| Message ID | 20231208033142.1673232-3-quan@os.amperecomputing.com |
|---|---|
| State | New |
| Headers | show |
| Series | i2c: aspeed: Late ack Tx done irqs and handle coalesced start with stop conditions | expand |
Hi Quan, [...] > - /* Ack all interrupts except for Rx done */ > - writel(irq_received & ~ASPEED_I2CD_INTR_RX_DONE, > - bus->base + ASPEED_I2C_INTR_STS_REG); > + > + /* > + * Early acking of INTR_RX_DONE and INTR_TX_[ACK|NAK] would indicate HW to > + * start receiving or sending new data, and this may cause a race condition > + * as the irq handler has not yet handled these irqs but is being acked. > + * Let's ack them late at the end of the irq handler when those are truly processed. > + */ > + irq_ack_last = ASPEED_I2CD_INTR_RX_DONE | ASPEED_I2CD_INTR_TX_ACK | ASPEED_I2CD_INTR_TX_NAK; > + writel(irq_received & ~irq_ack_last, bus->base + ASPEED_I2C_INTR_STS_REG); I like Andrews suggestion of having irq_ack_last as a define that is already negated, instead of negating it in the writel, which makes it a bit difficult to read. Besides, ack_last, as a name is not very meaningful, I'd rather call it irq_ack_rx_tx (or something similar). But I'm not going to block it for this, up to you if you want to send a new version. Reviewed-by: Andi Shyti <andi.shyti@kernel.org> Thanks, Andi
On 08/12/2023 11:00, Andrew Jeffery wrote: > On Fri, 2023-12-08 at 10:31 +0700, Quan Nguyen wrote: >> Commit 2be6b47211e1 ("i2c: aspeed: Acknowledge most interrupts early in >> interrupt handler") acknowledges most interrupts early before the slave >> irq handler is executed, except for the "Receive Done Interrupt status" >> which is acknowledged late in the interrupt. >> However, it has been observed that the early acknowledgment of "Transmit >> Done Interrupt Status" (with ACK or NACK) often causes the interrupt to >> be raised in READ REQUEST state, that shows the >> "Unexpected ACK on read request." complaint messages. >> >> Assuming that the "Transmit Done" interrupt should only be acknowledged >> once it is truly processed, this commit fixes that issue by acknowledging >> interrupts for both ACK and NACK cases late in the interrupt handler. >> >> Fixes: 2be6b47211e1 ("i2c: aspeed: Acknowledge most interrupts early in interrupt handler") >> Signed-off-by: Quan Nguyen <quan@os.amperecomputing.com> >> --- >> v3: >> + Fix the unconditinal write when ack the irqs [Andrew] >> + Refactor the code to enhance code readability [Quan] >> + Fix grammar in commit message [Quan] >> >> v2: >> + Split to separate series [Joel] >> + Added the Fixes line [Joel] >> + Fixed multiline comment [Joel] >> + Refactor irq clearing code [Joel, Guenter] >> + Revised commit message [Joel] >> + Revised commit message [Quan] >> + About a note to remind why the readl() should immediately follow the >> writel() to fix the race condition when clearing irq status from commit >> c926c87b8e36 ("i2c: aspeed: Avoid i2c interrupt status clear race >> condition"), I think it looks straight forward in this patch and decided >> not to add that note. [Joel] >> >> v1: >> + First introduced in >> https://lore.kernel.org/all/20210519074934.20712-1-quan@os.amperecomputing.com/ >> --- >> drivers/i2c/busses/i2c-aspeed.c | 21 +++++++++++++-------- >> 1 file changed, 13 insertions(+), 8 deletions(-) >> >> diff --git a/drivers/i2c/busses/i2c-aspeed.c b/drivers/i2c/busses/i2c-aspeed.c >> index 1c2a4f4c4e1b..967a26dd4ffa 100644 >> --- a/drivers/i2c/busses/i2c-aspeed.c >> +++ b/drivers/i2c/busses/i2c-aspeed.c >> @@ -617,13 +617,19 @@ static u32 aspeed_i2c_master_irq(struct aspeed_i2c_bus *bus, u32 irq_status) >> static irqreturn_t aspeed_i2c_bus_irq(int irq, void *dev_id) >> { >> struct aspeed_i2c_bus *bus = dev_id; >> - u32 irq_received, irq_remaining, irq_handled; >> + u32 irq_received, irq_remaining, irq_handled, irq_ack_last; > > `irq_ack_last` might be better as a macro, but you're probably saved by > the optimiser anyway. If there's another reason to do a v4 or others > are unhappy with it then consider fixing it, otherwise: > > Reviewed-by: Andrew Jeffery <andrew@codeconstruct.com.au> > I'll send out the v4 to use the defined macro instead as below: Thanks Andi, for the suggestion on the macro name. diff --git a/drivers/i2c/busses/i2c-aspeed.c b/drivers/i2c/busses/i2c-aspeed.c index 5511fd46a65e..0f67218cf04a 100644 --- a/drivers/i2c/busses/i2c-aspeed.c +++ b/drivers/i2c/busses/i2c-aspeed.c @@ -93,6 +93,10 @@ ASPEED_I2CD_INTR_RX_DONE | \ ASPEED_I2CD_INTR_TX_NAK | \ ASPEED_I2CD_INTR_TX_ACK) +#define ASPEED_I2CD_INTR_ACK_RX_TX \ + (ASPEED_I2CD_INTR_RX_DONE | \ + ASPEED_I2CD_INTR_TX_ACK | \ + ASPEED_I2CD_INTR_TX_NAK) Thanks for the review, - Quan
On 08/12/2023 10:31, Quan Nguyen wrote: > Commit 2be6b47211e1 ("i2c: aspeed: Acknowledge most interrupts early in > interrupt handler") acknowledges most interrupts early before the slave > irq handler is executed, except for the "Receive Done Interrupt status" > which is acknowledged late in the interrupt. > However, it has been observed that the early acknowledgment of "Transmit > Done Interrupt Status" (with ACK or NACK) often causes the interrupt to > be raised in READ REQUEST state, that shows the > "Unexpected ACK on read request." complaint messages. > > Assuming that the "Transmit Done" interrupt should only be acknowledged > once it is truly processed, this commit fixes that issue by acknowledging > interrupts for both ACK and NACK cases late in the interrupt handler. > > Fixes: 2be6b47211e1 ("i2c: aspeed: Acknowledge most interrupts early in interrupt handler") > Signed-off-by: Quan Nguyen <quan@os.amperecomputing.com> > --- > v3: > + Fix the unconditinal write when ack the irqs [Andrew] > + Refactor the code to enhance code readability [Quan] > + Fix grammar in commit message [Quan] > > v2: > + Split to separate series [Joel] > + Added the Fixes line [Joel] > + Fixed multiline comment [Joel] > + Refactor irq clearing code [Joel, Guenter] > + Revised commit message [Joel] > + Revised commit message [Quan] > + About a note to remind why the readl() should immediately follow the > writel() to fix the race condition when clearing irq status from commit > c926c87b8e36 ("i2c: aspeed: Avoid i2c interrupt status clear race > condition"), I think it looks straight forward in this patch and decided > not to add that note. [Joel] > > v1: > + First introduced in > https://lore.kernel.org/all/20210519074934.20712-1-quan@os.amperecomputing.com/ > --- > drivers/i2c/busses/i2c-aspeed.c | 21 +++++++++++++-------- > 1 file changed, 13 insertions(+), 8 deletions(-) > > diff --git a/drivers/i2c/busses/i2c-aspeed.c b/drivers/i2c/busses/i2c-aspeed.c > index 1c2a4f4c4e1b..967a26dd4ffa 100644 > --- a/drivers/i2c/busses/i2c-aspeed.c > +++ b/drivers/i2c/busses/i2c-aspeed.c > @@ -617,13 +617,19 @@ static u32 aspeed_i2c_master_irq(struct aspeed_i2c_bus *bus, u32 irq_status) > static irqreturn_t aspeed_i2c_bus_irq(int irq, void *dev_id) > { > struct aspeed_i2c_bus *bus = dev_id; > - u32 irq_received, irq_remaining, irq_handled; > + u32 irq_received, irq_remaining, irq_handled, irq_ack_last; > > spin_lock(&bus->lock); > irq_received = readl(bus->base + ASPEED_I2C_INTR_STS_REG); > - /* Ack all interrupts except for Rx done */ > - writel(irq_received & ~ASPEED_I2CD_INTR_RX_DONE, > - bus->base + ASPEED_I2C_INTR_STS_REG); > + > + /* > + * Early acking of INTR_RX_DONE and INTR_TX_[ACK|NAK] would indicate HW to > + * start receiving or sending new data, and this may cause a race condition > + * as the irq handler has not yet handled these irqs but is being acked. > + * Let's ack them late at the end of the irq handler when those are truly processed. > + */ > + irq_ack_last = ASPEED_I2CD_INTR_RX_DONE | ASPEED_I2CD_INTR_TX_ACK | ASPEED_I2CD_INTR_TX_NAK; > + writel(irq_received & ~irq_ack_last, bus->base + ASPEED_I2C_INTR_STS_REG); As there will be v4, I'd like to improve this line of code a bit. During debug, there are significant number of times when the irq_received has the value of 0x00000004 (INTR_RX_DONE) or 0x00000001 (INTR_TX_ACK). This makes the "irq_received & ~irq_ack_last" turn out to be 0. AFAIK, it does not make sense to keep repeatedly writing the 0 to the HW then try to re-read it with readl() every time. I'd like to change this HW access conditionally to avoid those unnecessary access. The change would looks like below: diff --git a/drivers/i2c/busses/i2c-aspeed.c b/drivers/i2c/busses/i2c-aspeed.c index 5511fd46a65e..0f67218cf04a 100644 --- a/drivers/i2c/busses/i2c-aspeed.c +++ b/drivers/i2c/busses/i2c-aspeed.c @@ -93,6 +93,10 @@ ASPEED_I2CD_INTR_RX_DONE | \ ASPEED_I2CD_INTR_TX_NAK | \ ASPEED_I2CD_INTR_TX_ACK) +#define ASPEED_I2CD_INTR_ACK_RX_TX \ + (ASPEED_I2CD_INTR_RX_DONE | \ + ASPEED_I2CD_INTR_TX_ACK | \ + ASPEED_I2CD_INTR_TX_NAK) /* 0x14 : I2CD Command/Status Register */ #define ASPEED_I2CD_SCL_LINE_STS BIT(18) @@ -622,10 +626,19 @@ static irqreturn_t aspeed_i2c_bus_irq(int irq, void *dev_id) spin_lock(&bus->lock); irq_received = readl(bus->base + ASPEED_I2C_INTR_STS_REG); - /* Ack all interrupts except for Rx done */ - writel(irq_received & ~ASPEED_I2CD_INTR_RX_DONE, - bus->base + ASPEED_I2C_INTR_STS_REG); - readl(bus->base + ASPEED_I2C_INTR_STS_REG); + + /* + * Early acking of INTR_RX_DONE and INTR_TX_[ACK|NAK] would + ... + */ + if (irq_received & ~ASPEED_I2CD_INTR_ACK_RX_TX) { + writel(irq_received & ~ASPEED_I2CD_INTR_ACK_RX_TX, + bus->base + ASPEED_I2C_INTR_STS_REG); + readl(bus->base + ASPEED_I2C_INTR_STS_REG); + } + irq_received &= ASPEED_I2CD_INTR_RECV_MASK; irq_remaining = irq_received; Thanks, - Quan
On 10/12/2023 03:44, Andi Shyti wrote: > Hi Quan, > > [...] > >> - /* Ack all interrupts except for Rx done */ >> - writel(irq_received & ~ASPEED_I2CD_INTR_RX_DONE, >> - bus->base + ASPEED_I2C_INTR_STS_REG); >> + >> + /* >> + * Early acking of INTR_RX_DONE and INTR_TX_[ACK|NAK] would indicate HW to >> + * start receiving or sending new data, and this may cause a race condition >> + * as the irq handler has not yet handled these irqs but is being acked. >> + * Let's ack them late at the end of the irq handler when those are truly processed. >> + */ >> + irq_ack_last = ASPEED_I2CD_INTR_RX_DONE | ASPEED_I2CD_INTR_TX_ACK | ASPEED_I2CD_INTR_TX_NAK; >> + writel(irq_received & ~irq_ack_last, bus->base + ASPEED_I2C_INTR_STS_REG); > > I like Andrews suggestion of having irq_ack_last as a define that > is already negated, instead of negating it in the writel, which > makes it a bit difficult to read. > Yes, but the it still need to negate again when do the write to late ack them later in the end of irq handler. So I'll keep the define as below in my v4: +#define ASPEED_I2CD_INTR_ACK_RX_TX \ + (ASPEED_I2CD_INTR_RX_DONE | \ + ASPEED_I2CD_INTR_TX_ACK | \ + ASPEED_I2CD_INTR_TX_NAK) The early ack will look like this: + writel(irq_received & ~ASPEED_I2CD_INTR_ACK_RX_TX, + bus->base + ASPEED_I2C_INTR_STS_REG); + readl(bus->base + ASPEED_I2C_INTR_STS_REG); And the late ack: - /* Ack Rx done */ - if (irq_received & ASPEED_I2CD_INTR_RX_DONE) { - writel(ASPEED_I2CD_INTR_RX_DONE, + if (irq_received & ASPEED_I2CD_INTR_ACK_RX_TX) { + writel(irq_received & ASPEED_I2CD_INTR_ACK_RX_TX, bus->base + ASPEED_I2C_INTR_STS_REG); readl(bus->base + ASPEED_I2C_INTR_STS_REG); } > Besides, ack_last, as a name is not very meaningful, I'd rather > call it irq_ack_rx_tx (or something similar). > > But I'm not going to block it for this, up to you if you want to > send a new version. > > Reviewed-by: Andi Shyti <andi.shyti@kernel.org> > Thanks, Andi for the comments. I will send out v4 to address those. - Quan
diff --git a/drivers/i2c/busses/i2c-aspeed.c b/drivers/i2c/busses/i2c-aspeed.c index 1c2a4f4c4e1b..967a26dd4ffa 100644 --- a/drivers/i2c/busses/i2c-aspeed.c +++ b/drivers/i2c/busses/i2c-aspeed.c @@ -617,13 +617,19 @@ static u32 aspeed_i2c_master_irq(struct aspeed_i2c_bus *bus, u32 irq_status) static irqreturn_t aspeed_i2c_bus_irq(int irq, void *dev_id) { struct aspeed_i2c_bus *bus = dev_id; - u32 irq_received, irq_remaining, irq_handled; + u32 irq_received, irq_remaining, irq_handled, irq_ack_last; spin_lock(&bus->lock); irq_received = readl(bus->base + ASPEED_I2C_INTR_STS_REG); - /* Ack all interrupts except for Rx done */ - writel(irq_received & ~ASPEED_I2CD_INTR_RX_DONE, - bus->base + ASPEED_I2C_INTR_STS_REG); + + /* + * Early acking of INTR_RX_DONE and INTR_TX_[ACK|NAK] would indicate HW to + * start receiving or sending new data, and this may cause a race condition + * as the irq handler has not yet handled these irqs but is being acked. + * Let's ack them late at the end of the irq handler when those are truly processed. + */ + irq_ack_last = ASPEED_I2CD_INTR_RX_DONE | ASPEED_I2CD_INTR_TX_ACK | ASPEED_I2CD_INTR_TX_NAK; + writel(irq_received & ~irq_ack_last, bus->base + ASPEED_I2C_INTR_STS_REG); readl(bus->base + ASPEED_I2C_INTR_STS_REG); irq_received &= ASPEED_I2CD_INTR_RECV_MASK; irq_remaining = irq_received; @@ -667,12 +673,11 @@ static irqreturn_t aspeed_i2c_bus_irq(int irq, void *dev_id) "irq handled != irq. expected 0x%08x, but was 0x%08x\n", irq_received, irq_handled); - /* Ack Rx done */ - if (irq_received & ASPEED_I2CD_INTR_RX_DONE) { - writel(ASPEED_I2CD_INTR_RX_DONE, - bus->base + ASPEED_I2C_INTR_STS_REG); + if (irq_received & irq_ack_last) { + writel(irq_received & irq_ack_last, bus->base + ASPEED_I2C_INTR_STS_REG); readl(bus->base + ASPEED_I2C_INTR_STS_REG); } + spin_unlock(&bus->lock); return irq_remaining ? IRQ_NONE : IRQ_HANDLED; }
Commit 2be6b47211e1 ("i2c: aspeed: Acknowledge most interrupts early in interrupt handler") acknowledges most interrupts early before the slave irq handler is executed, except for the "Receive Done Interrupt status" which is acknowledged late in the interrupt. However, it has been observed that the early acknowledgment of "Transmit Done Interrupt Status" (with ACK or NACK) often causes the interrupt to be raised in READ REQUEST state, that shows the "Unexpected ACK on read request." complaint messages. Assuming that the "Transmit Done" interrupt should only be acknowledged once it is truly processed, this commit fixes that issue by acknowledging interrupts for both ACK and NACK cases late in the interrupt handler. Fixes: 2be6b47211e1 ("i2c: aspeed: Acknowledge most interrupts early in interrupt handler") Signed-off-by: Quan Nguyen <quan@os.amperecomputing.com> --- v3: + Fix the unconditinal write when ack the irqs [Andrew] + Refactor the code to enhance code readability [Quan] + Fix grammar in commit message [Quan] v2: + Split to separate series [Joel] + Added the Fixes line [Joel] + Fixed multiline comment [Joel] + Refactor irq clearing code [Joel, Guenter] + Revised commit message [Joel] + Revised commit message [Quan] + About a note to remind why the readl() should immediately follow the writel() to fix the race condition when clearing irq status from commit c926c87b8e36 ("i2c: aspeed: Avoid i2c interrupt status clear race condition"), I think it looks straight forward in this patch and decided not to add that note. [Joel] v1: + First introduced in https://lore.kernel.org/all/20210519074934.20712-1-quan@os.amperecomputing.com/ --- drivers/i2c/busses/i2c-aspeed.c | 21 +++++++++++++-------- 1 file changed, 13 insertions(+), 8 deletions(-)