[RFC PATCH v2 0/2] staging: media: atomisp: Refactor bit logic helpers in vmem.c

[RFC PATCH v2 0/2] staging: media: atomisp: Refactor bit logic helpers in vmem.c

[Adrian Barnaś]

Refactor proposition for bit operation in vmem.c.
* Previous name for function "inv_subword()" for me is not telling what
function acctualy does - it clears bit specified by subword, so renamed
to clear_subword()
* Added a helper to create a proper bitmask for a subword, without using
GENMASK(end-1, start) which was claimed to be unsafe
* Simplified subword() and clear_subword() to be more readable.

Continuation of https://lore.kernel.org/linux-staging/20250902073841.2338568-1-abarnas@google.com/ 

Adrian Barnaś (2):
  staging: media: atomisp: Change name to better follow its behavior
  staging: media: atomisp: Simplify logic in vmem.c

 .../pci/hive_isp_css_common/host/vmem.c       | 19 ++++++++++++-------
 1 file changed, 12 insertions(+), 7 deletions(-)

-- 
2.51.0.355.g5224444f11-goog

[RFC PATCH v2 1/2] staging: media: atomisp: Change name to better follow its behavior

[Adrian Barnaś]

Change name to clearly states what function does.
Signed-off-by: Adrian Barnaś <abarnas@google.com>
---
 .../atomisp/pci/hive_isp_css_common/host/vmem.c  | 16 ++++++++--------
 1 file changed, 8 insertions(+), 8 deletions(-)

diff --git a/drivers/staging/media/atomisp/pci/hive_isp_css_common/host/vmem.c b/drivers/staging/media/atomisp/pci/hive_isp_css_common/host/vmem.c
index 547cc480c105..a3fe03216389 100644
--- a/drivers/staging/media/atomisp/pci/hive_isp_css_common/host/vmem.c
+++ b/drivers/staging/media/atomisp/pci/hive_isp_css_common/host/vmem.c
@@ -22,9 +22,9 @@ subword(unsigned long long w, unsigned int start, unsigned int end)
 	return (w & (((1ULL << (end - 1)) - 1) << 1 | 1)) >> start;
 }
 
-/* inverse subword bits move like this: MSB[xxxx____xxxx]LSB -> MSB[xxxx0000xxxx]LSB */
+/* clears subword bits like this: MSB[xxxx____xxxx]LSB -> MSB[xxxx0000xxxx]LSB */
 static inline unsigned long long
-inv_subword(unsigned long long w, unsigned int start, unsigned int end)
+clear_subword(unsigned long long w, unsigned int start, unsigned int end)
 {
 	return w & (~(((1ULL << (end - 1)) - 1) << 1 | 1) | ((1ULL << start) - 1));
 }
@@ -45,16 +45,16 @@ static void move_subword(unsigned long long *target, unsigned int target_bit,
 
 	if (subword_width + start_bit > uedge_bits) { /* overlap */
 		unsigned long long old_val1;
-		unsigned long long old_val0 = inv_subword(target[start_elem],
-							  start_bit, uedge_bits);
+		unsigned long long old_val0 = clear_subword(target[start_elem],
+							    start_bit, uedge_bits);
 
 		target[start_elem] = old_val0 | (src_subword << start_bit);
-		old_val1 = inv_subword(target[start_elem + 1], 0,
-				       subword_width + start_bit - uedge_bits);
+		old_val1 = clear_subword(target[start_elem + 1], 0,
+					 subword_width + start_bit - uedge_bits);
 		target[start_elem + 1] = old_val1 | (src_subword >> (uedge_bits - start_bit));
 	} else {
-		unsigned long long old_val = inv_subword(target[start_elem], start_bit,
-							 start_bit + subword_width);
+		unsigned long long old_val = clear_subword(target[start_elem], start_bit,
+							   start_bit + subword_width);
 
 		target[start_elem] = old_val | (src_subword << start_bit);
 	}
-- 
2.51.0.355.g5224444f11-goog

[RFC PATCH v2 2/2] staging: media: atomisp: Simplify logic in vmem.c

[Adrian Barnaś]

Add inline helper and simplify logic for subword operations in
pci/hive_isp_css_common/host/vmem.c.

Signed-off-by: Adrian Barnaś <abarnas@google.com>
---
 .../media/atomisp/pci/hive_isp_css_common/host/vmem.c    | 9 +++++++--
 1 file changed, 7 insertions(+), 2 deletions(-)

diff --git a/drivers/staging/media/atomisp/pci/hive_isp_css_common/host/vmem.c b/drivers/staging/media/atomisp/pci/hive_isp_css_common/host/vmem.c
index a3fe03216389..f11b0448ed83 100644
--- a/drivers/staging/media/atomisp/pci/hive_isp_css_common/host/vmem.c
+++ b/drivers/staging/media/atomisp/pci/hive_isp_css_common/host/vmem.c
@@ -15,18 +15,23 @@
 
 /* Copied from SDK: sim_semantics.c */
 
+static inline unsigned long long subword_bitmask(unsigned int start, unsigned int end)
+{
+	return GENMASK_ULL(end - 1, 0) & ~(GENMASK_ULL(start, 0) >> 1);
+}
+
 /* subword bits move like this:         MSB[____xxxx____]LSB -> MSB[00000000xxxx]LSB */
 static inline unsigned long long
 subword(unsigned long long w, unsigned int start, unsigned int end)
 {
-	return (w & (((1ULL << (end - 1)) - 1) << 1 | 1)) >> start;
+	return (w & GENMASK_ULL(end - 1, 0)) >> start;
 }
 
 /* clears subword bits like this: MSB[xxxx____xxxx]LSB -> MSB[xxxx0000xxxx]LSB */
 static inline unsigned long long
 clear_subword(unsigned long long w, unsigned int start, unsigned int end)
 {
-	return w & (~(((1ULL << (end - 1)) - 1) << 1 | 1) | ((1ULL << start) - 1));
+	return w & ~subword_bitmask(start, end);
 }
 
 #define uedge_bits (8 * sizeof(unsigned long long))
-- 
2.51.0.355.g5224444f11-goog

Re: [RFC PATCH v2 0/2] staging: media: atomisp: Refactor bit logic helpers in vmem.c

[Andy Shevchenko]

On Wed, Sep 03, 2025 at 09:27:52AM +0000, Adrian Barnaś wrote:
> Refactor proposition for bit operation in vmem.c.
> * Previous name for function "inv_subword()" for me is not telling what
> function acctualy does - it clears bit specified by subword, so renamed
> to clear_subword()
> * Added a helper to create a proper bitmask for a subword, without using
> GENMASK(end-1, start) which was claimed to be unsafe
> * Simplified subword() and clear_subword() to be more readable.
> 
> Continuation of https://lore.kernel.org/linux-staging/20250902073841.2338568-1-abarnas@google.com/ 

Thanks for the effort, but I think it's just a tip of the iceberg.
What we really need is to completely rewrite hive_sim_wide_unpack()
and hive_sim_wide_pack(). If we want to preserve arbitrary bit numbers
we ought to use bitmap API rather than that custom approach.

-- 
With Best Regards,
Andy Shevchenko