Skip to contentSkip to conversation

spl: ram: Do not memcpy() identical buffers

PATCH New Marek Vasut 3 participants 14 messages May 28, 2019 – Jul 12, 2019

lore
Share & tools
lore
#0 lore raw

PATCH 1/3 spl: ram: Do not memcpy() identical buffers

If the source and destination buffer address is identical, there is no need to memcpy() the content. Skip the memcpy() in such a case.

Cc Michal Simek <[email protected]>
Cc Tom Rini <[email protected]>
---
 common/spl/spl_ram.c | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)
common/spl/spl_ram.c +2 -1
index 954e91a004..2ef33f717e 100644
@@ -24,7 +24,8 @@ static ulong spl_ram_load_read(struct spl_load_info *load, ulong sector,
 {
 	debug("%s: sector %lx, count %lx, buf %lx\n",
 	      __func__, sector, count, (ulong)buf);
-	memcpy(buf, (void *)(CONFIG_SPL_LOAD_FIT_ADDRESS + sector), count);
+	if (buf != CONFIG_SPL_LOAD_FIT_ADDRESS + sector)
+		memcpy(buf, (void *)(CONFIG_SPL_LOAD_FIT_ADDRESS + sector), count);
 	return count;
 }
 
Signature · 1 line
2.20.1
#1 lore raw

PATCH 2/3 spl: ubi: Add support for loading simple fitImage

Add code to load simple fitImage from the kernel volume, the args volume is unnecessary.

Cc Ladislav Michl <[email protected]>
Cc Tom Rini <[email protected]>
---
 common/spl/spl_ubi.c | 22 +++++++++++++++++++++-
 1 file changed, 21 insertions(+), 1 deletion(-)
common/spl/spl_ubi.c +21 -1
index 67e5fadd7c..6addd46ece 100644
@@ -11,6 +11,14 @@
 #include <ubispl.h>
 #include <spl.h>
 
+static ulong spl_ubi_load_read(struct spl_load_info *load, ulong sector,
+			       ulong count, void *buf)
+{
+	if (buf != (void *)CONFIG_SPL_LOAD_FIT_ADDRESS + sector)
+		memcpy(buf, (void *)(CONFIG_SYS_LOAD_ADDR + sector), count);
+	return count;
+}
+
 int spl_ubi_load_image(struct spl_image_info *spl_image,
 		       struct spl_boot_device *bootdev)
 {
@@ -54,7 +62,19 @@ int spl_ubi_load_image(struct spl_image_info *spl_image,
 		ret = ubispl_load_volumes(&info, volumes, 2);
 		if (!ret) {
 			header = (struct image_header *)volumes[0].load_addr;
-			spl_parse_image_header(spl_image, header);
+
+			if (IS_ENABLED(CONFIG_SPL_LOAD_FIT) &&
+			    image_get_magic(header) == FDT_MAGIC) {
+				struct spl_load_info load;
+
+				debug("Found FIT\n");
+				load.bl_len = 1;
+				load.read = spl_ubi_load_read;
+				spl_load_simple_fit(spl_image, &load, 0, header);
+			} else {
+				spl_parse_image_header(spl_image, header);
+			}
+
 			puts("Linux loaded.\n");
 			goto out;
 		}
Signature · 1 line
2.20.1
#2 lore raw

PATCH 2/3 spl: ubi: Add support for loading simple fitImage

Quoted 40 lines · Marek Vasut

Add code to load simple fitImage from the kernel volume, the args volume is unnecessary.

Cc Ladislav Michl <[email protected]>
Cc Tom Rini <[email protected]>
---
 common/spl/spl_ubi.c | 22 +++++++++++++++++++++-
 1 file changed, 21 insertions(+), 1 deletion(-)
common/spl/spl_ubi.c +14 -1
index 67e5fadd7c..6addd46ece 100644
@@ -11,6 +11,14 @@
 #include <ubispl.h>
 #include <spl.h>
 
+static ulong spl_ubi_load_read(struct spl_load_info *load, ulong sector,
+			       ulong count, void *buf)
+{
+	if (buf != (void *)CONFIG_SPL_LOAD_FIT_ADDRESS + sector)
+		memcpy(buf, (void *)(CONFIG_SYS_LOAD_ADDR + sector), count);
+	return count;
+}
+
 int spl_ubi_load_image(struct spl_image_info *spl_image,
 		       struct spl_boot_device *bootdev)
 {
@@ -54,7 +62,19 @@ int spl_ubi_load_image(struct spl_image_info *spl_image,
 		ret = ubispl_load_volumes(&info, volumes, 2);
 		if (!ret) {
 			header = (struct image_header *)volumes[0].load_addr;
-			spl_parse_image_header(spl_image, header);
+
+			if (IS_ENABLED(CONFIG_SPL_LOAD_FIT) &&
+			    image_get_magic(header) == FDT_MAGIC) {
+				struct spl_load_info load;
+
+				debug("Found FIT\n");

common/spl/spl_ubi.c int spl_ubi_load_image(struct spl_image_info *spl_image, · line 80

Quoted 10 lines
common/spl/spl_ubi.c +7 -0
+				load.bl_len = 1;
+				load.read = spl_ubi_load_read;
+				spl_load_simple_fit(spl_image, &load, 0, header);
+			} else {
+				spl_parse_image_header(spl_image, header);
+			}
+
 			puts("Linux loaded.\n");
 			goto out;
 		}

This breaks building on igep00x0 and am335x_igep003x, sorry.

Signature · 8 lines
Tom
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 819 bytes
Desc: not available
URL: <http://lists.denx.de/pipermail/u-boot/attachments/20190712/4525bf96/attachment.sig>
#3 lore raw

PATCH 3/3 spl: fit: Place OS_BOOT DTB at CONFIG_SYS_SPL_ARGS_ADDR if defined

In reply to Marek Vasut

If SPL OS_BOOT is enabled and CONFIG_SYS_SPL_ARGS_ADDR is defined, place the kernel DTB at that location, so the rest of the OS_BOOT machinery in SPL has the DTB blob at the correct location.

---
 common/spl/spl_fit.c | 6 +++++-
 1 file changed, 5 insertions(+), 1 deletion(-)
common/spl/spl_fit.c +5 -1
index 87ecf0bb9e..e3312b3477 100644
@@ -291,7 +291,11 @@ static int spl_fit_append_fdt(struct spl_image_info *spl_image,
 	 * Read the device tree and place it after the image.
 	 * Align the destination address to ARCH_DMA_MINALIGN.
 	 */
+#if defined(CONFIG_SPL_OS_BOOT) && defined(CONFIG_SYS_SPL_ARGS_ADDR)
+	image_info.load_addr = CONFIG_SYS_SPL_ARGS_ADDR;
+#else
 	image_info.load_addr = spl_image->load_addr + spl_image->size;
+#endif
 	ret = spl_load_fit_image(info, sector, fit, base_offset, node,
 				 &image_info);
 
@@ -472,7 +476,7 @@ int spl_load_simple_fit(struct spl_image_info *spl_image,
 	 * Booting a next-stage U-Boot may require us to append the FDT.
 	 * We allow this to fail, as the U-Boot image might embed its FDT.
 	 */
-	if (spl_image->os == IH_OS_U_BOOT)
+	if (spl_image->os == IH_OS_U_BOOT || spl_image->os == IH_OS_LINUX)
 		spl_fit_append_fdt(spl_image, info, sector, fit,
 				   images, base_offset);
 
Signature · 1 line
2.20.1
#4 lore raw

PATCH 3/3 spl: fit: Place OS_BOOT DTB at CONFIG_SYS_SPL_ARGS_ADDR if defined

Quoted 27 lines · Marek Vasut

If SPL OS_BOOT is enabled and CONFIG_SYS_SPL_ARGS_ADDR is defined, place the kernel DTB at that location, so the rest of the OS_BOOT machinery in SPL has the DTB blob at the correct location.

---
 common/spl/spl_fit.c | 6 +++++-
 1 file changed, 5 insertions(+), 1 deletion(-)
common/spl/spl_fit.c +4 -0
index 87ecf0bb9e..e3312b3477 100644
@@ -291,7 +291,11 @@ static int spl_fit_append_fdt(struct spl_image_info *spl_image,
 	 * Read the device tree and place it after the image.
 	 * Align the destination address to ARCH_DMA_MINALIGN.
 	 */
+#if defined(CONFIG_SPL_OS_BOOT) && defined(CONFIG_SYS_SPL_ARGS_ADDR)
+	image_info.load_addr = CONFIG_SYS_SPL_ARGS_ADDR;
+#else
 	image_info.load_addr = spl_image->load_addr + spl_image->size;
+#endif
 	ret = spl_load_fit_image(info, sector, fit, base_offset, node,
 				 &image_info);
 

common/spl/spl_fit.c int spl_load_simple_fit(struct spl_image_info *spl_image, · line 481

Quoted 7 lines
common/spl/spl_fit.c +1 -1
@@ -472,7 +476,7 @@ int spl_load_simple_fit(struct spl_image_info *spl_image,
 	 * Booting a next-stage U-Boot may require us to append the FDT.
 	 * We allow this to fail, as the U-Boot image might embed its FDT.
 	 */
-	if (spl_image->os == IH_OS_U_BOOT)
+	if (spl_image->os == IH_OS_U_BOOT || spl_image->os == IH_OS_LINUX)
 		spl_fit_append_fdt(spl_image, info, sector, fit,
 				   images, base_offset);

As-is and I haven't tried to debug this yet, it breaks the U-Boot booting case for am335x_evm.

Signature · 8 lines
Tom
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 819 bytes
Desc: not available
URL: <http://lists.denx.de/pipermail/u-boot/attachments/20190712/446f8cd8/attachment.sig>
#5 lore raw

PATCH 1/3 spl: ram: Do not memcpy() identical buffers

In reply to Marek Vasut
Quoted 6 lines · Marek Vasut

If the source and destination buffer address is identical, there is no need to memcpy() the content. Skip the memcpy() in such a case.

Cc Michal Simek <[email protected]>
Cc Tom Rini <[email protected]>

Shouldn't memcpy catch that itself?

Signature · 8 lines
Tom
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 819 bytes
Desc: not available
URL: <http://lists.denx.de/pipermail/u-boot/attachments/20190527/19b85e66/attachment.sig>
#6 lore raw

PATCH 1/3 spl: ram: Do not memcpy() identical buffers

Quoted 11 lines · Tom Rini
Quoted 6 lines · Marek Vasut

If the source and destination buffer address is identical, there is no need to memcpy() the content. Skip the memcpy() in such a case.

Cc Michal Simek <[email protected]>
Cc Tom Rini <[email protected]>

Shouldn't memcpy catch that itself?

memcpy(3) says
The memcpy() function copies n bytes from memory area src to
memory area dest. The memory areas must not overlap. Use memmove(3) if the memory areas do overlap.

Signature · 2 lines
Best regards,
Marek Vasut
#7 lore raw

PATCH 1/3 spl: ram: Do not memcpy() identical buffers

Quoted 10 lines · Marek Vasut
Quoted 9 lines · Tom Rini
Quoted 6 lines · Marek Vasut

If the source and destination buffer address is identical, there is no need to memcpy() the content. Skip the memcpy() in such a case.

Cc Michal Simek <[email protected]>
Cc Tom Rini <[email protected]>
Quoted 6 lines
Quoted 2 lines

Shouldn't memcpy catch that itself?

memcpy(3) says
The memcpy() function copies n bytes from memory area src to
memory area dest. The memory areas must not overlap. Use memmove(3) if the memory areas do overlap.

OK, and shouldn't memcpy optimize that case? Does it usually?

Signature · 8 lines
Tom
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 819 bytes
Desc: not available
URL: <http://lists.denx.de/pipermail/u-boot/attachments/20190527/15958c78/attachment.sig>
#8 lore raw

PATCH 1/3 spl: ram: Do not memcpy() identical buffers

Quoted 11 lines · Tom Rini
Quoted 10 lines · Marek Vasut
Quoted 9 lines · Tom Rini
Quoted 6 lines · Marek Vasut

If the source and destination buffer address is identical, there is no need to memcpy() the content. Skip the memcpy() in such a case.

Cc Michal Simek <[email protected]>
Cc Tom Rini <[email protected]>
Quoted 8 lines
Quoted 6 lines
Quoted 2 lines

Shouldn't memcpy catch that itself?

memcpy(3) says
The memcpy() function copies n bytes from memory area src to
memory area dest. The memory areas must not overlap. Use memmove(3) if the memory areas do overlap.

OK, and shouldn't memcpy optimize that case? Does it usually?

As the manpage says "The memory areas must not overlap." , I would expect it does not have to ?

Signature · 2 lines
Best regards,
Marek Vasut
#9 lore raw

PATCH 1/3 spl: ram: Do not memcpy() identical buffers

Quoted 19 lines · Marek Vasut
Quoted 18 lines · Tom Rini
Quoted 16 lines · Marek Vasut
Quoted 11 lines · Tom Rini
Quoted 6 lines · Marek Vasut

If the source and destination buffer address is identical, there is no need to memcpy() the content. Skip the memcpy() in such a case.

Cc Michal Simek <[email protected]>
Cc Tom Rini <[email protected]>

Shouldn't memcpy catch that itself?

memcpy(3) says
The memcpy() function copies n bytes from memory area src to
memory area dest. The memory areas must not overlap. Use memmove(3) if the memory areas do overlap.

Quoted 4 lines
Quoted 1 line

OK, and shouldn't memcpy optimize that case? Does it usually?

As the manpage says "The memory areas must not overlap." , I would expect it does not have to ?

I guess I'm not being clear enough, sorry. Go look at how this is implemented in a few places please and report back to us. Someone else, or many someone else, have probably already figured out if optimizing this case in general, in memcpy, is a good idea or not. Thanks!

Signature · 8 lines
Tom
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 819 bytes
Desc: not available
URL: <http://lists.denx.de/pipermail/u-boot/attachments/20190527/e0e7fc28/attachment.sig>
#10 lore raw

PATCH 1/3 spl: ram: Do not memcpy() identical buffers

Quoted 22 lines · Tom Rini
Quoted 21 lines · Marek Vasut
Quoted 19 lines · Tom Rini
Quoted 10 lines · Marek Vasut
Quoted 9 lines · Tom Rini
Quoted 6 lines · Marek Vasut

If the source and destination buffer address is identical, there is no need to memcpy() the content. Skip the memcpy() in such a case.

Cc Michal Simek <[email protected]>
Cc Tom Rini <[email protected]>
Quoted 6 lines
Quoted 2 lines

Shouldn't memcpy catch that itself?

memcpy(3) says
The memcpy() function copies n bytes from memory area src to
memory area dest. The memory areas must not overlap. Use memmove(3) if the memory areas do overlap.

OK, and shouldn't memcpy optimize that case? Does it usually?

Quoted 7 lines
Quoted 2 lines

As the manpage says "The memory areas must not overlap." , I would expect it does not have to ?

I guess I'm not being clear enough, sorry. Go look at how this is implemented in a few places please and report back to us. Someone else, or many someone else, have probably already figured out if optimizing this case in general, in memcpy, is a good idea or not. Thanks!

If even [1] says the behavior is undefined, then what does it matter whether some implementation added an optimization for such undefined behavior? We cannot depend on it, can we ?

[1] https://pubs.opengroup.org/onlinepubs/009695399/functions/memcpy.html

Signature · 2 lines
Best regards,
Marek Vasut
#11 lore raw

PATCH 1/3 spl: ram: Do not memcpy() identical buffers

Quoted 31 lines · Marek Vasut
Quoted 22 lines · Tom Rini
Quoted 21 lines · Marek Vasut
Quoted 19 lines · Tom Rini
Quoted 10 lines · Marek Vasut
Quoted 9 lines · Tom Rini
Quoted 6 lines · Marek Vasut

If the source and destination buffer address is identical, there is no need to memcpy() the content. Skip the memcpy() in such a case.

Cc Michal Simek <[email protected]>
Cc Tom Rini <[email protected]>
Quoted 6 lines
Quoted 2 lines

Shouldn't memcpy catch that itself?

memcpy(3) says
The memcpy() function copies n bytes from memory area src to
memory area dest. The memory areas must not overlap. Use memmove(3) if the memory areas do overlap.

OK, and shouldn't memcpy optimize that case? Does it usually?

Quoted 7 lines
Quoted 2 lines

As the manpage says "The memory areas must not overlap." , I would expect it does not have to ?

I guess I'm not being clear enough, sorry. Go look at how this is implemented in a few places please and report back to us. Someone else, or many someone else, have probably already figured out if optimizing this case in general, in memcpy, is a good idea or not. Thanks!

Quoted 5 lines

If even [1] says the behavior is undefined, then what does it matter whether some implementation added an optimization for such undefined behavior? We cannot depend on it, can we ?

[1] https://pubs.opengroup.org/onlinepubs/009695399/functions/memcpy.html

Yes, yes it would be worth seeing if other groups that have looked into the performance impact here have also decided to optimize this undefined behavior or not, thanks.

Signature · 8 lines
Tom
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 819 bytes
Desc: not available
URL: <http://lists.denx.de/pipermail/u-boot/attachments/20190528/2057183a/attachment.sig>
#12 lore raw

PATCH 1/3 spl: ram: Do not memcpy() identical buffers

Quoted 27 lines · Tom Rini
Quoted 26 lines · Marek Vasut
Quoted 25 lines · Tom Rini
Quoted 13 lines · Marek Vasut
Quoted 12 lines · Tom Rini
Quoted 11 lines · Marek Vasut
Quoted 10 lines · Tom Rini
Quoted 6 lines · Marek Vasut

If the source and destination buffer address is identical, there is no need to memcpy() the content. Skip the memcpy() in such a case.

Cc Michal Simek <[email protected]>
Cc Tom Rini <[email protected]>

Shouldn't memcpy catch that itself?

Quoted 7 lines
Quoted 5 lines
Quoted 4 lines

memcpy(3) says
The memcpy() function copies n bytes from memory area src to
memory area dest. The memory areas must not overlap. Use memmove(3) if the memory areas do overlap.

OK, and shouldn't memcpy optimize that case? Does it usually?

As the manpage says "The memory areas must not overlap." , I would expect it does not have to ?

I guess I'm not being clear enough, sorry. Go look at how this is implemented in a few places please and report back to us. Someone else, or many someone else, have probably already figured out if optimizing this case in general, in memcpy, is a good idea or not. Thanks!

Quoted 8 lines
Quoted 5 lines

If even [1] says the behavior is undefined, then what does it matter whether some implementation added an optimization for such undefined behavior? We cannot depend on it, can we ?

[1] https://pubs.opengroup.org/onlinepubs/009695399/functions/memcpy.html

Yes, yes it would be worth seeing if other groups that have looked into the performance impact here have also decided to optimize this undefined behavior or not, thanks.

I don't think this is an optimization question, really. Calling memcpy with overlapping areas is an error. The result is explicitly undefined. It may well be that all the existing implementations effectively do nothing, either by explicit check or by actually copying the data over itself. However, to rely on that behavior is asking for trouble down the line. Undefined behavior is exactly that. Don't do it.

Quoted 5 lines

_______________________________________________
U-Boot mailing list
U-Boot at lists.denx.de
https://lists.denx.de/listinfo/u-boot

#13 lore raw

PATCH 1/3 spl: ram: Do not memcpy() identical buffers

In reply to Tom Rini
Quoted 36 lines · Tom Rini
Quoted 35 lines · Marek Vasut
Quoted 22 lines · Tom Rini
Quoted 21 lines · Marek Vasut
Quoted 19 lines · Tom Rini
Quoted 10 lines · Marek Vasut
Quoted 9 lines · Tom Rini
Quoted 6 lines · Marek Vasut

If the source and destination buffer address is identical, there is no need to memcpy() the content. Skip the memcpy() in such a case.

Cc Michal Simek <[email protected]>
Cc Tom Rini <[email protected]>
Quoted 6 lines
Quoted 2 lines

Shouldn't memcpy catch that itself?

memcpy(3) says
The memcpy() function copies n bytes from memory area src to
memory area dest. The memory areas must not overlap. Use memmove(3) if the memory areas do overlap.

OK, and shouldn't memcpy optimize that case? Does it usually?

Quoted 7 lines
Quoted 2 lines

As the manpage says "The memory areas must not overlap." , I would expect it does not have to ?

I guess I'm not being clear enough, sorry. Go look at how this is implemented in a few places please and report back to us. Someone else, or many someone else, have probably already figured out if optimizing this case in general, in memcpy, is a good idea or not. Thanks!

If even [1] says the behavior is undefined, then what does it matter whether some implementation added an optimization for such undefined behavior? We cannot depend on it, can we ?

Quoted 5 lines
Quoted 1 line

[1] https://pubs.opengroup.org/onlinepubs/009695399/functions/memcpy.html

Yes, yes it would be worth seeing if other groups that have looked into the performance impact here have also decided to optimize this undefined behavior or not, thanks.

I will just drop this patch, since U-Boot memcpy() implementation does this check. But let me be very clear here, that check is part of undefined behavior (!) and I don't think it's the right thing to do in memcpy() itself.

Signature · 2 lines
Best regards,
Marek Vasut