2015-01-31 00:14:44

by Peter Oh

[permalink] [raw]
Subject: [PATCH v2] ath10k: Replace ioread with mb to drain write buffer

Using ioread() to perform draining write buffer is excessive.
Use compact API, mb(), that intended to be used for the case.
It reduces total 14 CPU clocks per interrupt.

Signed-off-by: Peter Oh <[email protected]>
---
drivers/net/wireless/ath/ath10k/pci.c | 12 ++++--------
1 file changed, 4 insertions(+), 8 deletions(-)

diff --git a/drivers/net/wireless/ath/ath10k/pci.c b/drivers/net/wireless/ath/ath10k/pci.c
index e6972b0..f1e6980 100644
--- a/drivers/net/wireless/ath/ath10k/pci.c
+++ b/drivers/net/wireless/ath/ath10k/pci.c
@@ -353,10 +353,8 @@ static void ath10k_pci_disable_and_clear_legacy_irq(struct ath10k *ar)
ath10k_pci_write32(ar, SOC_CORE_BASE_ADDRESS + PCIE_INTR_CLR_ADDRESS,
PCIE_INTR_FIRMWARE_MASK | PCIE_INTR_CE_MASK_ALL);

- /* IMPORTANT: this extra read transaction is required to
- * flush the posted write buffer. */
- (void)ath10k_pci_read32(ar, SOC_CORE_BASE_ADDRESS +
- PCIE_INTR_ENABLE_ADDRESS);
+ /* drain write buffer */
+ mb();
}

static void ath10k_pci_enable_legacy_irq(struct ath10k *ar)
@@ -365,10 +363,8 @@ static void ath10k_pci_enable_legacy_irq(struct ath10k *ar)
PCIE_INTR_ENABLE_ADDRESS,
PCIE_INTR_FIRMWARE_MASK | PCIE_INTR_CE_MASK_ALL);

- /* IMPORTANT: this extra read transaction is required to
- * flush the posted write buffer. */
- (void)ath10k_pci_read32(ar, SOC_CORE_BASE_ADDRESS +
- PCIE_INTR_ENABLE_ADDRESS);
+ /* drain write buffer */
+ mb();
}

static inline const char *ath10k_pci_get_irq_method(struct ath10k *ar)
--
1.9.1



2015-02-06 09:05:29

by Jakub Kicinski

[permalink] [raw]
Subject: Re: [PATCH v2] ath10k: Replace ioread with mb to drain write buffer

On Fri, 30 Jan 2015 16:14:30 -0800, Peter Oh wrote:
> Using ioread() to perform draining write buffer is excessive.
> Use compact API, mb(), that intended to be used for the case.
> It reduces total 14 CPU clocks per interrupt.
>
> Signed-off-by: Peter Oh <[email protected]>

I have no idea what the code does but this change looks suspicious.
Usually the point of ioread() is to flush the interconnect buffers
while mb() ensures ordering only from the CPU perspective.

Could you provide the reason *why* flushing buffers is unnecessary
in the commit message?

2015-02-06 14:53:48

by Jakub Kicinski

[permalink] [raw]
Subject: Re: [PATCH v2] ath10k: Replace ioread with mb to drain write buffer

On Fri, 6 Feb 2015 09:58:46 +0100, Jakub KiciƄski wrote:
> On Fri, 30 Jan 2015 16:14:30 -0800, Peter Oh wrote:
> > Using ioread() to perform draining write buffer is excessive.
> > Use compact API, mb(), that intended to be used for the case.
> > It reduces total 14 CPU clocks per interrupt.
> >
> > Signed-off-by: Peter Oh <[email protected]>
>
> I have no idea what the code does but this change looks suspicious.
> Usually the point of ioread() is to flush the interconnect buffers
> while mb() ensures ordering only from the CPU perspective.
>
> Could you provide the reason *why* flushing buffers is unnecessary
> in the commit message?

I just noticed the discussion on the first version of the patch.
Sorry about the noise.