I have used O_DIRECT with aligned buffers of 4k size (the default linux page size). I have even tried fadvise calls according to Linus's suggestion of not using the O_DIRECT method. None of the above method causes the write call to fail and media errors to be propagated to my program. It is handled at the driver / kernel level (either by retrying / remapping the sector). Please advise. Thanks. On Thu, Mar 4, 2010 at 11:37 AM, Mike Hayward <hayward@xxxxxxxx> wrote: > I always take it for granted, but forgot to mention, you should also > use O_DIRECT to bypass the linux buffer cache. It often gets in the > way of error propagation since it is changing your io requests into > it's own page sized ios and will also "lie" to you about having > written your data in the first place since it's a write back cache. > > The point is you have to disable all the caches everywhere or the > error information will get absorbed by the caches. > > - Mike > -- 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