* [RFC PATCH] disk: Don't loop over MAX_SEARCH_PARTITIONS in part_create_block_devices()
@ 2023-01-10 8:00 Stefan Roese
2023-01-11 1:32 ` AKASHI Takahiro
0 siblings, 1 reply; 4+ messages in thread
From: Stefan Roese @ 2023-01-10 8:00 UTC (permalink / raw)
To: u-boot; +Cc: Simon Glass, AKASHI Takahiro, Heinrich Schuchardt
I've noticed that the first ext4 file loading from a MMC partition for
ZynqMP takes quite some time (~ 1 second). Debugging showed, that the
MMC driver reads the partition info 128 time (MAX_SEARCH_PARTITIONS)
resulting in this boot delay. To fix this, let's just end creating the
block drives in part_create_block_devices() when no more valid partition
is found. This reduces the first file reading from ~0.9s to ~0.3s.
Signed-off-by: Stefan Roese <sr@denx.de>
Cc: Simon Glass <sjg@chromium.org>
Cc: AKASHI Takahiro <takahiro.akashi@linaro.org>
Cc: Heinrich Schuchardt <xypron.glpk@gmx.de>
---
Reasoning for RFC:
I did not dig into the current disk / partition stuff too deeply, so I'm
not 100% sure that this patch does not break anything.
---
disk/disk-uclass.c | 2 +-
1 file changed, 1 insertion(+), 1 deletion(-)
diff --git a/disk/disk-uclass.c b/disk/disk-uclass.c
index d32747e2242d..2999f7285b5a 100644
--- a/disk/disk-uclass.c
+++ b/disk/disk-uclass.c
@@ -36,7 +36,7 @@ int part_create_block_devices(struct udevice *blk_dev)
/* Add devices for each partition */
for (count = 0, part = 1; part <= MAX_SEARCH_PARTITIONS; part++) {
if (part_get_info(desc, part, &info))
- continue;
+ break;
snprintf(devname, sizeof(devname), "%s:%d", blk_dev->name,
part);
--
2.39.0
^ permalink raw reply related [flat|nested] 4+ messages in thread
* Re: [RFC PATCH] disk: Don't loop over MAX_SEARCH_PARTITIONS in part_create_block_devices()
2023-01-10 8:00 [RFC PATCH] disk: Don't loop over MAX_SEARCH_PARTITIONS in part_create_block_devices() Stefan Roese
@ 2023-01-11 1:32 ` AKASHI Takahiro
2023-01-11 8:05 ` Stefan Roese
0 siblings, 1 reply; 4+ messages in thread
From: AKASHI Takahiro @ 2023-01-11 1:32 UTC (permalink / raw)
To: Stefan Roese; +Cc: u-boot, Simon Glass, Heinrich Schuchardt
Hi,
Thank you for catching this issue.
On Tue, Jan 10, 2023 at 09:00:38AM +0100, Stefan Roese wrote:
> I've noticed that the first ext4 file loading from a MMC partition for
> ZynqMP takes quite some time (~ 1 second). Debugging showed, that the
> MMC driver reads the partition info 128 time (MAX_SEARCH_PARTITIONS)
> resulting in this boot delay. To fix this, let's just end creating the
> block drives in part_create_block_devices() when no more valid partition
> is found. This reduces the first file reading from ~0.9s to ~0.3s.
>
> Signed-off-by: Stefan Roese <sr@denx.de>
> Cc: Simon Glass <sjg@chromium.org>
> Cc: AKASHI Takahiro <takahiro.akashi@linaro.org>
> Cc: Heinrich Schuchardt <xypron.glpk@gmx.de>
> ---
> Reasoning for RFC:
> I did not dig into the current disk / partition stuff too deeply, so I'm
> not 100% sure that this patch does not break anything.
I'm afraid that this fix won't work for all the partition types,
especially for those in which entries in a partition table can be sparsely filled,
or in other words, valid partition numbers may not always be contiguous
even if they don't reach a maximum number.
I somehow confirmed this against a GPT partition by using gdisk.
-Takahiro Akashi
> ---
> disk/disk-uclass.c | 2 +-
> 1 file changed, 1 insertion(+), 1 deletion(-)
>
> diff --git a/disk/disk-uclass.c b/disk/disk-uclass.c
> index d32747e2242d..2999f7285b5a 100644
> --- a/disk/disk-uclass.c
> +++ b/disk/disk-uclass.c
> @@ -36,7 +36,7 @@ int part_create_block_devices(struct udevice *blk_dev)
> /* Add devices for each partition */
> for (count = 0, part = 1; part <= MAX_SEARCH_PARTITIONS; part++) {
> if (part_get_info(desc, part, &info))
> - continue;
> + break;
> snprintf(devname, sizeof(devname), "%s:%d", blk_dev->name,
> part);
>
> --
> 2.39.0
>
^ permalink raw reply [flat|nested] 4+ messages in thread
* Re: [RFC PATCH] disk: Don't loop over MAX_SEARCH_PARTITIONS in part_create_block_devices()
2023-01-11 1:32 ` AKASHI Takahiro
@ 2023-01-11 8:05 ` Stefan Roese
2023-01-11 9:04 ` AKASHI Takahiro
0 siblings, 1 reply; 4+ messages in thread
From: Stefan Roese @ 2023-01-11 8:05 UTC (permalink / raw)
To: AKASHI Takahiro, u-boot, Simon Glass, Heinrich Schuchardt
On 1/11/23 02:32, AKASHI Takahiro wrote:
> Hi,
>
> Thank you for catching this issue.
>
> On Tue, Jan 10, 2023 at 09:00:38AM +0100, Stefan Roese wrote:
>> I've noticed that the first ext4 file loading from a MMC partition for
>> ZynqMP takes quite some time (~ 1 second). Debugging showed, that the
>> MMC driver reads the partition info 128 time (MAX_SEARCH_PARTITIONS)
>> resulting in this boot delay. To fix this, let's just end creating the
>> block drives in part_create_block_devices() when no more valid partition
>> is found. This reduces the first file reading from ~0.9s to ~0.3s.
>>
>> Signed-off-by: Stefan Roese <sr@denx.de>
>> Cc: Simon Glass <sjg@chromium.org>
>> Cc: AKASHI Takahiro <takahiro.akashi@linaro.org>
>> Cc: Heinrich Schuchardt <xypron.glpk@gmx.de>
>> ---
>> Reasoning for RFC:
>> I did not dig into the current disk / partition stuff too deeply, so I'm
>> not 100% sure that this patch does not break anything.
>
> I'm afraid that this fix won't work for all the partition types,
> especially for those in which entries in a partition table can be sparsely filled,
> or in other words, valid partition numbers may not always be contiguous
> even if they don't reach a maximum number.
So what is the maximum number? Where originates the
MAX_SEARCH_PARTITIONS = 128 from?
> I somehow confirmed this against a GPT partition by using gdisk.
How exactly did you confirm this?
Thanks,
Stefan
> -Takahiro Akashi
>
>> ---
>> disk/disk-uclass.c | 2 +-
>> 1 file changed, 1 insertion(+), 1 deletion(-)
>>
>> diff --git a/disk/disk-uclass.c b/disk/disk-uclass.c
>> index d32747e2242d..2999f7285b5a 100644
>> --- a/disk/disk-uclass.c
>> +++ b/disk/disk-uclass.c
>> @@ -36,7 +36,7 @@ int part_create_block_devices(struct udevice *blk_dev)
>> /* Add devices for each partition */
>> for (count = 0, part = 1; part <= MAX_SEARCH_PARTITIONS; part++) {
>> if (part_get_info(desc, part, &info))
>> - continue;
>> + break;
>> snprintf(devname, sizeof(devname), "%s:%d", blk_dev->name,
>> part);
>>
>> --
>> 2.39.0
>>
Viele Grüße,
Stefan Roese
--
DENX Software Engineering GmbH, Managing Director: Erika Unter
HRB 165235 Munich, Office: Kirchenstr.5, D-82194 Groebenzell, Germany
Phone: (+49)-8142-66989-51 Fax: (+49)-8142-66989-80 Email: sr@denx.de
^ permalink raw reply [flat|nested] 4+ messages in thread
* Re: [RFC PATCH] disk: Don't loop over MAX_SEARCH_PARTITIONS in part_create_block_devices()
2023-01-11 8:05 ` Stefan Roese
@ 2023-01-11 9:04 ` AKASHI Takahiro
0 siblings, 0 replies; 4+ messages in thread
From: AKASHI Takahiro @ 2023-01-11 9:04 UTC (permalink / raw)
To: Stefan Roese; +Cc: u-boot, Simon Glass, Heinrich Schuchardt
On Wed, Jan 11, 2023 at 09:05:05AM +0100, Stefan Roese wrote:
> On 1/11/23 02:32, AKASHI Takahiro wrote:
> > Hi,
> >
> > Thank you for catching this issue.
> >
> > On Tue, Jan 10, 2023 at 09:00:38AM +0100, Stefan Roese wrote:
> > > I've noticed that the first ext4 file loading from a MMC partition for
> > > ZynqMP takes quite some time (~ 1 second). Debugging showed, that the
> > > MMC driver reads the partition info 128 time (MAX_SEARCH_PARTITIONS)
> > > resulting in this boot delay. To fix this, let's just end creating the
> > > block drives in part_create_block_devices() when no more valid partition
> > > is found. This reduces the first file reading from ~0.9s to ~0.3s.
> > >
> > > Signed-off-by: Stefan Roese <sr@denx.de>
> > > Cc: Simon Glass <sjg@chromium.org>
> > > Cc: AKASHI Takahiro <takahiro.akashi@linaro.org>
> > > Cc: Heinrich Schuchardt <xypron.glpk@gmx.de>
> > > ---
> > > Reasoning for RFC:
> > > I did not dig into the current disk / partition stuff too deeply, so I'm
> > > not 100% sure that this patch does not break anything.
> >
> > I'm afraid that this fix won't work for all the partition types,
> > especially for those in which entries in a partition table can be sparsely filled,
> > or in other words, valid partition numbers may not always be contiguous
> > even if they don't reach a maximum number.
>
> So what is the maximum number? Where originates the
> MAX_SEARCH_PARTITIONS = 128 from?
commit bc314f8e5f9b ("cmd: part: list all 128 GPT partitions")
But GPT may accept arbitrary number of partitions.
The GPT header has "num_partition_entries" for the table size.
The max depends on partition types.
> > I somehow confirmed this against a GPT partition by using gdisk.
>
> How exactly did you confirm this?
Created partitions for, say, #3 and #5 by 'n' command.
-Takahiro Akashi
> Thanks,
> Stefan
>
> > -Takahiro Akashi
> >
> > > ---
> > > disk/disk-uclass.c | 2 +-
> > > 1 file changed, 1 insertion(+), 1 deletion(-)
> > >
> > > diff --git a/disk/disk-uclass.c b/disk/disk-uclass.c
> > > index d32747e2242d..2999f7285b5a 100644
> > > --- a/disk/disk-uclass.c
> > > +++ b/disk/disk-uclass.c
> > > @@ -36,7 +36,7 @@ int part_create_block_devices(struct udevice *blk_dev)
> > > /* Add devices for each partition */
> > > for (count = 0, part = 1; part <= MAX_SEARCH_PARTITIONS; part++) {
> > > if (part_get_info(desc, part, &info))
> > > - continue;
> > > + break;
> > > snprintf(devname, sizeof(devname), "%s:%d", blk_dev->name,
> > > part);
> > > --
> > > 2.39.0
> > >
>
> Viele Grüße,
> Stefan Roese
>
> --
> DENX Software Engineering GmbH, Managing Director: Erika Unter
> HRB 165235 Munich, Office: Kirchenstr.5, D-82194 Groebenzell, Germany
> Phone: (+49)-8142-66989-51 Fax: (+49)-8142-66989-80 Email: sr@denx.de
^ permalink raw reply [flat|nested] 4+ messages in thread
end of thread, other threads:[~2023-01-11 9:04 UTC | newest]
Thread overview: 4+ messages (download: mbox.gz follow: Atom feed
-- links below jump to the message on this page --
2023-01-10 8:00 [RFC PATCH] disk: Don't loop over MAX_SEARCH_PARTITIONS in part_create_block_devices() Stefan Roese
2023-01-11 1:32 ` AKASHI Takahiro
2023-01-11 8:05 ` Stefan Roese
2023-01-11 9:04 ` AKASHI Takahiro
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox