RISC-V nommu support v3
Subsystems (8)
Hi all,
below is a series to support nommu mode on RISC-V. For now this series just works under qemu with the qemu-virt platform, but Damien has also been able to get kernel based on this tree with additional driver hacks to work on the Kendryte KD210, but that will take a while to cleanup an upstream.
A git tree is available here:
git://git.infradead.org/users/hch/riscv.git riscv-nommu.3
Gitweb:
http://git.infradead.org/users/hch/riscv.git/shortlog/refs/heads/riscv-nommu.3
I've also pushed out a builtroot branch that can build a RISC-V nommu root filesystem here:
git://git.infradead.org/users/hch/buildroot.git riscv-nommu.2
Gitweb:
http://git.infradead.org/users/hch/buildroot.git/shortlog/refs/heads/riscv-nommu.2
Changes since v2:
- rebased to 5.3-rc
- remove the EFI image header for nommu builds
- set ARCH_SLAB_MINALIGN to ensure stack alignment in the flat binary loader
- minor comment improvement
- use #defines for more CSRs
Changes since v1:
- fixes so that a kernel with this series still work on builds with an IOMMU
- small clint cleanups
- the binfmt_flat base and buildroot now don't put arguments on the stack
PATCH 1/15 irqchip/sifive-plic: set max threshold for ignored handlers
When running in M-mode we still the S-mode plic handlers in the DT. Ignore them by setting the maximum threshold.
--- drivers/irqchip/irq-sifive-plic.c | 12 ++++++++++-- 1 file changed, 10 insertions(+), 2 deletions(-)
drivers/irqchip/irq-sifive-plic.c +10 -2
@@ -244,6 +244,7 @@ static int __init plic_init(struct device_node *node,struct plic_handler *handler;irq_hw_number_t hwirq;int cpu, hartid; +u32 threshold =0 ;if (of_irq_parse_one (node, i, &parent)) {pr_err ("failed to parse parent for context %d.\n" , i);
@@ -266,10 +267,16 @@ static int __init plic_init(struct device_node *node,continue ; } +/* +* When running in M-mode we need to ignore the S-mode handler. +* Here we assume it always comes later, but that might be a +* little fragile. +*/ handler =per_cpu_ptr (&plic_handlers, cpu);if (handler->present) {pr_warn ("handler already present for context %d.\n" , i); -continue ; + threshold =0xffffffff ; + goto done; } handler->present = true;
@@ -279,8 +286,9 @@ static int __init plic_init(struct device_node *node, handler->enable_base = plic_regs +ENABLE_BASE + i *ENABLE_PER_HART ; +done:/* priority must be > threshold to trigger an interrupt */ -writel (0 , handler->hart_base +CONTEXT_THRESHOLD ); +writel (threshold, handler->hart_base +CONTEXT_THRESHOLD );for (hwirq =1 ; hwirq <= nr_irqs; hwirq++)plic_toggle (handler, hwirq,0 ); nr_handlers++;
Signature · 1 line
2.20.1
Re: [PATCH 01/15] irqchip/sifive-plic: set max threshold for ignored handlers
Thomas, Jason, Marc,
Quoted 4 lines · Christoph Hellwig
When running in M-mode we still the S-mode plic handlers in the DT. Ignore them by setting the maximum threshold.
Signed-off-by Christoph Hellwig <[email protected]>
If you're happy with this, could one of you ack it so we can merge it with the rest of this series through the RISC-V tree?
thanks
- Paul
Quoted 49 lines
--- drivers/irqchip/irq-sifive-plic.c | 12 ++++++++++-- 1 file changed, 10 insertions(+), 2 deletions(-)drivers/irqchip/irq-sifive-plic.c +10
-2@@ -244,6 +244,7 @@ static int __init plic_init(struct device_node *node,struct plic_handler *handler;irq_hw_number_t hwirq;int cpu, hartid; +u32 threshold =0 ;if (of_irq_parse_one (node, i, &parent)) {pr_err ("failed to parse parent for context %d.\n" , i);@@ -266,10 +267,16 @@ static int __init plic_init(struct device_node *node,continue ; } +/* +* When running in M-mode we need to ignore the S-mode handler. +* Here we assume it always comes later, but that might be a +* little fragile. +*/ handler =per_cpu_ptr (&plic_handlers, cpu);if (handler->present) {pr_warn ("handler already present for context %d.\n" , i); -continue ; + threshold =0xffffffff ; + goto done; } handler->present = true;@@ -279,8 +286,9 @@ static int __init plic_init(struct device_node *node, handler->enable_base = plic_regs +ENABLE_BASE + i *ENABLE_PER_HART ; +done:/* priority must be > threshold to trigger an interrupt */ -writel (0 , handler->hart_base +CONTEXT_THRESHOLD ); +writel (threshold, handler->hart_base +CONTEXT_THRESHOLD );for (hwirq =1 ; hwirq <= nr_irqs; hwirq++)plic_toggle (handler, hwirq,0 ); nr_handlers++;Signature · 1 line
2.20.1
Marc Zyngier
Acked-byRe: [PATCH 01/15] irqchip/sifive-plic: set max threshold for ignored handlers
On Tue, 13 Aug 2019 18:44:02 +0100,
Paul Walmsley <[email protected]> wrote:
Hi Paul,
Quoted 6 lines
Thomas, Jason, Marc,
Quoted 1 line · Christoph Hellwig
When running in M-mode we still the S-mode plic handlers in the DT.
^^^^ missing word?
Quoted 6 lines
Quoted 3 lines
Ignore them by setting the maximum threshold.
Signed-off-by Christoph Hellwig <[email protected]>If you're happy with this, could one of you ack it so we can merge it with the rest of this series through the RISC-V tree?
Sure, no problem.
Thanks,
M.
Signature · 1 line
Jazz is not dead, it just smells funny.
PATCH 2/15 riscv: use CSR_SATP instead of the legacy sptbr name in switch_mm
In reply to Christoph HellwigSwitch to our own constant for the satp register instead of using the old name from a legacy version of the privileged spec.
--- arch/riscv/mm/context.c | 7 +------ 1 file changed, 1 insertion(+), 6 deletions(-)
arch/riscv/mm/context.c +1 -6
@@ -57,12 +57,7 @@ void switch_mm(struct mm_struct *prev, struct mm_struct *next,cpumask_clear_cpu (cpu,mm_cpumask (prev));cpumask_set_cpu (cpu,mm_cpumask (next)); -/* -* Use the old spbtr name instead of using the current satp -* name to support binutils 2.29 which doesn't know about the -* privileged ISA 1.10 yet. -*/ -csr_write (sptbr,virt_to_pfn (next->pgd) |SATP_MODE ); +csr_write (CSR_SATP ,virt_to_pfn (next->pgd) |SATP_MODE );local_flush_tlb_all ();flush_icache_deferred (next);
Signature · 1 line
2.20.1
Re: [PATCH 02/15] riscv: use CSR_SATP instead of the legacy sptbr name in switch_mm
Quoted 5 lines · Christoph Hellwig
Switch to our own constant for the satp register instead of using the old name from a legacy version of the privileged spec.
Signed-off-by Christoph Hellwig <[email protected]>Reviewed-by Atish Patra <[email protected]>
Didn't you want us to replace this with Bin Meng's patch?
https://lore.kernel.org/linux-riscv/[email protected]/
If so, probably best just to drop this one and state a dependency.
- Paul
Re: [PATCH 02/15] riscv: use CSR_SATP instead of the legacy sptbr name in switch_mm
Quoted 5 lines · Paul Walmsley
Quoted 3 lines · Christoph Hellwig
Switch to our own constant for the satp register instead of using the old name from a legacy version of the privileged spec.
Quoted 8 lines
Quoted 2 lines
Signed-off-by Christoph Hellwig <[email protected]>Reviewed-by Atish Patra <[email protected]>Didn't you want us to replace this with Bin Meng's patch?
https://lore.kernel.org/linux-riscv/[email protected]/
If so, probably best just to drop this one and state a dependency.
Either way is fine with me. But until you have a branch with either one applied I'm going to keep resending my patch, as random dependencies on uncommitted patches don't work.
Re: [PATCH 02/15] riscv: use CSR_SATP instead of the legacy sptbr name in switch_mm
Quoted 11 lines · Christoph Hellwig
Quoted 10 lines · Paul Walmsley
Quoted 5 lines · Christoph Hellwig
Switch to our own constant for the satp register instead of using the old name from a legacy version of the privileged spec.
Signed-off-by Christoph Hellwig <[email protected]>Reviewed-by Atish Patra <[email protected]>Didn't you want us to replace this with Bin Meng's patch?
Quoted 7 lines
Quoted 3 lines
https://lore.kernel.org/linux-riscv/[email protected]/
If so, probably best just to drop this one and state a dependency.
Either way is fine with me. But until you have a branch with either one applied I'm going to keep resending my patch, as random dependencies on uncommitted patches don't work.
If you're going to resend a patch, it's better to resend the other one that you've explicitly endorsed in favor of your own.
- Paul
Re: [PATCH 02/15] riscv: use CSR_SATP instead of the legacy sptbr name in switch_mm
In reply to Christoph HellwigQuoted 5 lines · Christoph Hellwig
Switch to our own constant for the satp register instead of using the old name from a legacy version of the privileged spec.
Signed-off-by Christoph Hellwig <[email protected]>Reviewed-by Atish Patra <[email protected]>
Dropping this one in favor of Bin Meng's patch per
https://lore.kernel.org/linux-riscv/[email protected]/
- Paul
PATCH 3/15 riscv: refactor the IPI code
In reply to Christoph HellwigThis prepare for adding native non-SBI IPI code.
--- arch/riscv/kernel/smp.c | 55 +++++++++++++++++++++++------------------ 1 file changed, 31 insertions(+), 24 deletions(-)
arch/riscv/kernel/smp.c +31 -24
@@ -78,13 +78,38 @@ static void ipi_stop(void)wait_for_interrupt (); } +static void send_ipi_mask (const struct cpumask *mask,enum ipi_message_type op) +{ +int cpuid, hartid; +struct cpumask hartid_mask; + +cpumask_clear (&hartid_mask); +mb (); +for_each_cpu (cpuid, mask) { +set_bit (op, &ipi_data[cpuid].bits); + hartid =cpuid_to_hartid_map (cpuid); +cpumask_set_cpu (hartid, &hartid_mask); + } +mb (); +sbi_send_ipi (cpumask_bits (&hartid_mask)); +} + +static void send_ipi_single (int cpu,enum ipi_message_type op) +{ +send_ipi_mask (cpumask_of (cpu), op); +} + +static inline void clear_ipi (void ) +{ +csr_clear (CSR_SIP ,SIE_SSIE ); +} +void riscv_software_interrupt (void ) {unsigned long *pending_ipis = &ipi_data[smp_processor_id ()].bits;unsigned long *stats = ipi_data[smp_processor_id ()].stats; -/* Clear pending IPI */ -csr_clear (CSR_SIP ,SIE_SSIE ); +clear_ipi ();while (true) {unsigned long ops;
@@ -118,23 +143,6 @@ void riscv_software_interrupt(void) } } -static void -send_ipi_message (const struct cpumask *to_whom,enum ipi_message_type operation) -{ -int cpuid, hartid; -struct cpumask hartid_mask; - -cpumask_clear (&hartid_mask); -mb (); -for_each_cpu (cpuid, to_whom) { -set_bit (operation, &ipi_data[cpuid].bits); - hartid =cpuid_to_hartid_map (cpuid); -cpumask_set_cpu (hartid, &hartid_mask); - } -mb (); -sbi_send_ipi (cpumask_bits (&hartid_mask)); -} -static const char *const ipi_names[] = { [IPI_RESCHEDULE ] ="Rescheduling interrupts" , [IPI_CALL_FUNC ] ="Function call interrupts" ,
@@ -156,12 +164,12 @@ void show_ipi_stats(struct seq_file *p, int prec)void arch_send_call_function_ipi_mask (struct cpumask *mask) { -send_ipi_message (mask,IPI_CALL_FUNC ); +send_ipi_mask (mask,IPI_CALL_FUNC ); }void arch_send_call_function_single_ipi (int cpu) { -send_ipi_message (cpumask_of (cpu),IPI_CALL_FUNC ); +send_ipi_single (cpu,IPI_CALL_FUNC ); }void smp_send_stop (void )
@@ -176,7 +184,7 @@ void smp_send_stop(void)if (system_state <=SYSTEM_RUNNING )pr_crit ("SMP: stopping secondary CPUs\n" ); -send_ipi_message (&mask,IPI_CPU_STOP ); +send_ipi_mask (&mask,IPI_CPU_STOP ); }/* Wait up to one second for other CPUs to stop */
@@ -191,6 +199,5 @@ void smp_send_stop(void)void smp_send_reschedule (int cpu) { -send_ipi_message (cpumask_of (cpu),IPI_RESCHEDULE ); +send_ipi_single (cpu,IPI_RESCHEDULE ); } -
Signature · 1 line
2.20.1
Re: [PATCH 03/15] riscv: refactor the IPI code
Quoted 3 lines · Christoph Hellwig
This prepare for adding native non-SBI IPI code.
Signed-off-by Christoph Hellwig <[email protected]>
Thanks, queued for v5.4-rc1.
- Paul
Re: [PATCH 03/15] riscv: refactor the IPI code
Quoted 7 lines · Paul Walmsley
Quoted 3 lines · Christoph Hellwig
This prepare for adding native non-SBI IPI code.
Signed-off-by Christoph Hellwig <[email protected]>Thanks, queued for v5.4-rc1.
Where did you queue it up? I can't find it anywhere in your tree, and I really need a baseline for the next iteration.
Re: [PATCH 03/15] riscv: refactor the IPI code
In reply to Paul WalmsleyQuoted 7 lines · Paul Walmsley
Quoted 3 lines · Christoph Hellwig
This prepare for adding native non-SBI IPI code.
Signed-off-by Christoph Hellwig <[email protected]>Thanks, queued for v5.4-rc1.
I still don't see it in any tree, so I'll resend it with the next version.
PATCH 4/15 riscv: abstract out CSR names for supervisor vs machine mode
In reply to Christoph HellwigMany of the privileged CSRs exist in a supervisor and machine version that are used very similarly. Provide a new X-naming layer so that we don't have to ifdef everywhere for M-mode Linux support.
Contains contributions from Damien Le Moal <[email protected]>.
--- arch/riscv/Kconfig | 4 ++ arch/riscv/include/asm/asm.h | 6 +++ arch/riscv/include/asm/csr.h | 58 ++++++++++++++++++++++++++-- arch/riscv/include/asm/irqflags.h | 12 +++--- arch/riscv/include/asm/processor.h | 2 +- arch/riscv/include/asm/ptrace.h | 16 ++++---- arch/riscv/include/asm/switch_to.h | 8 ++-- arch/riscv/kernel/asm-offsets.c | 8 ++-- arch/riscv/kernel/entry.S | 62 ++++++++++++++++-------------- arch/riscv/kernel/fpu.S | 8 ++-- arch/riscv/kernel/head.S | 12 +++--- arch/riscv/kernel/irq.c | 4 +- arch/riscv/kernel/process.c | 15 ++++---- arch/riscv/kernel/signal.c | 21 +++++----- arch/riscv/kernel/traps.c | 16 ++++---- arch/riscv/lib/uaccess.S | 12 +++--- arch/riscv/mm/extable.c | 4 +- arch/riscv/mm/fault.c | 6 +-- drivers/clocksource/timer-riscv.c | 8 ++-- drivers/irqchip/irq-sifive-plic.c | 4 +- 20 files changed, 177 insertions(+), 109 deletions(-)
arch/riscv/Kconfig +4 -0
@@ -55,6 +55,10 @@ config RISCV select ARCH_HAS_GIGANTIC_PAGE select ARCH_WANT_HUGE_PMD_SHARE if 64BIT +# set if we run in machine mode, cleared if we run in supervisor mode +config M_MODE + bool + config MMU def_bool y
arch/riscv/include/asm/asm.h +6 -0
@@ -65,4 +65,10 @@#error "Unexpected __SIZEOF_SHORT__"#endif +#ifdef CONFIG_M_MODE +# define Xret mret +#else +# define Xret sret +#endif +#endif /* _ASM_RISCV_ASM_H */
arch/riscv/include/asm/csr.h +55 -3
88 lines not on this page · raw message
arch/riscv/include/asm/irqflags.h +6 -6
@@ -13,31 +13,31 @@/* read interrupt enabled status */ static inline unsigned long arch_local_save_flags (void ) { -return csr_read (CSR_SSTATUS ); +return csr_read (CSR_XSTATUS ); }/* unconditionally enable interrupts */ static inline void arch_local_irq_enable (void ) { -csr_set (CSR_SSTATUS ,SR_SIE ); +csr_set (CSR_XSTATUS ,SR_XIE ); }/* unconditionally disable interrupts */ static inline void arch_local_irq_disable (void ) { -csr_clear (CSR_SSTATUS ,SR_SIE ); +csr_clear (CSR_XSTATUS ,SR_XIE ); }/* get status and disable interrupts */ static inline unsigned long arch_local_irq_save (void ) { -return csr_read_clear (CSR_SSTATUS ,SR_SIE ); +return csr_read_clear (CSR_XSTATUS ,SR_XIE ); }/* test flags */ static inline int arch_irqs_disabled_flags (unsigned long flags) { -return !(flags &SR_SIE ); +return !(flags &SR_XIE ); }/* test hardware interrupt enable bit */
@@ -49,7 +49,7 @@ static inline int arch_irqs_disabled(void)/* set interrupt enabled status */ static inline void arch_local_irq_restore (unsigned long flags) { -csr_set (CSR_SSTATUS , flags &SR_SIE ); +csr_set (CSR_XSTATUS , flags &SR_XIE ); }#endif /* _ASM_RISCV_IRQFLAGS_H */
arch/riscv/include/asm/processor.h +1 -1
@@ -42,7 +42,7 @@ struct thread_struct { ((struct pt_regs *)(task_stack_page (tsk) +THREAD_SIZE \ - ALIGN(sizeof(struct pt_regs), STACK_ALIGN))) -#define KSTK_EIP (tsk) (task_pt_regs(tsk)->sepc) +#define KSTK_EIP (tsk) (task_pt_regs(tsk)->xepc)#define KSTK_ESP (tsk) (task_pt_regs(tsk)->sp)
arch/riscv/include/asm/ptrace.h +8 -8
@@ -12,7 +12,7 @@ #ifndef __ASSEMBLY__ struct pt_regs { - unsigned long sepc; + unsigned long xepc; unsigned long ra; unsigned long sp; unsigned long gp;
@@ -44,10 +44,10 @@ struct pt_regs {unsigned long t4;unsigned long t5;unsigned long t6; -/* Supervisor CSRs */ -unsigned long sstatus; -unsigned long sbadaddr; -unsigned long scause; +/* Supervisor/Machine CSRs */ +unsigned long xstatus; +unsigned long xbadaddr; +unsigned long xcause;/* a0 value before the syscall */ unsigned long orig_a0; };
@@ -58,18 +58,18 @@ struct pt_regs {#define REG_FMT "%08lx"#endif -#define user_mode (regs) (((regs)->sstatus & SR_SPP) == 0) +#define user_mode (regs) (((regs)->xstatus & SR_XPP) == 0)/* Helpers for working with the instruction pointer */ static inline unsigned long instruction_pointer (struct pt_regs *regs) { -return regs->sepc; +return regs->xepc; }static inline void instruction_pointer_set (struct pt_regs *regs,unsigned long val) { - regs->sepc = val; + regs->xepc = val; }#define profile_pc (regs) instruction_pointer(regs)
arch/riscv/include/asm/switch_to.h +4 -4
@@ -16,13 +16,13 @@ extern void __fstate_restore(struct task_struct *restore_from);static inline void __fstate_clean (struct pt_regs *regs) { - regs->sstatus |= (regs->sstatus & ~(SR_FS )) |SR_FS_CLEAN ; + regs->xstatus |= (regs->xstatus & ~(SR_FS )) |SR_FS_CLEAN ; }static inline void fstate_save (struct task_struct *task,struct pt_regs *regs) { -if ((regs->sstatus &SR_FS ) ==SR_FS_DIRTY ) { +if ((regs->xstatus &SR_FS ) ==SR_FS_DIRTY ) {__fstate_save (task);__fstate_clean (regs); }
@@ -31,7 +31,7 @@ static inline void fstate_save(struct task_struct *task,static inline void fstate_restore (struct task_struct *task,struct pt_regs *regs) { -if ((regs->sstatus &SR_FS ) !=SR_FS_OFF ) { +if ((regs->xstatus &SR_FS ) !=SR_FS_OFF ) {__fstate_restore (task);__fstate_clean (regs); }
@@ -43,7 +43,7 @@ static inline void __switch_to_aux(struct task_struct *prev,struct pt_regs *regs; regs =task_pt_regs (prev); -if (unlikely (regs->sstatus &SR_SD )) +if (unlikely (regs->xstatus &SR_SD ))fstate_save (prev, regs);fstate_restore (next,task_pt_regs (next)); }
arch/riscv/kernel/asm-offsets.c +4 -4
@@ -71,7 +71,7 @@ void asm_offsets(void)OFFSET (TASK_THREAD_FCSR , task_struct, thread.fstate.fcsr);DEFINE (PT_SIZE ,sizeof (struct pt_regs )); -OFFSET (PT_SEPC , pt_regs, sepc); +OFFSET (PT_XEPC , pt_regs, xepc);OFFSET (PT_RA , pt_regs, ra);OFFSET (PT_FP , pt_regs, s0);OFFSET (PT_S0 , pt_regs, s0);
@@ -105,9 +105,9 @@ void asm_offsets(void)OFFSET (PT_T6 , pt_regs, t6);OFFSET (PT_GP , pt_regs, gp);OFFSET (PT_ORIG_A0 , pt_regs, orig_a0); -OFFSET (PT_SSTATUS , pt_regs, sstatus); -OFFSET (PT_SBADADDR , pt_regs, sbadaddr); -OFFSET (PT_SCAUSE , pt_regs, scause); +OFFSET (PT_XSTATUS , pt_regs, xstatus); +OFFSET (PT_XBADADDR , pt_regs, xbadaddr); +OFFSET (PT_XCAUSE , pt_regs, xcause);/* * THREAD_{F,X}* might be larger than a S-type offset can handle, but
arch/riscv/kernel/entry.S +34 -28
128 lines not on this page · raw message
arch/riscv/kernel/fpu.S +4 -4
@@ -23,7 +23,7 @@ ENTRY(__fstate_save) li a2, TASK_THREAD_F0 add a0, a0, a2 li t1, SR_FS - csrs sstatus, t1 + csrs CSR_XSTATUS, t1 frcsr t0 fsd f0, TASK_THREAD_F0_F0(a0) fsd f1, TASK_THREAD_F1_F0(a0)
@@ -58,7 +58,7 @@ ENTRY(__fstate_save) fsd f30, TASK_THREAD_F30_F0(a0) fsd f31, TASK_THREAD_F31_F0(a0) sw t0, TASK_THREAD_FCSR_F0(a0) - csrc sstatus, t1 + csrc CSR_XSTATUS, t1 ret ENDPROC(__fstate_save)
@@ -67,7 +67,7 @@ ENTRY(__fstate_restore) add a0, a0, a2 li t1, SR_FS lw t0, TASK_THREAD_FCSR_F0(a0) - csrs sstatus, t1 + csrs CSR_XSTATUS, t1 fld f0, TASK_THREAD_F0_F0(a0) fld f1, TASK_THREAD_F1_F0(a0) fld f2, TASK_THREAD_F2_F0(a0)
@@ -101,6 +101,6 @@ ENTRY(__fstate_restore) fld f30, TASK_THREAD_F30_F0(a0) fld f31, TASK_THREAD_F31_F0(a0) fscsr t0 - csrc sstatus, t1 + csrc CSR_XSTATUS, t1 ret ENDPROC(__fstate_restore)
arch/riscv/kernel/head.S +6 -6
@@ -47,8 +47,8 @@ ENTRY(_start) .global _start_kernel _start_kernel: /* Mask all interrupts */ - csrw CSR_SIE, zero - csrw CSR_SIP, zero + csrw CSR_XIE, zero + csrw CSR_XIP, zero /* Load the global pointer */ .option push
@@ -61,7 +61,7 @@ _start_kernel: * floating point in kernel space */ li t0, SR_FS - csrc sstatus, t0 + csrc CSR_XSTATUS, t0 /* Pick one hart to run the main boot sequence */ la a3, hart_lottery
@@ -111,7 +111,7 @@ relocate: /* Point stvec to virtual address of intruction after satp write */ la a2, 1f add a2, a2, a1 - csrw CSR_STVEC, a2 + csrw CSR_XTVEC, a2 /* Compute satp for kernel page tables, but don't load it yet */ srl a2, a0, PAGE_SHIFT
@@ -133,7 +133,7 @@ relocate: 1: /* Set trap vector to spin forever to help debug */ la a0, .Lsecondary_park - csrw CSR_STVEC, a0 + csrw CSR_XTVEC, a0 /* Reload the global pointer */ .option push
@@ -159,7 +159,7 @@ relocate: /* Set trap vector to spin forever to help debug */ la a3, .Lsecondary_park - csrw CSR_STVEC, a3 + csrw CSR_XTVEC, a3 slli a3, a0, LGREG la a1, __cpu_up_stack_pointer
arch/riscv/kernel/irq.c +2 -2
@@ -29,7 +29,7 @@ asmlinkage void __irq_entry do_IRQ(struct pt_regs *regs)struct pt_regs *old_regs =set_irq_regs (regs);irq_enter (); -switch (regs->scause & ~SCAUSE_IRQ_FLAG ) { +switch (regs->xcause & ~XCAUSE_IRQ_FLAG ) {case INTERRUPT_CAUSE_TIMER :riscv_timer_interrupt ();break ;
@@ -46,7 +46,7 @@ asmlinkage void __irq_entry do_IRQ(struct pt_regs *regs)handle_arch_irq (regs);break ;default : -pr_alert ("unexpected interrupt cause 0x%lx" , regs->scause); +pr_alert ("unexpected interrupt cause 0x%lx" , regs->xcause);BUG (); }irq_exit ();
arch/riscv/kernel/process.c +8 -7
@@ -34,7 +34,7 @@ void show_regs(struct pt_regs *regs)show_regs_print_info (KERN_DEFAULT );pr_cont ("sepc: " REG_FMT " ra : " REG_FMT " sp : " REG_FMT "\n" , - regs->sepc, regs->ra, regs->sp); + regs->xepc, regs->ra, regs->sp);pr_cont (" gp : " REG_FMT " tp : " REG_FMT " t0 : " REG_FMT "\n" , regs->gp, regs->tp, regs->t0); pr_cont(" t1 : " REG_FMT " t2 : " REG_FMT " s0 : " REG_FMT "\n",
@@ -56,17 +56,17 @@ void show_regs(struct pt_regs *regs)pr_cont (" t5 : " REG_FMT " t6 : " REG_FMT "\n" , regs->t5, regs->t6); -pr_cont ("sstatus: " REG_FMT " sbadaddr: " REG_FMT " scause: " REG_FMT "\n" , - regs->sstatus, regs->sbadaddr, regs->scause); +pr_cont ("status: " REG_FMT " badaddr: " REG_FMT " cause: " REG_FMT "\n" , + regs->xstatus, regs->xbadaddr, regs->xcause); }void start_thread (struct pt_regs *regs,unsigned long pc,unsigned long sp) { - regs->sstatus =SR_SPIE ; + regs->xstatus =SR_XPIE ;if (has_fpu) - regs->sstatus |=SR_FS_INITIAL ; - regs->sepc = pc; + regs->xstatus |=SR_FS_INITIAL ; + regs->xepc = pc; regs->sp = sp;set_fs (USER_DS ); }
@@ -101,7 +101,8 @@ int copy_thread(unsigned long clone_flags, unsigned long usp,const registerunsigned long gp __asm__ ("gp" );memset (childregs,0 ,sizeof (struct pt_regs )); childregs->gp = gp; - childregs->sstatus =SR_SPP |SR_SPIE ;/* Supervisor, irqs on */ +/* Supervisor/Machine, irqs on: */ + childregs->xstatus =SR_XPP |SR_XPIE ; p->thread.ra = (unsigned long )ret_from_kernel_thread; p->thread.s[0 ] = usp;/* fn */
arch/riscv/kernel/signal.c +10 -11
@@ -124,7 +124,7 @@ SYSCALL_DEFINE0(rt_sigreturn)pr_info_ratelimited ("%s[%d]: bad frame in %s: frame=%p pc=%p sp=%p\n" , task->comm,task_pid_nr (task), __func__, - frame, (void *)regs->sepc, (void *)regs->sp); + frame, (void *)regs->xepc, (void *)regs->sp); }force_sig (SIGSEGV );return 0 ;
@@ -199,7 +199,7 @@ static int setup_rt_frame(struct ksignal *ksig, sigset_t *set,* We always pass siginfo and mcontext, regardless of SA_SIGINFO, * since some things rely on this (e.g. glibc's debug/segfault.c). */ - regs->sepc = (unsigned long )ksig->ka.sa.sa_handler; + regs->xepc = (unsigned long )ksig->ka.sa.sa_handler; regs->sp = (unsigned long )frame; regs->a0 = ksig->sig;/* a0: signal number */ regs->a1 = (unsigned long )(&frame->info);/* a1: siginfo pointer */
@@ -208,7 +208,7 @@ static int setup_rt_frame(struct ksignal *ksig, sigset_t *set,#if DEBUG_SIG pr_info ("SIG deliver (%s:%d): sig=%d pc=%p ra=%p sp=%p\n" , current->comm,task_pid_nr (current), ksig->sig, - (void *)regs->sepc, (void *)regs->ra, frame); + (void *)regs->xepc, (void *)regs->ra, frame);#endif return 0 ;
@@ -220,10 +220,9 @@ static void handle_signal(struct ksignal *ksig, struct pt_regs *regs)int ret;/* Are we from a system call? */ -if (regs->scause ==EXC_SYSCALL ) { +if (regs->xcause ==EXC_SYSCALL ) {/* Avoid additional syscall restarting via ret_from_exception */ - regs->scause =-1UL ; - + regs->xcause =-1UL ;/* If so, check system call restarting.. */ switch (regs->a0) {case -ERESTART_RESTARTBLOCK :
@@ -239,7 +238,7 @@ static void handle_signal(struct ksignal *ksig, struct pt_regs *regs)/* fallthrough */ case -ERESTARTNOINTR : regs->a0 = regs->orig_a0; - regs->sepc -=0x4 ; + regs->xepc -=0x4 ;break ; } }
@@ -261,9 +260,9 @@ static void do_signal(struct pt_regs *regs) }/* Did we come from a system call? */ -if (regs->scause ==EXC_SYSCALL ) { +if (regs->xcause ==EXC_SYSCALL ) {/* Avoid additional syscall restarting via ret_from_exception */ - regs->scause =-1UL ; + regs->xcause =-1UL ;/* Restart the system call - no handlers present */ switch (regs->a0) {
@@ -271,12 +270,12 @@ static void do_signal(struct pt_regs *regs)case -ERESTARTSYS :case -ERESTARTNOINTR : regs->a0 = regs->orig_a0; - regs->sepc -=0x4 ; + regs->xepc -=0x4 ;break ;case -ERESTART_RESTARTBLOCK : regs->a0 = regs->orig_a0; regs->a7 = __NR_restart_syscall; - regs->sepc -=0x4 ; + regs->xepc -=0x4 ;break ; } }
arch/riscv/kernel/traps.c +8 -8
@@ -40,7 +40,7 @@ void die(struct pt_regs *regs, const char *str)print_modules ();show_regs (regs); - ret =notify_die (DIE_OOPS , str, regs,0 , regs->scause,SIGSEGV ); + ret =notify_die (DIE_OOPS , str, regs,0 , regs->xcause,SIGSEGV );bust_spinlocks (0 );add_taint (TAINT_DIE ,LOCKDEP_NOW_UNRELIABLE );
@@ -85,7 +85,7 @@ static void do_trap_error(struct pt_regs *regs, int signo, int code,#define DO_ERROR_INFO (name, signo, code, str) \ asmlinkage void name(struct pt_regs *regs) \ { \ - do_trap_error(regs, signo, code, regs->sepc, "Oops - " str); \ + do_trap_error(regs, signo, code, regs->xepc, "Oops - " str); \ } DO_ERROR_INFO(do_trap_unknown,
@@ -128,12 +128,12 @@ asmlinkage void do_trap_break(struct pt_regs *regs)if (!user_mode (regs)) {enum bug_trap_type type; - type =report_bug (regs->sepc, regs); + type =report_bug (regs->xepc, regs);switch (type) {case BUG_TRAP_TYPE_NONE :break ;case BUG_TRAP_TYPE_WARN : - regs->sepc +=get_break_insn_length (regs->sepc); + regs->xepc +=get_break_insn_length (regs->xepc);break ;case BUG_TRAP_TYPE_BUG :die (regs,"Kernel BUG" );
@@ -141,7 +141,7 @@ asmlinkage void do_trap_break(struct pt_regs *regs) }#endif /* CONFIG_GENERIC_BUG */ -force_sig_fault (SIGTRAP ,TRAP_BRKPT , (void __user *)(regs->sepc)); +force_sig_fault (SIGTRAP ,TRAP_BRKPT , (void __user *)(regs->xepc)); }#ifdef CONFIG_GENERIC_BUG
@@ -166,9 +166,9 @@ void __init trap_init(void)* Set sup0 scratch register to 0, indicating to exception vector * that we are presently executing in the kernel */ -csr_write (CSR_SSCRATCH ,0 ); +csr_write (CSR_XSCRATCH ,0 );/* Set the exception vector address */ -csr_write (CSR_STVEC , &handle_exception); +csr_write (CSR_XTVEC , &handle_exception);/* Enable all interrupts */ -csr_write (CSR_SIE ,-1 ); +csr_write (CSR_XIE ,-1 ); }
arch/riscv/lib/uaccess.S +6 -6
@@ -18,7 +18,7 @@ ENTRY(__asm_copy_from_user) /* Enable access to user memory */ li t6, SR_SUM - csrs sstatus, t6 + csrs CSR_XSTATUS, t6 add a3, a1, a2 /* Use word-oriented copy only if low-order bits match */
@@ -47,7 +47,7 @@ ENTRY(__asm_copy_from_user) 3: /* Disable access to user memory */ - csrc sstatus, t6 + csrc CSR_XSTATUS, t6 li a0, 0 ret 4: /* Edge case: unalignment */
@@ -72,7 +72,7 @@ ENTRY(__clear_user) /* Enable access to user memory */ li t6, SR_SUM - csrs sstatus, t6 + csrs CSR_XSTATUS, t6 add a3, a0, a1 addi t0, a0, SZREG-1
@@ -94,7 +94,7 @@ ENTRY(__clear_user) 3: /* Disable access to user memory */ - csrc sstatus, t6 + csrc CSR_XSTATUS, t6 li a0, 0 ret 4: /* Edge case: unalignment */
@@ -114,11 +114,11 @@ ENDPROC(__clear_user) /* Fixup code for __copy_user(10) and __clear_user(11) */ 10: /* Disable access to user memory */ - csrs sstatus, t6 + csrs CSR_XSTATUS, t6 mv a0, a2 ret 11: - csrs sstatus, t6 + csrs CSR_XSTATUS, t6 mv a0, a1 ret .previous
arch/riscv/mm/extable.c +2 -2
@@ -15,9 +15,9 @@ int fixup_exception(struct pt_regs *regs) {const struct exception_table_entry *fixup; - fixup =search_exception_tables (regs->sepc); + fixup =search_exception_tables (regs->xepc);if (fixup) { - regs->sepc = fixup->fixup; + regs->xepc = fixup->fixup;return 1 ; }return 0 ;
arch/riscv/mm/fault.c +3 -3
@@ -32,8 +32,8 @@ asmlinkage void do_page_fault(struct pt_regs *regs)int code =SEGV_MAPERR ;vm_fault_t fault; - cause = regs->scause; - addr = regs->sbadaddr; + cause = regs->xcause; + addr = regs->xbadaddr; tsk = current; mm = tsk->mm;
@@ -51,7 +51,7 @@ asmlinkage void do_page_fault(struct pt_regs *regs) goto vmalloc_fault;/* Enable interrupts if they were enabled in the parent context. */ -if (likely (regs->sstatus &SR_SPIE )) +if (likely (regs->xstatus &SR_XPIE ))local_irq_enable ();/*
drivers/clocksource/timer-riscv.c +4 -4
@@ -28,7 +28,7 @@static int riscv_clock_next_event (unsigned long delta,struct clock_event_device *ce) { -csr_set (sie,SIE_STIE ); +csr_set (CSR_XIE ,XIE_XTIE );sbi_set_timer (get_cycles64 () + delta);return 0 ; }
@@ -70,13 +70,13 @@ static int riscv_timer_starting_cpu(unsigned int cpu) ce->cpumask =cpumask_of (cpu);clockevents_config_and_register (ce, riscv_timebase,100 ,0x7fffffff ); -csr_set (sie,SIE_STIE ); +csr_set (CSR_XIE ,XIE_XTIE );return 0 ; }static int riscv_timer_dying_cpu (unsigned int cpu) { -csr_clear (sie,SIE_STIE ); +csr_clear (CSR_XIE ,XIE_XTIE );return 0 ; }
@@ -85,7 +85,7 @@ void riscv_timer_interrupt(void) {struct clock_event_device *evdev =this_cpu_ptr (&riscv_clock_event); -csr_clear (sie,SIE_STIE ); +csr_clear (CSR_XIE ,XIE_XTIE ); evdev->event_handler (evdev); }
drivers/irqchip/irq-sifive-plic.c +2 -2
@@ -179,7 +179,7 @@ static void plic_handle_irq(struct pt_regs *regs)WARN_ON_ONCE (!handler->present); -csr_clear (sie,SIE_SEIE ); +csr_clear (CSR_XIE ,XIE_XEIE );while ((hwirq =readl (claim))) {int irq =irq_find_mapping (plic_irqdomain, hwirq);
@@ -190,7 +190,7 @@ static void plic_handle_irq(struct pt_regs *regs)generic_handle_irq (irq);writel (hwirq, claim); } -csr_set (sie,SIE_SEIE ); +csr_set (CSR_XIE ,XIE_XEIE ); }/*
Signature · 7 lines
2.20.1 _______________________________________________ linux-riscv mailing list [email protected] http://lists.infradead.org/mailman/listinfo/linux-riscv
PATCH 5/15 riscv: improve the default power off implementation
In reply to Christoph HellwigOnly call the SBI code if we are not running in M mode, and if we didn't do the SBI call, or it didn't succeed call wfi in a loop to at least save some power.
--- arch/riscv/kernel/reset.c | 5 ++++- 1 file changed, 4 insertions(+), 1 deletion(-)
arch/riscv/kernel/reset.c +4 -1
@@ -8,8 +8,11 @@static void default_power_off (void ) { +#ifndef CONFIG_M_MODE sbi_shutdown (); -while (1 ); +#endif +while (1 ) +wait_for_interrupt (); }void (*pm_power_off)(void ) = default_power_off;
Signature · 7 lines
2.20.1 _______________________________________________ linux-riscv mailing list [email protected] http://lists.infradead.org/mailman/listinfo/linux-riscv
PATCH 6/15 riscv: provide a flat entry loader
In reply to Christoph HellwigThis allows just loading the kernel at a pre-set address without qemu going bonkers trying to map the ELF file.
--- arch/riscv/Makefile | 13 +++++++++---- arch/riscv/boot/Makefile | 7 ++++++- arch/riscv/boot/loader.S | 8 ++++++++ arch/riscv/boot/loader.lds | 14 ++++++++++++++ 4 files changed, 37 insertions(+), 5 deletions(-) create mode 100644 arch/riscv/boot/loader.S create mode 100644 arch/riscv/boot/loader.lds
arch/riscv/Makefile +9 -4
@@ -80,13 +80,18 @@ PHONY += vdso_install vdso_install: $(Q)$(MAKE) $(build)=arch/riscv/kernel/vdso $@ -all: Image.gz +ifeq ($(CONFIG_M_MODE),y) +KBUILD_IMAGE := $(boot)/loader +else +KBUILD_IMAGE := $(boot)/Image.gz +endif +BOOT_TARGETS := Image Image.gz loader -Image: vmlinux - $(Q)$(MAKE) $(build)=$(boot) $(boot)/$@ +all: $(notdir $(KBUILD_IMAGE)) -Image.%: Image +$(BOOT_TARGETS): vmlinux $(Q)$(MAKE) $(build)=$(boot) $(boot)/$@ + @$(kecho) ' Kernel: $(boot)/$@ is ready' zinstall install: $(Q)$(MAKE) $(build)=$(boot) $@
arch/riscv/boot/Makefile +6 -1
@@ -16,7 +16,7 @@ OBJCOPYFLAGS_Image :=-O binary -R .note -R .note.gnu.build-id -R .comment -S -targets := Image +targets := Image loader $(obj)/Image: vmlinux FORCE $(call if_changed,objcopy)
@@ -24,6 +24,11 @@ $(obj)/Image: vmlinux FORCE $(obj)/Image.gz: $(obj)/Image FORCE $(call if_changed,gzip) +loader.o: $(src)/loader.S $(obj)/Image + +$(obj)/loader: $(obj)/loader.o $(obj)/Image FORCE + $(Q)$(LD) -T $(src)/loader.lds -o $@ $(obj)/loader.o + install: $(CONFIG_SHELL) $(srctree)/$(src)/install.sh $(KERNELRELEASE) \ $(obj)/Image System.map "$(INSTALL_PATH)"
arch/riscv/boot/loader.S new file +8 -0
@@ -0,0 +1,8 @@ +// SPDX-License-Identifier: GPL-2.0 + + .align 4 + .section .payload, "ax", %progbits + .globl _start +_start: + .incbin "arch/riscv/boot/Image" +
arch/riscv/boot/loader.lds new file +14 -0
@@ -0,0 +1,14 @@ +/* SPDX-License-Identifier: GPL-2.0 */ + +OUTPUT_ARCH(riscv) +ENTRY(_start) + +SECTIONS +{ + . = 0x80000000; + + .payload : { + *(.payload) + . = ALIGN(8); + } +}
Signature · 1 line
2.20.1
PATCH 7/15 riscv: read the hart ID from mhartid on boot
In reply to Christoph HellwigFrom: Damien Le Moal <[email protected]>
When in M-Mode, we can use the mhartid CSR to get the ID of the running HART. Doing so, direct M-Mode boot without firmware is possible.
--- arch/riscv/include/asm/csr.h | 1 + arch/riscv/kernel/head.S | 8 ++++++++ 2 files changed, 9 insertions(+)
arch/riscv/include/asm/csr.h +1 -0
@@ -81,6 +81,7 @@#define SIE_SEIE (_AC(0x1, UL) << IRQ_S_EXT)/* symbolic CSR names: */ +#define CSR_MHARTID 0xf14#define CSR_MSTATUS 0x300#define CSR_MIE 0x304#define CSR_MTVEC 0x305
arch/riscv/kernel/head.S +8 -0
@@ -50,6 +50,14 @@ _start_kernel: csrw CSR_XIE, zero csrw CSR_XIP, zero +#ifdef CONFIG_M_MODE + /* + * The hartid in a0 is expected later on, and we have no firmware + * to hand it to us. + */ + csrr a0, CSR_MHARTID +#endif + /* Load the global pointer */ .option push .option norelax
Signature · 1 line
2.20.1
PATCH 8/15 riscv: provide native clint access for M-mode
In reply to Christoph HellwigRISC-V has the concept of a cpu level interrupt controller. Part of it is expose as bits in the status registers, and 2 new CSRs per privilege level in the instruction set, but the machanisms to trigger IPIs and timer events, as well as reading the actual timer value are not specified in the RISC-V spec but usually delegated to a block of MMIO registers. This patch adds support for those MMIO registers in the timer and IPI code. For now only the SiFive layout also supported by a few other implementations is supported, but the code should be easily extensible to others in the future.
--- arch/riscv/include/asm/clint.h | 40 +++++++++++++++++++++++++++ arch/riscv/include/asm/timex.h | 17 ++++++++++++ arch/riscv/kernel/Makefile | 1 + arch/riscv/kernel/clint.c | 45 +++++++++++++++++++++++++++++++ arch/riscv/kernel/setup.c | 2 ++ arch/riscv/kernel/smp.c | 24 +++++++++++++++++ arch/riscv/kernel/smpboot.c | 3 +++ drivers/clocksource/timer-riscv.c | 16 ++++++++--- 8 files changed, 144 insertions(+), 4 deletions(-) create mode 100644 arch/riscv/include/asm/clint.h create mode 100644 arch/riscv/kernel/clint.c
arch/riscv/include/asm/clint.h new file +40 -0
@@ -0,0 +1,40 @@ +// SPDX-License-Identifier: GPL-2.0 +#ifndef _ASM_CLINT_H +#define _ASM_CLINT_H 1 + +#include <linux/smp.h> + +#ifdef CONFIG_M_MODE +extern u32 __iomem *clint_ipi_base; +extern u64 __iomem *clint_time_val; +extern u64 __iomem *clint_time_cmp; + +void clint_init_boot_cpu (void ); + +static inline void clint_send_ipi (unsigned long hartid) +{ +writel (1 , clint_ipi_base + hartid); +} + +static inline void clint_clear_ipi (unsigned long hartid) +{ +writel (0 , clint_ipi_base + hartid); +} + +static inline u64 clint_read_timer (void ) +{ +return readq_relaxed (clint_time_val); +} + +static inline void clint_set_timer (unsigned long delta) +{ +writeq_relaxed (clint_read_timer () + delta, + clint_time_cmp +cpuid_to_hartid_map (smp_processor_id ())); +} + +#else +#define clint_init_boot_cpu () do { } while (0) +#define clint_clear_ipi (hartid) do { } while (0) +#endif /* CONFIG_M_MODE */ + +#endif /* _ASM_CLINT_H */
arch/riscv/include/asm/timex.h +17 -0
@@ -10,6 +10,22 @@typedef unsigned long cycles_t ; +#ifdef CONFIG_M_MODE + +#include <linux/io-64-nonatomic-lo-hi.h> +#include <asm/clint.h> + +static inline cycles_t get_cycles (void ) +{ +#ifdef CONFIG_64BIT +return readq_relaxed (clint_time_val); +#else +return readl_relaxed (clint_time_val); +#endif +} +#define get_cycles get_cycles + +#else /* CONFIG_M_MODE */ static inline cycles_t get_cycles_inline (void ) {cycles_t n;
@@ -40,6 +56,7 @@ static inline uint64_t get_cycles64(void)return ((u64 )hi <<32 ) | lo; }#endif +#endif /* CONFIG_M_MODE */ #define ARCH_HAS_READ_CURRENT_TIMER
arch/riscv/kernel/Makefile +1 -0
@@ -29,6 +29,7 @@ obj-y += vdso.o obj-y += cacheinfo.o obj-y += vdso/ +obj-$(CONFIG_M_MODE) += clint.o obj-$(CONFIG_FPU) += fpu.o obj-$(CONFIG_SMP) += smpboot.o obj-$(CONFIG_SMP) += smp.o
arch/riscv/kernel/clint.c new file +45 -0
45 lines not on this page · raw message
arch/riscv/kernel/setup.c +2 -0
@@ -17,6 +17,7 @@#include <linux/sched/task.h> #include <linux/swiotlb.h> +#include <asm/clint.h> #include <asm/setup.h> #include <asm/sections.h> #include <asm/pgtable.h>
@@ -65,6 +66,7 @@ void __init setup_arch(char **cmdline_p)setup_bootmem ();paging_init ();unflatten_device_tree (); +clint_init_boot_cpu ();#ifdef CONFIG_SWIOTLB swiotlb_init (1 );
arch/riscv/kernel/smp.c +24 -0
@@ -13,7 +13,9 @@#include <linux/sched.h> #include <linux/seq_file.h> #include <linux/delay.h> +#include <linux/io.h> +#include <asm/clint.h> #include <asm/sbi.h> #include <asm/tlbflush.h> #include <asm/cacheflush.h>
@@ -78,6 +80,27 @@ static void ipi_stop(void)wait_for_interrupt (); } +#ifdef CONFIG_M_MODE +static inline void send_ipi_single (int cpu,enum ipi_message_type op) +{ +set_bit (op, &ipi_data[cpu].bits); +clint_send_ipi (cpuid_to_hartid_map (cpu)); +} + +static inline void send_ipi_mask (const struct cpumask *mask, +enum ipi_message_type op) +{ +int cpu; + +for_each_cpu (cpu, mask) +send_ipi_single (cpu, op); +} + +static inline void clear_ipi (void ) +{ +clint_clear_ipi (cpuid_to_hartid_map (smp_processor_id ())); +} +#else /* CONFIG_M_MODE */ static void send_ipi_mask (const struct cpumask *mask,enum ipi_message_type op) {int cpuid, hartid;
@@ -103,6 +126,7 @@ static inline void clear_ipi(void) {csr_clear (CSR_SIP ,SIE_SSIE ); } +#endif /* CONFIG_M_MODE */ void riscv_software_interrupt (void ) {
arch/riscv/kernel/smpboot.c +3 -0
@@ -23,6 +23,7 @@#include <linux/of.h> #include <linux/sched/task_stack.h> #include <linux/sched/mm.h> +#include <asm/clint.h> #include <asm/irq.h> #include <asm/mmu_context.h> #include <asm/tlbflush.h>
@@ -132,6 +133,8 @@ asmlinkage void __init smp_callin(void) {struct mm_struct *mm = &init_mm; +clint_clear_ipi (cpuid_to_hartid_map (smp_processor_id ())); +/* All kernel threads share the same mm context. */ mmgrab (mm); current->active_mm = mm;
drivers/clocksource/timer-riscv.c +12 -4
@@ -24,12 +24,16 @@* operations on the current hart. There is guaranteed to be exactly one timer * per hart on all RISC-V systems. */ -static int riscv_clock_next_event (unsigned long delta,struct clock_event_device *ce) {csr_set (CSR_XIE ,XIE_XTIE ); + +#ifdef CONFIG_M_MODE +clint_set_timer (delta); +#else sbi_set_timer (get_cycles64 () + delta); +#endif return 0 ; }
@@ -45,14 +49,18 @@ static DEFINE_PER_CPU(struct clock_event_device, riscv_clock_event) = {* within one tick of each other, so while this could technically go * backwards when hopping between CPUs, practically it won't happen. */ -static unsigned long long riscv_clocksource_rdtime (struct clocksource *cs) +static u64 riscv_sched_clock (void ) { +#ifdef CONFIG_M_MODE +return clint_read_timer (); +#else return get_cycles64 (); +#endif } -static u64 riscv_sched_clock (void ) +static unsigned long long riscv_clocksource_rdtime (struct clocksource *cs) { -return get_cycles64 (); +return riscv_sched_clock (); }static struct clocksource riscv_clocksource = {
Signature · 1 line
2.20.1
Re: [PATCH 08/15] riscv: provide native clint access for M-mode
Quoted 20 lines · Christoph Hellwig
RISC-V has the concept of a cpu level interrupt controller. Part of it is expose as bits in the status registers, and 2 new CSRs per privilege level in the instruction set, but the machanisms to trigger IPIs and timer events, as well as reading the actual timer value are not specified in the RISC-V spec but usually delegated to a block of MMIO registers. This patch adds support for those MMIO registers in the timer and IPI code. For now only the SiFive layout also supported by a few other implementations is supported, but the code should be easily extensible to others in the future.
Signed-off-by Christoph Hellwig <[email protected]>arch/riscv/kernel/clint.c new file +8
-0+/* +* This is the layout used by the SiFive clint, which is also shared by the qemu +* virt platform, and the Kendryte KD210 at least. +*/ +#define CLINT_IPI_OFF 0 +#define CLINT_TIME_VAL_OFF 0xbff8 +#define CLINT_TIME_CMP_OFF 0x4000; +
arch/riscv/kernel/clint.c line 30
Quoted 10 lines
arch/riscv/kernel/clint.c new file +10
-0+u32 __iomem *clint_ipi_base; +u64 __iomem *clint_time_val; +u64 __iomem *clint_time_cmp; + +void clint_init_boot_cpu (void ) +{ +struct device_node *np; +void __iomem *base; + + np =of_find_compatible_node (NULL ,NULL ,"riscv,clint0" );
Since the MMIO layout is that of the SiFive clint, the compatible string should be specific to that. e.g. "sifive,clint". That way it will be possible to distinguish it from other implementations.
What exactly is the "0" suffix for? Is that a version number?
If that's a CPU index, then I don't think that's the right way to encode this unless the programming interface actually differs across CPUs. It would be better to use an explicit phandle to express the affinity.
Thanks,
Mark.
Re: [PATCH 08/15] riscv: provide native clint access for M-mode
Quoted 11 lines · Mark Rutland
arch/riscv/kernel/clint.c line 30
Quoted 1 line
arch/riscv/kernel/clint.c new file +1
-0+ np =of_find_compatible_node (NULL ,NULL ,"riscv,clint0" );Since the MMIO layout is that of the SiFive clint, the compatible string should be specific to that. e.g. "sifive,clint". That way it will be possible to distinguish it from other implementations.
What exactly is the "0" suffix for? Is that a version number?
If that's a CPU index, then I don't think that's the right way to encode this unless the programming interface actually differs across CPUs. It would be better to use an explicit phandle to express the affinity.
It isn't a cpu index, I suspect a version number. These show up in a lot of the early RISC-V DTs coming from the UCB/SiFive sphere. They've now spread everywhere unfortunately.
Re: [PATCH 08/15] riscv: provide native clint access for M-mode
Quoted 9 lines · Christoph Hellwig
Quoted 8 lines · Mark Rutland
arch/riscv/kernel/clint.c line 30
Quoted 1 line
arch/riscv/kernel/clint.c new file +1
-0+ np =of_find_compatible_node (NULL ,NULL ,"riscv,clint0" );Since the MMIO layout is that of the SiFive clint, the compatible string should be specific to that. e.g. "sifive,clint". That way it will be possible to distinguish it from other implementations.
What exactly is the "0" suffix for? Is that a version number?
Quoted 7 lines
Quoted 3 lines
If that's a CPU index, then I don't think that's the right way to encode this unless the programming interface actually differs across CPUs. It would be better to use an explicit phandle to express the affinity.
It isn't a cpu index, I suspect a version number. These show up in a lot of the early RISC-V DTs coming from the UCB/SiFive sphere. They've now spread everywhere unfortunately.
clint0 would be version 0 of the clint, with is the core-local interrupt controller in rocket chip. It should be "sifive,clint-1.0.0", not "riscv,clint0", as it's a SiFive widget. Unfortunately there are a lot of legacy device trees floating around, but I'm only considering what's been upstream to be actually part of the spec.
In this case the code should match on a "sifive,clint-1.0.0", and the device trees should be fixed up to match. We match on "riscv,plic0" for legacy systems, and I guess it makes sense to do something similar here.
Re: [PATCH 08/15] riscv: provide native clint access for M-mode
Quoted 9 lines · Palmer Dabbelt
clint0 would be version 0 of the clint, with is the core-local interrupt controller in rocket chip. It should be "sifive,clint-1.0.0", not "riscv,clint0", as it's a SiFive widget. Unfortunately there are a lot of legacy device trees floating around, but I'm only considering what's been upstream to be actually part of the spec.
In this case the code should match on a "sifive,clint-1.0.0", and the device trees should be fixed up to match. We match on "riscv,plic0" for legacy systems, and I guess it makes sense to do something similar here.
IFF we decided to change it I'd rather separate DT noes for the ipi bank vs timecmp register vs timeval to support variable layouts. The downside is that we can't just boot on unmodified upstream qemu, which has used the "riscv,clint0" for years.
Re: [PATCH 08/15] riscv: provide native clint access for M-mode
Quoted 7 lines · Christoph Hellwig
Quoted 6 lines · Palmer Dabbelt
clint0 would be version 0 of the clint, with is the core-local interrupt controller in rocket chip. It should be "sifive,clint-1.0.0", not "riscv,clint0", as it's a SiFive widget. Unfortunately there are a lot of legacy device trees floating around, but I'm only considering what's been upstream to be actually part of the spec.
Quoted 8 lines
Quoted 3 lines
In this case the code should match on a "sifive,clint-1.0.0", and the device trees should be fixed up to match. We match on "riscv,plic0" for legacy systems, and I guess it makes sense to do something similar here.
IFF we decided to change it I'd rather separate DT noes for the ipi bank vs timecmp register vs timeval to support variable layouts. The downside is that we can't just boot on unmodified upstream qemu, which has used the "riscv,clint0" for years.
Like I alluded to above, matching on "riscv,clint0" seems reasonable to me as it's a defacto standard -- we'll just have to make sure that if we ever end up with a RISC-V CLINT that the DT entry is something else.
As far as splitting the memory maps goes, I don't have a strong opinion but it seems like that'll introduce more complexity than it's worth.
Re: [PATCH 08/15] riscv: provide native clint access for M-mode
Quoted 12 lines · Palmer Dabbelt
Quoted 11 lines · Christoph Hellwig
Quoted 9 lines · Palmer Dabbelt
clint0 would be version 0 of the clint, with is the core-local interrupt controller in rocket chip. It should be "sifive,clint-1.0.0", not "riscv,clint0", as it's a SiFive widget. Unfortunately there are a lot of legacy device trees floating around, but I'm only considering what's been upstream to be actually part of the spec.
In this case the code should match on a "sifive,clint-1.0.0", and the device trees should be fixed up to match. We match on "riscv,plic0" for legacy systems, and I guess it makes sense to do something similar here.
Quoted 8 lines
Quoted 4 lines
IFF we decided to change it I'd rather separate DT noes for the ipi bank vs timecmp register vs timeval to support variable layouts. The downside is that we can't just boot on unmodified upstream qemu, which has used the "riscv,clint0" for years.
Like I alluded to above, matching on "riscv,clint0" seems reasonable to me as it's a defacto standard -- we'll just have to make sure that if we ever end up with a RISC-V CLINT that the DT entry is something else.
De facto, but not mandatory.
Quoted 4 lines
As far as splitting the memory maps goes, I don't have a strong opinion but it seems like that'll introduce more complexity than it's worth.
At least the splitting can keep reminding us and any new comers in the future that CLINT is not (yet) a must in RISC-V landscape. A previous discussion
FYI: ( https://lkml.org/lkml/2019/8/20/1361 )
Quoted 4 lines
_______________________________________________
linux-riscv mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/linux-riscv
_______________________________________________
linux-riscv mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/linux-riscv
Re: [PATCH 08/15] riscv: provide native clint access for M-mode
In reply to Christoph HellwigQuoted 3 lines · Christoph Hellwig
RISC-V has the concept of a cpu level interrupt controller. Part of it
is expose as bits in the status registers, and 2 new CSRs per
//is expose/is to expose/
Quoted 63 lines
privilege
level in the instruction set, but the machanisms to trigger IPIs and timer events, as well as reading the actual timer value are not specified in the RISC-V spec but usually delegated to a block of MMIO registers. This patch adds support for those MMIO registers in the timer and IPI code. For now only the SiFive layout also supported by a few other implementations is supported, but the code should be easily extensible to others in the future.Signed-off-by Christoph Hellwig <[email protected]>--- arch/riscv/include/asm/clint.h | 40 +++++++++++++++++++++++++++ arch/riscv/include/asm/timex.h | 17 ++++++++++++ arch/riscv/kernel/Makefile | 1 + arch/riscv/kernel/clint.c | 45 +++++++++++++++++++++++++++++++ arch/riscv/kernel/setup.c | 2 ++ arch/riscv/kernel/smp.c | 24 +++++++++++++++++ arch/riscv/kernel/smpboot.c | 3 +++ drivers/clocksource/timer-riscv.c | 16 ++++++++--- 8 files changed, 144 insertions(+), 4 deletions(-) create mode 100644 arch/riscv/include/asm/clint.h create mode 100644 arch/riscv/kernel/clint.carch/riscv/include/asm/clint.h new file +32
-0@@ -0,0 +1,40 @@ +// SPDX-License-Identifier: GPL-2.0 +#ifndef _ASM_CLINT_H +#define _ASM_CLINT_H 1 + +#include <linux/smp.h> + +#ifdef CONFIG_M_MODE +extern u32 __iomem *clint_ipi_base; +extern u64 __iomem *clint_time_val; +extern u64 __iomem *clint_time_cmp; + +void clint_init_boot_cpu(void); + +static inline void clint_send_ipi(unsigned long hartid) +{ + writel(1, clint_ipi_base + hartid); +} + +static inline void clint_clear_ipi(unsigned long hartid) +{ + writel(0, clint_ipi_base + hartid); +} + +static inline u64 clint_read_timer(void) +{ + return readq_relaxed(clint_time_val); +} + +static inline void clint_set_timer(unsigned long delta) +{ + writeq_relaxed(clint_read_timer() + delta, + clint_time_cmp +
Quoted 1 line
cpuid_to_hartid_map(smp_processor_id()));'
This is not compatible with 32 bit mode. IIRC, timecmp is a 64 bit on RV32 as well. Here is the implementation in OpenSBI.
https://github.com/riscv/opensbi/blob/master/lib/utils/sys/clint.c#L104
Quoted 19 lines
arch/riscv/include/asm/clint.h new file +8
-0+} + +#else +#define clint_init_boot_cpu () do { } while (0) +#define clint_clear_ipi (hartid) do { } while (0) +#endif /* CONFIG_M_MODE */ + +#endif /* _ASM_CLINT_H */ arch/riscv/include/asm/timex.h +2
-0@@ -10,6 +10,22 @@typedef unsigned long cycles_t ; +#ifdef CONFIG_M_MODE +
arch/riscv/include/asm/timex.h line 24
Quoted 10 lines
arch/riscv/include/asm/timex.h +10
-0+#include <linux/io-64-nonatomic-lo-hi.h> +#include <asm/clint.h> + +static inline cycles_t get_cycles (void ) +{ +#ifdef CONFIG_64BIT +return readq_relaxed (clint_time_val); +#else +return readl_relaxed (clint_time_val); +#endif
Same comment as above. Both RV32 & RV64 bit have 64 bit have 64 bit precission for timer val. You have to read 32 bits at a time and "or" them to get 64 bit value. Here is the implementation from OpenSBI
https://github.com/riscv/opensbi/blob/master/lib/utils/sys/clint.c#L69
Quoted 222 lines
arch/riscv/include/asm/timex.h +5
-0+} +#define get_cycles get_cycles + +#else /* CONFIG_M_MODE */ static inline cycles_t get_cycles_inline (void ) {cycles_t n;@@ -40,6 +56,7 @@ static inline uint64_t get_cycles64(void)return ((u64 )hi <<32 ) | lo; }#endif +#endif /* CONFIG_M_MODE */ #define ARCH_HAS_READ_CURRENT_TIMER arch/riscv/kernel/Makefile +1
-0@@ -29,6 +29,7 @@ obj-y += vdso.o obj-y += cacheinfo.o obj-y += vdso/ +obj-$(CONFIG_M_MODE) += clint.o obj-$(CONFIG_FPU) += fpu.o obj-$(CONFIG_SMP) += smpboot.o obj-$(CONFIG_SMP) += smp.oarch/riscv/kernel/clint.c new file +14
-0@@ -0,0 +1,45 @@ +// SPDX-License-Identifier: GPL-2.0 +/* +* Copyright (c) 2019 Christoph Hellwig. +*/ + +#include <linux/io.h> +#include <linux/of_address.h> +#include <linux/types.h> +#include <asm/csr.h> +#include <asm/irq.h> +#include <asm/timex.h> + +/* +* This is the layout used by the SiFive clint, which is also shared by the qemu
+ * virt platform, and the Kendryte KD210 at least.
+ */arch/riscv/kernel/clint.c new file +29
-0+#define CLINT_IPI_OFF 0 +#define CLINT_TIME_VAL_OFF 0xbff8 +#define CLINT_TIME_CMP_OFF 0x4000; + +u32 __iomem *clint_ipi_base; +u64 __iomem *clint_time_val; +u64 __iomem *clint_time_cmp; + +void clint_init_boot_cpu (void ) +{ +struct device_node *np; +void __iomem *base; + + np =of_find_compatible_node (NULL ,NULL ,"riscv,clint0" ); +if (!np) { +panic ("clint not found" ); +return ; + } + + base =of_iomap (np,0 ); +if (!base) +panic ("could not map CLINT" ); + + clint_ipi_base = base +CLINT_IPI_OFF ; + clint_time_val = base +CLINT_TIME_VAL_OFF ; + clint_time_cmp = base +CLINT_TIME_CMP_OFF ; + +clint_clear_ipi (boot_cpu_hartid); +}arch/riscv/kernel/setup.c +2
-0@@ -17,6 +17,7 @@#include <linux/sched/task.h> #include <linux/swiotlb.h> +#include <asm/clint.h> #include <asm/setup.h> #include <asm/sections.h> #include <asm/pgtable.h> @@ -65,6 +66,7 @@ void __init setup_arch(char **cmdline_p)setup_bootmem ();paging_init ();unflatten_device_tree (); +clint_init_boot_cpu ();#ifdef CONFIG_SWIOTLB swiotlb_init (1 );arch/riscv/kernel/smp.c +23
-0@@ -13,7 +13,9 @@#include <linux/sched.h> #include <linux/seq_file.h> #include <linux/delay.h> +#include <linux/io.h> +#include <asm/clint.h> #include <asm/sbi.h> #include <asm/tlbflush.h> #include <asm/cacheflush.h> @@ -78,6 +80,27 @@ static void ipi_stop(void)wait_for_interrupt (); } +#ifdef CONFIG_M_MODE +static inline void send_ipi_single (int cpu,enum ipi_message_type op) +{ +set_bit (op, &ipi_data[cpu].bits); +clint_send_ipi (cpuid_to_hartid_map (cpu)); +} + +static inline void send_ipi_mask (const struct cpumask *mask, +enum ipi_message_type op) +{ +int cpu; + +for_each_cpu (cpu, mask) +send_ipi_single (cpu, op); +} + +static inline void clear_ipi (void ) +{ +clint_clear_ipi (cpuid_to_hartid_map (smp_processor_id ())); +} +#else /* CONFIG_M_MODE */ static void send_ipi_mask (const struct cpumask *mask,enum ipi_message_type op){
arch/riscv/kernel/smp.c +1
-0int cpuid, hartid;@@ -103,6 +126,7 @@ static inline void clear_ipi(void) {csr_clear (CSR_SIP ,SIE_SSIE ); } +#endif /* CONFIG_M_MODE */ void riscv_software_interrupt (void ) {arch/riscv/kernel/smpboot.c +3
-0@@ -23,6 +23,7 @@#include <linux/of.h> #include <linux/sched/task_stack.h> #include <linux/sched/mm.h> +#include <asm/clint.h> #include <asm/irq.h> #include <asm/mmu_context.h> #include <asm/tlbflush.h> @@ -132,6 +133,8 @@ asmlinkage void __init smp_callin(void) {struct mm_struct *mm = &init_mm; +clint_clear_ipi (cpuid_to_hartid_map (smp_processor_id ())); +/* All kernel threads share the same mm context. */ mmgrab (mm); current->active_mm = mm;drivers/clocksource/timer-riscv.c +0
-0@@ -24,12 +24,16 @@* operations on the current hart. There is guaranteed to be exactly one timer
* per hart on all RISC-V systems.
*/drivers/clocksource/timer-riscv.c +12
-4-static int riscv_clock_next_event (unsigned long delta,struct clock_event_device *ce) {csr_set (CSR_XIE ,XIE_XTIE ); + +#ifdef CONFIG_M_MODE +clint_set_timer (delta); +#else sbi_set_timer (get_cycles64 () + delta); +#endif return 0 ; }@@ -45,14 +49,18 @@ static DEFINE_PER_CPU(struct clock_event_device,riscv_clock_event) = { * within one tick of each other, so while this could technically go * backwards when hopping between CPUs, practically it won't happen. */ -static unsigned long long riscv_clocksource_rdtime (struct clocksource *cs) +static u64 riscv_sched_clock (void ) { +#ifdef CONFIG_M_MODE +return clint_read_timer (); +#else return get_cycles64 (); +#endif } -static u64 riscv_sched_clock (void ) +static unsigned long long riscv_clocksource_rdtime (struct clocksource *cs) { -return get_cycles64 (); +return riscv_sched_clock ();}
static struct clocksource
riscv_clocksource= {
Signature · 2 lines
Regards, Atish
Re: [PATCH 08/15] riscv: provide native clint access for M-mode
Quoted 8 lines · Atish Patra
Quoted 5 lines
arch/riscv/include/asm/clint.h new file +4
-0+static inline void clint_set_timer (unsigned long delta) +{ + writeq_relaxed(clint_read_timer() + delta, + clint_time_cmp +cpuid_to_hartid_map(smp_processor_id()));'
This is not compatible with 32 bit mode. IIRC, timecmp is a 64 bit on RV32 as well. Here is the implementation in OpenSBI.
writeq alwasy writes 64-bit anyway, but the deltas is just 32-bit per the Linux clocksource API.
Quoted 11 lines
arch/riscv/include/asm/timex.h line 24
Quoted 7 lines
arch/riscv/include/asm/timex.h +7
-0+static inline cycles_t get_cycles (void ) +{ +#ifdef CONFIG_64BIT +return readq_relaxed (clint_time_val); +#else +return readl_relaxed (clint_time_val); +#endif Same comment as above. Both RV32 & RV64 bit have 64 bit have 64 bit precission for timer val. You have to read 32 bits at a time and "or" them to get 64 bit value. Here is the implementation from OpenSBI
But the Linux API is only going to read 32-bits of that, same as for the rdtime pseudo-instruction used by the current SBI-based code.
Note that I've reworked this area a bit for v4, which I'm going to send out soon, including cleanups to the existing code to make a few of these things more obvious:
http://git.infradead.org/users/hch/riscv.git/shortlog/refs/heads/riscv-nommu.4
PATCH 9/15 riscv: implement remote sfence.i natively for M-mode
In reply to Christoph HellwigThe RISC-V ISA only supports flushing the instruction cache for the local CPU core. For normal S-mode Linux remote flushing is offloaded to machine mode using ecalls, but for M-mode Linux we'll have to do it ourselves. Use the same implementation as all the existing open source SBI implementations by just doing an IPI to all remote cores to execute th sfence.i instruction on every live core.
--- arch/riscv/mm/cacheflush.c | 31 +++++++++++++++++++++++++++---- 1 file changed, 27 insertions(+), 4 deletions(-)
arch/riscv/mm/cacheflush.c +27 -4
@@ -10,10 +10,35 @@#include <asm/sbi.h> +#ifdef CONFIG_M_MODE +static void ipi_remote_fence_i (void *info) +{ +return local_flush_icache_all (); +} + +void flush_icache_all (void ) +{ +on_each_cpu (ipi_remote_fence_i,NULL ,1 ); +} + +static void flush_icache_cpumask (const cpumask_t *mask) +{ +on_each_cpu_mask (mask, ipi_remote_fence_i,NULL ,1 ); +} +#else /* CONFIG_M_MODE */ void flush_icache_all (void ) {sbi_remote_fence_i (NULL ); } +static void flush_icache_cpumask (const cpumask_t *mask) +{ +cpumask_t hmask; + +cpumask_clear (&hmask); +riscv_cpuid_to_hartid_mask (mask, &hmask); +sbi_remote_fence_i (hmask.bits); +} +#endif /* CONFIG_M_MODE */ /* * Performs an icache flush for the given MM context. RISC-V has no direct
@@ -28,7 +53,7 @@ void flush_icache_all(void)void flush_icache_mm (struct mm_struct *mm,bool local) {unsigned int cpu; -cpumask_t others, hmask, *mask; +cpumask_t others, *mask;preempt_disable ();
@@ -47,9 +72,7 @@ void flush_icache_mm(struct mm_struct *mm, bool local)cpumask_andnot (&others,mm_cpumask (mm),cpumask_of (cpu)); local |=cpumask_empty (&others);if (mm != current->active_mm || !local) { -cpumask_clear (&hmask); -riscv_cpuid_to_hartid_mask (&others, &hmask); -sbi_remote_fence_i (hmask.bits); +flush_icache_cpumask (&others); }else {/* * It's assumed that at least one strongly ordered operation is
Signature · 1 line
2.20.1
Re: [PATCH 09/15] riscv: implement remote sfence.i natively for M-mode
Quoted 76 lines · Christoph Hellwig
The RISC-V ISA only supports flushing the instruction cache for the local
CPU core. For normal S-mode Linux remote flushing is offloaded to machine mode using ecalls, but for M-mode Linux we'll have to do it ourselves. Use the same implementation as all the existing open source
SBI implementations by just doing an IPI to all remote cores to execute
th sfence.i instruction on every live core.Signed-off-by Christoph Hellwig <[email protected]>--- arch/riscv/mm/cacheflush.c | 31 +++++++++++++++++++++++++++---- 1 file changed, 27 insertions(+), 4 deletions(-)arch/riscv/mm/cacheflush.c +25
-0@@ -10,10 +10,35 @@#include <asm/sbi.h> +#ifdef CONFIG_M_MODE +static void ipi_remote_fence_i (void *info) +{ +return local_flush_icache_all (); +} + +void flush_icache_all (void ) +{ +on_each_cpu (ipi_remote_fence_i,NULL ,1 ); +} + +static void flush_icache_cpumask (const cpumask_t *mask) +{ +on_each_cpu_mask (mask, ipi_remote_fence_i,NULL ,1 ); +} +#else /* CONFIG_M_MODE */ void flush_icache_all (void ) {sbi_remote_fence_i (NULL ); } +static void flush_icache_cpumask (const cpumask_t *mask) +{ +cpumask_t hmask; + +cpumask_clear (&hmask); +riscv_cpuid_to_hartid_mask (mask, &hmask); +sbi_remote_fence_i (hmask.bits); +} +#endif /* CONFIG_M_MODE */ /* * Performs an icache flush for the given MM context. RISC-V has no direct
arch/riscv/mm/cacheflush.c +2
-4@@ -28,7 +53,7 @@ void flush_icache_all(void)void flush_icache_mm (struct mm_struct *mm,bool local) {unsigned int cpu; -cpumask_t others, hmask, *mask; +cpumask_t others, *mask;preempt_disable ();@@ -47,9 +72,7 @@ void flush_icache_mm(struct mm_struct *mm, bool local) cpumask_andnot(&others, mm_cpumask(mm), cpumask_of(cpu)); local |=cpumask_empty (&others);if (mm != current->active_mm || !local) { -cpumask_clear (&hmask); -riscv_cpuid_to_hartid_mask (&others, &hmask); -sbi_remote_fence_i (hmask.bits); +flush_icache_cpumask (&others); }else {/*
Quoted 2 lines
* It's assumed that at least one strongly ordered
operation is
Signature · 2 lines
Regards, Atish
PATCH 10/15 riscv: poison SBI calls for M-mode
In reply to Christoph HellwigThere is no SBI when we run in M-mode, so fail the compile for any code trying to use SBI calls.
--- arch/riscv/include/asm/sbi.h | 4 +++- 1 file changed, 3 insertions(+), 1 deletion(-)
arch/riscv/include/asm/sbi.h +3 -1
@@ -8,6 +8,7 @@#include <linux/types.h> +#ifndef CONFIG_M_MODE #define SBI_SET_TIMER 0#define SBI_CONSOLE_PUTCHAR 1#define SBI_CONSOLE_GETCHAR 2
@@ -94,4 +95,5 @@ static inline void sbi_remote_sfence_vma_asid(const unsigned long *hart_mask,SBI_CALL_4 (SBI_REMOTE_SFENCE_VMA_ASID , hart_mask, start, size, asid); } -#endif +#endif /* CONFIG_M_MODE */ +#endif /* _ASM_RISCV_SBI_H */
Signature · 1 line
2.20.1
Re: [PATCH 10/15] riscv: poison SBI calls for M-mode
Quoted 23 lines · Christoph Hellwig
There is no SBI when we run in M-mode, so fail the compile for any code
trying to use SBI calls.Signed-off-by Christoph Hellwig <[email protected]>--- arch/riscv/include/asm/sbi.h | 4 +++- 1 file changed, 3 insertions(+), 1 deletion(-)arch/riscv/include/asm/sbi.h +1
-0@@ -8,6 +8,7 @@#include <linux/types.h> +#ifndef CONFIG_M_MODE #define SBI_SET_TIMER 0#define SBI_CONSOLE_PUTCHAR 1#define SBI_CONSOLE_GETCHAR 2
arch/riscv/include/asm/sbi.h static inline void sbi_remote_sfence_vma_asid(const · line 99
Quoted 8 lines
arch/riscv/include/asm/sbi.h +2
-1@@ -94,4 +95,5 @@ static inline void sbi_remote_sfence_vma_asid(constunsigned long *hart_mask,SBI_CALL_4 (SBI_REMOTE_SFENCE_VMA_ASID ,hart_mask ,start ,size ,asid ); } -#endif +#endif /* CONFIG_M_MODE */ +#endif /* _ASM_RISCV_SBI_H */
Signature · 2 lines
Regards, Atish
PATCH 11/15 riscv: don't allow selecting SBI-based drivers for M-mode
In reply to Christoph HellwigFrom: Damien Le Moal <[email protected]>
Do not allow selecting SBI related options with MMU option not set.
--- drivers/tty/hvc/Kconfig | 2 +- drivers/tty/serial/Kconfig | 2 +- 2 files changed, 2 insertions(+), 2 deletions(-)
drivers/tty/hvc/Kconfig +1 -1
@@ -89,7 +89,7 @@ config HVC_DCC config HVC_RISCV_SBI bool "RISC-V SBI console support" - depends on RISCV + depends on RISCV && !M_MODE select HVC_DRIVER help This enables support for console output via RISC-V SBI calls, which
drivers/tty/serial/Kconfig +1 -1
@@ -88,7 +88,7 @@ config SERIAL_EARLYCON_ARM_SEMIHOST config SERIAL_EARLYCON_RISCV_SBI bool "Early console using RISC-V SBI" - depends on RISCV + depends on RISCV && !M_MODE select SERIAL_CORE select SERIAL_CORE_CONSOLE select SERIAL_EARLYCON
Signature · 7 lines
2.20.1 _______________________________________________ linux-riscv mailing list [email protected] http://lists.infradead.org/mailman/listinfo/linux-riscv
PATCH 12/15 riscv: use the correct interrupt levels for M-mode
In reply to Christoph HellwigThe numerical levels for External/Timer/Software interrupts differ between S-mode and M-mode.
--- arch/riscv/kernel/irq.c | 12 +++++++++--- 1 file changed, 9 insertions(+), 3 deletions(-)
arch/riscv/kernel/irq.c +9 -3
@@ -14,9 +14,15 @@/* * Possible interrupt causes: */ -#define INTERRUPT_CAUSE_SOFTWARE IRQ_S_SOFT -#define INTERRUPT_CAUSE_TIMER IRQ_S_TIMER -#define INTERRUPT_CAUSE_EXTERNAL IRQ_S_EXT +#ifdef CONFIG_M_MODE +# define INTERRUPT_CAUSE_SOFTWARE IRQ_M_SOFT +# define INTERRUPT_CAUSE_TIMER IRQ_M_TIMER +# define INTERRUPT_CAUSE_EXTERNAL IRQ_M_EXT +#else +# define INTERRUPT_CAUSE_SOFTWARE IRQ_S_SOFT +# define INTERRUPT_CAUSE_TIMER IRQ_S_TIMER +# define INTERRUPT_CAUSE_EXTERNAL IRQ_S_EXT +#endif /* CONFIG_M_MODE */ int arch_show_interrupts (struct seq_file *p,int prec) {
Signature · 1 line
2.20.1
PATCH 13/15 riscv: clear the instruction cache and all registers when booting
In reply to Christoph HellwigWhen we get booted we want a clear slate without any leaks from previous supervisors or the firmware. Flush the instruction cache and then clear all registers to known good values. This is really important for the upcoming nommu support that runs on M-mode, but can't really harm when running in S-mode either. Vaguely based on the concepts from opensbi.
--- arch/riscv/include/asm/csr.h | 1 + arch/riscv/kernel/head.S | 86 ++++++++++++++++++++++++++++++++++++ 2 files changed, 87 insertions(+)
arch/riscv/include/asm/csr.h +1 -0
@@ -83,6 +83,7 @@/* symbolic CSR names: */ #define CSR_MHARTID 0xf14#define CSR_MSTATUS 0x300 +#define CSR_MISA 0x301#define CSR_MIE 0x304#define CSR_MTVEC 0x305#define CSR_MSCRATCH 0x340
arch/riscv/kernel/head.S +86 -0
104 lines not on this page · raw message
Signature · 1 line
2.20.1
Re: [PATCH 13/15] riscv: clear the instruction cache and all registers when booting
Hi Christoph,
Quoted 5 lines · Christoph Hellwig
When we get booted we want a clear slate without any leaks from previous supervisors or the firmware. Flush the instruction cache and then clear all registers to known good values. This is really important for the upcoming nommu support that runs on M-mode, but can't really harm when running in S-mode either.
Sure.
Quoted 1 line
+#ifdef
CONFIG_FPU
But it doesn't really mean that the running system has an FPU given CONFIG_FPU enabled. Normally the existence of an FPU is checked in riscv_fill_hwcap by searching device tree, where the code looks like
bool has_fpu = false; // this one is global
...
#ifdef CONFIG_FPU
if (elf_hwcap & (COMPAT_HWCAP_ISA_F | COMPAT_HWCAP_ISA_D))
has_fpu = true;
#endif
Or does CONFIG_FPU have a more intuitive meaning when CONFIG_M_MODE is enabled?
Quoted 31 lines
arch/riscv/kernel/head.S +31
-0+ csrr t0, CSR_MISA + andi t0, t0, (COMPAT_HWCAP_ISA_F | COMPAT_HWCAP_ISA_D) + bnez t0, .Lreset_regs_done + + li t1, SR_FS + csrs CSR_XSTATUS, t1 + fmv.s.x f0, zero + fmv.s.x f1, zero + fmv.s.x f2, zero + fmv.s.x f3, zero + fmv.s.x f4, zero + fmv.s.x f5, zero + fmv.s.x f6, zero + fmv.s.x f7, zero + fmv.s.x f8, zero + fmv.s.x f9, zero + fmv.s.x f10, zero + fmv.s.x f11, zero + fmv.s.x f12, zero + fmv.s.x f13, zero + fmv.s.x f14, zero + fmv.s.x f15, zero + fmv.s.x f16, zero + fmv.s.x f17, zero + fmv.s.x f18, zero + fmv.s.x f19, zero + fmv.s.x f20, zero + fmv.s.x f21, zero + fmv.s.x f22, zero + fmv.s.x f23, zero + fmv.s.x f24, zero
arch/riscv/kernel/head.S relocate: · line 284
Quoted 10 lines
arch/riscv/kernel/head.S +10
-0+ fmv.s.x f25, zero + fmv.s.x f26, zero + fmv.s.x f27, zero + fmv.s.x f28, zero + fmv.s.x f29, zero + fmv.s.x f30, zero + fmv.s.x f31, zero + csrw fcsr, 0 + /* note that the caller must clear SR_FS */ +#endif /* CONFIG_FPU */
_______________________________________________
linux-riscv mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/linux-riscv
Re: [PATCH 13/15] riscv: clear the instruction cache and all registers when booting
Please ignore the previous mail, I must have missed this part of the patch,
Quoted 5 lines
arch/riscv/kernel/head.S relocate: · line 247
Quoted 5 lines
arch/riscv/kernel/head.S +4
-0+ csrr t0, CSR_MISA + andi t0, t0, (COMPAT_HWCAP_ISA_F | COMPAT_HWCAP_ISA_D) + bnez t0, .Lreset_regs_done +
In S-mode we were not able to obtain the ISA information in misa, but now the nommu port is in M-mode so this is rather straightforward.
_______________________________________________
linux-riscv mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/linux-riscv
Re: [PATCH 13/15] riscv: clear the instruction cache and all registers when booting
In reply to Alan KaoQuoted 13 lines · Alan Kao
But it doesn't really mean that the running system has an FPU given
CONFIG_FPUenabled. Normally the existence of an FPU is checked inriscv_fill_hwcapby searching device tree, where the code looks likebool
has_fpu= false; // this one is global
...
#ifdefCONFIG_FPU
if (elf_hwcap& (COMPAT_HWCAP_ISA_F|COMPAT_HWCAP_ISA_D))
has_fpu= true;
#endif
Quoted 1 line
Or does
CONFIG_FPUhave a more intuitive meaning whenCONFIG_M_MODEis enabled?
No, it doesn't..
Quoted 4 lines
arch/riscv/kernel/head.S relocate: · line 246
Quoted 4 lines
arch/riscv/kernel/head.S +3
-0+ csrr t0, CSR_MISA + andi t0, t0, (COMPAT_HWCAP_ISA_F | COMPAT_HWCAP_ISA_D) + bnez t0, .Lreset_regs_done
... which is why we have these few lines of code that check the caps returns from the misa CSR, similar to the elf_caps check quoted above.
PATCH 14/15 riscv: add nommu support
In reply to Christoph HellwigThe kernel runs in M-mode without using page tables, and thus can't run bare metal without help from additional firmware.
Most of the patch is just stubbing out code not needed without page tables, but there is an interesting detail in the signals implementation:
- The normal RISC-V syscall ABI only implements rt_sigreturn as VDSO entry point, but the ELF VDSO is not supported for nommu Linux. We instead copy the code to call the syscall onto the stack.
In addition to enabling the nommu code a new defconfig for a small kernel image that can run in nommu mode on qemu is also provided, to run a kernel in qemu you can use the following command line:
qemu-system-riscv64 -smp 2 -m 64 -machine virt -nographic \ -kernel arch/riscv/boot/loader \ -drive file=rootfs.ext2,format=raw,id=hd0 \ -device virtio-blk-device,drive=hd0
Contains contributions from Damien Le Moal <[email protected]>.
--- arch/riscv/Kconfig | 24 +++++--- arch/riscv/configs/nommu_virt_defconfig | 78 +++++++++++++++++++++++++ arch/riscv/include/asm/cache.h | 8 +++ arch/riscv/include/asm/elf.h | 4 +- arch/riscv/include/asm/futex.h | 6 ++ arch/riscv/include/asm/io.h | 4 ++ arch/riscv/include/asm/mmu.h | 3 + arch/riscv/include/asm/page.h | 10 +++- arch/riscv/include/asm/pgalloc.h | 2 + arch/riscv/include/asm/pgtable.h | 38 ++++++++---- arch/riscv/include/asm/tlbflush.h | 7 ++- arch/riscv/include/asm/uaccess.h | 4 ++ arch/riscv/kernel/Makefile | 3 +- arch/riscv/kernel/entry.S | 11 ++++ arch/riscv/kernel/head.S | 6 ++ arch/riscv/kernel/signal.c | 17 +++++- arch/riscv/lib/Makefile | 11 ++-- arch/riscv/mm/Makefile | 3 +- arch/riscv/mm/cacheflush.c | 2 + arch/riscv/mm/context.c | 2 + arch/riscv/mm/init.c | 13 ++++- 21 files changed, 218 insertions(+), 38 deletions(-) create mode 100644 arch/riscv/configs/nommu_virt_defconfig
arch/riscv/Kconfig +16 -8
@@ -26,13 +26,13 @@ config RISCV select GENERIC_IRQ_SHOW select GENERIC_PCI_IOMAP select GENERIC_SCHED_CLOCK - select GENERIC_STRNCPY_FROM_USER - select GENERIC_STRNLEN_USER + select GENERIC_STRNCPY_FROM_USER if MMU + select GENERIC_STRNLEN_USER if MMU select GENERIC_SMP_IDLE_THREAD select GENERIC_ATOMIC64 if !64BIT select HAVE_ARCH_AUDITSYSCALL select HAVE_MEMBLOCK_NODE_MAP - select HAVE_DMA_CONTIGUOUS + select HAVE_DMA_CONTIGUOUS if MMU select HAVE_FUTEX_CMPXCHG if FUTEX select HAVE_PERF_EVENTS select HAVE_SYSCALL_TRACEPOINTS
@@ -47,6 +47,7 @@ config RISCV select PCI_DOMAINS_GENERIC if PCI select PCI_MSI if PCI select RISCV_TIMER + select UACCESS_MEMCPY if !MMU select GENERIC_IRQ_MULTI_HANDLER select ARCH_HAS_PTE_SPECIAL select ARCH_HAS_MMIOWB
@@ -58,9 +59,14 @@ config RISCV # set if we run in machine mode, cleared if we run in supervisor mode config M_MODE bool + default y if !MMU config MMU - def_bool y + bool "MMU-based Paged Memory Management Support" + default y + help + Select if you want MMU-based virtualised addressing space + support by paged memory management. If unsure, say 'Y'. config ZONE_DMA32 bool
@@ -69,6 +75,7 @@ config ZONE_DMA32 config PAGE_OFFSET hex default 0xC0000000 if 32BIT && MAXPHYSMEM_2GB + default 0x80000000 if 64BIT && !MMU default 0xffffffff80000000 if 64BIT && MAXPHYSMEM_2GB default 0xffffffe000000000 if 64BIT && MAXPHYSMEM_128GB
@@ -102,7 +109,7 @@ config GENERIC_HWEIGHT def_bool y config FIX_EARLYCON_MEM - def_bool y + def_bool CONFIG_MMU config PGTABLE_LEVELS int
@@ -127,6 +134,7 @@ config ARCH_RV32I select GENERIC_LIB_ASHRDI3 select GENERIC_LIB_LSHRDI3 select GENERIC_LIB_UCMPDI2 + select MMU config ARCH_RV64I bool "RV64I"
@@ -135,9 +143,9 @@ config ARCH_RV64I select HAVE_FUNCTION_TRACER select HAVE_FUNCTION_GRAPH_TRACER select HAVE_FTRACE_MCOUNT_RECORD - select HAVE_DYNAMIC_FTRACE - select HAVE_DYNAMIC_FTRACE_WITH_REGS - select SWIOTLB + select HAVE_DYNAMIC_FTRACE if MMU + select HAVE_DYNAMIC_FTRACE_WITH_REGS if HAVE_DYNAMIC_FTRACE + select SWIOTLB if MMU endchoice
arch/riscv/configs/nommu_virt_defconfig new file +78 -0
78 lines not on this page · raw message
arch/riscv/include/asm/cache.h +8 -0
@@ -11,4 +11,12 @@#define L1_CACHE_BYTES (1 << L1_CACHE_SHIFT) +/* +* RISC-V requires the stack pointer to be 16-byte aligned, so ensure that +* the flat loader aligns it accordingly. +*/ +#ifndef CONFIG_MMU +#define ARCH_SLAB_MINALIGN 16 +#endif +#endif /* _ASM_RISCV_CACHE_H */
arch/riscv/include/asm/elf.h +2 -2
@@ -56,16 +56,16 @@ extern unsigned long elf_hwcap;*/ #define ELF_PLATFORM (NULL) +#ifdef CONFIG_MMU #define ARCH_DLINFO \ do { \ NEW_AUX_ENT(AT_SYSINFO_EHDR, \ (elf_addr_t)current->mm->context.vdso); \ } while (0) - -#define ARCH_HAS_SETUP_ADDITIONAL_PAGES struct linux_binprm ;extern int arch_setup_additional_pages (struct linux_binprm *bprm,int uses_interp); +#endif /* CONFIG_MMU */ #endif /* _ASM_RISCV_ELF_H */
arch/riscv/include/asm/futex.h +6 -0
@@ -12,6 +12,12 @@#include <linux/errno.h> #include <asm/asm.h> +/* We don't even really need the extable code, but for now keep it simple */ +#ifndef CONFIG_MMU +#define __enable_user_access () do { } while (0) +#define __disable_user_access () do { } while (0) +#endif +#define __futex_atomic_op (insn, ret, oldval, uaddr, oparg) \ { \ uintptr_t tmp; \
arch/riscv/include/asm/io.h +4 -0
@@ -14,6 +14,7 @@#include <linux/types.h> #include <asm/mmiowb.h> +#ifdef CONFIG_MMU extern void __iomem *ioremap (phys_addr_t offset,unsigned long size);/*
@@ -26,6 +27,9 @@ extern void __iomem *ioremap(phys_addr_t offset, unsigned long size);#define ioremap_wt (addr, size) ioremap((addr), (size))extern void iounmap (volatile void __iomem *addr); +#else +#define pgprot_noncached (x) (x) +#endif /* CONFIG_MMU */ /* Generic IO read/write. These perform native-endian accesses. */ #define __raw_writeb __raw_writeb
arch/riscv/include/asm/mmu.h +3 -0
@@ -10,6 +10,9 @@ #ifndef __ASSEMBLY__ typedef struct { +#ifndef CONFIG_MMU + unsigned long end_brk; +#endif void *vdso; #ifdef CONFIG_SMP /* A local icache flush is needed before user execution can resume. */
arch/riscv/include/asm/page.h +7 -3
@@ -88,8 +88,14 @@ typedef struct page *pgtable_t;#define PTE_FMT "%08lx"#endif +#ifdef CONFIG_MMU extern unsigned long va_pa_offset;extern unsigned long pfn_base; +#define ARCH_PFN_OFFSET (pfn_base) +#else +#define va_pa_offset 0 +#define ARCH_PFN_OFFSET (PAGE_OFFSET >> PAGE_SHIFT) +#endif /* CONFIG_MMU */ extern unsigned long max_low_pfn;extern unsigned long min_low_pfn;
@@ -111,9 +117,7 @@ extern unsigned long min_low_pfn;#define phys_to_page (paddr) (pfn_to_page(phys_to_pfn(paddr)))#define pfn_valid (pfn) \ - (((pfn) >= pfn_base) && (((pfn)-pfn_base) < max_mapnr)) - -#define ARCH_PFN_OFFSET (pfn_base) + (((pfn) >= ARCH_PFN_OFFSET) && (((pfn) - ARCH_PFN_OFFSET) < max_mapnr))#endif /* __ASSEMBLY__ */
arch/riscv/include/asm/pgalloc.h +2 -0
@@ -10,6 +10,7 @@#include <linux/mm.h> #include <asm/tlb.h> +#ifdef CONFIG_MMU #include <asm-generic/pgalloc.h> /* for pte_{alloc,free}_one */ static inline void pmd_populate_kernel(struct mm_struct *mm,
@@ -81,6 +82,7 @@ do { \pgtable_page_dtor (pte); \tlb_remove_page ((tlb), pte); \ } while (0) +#endif /* CONFIG_MMU */ static inline void check_pgt_cache (void ) {
arch/riscv/include/asm/pgtable.h +26 -12
@@ -24,6 +24,7 @@#include <asm/pgtable-32.h> #endif /* CONFIG_64BIT */ +#ifdef CONFIG_MMU /* Number of entries in the page global directory */ #define PTRS_PER_PGD (PAGE_SIZE / sizeof(pgd_t))/* Number of entries in the page table */
@@ -31,7 +32,6 @@/* Number of PGD entries that a user-mode program can use */ #define USER_PTRS_PER_PGD (TASK_SIZE / PGDIR_SIZE) -#define FIRST_USER_ADDRESS 0/* Page protection bits */ #define _PAGE_BASE (_PAGE_PRESENT | _PAGE_ACCESSED | _PAGE_USER)
@@ -83,13 +83,6 @@ extern pgd_t swapper_pg_dir[];#define __S110 PAGE_SHARED_EXEC#define __S111 PAGE_SHARED_EXEC -/* -* ZERO_PAGE is a global shared page that is always zero, -* used for zero-mapped memory areas, etc. -*/ -extern unsigned long empty_zero_page[PAGE_SIZE /sizeof (unsigned long )]; -#define ZERO_PAGE (vaddr) (virt_to_page(empty_zero_page)) -static inline int pmd_present (pmd_t pmd) {return (pmd_val (pmd) & (_PAGE_PRESENT | _PAGE_PROT_NONE));
@@ -403,6 +396,18 @@ static inline int ptep_clear_flush_young(struct vm_area_struct *vma,#define __pte_to_swp_entry (pte) ((swp_entry_t) { pte_val(pte) })#define __swp_entry_to_pte (x) ((pte_t) { (x).val }) +#define VMALLOC_SIZE (KERN_VIRT_SIZE >> 1) +#define VMALLOC_END (PAGE_OFFSET - 1) +#define VMALLOC_START (PAGE_OFFSET - VMALLOC_SIZE) + +#else /* CONFIG_MMU */ + +#define PAGE_KERNEL __pgprot(0) +#define swapper_pg_dir NULL +#define VMALLOC_START 0 + +#endif /* CONFIG_MMU */ +#ifdef CONFIG_FLATMEM #define kern_addr_valid (addr) (1)/* FIXME */ #endif
@@ -416,10 +421,7 @@ static inline void pgtable_cache_init(void)/* No page table caches to initialize */ } -#define VMALLOC_SIZE (KERN_VIRT_SIZE >> 1) -#define VMALLOC_END (PAGE_OFFSET - 1) -#define VMALLOC_START (PAGE_OFFSET - VMALLOC_SIZE) - +#ifdef CONFIG_MMU /* * Task size is 0x4000000000 for RV64 or 0xb800000 for RV32. * Note that PGDIR_SIZE must evenly divide TASK_SIZE.
@@ -429,6 +431,18 @@ static inline void pgtable_cache_init(void)#else #define TASK_SIZE VMALLOC_START#endif +#else /* CONFIG_MMU */ +#define TASK_SIZE 0xffffffffUL +#endif /* !CONFIG_MMU */ + +#define FIRST_USER_ADDRESS 0 + +/* +* ZERO_PAGE is a global shared page that is always zero, +* used for zero-mapped memory areas, etc. +*/ +extern unsigned long empty_zero_page[PAGE_SIZE /sizeof (unsigned long )]; +#define ZERO_PAGE (vaddr) (virt_to_page(empty_zero_page))#include <asm-generic/pgtable.h>
arch/riscv/include/asm/tlbflush.h +6 -1
@@ -10,6 +10,7 @@#include <linux/mm_types.h> #include <asm/smp.h> +#ifdef CONFIG_MMU /* * Flush entire local TLB. 'sfence.vma' implicitly fences with the instruction * cache as well, so a 'fence.i' is not necessary.
@@ -24,8 +25,12 @@ static inline void local_flush_tlb_page(unsigned long addr) { __asm__ __volatile__ ("sfence.vma %0" : :"r" (addr) :"memory" ); } +#else /* CONFIG_MMU */ +#define local_flush_tlb_all () do { } while (0) +#define local_flush_tlb_page (addr) do { } while (0) +#endif /* CONFIG_MMU */ -#ifndef CONFIG_SMP +#if !defined(CONFIG_SMP ) || !defined(CONFIG_MMU )#define flush_tlb_all () local_flush_tlb_all()#define flush_tlb_page (vma, addr) local_flush_tlb_page(addr)
arch/riscv/include/asm/uaccess.h +4 -0
@@ -11,6 +11,7 @@/* * User space memory access functions */ +#ifdef CONFIG_MMU #include <linux/errno.h> #include <linux/compiler.h> #include <linux/thread_info.h>
@@ -475,4 +476,7 @@ unsigned long __must_check clear_user(void __user *to, unsigned long n) __ret; \ }) +#else /* CONFIG_MMU */ +#include <asm-generic/uaccess.h> +#endif /* CONFIG_MMU */ #endif /* _ASM_RISCV_UACCESS_H */
arch/riscv/kernel/Makefile +1 -2
@@ -25,9 +25,8 @@ obj-y += time.o obj-y += traps.o obj-y += riscv_ksyms.o obj-y += stacktrace.o -obj-y += vdso.o obj-y += cacheinfo.o -obj-y += vdso/ +obj-$(CONFIG_MMU) += vdso.o vdso/ obj-$(CONFIG_M_MODE) += clint.o obj-$(CONFIG_FPU) += fpu.o
arch/riscv/kernel/entry.S +11 -0
@@ -370,6 +370,10 @@ ENTRY(__switch_to) ret ENDPROC(__switch_to) +#ifndef CONFIG_MMU +#define do_page_fault do_trap_unknown +#endif + .section ".rodata" /* Exception vector table */ ENTRY(excp_vect_table)
@@ -391,3 +395,10 @@ ENTRY(excp_vect_table) RISCV_PTR do_page_fault /* store page fault */ excp_vect_table_end: END(excp_vect_table) + +#ifndef CONFIG_MMU +ENTRY(__user_rt_sigreturn) + li a7, __NR_rt_sigreturn + scall +END(__user_rt_sigreturn) +#endif
arch/riscv/kernel/head.S +6 -0
@@ -104,8 +104,10 @@ clear_bss_done: la sp, init_thread_union + THREAD_SIZE mv a0, s1 call setup_vm +#ifdef CONFIG_MMU la a0, early_pg_dir call relocate +#endif /* CONFIG_MMU */ /* Restore C environment */ la tp, init_task
@@ -116,6 +118,7 @@ clear_bss_done: call parse_dtb tail start_kernel +#ifdef CONFIG_MMU relocate: /* Relocate return address */ li a1, PAGE_OFFSET
@@ -166,6 +169,7 @@ relocate: sfence.vma ret +#endif /* CONFIG_MMU */ .Lsecondary_start: #ifdef CONFIG_SMP
@@ -194,9 +198,11 @@ relocate: beqz tp, .Lwait_for_cpu_up fence +#ifdef CONFIG_MMU /* Enable virtual memory and relocate to virtual address */ la a0, swapper_pg_dir call relocate +#endif tail smp_callin #endif
arch/riscv/kernel/signal.c +16 -1
@@ -17,11 +17,16 @@#include <asm/switch_to.h> #include <asm/csr.h> +extern u32 __user_rt_sigreturn[2 ]; +#define DEBUG_SIG 0struct rt_sigframe {struct siginfo info;struct ucontext uc; +#ifndef CONFIG_MMU +u32 sigreturn_code[2 ]; +#endif };#ifdef CONFIG_FPU
@@ -166,7 +171,6 @@ static inline void __user *get_sigframe(struct ksignal *ksig,return (void __user *)sp; } -static int setup_rt_frame (struct ksignal *ksig,sigset_t *set,struct pt_regs *regs) {
@@ -189,8 +193,19 @@ static int setup_rt_frame(struct ksignal *ksig, sigset_t *set,return -EFAULT ;/* Set up to return from userspace. */ +#ifdef CONFIG_MMU regs->ra = (unsigned long )VDSO_SYMBOL ( current->mm->context.vdso, rt_sigreturn); +#else +/* +* For the nommu case we don't have a VDSO. Instead we push two +* instructions to call the rt_sigreturn syscall onto the user stack. +*/ +if (copy_to_user (&frame->sigreturn_code, __user_rt_sigreturn, +sizeof (frame->sigreturn_code))) +return -EFAULT ; + regs->ra = (unsigned long )&frame->sigreturn_code;; +#endif /* CONFIG_MMU */ /* * Set up registers for signal handler.
arch/riscv/lib/Makefile +5 -6
@@ -1,7 +1,6 @@ # SPDX-License-Identifier: GPL-2.0-only -lib-y += delay.o -lib-y += memcpy.o -lib-y += memset.o -lib-y += uaccess.o - -lib-$(CONFIG_64BIT) += tishift.o +lib-y += delay.o +lib-y += memcpy.o +lib-y += memset.o +lib-$(CONFIG_MMU) += uaccess.o +lib-$(CONFIG_64BIT) += tishift.o
arch/riscv/mm/Makefile +1 -2
@@ -6,9 +6,8 @@ CFLAGS_REMOVE_init.o = -pg endif obj-y += init.o -obj-y += fault.o obj-y += extable.o -obj-y += ioremap.o +obj-$(CONFIG_MMU) += fault.o ioremap.o obj-y += cacheflush.o obj-y += context.o obj-y += sifive_l2_cache.o
arch/riscv/mm/cacheflush.c +2 -0
@@ -90,6 +90,7 @@ void flush_icache_mm(struct mm_struct *mm, bool local)#endif /* CONFIG_SMP */ +#ifdef CONFIG_MMU void flush_icache_pte (pte_t pte) {struct page *page =pte_page (pte);
@@ -97,3 +98,4 @@ void flush_icache_pte(pte_t pte)if (!test_and_set_bit (PG_dcache_clean, &page->flags)) flush_icache_all(); } +#endif /* CONFIG_MMU */
arch/riscv/mm/context.c +2 -0
@@ -57,8 +57,10 @@ void switch_mm(struct mm_struct *prev, struct mm_struct *next,cpumask_clear_cpu (cpu,mm_cpumask (prev));cpumask_set_cpu (cpu,mm_cpumask (next)); +#ifdef CONFIG_MMU csr_write (CSR_SATP ,virt_to_pfn (next->pgd) |SATP_MODE );local_flush_tlb_all (); +#endif flush_icache_deferred (next); }
arch/riscv/mm/init.c +12 -1
@@ -23,6 +23,7 @@ unsigned long empty_zero_page[PAGE_SIZE / sizeof(unsigned long)]EXPORT_SYMBOL (empty_zero_page);extern char _start[]; +void *dtb_early_va;static void __init zone_sizes_init (void ) {
@@ -132,12 +133,12 @@ void __init setup_bootmem(void) } } +#ifdef CONFIG_MMU unsigned long va_pa_offset;EXPORT_SYMBOL (va_pa_offset);unsigned long pfn_base;EXPORT_SYMBOL (pfn_base); -void *dtb_early_va; pgd_t swapper_pg_dir[PTRS_PER_PGD] __page_aligned_bss; pgd_t trampoline_pg_dir[PTRS_PER_PGD] __page_aligned_bss; pte_t fixmap_pte[PTRS_PER_PTE] __page_aligned_bss;
@@ -438,6 +439,16 @@ static void __init setup_vm_final(void)csr_write (sptbr,PFN_DOWN (__pa (swapper_pg_dir)) |SATP_MODE );local_flush_tlb_all (); } +#else +asmlinkage void __init setup_vm(uintptr_t dtb_pa) +{ + dtb_early_va = (void *)dtb_pa; +} + +static inline void setup_vm_final (void ) +{ +} +#endif /* CONFIG_MMU */ void __init paging_init (void ) {
Signature · 1 line
2.20.1
PATCH 15/15 riscv: disable the EFI PECOFF header for M-mode
In reply to Christoph HellwigNo point in bloating the kernel image with a bootloader header if we run bare metal.
--- arch/riscv/kernel/head.S | 2 ++ 1 file changed, 2 insertions(+)
arch/riscv/kernel/head.S +2 -0
@@ -16,6 +16,7 @@ __INIT ENTRY(_start) +#ifndef CONFIG_M_MODE /* * Image header expected by Linux boot-loaders. The image header data * structure is described in asm/image.h.
@@ -47,6 +48,7 @@ ENTRY(_start) .global _start_kernel _start_kernel: +#endif /* CONFIG_M_MODE */ /* Mask all interrupts */ csrw CSR_XIE, zero csrw CSR_XIP, zero
Signature · 1 line
2.20.1
Re: [PATCH 15/15] riscv: disable the EFI PECOFF header for M-mode
Quoted 22 lines · Christoph Hellwig
No point in bloating the kernel image with a bootloader header if we run bare metal.
Signed-off-by Christoph Hellwig <[email protected]>--- arch/riscv/kernel/head.S | 2 ++ 1 file changed, 2 insertions(+)arch/riscv/kernel/head.S +1
-0@@ -16,6 +16,7 @@ __INIT ENTRY(_start) +#ifndef CONFIG_M_MODE /* * Image header expected by Linux boot-loaders. The imageheader data
arch/riscv/kernel/head.S +0
-0* structure is described in asm/image.h.
arch/riscv/kernel/head.S ENTRY(_start) · line 54
Quoted 7 lines
arch/riscv/kernel/head.S +1
-0@@ -47,6 +48,7 @@ ENTRY(_start) .global _start_kernel _start_kernel: +#endif /* CONFIG_M_MODE */ /* Mask all interrupts */ csrw CSR_XIE, zero csrw CSR_XIP, zero
Signature · 2 lines
Regards, Atish
Re: [PATCH 15/15] riscv: disable the EFI PECOFF header for M-mode
In reply to Christoph HellwigQuoted 6 lines · Christoph Hellwig
No point in bloating the kernel image with a bootloader header if we run bare metal.
I would say the same for S-mode. EFI booting should be an option, not a requirement. I have M-mode U-boot working with bootelf to start BBL, and at some point, I’m hoping we can have a M-mode linux kernel be the SBI provider for S-mode kernels, which seem most logical to me to start using the vmlinux elf binaries using something like kexec()
Quoted 34 lines
Signed-off-by Christoph Hellwig <[email protected]>---
arch/riscv/kernel/head.S| 2 ++1 file changed, 2 insertions(+)arch/riscv/kernel/head.S +0
-0@@ -16,6 +16,7 @@
__INITENTRY(_start)
+#ifndefCONFIG_M_MODE
/*
* Image header expected by Linux boot-loaders. The image header data
* structure is described inasm/image.h.arch/riscv/kernel/head.S +0
-0@@ -47,6 +48,7 @@ ENTRY(_start).global
_start_kernel_start_kernel:
+#endif /*CONFIG_M_MODE*/
/* Mask all interrupts */
csrwCSR_XIE, zero
csrwCSR_XIP, zeroSignature · 7 lines
2.20.1 _______________________________________________ linux-riscv mailing list [email protected] http://lists.infradead.org/mailman/listinfo/linux-riscv
Re: [PATCH 15/15] riscv: disable the EFI PECOFF header for M-mode
Quoted 4 lines · Troy Benjegerdes
Quoted 4 lines · Christoph Hellwig
Quoted 8 lines
Quoted 2 lines
No point in bloating the kernel image with a bootloader header if we run bare metal.
I would say the same for S-mode. EFI booting should be an option, not a requirement. I have M-mode U-boot working with bootelf to start BBL, and at some point, I’m hoping we can have a M-mode linux kernel be the SBI provider for S-mode kernels, which seem most logical to me to start using the vmlinux elf binaries using something like
kexec()
Strictly speaking we could just add another option for the header so that you could also disable it for S-mode. But then again the header is not very harmful, as you don't have to use it. It just eats up a little more kernel space. And as that aspace is very tight for my M-mode target (the Kendryte KD210) and it is totally pointless for M-mode I just remove it there.
The idea of using M-mode Linux as the SBI sounds cool.
Re: [PATCH 15/15] riscv: disable the EFI PECOFF header for M-mode
In reply to Troy BenjegerdesQuoted 7 lines · Troy Benjegerdes
Quoted 4 lines · Christoph Hellwig
No point in bloating the kernel image with a bootloader header if we run bare metal.
I would say the same for S-mode. EFI booting should be an option, not a requirement.
EFI booting is never a requirement on any board. When EFI stub will be added for kernel, it will be enabled with CONFIG_EFI_STUB only.
The current additional header is only 64 bytes and also required for booti in U-boot. So it shouldn't disabled for S-mode.
Disabling it for M-Mode Linux is okay because of memory constraint and M-Mode linux won't use U-boot anyways.
Quoted 3 lines
I have M-mode U-boot working with bootelf to start BBL, and at some point, I’m hoping we can have a M-mode linux kernel be the SBI provider for S-mode kernels,
Why do you want bloat a M-Mode software with Linux just for SBI implementation?
Using Linux as a last stage boot loader i.e. LinuxBoot may make sense though.
Quoted 42 lines
which seem most logical to me
to start using the vmlinux elf binaries using something likekexec()Quoted 20 lines
Signed-off-by Christoph Hellwig <[email protected]>---
arch/riscv/kernel/head.S| 2 ++1 file changed, 2 insertions(+)arch/riscv/kernel/head.S +0
-0@@ -16,6 +16,7 @@
__INITENTRY(_start)
+#ifndefCONFIG_M_MODE/* * Image header expected by Linux boot-loaders. The imageheader data
* structure is described inasm/image.h.arch/riscv/kernel/head.S +0
-0@@ -47,6 +48,7 @@ ENTRY(_start)Quoted 14 lines
.global
_start_kernel_start_kernel:
+#endif /*CONFIG_M_MODE*/
/* Mask all interrupts */
csrwCSR_XIE, zero
csrwCSR_XIP, zeroSignature · 7 lines
2.20.1 _______________________________________________ linux-riscv mailing list [email protected] http://lists.infradead.org/mailman/listinfo/linux-riscv_______________________________________________
linux-riscv mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/linux-riscv
Signature · 2 lines
Regards, Atish
Re: [PATCH 15/15] riscv: disable the EFI PECOFF header for M-mode
Quoted 26 lines · Atish Patra
Quoted 7 lines · Troy Benjegerdes
Quoted 4 lines · Christoph Hellwig
No point in bloating the kernel image with a bootloader header if we run bare metal.
I would say the same for S-mode. EFI booting should be an option, not a requirement.
EFI booting is never a requirement on any board. When EFI stub will be added for kernel, it will be enabled with
CONFIG_EFI_STUBonly.The current additional header is only 64 bytes and also required for booti in U-boot. So it shouldn't disabled for S-mode.
Disabling it for M-Mode Linux is okay because of memory constraint and M-Mode linux won't use U-boot anyways.
Quoted 3 lines
I have M-mode U-boot working with bootelf to start BBL, and at some point, I’m hoping we can have a M-mode linux kernel be the SBI provider for S-mode kernels,
Quoted 6 lines
Why do you want bloat a M-Mode software with Linux just for SBI implementation?
Using Linux as a last stage boot loader i.e. LinuxBoot may make sense though.
Boot time, and ease of development, and simplified system management.
Having M-mode linux as a supervisor/boot kernel can get us to responding to HTTPS/SSH/etc requests within seconds of power-on, while the ‘boot’ kernel can be loading guest S-mode kernels from things like NVME flash drives that are going to be a lot more code and development to support in U-boot or any other non-linux dedicated boot loader.
There’s also a very strong security argument, as Linux is going to get the largest and broadest security review, and will likely get software updates a lot faster than dedicated boot firmwares will.
Another reason would be sharing the same kernel binary (elf file) for both M-mode, and S-mode, and using the device tree passed to each to specify which mode it should be running it. There are probably a bunch of gotchas with this idea, and even so I suspect someone will decide to go ahead and just do it eventually because it could make testing, validation, and security updates a lot easier from an operational/deployment point of view.
Linuxbios convinced me that if you want to do a really large cluster, you can build, manage, and run such a thing with fewer people and engineering cost than if you have all these extra layers of boot firmware that require some company to have firmware engineers and lots of extra system testing on the firmware.
RE: [PATCH 15/15] riscv: disable the EFI PECOFF header for M-mode
Quoted 45 lines
-----Original Message-----
From: [email protected] <linux-kernel- [email protected]> On Behalf Of Troy Benjegerdes
Sent: Wednesday, August 21, 2019 11:25 PM
To: Atish Patra <[email protected]>
Cc: [email protected]; [email protected]; [email protected]; Damien Le Moal <[email protected]>; linux- [email protected]; [email protected]
Subject: Re: [PATCH 15/15] riscv: disable the EFI PECOFF header for M-modeQuoted 24 lines · Atish Patra
Quoted 7 lines · Troy Benjegerdes
Quoted 4 lines · Christoph Hellwig
No point in bloating the kernel image with a bootloader header if we run bare metal.
I would say the same for S-mode. EFI booting should be an option, not a requirement.
EFI booting is never a requirement on any board. When EFI stub will be added for kernel, it will be enabled with
CONFIG_EFI_STUBonly.The current additional header is only 64 bytes and also required for booti in U-boot. So it shouldn't disabled for S-mode.
Disabling it for M-Mode Linux is okay because of memory constraint and M-Mode linux won't use U-boot anyways.
Quoted 3 lines
I have M-mode U-boot working with bootelf to start BBL, and at some point, I’m hoping we can have a M-mode linux kernel be the SBI provider for S-mode kernels,
Quoted 6 lines
Why do you want bloat a M-Mode software with Linux just for SBI implementation?
Using Linux as a last stage boot loader i.e. LinuxBoot may make sense though.
Quoted 7 lines
Boot time, and ease of development, and simplified system management.
Having M-mode linux as a supervisor/boot kernel can get us to responding to HTTPS/SSH/etc requests within seconds of power-on, while the ‘boot’ kernel can be loading guest S-mode kernels from things like NVME flash drives that are going to be a lot more code and development to support in U- boot or any other non-linux dedicated boot loader.
I don't see why these things cannot be achieved in existing open-source bootloaders. In fact, U-boot already has "Falcon" mode for fast booting.
Quoted 4 lines
There’s also a very strong security argument, as Linux is going to get the largest and broadest security review, and will likely get software updates a lot faster than dedicated boot firmwares will.
For security, we have to get SW certified with various something like ISO2626 standard. This is very common practice in Automotive industry. To achieve such a certification for any SW, the size of code base is very very important.
Due to this reason, even today Linux (and other big open-source project) are very difficult to be security certified.
Quoted 8 lines
Another reason would be sharing the same kernel binary (elf file) for both M-mode, and S-mode, and using the device tree passed to each to specify which mode it should be running it. There are probably a bunch of gotchas with this idea, and even so I suspect someone will decide to go ahead and just do it eventually because it could make testing, validation, and security updates a lot easier from an operational/deployment point of view.
Quoted 5 lines
Linuxbios convinced me that if you want to do a really large cluster, you can build, manage, and run such a thing with fewer people and engineering cost than if you have all these extra layers of boot firmware that require some company to have firmware engineers and lots of extra system testing on the firmware.
I don't by this last argument. These days it's just very few folks doing firmware, bootloader, and Linux porting for any new SOC (any architecture). Most of the things are already there in various open-source project so same person can easily contribute to various projects.
Regards,
Anup
Re: [PATCH 15/15] riscv: disable the EFI PECOFF header for M-mode
Quoted 65 lines · Anup Patel
Quoted 46 lines
-----Original Message-----
From: [email protected] <linux-kernel- [email protected]> On Behalf Of Troy Benjegerdes
Sent: Wednesday, August 21, 2019 11:25 PM
To: Atish Patra <[email protected]>
Cc: [email protected]; [email protected]; [email protected]; Damien Le Moal <[email protected]>; linux- [email protected]; [email protected]
Subject: Re: [PATCH 15/15] riscv: disable the EFI PECOFF header for M-modeQuoted 24 lines · Atish Patra
Quoted 7 lines · Troy Benjegerdes
Quoted 4 lines · Christoph Hellwig
No point in bloating the kernel image with a bootloader header if we run bare metal.
I would say the same for S-mode. EFI booting should be an option, not a requirement.
EFI booting is never a requirement on any board. When EFI stub will be added for kernel, it will be enabled with
CONFIG_EFI_STUBonly.The current additional header is only 64 bytes and also required for booti in U-boot. So it shouldn't disabled for S-mode.
Disabling it for M-Mode Linux is okay because of memory constraint and M-Mode linux won't use U-boot anyways.
Quoted 3 lines
I have M-mode U-boot working with bootelf to start BBL, and at some point, I’m hoping we can have a M-mode linux kernel be the SBI provider for S-mode kernels,
Quoted 6 lines
Why do you want bloat a M-Mode software with Linux just for SBI implementation?
Using Linux as a last stage boot loader i.e. LinuxBoot may make sense though.
Quoted 7 lines
Boot time, and ease of development, and simplified system management.
Having M-mode linux as a supervisor/boot kernel can get us to responding to HTTPS/SSH/etc requests within seconds of power-on, while the ‘boot’ kernel can be loading guest S-mode kernels from things like NVME flash drives that are going to be a lot more code and development to support in U- boot or any other non-linux dedicated boot loader.
I don't see why these things cannot be achieved in existing open-source bootloaders. In fact, U-boot already has "Falcon" mode for fast booting.
Quoted 4 lines
There’s also a very strong security argument, as Linux is going to get the largest and broadest security review, and will likely get software updates a lot faster than dedicated boot firmwares will.
Quoted 6 lines
For security, we have to get SW certified with various something like ISO2626 standard. This is very common practice in Automotive industry. To achieve such a certification for any SW, the size of code base is very very important.
Due to this reason, even today Linux (and other big open-source project) are very difficult to be security certified.
There’s security certified, and then there’s what I personally consider secure.
The second category is code that I know is widely audited by lots of people, and gets quickly updated when there is a problem. I like U-boot, and I think its a great solution for industry, it’s just not the only solution that could be used.
Quoted 15 lines
Quoted 9 lines
Another reason would be sharing the same kernel binary (elf file) for both M-mode, and S-mode, and using the device tree passed to each to specify which mode it should be running it. There are probably a bunch of gotchas with this idea, and even so I suspect someone will decide to go ahead and just do it eventually because it could make testing, validation, and security updates a lot easier from an operational/deployment point of view.
Quoted 5 lines
Linuxbios convinced me that if you want to do a really large cluster, you can build, manage, and run such a thing with fewer people and engineering cost than if you have all these extra layers of boot firmware that require some company to have firmware engineers and lots of extra system testing on the firmware.
Quoted 7 lines
I don't by this last argument. These days it's just very few folks doing firmware, bootloader, and Linux porting for any new SOC (any architecture). Most of the things are already there in various open-source project so same person can easily contribute to various projects.
Regards,
Anup
What I see though is we’re duplicating code and work between bootloaders and kernel, for example the SPI-NOR code, and if it was all linux, it would be one driver model to learn/remember/track, and one place to fix things.
U-boot is great because you can boot other !linux things (like FreeBSD), however if I was purpose building a linux cluster, I would want to be running linux as early as possible so I can use linux scripting in bash/go/python and talk to the queue/workload manager over a native high performance network instead of the extremely limited ‘hush’ shell and having to discover which user image to boot with something old and slow like dhcp/tftp/etc.