clocksource/arm_arch_timer: Add basic ARMv8.6 support

PATCH Superseded Marc Zyngier 2 participants 20 messages Sep 22, 2021 – Sep 30, 2021

Subsystems (4)

Arm (32-bit) arm64 Clocksource, clockevent drivers Arm architected timer driver

lore

Share & tools

lore

Marc Zyngier

author maintainer

2021-09-22 21:19 UTC #0 lore raw

v2 cover clocksource/arm_arch_timer: Add basic ARMv8.6 support

This is v2 of the series enabling ARMv8.6 support for timer subsystem, and was prompted by a discussion with Oliver around the fact that an ARMv8.6 implementation must have a 1GHz counter, which leads to a number of things to break in the timer code:

- the counter rollover can come pretty quickly as we only advertise a 56bit counter,
- the maximum timer delta can be remarkably small, as we use the countdown interface which is limited to 32bit...

Thankfully, there is a way out: we can compute the minimal width of the counter based on the guarantees that the architecture gives us, and we can use the 64bit comparator interface instead of the countdown to program the timer.

Finally, we start making use of the ARMv8.6 ECV features by switching accesses to the counters to a self-synchronising register, removing the need for an ISB. Hopefully, implementations will *not* just stick an invisible ISB there...

A side effect of the switch to CVAL is that XGene-1 breaks. I have added a workaround to keep it alive.

I have added Oliver's original patch[0] to the series and tweaked a couple of things. Blame me if I broke anything.

The whole things has been tested on Juno (sysreg + MMIO timers), XGene-1 (broken sysreg timers), FVP (FEAT_ECV, CNT*CTSS_EL0).

* From v1:
- New patch adding a bunch of BUILD_BUG()s for register accesses we don't expect. This makes subsequent patches much simpler.
- New patch moving the ISBs for workaround in a way that makes more sense for the self-synchronising accessors.
- Rework the XGene-1 workaround to rely solely on MIDR.
- Split the CNTVCTSS trap handling in its own patch.
- Rebased on 5.15-rc2
- Collected RBs, with thanks.

[0] https://lore.kernel.org/r/[email protected]
[1] https://lore.kernel.org/r/[email protected]

Marc Zyngier (15):
clocksource/arm_arch_timer: Add build-time guards for unhandled
register accesses
clocksource/arm_arch_timer: Drop CNT*_TVAL read accessors clocksource/arm_arch_timer: Extend write side of timer register
accessors to u64
clocksource/arm_arch_timer: Move system register timer programming
over to CVAL
clocksource/arm_arch_timer: Move drop _tval from erratum function
names
clocksource/arm_arch_timer: Fix MMIO base address vs callback ordering
issue
clocksource/arm_arch_timer: Move MMIO timer programming over to CVAL clocksource/arm_arch_timer: Advertise 56bit timer to the core code clocksource/arm_arch_timer: Work around broken CVAL implementations clocksource/arm_arch_timer: Remove any trace of the TVAL programming
interface
clocksource/arm_arch_timer: Drop unnecessary ISB on CVAL programming clocksource/arch_arm_timer: Move workaround synchronisation around
arm64: Add a capability for FEAT_ECV
arm64: Add CNT{P,V}CTSS_EL0 alternatives to cnt{p,v}ct_el0
arm64: Add handling of CNTVCTSS traps

Oliver Upton (1):
clocksource/arm_arch_timer: Fix masking for high freq counters

 arch/arm/include/asm/arch_timer.h    |  37 ++--
 arch/arm64/include/asm/arch_timer.h  |  78 +++++----
 arch/arm64/include/asm/esr.h         |   6 +
 arch/arm64/include/asm/sysreg.h      |   3 +
 arch/arm64/kernel/cpufeature.c       |  10 ++
 arch/arm64/kernel/traps.c            |  11 ++
 arch/arm64/tools/cpucaps             |   1 +
 drivers/clocksource/arm_arch_timer.c | 243 ++++++++++++++++-----------
 include/clocksource/arm_arch_timer.h |   2 +-
 9 files changed, 245 insertions(+), 146 deletions(-)

Signature · 1 line

2.30.2

Marc Zyngier

author maintainer

2021-09-22 21:19 UTC #1 lore raw

PATCH v2 1/16 clocksource/arm_arch_timer: Add build-time guards for unhandled register accesses

As we are about to change the registers that are used by the driver, start by adding build-time checks to ensure that we always handle all registers and access modes.

Signed-off-by Marc Zyngier <[email protected]>

---
 arch/arm/include/asm/arch_timer.h    | 12 ++++++++++++
 arch/arm64/include/asm/arch_timer.h  | 13 ++++++++++++-
 drivers/clocksource/arm_arch_timer.c |  8 ++++++++
 3 files changed, 32 insertions(+), 1 deletion(-)

arch/arm/include/asm/arch_timer.h +12 -0

index 99175812d903..a5d27cff28fa 100644
@@ -34,6 +34,8 @@ void arch_timer_reg_write_cp15(int access, enum arch_timer_reg reg, u32 val)
 		case ARCH_TIMER_REG_TVAL:
 			asm volatile("mcr p15, 0, %0, c14, c2, 0" : : "r" (val));
 			break;
+		default:
+			BUILD_BUG();
 		}
 	} else if (access == ARCH_TIMER_VIRT_ACCESS) {
 		switch (reg) {

@@ -43,7 +45,11 @@ void arch_timer_reg_write_cp15(int access, enum arch_timer_reg reg, u32 val)
 		case ARCH_TIMER_REG_TVAL:
 			asm volatile("mcr p15, 0, %0, c14, c3, 0" : : "r" (val));
 			break;
+		default:
+			BUILD_BUG();
 		}
+	} else {
+		BUILD_BUG();
 	}
 
 	isb();

@@ -62,6 +68,8 @@ u32 arch_timer_reg_read_cp15(int access, enum arch_timer_reg reg)
 		case ARCH_TIMER_REG_TVAL:
 			asm volatile("mrc p15, 0, %0, c14, c2, 0" : "=r" (val));
 			break;
+		default:
+			BUILD_BUG();
 		}
 	} else if (access == ARCH_TIMER_VIRT_ACCESS) {
 		switch (reg) {

@@ -71,7 +79,11 @@ u32 arch_timer_reg_read_cp15(int access, enum arch_timer_reg reg)
 		case ARCH_TIMER_REG_TVAL:
 			asm volatile("mrc p15, 0, %0, c14, c3, 0" : "=r" (val));
 			break;
+		default:
+			BUILD_BUG();
 		}
+	} else {
+		BUILD_BUG();
 	}
 
 	return val;

arch/arm64/include/asm/arch_timer.h +12 -1

index 88d20f04c64a..fa12ea4a9812 100644
@@ -112,6 +112,8 @@ void arch_timer_reg_write_cp15(int access, enum arch_timer_reg reg, u32 val)
 		case ARCH_TIMER_REG_TVAL:
 			write_sysreg(val, cntp_tval_el0);
 			break;
+		default:
+			BUILD_BUG();
 		}
 	} else if (access == ARCH_TIMER_VIRT_ACCESS) {
 		switch (reg) {

@@ -121,7 +123,11 @@ void arch_timer_reg_write_cp15(int access, enum arch_timer_reg reg, u32 val)
 		case ARCH_TIMER_REG_TVAL:
 			write_sysreg(val, cntv_tval_el0);
 			break;
+		default:
+			BUILD_BUG();
 		}
+	} else {
+		BUILD_BUG();
 	}
 
 	isb();

@@ -136,6 +142,8 @@ u32 arch_timer_reg_read_cp15(int access, enum arch_timer_reg reg)
 			return read_sysreg(cntp_ctl_el0);
 		case ARCH_TIMER_REG_TVAL:
 			return arch_timer_reg_read_stable(cntp_tval_el0);
+		default:
+			BUILD_BUG();
 		}
 	} else if (access == ARCH_TIMER_VIRT_ACCESS) {
 		switch (reg) {

@@ -143,10 +151,13 @@ u32 arch_timer_reg_read_cp15(int access, enum arch_timer_reg reg)
 			return read_sysreg(cntv_ctl_el0);
 		case ARCH_TIMER_REG_TVAL:
 			return arch_timer_reg_read_stable(cntv_tval_el0);
+		default:
+			BUILD_BUG();
 		}
 	}
 
-	BUG();
+	BUILD_BUG();
+	unreachable();
 }
 
 static inline u32 arch_timer_get_cntfrq(void)

drivers/clocksource/arm_arch_timer.c +8 -0

index be6d741d404c..3b7d46d9db73 100644
@@ -112,6 +112,8 @@ void arch_timer_reg_write(int access, enum arch_timer_reg reg, u32 val,
 		case ARCH_TIMER_REG_TVAL:
 			writel_relaxed(val, timer->base + CNTP_TVAL);
 			break;
+		default:
+			BUILD_BUG();
 		}
 	} else if (access == ARCH_TIMER_MEM_VIRT_ACCESS) {
 		struct arch_timer *timer = to_arch_timer(clk);

@@ -122,6 +124,8 @@ void arch_timer_reg_write(int access, enum arch_timer_reg reg, u32 val,
 		case ARCH_TIMER_REG_TVAL:
 			writel_relaxed(val, timer->base + CNTV_TVAL);
 			break;
+		default:
+			BUILD_BUG();
 		}
 	} else {
 		arch_timer_reg_write_cp15(access, reg, val);

@@ -143,6 +147,8 @@ u32 arch_timer_reg_read(int access, enum arch_timer_reg reg,
 		case ARCH_TIMER_REG_TVAL:
 			val = readl_relaxed(timer->base + CNTP_TVAL);
 			break;
+		default:
+			BUILD_BUG();
 		}
 	} else if (access == ARCH_TIMER_MEM_VIRT_ACCESS) {
 		struct arch_timer *timer = to_arch_timer(clk);

@@ -153,6 +159,8 @@ u32 arch_timer_reg_read(int access, enum arch_timer_reg reg,
 		case ARCH_TIMER_REG_TVAL:
 			val = readl_relaxed(timer->base + CNTV_TVAL);
 			break;
+		default:
+			BUILD_BUG();
 		}
 	} else {
 		val = arch_timer_reg_read_cp15(access, reg);

Signature · 1 line

2.30.2

Marc Zyngier

author maintainer

2021-09-22 21:19 UTC #2 lore raw

PATCH v2 2/16 clocksource/arm_arch_timer: Drop CNT*_TVAL read accessors

In reply to Marc Zyngier

The arch timer driver never reads the various TVAL registers, only writes to them. It is thus pointless to provide accessors for them and to implement errata workarounds.

Drop these read-side accessors, and add a couple of BUG() statements for the time being. These statements will be removed further down the line.

Reviewed-by Oliver Upton <[email protected]>

Reviewed-by Mark Rutland <[email protected]>

Tested-by Mark Rutland <[email protected]>

Signed-off-by Marc Zyngier <[email protected]>

---
 arch/arm/include/asm/arch_timer.h    |  6 ----
 arch/arm64/include/asm/arch_timer.h  | 17 -----------
 drivers/clocksource/arm_arch_timer.c | 44 ----------------------------
 3 files changed, 67 deletions(-)

arch/arm/include/asm/arch_timer.h +0 -6

index a5d27cff28fa..7d757085c61a 100644
@@ -65,9 +65,6 @@ u32 arch_timer_reg_read_cp15(int access, enum arch_timer_reg reg)
 		case ARCH_TIMER_REG_CTRL:
 			asm volatile("mrc p15, 0, %0, c14, c2, 1" : "=r" (val));
 			break;
-		case ARCH_TIMER_REG_TVAL:
-			asm volatile("mrc p15, 0, %0, c14, c2, 0" : "=r" (val));
-			break;
 		default:
 			BUILD_BUG();
 		}

@@ -76,9 +73,6 @@ u32 arch_timer_reg_read_cp15(int access, enum arch_timer_reg reg)
 		case ARCH_TIMER_REG_CTRL:
 			asm volatile("mrc p15, 0, %0, c14, c3, 1" : "=r" (val));
 			break;
-		case ARCH_TIMER_REG_TVAL:
-			asm volatile("mrc p15, 0, %0, c14, c3, 0" : "=r" (val));
-			break;
 		default:
 			BUILD_BUG();
 		}

arch/arm64/include/asm/arch_timer.h +0 ~~-17~~

index fa12ea4a9812..8332fcfb08e8 100644
@@ -52,8 +52,6 @@ struct arch_timer_erratum_workaround {
 	enum arch_timer_erratum_match_type match_type;
 	const void *id;
 	const char *desc;
-	u32 (*read_cntp_tval_el0)(void);
-	u32 (*read_cntv_tval_el0)(void);
 	u64 (*read_cntpct_el0)(void);
 	u64 (*read_cntvct_el0)(void);
 	int (*set_next_event_phys)(unsigned long, struct clock_event_device *);

@@ -64,17 +62,6 @@ struct arch_timer_erratum_workaround {
 DECLARE_PER_CPU(const struct arch_timer_erratum_workaround *,
 		timer_unstable_counter_workaround);
 
-/* inline sysreg accessors that make erratum_handler() work */
-static inline notrace u32 arch_timer_read_cntp_tval_el0(void)
-{
-	return read_sysreg(cntp_tval_el0);
-}
-
-static inline notrace u32 arch_timer_read_cntv_tval_el0(void)
-{
-	return read_sysreg(cntv_tval_el0);
-}
-
 static inline notrace u64 arch_timer_read_cntpct_el0(void)
 {
 	return read_sysreg(cntpct_el0);

@@ -140,8 +127,6 @@ u32 arch_timer_reg_read_cp15(int access, enum arch_timer_reg reg)
 		switch (reg) {
 		case ARCH_TIMER_REG_CTRL:
 			return read_sysreg(cntp_ctl_el0);
-		case ARCH_TIMER_REG_TVAL:
-			return arch_timer_reg_read_stable(cntp_tval_el0);
 		default:
 			BUILD_BUG();
 		}

@@ -149,8 +134,6 @@ u32 arch_timer_reg_read_cp15(int access, enum arch_timer_reg reg)
 		switch (reg) {
 		case ARCH_TIMER_REG_CTRL:
 			return read_sysreg(cntv_ctl_el0);
-		case ARCH_TIMER_REG_TVAL:
-			return arch_timer_reg_read_stable(cntv_tval_el0);
 		default:
 			BUILD_BUG();
 		}

drivers/clocksource/arm_arch_timer.c +0 ~~-44~~

98 lines not on this page · raw message

Signature · 1 line

2.30.2

Marc Zyngier

author maintainer

2021-09-22 21:20 UTC #3 lore raw

PATCH v2 3/16 clocksource/arm_arch_timer: Extend write side of timer register accessors to u64

In reply to Marc Zyngier

The various accessors for the timer sysreg and MMIO registers are currently hardwired to 32bit. However, we are about to introduce the use of the CVAL registers, which require a 64bit access.

Upgrade the write side of the accessors to take a 64bit value (the read side is left untouched as we don't plan to ever read back any of these registers).

No functional change expected.

Reviewed-by Oliver Upton <[email protected]>

Reviewed-by Mark Rutland <[email protected]>

Tested-by Mark Rutland <[email protected]>

Signed-off-by Marc Zyngier <[email protected]>

---
 arch/arm/include/asm/arch_timer.h    | 10 +++++-----
 arch/arm64/include/asm/arch_timer.h  |  2 +-
 drivers/clocksource/arm_arch_timer.c | 10 +++++-----
 3 files changed, 11 insertions(+), 11 deletions(-)

arch/arm/include/asm/arch_timer.h +5 -5

index 7d757085c61a..1482e70da7d3 100644
@@ -24,15 +24,15 @@ int arch_timer_arch_init(void);
  * the code. At least it does so with a recent GCC (4.6.3).
  */
 static __always_inline
-void arch_timer_reg_write_cp15(int access, enum arch_timer_reg reg, u32 val)
+void arch_timer_reg_write_cp15(int access, enum arch_timer_reg reg, u64 val)
 {
 	if (access == ARCH_TIMER_PHYS_ACCESS) {
 		switch (reg) {
 		case ARCH_TIMER_REG_CTRL:
-			asm volatile("mcr p15, 0, %0, c14, c2, 1" : : "r" (val));
+			asm volatile("mcr p15, 0, %0, c14, c2, 1" : : "r" ((u32)val));
 			break;
 		case ARCH_TIMER_REG_TVAL:
-			asm volatile("mcr p15, 0, %0, c14, c2, 0" : : "r" (val));
+			asm volatile("mcr p15, 0, %0, c14, c2, 0" : : "r" ((u32)val));
 			break;
 		default:
 			BUILD_BUG();

@@ -40,10 +40,10 @@ void arch_timer_reg_write_cp15(int access, enum arch_timer_reg reg, u32 val)
 	} else if (access == ARCH_TIMER_VIRT_ACCESS) {
 		switch (reg) {
 		case ARCH_TIMER_REG_CTRL:
-			asm volatile("mcr p15, 0, %0, c14, c3, 1" : : "r" (val));
+			asm volatile("mcr p15, 0, %0, c14, c3, 1" : : "r" ((u32)val));
 			break;
 		case ARCH_TIMER_REG_TVAL:
-			asm volatile("mcr p15, 0, %0, c14, c3, 0" : : "r" (val));
+			asm volatile("mcr p15, 0, %0, c14, c3, 0" : : "r" ((u32)val));
 			break;
 		default:
 			BUILD_BUG();

arch/arm64/include/asm/arch_timer.h +1 -1

index 8332fcfb08e8..43f827b680d0 100644
@@ -89,7 +89,7 @@ static inline notrace u64 arch_timer_read_cntvct_el0(void)
  * the code.
  */
 static __always_inline
-void arch_timer_reg_write_cp15(int access, enum arch_timer_reg reg, u32 val)
+void arch_timer_reg_write_cp15(int access, enum arch_timer_reg reg, u64 val)
 {
 	if (access == ARCH_TIMER_PHYS_ACCESS) {
 		switch (reg) {

drivers/clocksource/arm_arch_timer.c +5 -5

index 67bdc7288f59..a49bcefaa370 100644
@@ -100,17 +100,17 @@ early_param("clocksource.arm_arch_timer.evtstrm", early_evtstrm_cfg);
  */
 
 static __always_inline
-void arch_timer_reg_write(int access, enum arch_timer_reg reg, u32 val,
+void arch_timer_reg_write(int access, enum arch_timer_reg reg, u64 val,
 			  struct clock_event_device *clk)
 {
 	if (access == ARCH_TIMER_MEM_PHYS_ACCESS) {
 		struct arch_timer *timer = to_arch_timer(clk);
 		switch (reg) {
 		case ARCH_TIMER_REG_CTRL:
-			writel_relaxed(val, timer->base + CNTP_CTL);
+			writel_relaxed((u32)val, timer->base + CNTP_CTL);
 			break;
 		case ARCH_TIMER_REG_TVAL:
-			writel_relaxed(val, timer->base + CNTP_TVAL);
+			writel_relaxed((u32)val, timer->base + CNTP_TVAL);
 			break;
 		default:
 			BUILD_BUG();

@@ -119,10 +119,10 @@ void arch_timer_reg_write(int access, enum arch_timer_reg reg, u32 val,
 		struct arch_timer *timer = to_arch_timer(clk);
 		switch (reg) {
 		case ARCH_TIMER_REG_CTRL:
-			writel_relaxed(val, timer->base + CNTV_CTL);
+			writel_relaxed((u32)val, timer->base + CNTV_CTL);
 			break;
 		case ARCH_TIMER_REG_TVAL:
-			writel_relaxed(val, timer->base + CNTV_TVAL);
+			writel_relaxed((u32)val, timer->base + CNTV_TVAL);
 			break;
 		default:
 			BUILD_BUG();

Signature · 1 line

2.30.2

Marc Zyngier

author maintainer

2021-09-22 21:20 UTC #4 lore raw

PATCH v2 4/16 clocksource/arm_arch_timer: Move system register timer programming over to CVAL

In reply to Marc Zyngier

In order to cope better with high frequency counters, move the programming of the timers from the countdown timer (TVAL) over to the comparator (CVAL).

The programming model is slightly different, as we now need to read the current counter value to have an absolute deadline instead of a relative one.

There is a small overhead to this change, which we will address in the following patches.

Reviewed-by Oliver Upton <[email protected]>

Reviewed-by Mark Rutland <[email protected]>

Tested-by Mark Rutland <[email protected]>

Signed-off-by Marc Zyngier <[email protected]>

---
 arch/arm/include/asm/arch_timer.h    |  8 ++++----
 arch/arm64/include/asm/arch_timer.h  | 10 +++++-----
 drivers/clocksource/arm_arch_timer.c | 26 +++++++++++++++++++++++---
 include/clocksource/arm_arch_timer.h |  1 +
 4 files changed, 33 insertions(+), 12 deletions(-)

arch/arm/include/asm/arch_timer.h +4 -4

index 1482e70da7d3..a9b2b721c7f9 100644
@@ -31,8 +31,8 @@ void arch_timer_reg_write_cp15(int access, enum arch_timer_reg reg, u64 val)
 		case ARCH_TIMER_REG_CTRL:
 			asm volatile("mcr p15, 0, %0, c14, c2, 1" : : "r" ((u32)val));
 			break;
-		case ARCH_TIMER_REG_TVAL:
-			asm volatile("mcr p15, 0, %0, c14, c2, 0" : : "r" ((u32)val));
+		case ARCH_TIMER_REG_CVAL:
+			asm volatile("mcrr p15, 2, %Q0, %R0, c14" : : "r" (val));
 			break;
 		default:
 			BUILD_BUG();

@@ -42,8 +42,8 @@ void arch_timer_reg_write_cp15(int access, enum arch_timer_reg reg, u64 val)
 		case ARCH_TIMER_REG_CTRL:
 			asm volatile("mcr p15, 0, %0, c14, c3, 1" : : "r" ((u32)val));
 			break;
-		case ARCH_TIMER_REG_TVAL:
-			asm volatile("mcr p15, 0, %0, c14, c3, 0" : : "r" ((u32)val));
+		case ARCH_TIMER_REG_CVAL:
+			asm volatile("mcrr p15, 3, %Q0, %R0, c14" : : "r" (val));
 			break;
 		default:
 			BUILD_BUG();

arch/arm64/include/asm/arch_timer.h +5 -5

index 43f827b680d0..4f4aa13dd01e 100644
@@ -96,8 +96,8 @@ void arch_timer_reg_write_cp15(int access, enum arch_timer_reg reg, u64 val)
 		case ARCH_TIMER_REG_CTRL:
 			write_sysreg(val, cntp_ctl_el0);
 			break;
-		case ARCH_TIMER_REG_TVAL:
-			write_sysreg(val, cntp_tval_el0);
+		case ARCH_TIMER_REG_CVAL:
+			write_sysreg(val, cntp_cval_el0);
 			break;
 		default:
 			BUILD_BUG();

@@ -107,8 +107,8 @@ void arch_timer_reg_write_cp15(int access, enum arch_timer_reg reg, u64 val)
 		case ARCH_TIMER_REG_CTRL:
 			write_sysreg(val, cntv_ctl_el0);
 			break;
-		case ARCH_TIMER_REG_TVAL:
-			write_sysreg(val, cntv_tval_el0);
+		case ARCH_TIMER_REG_CVAL:
+			write_sysreg(val, cntv_cval_el0);
 			break;
 		default:
 			BUILD_BUG();

@@ -121,7 +121,7 @@ void arch_timer_reg_write_cp15(int access, enum arch_timer_reg reg, u64 val)
 }
 
 static __always_inline
-u32 arch_timer_reg_read_cp15(int access, enum arch_timer_reg reg)
+u64 arch_timer_reg_read_cp15(int access, enum arch_timer_reg reg)
 {
 	if (access == ARCH_TIMER_PHYS_ACCESS) {
 		switch (reg) {

drivers/clocksource/arm_arch_timer.c +23 -3

index a49bcefaa370..322165468edf 100644
@@ -691,10 +691,18 @@ static __always_inline void set_next_event(const int access, unsigned long evt,
 					   struct clock_event_device *clk)
 {
 	unsigned long ctrl;
+	u64 cnt;
+
 	ctrl = arch_timer_reg_read(access, ARCH_TIMER_REG_CTRL, clk);
 	ctrl |= ARCH_TIMER_CTRL_ENABLE;
 	ctrl &= ~ARCH_TIMER_CTRL_IT_MASK;
-	arch_timer_reg_write(access, ARCH_TIMER_REG_TVAL, evt, clk);
+
+	if (access == ARCH_TIMER_PHYS_ACCESS)
+		cnt = __arch_counter_get_cntpct();
+	else
+		cnt = __arch_counter_get_cntvct();
+
+	arch_timer_reg_write(access, ARCH_TIMER_REG_CVAL, evt + cnt, clk);
 	arch_timer_reg_write(access, ARCH_TIMER_REG_CTRL, ctrl, clk);
 }

@@ -712,17 +720,29 @@ static int arch_timer_set_next_event_phys(unsigned long evt,
 	return 0;
 }
 
+static __always_inline void set_next_event_mem(const int access, unsigned long evt,
+					   struct clock_event_device *clk)
+{
+	unsigned long ctrl;
+	ctrl = arch_timer_reg_read(access, ARCH_TIMER_REG_CTRL, clk);
+	ctrl |= ARCH_TIMER_CTRL_ENABLE;
+	ctrl &= ~ARCH_TIMER_CTRL_IT_MASK;
+
+	arch_timer_reg_write(access, ARCH_TIMER_REG_TVAL, evt, clk);
+	arch_timer_reg_write(access, ARCH_TIMER_REG_CTRL, ctrl, clk);
+}
+
 static int arch_timer_set_next_event_virt_mem(unsigned long evt,
 					      struct clock_event_device *clk)
 {
-	set_next_event(ARCH_TIMER_MEM_VIRT_ACCESS, evt, clk);
+	set_next_event_mem(ARCH_TIMER_MEM_VIRT_ACCESS, evt, clk);
 	return 0;
 }
 
 static int arch_timer_set_next_event_phys_mem(unsigned long evt,
 					      struct clock_event_device *clk)
 {
-	set_next_event(ARCH_TIMER_MEM_PHYS_ACCESS, evt, clk);
+	set_next_event_mem(ARCH_TIMER_MEM_PHYS_ACCESS, evt, clk);
 	return 0;
 }

include/clocksource/arm_arch_timer.h +1 -0

index 73c7139c866f..d59537afb29d 100644
@@ -25,6 +25,7 @@
 enum arch_timer_reg {
 	ARCH_TIMER_REG_CTRL,
 	ARCH_TIMER_REG_TVAL,
+	ARCH_TIMER_REG_CVAL,
 };
 
 enum arch_timer_ppi_nr {

Signature · 1 line

2.30.2

Marc Zyngier

author maintainer

2021-09-22 21:20 UTC #5 lore raw

PATCH v2 5/16 clocksource/arm_arch_timer: Move drop _tval from erratum function names

In reply to Marc Zyngier

The '_tval' name in the erratum handling function names doesn't make much sense anymore (and they were using CVAL the first place).

Drop the _tval tag.

Reviewed-by Oliver Upton <[email protected]>

Reviewed-by Mark Rutland <[email protected]>

Tested-by Mark Rutland <[email protected]>

Signed-off-by Marc Zyngier <[email protected]>

---
 drivers/clocksource/arm_arch_timer.c | 26 +++++++++++++-------------
 1 file changed, 13 insertions(+), 13 deletions(-)

drivers/clocksource/arm_arch_timer.c +13 ~~-13~~

index 322165468edf..8afe8c814eba 100644
@@ -369,7 +369,7 @@ EXPORT_SYMBOL_GPL(timer_unstable_counter_workaround);
 
 static atomic_t timer_unstable_counter_workaround_in_use = ATOMIC_INIT(0);
 
-static void erratum_set_next_event_tval_generic(const int access, unsigned long evt,
+static void erratum_set_next_event_generic(const int access, unsigned long evt,
 						struct clock_event_device *clk)
 {
 	unsigned long ctrl;

@@ -390,17 +390,17 @@ static void erratum_set_next_event_tval_generic(const int access, unsigned long
 	arch_timer_reg_write(access, ARCH_TIMER_REG_CTRL, ctrl, clk);
 }
 
-static __maybe_unused int erratum_set_next_event_tval_virt(unsigned long evt,
+static __maybe_unused int erratum_set_next_event_virt(unsigned long evt,
 					    struct clock_event_device *clk)
 {
-	erratum_set_next_event_tval_generic(ARCH_TIMER_VIRT_ACCESS, evt, clk);
+	erratum_set_next_event_generic(ARCH_TIMER_VIRT_ACCESS, evt, clk);
 	return 0;
 }
 
-static __maybe_unused int erratum_set_next_event_tval_phys(unsigned long evt,
+static __maybe_unused int erratum_set_next_event_phys(unsigned long evt,
 					    struct clock_event_device *clk)
 {
-	erratum_set_next_event_tval_generic(ARCH_TIMER_PHYS_ACCESS, evt, clk);
+	erratum_set_next_event_generic(ARCH_TIMER_PHYS_ACCESS, evt, clk);
 	return 0;
 }

@@ -412,8 +412,8 @@ static const struct arch_timer_erratum_workaround ool_workarounds[] = {
 		.desc = "Freescale erratum a005858",
 		.read_cntpct_el0 = fsl_a008585_read_cntpct_el0,
 		.read_cntvct_el0 = fsl_a008585_read_cntvct_el0,
-		.set_next_event_phys = erratum_set_next_event_tval_phys,
-		.set_next_event_virt = erratum_set_next_event_tval_virt,
+		.set_next_event_phys = erratum_set_next_event_phys,
+		.set_next_event_virt = erratum_set_next_event_virt,
 	},
 #endif
 #ifdef CONFIG_HISILICON_ERRATUM_161010101

@@ -423,8 +423,8 @@ static const struct arch_timer_erratum_workaround ool_workarounds[] = {
 		.desc = "HiSilicon erratum 161010101",
 		.read_cntpct_el0 = hisi_161010101_read_cntpct_el0,
 		.read_cntvct_el0 = hisi_161010101_read_cntvct_el0,
-		.set_next_event_phys = erratum_set_next_event_tval_phys,
-		.set_next_event_virt = erratum_set_next_event_tval_virt,
+		.set_next_event_phys = erratum_set_next_event_phys,
+		.set_next_event_virt = erratum_set_next_event_virt,
 	},
 	{
 		.match_type = ate_match_acpi_oem_info,

@@ -432,8 +432,8 @@ static const struct arch_timer_erratum_workaround ool_workarounds[] = {
 		.desc = "HiSilicon erratum 161010101",
 		.read_cntpct_el0 = hisi_161010101_read_cntpct_el0,
 		.read_cntvct_el0 = hisi_161010101_read_cntvct_el0,
-		.set_next_event_phys = erratum_set_next_event_tval_phys,
-		.set_next_event_virt = erratum_set_next_event_tval_virt,
+		.set_next_event_phys = erratum_set_next_event_phys,
+		.set_next_event_virt = erratum_set_next_event_virt,
 	},
 #endif
 #ifdef CONFIG_ARM64_ERRATUM_858921

@@ -452,8 +452,8 @@ static const struct arch_timer_erratum_workaround ool_workarounds[] = {
 		.desc = "Allwinner erratum UNKNOWN1",
 		.read_cntpct_el0 = sun50i_a64_read_cntpct_el0,
 		.read_cntvct_el0 = sun50i_a64_read_cntvct_el0,
-		.set_next_event_phys = erratum_set_next_event_tval_phys,
-		.set_next_event_virt = erratum_set_next_event_tval_virt,
+		.set_next_event_phys = erratum_set_next_event_phys,
+		.set_next_event_virt = erratum_set_next_event_virt,
 	},
 #endif
 #ifdef CONFIG_ARM64_ERRATUM_1418040

Signature · 1 line

2.30.2

Marc Zyngier

author maintainer

2021-09-22 21:20 UTC #6 lore raw

PATCH v2 6/16 clocksource/arm_arch_timer: Fix MMIO base address vs callback ordering issue

In reply to Marc Zyngier

The MMIO timer base address gets published after we have registered the callbacks and the interrupt handler, which is... a bit dangerous.

Fix this by moving the base address publication to the point where we register the timer, and expose a pointer to the timer structure itself rather than a naked value.

Reviewed-by Oliver Upton <[email protected]>

Reviewed-by Mark Rutland <[email protected]>

Signed-off-by Marc Zyngier <[email protected]>

---
 drivers/clocksource/arm_arch_timer.c | 27 +++++++++++++--------------
 1 file changed, 13 insertions(+), 14 deletions(-)

drivers/clocksource/arm_arch_timer.c +13 ~~-14~~

index 8afe8c814eba..bede10f67f9a 100644
@@ -54,13 +54,13 @@
 
 static unsigned arch_timers_present __initdata;
 
-static void __iomem *arch_counter_base __ro_after_init;
-
 struct arch_timer {
 	void __iomem *base;
 	struct clock_event_device evt;
 };
 
+static struct arch_timer *arch_timer_mem __ro_after_init;
+
 #define to_arch_timer(e) container_of(e, struct arch_timer, evt)
 
 static u32 arch_timer_rate __ro_after_init;

@@ -973,9 +973,9 @@ static u64 arch_counter_get_cntvct_mem(void)
 	u32 vct_lo, vct_hi, tmp_hi;
 
 	do {
-		vct_hi = readl_relaxed(arch_counter_base + CNTVCT_HI);
-		vct_lo = readl_relaxed(arch_counter_base + CNTVCT_LO);
-		tmp_hi = readl_relaxed(arch_counter_base + CNTVCT_HI);
+		vct_hi = readl_relaxed(arch_timer_mem->base + CNTVCT_HI);
+		vct_lo = readl_relaxed(arch_timer_mem->base + CNTVCT_LO);
+		tmp_hi = readl_relaxed(arch_timer_mem->base + CNTVCT_HI);
 	} while (vct_hi != tmp_hi);
 
 	return ((u64) vct_hi << 32) | vct_lo;

@@ -1166,25 +1166,25 @@ static int __init arch_timer_mem_register(void __iomem *base, unsigned int irq)
 {
 	int ret;
 	irq_handler_t func;
-	struct arch_timer *t;
 
-	t = kzalloc(sizeof(*t), GFP_KERNEL);
-	if (!t)
+	arch_timer_mem = kzalloc(sizeof(*arch_timer_mem), GFP_KERNEL);
+	if (!arch_timer_mem)
 		return -ENOMEM;
 
-	t->base = base;
-	t->evt.irq = irq;
-	__arch_timer_setup(ARCH_TIMER_TYPE_MEM, &t->evt);
+	arch_timer_mem->base = base;
+	arch_timer_mem->evt.irq = irq;
+	__arch_timer_setup(ARCH_TIMER_TYPE_MEM, &arch_timer_mem->evt);
 
 	if (arch_timer_mem_use_virtual)
 		func = arch_timer_handler_virt_mem;
 	else
 		func = arch_timer_handler_phys_mem;
 
-	ret = request_irq(irq, func, IRQF_TIMER, "arch_mem_timer", &t->evt);
+	ret = request_irq(irq, func, IRQF_TIMER, "arch_mem_timer", &arch_timer_mem->evt);
 	if (ret) {
 		pr_err("Failed to request mem timer irq\n");
-		kfree(t);
+		kfree(arch_timer_mem);
+		arch_timer_mem = NULL;
 	}
 
 	return ret;

@@ -1442,7 +1442,6 @@ arch_timer_mem_frame_register(struct arch_timer_mem_frame *frame)
 		return ret;
 	}
 
-	arch_counter_base = base;
 	arch_timers_present |= ARCH_TIMER_TYPE_MEM;
 
 	return 0;

Signature · 1 line

2.30.2

Marc Zyngier

author maintainer

2021-09-22 21:20 UTC #7 lore raw

PATCH v2 7/16 clocksource/arm_arch_timer: Move MMIO timer programming over to CVAL

In reply to Marc Zyngier

Similarily to the sysreg-based timer, move the MMIO over to using the CVAL registers instead of TVAL. Note that there is no warranty that the 64bit MMIO access will be atomic, but the timer is always disabled at the point where we program CVAL.

Signed-off-by Marc Zyngier <[email protected]>

---
 arch/arm/include/asm/arch_timer.h    |  1 +
 drivers/clocksource/arm_arch_timer.c | 50 +++++++++++++++++++++-------
 2 files changed, 39 insertions(+), 12 deletions(-)

arch/arm/include/asm/arch_timer.h +1 -0

index a9b2b721c7f9..9f4b895b78f7 100644
@@ -7,6 +7,7 @@
 #include <asm/hwcap.h>
 #include <linux/clocksource.h>
 #include <linux/init.h>
+#include <linux/io-64-nonatomic-lo-hi.h>
 #include <linux/types.h>
 
 #include <clocksource/arm_arch_timer.h>

drivers/clocksource/arm_arch_timer.c +38 ~~-12~~

91 lines not on this page · raw message

Signature · 1 line

2.30.2

Marc Zyngier

author maintainer

2021-09-22 21:20 UTC #8 lore raw

PATCH v2 8/16 clocksource/arm_arch_timer: Advertise 56bit timer to the core code

In reply to Marc Zyngier

Proudly tell the code code that we have a timer able to handle 56 bits deltas.

Signed-off-by Marc Zyngier <[email protected]>

---
 drivers/clocksource/arm_arch_timer.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

drivers/clocksource/arm_arch_timer.c +1 -1

index f4db3a65bc79..36e091412151 100644
@@ -834,7 +834,7 @@ static void __arch_timer_setup(unsigned type,
 
 	clk->set_state_shutdown(clk);
 
-	clockevents_config_and_register(clk, arch_timer_rate, 0xf, 0x7fffffff);
+	clockevents_config_and_register(clk, arch_timer_rate, 0xf, CLOCKSOURCE_MASK(56));
 }
 
 static void arch_timer_evtstrm_enable(int divider)

Signature · 1 line

2.30.2

Marc Zyngier

author maintainer

2021-09-22 21:20 UTC #9 lore raw

PATCH v2 9/16 clocksource/arm_arch_timer: Work around broken CVAL implementations

In reply to Marc Zyngier

The Applied Micro XGene-1 SoC has a busted implementation of the CVAL register: it looks like it is based on TVAL instead of the other way around. The net effect of this implementation blunder is that the maximum deadline you can program in the timer is 32bit wide.

Use a MIDR check to notice the broken CPU, and reduce the width of the timer to 32bit.

Signed-off-by Marc Zyngier <[email protected]>

---
 drivers/clocksource/arm_arch_timer.c | 28 +++++++++++++++++++++++++++-
 1 file changed, 27 insertions(+), 1 deletion(-)

drivers/clocksource/arm_arch_timer.c +27 -1

index 36e091412151..ef3f83865dcd 100644
@@ -780,9 +780,32 @@ static int arch_timer_set_next_event_phys_mem(unsigned long evt,
 	return 0;
 }
 
+static u64 __arch_timer_check_delta(void)
+{
+#ifdef CONFIG_ARM64
+	const struct midr_range broken_cval_midrs[] = {
+		/*
+		 * XGene-1 implements CVAL in terms of TVAL, meaning
+		 * that the maximum timer range is 32bit. Shame on them.
+		 */
+		MIDR_ALL_VERSIONS(MIDR_CPU_MODEL(ARM_CPU_IMP_APM,
+						 APM_CPU_PART_POTENZA)),
+		{},
+	};
+
+	if (is_midr_in_range_list(read_cpuid_id(), broken_cval_midrs)) {
+		pr_warn_once("Broken CNTx_CVAL_EL1, limiting width to 32bits");
+		return CLOCKSOURCE_MASK(32);
+	}
+#endif
+	return CLOCKSOURCE_MASK(56);
+}
+
 static void __arch_timer_setup(unsigned type,
 			       struct clock_event_device *clk)
 {
+	u64 max_delta;
+
 	clk->features = CLOCK_EVT_FEAT_ONESHOT;
 
 	if (type == ARCH_TIMER_TYPE_CP15) {

@@ -814,6 +837,7 @@ static void __arch_timer_setup(unsigned type,
 		}
 
 		clk->set_next_event = sne;
+		max_delta = __arch_timer_check_delta();
 	} else {
 		clk->features |= CLOCK_EVT_FEAT_DYNIRQ;
 		clk->name = "arch_mem_timer";

@@ -830,11 +854,13 @@ static void __arch_timer_setup(unsigned type,
 			clk->set_next_event =
 				arch_timer_set_next_event_phys_mem;
 		}
+
+		max_delta = CLOCKSOURCE_MASK(56);
 	}
 
 	clk->set_state_shutdown(clk);
 
-	clockevents_config_and_register(clk, arch_timer_rate, 0xf, CLOCKSOURCE_MASK(56));
+	clockevents_config_and_register(clk, arch_timer_rate, 0xf, max_delta);
 }
 
 static void arch_timer_evtstrm_enable(int divider)

Signature · 1 line

2.30.2

Marc Zyngier

author maintainer

2021-09-22 21:27 UTC #10 lore raw

PATCH v2 10/16 clocksource/arm_arch_timer: Remove any trace of the TVAL programming interface

In reply to Marc Zyngier

TVAL usage is now long gone, get rid of the leftovers.

Signed-off-by Marc Zyngier <[email protected]>

---
 drivers/clocksource/arm_arch_timer.c | 8 --------
 include/clocksource/arm_arch_timer.h | 1 -
 2 files changed, 9 deletions(-)

drivers/clocksource/arm_arch_timer.c +0 -8

index ef3f83865dcd..6e20bc12dc35 100644
@@ -48,10 +48,8 @@
 #define CNTPCT_LO	0x08
 #define CNTFRQ		0x10
 #define CNTP_CVAL_LO	0x20
-#define CNTP_TVAL	0x28
 #define CNTP_CTL	0x2c
 #define CNTV_CVAL_LO	0x30
-#define CNTV_TVAL	0x38
 #define CNTV_CTL	0x3c
 
 static unsigned arch_timers_present __initdata;

@@ -111,9 +109,6 @@ void arch_timer_reg_write(int access, enum arch_timer_reg reg, u64 val,
 		case ARCH_TIMER_REG_CTRL:
 			writel_relaxed((u32)val, timer->base + CNTP_CTL);
 			break;
-		case ARCH_TIMER_REG_TVAL:
-			writel_relaxed((u32)val, timer->base + CNTP_TVAL);
-			break;
 		case ARCH_TIMER_REG_CVAL:
 			/*
 			 * Not guaranteed to be atomic, so the timer

@@ -130,9 +125,6 @@ void arch_timer_reg_write(int access, enum arch_timer_reg reg, u64 val,
 		case ARCH_TIMER_REG_CTRL:
 			writel_relaxed((u32)val, timer->base + CNTV_CTL);
 			break;
-		case ARCH_TIMER_REG_TVAL:
-			writel_relaxed((u32)val, timer->base + CNTV_TVAL);
-			break;
 		case ARCH_TIMER_REG_CVAL:
 			/* Same restriction as above */
 			writeq_relaxed(val, timer->base + CNTV_CVAL_LO);

include/clocksource/arm_arch_timer.h +0 -1

index d59537afb29d..e715bdb720d5 100644
@@ -24,7 +24,6 @@
 
 enum arch_timer_reg {
 	ARCH_TIMER_REG_CTRL,
-	ARCH_TIMER_REG_TVAL,
 	ARCH_TIMER_REG_CVAL,
 };

Signature · 1 line

2.30.2

Marc Zyngier

author maintainer

2021-09-22 21:27 UTC #11 lore raw

PATCH v2 11/16 clocksource/arm_arch_timer: Drop unnecessary ISB on CVAL programming

In reply to Marc Zyngier

Switching from TVAL to CVAL has a small drawback: we need an ISB before reading the counter. We cannot get rid of it, but we can instead remove the one that comes just after writing to CVAL.

This reduces the number of ISBs from 3 to 2 when programming the timer.

Signed-off-by Marc Zyngier <[email protected]>

---
 arch/arm/include/asm/arch_timer.h   | 4 ++--
 arch/arm64/include/asm/arch_timer.h | 4 ++--
 2 files changed, 4 insertions(+), 4 deletions(-)

arch/arm/include/asm/arch_timer.h +2 -2

index 9f4b895b78f7..bb129b6d2366 100644
@@ -31,6 +31,7 @@ void arch_timer_reg_write_cp15(int access, enum arch_timer_reg reg, u64 val)
 		switch (reg) {
 		case ARCH_TIMER_REG_CTRL:
 			asm volatile("mcr p15, 0, %0, c14, c2, 1" : : "r" ((u32)val));
+			isb();
 			break;
 		case ARCH_TIMER_REG_CVAL:
 			asm volatile("mcrr p15, 2, %Q0, %R0, c14" : : "r" (val));

@@ -42,6 +43,7 @@ void arch_timer_reg_write_cp15(int access, enum arch_timer_reg reg, u64 val)
 		switch (reg) {
 		case ARCH_TIMER_REG_CTRL:
 			asm volatile("mcr p15, 0, %0, c14, c3, 1" : : "r" ((u32)val));
+			isb();
 			break;
 		case ARCH_TIMER_REG_CVAL:
 			asm volatile("mcrr p15, 3, %Q0, %R0, c14" : : "r" (val));

@@ -52,8 +54,6 @@ void arch_timer_reg_write_cp15(int access, enum arch_timer_reg reg, u64 val)
 	} else {
 		BUILD_BUG();
 	}
-
-	isb();
 }
 
 static __always_inline

arch/arm64/include/asm/arch_timer.h +2 -2

index 4f4aa13dd01e..b8000ef71a2c 100644
@@ -95,6 +95,7 @@ void arch_timer_reg_write_cp15(int access, enum arch_timer_reg reg, u64 val)
 		switch (reg) {
 		case ARCH_TIMER_REG_CTRL:
 			write_sysreg(val, cntp_ctl_el0);
+			isb();
 			break;
 		case ARCH_TIMER_REG_CVAL:
 			write_sysreg(val, cntp_cval_el0);

@@ -106,6 +107,7 @@ void arch_timer_reg_write_cp15(int access, enum arch_timer_reg reg, u64 val)
 		switch (reg) {
 		case ARCH_TIMER_REG_CTRL:
 			write_sysreg(val, cntv_ctl_el0);
+			isb();
 			break;
 		case ARCH_TIMER_REG_CVAL:
 			write_sysreg(val, cntv_cval_el0);

@@ -116,8 +118,6 @@ void arch_timer_reg_write_cp15(int access, enum arch_timer_reg reg, u64 val)
 	} else {
 		BUILD_BUG();
 	}
-
-	isb();
 }
 
 static __always_inline

Signature · 1 line

2.30.2

Marc Zyngier

author maintainer

2021-09-22 21:26 UTC #12 lore raw

PATCH v2 12/16 clocksource/arm_arch_timer: Fix masking for high freq counters

In reply to Marc Zyngier

From: Oliver Upton <[email protected]>

Unfortunately, the architecture provides no means to determine the bit width of the system counter. However, we do know the following from the specification:

- the system counter is at least 56 bits wide
- Roll-over time of not less than 40 years

To date, the arch timer driver has depended on the first property, assuming any system counter to be 56 bits wide and masking off the rest. However, combining a narrow clocksource mask with a high frequency counter could result in prematurely wrapping the system counter by a significant margin. For example, a 56 bit wide, 1GHz system counter would wrap in a mere 2.28 years!

This is a problem for two reasons: v8.6+ implementations are required to provide a 64 bit, 1GHz system counter. Furthermore, before v8.6, implementers may select a counter frequency of their choosing.

Fix the issue by deriving a valid clock mask based on the second property from above. Set the floor at 56 bits, since we know no system counter is narrower than that.

Signed-off-by Oliver Upton <[email protected]>

Reviewed-by Linus Walleij <[email protected]>

[maz: fixed width computation not to lose the last bit, added
max delta generation for the timer]

Signed-off-by Marc Zyngier <[email protected]>

Link https://lore.kernel.org/r/[email protected] open here

---
 drivers/clocksource/arm_arch_timer.c | 34 ++++++++++++++++++++++++----
 1 file changed, 29 insertions(+), 5 deletions(-)

drivers/clocksource/arm_arch_timer.c +29 -5

index 6e20bc12dc35..9a04eacc4412 100644
@@ -52,6 +52,12 @@
 #define CNTV_CVAL_LO	0x30
 #define CNTV_CTL	0x3c
 
+/*
+ * The minimum amount of time a generic counter is guaranteed to not roll over
+ * (40 years)
+ */
+#define MIN_ROLLOVER_SECS	(40ULL * 365 * 24 * 3600)
+
 static unsigned arch_timers_present __initdata;
 
 struct arch_timer {

@@ -95,6 +101,22 @@ static int __init early_evtstrm_cfg(char *buf)
 }
 early_param("clocksource.arm_arch_timer.evtstrm", early_evtstrm_cfg);
 
+/*
+ * Makes an educated guess at a valid counter width based on the Generic Timer
+ * specification. Of note:
+ *   1) the system counter is at least 56 bits wide
+ *   2) a roll-over time of not less than 40 years
+ *
+ * See 'ARM DDI 0487G.a D11.1.2 ("The system counter")' for more details.
+ */
+static int arch_counter_get_width(void)
+{
+	u64 min_cycles = MIN_ROLLOVER_SECS * arch_timer_rate;
+
+	/* guarantee the returned width is within the valid range */
+	return clamp_val(ilog2(min_cycles - 1) + 1, 56, 64);
+}
+
 /*
  * Architected system timer support.
  */

@@ -212,13 +234,11 @@ static struct clocksource clocksource_counter = {
 	.id	= CSID_ARM_ARCH_COUNTER,
 	.rating	= 400,
 	.read	= arch_counter_read,
-	.mask	= CLOCKSOURCE_MASK(56),
 	.flags	= CLOCK_SOURCE_IS_CONTINUOUS,
 };
 
 static struct cyclecounter cyclecounter __ro_after_init = {
 	.read	= arch_counter_read_cc,
-	.mask	= CLOCKSOURCE_MASK(56),
 };
 
 struct ate_acpi_oem_info {

@@ -790,7 +810,7 @@ static u64 __arch_timer_check_delta(void)
 		return CLOCKSOURCE_MASK(32);
 	}
 #endif
-	return CLOCKSOURCE_MASK(56);
+	return CLOCKSOURCE_MASK(arch_counter_get_width());
 }
 
 static void __arch_timer_setup(unsigned type,

@@ -1035,6 +1055,7 @@ struct arch_timer_kvm_info *arch_timer_get_kvm_info(void)
 static void __init arch_counter_register(unsigned type)
 {
 	u64 start_count;
+	int width;
 
 	/* Register the CP15 based counter if we have one */
 	if (type & ARCH_TIMER_TYPE_CP15) {

@@ -1059,6 +1080,10 @@ static void __init arch_counter_register(unsigned type)
 		arch_timer_read_counter = arch_counter_get_cntvct_mem;
 	}
 
+	width = arch_counter_get_width();
+	clocksource_counter.mask = CLOCKSOURCE_MASK(width);
+	cyclecounter.mask = CLOCKSOURCE_MASK(width);
+
 	if (!arch_counter_suspend_stop)
 		clocksource_counter.flags |= CLOCK_SOURCE_SUSPEND_NONSTOP;
 	start_count = arch_timer_read_counter();

@@ -1068,8 +1093,7 @@ static void __init arch_counter_register(unsigned type)
 	timecounter_init(&arch_timer_kvm_info.timecounter,
 			 &cyclecounter, start_count);
 
-	/* 56 bits minimum, so we assume worst case rollover */
-	sched_clock_register(arch_timer_read_counter, 56, arch_timer_rate);
+	sched_clock_register(arch_timer_read_counter, width, arch_timer_rate);
 }
 
 static void arch_timer_stop(struct clock_event_device *clk)

Signature · 1 line

2.30.2

Marc Zyngier

author maintainer

2021-09-22 21:27 UTC #13 lore raw

PATCH v2 13/16 clocksource/arch_arm_timer: Move workaround synchronisation around

In reply to Marc Zyngier

We currently handle synchronisation when workarounds are enabled by having an ISB in the __arch_counter_get_cnt?ct_stable() helpers.

WHile this works, this prevents us from relaxing this synchronisation.

Instead, move it closer to the point where the synchronisation is actually needed. Further patches will subsequently relax this.

Signed-off-by Marc Zyngier <[email protected]>

---
 arch/arm64/include/asm/arch_timer.h | 6 +++---
 1 file changed, 3 insertions(+), 3 deletions(-)

arch/arm64/include/asm/arch_timer.h +3 -3

index b8000ef71a2c..519ac1f7f859 100644
@@ -32,7 +32,7 @@
 	({								\
 		const struct arch_timer_erratum_workaround *__wa;	\
 		__wa = __this_cpu_read(timer_unstable_counter_workaround); \
-		(__wa && __wa->h) ? __wa->h : arch_timer_##h;		\
+		(__wa && __wa->h) ? ({ isb(); __wa->h;}) : arch_timer_##h; \
 	})
 
 #else

@@ -64,11 +64,13 @@ DECLARE_PER_CPU(const struct arch_timer_erratum_workaround *,
 
 static inline notrace u64 arch_timer_read_cntpct_el0(void)
 {
+	isb();
 	return read_sysreg(cntpct_el0);
 }
 
 static inline notrace u64 arch_timer_read_cntvct_el0(void)
 {
+	isb();
 	return read_sysreg(cntvct_el0);
 }

@@ -163,7 +165,6 @@ static __always_inline u64 __arch_counter_get_cntpct_stable(void)
 {
 	u64 cnt;
 
-	isb();
 	cnt = arch_timer_reg_read_stable(cntpct_el0);
 	arch_counter_enforce_ordering(cnt);
 	return cnt;

@@ -183,7 +184,6 @@ static __always_inline u64 __arch_counter_get_cntvct_stable(void)
 {
 	u64 cnt;
 
-	isb();
 	cnt = arch_timer_reg_read_stable(cntvct_el0);
 	arch_counter_enforce_ordering(cnt);
 	return cnt;

Signature · 1 line

2.30.2

Marc Zyngier

author maintainer

2021-09-22 21:26 UTC #14 lore raw

PATCH v2 14/16 arm64: Add a capability for FEAT_ECV

In reply to Marc Zyngier

Add a new capability to detect the Enhanced Counter Virtualization feature (FEAT_ECV).

Reviewed-by Oliver Upton <[email protected]>

Signed-off-by Marc Zyngier <[email protected]>

---
 arch/arm64/kernel/cpufeature.c | 10 ++++++++++
 arch/arm64/tools/cpucaps       |  1 +
 2 files changed, 11 insertions(+)

arch/arm64/kernel/cpufeature.c +10 -0

index f8a3067d10c6..26b11ce8fff6 100644
@@ -1926,6 +1926,16 @@ static const struct arm64_cpu_capabilities arm64_features[] = {
 		.sign = FTR_UNSIGNED,
 		.min_field_value = 1,
 	},
+	{
+		.desc = "Enhanced Counter Virtualization",
+		.capability = ARM64_HAS_ECV,
+		.type = ARM64_CPUCAP_SYSTEM_FEATURE,
+		.matches = has_cpuid_feature,
+		.sys_reg = SYS_ID_AA64MMFR0_EL1,
+		.field_pos = ID_AA64MMFR0_ECV_SHIFT,
+		.sign = FTR_UNSIGNED,
+		.min_field_value = 1,
+	},
 #ifdef CONFIG_ARM64_PAN
 	{
 		.desc = "Privileged Access Never",

arch/arm64/tools/cpucaps +1 -0

index 49305c2e6dfd..7a7c58acd8f0 100644
@@ -18,6 +18,7 @@ HAS_CRC32
 HAS_DCPODP
 HAS_DCPOP
 HAS_E0PD
+HAS_ECV
 HAS_EPAN
 HAS_GENERIC_AUTH
 HAS_GENERIC_AUTH_ARCH

Signature · 1 line

2.30.2

Will Deacon

maintainer

2021-09-29 16:04 UTC #15 lore raw

Re: [PATCH v2 14/16] arm64: Add a capability for FEAT_ECV

Quoted 18 lines · Marc Zyngier

Add a new capability to detect the Enhanced Counter Virtualization feature (FEAT_ECV).
Reviewed-by Oliver Upton <[email protected]>
Signed-off-by Marc Zyngier <[email protected]>
---
 arch/arm64/kernel/cpufeature.c | 10 ++++++++++
 arch/arm64/tools/cpucaps       |  1 +
 2 files changed, 11 insertions(+)
arch/arm64/kernel/cpufeature.c +0 -0
index f8a3067d10c6..26b11ce8fff6 100644
@@ -1926,6 +1926,16 @@ static const struct arm64_cpu_capabilities arm64_features[] = {
 		.sign = FTR_UNSIGNED,
 		.min_field_value = 1,
 	},

arch/arm64/kernel/cpufeature.c static const struct arm64_cpu_capabilities arm64_features[] = { · line 1938

Quoted 10 lines

arch/arm64/kernel/cpufeature.c +10 -0

+	{
+		.desc = "Enhanced Counter Virtualization",
+		.capability = ARM64_HAS_ECV,
+		.type = ARM64_CPUCAP_SYSTEM_FEATURE,
+		.matches = has_cpuid_feature,
+		.sys_reg = SYS_ID_AA64MMFR0_EL1,
+		.field_pos = ID_AA64MMFR0_ECV_SHIFT,
+		.sign = FTR_UNSIGNED,
+		.min_field_value = 1,
+	},

Could we add a HWCAP for this and change the field to FTR_VISIBLE, please? I know most users of the counter are indirected via the vDSO, but there are some users out there using the counter directly and it would save them having to probe via SIGILL if there was a hwcap available.

Cheers,

Will

Marc Zyngier

author maintainer

2021-09-30 07:43 UTC #16 lore raw

Re: [PATCH v2 14/16] arm64: Add a capability for FEAT_ECV

Quoted 27 lines · Will Deacon

Quoted 25 lines · Marc Zyngier

Add a new capability to detect the Enhanced Counter Virtualization feature (FEAT_ECV).

Reviewed-by Oliver Upton <[email protected]>

Signed-off-by Marc Zyngier <[email protected]>

---
 arch/arm64/kernel/cpufeature.c | 10 ++++++++++
 arch/arm64/tools/cpucaps       |  1 +
 2 files changed, 11 insertions(+)

arch/arm64/kernel/cpufeature.c +7 -0

index f8a3067d10c6..26b11ce8fff6 100644
@@ -1926,6 +1926,16 @@ static const struct arm64_cpu_capabilities arm64_features[] = {
 		.sign = FTR_UNSIGNED,
 		.min_field_value = 1,
 	},
+	{
+		.desc = "Enhanced Counter Virtualization",
+		.capability = ARM64_HAS_ECV,
+		.type = ARM64_CPUCAP_SYSTEM_FEATURE,
+		.matches = has_cpuid_feature,
+		.sys_reg = SYS_ID_AA64MMFR0_EL1,
+		.field_pos = ID_AA64MMFR0_ECV_SHIFT,

Quoted 8 lines

arch/arm64/kernel/cpufeature.c static const struct arm64_cpu_capabilities arm64_features[] = { · line 1938
Quoted 3 lines
arch/arm64/kernel/cpufeature.c +3 -0
+		.sign = FTR_UNSIGNED,
+		.min_field_value = 1,
+	},
Could we add a HWCAP for this and change the field to FTR_VISIBLE, please? I know most users of the counter are indirected via the vDSO, but there are some users out there using the counter directly and it would save them having to probe via SIGILL if there was a hwcap available.

Fair enough, I'll add that.

The problem of the vdso remains though, and is by far the most common user of the feature. Any idea on how to handle it? Patching the vdso is ugly, and I'd rather avoid it.

I briefly looked at using ifunc, but it is likely that the indirection would add an extra cost. Are we OK with that?

M.

Signature · 1 line

Without deviation from the norm, progress is not possible.

Will Deacon

maintainer

2021-09-30 08:31 UTC #17 lore raw

Re: [PATCH v2 14/16] arm64: Add a capability for FEAT_ECV

Quoted 38 lines · Marc Zyngier

Quoted 27 lines · Will Deacon
Quoted 25 lines · Marc Zyngier
Add a new capability to detect the Enhanced Counter Virtualization feature (FEAT_ECV).
Reviewed-by Oliver Upton <[email protected]>
Signed-off-by Marc Zyngier <[email protected]>
---
 arch/arm64/kernel/cpufeature.c | 10 ++++++++++
 arch/arm64/tools/cpucaps       |  1 +
 2 files changed, 11 insertions(+)
arch/arm64/kernel/cpufeature.c +7 -0
index f8a3067d10c6..26b11ce8fff6 100644
@@ -1926,6 +1926,16 @@ static const struct arm64_cpu_capabilities arm64_features[] = {
 		.sign = FTR_UNSIGNED,
 		.min_field_value = 1,
 	},
+	{
+		.desc = "Enhanced Counter Virtualization",
+		.capability = ARM64_HAS_ECV,
+		.type = ARM64_CPUCAP_SYSTEM_FEATURE,
+		.matches = has_cpuid_feature,
+		.sys_reg = SYS_ID_AA64MMFR0_EL1,
+		.field_pos = ID_AA64MMFR0_ECV_SHIFT,
Quoted 8 lines
arch/arm64/kernel/cpufeature.c static const struct arm64_cpu_capabilities arm64_features[] = { · line 1938
Quoted 3 lines
arch/arm64/kernel/cpufeature.c +3 -0
+		.sign = FTR_UNSIGNED,
+		.min_field_value = 1,
+	},
Could we add a HWCAP for this and change the field to FTR_VISIBLE, please? I know most users of the counter are indirected via the vDSO, but there are some users out there using the counter directly and it would save them having to probe via SIGILL if there was a hwcap available.

Quoted 1 line

Fair enough, I'll add that.

Thanks!

Quoted 6 lines

The problem of the vdso remains though, and is by far the most common user of the feature. Any idea on how to handle it? Patching the vdso is ugly, and I'd rather avoid it.
I briefly looked at using ifunc, but it is likely that the indirection would add an extra cost. Are we OK with that?

The vDSO is still miles faster than a system call, so I'd be inclined to leave it as-is for the time being. I suspect that use-cases which can't stomach the cost of the ISB can't stomach the cost of the vDSO at all.

Will

Marc Zyngier

author maintainer

2021-09-22 21:27 UTC #18 lore raw

PATCH v2 15/16 arm64: Add CNT{P,V}CTSS_EL0 alternatives to cnt{p,v}ct_el0

In reply to Marc Zyngier

CNTPCTSS_EL0 and CNTVCTSS_EL0 are alternatives to the usual CNTPCT_EL0 and CNTVCT_EL0 that do not require a previous ISB to be synchronised (SS stands for Self-Synchronising).

Use the ARM64_HAS_ECV capability to control alternative sequences that switch to these low(er)-cost primitives. Note that the counter access in the VDSO is for now left alone until we decide whether we want to allow this.

Signed-off-by Marc Zyngier <[email protected]>

---
 arch/arm64/include/asm/arch_timer.h | 32 +++++++++++++++++++++--------
 arch/arm64/include/asm/sysreg.h     |  3 +++
 2 files changed, 27 insertions(+), 8 deletions(-)

arch/arm64/include/asm/arch_timer.h +24 -8

index 519ac1f7f859..33a08fff0f06 100644
@@ -64,14 +64,26 @@ DECLARE_PER_CPU(const struct arch_timer_erratum_workaround *,
 
 static inline notrace u64 arch_timer_read_cntpct_el0(void)
 {
-	isb();
-	return read_sysreg(cntpct_el0);
+	u64 cnt;
+
+	asm volatile(ALTERNATIVE("isb\n mrs %x0, cntpct_el0",
+				 "nop\n" __mrs_s("%x0", SYS_CNTPCTSS_EL0),
+				 ARM64_HAS_ECV)
+		     : "=r" (cnt));
+
+	return cnt;
 }
 
 static inline notrace u64 arch_timer_read_cntvct_el0(void)
 {
-	isb();
-	return read_sysreg(cntvct_el0);
+	u64 cnt;
+
+	asm volatile(ALTERNATIVE("isb\n mrs %x0, cntvct_el0",
+				 "nop\n" __mrs_s("%x0", SYS_CNTVCTSS_EL0),
+				 ARM64_HAS_ECV)
+		     : "=r" (cnt));
+
+	return cnt;
 }
 
 #define arch_timer_reg_read_stable(reg)					\

@@ -174,8 +186,10 @@ static __always_inline u64 __arch_counter_get_cntpct(void)
 {
 	u64 cnt;
 
-	isb();
-	cnt = read_sysreg(cntpct_el0);
+	asm volatile(ALTERNATIVE("isb\n mrs %x0, cntpct_el0",
+				 "nop\n" __mrs_s("%x0", SYS_CNTPCTSS_EL0),
+				 ARM64_HAS_ECV)
+		     : "=r" (cnt));
 	arch_counter_enforce_ordering(cnt);
 	return cnt;
 }

@@ -193,8 +207,10 @@ static __always_inline u64 __arch_counter_get_cntvct(void)
 {
 	u64 cnt;
 
-	isb();
-	cnt = read_sysreg(cntvct_el0);
+	asm volatile(ALTERNATIVE("isb\n mrs %x0, cntvct_el0",
+				 "nop\n" __mrs_s("%x0", SYS_CNTVCTSS_EL0),
+				 ARM64_HAS_ECV)
+		     : "=r" (cnt));
 	arch_counter_enforce_ordering(cnt);
 	return cnt;
 }

arch/arm64/include/asm/sysreg.h +3 -0

index b268082d67ed..5ce70c034d37 100644
@@ -507,6 +507,9 @@
 
 #define SYS_CNTFRQ_EL0			sys_reg(3, 3, 14, 0, 0)
 
+#define SYS_CNTPCTSS_EL0		sys_reg(3, 3, 14, 0, 5)
+#define SYS_CNTVCTSS_EL0		sys_reg(3, 3, 14, 0, 6)
+
 #define SYS_CNTP_TVAL_EL0		sys_reg(3, 3, 14, 2, 0)
 #define SYS_CNTP_CTL_EL0		sys_reg(3, 3, 14, 2, 1)
 #define SYS_CNTP_CVAL_EL0		sys_reg(3, 3, 14, 2, 2)

Signature · 1 line

2.30.2

Marc Zyngier

author maintainer

2021-09-22 21:26 UTC #19 lore raw

PATCH v2 16/16 arm64: Add handling of CNTVCTSS traps

In reply to Marc Zyngier

Since CNTVCTSS obey the same control bits as CNTVCT, add the necessary decoding to the hook table. Note that there is no known user of this at the moment.

Signed-off-by Marc Zyngier <[email protected]>

---
 arch/arm64/include/asm/esr.h |  6 ++++++
 arch/arm64/kernel/traps.c    | 11 +++++++++++
 2 files changed, 17 insertions(+)

arch/arm64/include/asm/esr.h +6 -0

index 29f97eb3dad4..a305ce256090 100644
@@ -227,6 +227,9 @@
 #define ESR_ELx_SYS64_ISS_SYS_CNTVCT	(ESR_ELx_SYS64_ISS_SYS_VAL(3, 3, 2, 14, 0) | \
 					 ESR_ELx_SYS64_ISS_DIR_READ)
 
+#define ESR_ELx_SYS64_ISS_SYS_CNTVCTSS	(ESR_ELx_SYS64_ISS_SYS_VAL(3, 3, 6, 14, 0) | \
+					 ESR_ELx_SYS64_ISS_DIR_READ)
+
 #define ESR_ELx_SYS64_ISS_SYS_CNTFRQ	(ESR_ELx_SYS64_ISS_SYS_VAL(3, 3, 0, 14, 0) | \
 					 ESR_ELx_SYS64_ISS_DIR_READ)

@@ -317,6 +320,9 @@
 #define ESR_ELx_CP15_64_ISS_SYS_CNTVCT	(ESR_ELx_CP15_64_ISS_SYS_VAL(1, 14) | \
 					 ESR_ELx_CP15_64_ISS_DIR_READ)
 
+#define ESR_ELx_CP15_64_ISS_SYS_CNTVCTSS (ESR_ELx_CP15_64_ISS_SYS_VAL(9, 14) | \
+					 ESR_ELx_CP15_64_ISS_DIR_READ)
+
 #define ESR_ELx_CP15_32_ISS_SYS_CNTFRQ	(ESR_ELx_CP15_32_ISS_SYS_VAL(0, 0, 14, 0) |\
 					 ESR_ELx_CP15_32_ISS_DIR_READ)

arch/arm64/kernel/traps.c +11 -0

index b03e383d944a..16710ca55fbb 100644
@@ -653,6 +653,12 @@ static const struct sys64_hook sys64_hooks[] = {
 		.esr_val = ESR_ELx_SYS64_ISS_SYS_CNTVCT,
 		.handler = cntvct_read_handler,
 	},
+	{
+		/* Trap read access to CNTVCTSS_EL0 */
+		.esr_mask = ESR_ELx_SYS64_ISS_SYS_OP_MASK,
+		.esr_val = ESR_ELx_SYS64_ISS_SYS_CNTVCTSS,
+		.handler = cntvct_read_handler,
+	},
 	{
 		/* Trap read access to CNTFRQ_EL0 */
 		.esr_mask = ESR_ELx_SYS64_ISS_SYS_OP_MASK,

@@ -729,6 +735,11 @@ static const struct sys64_hook cp15_64_hooks[] = {
 		.esr_val = ESR_ELx_CP15_64_ISS_SYS_CNTVCT,
 		.handler = compat_cntvct_read_handler,
 	},
+	{
+		.esr_mask = ESR_ELx_CP15_64_ISS_SYS_MASK,
+		.esr_val = ESR_ELx_CP15_64_ISS_SYS_CNTVCTSS,
+		.handler = compat_cntvct_read_handler,
+	},
 	{},
 };

Signature · 1 line

2.30.2