Skip to content

Apple's open-source XNU kernel on Android phones: its own loader and drivers, Metal on the Adreno. Runs from RAM; a reboot gives the phone back to Android.

C

The GPU driver came up and air_run died reaching it: a zeroed pipe_screen_config

vcvkk694629d13ca8
diff --git a/docs/air.md b/docs/air.md
index c32fc20..9e0dd3a 100644
--- a/docs/air.md
+++ b/docs/air.md
@@ -485,9 +485,38 @@ own firmware, which is not in the repository. **Not the driver refusing work: no
exactly what that first line is for: the pixel was right, and reporting it as a result on the
Adreno would have been false. Read the line before the numbers.
-What is *not* known until it is tried is whether the kernel's GPU driver and the submit path
-behave for a draw that does not come from CoreAnimation: the GLES path is that driver's only
-consumer today. `rc.jobs.air` asks with the GPU idle; `rc.jobs.air2` asks again after SpringBoard
+**Second run, with the firmware in place: the GPU driver came up and `air_run` died reaching it.**
+`logs/233855-run.log` has the Adreno starting properly — the GMU, the zap shader through
+TrustZone, the SMMU — and then nothing from `air_run` but one line, which is the whole reason
+that line exists:
+
+```
+air_run: signal 11 at 0x10, pc 0x10037e404, lr 0x1002829b0,
+ called from 0x1002829b0 0x1000a139c 0x1000077e8 0x1c3fef344 (unslid)
+```
+
+Joined against `llvm-nm` of the binary by nearest preceding symbol, that is
+
+```
+pc driParseConfigFiles + 0x54 <- the fault, at address 0x10
+lr fd_screen_create + 0x344
+ u_pipe_screen_lookup_or_create + 0xa4
+ main + 0x1f8
+```
+
+`fd_screen_create` passes the `pipe_screen_config` straight to `driParseConfigFiles`, which
+dereferences it; this program handed it a zeroed struct. The port's own
+`osmesa_create_gpu_screen()` fills it in with `driParseOptionInfo` and `driParseConfigFiles` over
+`driinfo_freedreno.h`, and that sequence — the one CoreAnimation goes through on this device every
+day — is now what `air_run` uses. **A machine with no `msm` device never reaches that code**, so
+no amount of running it here would have found it; it took the first contact with real hardware.
+
+Without the fault handler there would have been nothing at all to go on: iOS answers a crash here
+with "Failed to send exception EXC_CORPSE_NOTIFY" and no report. Four lines of borrowed code
+turned a silent exit into an exact address.
+
+What is *still* not known is whether the kernel's GPU driver and the submit path behave for a draw
+that does not come from CoreAnimation: the GLES path is that driver's only consumer today. `rc.jobs.air` asks with the GPU idle; `rc.jobs.air2` asks again after SpringBoard
is up and holding it. A refusal is a finding the whole port needs, not just this frontend, so
either answer is worth the run.
diff --git a/tools/mesa/air/air_run.c b/tools/mesa/air/air_run.c
index 93faea1..ee36e2b 100644
--- a/tools/mesa/air/air_run.c
+++ b/tools/mesa/air/air_run.c
@@ -39,6 +39,8 @@
#include "null/null_sw_winsys.h"
#ifdef GALLIUM_FREEDRENO
#include "freedreno/drm/freedreno_drm_public.h"
+#include "util/driconf.h"
+#include "util/xmlconfig.h"
#include <xf86drm.h>
#endif
#include "util/format/u_format.h"
@@ -234,7 +236,25 @@ main(int argc, char **argv)
if (!debug_get_bool_option("AIR_RUN_SOFTWARE", false)) {
int fd = drmOpenWithType("msm", NULL, DRM_NODE_RENDER);
if (fd >= 0) {
+ /* The driver option cache has to be real, not a zeroed struct:
+ * fd_screen_create passes it straight to driParseConfigFiles, which
+ * dereferences it. A zeroed one died at address 0x10 on the phone --
+ * the first thing this program ever did on a machine that has a GPU,
+ * and invisible on any machine that has not. This is the sequence
+ * osmesa_create_gpu_screen() uses in tools/mesa/freedreno-darwin.patch,
+ * which is the one CoreAnimation goes through on this device daily.
+ */
+ static const driOptionDescription freedreno_options[] = {
+#include "freedreno/driinfo_freedreno.h"
+ };
+ static driOptionCache info, cache;
static struct pipe_screen_config config;
+
+ driParseOptionInfo(&info, freedreno_options, ARRAY_SIZE(freedreno_options));
+ driParseConfigFiles(&cache, &info, 0, "msm", NULL, NULL, NULL, 0, NULL, 0);
+ config.options = &cache;
+ config.options_info = &info;
+
screen = fd_drm_screen_create_renderonly(fd, NULL, &config);
if (!screen)
fprintf(stderr, "air_run: the msm device opened but gave no screen\n");