time I will play with it a bit. > > Thanks everyone for your valuable input! > //richard > Hyperstone GmbH | Reichenaustr. 39a | 78467 Konstanz Managing Director: Dr. Jan Peter Berns. Commercial register of local courts: Freiburg HRB381782
