realtek: don't disable MIPS counter on secondary VPEs

After observation that timer interrupt 7 always fires on secondary VPEs
the counter was disabled in the startup code. This is a bad idea when
building the kernel with jitterentropy. To generate entropy it makes use
of function random_get_entropy(). On MIPS architecture this simply reads
the counter register on the current core. With a disabled counter it
always returns the same value and the entropy initialization stalls the
core if it runs on a secondary VPE. See backtrace

[   21.736246] rcu: INFO: rcu_sched self-detected stall on CPU
[   21.736246] rcu: INFO: rcu_sched self-detected stall on CPU
[   21.748594] rcu:     1-....: (2100 ticks this GP) idle=064c/1/0x40000002 softirq=7/7 fqs=1050
[   21.748594] rcu:     1-....: (2100 ticks this GP) idle=064c/1/0x40000002 softirq=7/7 fqs=1050
[   21.766871] rcu:     (t=2102 jiffies g=-1187 q=25 ncpus=2)
[   21.766871] rcu:     (t=2102 jiffies g=-1187 q=25 ncpus=2)
[   21.778429] CPU: 1 UID: 0 PID: 1 Comm: swapper/0 Not tainted 6.12.39 #482
[   21.778429] CPU: 1 UID: 0 PID: 1 Comm: swapper/0 Not tainted 6.12.39 #482
[   21.778461] Hardware name: Zyxel GS1900-48
[   21.778461] Hardware name: Zyxel GS1900-48
...
[   21.779757] [<8029b968>] jent_measure_jitter+0xc8/0x10c
[   21.779757] [<8029b968>] jent_measure_jitter+0xc8/0x10c
[   21.779779] [<8029b9e8>] jent_gen_entropy+0x3c/0xb0
[   21.779779] [<8029b9e8>] jent_gen_entropy+0x3c/0xb0
[   21.779800] [<8029bcc0>] jent_entropy_collector_alloc+0x104/0x118
[   21.779800] [<8029bcc0>] jent_entropy_collector_alloc+0x104/0x118
[   21.779822] [<8029bd6c>] jent_entropy_init+0x4c/0x2ec
[   21.779822] [<8029bd6c>] jent_entropy_init+0x4c/0x2ec
[   21.779844] [<8086f184>] jent_mod_init+0x58/0xac
[   21.779844] [<8086f184>] jent_mod_init+0x58/0xac
[   21.779865] [<80100200>] do_one_initcall+0x70/0x250
[   21.779865] [<80100200>] do_one_initcall+0x70/0x250
[   21.779883] [<8085c018>] kernel_init_freeable+0x1f0/0x280
[   21.779883] [<8085c018>] kernel_init_freeable+0x1f0/0x280
[   21.779905] [<8067cba4>] kernel_init+0x20/0xb0
[   21.779905] [<8067cba4>] kernel_init+0x20/0xb0
[   21.779926] [<80101158>] ret_from_kernel_thread+0x14/0x1c
[   21.779926] [<80101158>] ret_from_kernel_thread+0x14/0x1c

This bit of entropy is helpful on these low end devices. Reenable the
counter and simply disable the interrupt.

Fixes: b7aab1958591e6290 ("realtek: SMP handling of R4K timer interrupts")
Reported-by: Sebastian Gottschall <s.gottschall@dd-wrt.com>
Signed-off-by: Markus Stockhausen <markus.stockhausen@gmx.de>
Link: https://github.com/openwrt/openwrt/pull/19499
Signed-off-by: Hauke Mehrtens <hauke@hauke-m.de>

diff --git a/target/linux/realtek/files-6.12/arch/mips/rtl838x/prom.c b/target/linux/realtek/files-6.12/arch/mips/rtl838x/prom.c
index c2c4b09d1d140036660169da693bbff0b0612224..68b4e5c14494d71757989ac9b8adea177fd5f35c 100644 (file)
--- a/target/linux/realtek/files-6.12/arch/mips/rtl838x/prom.c
+++ b/target/linux/realtek/files-6.12/arch/mips/rtl838x/prom.c
@@ -13,6 +13,7 @@
 #include <asm/mips-cps.h>
 #include <asm/prom.h>
 #include <asm/smp-ops.h>
+#include <linux/smp.h>
 
 #include <mach-rtl83xx.h>
 
@@ -41,13 +42,13 @@ static void rtlsmp_finish(void)
 {
        /* These devices are low on resources. There might be the chance that CEVT_R4K is
         * not enabled in kernel build. Nevertheless the timer and interrupt 7 might be
-        * active by default after startup of secondary VPE. With no registered handler
-        * that leads to continuous unhandeled interrupts. In this case disable counting
-        * (DC) in the core and confirm a pending interrupt.
+        * active by default after startup of secondary VPEs. With no registered handler
+        * that leads to continuous unhandeled interrupts. Disable it but keep the counter
+        * running so it can still be used as an entropy source.
         */
        if (!IS_ENABLED(CONFIG_CEVT_R4K)) {
-               write_c0_cause(read_c0_cause() | CAUSEF_DC);
-               write_c0_compare(0);
+               write_c0_status(read_c0_status() & ~CAUSEF_IP7);
+               write_c0_compare(read_c0_count() - 1);
        }
 
        local_irq_enable();