From 8f9ed93d09a97444733d492a3bbf66bcb786a777 Mon Sep 17 00:00:00 2001 From: Douglas Anderson Date: Mon, 15 Jun 2020 20:28:59 +0300 Subject: [PATCH] ath10k: Wait until copy complete is actually done before completing On wcn3990 we have "per_ce_irq = true". That makes the ath10k_ce_interrupt_summary() function always return 0xfff. The ath10k_ce_per_engine_service_any() function will see this and think that _all_ copy engines have an interrupt. Without checking, the ath10k_ce_per_engine_service() assumes that if it's called that the "copy complete" (cc) interrupt fired. This combination seems bad. Let's add a check to make sure that the "copy complete" interrupt actually fired in ath10k_ce_per_engine_service(). This might fix a hard-to-reproduce failure where it appears that the copy complete handlers run before the copy is really complete. Specifically a symptom was that we were seeing this on a Qualcomm sc7180 board: arm-smmu 15000000.iommu: Unhandled context fault: fsr=0x402, iova=0x7fdd45780, fsynr=0x30003, cbfrsynra=0xc1, cb=10 Even on platforms that don't have wcn3990 this still seems like it would be a sane thing to do. Specifically the current IRQ handler comments indicate that there might be other misc interrupt sources firing that need to be cleared. If one of those sources was the one that caused the IRQ handler to be called it would also be important to double-check that the interrupt we cared about actually fired. Tested-on: WCN3990 SNOC WLAN.HL.3.2.2-00490-QCAHLSWMTPL-1 Signed-off-by: Douglas Anderson Signed-off-by: Kalle Valo Link: https://lore.kernel.org/r/20200609082015.1.Ife398994e5a0a6830e4d4a16306ef36e0144e7ba@changeid --- drivers/net/wireless/ath/ath10k/ce.c | 30 +++++++++++++++++++++--------- 1 file changed, 21 insertions(+), 9 deletions(-) diff --git a/drivers/net/wireless/ath/ath10k/ce.c b/drivers/net/wireless/ath/ath10k/ce.c index 294fbc1..ffdd4b9 100644 --- a/drivers/net/wireless/ath/ath10k/ce.c +++ b/drivers/net/wireless/ath/ath10k/ce.c @@ -481,6 +481,15 @@ static inline void ath10k_ce_engine_int_status_clear(struct ath10k *ar, ath10k_ce_write32(ar, ce_ctrl_addr + wm_regs->addr, mask); } +static inline bool ath10k_ce_engine_int_status_check(struct ath10k *ar, + u32 ce_ctrl_addr, + unsigned int mask) +{ + struct ath10k_hw_ce_host_wm_regs *wm_regs = ar->hw_ce_regs->wm_regs; + + return ath10k_ce_read32(ar, ce_ctrl_addr + wm_regs->addr) & mask; +} + /* * Guts of ath10k_ce_send. * The caller takes responsibility for any needed locking. @@ -1301,19 +1310,22 @@ void ath10k_ce_per_engine_service(struct ath10k *ar, unsigned int ce_id) spin_lock_bh(&ce->ce_lock); - /* Clear the copy-complete interrupts that will be handled here. */ - ath10k_ce_engine_int_status_clear(ar, ctrl_addr, - wm_regs->cc_mask); + if (ath10k_ce_engine_int_status_check(ar, ctrl_addr, + wm_regs->cc_mask)) { + /* Clear before handling */ + ath10k_ce_engine_int_status_clear(ar, ctrl_addr, + wm_regs->cc_mask); - spin_unlock_bh(&ce->ce_lock); + spin_unlock_bh(&ce->ce_lock); - if (ce_state->recv_cb) - ce_state->recv_cb(ce_state); + if (ce_state->recv_cb) + ce_state->recv_cb(ce_state); - if (ce_state->send_cb) - ce_state->send_cb(ce_state); + if (ce_state->send_cb) + ce_state->send_cb(ce_state); - spin_lock_bh(&ce->ce_lock); + spin_lock_bh(&ce->ce_lock); + } /* * Misc CE interrupts are not being handled, but still need -- 2.7.4