ARM: mx7: psci: fix suspend/resume e10133 workaround
Subsystems (3)
PATCH 1/1 ARM: mx7: psci: fix suspend/resume e10133 workaround
The e10133 workaround was broken in two places:
- The code intended to temporarily mask all interrupts in GPC_IMRx_CORE0. While the old register values were saved, the actual masking was missing.
- imx_udelay() expects the system counter to run at its base frequency, but the system counter is switched to a lower frequency earlier in psci_system_suspend(), leading to a much longer delay than intended. Replace the call with an equivalent loop (linux-imx 5.15 does the same)
This fixes the SoC hanging forever when there was already a wakeup IRQ pending while suspending.
--- arch/arm/mach-imx/mx7/psci-mx7.c | 9 ++++++--- 1 file changed, 6 insertions(+), 3 deletions(-)
arch/arm/mach-imx/mx7/psci-mx7.c +6 -3
@@ -643,8 +643,10 @@ __secure void psci_system_suspend(u32 __always_unused function_id,/* disable GIC distributor */ writel (0 ,GIC400_ARB_BASE_ADDR +GIC_DIST_OFFSET ); -for (i =0 ; i <4 ; i++) +for (i =0 ; i <4 ; i++) { gpc_mask[i] =readl (GPC_IPS_BASE_ADDR +GPC_IMR1_CORE0 + i *4 ); +writel (~0 ,GPC_IPS_BASE_ADDR +GPC_IMR1_CORE0 + i *4 ); + }/* * enable the RBC bypass counter here
@@ -668,7 +670,7 @@ __secure void psci_system_suspend(u32 __always_unused function_id,writel (gpc_mask[i],GPC_IPS_BASE_ADDR +GPC_IMR1_CORE0 + i *4 );/* -* now delay for a short while (3usec) +* now delay for a short while (~3usec) * ARM is at 1GHz at this point * so a short loop should be enough. * this delay is required to ensure that
@@ -677,7 +679,8 @@ __secure void psci_system_suspend(u32 __always_unused function_id,* or in case an interrupt arrives just * as ARM is about to assert DSM_request. */ -imx_udelay (3 ); +for (i =0 ; i <2000 ; i++) + asmvolatile ("" );/* save resume entry and sp in CPU0 GPR registers */ asmvolatile ("mov %0, sp" :"=r" (val));
Signature · 1 line
2.25.1
Quoted 10 lines · Matthias Schiffer
The e10133 workaround was broken in two places:
- The code intended to temporarily mask all interrupts in
GPC_IMRx_CORE0. While the old register values were saved, the actual masking was missing.
-imx_udelay()expects the system counter to run at its base frequency, but the system counter is switched to a lower frequency earlier inpsci_system_suspend(), leading to a much longer delay than intended. Replace the call with an equivalent loop (linux-imx 5.15 does the same)
Quoted 5 lines
This fixes the SoC hanging forever when there was already a wakeup IRQ pending while suspending.
Fixes 57b620255e ("imx: mx7: add system suspend/resume support")Signed-off-by Matthias Schiffer <[email protected]>
Ping, any feedback on this?
I'm not entirely sure anymore if this solution is adequate, as the duration of the loop depends on the CPU clock frequency and cache enable/disable state. Maybe a modified imx_udelay() that accounts for the changed system counter configuration would be necessary after all?
Quoted 39 lines
--- arch/arm/mach-imx/mx7/psci-mx7.c | 9 ++++++--- 1 file changed, 6 insertions(+), 3 deletions(-)arch/arm/mach-imx/mx7/psci-mx7.c +6
-3@@ -643,8 +643,10 @@ __secure void psci_system_suspend(u32 __always_unused function_id,/* disable GIC distributor */ writel (0 ,GIC400_ARB_BASE_ADDR +GIC_DIST_OFFSET ); -for (i =0 ; i <4 ; i++) +for (i =0 ; i <4 ; i++) { gpc_mask[i] =readl (GPC_IPS_BASE_ADDR +GPC_IMR1_CORE0 + i *4 ); +writel (~0 ,GPC_IPS_BASE_ADDR +GPC_IMR1_CORE0 + i *4 ); + }/* * enable the RBC bypass counter here @@ -668,7 +670,7 @@ __secure void psci_system_suspend(u32 __always_unused function_id,writel (gpc_mask[i],GPC_IPS_BASE_ADDR +GPC_IMR1_CORE0 + i *4 );/* -* now delay for a short while (3usec) +* now delay for a short while (~3usec) * ARM is at 1GHz at this point * so a short loop should be enough. * this delay is required to ensure that @@ -677,7 +679,8 @@ __secure void psci_system_suspend(u32 __always_unused function_id,* or in case an interrupt arrives just * as ARM is about to assert DSM_request. */ -imx_udelay (3 ); +for (i =0 ; i <2000 ; i++) + asmvolatile ("" );/* save resume entry and sp in CPU0 GPR registers */ asmvolatile ("mov %0, sp" :"=r" (val));
PATCH 1/1 ARM: mx7: psci: fix suspend/resume e10133 workaround
In reply to Matthias SchifferQuoted 10 lines
The e10133 workaround was broken in two places:
- The code intended to temporarily mask all interrupts inGPC_IMRx_CORE0. While the old register values were saved, the actual masking was missing.
-imx_udelay()expects the system counter to run at its base frequency, but the system counter is switched to a lower frequency earlier inpsci_system_suspend(), leading to a much longer delay than intended. Replace the call with an equivalent loop (linux-imx 5.15 does the same)
This fixes the SoC hanging forever when there was already a wakeup IRQ pending while suspending.
Quoted 2 lines
Fixes 57b620255e ("imx: mx7: add system suspend/resume support")Signed-off-by Matthias Schiffer <[email protected]>
Applied to u-boot-imx, master, thanks !
Best regards,
Stefano Babic
Signature · 5 lines
===================================================================== DENX Software Engineering GmbH, Managing Director: Wolfgang Denk HRB 165235 Munich, Office: Kirchenstr.5, D-82194 Groebenzell, Germany Phone: +49-8142-66989-53 Fax: +49-8142-66989-80 Email: [email protected] =====================================================================