Skip to contentSkip to conversation

env_mmc: fix buffer allocation for armv7

PATCH In review Markus Niebel 3 participants 12 messages Oct 4, 2013 – Oct 14, 2013

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 env_mmc: fix buffer allocation for armv7

From: Markus Niebel <[email protected]>

commit d196bd880347373237d73e0d115b4d51c68cf2ad adds redundand environment to mmc. The usage of malloc in env_relocate_spec triggers cache errors on armv7.

Tested on a not mainlined i.MX53 board:

Board: TQMa53
I2C:   ready
DRAM:  512 MiB
MMC:   FSL_SDHC: 0, FSL_SDHC: 1
ERROR: v7_dcache_inval_range - start address is not aligned - 0x8f57c2d8
ERROR: v7_dcache_inval_range - stop address is not aligned - 0x8f57e2d8
ERROR: v7_dcache_inval_range - start address is not aligned - 0x8f57e2e0
ERROR: v7_dcache_inval_range - stop address is not aligned - 0x8f5802e0

Using default environment

---
 common/env_mmc.c |    9 ++++-----
 1 file changed, 4 insertions(+), 5 deletions(-)
common/env_mmc.c +4 -5
index 65aafa9..204d23b 100644
@@ -192,11 +192,12 @@ void env_relocate_spec(void)
 	u32 offset1, offset2;
 	int read1_fail = 0, read2_fail = 0;
 	int crc1_ok = 0, crc2_ok = 0;
-	env_t *ep, *tmp_env1, *tmp_env2;
+	env_t *ep;
 	int ret;
 
-	tmp_env1 = (env_t *)malloc(CONFIG_ENV_SIZE);
-	tmp_env2 = (env_t *)malloc(CONFIG_ENV_SIZE);
+	ALLOC_CACHE_ALIGN_BUFFER(env_t, tmp_env1, 1);
+	ALLOC_CACHE_ALIGN_BUFFER(env_t, tmp_env2, 1);
+
 	if (tmp_env1 == NULL || tmp_env2 == NULL) {
 		puts("Can't allocate buffers for environment\n");
 		ret = 1;
@@ -266,8 +267,6 @@ err:
 	if (ret)
 		set_default_env(NULL);
 
-	free(tmp_env1);
-	free(tmp_env2);
 #endif
 }
 #else /* ! CONFIG_ENV_OFFSET_REDUND */
Signature · 1 line
1.7.9.5
#1 lore raw

PATCH 1/1 env_mmc: fix buffer allocation for armv7

Quoted 16 lines · Markus Niebel

From: Markus Niebel <[email protected]>

commit d196bd880347373237d73e0d115b4d51c68cf2ad adds redundand environment to mmc. The usage of malloc in env_relocate_spec triggers cache errors on armv7.

Tested on a not mainlined i.MX53 board:

Board: TQMa53
I2C:   ready
DRAM:  512 MiB
MMC:   FSL_SDHC: 0, FSL_SDHC: 1
ERROR: v7_dcache_inval_range - start address is not aligned - 0x8f57c2d8
ERROR: v7_dcache_inval_range - stop address is not aligned - 0x8f57e2d8
ERROR: v7_dcache_inval_range - start address is not aligned - 0x8f57e2e0
ERROR: v7_dcache_inval_range - stop address is not aligned - 0x8f5802e0
Quoted 3 lines

Using default environment

I really don't like this. We're now allocating for example 256KiB on the stack, rather than malloc. I posted a patch recently to convert the non-redundant case to malloc instead for this reason. I believe the answer is we need to be using memalign here, like common/bouncebuf.c::bounce_buffer_start does. Can you do this? If not, can you test a patch? Thanks.

Signature · 8 lines
Tom
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 836 bytes
Desc: Digital signature
URL: <http://lists.denx.de/pipermail/u-boot/attachments/20131004/c09e8ca6/attachment.pgp>
#2 lore raw

PATCH 1/1 env_mmc: fix buffer allocation for armv7

Dear Tom Rini,

Quoted 7 lines · In message <20131004170203.GL15917@bill-the-cat> you wrote:

I really don't like this. We're now allocating for example 256KiB on the stack, rather than malloc. I posted a patch recently to convert the non-redundant case to malloc instead for this reason. I believe the answer is we need to be using memalign here, like common/bouncebuf.c::bounce_buffer_start does. Can you do this? If not, can you test a patch? Thanks.

What exactly don't you like in using the stack for temporary data buffers? That's what it has been invented for. Using malloc() is only useful when the allocated buffers neet to be kept across file scope, which appears not to be the case here.

For file scope buffers, usign the satck is the most efficient and preferred approach - it's fast and results in minimal (virtually no) code.

Why do you hesitate to use the stack?

Best regards,

Wolfgang Denk

Signature · 6 lines
DENX Software Engineering GmbH,     MD: Wolfgang Denk & Detlev Zundel
HRB 165235 Munich, Office: Kirchenstr.5, D-82194 Groebenzell, Germany
Phone: (+49)-8142-66989-10 Fax: (+49)-8142-66989-80 Email: wd at denx.de
Humanity has the  stars  in  its  future,  and  that  future  is  too
important  to be lost under the burden of juvenile folly and ignorant
superstition.                                          - Isaac Asimov
#3 lore raw

PATCH 1/1 env_mmc: fix buffer allocation for armv7

Quoted 16 lines · Wolfgang Denk

Dear Tom Rini,

Quoted 7 lines · In message <20131004170203.GL15917@bill-the-cat> you wrote:

I really don't like this. We're now allocating for example 256KiB on the stack, rather than malloc. I posted a patch recently to convert the non-redundant case to malloc instead for this reason. I believe the answer is we need to be using memalign here, like common/bouncebuf.c::bounce_buffer_start does. Can you do this? If not, can you test a patch? Thanks.

What exactly don't you like in using the stack for temporary data buffers? That's what it has been invented for. Using malloc() is only useful when the allocated buffers neet to be kept across file scope, which appears not to be the case here.

Quoted 5 lines

For file scope buffers, usign the satck is the most efficient and preferred approach - it's fast and results in minimal (virtually no) code.

Why do you hesitate to use the stack?

Mainly to allow us to work in restricted stack areas like SPL where we simply may not have that much space available.

Signature · 8 lines
Tom
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 836 bytes
Desc: Digital signature
URL: <http://lists.denx.de/pipermail/u-boot/attachments/20131006/c2026da9/attachment.pgp>
#4 lore raw

PATCH 1/1 env_mmc: fix buffer allocation for armv7

Dear Tom,

Quoted 5 lines · In message <20131006204214.GO15917@bill-the-cat> you wrote:
Quoted 2 lines

Why do you hesitate to use the stack?

Mainly to allow us to work in restricted stack areas like SPL where we simply may not have that much space available.

But malloc() is drawing from the very same resource as the stack, even more so: with a buffer on the stack, the memory isfreed completeky upon return from the fucntion, with no reminders left. With malloc() you need to statically allocate the malloc pool size for the whole runtime, and allocating the buffers may fragment tha area, so even after freeing the buffers there is less space left for other purposes.

Especially in memory-tight situations you want to avoid malloc().

Best regards,

Wolfgang Denk

Signature · 5 lines
DENX Software Engineering GmbH,     MD: Wolfgang Denk & Detlev Zundel
HRB 165235 Munich, Office: Kirchenstr.5, D-82194 Groebenzell, Germany
Phone: (+49)-8142-66989-10 Fax: (+49)-8142-66989-80 Email: wd at denx.de
It's all Klatchian to me.
        - Terry Pratchett & Stephen Briggs, _The Discworld Companion_
#5 lore raw

PATCH 1/1 env_mmc: fix buffer allocation for armv7

Quoted 9 lines · Wolfgang Denk

Dear Tom,

Quoted 5 lines · In message <20131006204214.GO15917@bill-the-cat> you wrote:
Quoted 2 lines

Why do you hesitate to use the stack?

Mainly to allow us to work in restricted stack areas like SPL where we simply may not have that much space available.

Quoted 8 lines

But malloc() is drawing from the very same resource as the stack, even more so: with a buffer on the stack, the memory isfreed completeky upon return from the fucntion, with no reminders left. With malloc() you need to statically allocate the malloc pool size for the whole runtime, and allocating the buffers may fragment tha area, so even after freeing the buffers there is less space left for other purposes.

Especially in memory-tight situations you want to avoid malloc().

Ah, but in these cases we don't have stack room, period. We have a malloc pool. So unless we make SPL move its stack pointer into DDR from wherever we set the initial one to, the other option here is to just restrict real env support to NOR (and we already don't allow embedded env, since that's embedded within U-Boot, not SPL).

Quoted 15 lines

Best regards,

Wolfgang Denk

Signature · 9 lines
DENX Software Engineering GmbH,     MD: Wolfgang Denk & Detlev Zundel
HRB 165235 Munich, Office: Kirchenstr.5, D-82194 Groebenzell, Germany
Phone: (+49)-8142-66989-10 Fax: (+49)-8142-66989-80 Email: wd at denx.de
It's all Klatchian to me.
        - Terry Pratchett & Stephen Briggs, _The Discworld Companion_
_______________________________________________
U-Boot mailing list
U-Boot at lists.denx.de
http://lists.denx.de/mailman/listinfo/u-boot
Signature · 8 lines
Tom
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 836 bytes
Desc: Digital signature
URL: <http://lists.denx.de/pipermail/u-boot/attachments/20131007/f878ad42/attachment.pgp>
#6 lore raw

PATCH 1/1 env_mmc: fix buffer allocation for armv7

Dear Tom,

Quoted 8 lines · In message <20131007122020.GT15917@bill-the-cat> you wrote:
Quoted 8 lines

But malloc() is drawing from the very same resource as the stack, even more so: with a buffer on the stack, the memory isfreed completeky upon return from the fucntion, with no reminders left. With malloc() you need to statically allocate the malloc pool size for the whole runtime, and allocating the buffers may fragment tha area, so even after freeing the buffers there is less space left for other purposes.

Quoted 7 lines
Quoted 1 line

Especially in memory-tight situations you want to avoid malloc().

Ah, but in these cases we don't have stack room, period. We have a malloc pool. So unless we make SPL move its stack pointer into DDR from wherever we set the initial one to, the other option here is to just restrict real env support to NOR (and we already don't allow embedded env, since that's embedded within U-Boot, not SPL).

Well, if we have DDR such that we can use it for the malloc arena, we also should use it for the stack. Or is there a good reason for not doing this? It would solve all these issues at the root...

Best regards,

Wolfgang Denk

Signature · 5 lines
DENX Software Engineering GmbH,     MD: Wolfgang Denk & Detlev Zundel
HRB 165235 Munich, Office: Kirchenstr.5, D-82194 Groebenzell, Germany
Phone: (+49)-8142-66989-10 Fax: (+49)-8142-66989-80 Email: wd at denx.de
Q:  Do you know what the death rate around here is?
A:  One per person.
#7 lore raw

PATCH 1/1 env_mmc: fix buffer allocation for armv7

Quoted 19 lines · Wolfgang Denk

Dear Tom,

Quoted 8 lines · In message <20131007122020.GT15917@bill-the-cat> you wrote:
Quoted 8 lines

But malloc() is drawing from the very same resource as the stack, even more so: with a buffer on the stack, the memory isfreed completeky upon return from the fucntion, with no reminders left. With malloc() you need to statically allocate the malloc pool size for the whole runtime, and allocating the buffers may fragment tha area, so even after freeing the buffers there is less space left for other purposes.

Quoted 7 lines
Quoted 1 line

Especially in memory-tight situations you want to avoid malloc().

Ah, but in these cases we don't have stack room, period. We have a malloc pool. So unless we make SPL move its stack pointer into DDR from wherever we set the initial one to, the other option here is to just restrict real env support to NOR (and we already don't allow embedded env, since that's embedded within U-Boot, not SPL).

Quoted 3 lines

Well, if we have DDR such that we can use it for the malloc arena, we also should use it for the stack. Or is there a good reason for not doing this? It would solve all these issues at the root...

Making SPL more complex for everyone? We would need to do a fairly good-sized re-jigger of SPL to setup and swap around stack pointers like we do in full U-Boot.

Signature · 8 lines
Tom
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 836 bytes
Desc: Digital signature
URL: <http://lists.denx.de/pipermail/u-boot/attachments/20131008/242ff7d8/attachment.pgp>
#8 lore raw

PATCH 1/1 env_mmc: fix buffer allocation for armv7

Dear Tom,

Quoted 8 lines · In message <20131008134456.GB15917@bill-the-cat> you wrote:
Quoted 4 lines

Well, if we have DDR such that we can use it for the malloc arena, we also should use it for the stack. Or is there a good reason for not doing this? It would solve all these issues at the root...

Making SPL more complex for everyone? We would need to do a fairly good-sized re-jigger of SPL to setup and swap around stack pointers like we do in full U-Boot.

Hm, I'm not convinced. As proposed, we make the code bigger, less efficient, more error prone and more ugly for everyone, not only for SPL users. Aslo, this might not be the only place where buffers or such may be kept on the stack. I hope you don't want to change all these?

Really, if we have the resources, we should use them. If RAM is abailable, it should also be used for the stack. Just using it for malloc() is neither fish nor fowl.

Best regards,

Wolfgang Denk

Signature · 4 lines
DENX Software Engineering GmbH,     MD: Wolfgang Denk & Detlev Zundel
HRB 165235 Munich, Office: Kirchenstr.5, D-82194 Groebenzell, Germany
Phone: (+49)-8142-66989-10 Fax: (+49)-8142-66989-80 Email: wd at denx.de
"Have you lived in this village all your life?"        "No, not yet."
#9 lore raw

PATCH 1/1 env_mmc: fix buffer allocation for armv7

Quoted 18 lines · Wolfgang Denk

Dear Tom,

Quoted 8 lines · In message <20131008134456.GB15917@bill-the-cat> you wrote:
Quoted 4 lines

Well, if we have DDR such that we can use it for the malloc arena, we also should use it for the stack. Or is there a good reason for not doing this? It would solve all these issues at the root...

Making SPL more complex for everyone? We would need to do a fairly good-sized re-jigger of SPL to setup and swap around stack pointers like we do in full U-Boot.

Hm, I'm not convinced. As proposed, we make the code bigger, less efficient, more error prone and more ugly for everyone, not only for SPL users. Aslo, this might not be the only place where buffers or such may be kept on the stack. I hope you don't want to change all these?

Quoted 3 lines

Really, if we have the resources, we should use them. If RAM is abailable, it should also be used for the stack. Just using it for malloc() is neither fish nor fowl.

I'll ceed the point and re-work things on my series then. Markus, your patch is good as-is and I shall pick it up shortly.

Signature · 8 lines
Tom
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 836 bytes
Desc: Digital signature
URL: <http://lists.denx.de/pipermail/u-boot/attachments/20131011/1b527155/attachment.pgp>
#10 lore raw

PATCH 1/1 env_mmc: fix buffer allocation for armv7

In reply to Tom Rini

Hello,

Am 04.10.2013 19:02, wrote Tom Rini:

Quoted 20 lines · Markus Niebel
Quoted 19 lines

From: Markus Niebel <[email protected]>

commit d196bd880347373237d73e0d115b4d51c68cf2ad adds redundand environment to mmc. The usage of malloc in env_relocate_spec triggers cache errors on armv7.

Tested on a not mainlined i.MX53 board:

Board: TQMa53
I2C:   ready
DRAM:  512 MiB
MMC:   FSL_SDHC: 0, FSL_SDHC: 1
ERROR: v7_dcache_inval_range - start address is not aligned - 0x8f57c2d8
ERROR: v7_dcache_inval_range - stop address is not aligned - 0x8f57e2d8
ERROR: v7_dcache_inval_range - start address is not aligned - 0x8f57e2e0
ERROR: v7_dcache_inval_range - stop address is not aligned - 0x8f5802e0

Using default environment

Quoted 9 lines
Quoted 1 line

I really don't like this. We're now allocating for example 256KiB on the stack, rather than malloc. I posted a patch recently to convert the non-redundant case to malloc instead for this reason. I believe the answer is we need to be using memalign here, like common/bouncebuf.c::bounce_buffer_start does. Can you do this? If not, can you test a patch? Thanks.

Thanks for reviewing. I will just follow the ongoing discussion and wait for consensus on how to fix the error before I do tests.

Markus

#11 lore raw
In reply to Markus Niebel
Quoted 16 lines · Markus Niebel

From: Markus Niebel <[email protected]>

commit d196bd880347373237d73e0d115b4d51c68cf2ad adds redundand environment to mmc. The usage of malloc in env_relocate_spec triggers cache errors on armv7.

Tested on a not mainlined i.MX53 board:

Board: TQMa53
I2C:   ready
DRAM:  512 MiB
MMC:   FSL_SDHC: 0, FSL_SDHC: 1
ERROR: v7_dcache_inval_range - start address is not aligned - 0x8f57c2d8
ERROR: v7_dcache_inval_range - stop address is not aligned - 0x8f57e2d8
ERROR: v7_dcache_inval_range - start address is not aligned - 0x8f57e2e0
ERROR: v7_dcache_inval_range - stop address is not aligned - 0x8f5802e0
Quoted 3 lines

Using default environment

Applied to u-boot/master, thanks!

Signature · 8 lines
Tom
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 836 bytes
Desc: Digital signature
URL: <http://lists.denx.de/pipermail/u-boot/attachments/20131014/1a434e44/attachment.pgp>