On Fri, Jan 19, 2024 at 11:11:44AM -0800, Bjorn Andersson wrote:
> On Thu, Jan 18, 2024 at 06:58:19PM -0500, Eric Chanudet wrote:
> > On Thu, Jan 18, 2024 at 09:27:11PM +0530, Ninad Naik wrote:
> > > New memory map layout changes (by Qualcomm firmware) have brought
> > > in updates to base addresses and/or size for different memory regions
> > > like cpcucp_fw, tz-stat, and also introduces new memory regions for
> > > resource manager firmware. This change brings in these corresponding
> > > memory map updates to the SA8775P SoC device tree.
> > >
> > > Signed-off-by: Ninad Naik <quic_ninanaik@xxxxxxxxxxx>
> >
> > With next-20240118, without this patch, running "./memtester 32G"[1]
> > crashed the board quickly during the mlock:
> >
> > [ 42.144892] Internal error: synchronous external abort: 0000000096000010 [#1] PREEMPT SMP
>
> Sounds like just passing "memtest=1" on the kernel command line (with
> CONFIG_MEMTEST=y) would trip this...
Actually, this does a better job, not having to deal with the
oom-killer, and picks up the following missing region with this patch:
[ 0.000000] early_memtest: # of tests: 1
[ 0.000000] 0x0000000090880000 - 0x00000000908b0000 pattern 0000000000000000
[ 0.000000] 0x00000000908c1000 - 0x00000000908f0000 pattern 0000000000000000
[ 0.000000] 0x0000000090b00000 - 0x0000000090c00000 pattern 0000000000000000
[ 0.000000] Internal error: synchronous external abort: 0000000096000010 [#1] PREEMPT SMP
> > > @@ -373,8 +378,43 @@ smem_mem: smem@90900000 {
> > > hwlocks = <&tcsr_mutex 3>;
> > > };
> > >
> > > - cpucp_fw_mem: cpucp-fw@90b00000 {
> > > - reg = <0x0 0x90b00000 0x0 0x100000>;
> > > + tz_sail_mailbox_mem: tz-sail-mailbox@90c00000 {
> > > + reg = <0x0 0x90c00000 0x0 0x100000>;
> > > + no-map;
> > > + };
Re-introducing 0x90b00000-0x90bfffff as no-map passes the kernel
memtest=1 and boots with the firmware I have.
My previous test was incomplete. Please disregard the Tested-by.
--
Eric Chanudet
