Skip to contentSkip to conversation

drm/komeda: use bitmap API to convert U32 to bitmap

PATCH In review Yury Norov 4 participants 7 messages Dec 28, 2020 – Dec 29, 2020

Subsystems (4)
lore
Share & tools
lore

Yury Norov

author
#0 lore raw

PATCH 1/1 drm/komeda: use bitmap API to convert U32 to bitmap

The commit be3e477effba636ad25 ("drm/komeda: Fix bit check to import to value of proper type") fixes possible out-of-bound issue related to find_first_bit() usage, but does not address the endianness problem.

We can use bitmap_from_arr32() here.

Since I have no hardware, the patch is compile-testes only. Carsten, could you please test it and consider including into your tree?

---
 drivers/gpu/drm/arm/display/komeda/komeda_pipeline.c | 5 +++--
 1 file changed, 3 insertions(+), 2 deletions(-)
drivers/gpu/drm/arm/display/komeda/komeda_pipeline.c +3 -2
index 719a79728e24..27968215e41d 100644
@@ -136,11 +136,12 @@ struct komeda_component *
 komeda_pipeline_get_first_component(struct komeda_pipeline *pipe,
 				    u32 comp_mask)
 {
+	DECLARE_BITMAP(comp_mask_local, 32);
 	struct komeda_component *c = NULL;
-	unsigned long comp_mask_local = (unsigned long)comp_mask;
 	int id;
 
-	id = find_first_bit(&comp_mask_local, 32);
+	bitmap_from_arr32(comp_mask_local, &comp_mask, 32);
+	id = find_first_bit(comp_mask_local, 32);
 	if (id < 32)
 		c = komeda_pipeline_get_component(pipe, id);
 
Signature · 1 line
2.25.1
#1 lore raw
Quoted 4 lines · Yury Norov

The commit be3e477effba636ad25 ("drm/komeda: Fix bit check to import to value of proper type") fixes possible out-of-bound issue related to find_first_bit() usage, but does not address the endianness problem.

Hmm... Can you elaborate?

...

drivers/gpu/drm/arm/display/komeda/komeda_pipeline.c struct komeda_component *

Quoted 3 lines
drivers/gpu/drm/arm/display/komeda/komeda_pipeline.c +0 -1
 				    u32 comp_mask)
 
-	unsigned long comp_mask_local = (unsigned long)comp_mask;

Here we convert u32 to unsigned long (LSB is kept LSB since it happens in native endianess).

drivers/gpu/drm/arm/display/komeda/komeda_pipeline.c struct komeda_component * · line 143

Quoted 1 line
drivers/gpu/drm/arm/display/komeda/komeda_pipeline.c +0 -1
-	id = find_first_bit(&comp_mask_local, 32);

Here it takes an address to unsigned long and tries only lower 32 bits.

Are you telling that find_first_bit() has an issue?

Signature · 2 lines
With Best Regards,
Andy Shevchenko
author
#2 lore raw
Quoted 15 lines · Andy Shevchenko
Quoted 4 lines · Yury Norov

The commit be3e477effba636ad25 ("drm/komeda: Fix bit check to import to value of proper type") fixes possible out-of-bound issue related to find_first_bit() usage, but does not address the endianness problem.

Hmm... Can you elaborate?

...

Quoted 3 lines
                                  u32 comp_mask)

- unsigned long comp_mask_local = (unsigned long)comp_mask;

Quoted 8 lines

Here we convert u32 to unsigned long (LSB is kept LSB since it happens in native endianess).

Quoted 1 line

- id = find_first_bit(&comp_mask_local, 32);

Here it takes an address to unsigned long and tries only lower 32 bits.

Are you telling that find_first_bit() has an issue?

It seems you're right, there's no issue with endianness in existing code. In fact, the line

Quoted 1 line
Quoted 1 line

- unsigned long comp_mask_local = (unsigned long)comp_mask;

is an opencoded version of bitmap_from_arr32(dst, src, 32).

Maybe it would be better to use the bitmap API here, but existing code is correct. Sorry for the noise.

Yury

#3 lore raw
Quoted 16 lines · Yury Norov
Quoted 14 lines · Andy Shevchenko
Quoted 4 lines · Yury Norov

The commit be3e477effba636ad25 ("drm/komeda: Fix bit check to import to value of proper type") fixes possible out-of-bound issue related to find_first_bit() usage, but does not address the endianness problem.

Hmm... Can you elaborate?

...

Quoted 2 lines
                                   u32 comp_mask)

- unsigned long comp_mask_local = (unsigned long)comp_mask;

Here we convert u32 to unsigned long (LSB is kept LSB since it happens in native endianess).

Quoted 6 lines
Quoted 4 lines
Quoted 1 line

- id = find_first_bit(&comp_mask_local, 32);

Here it takes an address to unsigned long and tries only lower 32 bits.

Are you telling that find_first_bit() has an issue?

It seems you're right, there's no issue with endianness in existing code. In fact, the line

Indeed Andy covered this. Take LSB32 with the cast to "local on-stack long" and possible extend upper 32bits with 0's if needed (64bit longs).

Quoted 5 lines
Quoted 1 line
Quoted 1 line

- unsigned long comp_mask_local = (unsigned long)comp_mask;

is an opencoded version of bitmap_from_arr32(dst, src, 32).

Maybe it would be better to use the bitmap API here, but existing code is correct. Sorry for the noise.

While your code is seemingly also valid (I can check to be sure with KASAN if you want still), it does seem a little less "nice to read" with more lines of code for the same work. Is it worth making the code a little longer here as it's not actually fixing anything to do it this other way? DECLARE_BITMAP() is a bit of an obscure way to declare a single unsigned long (in this case) where the compiler does the right thing anyway with a simple assign+cast making it easier to read/follow IMHO.

IMPORTANT NOTICE: The contents of this email and any attachments are confidential and may also be privileged. If you are not the intended recipient, please notify the sender immediately and do not disclose the contents to any other person, use it for any purpose, or store or copy the information in any medium. Thank you.

#4 lore raw
Quoted 33 lines · Carsten Haitzler
Quoted 16 lines · Yury Norov
Quoted 14 lines · Andy Shevchenko
Quoted 4 lines · Yury Norov

The commit be3e477effba636ad25 ("drm/komeda: Fix bit check to import to value of proper type") fixes possible out-of-bound issue related to find_first_bit() usage, but does not address the endianness problem.

Hmm... Can you elaborate?

...

Quoted 2 lines
                                   u32 comp_mask)

- unsigned long comp_mask_local = (unsigned long)comp_mask;

Here we convert u32 to unsigned long (LSB is kept LSB since it happens in native endianess).

Quoted 6 lines
Quoted 4 lines
Quoted 1 line

- id = find_first_bit(&comp_mask_local, 32);

Here it takes an address to unsigned long and tries only lower 32 bits.

Are you telling that find_first_bit() has an issue?

It seems you're right, there's no issue with endianness in existing code. In fact, the line

Indeed Andy covered this. Take LSB32 with the cast to "local on-stack long" and possible extend upper 32bits with 0's if needed (64bit longs).

Quoted 5 lines
Quoted 1 line
Quoted 1 line

- unsigned long comp_mask_local = (unsigned long)comp_mask;

is an opencoded version of bitmap_from_arr32(dst, src, 32).

Maybe it would be better to use the bitmap API here, but existing code is correct. Sorry for the noise.

Quoted 7 lines

While your code is seemingly also valid (I can check to be sure with KASAN if you want still), it does seem a little less "nice to read" with more lines of code for the same work. Is it worth making the code a little longer here as it's not actually fixing anything to do it this other way? DECLARE_BITMAP() is a bit of an obscure way to declare a single unsigned long (in this case) where the compiler does the right thing anyway with a simple assign+cast making it easier to read/follow IMHO.

What we can do is to declare BITS_PER_U32.
Also we can pay attention on these definitions:
https://elixir.bootlin.com/linux/latest/source/kernel/bpf/btf.c#L168
https://elixir.bootlin.com/linux/latest/source/security/selinux/ss/ebitmap.c#L27

And define BITMAP_FROM_U32() macro.

Then It would be written like

DECLARE_BITMAP(comp_mask_local, BITS_PER_U32) = BITMAP_FROM_U32(comp_mask);

But this is a bit verbose.

Also, it can be something like DECLARE_BITMAP_U32(...) = BITMAP_FROM_U32(...);

Quoted 1 line

IMPORTANT NOTICE: The contents of this email and any attachments are confidential and may also be privileged. If you are not the intended recipient, please notify the sender immediately and do not disclose the contents to any other person, use it for any purpose, or store or copy the information in any medium. Thank you.

Hmm... you probably have to get rid of this footer.

Signature · 2 lines
With Best Regards,
Andy Shevchenko
author
#5 lore raw
Quoted 67 lines · Andy Shevchenko
Quoted 33 lines · Carsten Haitzler
Quoted 16 lines · Yury Norov
Quoted 14 lines · Andy Shevchenko
Quoted 4 lines · Yury Norov

The commit be3e477effba636ad25 ("drm/komeda: Fix bit check to import to value of proper type") fixes possible out-of-bound issue related to find_first_bit() usage, but does not address the endianness problem.

Hmm... Can you elaborate?

...

Quoted 2 lines
                                   u32 comp_mask)

- unsigned long comp_mask_local = (unsigned long)comp_mask;

Here we convert u32 to unsigned long (LSB is kept LSB since it happens in native endianess).

Quoted 6 lines
Quoted 4 lines
Quoted 1 line

- id = find_first_bit(&comp_mask_local, 32);

Here it takes an address to unsigned long and tries only lower 32 bits.

Are you telling that find_first_bit() has an issue?

It seems you're right, there's no issue with endianness in existing code. In fact, the line

Indeed Andy covered this. Take LSB32 with the cast to "local on-stack long" and possible extend upper 32bits with 0's if needed (64bit longs).

Quoted 5 lines
Quoted 1 line
Quoted 1 line

- unsigned long comp_mask_local = (unsigned long)comp_mask;

is an opencoded version of bitmap_from_arr32(dst, src, 32).

Maybe it would be better to use the bitmap API here, but existing code is correct. Sorry for the noise.

Quoted 7 lines

While your code is seemingly also valid (I can check to be sure with KASAN if you want still), it does seem a little less "nice to read" with more lines of code for the same work. Is it worth making the code a little longer here as it's not actually fixing anything to do it this other way? DECLARE_BITMAP() is a bit of an obscure way to declare a single unsigned long (in this case) where the compiler does the right thing anyway with a simple assign+cast making it easier to read/follow IMHO.

What we can do is to declare BITS_PER_U32.
Also we can pay attention on these definitions:
https://elixir.bootlin.com/linux/latest/source/kernel/bpf/btf.c#L168
https://elixir.bootlin.com/linux/latest/source/security/selinux/ss/ebitmap.c#L27

And define BITMAP_FROM_U32() macro.

Then It would be written like

DECLARE_BITMAP(comp_mask_local, BITS_PER_U32) = BITMAP_FROM_U32(comp_mask);

But this is a bit verbose.

Also, it can be something like DECLARE_BITMAP_U32(...) = BITMAP_FROM_U32(...);

Quoted 1 line

IMPORTANT NOTICE: The contents of this email and any attachments are confidential and may also be privileged. If you are not the intended recipient, please notify the sender immediately and do not disclose the contents to any other person, use it for any purpose, or store or copy the information in any medium. Thank you.

Hmm... you probably have to get rid of this footer.

--
With Best Regards,
Andy Shevchenko

author
#6 lore raw
Quoted 47 lines · Yury Norov
Quoted 44 lines · Andy Shevchenko
Quoted 40 lines · Carsten Haitzler
Quoted 16 lines · Yury Norov
Quoted 14 lines · Andy Shevchenko
Quoted 4 lines · Yury Norov

The commit be3e477effba636ad25 ("drm/komeda: Fix bit check to import to value of proper type") fixes possible out-of-bound issue related to find_first_bit() usage, but does not address the endianness problem.

Hmm... Can you elaborate?

...

Quoted 2 lines
                                   u32 comp_mask)

- unsigned long comp_mask_local = (unsigned long)comp_mask;

Here we convert u32 to unsigned long (LSB is kept LSB since it happens in native endianess).

Quoted 6 lines
Quoted 4 lines
Quoted 1 line

- id = find_first_bit(&comp_mask_local, 32);

Here it takes an address to unsigned long and tries only lower 32 bits.

Are you telling that find_first_bit() has an issue?

It seems you're right, there's no issue with endianness in existing code. In fact, the line

Indeed Andy covered this. Take LSB32 with the cast to "local on-stack long" and possible extend upper 32bits with 0's if needed (64bit longs).

Quoted 5 lines
Quoted 1 line
Quoted 1 line

- unsigned long comp_mask_local = (unsigned long)comp_mask;

is an opencoded version of bitmap_from_arr32(dst, src, 32).

Maybe it would be better to use the bitmap API here, but existing code is correct. Sorry for the noise.

While your code is seemingly also valid (I can check to be sure with KASAN if you want still), it does seem a little less "nice to read" with more lines of code for the same work. Is it worth making the code a little longer here as it's not actually fixing anything to do it this other way? DECLARE_BITMAP() is a bit of an obscure way to declare a single unsigned long (in this case) where the compiler does the right thing anyway with a simple assign+cast making it easier to read/follow IMHO.

Quoted 4 lines
Quoted 4 lines

What we can do is to declare BITS_PER_U32.
Also we can pay attention on these definitions:
https://elixir.bootlin.com/linux/latest/source/kernel/bpf/btf.c#L168
https://elixir.bootlin.com/linux/latest/source/security/selinux/ss/ebitmap.c#L27

On the other hand, fixed width types are designed especially to contain a specific
number of bits. I don't understand why BITS_PER_U32 is any better than just 32 ...

Quoted 9 lines
Quoted 9 lines

And define BITMAP_FROM_U32() macro.

Then It would be written like

DECLARE_BITMAP(comp_mask_local, BITS_PER_U32) = BITMAP_FROM_U32(comp_mask);

But this is a bit verbose.

Also, it can be something like DECLARE_BITMAP_U32(...) = BITMAP_FROM_U32(...);

People often do things like this:
https://elixir.bootlin.com/linux/latest/source/drivers/pci/controller/pcie-hisi-error.c#L199

Maybe it's worth adding a shorthand for it, like CREATE_BITMAP32(name, val) and CREATE_BITMAP64(name, val)?