diff mbox

[1/2] iommu/mtk: Avoid redundant TLB syncs locally

Message ID 367e86d1e73867eaf0467732c10c7dc74380f97c.1499360104.git.robin.murphy@arm.com (mailing list archive)
State New, archived
Headers show

Commit Message

Robin Murphy July 6, 2017, 4:55 p.m. UTC
Under certain circumstances, the io-pgtable code may end up issuing two
TLB sync operations without any intervening invalidations. This goes
badly for the M4U hardware, since it means the second sync ends up
polling for a non-existent operation to finish, and as a result times
out and warns. The io_pgtable_tlb_* helpers implement a high-level
optimisation to avoid issuing the second sync at all in such cases, but
in order to work correctly that requires all pagetable operations to be
serialised under a lock, thus is no longer applicable to all io-pgtable
users.

Since we're the only user actually relying on this flag for correctness,
let's reimplement it locally to avoid the headache of trying to make the
high-level version concurrency-safe for other users.

CC: Yong Wu <yong.wu@mediatek.com>
CC: Matthias Brugger <matthias.bgg@gmail.com>
Signed-off-by: Robin Murphy <robin.murphy@arm.com>
---
 drivers/iommu/mtk_iommu.c | 6 ++++++
 drivers/iommu/mtk_iommu.h | 1 +
 2 files changed, 7 insertions(+)

Comments

Yong Wu (吴勇) July 7, 2017, 3:06 a.m. UTC | #1
On Thu, 2017-07-06 at 17:55 +0100, Robin Murphy wrote:
> Under certain circumstances, the io-pgtable code may end up issuing two
> TLB sync operations without any intervening invalidations. This goes
> badly for the M4U hardware, since it means the second sync ends up
> polling for a non-existent operation to finish, and as a result times
> out and warns. The io_pgtable_tlb_* helpers implement a high-level
> optimisation to avoid issuing the second sync at all in such cases, but
> in order to work correctly that requires all pagetable operations to be
> serialised under a lock, thus is no longer applicable to all io-pgtable
> users.
> 
> Since we're the only user actually relying on this flag for correctness,
> let's reimplement it locally to avoid the headache of trying to make the
> high-level version concurrency-safe for other users.
> 
> CC: Yong Wu <yong.wu@mediatek.com>
> CC: Matthias Brugger <matthias.bgg@gmail.com>
> Signed-off-by: Robin Murphy <robin.murphy@arm.com>

Thanks Robin.

Tested-by: Yong Wu <yong.wu@mediatek.com>

> ---
>  drivers/iommu/mtk_iommu.c | 6 ++++++
>  drivers/iommu/mtk_iommu.h | 1 +
>  2 files changed, 7 insertions(+)
> 
> diff --git a/drivers/iommu/mtk_iommu.c b/drivers/iommu/mtk_iommu.c
> index 5d14cd15198d..91c6d367ab35 100644
> --- a/drivers/iommu/mtk_iommu.c
> +++ b/drivers/iommu/mtk_iommu.c
> @@ -129,6 +129,7 @@ static void mtk_iommu_tlb_add_flush_nosync(unsigned long iova, size_t size,
>  	writel_relaxed(iova, data->base + REG_MMU_INVLD_START_A);
>  	writel_relaxed(iova + size - 1, data->base + REG_MMU_INVLD_END_A);
>  	writel_relaxed(F_MMU_INV_RANGE, data->base + REG_MMU_INVALIDATE);
> +	data->tlb_flush_active = true;
>  }
>  
>  static void mtk_iommu_tlb_sync(void *cookie)
> @@ -137,6 +138,10 @@ static void mtk_iommu_tlb_sync(void *cookie)
>  	int ret;
>  	u32 tmp;
>  
> +	/* Avoid timing out if there's nothing to wait for */
> +	if (!data->tlb_flush_active)
> +		return;
> +
>  	ret = readl_poll_timeout_atomic(data->base + REG_MMU_CPE_DONE, tmp,
>  					tmp != 0, 10, 100000);
>  	if (ret) {
> @@ -146,6 +151,7 @@ static void mtk_iommu_tlb_sync(void *cookie)
>  	}
>  	/* Clear the CPE status */
>  	writel_relaxed(0, data->base + REG_MMU_CPE_DONE);
> +	data->tlb_flush_active = false;
>  }
>  
>  static const struct iommu_gather_ops mtk_iommu_gather_ops = {
> diff --git a/drivers/iommu/mtk_iommu.h b/drivers/iommu/mtk_iommu.h
> index 2a28eadeea0e..c06cc91b5d9a 100644
> --- a/drivers/iommu/mtk_iommu.h
> +++ b/drivers/iommu/mtk_iommu.h
> @@ -47,6 +47,7 @@ struct mtk_iommu_data {
>  	struct iommu_group		*m4u_group;
>  	struct mtk_smi_iommu		smi_imu;      /* SMI larb iommu info */
>  	bool                            enable_4GB;
> +	bool				tlb_flush_active;
>  
>  	struct iommu_device		iommu;
>  };
diff mbox

Patch

diff --git a/drivers/iommu/mtk_iommu.c b/drivers/iommu/mtk_iommu.c
index 5d14cd15198d..91c6d367ab35 100644
--- a/drivers/iommu/mtk_iommu.c
+++ b/drivers/iommu/mtk_iommu.c
@@ -129,6 +129,7 @@  static void mtk_iommu_tlb_add_flush_nosync(unsigned long iova, size_t size,
 	writel_relaxed(iova, data->base + REG_MMU_INVLD_START_A);
 	writel_relaxed(iova + size - 1, data->base + REG_MMU_INVLD_END_A);
 	writel_relaxed(F_MMU_INV_RANGE, data->base + REG_MMU_INVALIDATE);
+	data->tlb_flush_active = true;
 }
 
 static void mtk_iommu_tlb_sync(void *cookie)
@@ -137,6 +138,10 @@  static void mtk_iommu_tlb_sync(void *cookie)
 	int ret;
 	u32 tmp;
 
+	/* Avoid timing out if there's nothing to wait for */
+	if (!data->tlb_flush_active)
+		return;
+
 	ret = readl_poll_timeout_atomic(data->base + REG_MMU_CPE_DONE, tmp,
 					tmp != 0, 10, 100000);
 	if (ret) {
@@ -146,6 +151,7 @@  static void mtk_iommu_tlb_sync(void *cookie)
 	}
 	/* Clear the CPE status */
 	writel_relaxed(0, data->base + REG_MMU_CPE_DONE);
+	data->tlb_flush_active = false;
 }
 
 static const struct iommu_gather_ops mtk_iommu_gather_ops = {
diff --git a/drivers/iommu/mtk_iommu.h b/drivers/iommu/mtk_iommu.h
index 2a28eadeea0e..c06cc91b5d9a 100644
--- a/drivers/iommu/mtk_iommu.h
+++ b/drivers/iommu/mtk_iommu.h
@@ -47,6 +47,7 @@  struct mtk_iommu_data {
 	struct iommu_group		*m4u_group;
 	struct mtk_smi_iommu		smi_imu;      /* SMI larb iommu info */
 	bool                            enable_4GB;
+	bool				tlb_flush_active;
 
 	struct iommu_device		iommu;
 };