2019-06-20 04:15:55

by Alastair D'Silva

[permalink] [raw]
Subject: [PATCH v2] ocxl: Allow contexts to be attached with a NULL mm

From: Alastair D'Silva <[email protected]>

If an OpenCAPI context is to be used directly by a kernel driver, there
may not be a suitable mm to use.

The patch makes the mm parameter to ocxl_context_attach optional.

Signed-off-by: Alastair D'Silva <[email protected]>
---
arch/powerpc/mm/book3s64/radix_tlb.c | 5 +++++
drivers/misc/ocxl/context.c | 9 ++++++---
drivers/misc/ocxl/link.c | 28 ++++++++++++++++++++++++----
3 files changed, 35 insertions(+), 7 deletions(-)

diff --git a/arch/powerpc/mm/book3s64/radix_tlb.c b/arch/powerpc/mm/book3s64/radix_tlb.c
index bb9835681315..ce8a77fae6a7 100644
--- a/arch/powerpc/mm/book3s64/radix_tlb.c
+++ b/arch/powerpc/mm/book3s64/radix_tlb.c
@@ -666,6 +666,11 @@ EXPORT_SYMBOL(radix__flush_tlb_page);
#define radix__flush_all_mm radix__local_flush_all_mm
#endif /* CONFIG_SMP */

+/*
+ * If kernel TLBIs ever become local rather than global, then
+ * drivers/misc/ocxl/link.c:ocxl_link_add_pe will need some work, as it
+ * assumes kernel TLBIs are global.
+ */
void radix__flush_tlb_kernel_range(unsigned long start, unsigned long end)
{
_tlbie_pid(0, RIC_FLUSH_ALL);
diff --git a/drivers/misc/ocxl/context.c b/drivers/misc/ocxl/context.c
index bab9c9364184..994563a078eb 100644
--- a/drivers/misc/ocxl/context.c
+++ b/drivers/misc/ocxl/context.c
@@ -69,6 +69,7 @@ static void xsl_fault_error(void *data, u64 addr, u64 dsisr)
int ocxl_context_attach(struct ocxl_context *ctx, u64 amr, struct mm_struct *mm)
{
int rc;
+ unsigned long pidr = 0;

// Locks both status & tidr
mutex_lock(&ctx->status_mutex);
@@ -77,9 +78,11 @@ int ocxl_context_attach(struct ocxl_context *ctx, u64 amr, struct mm_struct *mm)
goto out;
}

- rc = ocxl_link_add_pe(ctx->afu->fn->link, ctx->pasid,
- mm->context.id, ctx->tidr, amr, mm,
- xsl_fault_error, ctx);
+ if (mm)
+ pidr = mm->context.id;
+
+ rc = ocxl_link_add_pe(ctx->afu->fn->link, ctx->pasid, pidr, ctx->tidr,
+ amr, mm, xsl_fault_error, ctx);
if (rc)
goto out;

diff --git a/drivers/misc/ocxl/link.c b/drivers/misc/ocxl/link.c
index cce5b0d64505..58d111afd9f6 100644
--- a/drivers/misc/ocxl/link.c
+++ b/drivers/misc/ocxl/link.c
@@ -224,6 +224,17 @@ static irqreturn_t xsl_fault_handler(int irq, void *data)
ack_irq(spa, ADDRESS_ERROR);
return IRQ_HANDLED;
}
+
+ if (!pe_data->mm) {
+ /*
+ * translation fault from a kernel context - an OpenCAPI
+ * device tried to access a bad kernel address
+ */
+ rcu_read_unlock();
+ pr_warn("Unresolved OpenCAPI xsl fault in kernel context\n");
+ ack_irq(spa, ADDRESS_ERROR);
+ return IRQ_HANDLED;
+ }
WARN_ON(pe_data->mm->context.id != pid);

if (mmget_not_zero(pe_data->mm)) {
@@ -523,7 +534,13 @@ int ocxl_link_add_pe(void *link_handle, int pasid, u32 pidr, u32 tidr,
pe->amr = cpu_to_be64(amr);
pe->software_state = cpu_to_be32(SPA_PE_VALID);

- mm_context_add_copro(mm);
+ /*
+ * For user contexts, register a copro so that TLBIs are seen
+ * by the nest MMU. If we have a kernel context, TLBIs are
+ * already global.
+ */
+ if (mm)
+ mm_context_add_copro(mm);
/*
* Barrier is to make sure PE is visible in the SPA before it
* is used by the device. It also helps with the global TLBI
@@ -546,7 +563,8 @@ int ocxl_link_add_pe(void *link_handle, int pasid, u32 pidr, u32 tidr,
* have a reference on mm_users. Incrementing mm_count solves
* the problem.
*/
- mmgrab(mm);
+ if (mm)
+ mmgrab(mm);
trace_ocxl_context_add(current->pid, spa->spa_mem, pasid, pidr, tidr);
unlock:
mutex_unlock(&spa->spa_lock);
@@ -652,8 +670,10 @@ int ocxl_link_remove_pe(void *link_handle, int pasid)
if (!pe_data) {
WARN(1, "Couldn't find pe data when removing PE\n");
} else {
- mm_context_remove_copro(pe_data->mm);
- mmdrop(pe_data->mm);
+ if (pe_data->mm) {
+ mm_context_remove_copro(pe_data->mm);
+ mmdrop(pe_data->mm);
+ }
kfree_rcu(pe_data, rcu);
}
unlock:
--
2.21.0


2019-06-20 05:25:54

by Nicholas Piggin

[permalink] [raw]
Subject: Re: [PATCH v2] ocxl: Allow contexts to be attached with a NULL mm

Alastair D'Silva's on June 20, 2019 2:12 pm:
> From: Alastair D'Silva <[email protected]>
>
> If an OpenCAPI context is to be used directly by a kernel driver, there
> may not be a suitable mm to use.
>
> The patch makes the mm parameter to ocxl_context_attach optional.
>
> Signed-off-by: Alastair D'Silva <[email protected]>

Yeah I don't think you need to manage a kernel context explicitly
because it will always be flushed with tlbie, comment helps. For
the powerpc/mm bit,

Acked-by: Nicholas Piggin <[email protected]>

2019-06-20 09:21:27

by Frederic Barrat

[permalink] [raw]
Subject: Re: [PATCH v2] ocxl: Allow contexts to be attached with a NULL mm



Le 20/06/2019 à 06:12, Alastair D'Silva a écrit :
> From: Alastair D'Silva <[email protected]>
>
> If an OpenCAPI context is to be used directly by a kernel driver, there
> may not be a suitable mm to use.
>
> The patch makes the mm parameter to ocxl_context_attach optional.
>
> Signed-off-by: Alastair D'Silva <[email protected]>
> ---

Thanks for the update.
Acked-by: Frederic Barrat <[email protected]>



> arch/powerpc/mm/book3s64/radix_tlb.c | 5 +++++
> drivers/misc/ocxl/context.c | 9 ++++++---
> drivers/misc/ocxl/link.c | 28 ++++++++++++++++++++++++----
> 3 files changed, 35 insertions(+), 7 deletions(-)
>
> diff --git a/arch/powerpc/mm/book3s64/radix_tlb.c b/arch/powerpc/mm/book3s64/radix_tlb.c
> index bb9835681315..ce8a77fae6a7 100644
> --- a/arch/powerpc/mm/book3s64/radix_tlb.c
> +++ b/arch/powerpc/mm/book3s64/radix_tlb.c
> @@ -666,6 +666,11 @@ EXPORT_SYMBOL(radix__flush_tlb_page);
> #define radix__flush_all_mm radix__local_flush_all_mm
> #endif /* CONFIG_SMP */
>
> +/*
> + * If kernel TLBIs ever become local rather than global, then
> + * drivers/misc/ocxl/link.c:ocxl_link_add_pe will need some work, as it
> + * assumes kernel TLBIs are global.
> + */
> void radix__flush_tlb_kernel_range(unsigned long start, unsigned long end)
> {
> _tlbie_pid(0, RIC_FLUSH_ALL);
> diff --git a/drivers/misc/ocxl/context.c b/drivers/misc/ocxl/context.c
> index bab9c9364184..994563a078eb 100644
> --- a/drivers/misc/ocxl/context.c
> +++ b/drivers/misc/ocxl/context.c
> @@ -69,6 +69,7 @@ static void xsl_fault_error(void *data, u64 addr, u64 dsisr)
> int ocxl_context_attach(struct ocxl_context *ctx, u64 amr, struct mm_struct *mm)
> {
> int rc;
> + unsigned long pidr = 0;
>
> // Locks both status & tidr
> mutex_lock(&ctx->status_mutex);
> @@ -77,9 +78,11 @@ int ocxl_context_attach(struct ocxl_context *ctx, u64 amr, struct mm_struct *mm)
> goto out;
> }
>
> - rc = ocxl_link_add_pe(ctx->afu->fn->link, ctx->pasid,
> - mm->context.id, ctx->tidr, amr, mm,
> - xsl_fault_error, ctx);
> + if (mm)
> + pidr = mm->context.id;
> +
> + rc = ocxl_link_add_pe(ctx->afu->fn->link, ctx->pasid, pidr, ctx->tidr,
> + amr, mm, xsl_fault_error, ctx);
> if (rc)
> goto out;
>
> diff --git a/drivers/misc/ocxl/link.c b/drivers/misc/ocxl/link.c
> index cce5b0d64505..58d111afd9f6 100644
> --- a/drivers/misc/ocxl/link.c
> +++ b/drivers/misc/ocxl/link.c
> @@ -224,6 +224,17 @@ static irqreturn_t xsl_fault_handler(int irq, void *data)
> ack_irq(spa, ADDRESS_ERROR);
> return IRQ_HANDLED;
> }
> +
> + if (!pe_data->mm) {
> + /*
> + * translation fault from a kernel context - an OpenCAPI
> + * device tried to access a bad kernel address
> + */
> + rcu_read_unlock();
> + pr_warn("Unresolved OpenCAPI xsl fault in kernel context\n");
> + ack_irq(spa, ADDRESS_ERROR);
> + return IRQ_HANDLED;
> + }
> WARN_ON(pe_data->mm->context.id != pid);
>
> if (mmget_not_zero(pe_data->mm)) {
> @@ -523,7 +534,13 @@ int ocxl_link_add_pe(void *link_handle, int pasid, u32 pidr, u32 tidr,
> pe->amr = cpu_to_be64(amr);
> pe->software_state = cpu_to_be32(SPA_PE_VALID);
>
> - mm_context_add_copro(mm);
> + /*
> + * For user contexts, register a copro so that TLBIs are seen
> + * by the nest MMU. If we have a kernel context, TLBIs are
> + * already global.
> + */
> + if (mm)
> + mm_context_add_copro(mm);
> /*
> * Barrier is to make sure PE is visible in the SPA before it
> * is used by the device. It also helps with the global TLBI
> @@ -546,7 +563,8 @@ int ocxl_link_add_pe(void *link_handle, int pasid, u32 pidr, u32 tidr,
> * have a reference on mm_users. Incrementing mm_count solves
> * the problem.
> */
> - mmgrab(mm);
> + if (mm)
> + mmgrab(mm);
> trace_ocxl_context_add(current->pid, spa->spa_mem, pasid, pidr, tidr);
> unlock:
> mutex_unlock(&spa->spa_lock);
> @@ -652,8 +670,10 @@ int ocxl_link_remove_pe(void *link_handle, int pasid)
> if (!pe_data) {
> WARN(1, "Couldn't find pe data when removing PE\n");
> } else {
> - mm_context_remove_copro(pe_data->mm);
> - mmdrop(pe_data->mm);
> + if (pe_data->mm) {
> + mm_context_remove_copro(pe_data->mm);
> + mmdrop(pe_data->mm);
> + }
> kfree_rcu(pe_data, rcu);
> }
> unlock:
>

2019-06-21 00:23:29

by Andrew Donnellan

[permalink] [raw]
Subject: Re: [PATCH v2] ocxl: Allow contexts to be attached with a NULL mm

On 20/6/19 2:12 pm, Alastair D'Silva wrote:
> From: Alastair D'Silva <[email protected]>
>
> If an OpenCAPI context is to be used directly by a kernel driver, there
> may not be a suitable mm to use.
>
> The patch makes the mm parameter to ocxl_context_attach optional.
>
> Signed-off-by: Alastair D'Silva <[email protected]>

Acked-by: Andrew Donnellan <[email protected]>

> ---
> arch/powerpc/mm/book3s64/radix_tlb.c | 5 +++++
> drivers/misc/ocxl/context.c | 9 ++++++---
> drivers/misc/ocxl/link.c | 28 ++++++++++++++++++++++++----
> 3 files changed, 35 insertions(+), 7 deletions(-)
>
> diff --git a/arch/powerpc/mm/book3s64/radix_tlb.c b/arch/powerpc/mm/book3s64/radix_tlb.c
> index bb9835681315..ce8a77fae6a7 100644
> --- a/arch/powerpc/mm/book3s64/radix_tlb.c
> +++ b/arch/powerpc/mm/book3s64/radix_tlb.c
> @@ -666,6 +666,11 @@ EXPORT_SYMBOL(radix__flush_tlb_page);
> #define radix__flush_all_mm radix__local_flush_all_mm
> #endif /* CONFIG_SMP */
>
> +/*
> + * If kernel TLBIs ever become local rather than global, then
> + * drivers/misc/ocxl/link.c:ocxl_link_add_pe will need some work, as it
> + * assumes kernel TLBIs are global.
> + */
> void radix__flush_tlb_kernel_range(unsigned long start, unsigned long end)
> {
> _tlbie_pid(0, RIC_FLUSH_ALL);
> diff --git a/drivers/misc/ocxl/context.c b/drivers/misc/ocxl/context.c
> index bab9c9364184..994563a078eb 100644
> --- a/drivers/misc/ocxl/context.c
> +++ b/drivers/misc/ocxl/context.c
> @@ -69,6 +69,7 @@ static void xsl_fault_error(void *data, u64 addr, u64 dsisr)
> int ocxl_context_attach(struct ocxl_context *ctx, u64 amr, struct mm_struct *mm)
> {
> int rc;
> + unsigned long pidr = 0;
>
> // Locks both status & tidr
> mutex_lock(&ctx->status_mutex);
> @@ -77,9 +78,11 @@ int ocxl_context_attach(struct ocxl_context *ctx, u64 amr, struct mm_struct *mm)
> goto out;
> }
>
> - rc = ocxl_link_add_pe(ctx->afu->fn->link, ctx->pasid,
> - mm->context.id, ctx->tidr, amr, mm,
> - xsl_fault_error, ctx);
> + if (mm)
> + pidr = mm->context.id;
> +
> + rc = ocxl_link_add_pe(ctx->afu->fn->link, ctx->pasid, pidr, ctx->tidr,
> + amr, mm, xsl_fault_error, ctx);
> if (rc)
> goto out;
>
> diff --git a/drivers/misc/ocxl/link.c b/drivers/misc/ocxl/link.c
> index cce5b0d64505..58d111afd9f6 100644
> --- a/drivers/misc/ocxl/link.c
> +++ b/drivers/misc/ocxl/link.c
> @@ -224,6 +224,17 @@ static irqreturn_t xsl_fault_handler(int irq, void *data)
> ack_irq(spa, ADDRESS_ERROR);
> return IRQ_HANDLED;
> }
> +
> + if (!pe_data->mm) {
> + /*
> + * translation fault from a kernel context - an OpenCAPI
> + * device tried to access a bad kernel address
> + */
> + rcu_read_unlock();
> + pr_warn("Unresolved OpenCAPI xsl fault in kernel context\n");
> + ack_irq(spa, ADDRESS_ERROR);
> + return IRQ_HANDLED;
> + }
> WARN_ON(pe_data->mm->context.id != pid);
>
> if (mmget_not_zero(pe_data->mm)) {
> @@ -523,7 +534,13 @@ int ocxl_link_add_pe(void *link_handle, int pasid, u32 pidr, u32 tidr,
> pe->amr = cpu_to_be64(amr);
> pe->software_state = cpu_to_be32(SPA_PE_VALID);
>
> - mm_context_add_copro(mm);
> + /*
> + * For user contexts, register a copro so that TLBIs are seen
> + * by the nest MMU. If we have a kernel context, TLBIs are
> + * already global.
> + */
> + if (mm)
> + mm_context_add_copro(mm);
> /*
> * Barrier is to make sure PE is visible in the SPA before it
> * is used by the device. It also helps with the global TLBI
> @@ -546,7 +563,8 @@ int ocxl_link_add_pe(void *link_handle, int pasid, u32 pidr, u32 tidr,
> * have a reference on mm_users. Incrementing mm_count solves
> * the problem.
> */
> - mmgrab(mm);
> + if (mm)
> + mmgrab(mm);
> trace_ocxl_context_add(current->pid, spa->spa_mem, pasid, pidr, tidr);
> unlock:
> mutex_unlock(&spa->spa_lock);
> @@ -652,8 +670,10 @@ int ocxl_link_remove_pe(void *link_handle, int pasid)
> if (!pe_data) {
> WARN(1, "Couldn't find pe data when removing PE\n");
> } else {
> - mm_context_remove_copro(pe_data->mm);
> - mmdrop(pe_data->mm);
> + if (pe_data->mm) {
> + mm_context_remove_copro(pe_data->mm);
> + mmdrop(pe_data->mm);
> + }
> kfree_rcu(pe_data, rcu);
> }
> unlock:
>

--
Andrew Donnellan OzLabs, ADL Canberra
[email protected] IBM Australia Limited