Skip to content

Releases

webrte-1.022

Choose a tag to compare

@github-actions github-actions released this 31 Aug 23:56
557eea8
Merge pull request #2 from Ahghahakahaj/support/13.52-fixall

Add 13.52 offsets header (installer/include/1352_offsets.h)

webrte-1.09

Choose a tag to compare

@github-actions github-actions released this 28 Aug 01:33
Do not install the trap_fatal hook on 13.00

install_hooks wrote three bytes and an absolute jmp at kernbase+0x1718D8, an
offset hardcoded for one firmware that never came from a per-firmware table.
On 13.00 that address is 0x68 bytes inside an unrelated function, on a
mov/test/jnz sequence. Overwriting it meant every thread that ran that
function jumped into hook_trap_fatal with the wrong registers.

That is what powered the console off whenever an app was launched: the crash
came from the kernel executing its own corrupted code, with the payload idle
and no request in flight, which is why throttling the dashboard and locking
the allproc walks did not help.

The hook only formats crash information into klog and nothing depends on it,
so it is skipped where the offset has not been checked rather than patched
blind. Every other firmware keeps the behaviour it had.

webrte-1.08

Choose a tag to compare

@github-actions github-actions released this 28 Aug 01:22
Take allproc_lock shared around every walk of the process list

proc_find_by_name, proc_find_by_pid and sys_proc_list all walked allproc with
no lock held. Launching an app while a walk was in flight powered the console
off, which is the list being modified underneath the walker.

Each walk now acquires allproc_lock shared once and releases it once, with no
return in between - leaking it would block every fork and exit on the system,
which is a hang rather than a reboot. Firmwares without the offsets mapped keep
the old unlocked behaviour, since the pointers are NULL there and the helpers
do nothing.

This closes the window during the walk. It does not make the returned proc
pointer safe to use after the lock is dropped; that needs PHOLD/PRELE.

webrte-1.07

Choose a tag to compare

@github-actions github-actions released this 28 Aug 01:07
Fix the kernel pointer test rejecting every proc

The guard added in the previous commit used 0xFFFFFF8000000000, which is one
hex digit away from KERNBASE and far above where proc structs actually live.
They are malloc'd into the kernel heap - 0xFFFFF8.. in the direct map or
0xFFFFFE.. in the kernel map - so every real pointer failed the test, the walk
never started, and the installer reported "could not find SceShellCore".

The right bound is the start of the upper canonical half, 0xFFFF800000000000:
it accepts heap and image addresses alike while still rejecting NULL and any
userland pointer.

webrte-1.06

Choose a tag to compare

@github-actions github-actions released this 28 Aug 01:00
Stop the allproc walk from following a dangling pointer

proc_find_by_pid and proc_find_by_name walk allproc with no lock held and no
reference taken on what they return. When a process exits mid-walk the next
p_forw is freed memory, and dereferencing it panics the kernel - which is why
touching a long-lived process like SceShellCore is fine while touching an app
that is starting and stopping takes the console down.

The lock cannot be taken from here, so this does not remove the race. It does
stop the walk from running off into unmapped memory: each link is checked for
a plausible kernel address before it is dereferenced, and the walk is bounded
so a corrupted list terminates instead of running forever.

Callers should still avoid hammering these paths.

webrte-1.05

Choose a tag to compare

@github-actions github-actions released this 28 Aug 00:43
Expose process control and kernel inspection

/signal?pid=&sig=   any signal 1-31 through the same kill() pause/resume use
/thrinfo?pid=       lwpid, priority and thread name via SYS_PROC_THRINFO
/kernbase           kernel base via sys_kern_base
/kread              kernel memory via sys_kern_rw, capped at 4 KB per request

All four were already implemented in the kdebugger and simply never reached
the HTTP layer. Signals outside 1-31 and pid < 1 are rejected so a stray
request cannot become an arbitrary syscall argument, and kernel reads are
bounded because a bad one faults the whole console.

webrte-1.04

Choose a tag to compare

@github-actions github-actions released this 28 Aug 00:37
Point /sensors at the nodes that actually exist on 13.00

Probing a live console showed Sony strips most of the stock FreeBSD tree:
hw.physmem, vm.stats.*, kern.cp_time and the whole hw.acpi.thermal branch
return ENOENT. What is present: dev.cpu.0.freq and freq_levels, hw.model,
hw.availpages, vm.budgets.mlock_*, vm.swap_*, machdep.liverpool.icc_max and
telemetry, machdep.tsc_freq, kern.version.

Adds sceKernelGetSocSensorTemperature, whose signature was taken from the
disassembly rather than guessed, and sceKernelGetDirectMemorySize for the
memory total. Printable sysctls now come back as strings instead of hex.

Fan duty stays absent: the icc_fan device registers no sysctl, so reading it
needs kernel-side ICC access rather than a userland read.

webrte-1.03

Choose a tag to compare

@github-actions github-actions released this 28 Aug 00:29
Credit 13.00 support in the on-screen notification

webrte by golden
Updated By EchoStretch
Developed & FW 13.00 support by MasterMaind

webrte-1.02

Choose a tag to compare

@github-actions github-actions released this 28 Aug 00:22
Add read-only telemetry endpoints

/sysctl?name=  reads any sysctl node and returns size, int, str and hex
/sensors       bundles CPU temperature with the sysctls a devkit-style
               readout needs: frequency, memory, thermal zone, fan duty,
               per-CPU tick counters

Both go through sysctlbyname and sceKernelGetCpuTemperature, which libPS4
already resolves in initKernel(), so no new symbol resolution is involved.
Reads only; nothing here writes to the console.

webrte-1.019

Choose a tag to compare

@github-actions github-actions released this 28 Aug 13:10
Move the dashboard to port 8181

8080 is the most contended port on a development machine; anything from a
local web server to a docker mapping will already own it.