I am using 4k aligned buffers for writing and reading. Kernel / driver catches the error during the write cycle and I can get the error messages about the media being bad or sector i/o errors. But it is not propagated to the program and write always passes (even in the case of the device being out of control. (i.e., the device fails to respond to any further open / read / write queries and inaccessible from the core). Isn't the error has to be notified to the program that makes the call? Reading is a completely different scenario and I am disabling the read-ahead cache completely with fadvise call. hdparm is good, but I don't want to use the internal ATA SECURE ERASE because I can never get the amount of bad sectors the drive had. On Thu, Mar 4, 2010 at 9:41 AM, Mark Lord <kernel@xxxxxxxxxxxx> wrote: > On 03/04/10 09:17, Greg Freemyer wrote: > .. >> >> I think / suspect your major problem is you say above that you use a >> 512-byte buffer to wipe with. The kernel is using 4K pages. So when >> you write to a 4K section of the drive for the first time, the kernel >> implements read-modify-write logic. >> >> Your i/o failures are almost certainly on the read cycle of the above, >> not the write cycle. You need to move to 4K buffers and you need to >> ensure your 4K writes are aligned with how the kernel is working with >> the disk. ie. You need your 4K buffer to perfectly align with the >> kernels 4K block handling so you never have a read-modify-write cycle. > > .. > > You'll also need to disable Linux read-ahead for the drive, > or it may try reading beyond even the 4KB block. > > But really.. isn't "hdparm --security-erase NULL /dev/sdX" good enough ??? > > Cheers > -- To unsubscribe, send a message with 'unsubscribe linux-mm' in the body to majordomo@xxxxxxxxxx For more info on Linux MM, see: http://www.linux-mm.org/ . Don't email: <a href