llext: Full ARM ELF relocation support #70452

selescop · 2024-03-19T16:49:36Z

Adds support for all relocation types produced by GCC on ARM platform using partial linking (-r flag) or shared link (-fpic and -shared flag).

Subset of LLEXT full ARM relocation support (adding executable) PR#70171

pillo79

Some minor nits and one big caveat for which I'm hesitant to approve this PR, even though it LGTM from the technical side: I honestly do not know what other rewrites will be needed to avoid future licensing issues.

arch/arm/core/elf.c

include/zephyr/llext/llext.h

subsys/llext/llext.c

arch/arm/core/elf.c

include/zephyr/llext/elf.h

arch/arm/core/elf.c

tests/subsys/llext/simple/src/movwmovt_ext.c

include/zephyr/llext/llext.h

selescop · 2024-03-21T13:34:04Z

Some minor nits and one big caveat for which I'm hesitant to approve this PR, even though it LGTM from the technical side: I honestly do not know what other rewrites will be needed to avoid future licensing issues.

Don't want to cause trouble here with the licensing issue. Tried to rewrite it the Zephyr way as @bjarki-trackunit asked but might not be sufficient enough. Can we ask for a second opinion here? Maybe @nashif?

bjarki-andreasen · 2024-03-21T14:32:03Z

Some minor nits and one big caveat for which I'm hesitant to approve this PR, even though it LGTM from the technical side: I honestly do not know what other rewrites will be needed to avoid future licensing issues.

Don't want to cause trouble here with the licensing issue. Tried to rewrite it the Zephyr way as @bjarki-trackunit asked but might not be sufficient enough. Can we ask for a second opinion here? Maybe @nashif?

I think we should rewrite all the relocations, in zephyr style, using bitfields and uints instead of manually doing bitshifting and handling endianess.

I have rewritten the relocations properly for the ARM_THM_CALL reloc like this


struct arm_bl {
	/* Upper */
	uint16_t imm10h : 10;
	uint16_t s : 1;
	uint16_t const0 : 5;
	/* Lower */
	uint16_t imm11l : 11;
	uint16_t j2 : 1;
	uint16_t const1 : 1;
	uint16_t j1 : 1;
	uint16_t const2 : 2;
};

/*
 * Addend shifted down 1 bit for BL
 * Addend shifted down 2 bits for BLX
 * See bit 12 of BL/BLX lower half-word (const1 in struct arm_bl)
 */
struct arm_bl_addend {
	uint32_t imm11l : 11;
	uint32_t imm10h : 10;
	uint32_t i2 : 1;
	uint32_t i1 : 1;
	uint32_t s : 1;
	uint32_t const0 : 8;
};

static int32_t arm_bl_decode_addend(uintptr_t addr)
{
	const struct arm_bl *bl = (struct arm_bl *)addr;
	int32_t addend = bl->s ? UINT32_MAX : 0;
	struct arm_bl_addend *decoded = (struct arm_bl_addend *)&addend;

	decoded->i1 = ~(bl->j1 ^ bl->s);
	decoded->i2 = ~(bl->j2 ^ bl->s);
	decoded->imm10h = bl->imm10h;
	decoded->imm11l = bl->imm11l;

	/* Shift up addend to aling with unit, returned addend is in bytes */
	return bl->const1 ? addend << 1 : addend << 2;
}

static void arm_bl_encode_addend(uintptr_t addr, int32_t addend)
{
	struct arm_bl *bl = (struct arm_bl *)addr;
	const struct arm_bl_addend *decoded = (struct arm_bl_addend *)&addend;

	/* Shift down addend to align with struct arm_bl_addend encoding */
	addend = bl->const1 ? addend >> 1 : addend >> 2;

	/* Encode addend */
	bl->s = decoded->s;
	bl->imm10h = decoded->imm10h;
	bl->j1 = decoded->s ^ (!decoded->i1);
	bl->j2 = decoded->s ^ (!decoded->i2);
	bl->imm11l = decoded->imm11l;
}

...
	case R_ARM_THM_CALL:
		int32_t branc_offset = (((int32_t)opval) - ((int32_t)opaddr));

		branc_offset += arm_bl_decode_addend(opaddr);

		arm_bl_encode_addend(opaddr, branc_offset);
		break;

It is way easier to read, way safer, and not copying linux, compared to the rather messy implementation in Linux of the same relocation :)

selescop · 2024-03-21T14:56:58Z

It is way easier to read, way safer, and not copying linux, compared to the rather messy implementation in Linux of the same relocation :)

I've already tried bitfield in structure in a past project and I've learnt that it shall not be used to map hardware. There is no warranty that the bits you defined will actually be set where you think. It is compiler/implementation depend. I remember reading it in GCC documentation but cannot get my hands on it... Here I found something close enough on the web:

In addition to that, C99 §6.7.2.1, paragraph 10 says: "The order of allocation of bit-fields within a unit (high-order to low-order or low-order to high-order) is implementation-defined."

Bit shifting/masking may seems messy but it is the only way to go.

teburd

I still like that we're adding the relocations that have been missing, particularly the PC relative jumps. -mlong-call was never intended to be the end game, the end game is to have normally constructed statically linked elfs load. That means that a jump table (I think arm calls them veneers) almost certainly will be needed. This feels like it gets us one step closer which is great.

Overall I think I'd like to see more tests exercising and showing these relocation at work. I'm happy to see the movw/movt setup. Lets get more of those please. I'd also like to see the code better match the docs in how the operation looks. E.g. for R_ARM_ABS32 the operation is defined in the docs as (S + A) | T

Is there a reason we cannot have effectively that exact operation in C code here, e.g. *op = (sym_addr + addend) | thumb_target;

The closer to the docs the code looks, the easier it is to see that we're doing the right things imho.

We're going to need to really take our time and comb over this I feel like given this is the second round where directly copied comments/code were noted.

arch/arm/core/elf.c

teburd · 2024-03-21T14:23:19Z

arch/arm/core/elf.c

+#define PREL31_LOWER_BOUNDARY ((int32_t)-0x40000000)
+#define THM_JUMP_UPPER_BOUNDARY ((int32_t)0xff000000)
+#define THM_JUMP_LOWER_BOUNDARY ((int32_t)0x01000000)
+#define MASK_V4BX_RM_COND 0xf000000f


Does Zephyr support armv4 at all? Whats the use case for the v4bx relocation?

Well, from what I know, newer ARM architecture are based on the previous ones. So, if Zephyr supports v8, it means that it support v4, 5 ,6 and 7 I guess

arch/arm/core/elf.c

teburd · 2024-03-21T14:29:55Z

arch/arm/core/elf.c

 	case R_ARM_ABS32:
-		/* Add the addend stored at opaddr to opval */
-		opval += *((uint32_t *)opaddr);
+		/* fall-through */


like wise a stack of cases is clear enough to designate a common branch, no need to comment its a fall-through

/* fall-through */ is for newer gcc's feature.
you'll get a warning/error if your 'case' does not have a 'break'. It is meant to force developers to explicit their code.

That's insane, this is used all over

case X: case Y: case Z: useful_work(); break;

Should absolutely not be warned on?

lol, totally agree. I had the same reaction but welcome to GCC 7 (check for fallthrough or comments)
So either you add a new warning suppressor according to GCC version or you already adapt your code

Is that true only if the case has a body but no break?

The docs suggest that to be the case

https://gcc.gnu.org/onlinedocs/gcc/Warning-Options.html#index-Wimplicit-fallthrough_003d

Well, I didn't find it in the doc but it's true that examples always have bodies... need to test

I agree, please do remove the fall through comments where there is no body.

teburd · 2024-03-21T14:30:19Z

arch/arm/core/elf.c

+		*(uint32_t *)loc |= OPCODE2ARMMEM(MASK_V4BX_NOT_RM_COND);
+		break;
+
+	case R_ARM_PREL31:


When is this relocation needed/used? Do we have a test covering it?

In the 5th commit of the PR#70171, there are tests.
I'll add more code to trigger most of the relocations and create a smaller PR when/if the previous ones are merged (otherwise they cannot work)

arch/arm/core/elf.c

teburd · 2024-03-21T14:50:06Z

arch/arm/core/elf.c

-		opval += *((uint32_t *)opaddr);
+		/* fall-through */
+	case R_ARM_TARGET1:
+		*(uint32_t *)loc += sym_base_addr;


Do we need to account for STT_FUNC and Thumb here? The arm docs suggest this should be (S + A) | T

It might also be clearer, at least when comparing with the docs to assign an addend variable here.

e.g. something like...

uint32_t t = (is_func(sym) && is_thumb(loc)) ? 1 : 0; uint32_t addend = *(uint32_t *)loc; *(uint32_t *)loc = (sym_addr + addend) | t;

Much closer matches the arm docs which I think personally would be helpful when looking to see "is this relocation doing what the docs say?"

Update: I get this is what was here before, but since this is rewriting much of it maybe this is the time to make it look like the docs

Here is an answer from my technical expert (what he meant is that is already done, but in several stages):

As in '(S + A) | T', the '| T' does mean that we must set the bit0 if we are in Thumb & a function.
This is done by LLEXT while we are computing 'link_addr' in 'llext_link' which becomes 'sym_addr' in 'arch_elf_relocate()'

else if (ELF_ST_TYPE(sym.st_info) == STT_SECTION ||
ELF_ST_TYPE(sym.st_info) == STT_FUNC ||
ELF_ST_TYPE(sym.st_info) == STT_OBJECT) {
/* Link address is relative to the start of the section */
link_addr = (uintptr_t)ext->mem[ldr->sect_map[sym.st_shndx]]
+ sym.st_value;

The internal extension symbols are handled by the + sym.st_value;
The external ones:

else if (sym.st_shndx == SHN_UNDEF) {
/* If symbol is undefined, then we need to look it up */
link_addr = (uintptr_t)llext_find_sym(NULL, name);

'llext_find_sym()' returns the correct address because it has been register via the EXPORT_SYNBOL macro which actually handles the bit0 in Thumb.
Here is an example with the movwmovt.llext test:

For the FUNC type 'test_func' symbol, 'sym.st_value' equals 1 but 0 for the others. Same for 'test_entry' where 'sym.st_value' is odd.

selescop · 2024-03-21T21:59:40Z

@teburd it might be the good place to ask such question but, is there anyway to ask the CI/CD (twister) to reset its workspace?
Because, when I locally build the test, I do have everything PASSED or SKIPPED but not on the server. I don't know why but it seems that the movwmovt.inc file is not auto-generated on the server, so then it fails to be included...
I cannot figured out what's different between local and server... same repo, same branch, same modules versions (west update), same twister command.... any idea?!

pillo79 · 2024-03-22T08:48:25Z

same repo, same branch, same modules versions (west update), same twister command.... any idea?!

I think the server auto-merges with main before the build. I see the failing tests are llext.simple.loader_build which does not seem to be present in your branch - the merge probably decided to keep your ext code outside the new #ifdef here.

A rebase and manual fix of this merge "non-conflict" should solve this 🙂

selescop · 2024-04-02T15:23:37Z

Some minor nits about data locality and passing by reference when unnecessary, looking pretty good to me.

You're right for 'offset'. I move it locally from the global switch() into each handler functions.
For the 'upper' & 'lower', it is not the same as they are reference in both _decode() and _reloc() functions. They must be the same. I set these 2 functions as 'inline' so the compiler will, hopefully, optimized them and function calls/pointers will no longer be needed it. Is this fine for you?

It might avoid some casting by also doing the initial cast form uintptr_t to uint32_t* in the top level arch_elf_relocate as you know it's always a pointer to a 32bit word here

Well, I'd prefer keep it as uintptr as, as you said, it is a global function. It could used in another architecture where the pointer are 64bit long (cast will be needed)

pillo79

Sorry for the delay and thanks for your dedication. Looks good, I've added a few comments below!

pillo79 · 2024-03-26T14:41:41Z

arch/arm/core/elf.c

 	case R_ARM_ABS32:
-		/* Add the addend stored at opaddr to opval */
-		opval += *((uint32_t *)opaddr);
+		/* fall-through */


I agree, please do remove the fall through comments where there is no body.

arch/arm/core/elf.c

tests/subsys/llext/simple/src/movwmovt_ext.c

pillo79 · 2024-04-03T09:49:41Z

arch/arm/core/elf.c

+		((offset & (MASK_THM_MOV_IMM4<<SHIFT_THM_MOV_IMM4)) >> SHIFT_THM_MOV_IMM4) |
+		((offset & (MASK_THM_MOV_I<<SHIFT_THM_MOV_I)) >> SHIFT_THM_MOV_I));


... here, with absolute shifts, this could be rewritten as something like

Suggested change

((offset & (MASK_THM_MOV_IMM4<<SHIFT_THM_MOV_IMM4)) >> SHIFT_THM_MOV_IMM4) |

((offset & (MASK_THM_MOV_I<<SHIFT_THM_MOV_I)) >> SHIFT_THM_MOV_I));

(FIELD_PREP(MASK_THM_MOV_IMM4, offset >> SHIFT_THM_MOV_IMM4) |

(FIELD_PREP(MASK_THM_MOV_I, offset >> SHIFT_THM_MOV_I));

(please double check, I haven't tested this code)

I've tried to address all your requests but I don't get this one. I checked the FIELD_PREP()/FIELD_GET() and I got scared :) They contains division/multiplication and, in my case, they would be used with a variable so... I don't know how the compiler could optimize it...
@pillo79 Do we really need to use these macros here?!

Sorry for the delay, missed this reply last week! Not a deal breaker, it was always in the same spirit of using "native" Zephyr tools. These macros use a maths trick (multiplication / division by a power of 2 - the LSB_GET returns the lowest bit in the bitmask) to actually shift numbers to/from offset 0; the compiler turns this into a constant shift operation, so it has no runtime cost.
The macros also automatically chop bits outside the mask you give them, so you can avoid some typing.

The current code uses the following numbers:

#define MASK_THM_MOV_I BIT(10) #define MASK_THM_MOV_IMM4 GENMASK(3, 0) #define SHIFT_THM_MOV_I 1 #define SHIFT_THM_MOV_IMM4 12 ... ((offset & (MASK_THM_MOV_IMM4<<SHIFT_THM_MOV_IMM4)) >> SHIFT_THM_MOV_IMM4) | ((offset & (MASK_THM_MOV_I<<SHIFT_THM_MOV_I)) >> SHIFT_THM_MOV_I)

I would rewrite it as:

#define MASK_THM_MOV_I BIT(10) #define MASK_THM_MOV_IMM4 GENMASK(3, 0) #define SHIFT_THM_MOV_I 11 // <-- #define SHIFT_THM_MOV_IMM4 12 ... FIELD_PREP(MASK_THM_MOV_IMM4, offset >> SHIFT_THM_MOV_IMM4) | FIELD_PREP(MASK_THM_MOV_I, offset >> SHIFT_THM_MOV_I)

Note how the shift for MOV_I is 11, which I find easier to check and maintain than "the position of the mask, plus or minus 1 depending on the direction" (since some have inverted shifts!).
(the one for MOV_IMM4 gets lucky as the mask is zero-based, so the shift does not change 🙂)

pillo79 · 2024-04-03T09:50:45Z

arch/arm/core/elf.c

+	offset = ((upper & MASK_THM_MOV_IMM4) << SHIFT_THM_MOV_IMM4) |
+		((upper & MASK_THM_MOV_I) << SHIFT_THM_MOV_I) |
+		((lower & MASK_THM_MOV_IMM3) >> SHIFT_THM_MOV_IMM3) | (lower & MASK_THM_MOV_IMM8);


IMHO the way shifts are written in this file really makes them rather arcane - you have to read both the MASK_ and the SHIFT_ symbols and do some mental gym to understand where each term ends up in the final integer.
I wish bit shifts like this were written in Zephyr style, for example like this:

Suggested change

offset = ((upper & MASK_THM_MOV_IMM4) << SHIFT_THM_MOV_IMM4) |

((upper & MASK_THM_MOV_I) << SHIFT_THM_MOV_I) |

((lower & MASK_THM_MOV_IMM3) >> SHIFT_THM_MOV_IMM3) | (lower & MASK_THM_MOV_IMM8);

*offset = (FIELD_GET(MASK_THM_MOV_IMM4, upper) << SHIFT_THM_MOV_IMM4) |

(FIELD_GET(MASK_THM_MOV_I, upper) << SHIFT_THM_MOV_I) |

(FIELD_GET(MASK_THM_MOV_IMM3, lower) << SHIFT_THM_MOV_IMM3) |

(FIELD_GET(MASK_THM_MOV_IMM8, lower) << SHIFT_THM_MOV_IMM8);

where now the shifts are absolute and not related to the masks. The advantage is even clearer in the reverse operation below...

tests/subsys/llext/simple/CMakeLists.txt

teburd

Looks good to me, thank you for all the hard work

selescop · 2024-04-09T19:17:15Z

Thank you!
What should I do next to get it merged?

kartben · 2024-04-09T19:36:33Z

Thank you!
What should I do next to get it merged?

Just wait for one of our release engineers to merge it since it now has all the approvals :) It should happen virtually anytime now!

MaureenHelm · 2024-04-09T21:36:24Z

Thank you!
What should I do next to get it merged?

Just wait for one of our release engineers to merge it since it now has all the approvals :) It should happen virtually anytime now!

There's a merge conflict, so it needs to be rebased first.

selescop · 2024-04-10T08:56:27Z

There's a merge conflict, so it needs to be rebased first.

I used the graphic tool to fix the conflict, hope it is the way to go...

Adds support for all relocation type produced by GCC on ARM platform using partial linking (-r flag) or shared link (-fpic and -shared flag). Signed-off-by: Cedric Lescop <cedric.lescop@se.com>

Arch arm relocatate test covers: R_ARM_ABS32: all tests decode_thm_jumps -> relative jmp extension decode_thm_movs -> movwmovt extension Signed-off-by: Cedric Lescop <cedric.lescop@se.com>

pillo79

Thanks for the updates! The FIELD_ macros are just a preference, but it's probably either now or never 🤭, so I had to ask!

pillo79 · 2024-04-10T17:10:10Z

arch/arm/core/elf.c

+		((offset & (MASK_THM_MOV_IMM4<<SHIFT_THM_MOV_IMM4)) >> SHIFT_THM_MOV_IMM4) |
+		((offset & (MASK_THM_MOV_I<<SHIFT_THM_MOV_I)) >> SHIFT_THM_MOV_I));


Sorry for the delay, missed this reply last week! Not a deal breaker, it was always in the same spirit of using "native" Zephyr tools. These macros use a maths trick (multiplication / division by a power of 2 - the LSB_GET returns the lowest bit in the bitmask) to actually shift numbers to/from offset 0; the compiler turns this into a constant shift operation, so it has no runtime cost.
The macros also automatically chop bits outside the mask you give them, so you can avoid some typing.

The current code uses the following numbers:

#define MASK_THM_MOV_I BIT(10) #define MASK_THM_MOV_IMM4 GENMASK(3, 0) #define SHIFT_THM_MOV_I 1 #define SHIFT_THM_MOV_IMM4 12 ... ((offset & (MASK_THM_MOV_IMM4<<SHIFT_THM_MOV_IMM4)) >> SHIFT_THM_MOV_IMM4) | ((offset & (MASK_THM_MOV_I<<SHIFT_THM_MOV_I)) >> SHIFT_THM_MOV_I)

I would rewrite it as:

#define MASK_THM_MOV_I BIT(10) #define MASK_THM_MOV_IMM4 GENMASK(3, 0) #define SHIFT_THM_MOV_I 11 // <-- #define SHIFT_THM_MOV_IMM4 12 ... FIELD_PREP(MASK_THM_MOV_IMM4, offset >> SHIFT_THM_MOV_IMM4) | FIELD_PREP(MASK_THM_MOV_I, offset >> SHIFT_THM_MOV_I)

Note how the shift for MOV_I is 11, which I find easier to check and maintain than "the position of the mask, plus or minus 1 depending on the direction" (since some have inverted shifts!).
(the one for MOV_IMM4 gets lucky as the mask is zero-based, so the shift does not change 🙂)

teburd · 2024-04-10T17:34:00Z

I tend to agree with @pillo79 on the readability on using the FIELD_PREP() macro. Happy to re-approve if changed to use FIELD_PREP() and would prefer this be done. Having said that this is a lot of great work and changing to FIELD_PREP may be done immediately as a follow up or in this PR.

selescop · 2024-04-10T17:39:42Z

I tend to agree with @pillo79 on the readability on using the FIELD_PREP() macro. Happy to re-approve if changed to use FIELD_PREP() and would prefer this be done. Having said that this is a lot of great work and changing to FIELD_PREP may be done immediately as a follow up or in this PR.

Roger that. Just let me do another PR since I still have "a lot" to propose from the PR#70171, I'll add it to the list. I need a "small" victory here 😇

teburd · 2024-04-10T17:42:50Z

I tend to agree with @pillo79 on the readability on using the FIELD_PREP() macro. Happy to re-approve if changed to use FIELD_PREP() and would prefer this be done. Having said that this is a lot of great work and changing to FIELD_PREP may be done immediately as a follow up or in this PR.

Roger that. Just let me do another PR since I still have "a lot" to propose from the PR#70171, I'll add it to the list. I need a "small" victory here 😇

No worries, I get it

github-actions · 2024-04-10T18:14:01Z

Hi @selescop!
Congratulations on getting your very first Zephyr pull request merged 🎉🥳. This is a fantastic achievement, and we're thrilled to have you as part of our community!

To celebrate this milestone and showcase your contribution, we'd love to award you the Zephyr Technical Contributor badge. If you're interested, please claim your badge by filling out this form: Claim Your Zephyr Badge.

Thank you for your valuable input, and we look forward to seeing more of your contributions in the future! 🪁

zephyrbot added area: Linkable Loadable Extensions area: ARM ARM (32-bit) Architecture labels Mar 19, 2024

zephyrbot requested review from bbolen, carlocaione, galak, ithinuel, lyakh, MaureenHelm, microbuilder, pillo79, stephanosio and teburd March 19, 2024 16:50

zephyrbot assigned teburd Mar 19, 2024

selescop force-pushed the llext_arm_elf_relocations branch from 6a07793 to d57622d Compare March 20, 2024 17:10

pillo79 reviewed Mar 21, 2024

View reviewed changes

kartben previously requested changes Mar 21, 2024

View reviewed changes

include/zephyr/llext/llext.h Show resolved Hide resolved

include/zephyr/llext/llext.h Outdated Show resolved Hide resolved

include/zephyr/llext/llext.h Outdated Show resolved Hide resolved

selescop force-pushed the llext_arm_elf_relocations branch 2 times, most recently from 0a104ba to 97446f7 Compare March 21, 2024 13:25

selescop force-pushed the llext_arm_elf_relocations branch from 97446f7 to a57b0a9 Compare March 21, 2024 13:37

teburd requested changes Mar 21, 2024

View reviewed changes

selescop force-pushed the llext_arm_elf_relocations branch from a57b0a9 to ed7ffe0 Compare March 21, 2024 21:55

selescop force-pushed the llext_arm_elf_relocations branch from ed7ffe0 to fc85b55 Compare March 22, 2024 12:21

selescop requested review from mmahadevan108, dleach02 and sylvioalves as code owners March 22, 2024 12:21

zephyrbot requested review from microbuilder and stephanosio April 2, 2024 15:22

pillo79 reviewed Apr 3, 2024

View reviewed changes

selescop force-pushed the llext_arm_elf_relocations branch from 7a8da39 to af37474 Compare April 3, 2024 15:09

selescop requested a review from teburd April 4, 2024 16:37

selescop force-pushed the llext_arm_elf_relocations branch 2 times, most recently from f36e8fd to 4eb4f35 Compare April 5, 2024 12:03

teburd previously approved these changes Apr 9, 2024

View reviewed changes

bjarki-andreasen previously approved these changes Apr 9, 2024

View reviewed changes

selescop dismissed stale reviews from bjarki-andreasen and teburd via 8771564 April 10, 2024 08:53

selescop requested review from teburd and bjarki-andreasen April 10, 2024 08:56

selescop added 2 commits April 10, 2024 11:05

llext: Full ARM ELF relocation support

77ac5e1

Adds support for all relocation type produced by GCC on ARM platform using partial linking (-r flag) or shared link (-fpic and -shared flag). Signed-off-by: Cedric Lescop <cedric.lescop@se.com>

llext: add movt/movw relocation test

da2d3f0

Arch arm relocatate test covers: R_ARM_ABS32: all tests decode_thm_jumps -> relative jmp extension decode_thm_movs -> movwmovt extension Signed-off-by: Cedric Lescop <cedric.lescop@se.com>

selescop force-pushed the llext_arm_elf_relocations branch from 8771564 to da2d3f0 Compare April 10, 2024 09:10

teburd approved these changes Apr 10, 2024

View reviewed changes

pillo79 approved these changes Apr 10, 2024

View reviewed changes

nashif merged commit b573f44 into zephyrproject-rtos:main Apr 10, 2024
31 checks passed

marc-hb mentioned this pull request Jun 10, 2024

LLEXT full ARM relocation support (adding executable) #70171

Closed

henrikbrixandersen added the area: llext Linkable Loadable Extensions label Jul 8, 2024

		((offset & (MASK_THM_MOV_IMM4<<SHIFT_THM_MOV_IMM4)) >> SHIFT_THM_MOV_IMM4) \|
		((offset & (MASK_THM_MOV_I<<SHIFT_THM_MOV_I)) >> SHIFT_THM_MOV_I));

-	offset = ((upper & MASK_THM_MOV_IMM4) << SHIFT_THM_MOV_IMM4) |
-		((upper & MASK_THM_MOV_I) << SHIFT_THM_MOV_I) |
-		((lower & MASK_THM_MOV_IMM3) >> SHIFT_THM_MOV_IMM3) | (lower & MASK_THM_MOV_IMM8);
+	*offset = (FIELD_GET(MASK_THM_MOV_IMM4, upper) << SHIFT_THM_MOV_IMM4) |
+		(FIELD_GET(MASK_THM_MOV_I, upper) << SHIFT_THM_MOV_I) |
+		(FIELD_GET(MASK_THM_MOV_IMM3, lower) << SHIFT_THM_MOV_IMM3) |
+		(FIELD_GET(MASK_THM_MOV_IMM8, lower) << SHIFT_THM_MOV_IMM8);

llext: Full ARM ELF relocation support #70452

llext: Full ARM ELF relocation support #70452

Conversation

selescop commented Mar 19, 2024

pillo79 left a comment

Choose a reason for hiding this comment

selescop commented Mar 21, 2024

bjarki-andreasen commented Mar 21, 2024

selescop commented Mar 21, 2024

teburd left a comment • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

teburd Mar 21, 2024 • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

teburd Mar 21, 2024 • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

teburd Mar 21, 2024 • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

teburd Mar 21, 2024 • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

selescop commented Mar 21, 2024

pillo79 commented Mar 22, 2024

selescop commented Apr 2, 2024 • edited Loading

pillo79 left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

selescop Apr 3, 2024 • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

teburd left a comment

Choose a reason for hiding this comment

selescop commented Apr 9, 2024

kartben commented Apr 9, 2024

MaureenHelm commented Apr 9, 2024

selescop commented Apr 10, 2024

pillo79 left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

teburd commented Apr 10, 2024 • edited Loading

selescop commented Apr 10, 2024

teburd commented Apr 10, 2024

github-actions bot commented Apr 10, 2024

teburd left a comment •

edited

Loading

teburd Mar 21, 2024 •

edited

Loading

teburd Mar 21, 2024 •

edited

Loading

teburd Mar 21, 2024 •

edited

Loading

teburd Mar 21, 2024 •

edited

Loading

selescop commented Apr 2, 2024 •

edited

Loading

selescop Apr 3, 2024 •

edited

Loading

teburd commented Apr 10, 2024 •

edited

Loading