[DeathCamel58]

Follow-up to #268, from the testing in #266.

rounded_num() counts the halvings it takes to reach zero and returns 1 << count, so an exact power of two comes back doubled: rounded_num(512) is 1024. A kernel MemTotal is never a power of two, so this never showed. Since #268, though, the Ambarella total is the top of the DSP carve-out, and on the S3L camera in #266 that is exactly 512 MiB (0x07800000 + 0x18800000). ambarella_totalmem() returns the right 524288 KiB; the doubling happens in ram.c's rounding.

The fix returns n unchanged when it is already a power of two. backup.c rounds its total with the same function and gets the same fix.

Testing

• ambarella_test gains checks for the 512M case, a 114940 kB MemTotal (112 → 128), 513 → 1024 and 1 → 1. Without the fix the 512M and 1 cases fail; with it all pass.
• On the camera from SoC not recognised: Ambarella S3L (Amcrest IP4M-1026E / Dahua IPC-HFW1430SN) #266 (Amcrest IP4M-1026E, Ambarella S3L, 512 MiB): arm32 built with the CI toolchain (toolchain.hisilicon-hi3516cv100, musl) and packed with UPX 4.2.3 as in release.yml:

	latest (7c51a7e)	this PR
Stock Dahua 3.10.73 (IAVMEM from the boot log)	total: 1024M, media: 392M	total: 512M, media: 392M
OpenIPC 3.10.104 (/proc/device-tree/iavmem/reg)	total: 1024M, media: 392M	total: 512M, media: 392M

[qodo-free-for-open-source-projects]

PR Summary by Qodo

Preserve RAM totals that are already powers of two

🐞 Bug fix 🧪 Tests 🕐 Less than 10 minutes

AI Description

• Keep exact power-of-two RAM totals unchanged instead of doubling them.
• Correct the displayed total for Ambarella S3L cameras with 512 MiB of RAM.
• Add regression checks for exact powers and totals that still need rounding.

Diagram

graph TD
  H["HAL total memory"] --> M["MiB conversion"] --> D{"Power of two?"}
  D -->|yes| K["Keep total"] --> O["RAM and backup"]
  D -->|no| R["Round upward"] --> O

Loading

High-Level Assessment

Keep the fix in the shared rounding helper: it corrects both RAM reporting and backup generation without changing Ambarella memory detection or duplicating logic at the call sites.

Files changed (2) +11 / -0

Bug fix (1) +3 / -0

common.cPreserve power-of-two inputs in rounded_num +3/-0

Preserve power-of-two inputs in rounded_num

• Returns a nonzero input unchanged when it is already a power of two. Other inputs retain the existing rounding behavior, correcting both RAM and backup callers.

src/hal/common.c

Tests (1) +8 / -0

ambarella_test.cCover exact and non-exact RAM rounding +8/-0

Cover exact and non-exact RAM rounding

• Adds checks for the S3L camera’s 512 MiB total and for 1 MiB, both of which must remain unchanged. Also verifies that 112 and 513 MiB still round up.

src/hal/ambarella_test.c

[qodo-free-for-open-source-projects]

Code Review by Qodo

🐞 Bugs (0) 📘 Rule violations (0) 📎 Requirement gaps (0)

Great, no issues found!

Qodo reviewed your code and found no material issues that require review

Tip of the day

💡 Did you know, you can switch off images and animations for a plain-text comment

More tips ↗ | Customize Qodo ↗ | Qodo docs ↗

[openipc-ai]

This is correct, thanks for finding it. I wrote in #268 that this camera should report 512M without checking the rounding in ram.c, and I'm glad your test caught it.

Verified at this commit:

• rounded_num() has two callers, ram.c and backup.c.
• Across 0 to 4999, the only values that change are exact powers of two; everything else rounds as before. 0 still gives 1.
• ambarella_test, anyka_test, longse_test, reginfo_test, sensors_test and hal_hisi_test all pass.
• With the fix removed, the 512M and 1 checks fail, as you describe.

Behaviour change to be aware of: this isn't Ambarella-only. Any HAL whose total lands exactly on a power of two was doubled too, and now reports the true figure. For backup.c, that is the value it writes when setTotalMem is set. Both are the correct result.