Build H2 with Picolibc by quic-k · Pull Request #10 · qualcomm/hexagon-hypervisor

quic-k · 2026-04-20T09:45:33Z

No description provided.

colmode

Looks good overall, but we might want to merge this after erich's makefile changes rather than the other way around.

quic-k · 2026-05-05T14:28:55Z

+	r1 = #__boot_net_phys_offset__
+	memw(r1) = r0
+	jump _start          // picolibc CRT entry (replaces hexagon_pre_main)
+


we can, but since we are not including any C header, we don't have __PICOLIBC__
so for simplicity I created 2 files

maybe I will have to use the PICOLIBC=1 flag to define a macro, which we can use
or do you have something else in mind?

"PICOLIBC=1 flag to define a macro", yes. But since this is all transitional and we'll eventually use only pico I think that leaving it as is would be fine.

quic-k · 2026-05-05T14:28:55Z

+	r1 = #__boot_net_phys_offset__
+	memw(r1) = r0
+	jump _start          // picolibc CRT entry (replaces hexagon_pre_main)
+


we can, but since we are not including any C header, we don't have __PICOLIBC__
so for simplicity I created 2 files

maybe I will have to use the PICOLIBC=1 flag to define a macro, which we can use
or do you have something else in mind?

quic-k · 2026-05-05T14:32:03Z

Looks good overall, but we might want to merge this after erich's makefile changes rather than the other way around.

thanks for the review, ping me whenever that is merged, I will have to resolve merge conflicts

Also, we don't yet have a way to run unit tests with pico since those tools aren't on github. Best to run make testall locally and then comment here when it passes.

sure, so far I have built H2 successfully and used the install artifacts to run Picolibc testsuite which passes
will look into getting H2 unit tests working with Picolibc

Bryan Bayerdorffer (bryanb-h2) · 2026-05-05T14:39:46Z

Looks good overall, but we might want to merge this after erich's makefile changes rather than the other way around.

Hey, look; our first comment from a non-qc account :)

quic-k · 2026-05-07T10:34:33Z

why is hexagon_standalone.h being used? Bryan Bayerdorffer (@bryanb-h2)

hexagon-hypervisor/kernel/util/stmode/test/test.c

Line 12 in f74ebc7

#include <hexagon_standalone.h>

hexagon-hypervisor/kernel/util/spinlock/test/test.c

Line 10 in f74ebc7

#include <hexagon_standalone.h>

Bryan Bayerdorffer (bryanb-h2) · 2026-05-07T10:48:04Z

why is hexagon_standalone.h being used? Bryan Bayerdorffer (@bryanb-h2)

hexagon-hypervisor/kernel/util/stmode/test/test.c

Line 12 in f74ebc7

#include <hexagon_standalone.h>

hexagon-hypervisor/kernel/util/spinlock/test/test.c

Line 10 in f74ebc7

#include <hexagon_standalone.h>

Because most of the unit tests run with standalone with a test harness for testing only one function -- avoids booting the kernel to simplify debug. We need a substitute for standalone but we haven't addressed this yet. The unit tests don't even use features of standalone. Using the default dinkum build is just a convenient way to link with crt0 and get to main(). We actually have a min_crt0 in the h2 repo. Just need to dust that off and change how we link unit tests.

Here it is, entirely:

/*

Copyright (c) Qualcomm Technologies, Inc. and/or its subsidiaries.
SPDX-License-Identifier: BSD-3-Clause-Clear
*/

.section .start,"awx",progbits

.global _start

.global hexagon_pre_main

#_start:

jump start

hexagon_pre_main:
r0 = ##SDA_BASE
gp = r0
r29 = ##stack
{
r0 = r29
r1 = #1
call main
}

jump start

.data
.p2align 8
.space 1024

stack:
.word stack
.word 0

quic-k · 2026-05-07T11:10:11Z

Because most of the unit tests run with standalone with a test harness for testing only one function -- avoids booting the kernel to simplify debug.

yeah, for debugging that is better, but booter tests would have been easier for me to get passing

We need a substitute for standalone but we haven't addressed this yet. The unit tests don't even use features of standalone. Using the default dinkum build is just a convenient way to link with crt0 and get to main(). We actually have a min_crt0 in the h2 repo. Just need to dust that off and change how we link unit tests.

is this planned to be done soon? or am I supposed to do this in this PR? maybe I could but this sounds like lot of debugging
:\

Bryan Bayerdorffer (bryanb-h2) · 2026-05-07T11:26:09Z

Because most of the unit tests run with standalone with a test harness for testing only one function -- avoids booting the kernel to simplify debug.

yeah, for debugging that is better, but booter tests would have been easier for me to get passing

We need a substitute for standalone but we haven't addressed this yet. The unit tests don't even use features of standalone. Using the default dinkum build is just a convenient way to link with crt0 and get to main(). We actually have a min_crt0 in the h2 repo. Just need to dust that off and change how we link unit tests.

is this planned to be done soon? or am I supposed to do this in this PR? maybe I could but this sounds like lot of debugging :\

Not unless you're really keen to do it. I just filed QDSP-57392 for this.

.

Bryan Bayerdorffer (bryanb-h2)

The merge conflicts with master were complicated in some places. Please check if I screwed something up.

Bryan Bayerdorffer (bryanb-h2) · 2026-05-19T20:05:52Z

 	}

-	if (NULL == (pmap->base.raw = (h2_u32_t)realloc((void *)(pmap->base.raw), sizeof(H2K_linear_fmt_t) * (end + npages + 1)))) {
+	if (0 == (pmap->base.raw = (h2_u32_t)realloc((void *)(pmap->base.raw), sizeof(H2K_linear_fmt_t) * (end + npages + 1)))) {


booter.c:673:11: warning: comparison between pointer and integer ('void *' and 'h2_u32_t' (aka 'unsigned int')) [-Wpointer-integer-compare] 673 | if (NULL == (pmap->base.raw = (h2_u32_t)realloc((void *)(pmap->base.raw), sizeof(H2K_linear_fmt_t) * (end + npages + 1)))) { | ~~~~ ^ ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

So pico defines NULL differently? I thought that def was part of the tools and we don't get the same warning when building with the existing dinkum tools.

in Pico toolchain NULL in preprocessed to ((void*)0) (comes from clang's stddef.h)
existing tools might be just defining NULL to 0

Ok, then the right thing to do is probably this, which works with dinkum also:

diff --git a/booter/booter.c b/booter/booter.c
index 071702bc6..93fcf9adc 100644
--- a/booter/booter.c
+++ b/booter/booter.c
@@ -670,7 +670,7 @@ void add_linear_trans(unsigned int idx, unsigned long va, unsigned long long pa,
vm_params[idx].pmap_added = 1;
}

if (NULL == (pmap->base.raw = (h2_u32_t)realloc((void *)(pmap->base.raw), sizeof(H2K_linear_fmt_t) * (end + npages + 1)))) {

if (NULL == (void *)(pmap->base.raw = (h2_u32_t)realloc((void *)(pmap->base.raw), sizeof(H2K_linear_fmt_t) * (end + npages + 1)))) {
error("realloc pmap->base", NULL);
}
base = (H2K_linear_fmt_t *)(pmap->base.raw);

Is there a reason int types are used for pointers?

Arguably H2K_offset_t should union a H2K_linear_fmt_t * so that it can be used without the funny casts here. But all this code is kind of "wrong" in the sense that we shouldn't really have the guest's initial translations allocating in booter's heap. So all of this needs to be reworked and we can fix this type of thing then.

Bryan Bayerdorffer (bryanb-h2) · 2026-05-19T20:07:13Z


-CFLAGS = $(OPTIMIZE) -mv$(TOOLARCH) -DARCHV=$(ARCHV) -Wall -Werror -Wno-builtin-requires-header -g -I. -I$(INSTALLPATH)/include -I$(KERNEL_BUILD_DIR)/include
-LDFLAGS = -L$(INSTALLPATH)/lib -moslib=h2 -Qunused-arguments
+ifeq ($(PICOLIBC),1)


How about pulling all these if pico sections into one make include file?

sure, I could do something like that
but maybe I should wait till H2 unit tests are cleaned up, so I don't have to rebase multiple times?

quic-k

I have rebased on master

quic-k · 2026-05-20T11:24:52Z

 	}

-	if (NULL == (pmap->base.raw = (h2_u32_t)realloc((void *)(pmap->base.raw), sizeof(H2K_linear_fmt_t) * (end + npages + 1)))) {
+	if (0 == (pmap->base.raw = (h2_u32_t)realloc((void *)(pmap->base.raw), sizeof(H2K_linear_fmt_t) * (end + npages + 1)))) {


booter.c:673:11: warning: comparison between pointer and integer ('void *' and 'h2_u32_t' (aka 'unsigned int')) [-Wpointer-integer-compare] 673 | if (NULL == (pmap->base.raw = (h2_u32_t)realloc((void *)(pmap->base.raw), sizeof(H2K_linear_fmt_t) * (end + npages + 1)))) { | ~~~~ ^ ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

quic-k · 2026-05-20T11:27:01Z


-CFLAGS = $(OPTIMIZE) -mv$(TOOLARCH) -DARCHV=$(ARCHV) -Wall -Werror -Wno-builtin-requires-header -g -I. -I$(INSTALLPATH)/include -I$(KERNEL_BUILD_DIR)/include
-LDFLAGS = -L$(INSTALLPATH)/lib -moslib=h2 -Qunused-arguments
+ifeq ($(PICOLIBC),1)


sure, I could do something like that
but maybe I should wait till H2 unit tests are cleaned up, so I don't have to rebase multiple times?

Signed-off-by: Kushal Pal <kushpal@qti.qualcomm.com>

Disable the remaining tests for now Signed-off-by: Kushal Pal <kushpal@qti.qualcomm.com>

quic-k requested review from andreykarpenko-qc and Bryan Bayerdorffer (bryanb-h2) April 20, 2026 09:45

colmode previously requested changes May 5, 2026

View reviewed changes

quic-k commented May 5, 2026

View reviewed changes

Bryan Bayerdorffer (bryanb-h2) requested changes May 19, 2026

View reviewed changes

quic-k force-pushed the build-with-pico branch from 786255c to d68962a Compare May 20, 2026 11:21

quic-k commented May 20, 2026

View reviewed changes

quic-k force-pushed the build-with-pico branch 3 times, most recently from 7e6b53e to 903038c Compare June 11, 2026 09:07

quic-k force-pushed the build-with-pico branch from 903038c to f20a720 Compare June 22, 2026 08:29

github-actions Bot added the untested Mark untested PRs label Jun 22, 2026

Build H2 with Picolibc

0277755

Signed-off-by: Kushal Pal <kushpal@qti.qualcomm.com>

quic-k force-pushed the build-with-pico branch 5 times, most recently from 2f37277 to d68c015 Compare June 29, 2026 13:44

Fix unit tests for Picolibc

113c5f8

Disable the remaining tests for now Signed-off-by: Kushal Pal <kushpal@qti.qualcomm.com>

quic-k force-pushed the build-with-pico branch from d68c015 to 113c5f8 Compare June 29, 2026 14:17

Uh oh!

Conversation

quic-k commented Apr 20, 2026

Uh oh!

colmode left a comment

Choose a reason for hiding this comment

Uh oh!

quic-k May 5, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

quic-k May 5, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

quic-k commented May 5, 2026

Uh oh!

Bryan Bayerdorffer (bryanb-h2) commented May 5, 2026

Uh oh!

quic-k commented May 7, 2026

Uh oh!

Bryan Bayerdorffer (bryanb-h2) commented May 7, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

.global _start

jump start

jump start

Uh oh!

quic-k commented May 7, 2026

Uh oh!

Bryan Bayerdorffer (bryanb-h2) commented May 7, 2026

Uh oh!

Bryan Bayerdorffer (bryanb-h2) left a comment

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!

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!

quic-k left a comment

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!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

4 participants

quic-k May 5, 2026 •

edited

Loading

quic-k May 5, 2026 •

edited

Loading

Bryan Bayerdorffer (bryanb-h2) commented May 7, 2026 •

edited

Loading