| Message ID | 76d83c2b-fd2c-35c1-cfd9-883037167716@siemens.com |
|---|---|
| State | Not Applicable |
| Delegated to: | Tom Rini |
| Headers | show |
| Series | [U-Boot] drivers: block: Allow to cache 4K blocks by default | expand |
On 12/09/2018 09:55 AM, Jan Kiszka wrote: > The default settings of the block cache so far only allow to hold single > 1K blocks. However, larger filesystems tend to use 4K. Failing to cache > those massively degrades access performance unless you manually tune the > cache first. This is not desirable. > > Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com> I submitted the same patch before [1], oddly enough it was not picked by Tom. [1] http://u-boot.10912.n7.nabble.com/PATCH-RFC-blk-Increase-cache-element-size-td337975.html > --- > > My "load mmc is slow" problem reappeared on AM65xx after growing the > rootfs to almost 1 GB - despite an active block cache. > > drivers/block/blkcache.c | 2 +- > 1 file changed, 1 insertion(+), 1 deletion(-) > > diff --git a/drivers/block/blkcache.c b/drivers/block/blkcache.c > index 294511fcdb..1fa64989d3 100644 > --- a/drivers/block/blkcache.c > +++ b/drivers/block/blkcache.c > @@ -24,7 +24,7 @@ struct block_cache_node { > static LIST_HEAD(block_cache); > > static struct block_cache_stats _stats = { > - .max_blocks_per_entry = 2, > + .max_blocks_per_entry = 8, > .max_entries = 32 > }; > >
On 10.12.18 05:13, Marek Vasut wrote: > On 12/09/2018 09:55 AM, Jan Kiszka wrote: >> The default settings of the block cache so far only allow to hold single >> 1K blocks. However, larger filesystems tend to use 4K. Failing to cache >> those massively degrades access performance unless you manually tune the >> cache first. This is not desirable. >> >> Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com> > > I submitted the same patch before [1], oddly enough it was not picked by > Tom. Maybe because the follow-up discussion felt like the topic didn't settle? I would vote for resolving this 4K think first, because it's very real. OTW: You patch should go in first of all, then better solutions can be developed on top. Jan > > [1] > http://u-boot.10912.n7.nabble.com/PATCH-RFC-blk-Increase-cache-element-size-td337975.html > >> --- >> >> My "load mmc is slow" problem reappeared on AM65xx after growing the >> rootfs to almost 1 GB - despite an active block cache. >> >> drivers/block/blkcache.c | 2 +- >> 1 file changed, 1 insertion(+), 1 deletion(-) >> >> diff --git a/drivers/block/blkcache.c b/drivers/block/blkcache.c >> index 294511fcdb..1fa64989d3 100644 >> --- a/drivers/block/blkcache.c >> +++ b/drivers/block/blkcache.c >> @@ -24,7 +24,7 @@ struct block_cache_node { >> static LIST_HEAD(block_cache); >> >> static struct block_cache_stats _stats = { >> - .max_blocks_per_entry = 2, >> + .max_blocks_per_entry = 8, >> .max_entries = 32 >> }; >> >> > >
On Mon, Dec 10, 2018 at 05:13:58AM +0100, Marek Vasut wrote: > On 12/09/2018 09:55 AM, Jan Kiszka wrote: > > The default settings of the block cache so far only allow to hold single > > 1K blocks. However, larger filesystems tend to use 4K. Failing to cache > > those massively degrades access performance unless you manually tune the > > cache first. This is not desirable. > > > > Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com> > > I submitted the same patch before [1], oddly enough it was not picked by > Tom. > > [1] > http://u-boot.10912.n7.nabble.com/PATCH-RFC-blk-Increase-cache-element-size-td337975.html Based on a quick re-read, I had filed that patch as "Marek says he'll look into a more complex but general scheme". If that's not the case then yes, we can just do that. Thanks!
On 12/10/2018 01:28 PM, Tom Rini wrote: > On Mon, Dec 10, 2018 at 05:13:58AM +0100, Marek Vasut wrote: >> On 12/09/2018 09:55 AM, Jan Kiszka wrote: >>> The default settings of the block cache so far only allow to hold single >>> 1K blocks. However, larger filesystems tend to use 4K. Failing to cache >>> those massively degrades access performance unless you manually tune the >>> cache first. This is not desirable. >>> >>> Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com> >> >> I submitted the same patch before [1], oddly enough it was not picked by >> Tom. >> >> [1] >> http://u-boot.10912.n7.nabble.com/PATCH-RFC-blk-Increase-cache-element-size-td337975.html > > Based on a quick re-read, I had filed that patch as "Marek says he'll > look into a more complex but general scheme". If that's not the case > then yes, we can just do that. Thanks! I still don't see the original patch in u-boot/master
On 20.12.18 09:45, Marek Vasut wrote: > On 12/10/2018 01:28 PM, Tom Rini wrote: >> On Mon, Dec 10, 2018 at 05:13:58AM +0100, Marek Vasut wrote: >>> On 12/09/2018 09:55 AM, Jan Kiszka wrote: >>>> The default settings of the block cache so far only allow to hold single >>>> 1K blocks. However, larger filesystems tend to use 4K. Failing to cache >>>> those massively degrades access performance unless you manually tune the >>>> cache first. This is not desirable. >>>> >>>> Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com> >>> >>> I submitted the same patch before [1], oddly enough it was not picked by >>> Tom. >>> >>> [1] >>> http://u-boot.10912.n7.nabble.com/PATCH-RFC-blk-Increase-cache-element-size-td337975.html >> >> Based on a quick re-read, I had filed that patch as "Marek says he'll >> look into a more complex but general scheme". If that's not the case >> then yes, we can just do that. Thanks! > > I still don't see the original patch in u-boot/master > Ping. Some form of this change should go into the next release, IMHO. Jan
On Thu, Jan 03, 2019 at 04:42:08PM +0100, Jan Kiszka wrote: > On 20.12.18 09:45, Marek Vasut wrote: > >On 12/10/2018 01:28 PM, Tom Rini wrote: > >>On Mon, Dec 10, 2018 at 05:13:58AM +0100, Marek Vasut wrote: > >>>On 12/09/2018 09:55 AM, Jan Kiszka wrote: > >>>>The default settings of the block cache so far only allow to hold single > >>>>1K blocks. However, larger filesystems tend to use 4K. Failing to cache > >>>>those massively degrades access performance unless you manually tune the > >>>>cache first. This is not desirable. > >>>> > >>>>Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com> > >>> > >>>I submitted the same patch before [1], oddly enough it was not picked by > >>>Tom. > >>> > >>>[1] > >>>http://u-boot.10912.n7.nabble.com/PATCH-RFC-blk-Increase-cache-element-size-td337975.html > >> > >>Based on a quick re-read, I had filed that patch as "Marek says he'll > >>look into a more complex but general scheme". If that's not the case > >>then yes, we can just do that. Thanks! > > > >I still don't see the original patch in u-boot/master > > > > Ping. Some form of this change should go into the next release, IMHO. So, this initially tripped over some EFI related issue, which has been fixed. But at this point I'm going to pull it in from v2019.04 instead.
diff --git a/drivers/block/blkcache.c b/drivers/block/blkcache.c index 294511fcdb..1fa64989d3 100644 --- a/drivers/block/blkcache.c +++ b/drivers/block/blkcache.c @@ -24,7 +24,7 @@ struct block_cache_node { static LIST_HEAD(block_cache); static struct block_cache_stats _stats = { - .max_blocks_per_entry = 2, + .max_blocks_per_entry = 8, .max_entries = 32 };
The default settings of the block cache so far only allow to hold single 1K blocks. However, larger filesystems tend to use 4K. Failing to cache those massively degrades access performance unless you manually tune the cache first. This is not desirable. Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com> --- My "load mmc is slow" problem reappeared on AM65xx after growing the rootfs to almost 1 GB - despite an active block cache. drivers/block/blkcache.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)