2009-09-01 12:43:54

by Nikanth Karthikesan

Subject: [PATCH] Allow changing max_sectors_kb above the default 512

The patch "block: Use accessor functions for queue limits"
(ae03bf639a5027d27270123f5f6e3ee6a412781d) changed queue_max_sectors_store()
to use blk_queue_max_sectors() instead of directly assigning the value.

But blk_queue_max_sectors() differs a bit
1. It sets both max_sectors_kb, and max_hw_sectors_kb
2. Never allows one to change max_sectors_kb above BLK_DEF_MAX_SECTORS. If one
specifies a value greater then max_hw_sectors is set to that value but
max_sectors is set to BLK_DEF_MAX_SECTORS

I am not sure whether blk_queue_max_sectors() should be changed, as it seems
to be that way for a long time. And there may be callers dependent on that
behaviour.

This patch simply reverts to the older way of directly assigning the value to
max_sectors as it was before.

Signed-off-by: Nikanth Karthikesan <[email protected]>

---

diff --git a/block/blk-sysfs.c b/block/blk-sysfs.c
index 418d636..d3aa2aa 100644
--- a/block/blk-sysfs.c
+++ b/block/blk-sysfs.c
@@ -133,7 +133,7 @@ queue_max_sectors_store(struct request_queue *q, const char *page, size_t count)
return -EINVAL;

spin_lock_irq(q->queue_lock);
- blk_queue_max_sectors(q, max_sectors_kb << 1);
+ q->limits.max_sectors = max_sectors_kb << 1;
spin_unlock_irq(q->queue_lock);

return ret;

2009-09-01 20:39:29

by Jens Axboe

Subject: Re: [PATCH] Allow changing max_sectors_kb above the default 512

On Tue, Sep 01 2009, Nikanth Karthikesan wrote:
> The patch "block: Use accessor functions for queue limits"
> (ae03bf639a5027d27270123f5f6e3ee6a412781d) changed queue_max_sectors_store()
> to use blk_queue_max_sectors() instead of directly assigning the value.
>
> But blk_queue_max_sectors() differs a bit
> 1. It sets both max_sectors_kb, and max_hw_sectors_kb
> 2. Never allows one to change max_sectors_kb above BLK_DEF_MAX_SECTORS. If one
> specifies a value greater then max_hw_sectors is set to that value but
> max_sectors is set to BLK_DEF_MAX_SECTORS
>
> I am not sure whether blk_queue_max_sectors() should be changed, as it seems
> to be that way for a long time. And there may be callers dependent on that
> behaviour.
>
> This patch simply reverts to the older way of directly assigning the value to
> max_sectors as it was before.
>
> Signed-off-by: Nikanth Karthikesan <[email protected]>

Good catch! Applied for 2.6.31.

--
Jens Axboe