Re: [PATCH] RISC-V: Check clint_time_val before use

[Damien Le Moal <Damien.LeMoal@wdc.com>]

On Sat, 2020-09-26 at 09:31 +0000, Anup Patel wrote:
> > -----Original Message-----
> > From: Damien Le Moal <Damien.LeMoal@wdc.com>
> > Sent: 26 September 2020 14:55
> > To: paul.walmsley@sifive.com; palmer@dabbelt.com;
> > palmerdabbelt@google.com; Anup Patel <Anup.Patel@wdc.com>;
> > aou@eecs.berkeley.edu
> > Cc: anup@brainfault.org; linux-riscv@lists.infradead.org; Atish Patra
> > <Atish.Patra@wdc.com>; Alistair Francis <Alistair.Francis@wdc.com>; linux-
> > kernel@vger.kernel.org
> > Subject: Re: [PATCH] RISC-V: Check clint_time_val before use
> > 
> > On Sat, 2020-09-26 at 12:57 +0530, Anup Patel wrote:
> > > The NoMMU kernel is broken for QEMU virt machine from Linux-5.9-rc6
> > > because the get_cycles() and friends are called very early from
> > > rand_initialize() before CLINT driver is probed. To fix this, we
> > > should check clint_time_val before use in get_cycles() and friends.
> > > 
> > > Fixes: d5be89a8d118 ("RISC-V: Resurrect the MMIO timer implementation
> > > for M-mode systems")
> > > Signed-off-by: Anup Patel <anup.patel@wdc.com>
> > > ---
> > >  arch/riscv/include/asm/timex.h | 12 +++++++++---
> > >  1 file changed, 9 insertions(+), 3 deletions(-)
> > > 
> > > diff --git a/arch/riscv/include/asm/timex.h
> > > b/arch/riscv/include/asm/timex.h index 7f659dda0032..52b42bb1602c
> > > 100644
> > > --- a/arch/riscv/include/asm/timex.h
> > > +++ b/arch/riscv/include/asm/timex.h
> > > @@ -17,18 +17,24 @@ typedef unsigned long cycles_t;  #ifdef
> > > CONFIG_64BIT  static inline cycles_t get_cycles(void)  {
> > > -	return readq_relaxed(clint_time_val);
> > > +	if (clint_time_val)
> > > +		return readq_relaxed(clint_time_val);
> > > +	return 0;
> > >  }
> > >  #else /* !CONFIG_64BIT */
> > >  static inline u32 get_cycles(void)
> > >  {
> > > -	return readl_relaxed(((u32 *)clint_time_val));
> > > +	if (clint_time_val)
> > > +		return readl_relaxed(((u32 *)clint_time_val));
> > > +	return 0;
> > >  }
> > >  #define get_cycles get_cycles
> > > 
> > >  static inline u32 get_cycles_hi(void)  {
> > > -	return readl_relaxed(((u32 *)clint_time_val) + 1);
> > > +	if (clint_time_val)
> > > +		return readl_relaxed(((u32 *)clint_time_val) + 1);
> > > +	return 0
> > >  }
> > >  #define get_cycles_hi get_cycles_hi
> > >  #endif /* CONFIG_64BIT */
> > 
> > Applying this on top of rc6, I now get a hang on Kendryte boot...
> > No problems without the patch on the other hand.
> 
> Not sure about the issue with Kendryte but I get a crash on
> QEMU virt machine without this patch.

With this applied in addition to your patch, it works.

diff --git a/drivers/clocksource/timer-clint.c b/drivers/clocksource/timer-
clint.c
index d17367dee02c..8dbec85979fd 100644
--- a/drivers/clocksource/timer-clint.c
+++ b/drivers/clocksource/timer-clint.c
@@ -37,7 +37,7 @@ static unsigned long clint_timer_freq;
 static unsigned int clint_timer_irq;
 
 #ifdef CONFIG_RISCV_M_MODE
-u64 __iomem *clint_time_val;
+u64 __iomem *clint_time_val = NULL;
 #endif
 
 static void clint_send_ipi(const struct cpumask *target)

-- 
Damien Le Moal
Western Digital