Cortex-m MPU subregions for >64KB apps by tyler-potyondy · Pull Request #103 · tock/elf2tab

tyler-potyondy · 2026-01-15T21:39:58Z

Rounding to the nearest power of two for cortex-m apps results in a high amount of flash wasting for large applications. This adds logic to take advantage of MPU subregions for apps larger than 64kB.

Example

For an ~80kB application the tab was padded to 128kB before these changes.

Application size report for arch family cortex-m:
   text	   data	    bss	    dec	    hex	filename
  81564	   5152	   6168	  92884	  16ad4	build/cortex-m0/cortex-m0.elf
  76440	   5128	   6168	  87736	  156b8	build/cortex-m3/cortex-m3.elf
  76608	   5128	   6168	  87904	  15760	build/cortex-m4/cortex-m4.elf
  76656	   5128	   6168	  87952	  15790	build/cortex-m7/cortex-m7.elf
 311268	  20536	  24672	 356476	  5707c	(TOTALS)

Before Changes

cortex-m4:
  TBF version           : 2                       [0x0  ]
  header_size           :         76         0x4c
  total_size            :     131072      0x20000 <==== TOTAL APP SIZE IN FLASH
  checksum              :              0x6148427d
  flags                 :          1          0x1
    enabled             : Yes
    sticky              : No
  TLV: Main (1)                                   [0x10 ]
    init_fn_offset      :         49         0x31
    protected_size      :          0          0x0
    minimum_ram_size    :      16436       0x4034
  TLV: Program (9)                                [0x20 ]
    init_fn_offset      :         49         0x31
    protected_size      :          0          0x0
    minimum_ram_size    :      16436       0x4034
    binary_end_offset   :      85092      0x14c64
    app_version         :          0          0x0
  TLV: Package Name (3)                           [0x38 ]
    package_name        : lorawan
  TLV: Kernel Version (8)                         [0x44 ]
    kernel_major        : 2
    kernel_minor        : 0
    kernel version      : ^2.0

TBF Footers
  Footer
    footer_size         :      45980       0xb39c
  Footer TLV: Credentials (128)                   [0x0  ]
    Type: Reserved (0)
    Length: 45972 <========= PADDING ADDED

After Changes

cortex-m4:
  TBF version           : 2                       [0x0  ]
  header_size           :         76         0x4c
  total_size            :      98304      0x18000 <===== TOTAL APP SIZE IN FLASH
  checksum              :              0x614bc30d
  flags                 :          1          0x1
    enabled             : Yes
    sticky              : No
  TLV: Main (1)                                   [0x10 ]
    init_fn_offset      :         49         0x31
    protected_size      :          0          0x0
    minimum_ram_size    :      13344       0x3420
  TLV: Program (9)                                [0x20 ]
    init_fn_offset      :         49         0x31
    protected_size      :          0          0x0
    minimum_ram_size    :      13344       0x3420
    binary_end_offset   :      85268      0x14d14
    app_version         :          0          0x0
  TLV: Package Name (3)                           [0x38 ]
    package_name        : lorawan
  TLV: Kernel Version (8)                         [0x44 ]
    kernel_major        : 2
    kernel_minor        : 0
    kernel version      : ^2.0

TBF Footers
  Footer
    footer_size         :      13036       0x32ec
  Footer TLV: Credentials (128)                   [0x0  ]
    Type: Reserved (0)
    Length: 13028 <============== PADDING ADDED

bradjc · 2026-01-16T20:43:09Z

+        /// To prevent wasted space, only pad to:
+        /// (X - 1) * (TotalSizePowerOfTwo / 8) < size < X * (TotalSizePowerOfTwo)
+        MaskedPowerOfTwo,


I'm not sure I understand this.

The version below makes sense to me, but multiplying TotalSizePowerOfTwo by X doesn't.

Suggested change

/// To prevent wasted space, only pad to:

/// (X - 1) * (TotalSizePowerOfTwo / 8) < size < X * (TotalSizePowerOfTwo)

MaskedPowerOfTwo,

/// To prevent wasted space, only pad to the smallest integer `X` where

/// `X * (TotalSizePowerOfTwo / 8) > size`.

MaskedPowerOfTwo,

This was a mistake. It should be:

(X - 1) * (TotalSizePowerOfTwo / 8) < size < X * (TotalSizePowerOfTwo / 8)

bradjc · 2026-01-16T20:44:36Z

+    /// Specify how elf2tab should add trailing padding to the end of the TBF
+    /// file.
+    enum TrailingPadding {


Why did this move in the file?

I moved it so all the comments are grouped to where the math is happening for this. With it being an enum def it makes sense for this to be where it was. I can revert this.

Looking closer, the enum doesn't need to move but the let trailing_padding ... block needs to move lower since we need to know the binary size (to decide which padding to do based on the size).

bradjc · 2026-01-16T20:49:10Z

+                // Pad to the nearest multiple of (power of two / 8).
+                if target_size > binary_index {
+                    target_size - binary_index
+                } else {
+                    0
+                }


Suggested change

// Pad to the nearest multiple of (power of two / 8).

if target_size > binary_index {

target_size - binary_index

} else {

0

}

// Pad to the nearest multiple of (power of two / 8).

target_size - binary_index

Takes advantage of subregions to lower the wasted space.

lschuermann · 2026-01-22T13:50:06Z

+        /// To prevent wasted space, only pad to:
+        /// (X - 1) * (TotalSizePowerOfTwo / 8) < size < X * (TotalSizePowerOfTwo / 8)
+        MaskedPowerOfTwo,


Is this supported for an unmodified, old kernel that doesn't know about these elf2tab changes yet?

I think it should be. The only issue I forsee would be if the kernel can't assign an MPU config for this app size (in which case the kernel wouldn't load the app).

With that being said, the MaskedPowerOfTwo should always be able to be assigned an MPU config since it is just using subregions to mask out some of the wasted power of two region size.

The only issue I forsee would be if the kernel can't assign an MPU config for this app size (in which case the kernel wouldn't load the app).

That's exactly what I'm concerned about.

With that being said, the MaskedPowerOfTwo should always be able to be assigned an MPU config since it is just using subregions to mask out some of the wasted power of two region size.

Got it. I'm not familiar with the Cortex-M MPU implementation, and so I wasn't sure whether that needs adjustment to support this new MaskedPowerOfTwo mode. If that works with the one that's upstream already (and hence won't break backwards-compatibility of elf2tab), then I'm conceptually on board with this change.

ZhekaS · 2026-02-26T15:39:15Z

Looks like this PR is crossing with mine #102
Wouldn't the proposed solution work for this issue as well?

tyler-potyondy · 2026-04-01T23:48:49Z

Just following up on this...

I agree with @ZhekaS that this has some overlap with #102. It seems #102 could also be used to fix this problem.

The padding/alignment to powers of 2 has been particularly wasteful of space on cortex-m so it would be nice to have a default solution that helps in the common case without needing to pass another flag.

tyler-potyondy force-pushed the near-power-of-two branch 2 times, most recently from 9f5b5ac to ef36f4f Compare January 15, 2026 21:41

bradjc reviewed Jan 16, 2026

View reviewed changes

Add new power-of-two masking logic for >64KB apps

54ff1d6

Takes advantage of subregions to lower the wasted space.

tyler-potyondy force-pushed the near-power-of-two branch from ef36f4f to 54ff1d6 Compare January 16, 2026 22:15

bradjc approved these changes Jan 21, 2026

View reviewed changes

lschuermann reviewed Jan 22, 2026

View reviewed changes

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Cortex-m MPU subregions for >64KB apps#103

Cortex-m MPU subregions for >64KB apps#103
tyler-potyondy wants to merge 1 commit intotock:masterfrom
tyler-potyondy:near-power-of-two

tyler-potyondy commented Jan 15, 2026

Uh oh!

bradjc Jan 16, 2026

Uh oh!

tyler-potyondy Jan 16, 2026

Uh oh!

bradjc Jan 16, 2026

Uh oh!

tyler-potyondy Jan 16, 2026

Uh oh!

tyler-potyondy Jan 16, 2026

Uh oh!

Uh oh!

bradjc Jan 16, 2026

Uh oh!

lschuermann Jan 22, 2026

Uh oh!

tyler-potyondy Jan 22, 2026

Uh oh!

lschuermann Jan 23, 2026

Uh oh!

ZhekaS commented Feb 26, 2026

Uh oh!

tyler-potyondy commented Apr 1, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

4 participants

Conversation

tyler-potyondy commented Jan 15, 2026

Example

Before Changes

After Changes

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

ZhekaS commented Feb 26, 2026

Uh oh!

tyler-potyondy commented Apr 1, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

4 participants