| Message ID | 1336465602-12627-1-git-send-email-richard@nod.at |
|---|---|
| State | New, archived |
| Headers | show |
On Tue, 2012-05-08 at 10:26 +0200, Richard Weinberger wrote: > The comment above yield()'s definition says: > * If you want to use yield() to wait for something, use wait_event(). > * If you want to use yield() to be 'nice' for others, use cond_resched(). > * If you still want to use yield(), do not! Oh wow, someone actually reads comments, sweet!!
On Tue, 2012-05-08 at 10:26 +0200, Richard Weinberger wrote: > The comment above yield()'s definition says: > * If you want to use yield() to wait for something, use wait_event(). > * If you want to use yield() to be 'nice' for others, use cond_resched(). > * If you still want to use yield(), do not! > > The yield() usage of UBI really looks like the "be nice for others" case, > so use cond_resched(). > > Signed-off-by: Richard Weinberger <richard@nod.at> > CC: peterz@infradead.org I guess we would better sleep for a couple of hundreds of milliseconds instead. The purpose was different. If we have I/O error, we hope it is a transient failure. E.g., the HW is temporary unavailable because of an internal issue. And with yield() we hoped to schedule away for longer time than usual and let other processes which may affect that HW go forward and do something. But whole this thing does not have any solid real-case basis, just something which looked sane to do to me. Someone on the list also questioned the entire re-tries, but he did not care enough to send a tested patch. But this is a separate and independent issue.
Am 08.05.2012 11:02, schrieb Artem Bityutskiy: > On Tue, 2012-05-08 at 10:26 +0200, Richard Weinberger wrote: >> The comment above yield()'s definition says: >> * If you want to use yield() to wait for something, use wait_event(). >> * If you want to use yield() to be 'nice' for others, use cond_resched(). >> * If you still want to use yield(), do not! >> >> The yield() usage of UBI really looks like the "be nice for others" case, >> so use cond_resched(). >> >> Signed-off-by: Richard Weinberger <richard@nod.at> >> CC: peterz@infradead.org > > I guess we would better sleep for a couple of hundreds of milliseconds > instead. So, you'd prefer a msleep_interruptible(500) instead of yield() in this case? > The purpose was different. If we have I/O error, we hope it is a > transient failure. E.g., the HW is temporary unavailable because of an > internal issue. And with yield() we hoped to schedule away for longer > time than usual and let other processes which may affect that HW go > forward and do something. Okay, yield() is clearly the wrong choice here. cond_resched() is better. Thanks, //richard
On Tue, 2012-05-08 at 16:23 +0300, Artem Bityutskiy wrote: > On Tue, 2012-05-08 at 12:59 +0200, Richard Weinberger wrote: > > > The purpose was different. If we have I/O error, we hope it is a > > > transient failure. E.g., the HW is temporary unavailable because of an > > > internal issue. And with yield() we hoped to schedule away for longer > > > time than usual and let other processes which may affect that HW go > > > forward and do something. > > > > Okay, yield() is clearly the wrong choice here. > > cond_resched() is better. > > I do not think it is better. Yes, for I/O I'd suggest > msleep_interruptible(200) instead. At least document the intent of that random sleep. Otherwise we'll be having this same discussion again in a few years or so ;-)
On Tue, 2012-05-08 at 12:59 +0200, Richard Weinberger wrote: > > The purpose was different. If we have I/O error, we hope it is a > > transient failure. E.g., the HW is temporary unavailable because of an > > internal issue. And with yield() we hoped to schedule away for longer > > time than usual and let other processes which may affect that HW go > > forward and do something. > > Okay, yield() is clearly the wrong choice here. > cond_resched() is better. I do not think it is better. Yes, for I/O I'd suggest msleep_interruptible(200) instead.
diff --git a/drivers/mtd/ubi/io.c b/drivers/mtd/ubi/io.c index 43f1a00..a125ef6 100644 --- a/drivers/mtd/ubi/io.c +++ b/drivers/mtd/ubi/io.c @@ -192,7 +192,7 @@ retry: dbg_io("error %d%s while reading %d bytes from PEB " "%d:%d, read only %zd bytes, retry", err, errstr, len, pnum, offset, read); - yield(); + cond_resched(); goto retry; } @@ -366,7 +366,7 @@ retry: if (retries++ < UBI_IO_RETRIES) { dbg_io("error %d while erasing PEB %d, retry", err, pnum); - yield(); + cond_resched(); goto retry; } ubi_err("cannot erase PEB %d, error %d", pnum, err); @@ -384,7 +384,7 @@ retry: if (ei.state == MTD_ERASE_FAILED) { if (retries++ < UBI_IO_RETRIES) { dbg_io("error while erasing PEB %d, retry", pnum); - yield(); + cond_resched(); goto retry; } ubi_err("cannot erase PEB %d", pnum); diff --git a/drivers/mtd/ubi/wl.c b/drivers/mtd/ubi/wl.c index 7c1a9bf..f2f9056 100644 --- a/drivers/mtd/ubi/wl.c +++ b/drivers/mtd/ubi/wl.c @@ -1243,7 +1243,7 @@ retry: */ spin_unlock(&ubi->wl_lock); dbg_wl("the PEB %d is not in proper tree, retry", pnum); - yield(); + cond_resched(); goto retry; }
The comment above yield()'s definition says: * If you want to use yield() to wait for something, use wait_event(). * If you want to use yield() to be 'nice' for others, use cond_resched(). * If you still want to use yield(), do not! The yield() usage of UBI really looks like the "be nice for others" case, so use cond_resched(). Signed-off-by: Richard Weinberger <richard@nod.at> CC: peterz@infradead.org --- drivers/mtd/ubi/io.c | 6 +++--- drivers/mtd/ubi/wl.c | 2 +- 2 files changed, 4 insertions(+), 4 deletions(-)