Skip to contentSkip to conversation

disk: Don't loop over MAX_SEARCH_PARTITIONS in part_create_block_devices()

RFC PATCH In review Stefan Roese 2 participants 4 messages Jan 10, 2023 – Jan 11, 2023

lore
Share & tools
lore
#0 lore raw

RFC PATCH 1/1 disk: Don't loop over MAX_SEARCH_PARTITIONS in part_create_block_devices()

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.

Cc Simon Glass <[email protected]>
Cc AKASHI Takahiro <[email protected]>
Cc Heinrich Schuchardt <[email protected]>

---
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(-)
disk/disk-uclass.c +1 -1
index d32747e2242d..2999f7285b5a 100644
@@ -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);
 
Signature · 1 line
2.39.0
#1 lore raw

Hi,

Thank you for catching this issue.

Quoted 7 lines · Stefan Roese

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.

Quoted 8 lines
Cc Simon Glass <[email protected]>
Cc AKASHI Takahiro <[email protected]>
Cc Heinrich Schuchardt <[email protected]>

---
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

Quoted 20 lines
---
 disk/disk-uclass.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
disk/disk-uclass.c +1 -1
index d32747e2242d..2999f7285b5a 100644
@@ -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);
 
Signature · 1 line
2.39.0
#2 lore raw
Quoted 21 lines · AKASHI Takahiro

Hi,

Thank you for catching this issue.

Quoted 7 lines · Stefan Roese

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.

Quoted 8 lines
Cc Simon Glass <[email protected]>
Cc AKASHI Takahiro <[email protected]>
Cc Heinrich Schuchardt <[email protected]>

---
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.

Quoted 4 lines

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?

Quoted 1 line

I somehow confirmed this against a GPT partition by using gdisk.

How exactly did you confirm this?

Thanks,
Stefan

Quoted 19 lines

-Takahiro Akashi

Quoted 17 lines
---
  disk/disk-uclass.c | 2 +-
  1 file changed, 1 insertion(+), 1 deletion(-)
disk/disk-uclass.c +1 -1
index d32747e2242d..2999f7285b5a 100644
@@ -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);
 
Quoted 3 lines
Quoted 3 lines
Signature · 1 line
2.39.0

Viele Grüße,
Stefan Roese

Signature · 3 lines
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: [email protected]
#3 lore raw
Quoted 22 lines · Stefan Roese
Quoted 21 lines · AKASHI Takahiro

Hi,

Thank you for catching this issue.

Quoted 15 lines · Stefan Roese

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.

Cc Simon Glass <[email protected]>
Cc AKASHI Takahiro <[email protected]>
Cc Heinrich Schuchardt <[email protected]>

---
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.

Quoted 7 lines
Quoted 4 lines

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.

Quoted 3 lines
Quoted 1 line

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

Quoted 32 lines

Thanks,
Stefan

Quoted 6 lines

-Takahiro Akashi

Quoted 4 lines
---
  disk/disk-uclass.c | 2 +-
  1 file changed, 1 insertion(+), 1 deletion(-)
Quoted 15 lines
Quoted 15 lines
disk/disk-uclass.c +1 -1
index d32747e2242d..2999f7285b5a 100644
@@ -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);
Signature · 1 line
2.39.0

Viele Grüße,
Stefan Roese

Signature · 3 lines
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: [email protected]