[PATCH v3] add icswx support
Michael Neuling
mikey at neuling.org
Tue Nov 2 09:23:01 EST 2010
In message <1287547340.14049.25.camel at flin.austin.ibm.com> you wrote:
> icswx is a PowerPC co-processor instruction to send data to a
> co-processor. On Book-S processors the LPAR_ID and process ID (PID) of
> the owning process are registered in the window context of the
> co-processor at initial time. When the icswx instruction is executed,
> the L2 generates a cop-reg transaction on PowerBus. The transaction has
> no address and the processor does not perform an MMU access to
> authenticate the transaction. The coprocessor compares the LPAR_ID and
> the PID included in the transaction and the LPAR_ID and PID held in the
> window context to determine if the process is authorized to generate the
> transaction.
>
> The OS needs to assign a 16-bit PID for the process. This cop-PID needs
> to be updated during context switch. The cop-PID needs to be destroyed
> when the context is destroyed.
>
> Change log from v2:
> - Make the code a CPU feature and return -NODEV if CPU doesn't have
> icswx co-processor instruction.
> - Change the goto loop in use_cop() into a do-while loop.
> - Change context destroy code into a new destroy_context_acop() function
> and #define it based on CONFIG_ICSWX.
> - Remove mmput() from drop_cop().
> - Fix some TAB/space problems.
>
> Signed-off-by: Sonny Rao <sonnyrao at linux.vnet.ibm.com>
> Signed-off-by: Tseng-Hui (Frank) Lin <thlin at linux.vnet.ibm.com>
>
> ---
> arch/powerpc/include/asm/cputable.h | 4 +-
> arch/powerpc/include/asm/mmu-hash64.h | 5 ++
> arch/powerpc/include/asm/mmu_context.h | 6 ++
> arch/powerpc/include/asm/reg.h | 11 +++
> arch/powerpc/include/asm/reg_booke.h | 3 -
> arch/powerpc/mm/mmu_context_hash64.c | 109
> ++++++++++++++++++++++++++++++++
> arch/powerpc/platforms/Kconfig.cputype | 17 +++++
> 7 files changed, 151 insertions(+), 4 deletions(-)
>
> diff --git a/arch/powerpc/include/asm/cputable.h
> b/arch/powerpc/include/asm/cputable.h
> index 3a40a99..bbb4e2c 100644
> --- a/arch/powerpc/include/asm/cputable.h
> +++ b/arch/powerpc/include/asm/cputable.h
> @@ -198,6 +198,7 @@ extern const char *powerpc_base_platform;
> #define CPU_FTR_CP_USE_DCBTZ LONG_ASM_CONST(0x0040000000000000)
> #define CPU_FTR_UNALIGNED_LD_STD LONG_ASM_CONST(0x0080000000000000)
> #define CPU_FTR_ASYM_SMT LONG_ASM_CONST(0x0100000000000000)
> +#define CPU_FTR_ICSWX LONG_ASM_CONST(0x02000000000000
00)
>
> #ifndef __ASSEMBLY__
>
> @@ -413,7 +414,8 @@ extern const char *powerpc_base_platform;
> CPU_FTR_MMCRA | CPU_FTR_SMT | \
> CPU_FTR_COHERENT_ICACHE | CPU_FTR_LOCKLESS_TLBIE | \
> CPU_FTR_PURR | CPU_FTR_SPURR | CPU_FTR_REAL_LE | \
> - CPU_FTR_DSCR | CPU_FTR_SAO | CPU_FTR_ASYM_SMT)
> + CPU_FTR_DSCR | CPU_FTR_SAO | CPU_FTR_ASYM_SMT | \
> + CPU_FTR_ICSWX)
You patch is against an old kernel. POWER7 doesn't look like this
anymore. Please update to newer kernel. I'd suggest patching against
linux-next.
We probabbly need to set a cpu_user_feature too so userspace knows it
can use this. We need to create a PPC_FEATURE_HAS_ICSWX. Frank: just
follow the bouncing ball on PPC_FEATURE_HAS_VSX.
> #define CPU_FTRS_CELL (CPU_FTR_USE_TB | CPU_FTR_LWSYNC | \
> CPU_FTR_PPCAS_ARCH_V2 | CPU_FTR_CTRL | \
> CPU_FTR_ALTIVEC_COMP | CPU_FTR_MMCRA | CPU_FTR_SMT | \
> diff --git a/arch/powerpc/include/asm/mmu-hash64.h
> b/arch/powerpc/include/asm/mmu-hash64.h
> index acac35d..6c1ab90 100644
> --- a/arch/powerpc/include/asm/mmu-hash64.h
> +++ b/arch/powerpc/include/asm/mmu-hash64.h
> @@ -423,6 +423,11 @@ typedef struct {
> #ifdef CONFIG_PPC_SUBPAGE_PROT
> struct subpage_prot_table spt;
> #endif /* CONFIG_PPC_SUBPAGE_PROT */
> +#ifdef CONFIG_ICSWX
> + unsigned long acop; /* mask of enabled coprocessor types */
> +#define HASH64_MAX_PID (0xFFFF)
> + unsigned int acop_pid; /* pid value used with coprocessors */
> +#endif /* CONFIG_ICSWX */
> } mm_context_t;
>
>
> diff --git a/arch/powerpc/include/asm/mmu_context.h
> b/arch/powerpc/include/asm/mmu_context.h
> index 81fb412..88118de 100644
> --- a/arch/powerpc/include/asm/mmu_context.h
> +++ b/arch/powerpc/include/asm/mmu_context.h
> @@ -80,6 +80,12 @@ static inline void switch_mm(struct mm_struct *prev,
> struct mm_struct *next,
>
> #define deactivate_mm(tsk,mm) do { } while (0)
>
> +#ifdef CONFIG_ICSWX
> +extern void switch_cop(struct mm_struct *next);
> +extern int use_cop(unsigned long acop, struct mm_struct *mm);
> +extern void drop_cop(unsigned long acop, struct mm_struct *mm);
> +#endif /* CONFIG_ICSWX */
> +
> /*
> * After we have set current->mm to a new value, this activates
> * the context for the new mm so we see the new mappings.
> diff --git a/arch/powerpc/include/asm/reg.h
> b/arch/powerpc/include/asm/reg.h
> index ff0005eec..b86d876 100644
> --- a/arch/powerpc/include/asm/reg.h
> +++ b/arch/powerpc/include/asm/reg.h
> @@ -170,8 +170,19 @@
> #define SPEFSCR_FRMC 0x00000003 /* Embedded FP rounding mode co
ntrol
> */
>
> /* Special Purpose Registers (SPRNs)*/
> +
> +#ifdef CONFIG_40x
> +#define SPRN_PID 0x3B1 /* Process ID */
> +#else
> +#define SPRN_PID 0x030 /* Process ID */
> +#ifdef CONFIG_BOOKE
> +#define SPRN_PID0 SPRN_PID/* Process ID Register 0 */
> +#endif
> +#endif
> +
> #define SPRN_CTR 0x009 /* Count Register */
> #define SPRN_DSCR 0x11
> +#define SPRN_ACOP 0x1F /* Available Coprocessor Register */
> #define SPRN_CTRLF 0x088
> #define SPRN_CTRLT 0x098
> #define CTRL_CT 0xc0000000 /* current thread */
> diff --git a/arch/powerpc/include/asm/reg_booke.h
> b/arch/powerpc/include/asm/reg_booke.h
> index 667a498..5b0c781 100644
> --- a/arch/powerpc/include/asm/reg_booke.h
> +++ b/arch/powerpc/include/asm/reg_booke.h
> @@ -150,8 +150,6 @@
> * or IBM 40x.
> */
> #ifdef CONFIG_BOOKE
> -#define SPRN_PID 0x030 /* Process ID */
> -#define SPRN_PID0 SPRN_PID/* Process ID Register 0 */
> #define SPRN_CSRR0 0x03A /* Critical Save and Restore Register 0 */
> #define SPRN_CSRR1 0x03B /* Critical Save and Restore Register 1 */
> #define SPRN_DEAR 0x03D /* Data Error Address Register */
> @@ -168,7 +166,6 @@
> #define SPRN_TCR 0x154 /* Timer Control Register */
> #endif /* Book E */
> #ifdef CONFIG_40x
> -#define SPRN_PID 0x3B1 /* Process ID */
> #define SPRN_DBCR1 0x3BD /* Debug Control Register 1 */
> #define SPRN_ESR 0x3D4 /* Exception Syndrome Register */
> #define SPRN_DEAR 0x3D5 /* Data Error Address Register */
> diff --git a/arch/powerpc/mm/mmu_context_hash64.c
> b/arch/powerpc/mm/mmu_context_hash64.c
> index 2535828..6ef6ce2 100644
> --- a/arch/powerpc/mm/mmu_context_hash64.c
> +++ b/arch/powerpc/mm/mmu_context_hash64.c
> @@ -18,6 +18,7 @@
> #include <linux/mm.h>
> #include <linux/spinlock.h>
> #include <linux/idr.h>
> +#include <linux/percpu.h>
> #include <linux/module.h>
> #include <linux/gfp.h>
>
> @@ -26,6 +27,113 @@
> static DEFINE_SPINLOCK(mmu_context_lock);
> static DEFINE_IDA(mmu_context_ida);
>
> +#ifdef CONFIG_ICSWX
> +static DEFINE_SPINLOCK(mmu_context_acop_lock);
> +static DEFINE_IDA(cop_ida);
> +
> +/* Lazy switch the ACOP register */
> +static DEFINE_PER_CPU(unsigned long, acop_reg);
Do we really need to do this lazily? Do you have any performance
numbers to say this is needed?
Why do the ACOP lazily and no the PID?
> +
> +void switch_cop(struct mm_struct *next)
> +{
> + if (!cpu_has_feature(CPU_FTR_ICSWX))
> + return;
> +
> + mtspr(SPRN_PID, next->context.acop_pid);
> + if (next->context.acop_pid &&
> + __get_cpu_var(acop_reg) != next->context.acop) {
> + mtspr(SPRN_ACOP, next->context.acop);
> + __get_cpu_var(acop_reg) = next->context.acop;
> + }
Does SPRN_PID need to be switch on normal context in __switch_to? What
happens when a CPU context switches to another process that is also
doing ICSWX? Don't we need to update the PID so if that process does an
ICSWX it sends the correct PID?
What context is this going to get called in? Could we get pre-empted
here and end up doing one of these and not the other and then coming
back in again? You use a lock later when updating ACP, but not here?
Why is that?
Again, what is the justfication for the lazy switching complexity?
Also, I'd like to see an example user for this interface. You created
three new interfaces but provided no examples and no documentation on
how to use them:
+EXPORT_SYMBOL(switch_cop);
+EXPORT_SYMBOL(use_cop);
+EXPORT_SYMBOL(drop_cop);
Please document how to use these.
> +}
> +EXPORT_SYMBOL(switch_cop);
> +
> +int use_cop(unsigned long acop, struct mm_struct *mm)
> +{
> + int acop_pid;
> + int err;
> +
> + if (!cpu_has_feature(CPU_FTR_ICSWX))
> + return -ENODEV;
> +
> + if (!mm)
> + return -EINVAL;
> +
> + if (!mm->context.acop_pid) {
> + if (!ida_pre_get(&cop_ida, GFP_KERNEL))
> + return -ENOMEM;
> + do {
> + spin_lock(&mmu_context_acop_lock);
> + err = ida_get_new_above(&cop_ida, 1, &acop_pid);
> + spin_unlock(&mmu_context_acop_lock);
> + } while (err == -EAGAIN);
ida_get_new_above says this:
* If memory is required, it will return %-EAGAIN, you should unlock
* and go back to the ida_pre_get() call. If the ida is full, it will
* return %-ENOSPC.
You are not going back to ida_pre_get().
> +
> + if (err)
> + return err;
> +
> + if (acop_pid > HASH64_MAX_PID) {
> + spin_lock(&mmu_context_acop_lock);
> + ida_remove(&cop_ida, acop_pid);
> + spin_unlock(&mmu_context_acop_lock);
> + return -EBUSY;
> + }
> + mm->context.acop_pid = acop_pid;
> + if (mm == current->active_mm)
> + mtspr(SPRN_PID, mm->context.acop_pid);
> + }
> + spin_lock(&mmu_context_acop_lock);
> + mm->context.acop |= acop;
> + spin_unlock(&mmu_context_acop_lock);
> +
> + get_cpu_var(acop_reg) = mm->context.acop;
> + if (mm == current->active_mm)
> + mtspr(SPRN_ACOP, mm->context.acop);
> + put_cpu_var(acop_reg);
> +
> + return mm->context.acop_pid;
> +}
> +EXPORT_SYMBOL(use_cop);
> +
> +void drop_cop(unsigned long acop, struct mm_struct *mm)
> +{
> + if (!cpu_has_feature(CPU_FTR_ICSWX))
> + return;
> +
> + if (WARN_ON(!mm))
> + return;
> +
> + spin_lock(&mmu_context_acop_lock);
> + mm->context.acop &= ~acop;
> + spin_unlock(&mmu_context_acop_lock);
> + if (!mm->context.acop) {
> + spin_lock(&mmu_context_acop_lock);
> + ida_remove(&cop_ida, mm->context.acop_pid);
> + spin_unlock(&mmu_context_acop_lock);
> + mm->context.acop_pid = 0;
Is the locking right here? Should this last line be inside the lock?
> + if (mm == current->active_mm)
> + mtspr(SPRN_PID, mm->context.acop_pid);
PID will now be 0. What happens in a userspace program does a ICSWX
with PID 0?
> + } else {
> + get_cpu_var(acop_reg) = mm->context.acop;
> + if (mm == current->active_mm)
> + mtspr(SPRN_ACOP, mm->context.acop);
> + put_cpu_var(acop_reg);
> + }
> +}
> +EXPORT_SYMBOL(drop_cop);
> +
> +static void destroy_context_acop(struct mm_struct *mm)
> +{
> + if (mm->context.acop_pid) {
> + spin_lock(&mmu_context_acop_lock);
> + ida_remove(&cop_ida, mm->context.acop_pid);
> + spin_unlock(&mmu_context_acop_lock);
> + }
> +}
> +
> +#else
> +#define destroy_context_acop(mm)
> +#endif /* CONFIG_ICSWX */
> +
> /*
> * The proto-VSID space has 2^35 - 1 segments available for user
> mappings.
> * Each segment contains 2^28 bytes. Each context maps 2^44 bytes,
> @@ -93,6 +201,7 @@ EXPORT_SYMBOL_GPL(__destroy_context);
>
> void destroy_context(struct mm_struct *mm)
> {
> + destroy_context_acop(mm);
> __destroy_context(mm->context.id);
> subpage_prot_free(mm);
> mm->context.id = NO_CONTEXT;
> diff --git a/arch/powerpc/platforms/Kconfig.cputype
> b/arch/powerpc/platforms/Kconfig.cputype
> index d361f81..7678e29 100644
> --- a/arch/powerpc/platforms/Kconfig.cputype
> +++ b/arch/powerpc/platforms/Kconfig.cputype
> @@ -220,6 +220,23 @@ config VSX
>
> If in doubt, say Y here.
>
> +config ICSWX
> + bool "Support for PowerPC icswx co-processor instruction"
> + depends on POWER4
> + default n
> + ---help---
> +
> + Enabling this option to turn on the PowerPC icswx co-processor
> + instruction support for POWER7 or newer processors.
> + This option is only useful if you have a processor that supports
> + icswx co-processor instruction. It does not have any effect on
> + processors without icswx co-processor instruction.
> +
> + This support slightly increases kernel memory usage.
> +
> + Say N if you do not have a PowerPC processor supporting icswx
> + instruction and a PowerPC co-processor.
> +
> config SPE
> bool "SPE Support"
> depends on E200 || (E500 && !PPC_E500MC)
>
>
> _______________________________________________
> Linuxppc-dev mailing list
> Linuxppc-dev at lists.ozlabs.org
> https://lists.ozlabs.org/listinfo/linuxppc-dev
>
More information about the Linuxppc-dev
mailing list