Skip to contentSkip to conversation

firmware: make sure builtin firmware is page alignment

PATCH In review Zhang Ning 2 participants 9 messages Aug 3, 2018 – Aug 7, 2018

lore
Share & tools
lore

Zhang Ning

author
#0 lore raw

PATCH 1/1 firmware: make sure builtin firmware is page alignment

when firmware is in filesystem, request_firmware will load it, and copy it to vmalloc memory, that is page align memory.

but when firmware is builtin, it is 8 bytes or 4 bytes alignment.

make sure builtin firmware is page algnment, that can simplify algorithm to handle firmware.

---
 firmware/Makefile | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)
firmware/Makefile +2 -1
index 29641383e136..d7bfce56378f 100644
@@ -16,10 +16,11 @@ quiet_cmd_fwbin = MK_FW   $@
 				firmware/%.gen.S,%,$@))))";		     \
 		  ASM_WORD=$(if $(CONFIG_64BIT),.quad,.long);		     \
 		  ASM_ALIGN=$(if $(CONFIG_64BIT),3,2);			     \
+		  ASM_FW_ALIGN=12;			  	   	     \
 		  PROGBITS=$(if $(CONFIG_ARM),%,@)progbits;		     \
 		  echo "/* Generated by firmware/Makefile */"		> $@;\
 		  echo "    .section .rodata"				>>$@;\
-		  echo "    .p2align $${ASM_ALIGN}"			>>$@;\
+		  echo "    .p2align $${ASM_FW_ALIGN}"			>>$@;\
 		  echo "_fw_$${FWSTR}_bin:"				>>$@;\
 		  echo "    .incbin \"$(2)\""				>>$@;\
 		  echo "_fw_end:"					>>$@;\
Signature · 1 line
2.18.0
#1 lore raw
Quoted 7 lines · Zhang Ning

when firmware is in filesystem, request_firmware will load it, and copy it to vmalloc memory, that is page align memory.

but when firmware is builtin, it is 8 bytes or 4 bytes alignment.

make sure builtin firmware is page algnment, that can simplify algorithm to handle firmware.

How is it simplified? I don't see any such change like that here :(

#2 lore raw
Quoted 4 lines · 在 2018-08-03五的 07:39 +0200,Greg KH写道:
Quoted 3 lines · Zhang Ning

when firmware is in filesystem, request_firmware will load it, and copy it to vmalloc memory, that is page align memory.

Quoted 8 lines
Quoted 5 lines

but when firmware is builtin, it is 8 bytes or 4 bytes alignment.

make sure builtin firmware is page algnment, that can simplify algorithm
to handle firmware.

How is it simplified?  I don't see any such change like that here :(

Thank you for review this patch.

When driver handles its firmware based on page, like below:

	struct page *p;
	p = vmalloc_to_page(fw->data);  // for filesystem firmware
	p = virt_to_page(fw->data); // for builtin firmware

but if builtin firmware is not page alignment, page pointer for builtin firmware is wrong, it contains memory not belong to firmware. drivers has to use additional code to handle this.

if builtin firmware is also page alignment, no need additional code to handle builtin firmware. simplified.

BR.
Ning.

#3 lore raw
Quoted 21 lines · Ning A
Quoted 4 lines · 在 2018-08-03五的 07:39 +0200,Greg KH写道:
Quoted 3 lines · Zhang Ning

when firmware is in filesystem, request_firmware will load it, and copy it to vmalloc memory, that is page align memory.

Quoted 8 lines
Quoted 5 lines

but when firmware is builtin, it is 8 bytes or 4 bytes alignment.

make sure builtin firmware is page algnment, that can simplify algorithm
to handle firmware.

How is it simplified?  I don't see any such change like that here :(

Thank you for review this patch.

When driver handles its firmware based on page, like below:

	struct page *p;
	p = vmalloc_to_page(fw->data);  // for filesystem firmware
	p = virt_to_page(fw->data); // for builtin firmware
Quoted 6 lines

but if builtin firmware is not page alignment, page pointer for builtin firmware is wrong, it contains memory not belong to firmware. drivers has to use additional code to handle this.

if builtin firmware is also page alignment, no need additional code to handle builtin firmware. simplified.

But you did not change anything like this in your code, so why would I know this?

Please fix this up submit this properly.

greg k-h

#4 lore raw
Quoted 31 lines · 在 2018-08-03五的 12:31 +0200,[email protected]写道:
Quoted 30 lines · Ning A
Quoted 7 lines · 在 2018-08-03五的 07:39 +0200,Greg KH写道:
Quoted 6 lines · Zhang Ning

when firmware is in filesystem, request_firmware will load it, and copy it to vmalloc memory, that is page align memory.

but when firmware is builtin, it is 8 bytes or 4 bytes alignment.

Quoted 7 lines
Quoted 3 lines

make sure builtin firmware is page algnment, that can simplify algorithm
to handle firmware.

How is it simplified?  I don't see any such change like that here :(

Thank you for review this patch.

When driver handles its firmware based on  page, like below:

	struct page *p;
	p = vmalloc_to_page(fw->data);  // for filesystem firmware
	p = virt_to_page(fw->data); // for builtin firmware

but if builtin firmware is not page alignment, page pointer for builtin
firmware is wrong, it contains memory not belong to firmware. drivers
has to use additional code to handle this.

Quoted 7 lines
Quoted 3 lines

if builtin firmware is also page alignment, no need additional code to
handle builtin firmware. simplified.

But you did not change anything like this in your code, so why would I
know this?

I understand it is very difficult to review this patch without context. The driver is not opensource, I can't show the patch for driver.

this patch changes kernel common code, and it has value to upstream, so I submit this patch for review and also look forward some advises from community. Once we decide to opensource, the driver changes will be there.

Quoted 3 lines

Please fix this up submit this properly.

greg k-h

#5 lore raw
Quoted 49 lines · Ning A
Quoted 39 lines · 在 2018-08-03五的 12:31 +0200,[email protected]写道:
Quoted 33 lines · Ning A
Quoted 8 lines · 在 2018-08-03五的 07:39 +0200,Greg KH写道:
Quoted 7 lines · Zhang Ning

when firmware is in filesystem, request_firmware will load it,
and copy it to vmalloc memory, that is page align memory.

but when firmware is builtin, it is 8 bytes or 4 bytes alignment.

Quoted 9 lines
Quoted 4 lines

make sure builtin firmware is page algnment, that can simplify
algorithm
to handle firmware.

How is it simplified?  I don't see any such change like that here
:(

Thank you for review this patch.

When driver handles its firmware based on  page, like below:

	struct page *p;
	p = vmalloc_to_page(fw->data);  // for filesystem firmware
	p = virt_to_page(fw->data); // for builtin firmware

but if builtin firmware is not page alignment, page pointer for builtin
firmware is wrong, it contains memory not belong to firmware. drivers
has to use additional code to handle this.

Quoted 4 lines

if builtin firmware is also page alignment, no need additional code
to
handle builtin firmware. simplified.

Quoted 4 lines

But you did not change anything like this in your code, so why would
I
know this?

I understand it is very difficult to review this patch without context.
The driver is not opensource, I can't show the patch for driver.

Quoted 8 lines

this patch changes kernel common code, and it has value to upstream, so
I submit this patch for review and also look forward some advises from
community. Once we decide to opensource, the driver changes will be there.

as said our driver handles firmware based on page, and if firmware is builtin, page point from virt_to_page(fw->data) contains memory doesn't belong to firmware. there are two way to fix it.

1, copy builtin firmware to vmalloc memory, this is additional work to handle firmware. I create a wrap function for firmware request.

int request_XYZ_fw(const struct firmware **firmware_p, const char *name,
 struct device *device)

{
	const struct firmware *fw;
	struct firmware *tmp;
	int ret;

	ret = request_firmware(&fw, name, device);

	if (ret < 0)
		return ret;

	if (is_vmalloc_addr(fw->data))
		*firmware_p = fw;
	else {
		tmp = (struct firmware *)kzalloc(sizeof(struct
firmware), GFP_KERNEL);
		if (!tmp)
			return -ENOMEM;
		tmp->size = fw->size;
		tmp->data = vmalloc(fw->size);
		memcpy(tmp->data, fw->data, fw->size);
		*firmware_p = tmp;
	}
	return ret;
}

2, make builtin firmware is also page alignment. the change in driver code is to check memory type of fw->data, and use different XXX_to_page function.

struct page *p;
if (is_vmalloc_addr(fw->data)) // new

	p = vmalloc_to_page(fw->data);
else                            // new
	p = virt_to_page(fw->data); // new

compare to solution, solution 2 is simple, this is what I said simplified. and solution 2 also save memory (I don't know the affect change memory alignment on memory usage.)

Hi, Greg
is this easier for you to review?

BR.
Ning.

Quoted 3 lines

Please fix this up submit this properly.

greg k-h

#6 lore raw
In reply to Zhang, Ning A
Quoted 36 lines · Ning A
Quoted 35 lines · 在 2018-08-03五的 12:31 +0200,[email protected]写道:
Quoted 33 lines · Ning A
Quoted 7 lines · 在 2018-08-03五的 07:39 +0200,Greg KH写道:
Quoted 6 lines · Zhang Ning

when firmware is in filesystem, request_firmware will load it, and copy it to vmalloc memory, that is page align memory.

but when firmware is builtin, it is 8 bytes or 4 bytes alignment.

Quoted 7 lines
Quoted 3 lines

make sure builtin firmware is page algnment, that can simplify algorithm
to handle firmware.

How is it simplified?  I don't see any such change like that here :(

Thank you for review this patch.

When driver handles its firmware based on  page, like below:

	struct page *p;
	p = vmalloc_to_page(fw->data);  // for filesystem firmware
	p = virt_to_page(fw->data); // for builtin firmware

but if builtin firmware is not page alignment, page pointer for builtin
firmware is wrong, it contains memory not belong to firmware. drivers
has to use additional code to handle this.

if builtin firmware is also page alignment, no need additional code to
handle builtin firmware. simplified.

Quoted 6 lines
Quoted 3 lines

But you did not change anything like this in your code, so why would I
know this?

I understand it is very difficult to review this patch without context. The driver is not opensource, I can't show the patch for driver.

Then I can not accept your patch. Go talk to your corporate lawyers about changing core kernel code for a closed source driver and what that implies about that closed driver (hint, your driver can not be closed...) :)

Then come back with a proper open sourced driver, that's the only way Linux drivers can be written sorry.

best of luck,

greg k-h

#7 lore raw
Quoted 51 lines · 在 2018-08-06一的 16:05 +0200,[email protected]写道:
Quoted 41 lines · Ning A
Quoted 40 lines · 在 2018-08-03五的 12:31 +0200,[email protected]写道:
Quoted 38 lines · Ning A
Quoted 8 lines · 在 2018-08-03五的 07:39 +0200,Greg KH写道:
Quoted 7 lines · Zhang Ning

when firmware is in filesystem, request_firmware will load it,
and copy it to vmalloc memory, that is page align memory.

but when firmware is builtin, it is 8 bytes or 4 bytes alignment.

Quoted 9 lines
Quoted 4 lines

make sure builtin firmware is page algnment, that can simplify
algorithm
to handle firmware.

How is it simplified?  I don't see any such change like that here
:(

Thank you for review this patch.

When driver handles its firmware based on  page, like below:

	struct page *p;
	p = vmalloc_to_page(fw->data);  // for filesystem
firmware
	p = virt_to_page(fw->data); // for builtin firmware

but if builtin firmware is not page alignment, page pointer for builtin
firmware is wrong, it contains memory not belong to firmware. drivers
has to use additional code to handle this.

if builtin firmware is also page alignment, no need additional code
to
handle builtin firmware. simplified.

Quoted 8 lines
Quoted 4 lines

But you did not change anything like this in your code, so why would
I
know this?

I understand it is very difficult to review this patch without context.
The driver is not opensource, I can't show the patch for driver.

Quoted 5 lines

Then I can not accept your patch.  Go talk to your corporate lawyers about changing core kernel code for a closed source driver and what that
implies about that closed driver (hint, your driver can not be closed...) :)

It's very sad, but anyway, thank you for your review. your review will definitely help us improve our work.

BR.
Ning.

Quoted 8 lines

Then come back with a proper open sourced driver, that's the only way Linux drivers can be written sorry.

best of luck,

greg k-h

#8 lore raw
Quoted 52 lines · Ning A
Quoted 51 lines · 在 2018-08-06一的 16:05 +0200,[email protected]写道:
Quoted 49 lines · Ning A
Quoted 40 lines · 在 2018-08-03五的 12:31 +0200,[email protected]写道:
Quoted 34 lines · Ning A
Quoted 8 lines · 在 2018-08-03五的 07:39 +0200,Greg KH写道:
Quoted 7 lines · Zhang Ning

when firmware is in filesystem, request_firmware will load it,
and copy it to vmalloc memory, that is page align memory.

but when firmware is builtin, it is 8 bytes or 4 bytes alignment.

Quoted 9 lines
Quoted 4 lines

make sure builtin firmware is page algnment, that can simplify
algorithm
to handle firmware.

How is it simplified?  I don't see any such change like that here
:(

Thank you for review this patch.

When driver handles its firmware based on  page, like below:

	struct page *p;
	p = vmalloc_to_page(fw->data);  // for filesystem
firmware
	p = virt_to_page(fw->data); // for builtin firmware

but if builtin firmware is not page alignment, page pointer for builtin
firmware is wrong, it contains memory not belong to firmware. drivers
has to use additional code to handle this.

Quoted 4 lines

if builtin firmware is also page alignment, no need additional code
to
handle builtin firmware. simplified.

Quoted 4 lines

But you did not change anything like this in your code, so why would
I
know this?

I understand it is very difficult to review this patch without context.
The driver is not opensource, I can't show the patch for driver.

Quoted 7 lines
Quoted 5 lines

Then I can not accept your patch.  Go talk to your corporate lawyers about changing core kernel code for a closed source driver and what that
implies about that closed driver (hint, your driver can not be closed...) :)

It's very sad,

"sad"? Again, please go discuss this with your corporate lawyers before thinking that this is even something that is possible to do. Hint, if I _were_ to accept this, they would be _VERY_ upset at both me, and you.

I am trying to _save_ you problems, please realize this.

greg k-h