On 03/07/17 12:50, Jassi Brar wrote: > On 3 July 2017 at 15:05, Sudeep Holla <sudeep.ho...@arm.com> wrote: >> >> >> On 03/07/17 09:57, Sudeep Holla wrote: >>> >>> >>> On 01/07/17 12:25, Jassi Brar wrote: >>>> On Fri, Jun 30, 2017 at 8:44 PM, Sudeep Holla <sudeep.ho...@arm.com> wrote: >>>>> Currently the mailbox framework sets txdone_method to TXDONE_BY_POLL if >>>>> the controller sets txdone_by_poll. However some clients can have a >>>>> mechanism to do TXDONE_BY_ACK which they can specify by knows_txdone. >>>>> However, we endup setting both TXDONE_BY_POLL and TXDONE_BY_ACK in that >>>>> case. In such scenario, we may end up with below warnings as the tx >>>>> ticker is run both by mailbox framework and the client. >>>>> >>>>> WARNING: CPU: 1 PID: 0 at kernel/time/hrtimer.c:805 >>>>> hrtimer_forward+0x88/0xd8 >>>>> CPU: 1 PID: 0 Comm: swapper/1 Not tainted 4.12.0-rc5 #242 >>>>> Hardware name: ARM LTD ARM Juno Development Platform >>>>> task: ffff8009768ca700 task.stack: ffff8009768f8000 >>>>> PC is at hrtimer_forward+0x88/0xd8 >>>>> LR is at txdone_hrtimer+0xd4/0xf8 >>>>> Call trace: >>>>> hrtimer_forward+0x88/0xd8 >>>>> __hrtimer_run_queues+0xe4/0x158 >>>>> hrtimer_interrupt+0xa4/0x220 >>>>> arch_timer_handler_phys+0x30/0x40 >>>>> handle_percpu_devid_irq+0x78/0x130 >>>>> generic_handle_irq+0x24/0x38 >>>>> __handle_domain_irq+0x5c/0xb8 >>>>> gic_handle_irq+0x54/0xa8 >>>>> >>>>> This patch fixes the issue by resetting TXDONE_BY_POLL if client has set >>>>> knows_txdone. >>>>> >>>>> Cc: Alexey Klimov <alexey.kli...@arm.com> >>>>> Cc: Jassi Brar <jaswinder.si...@linaro.org> >>>>> Signed-off-by: Sudeep Holla <sudeep.ho...@arm.com> >>>>> --- >>>>> drivers/mailbox/mailbox.c | 2 +- >>>>> 1 file changed, 1 insertion(+), 1 deletion(-) >>>>> >>>>> diff --git a/drivers/mailbox/mailbox.c b/drivers/mailbox/mailbox.c >>>>> index 537f4f6d009b..8da30f833262 100644 >>>>> --- a/drivers/mailbox/mailbox.c >>>>> +++ b/drivers/mailbox/mailbox.c >>>>> @@ -351,7 +351,7 @@ struct mbox_chan *mbox_request_channel(struct >>>>> mbox_client *cl, int index) >>>>> init_completion(&chan->tx_complete); >>>>> >>>>> if (chan->txdone_method == TXDONE_BY_POLL && cl->knows_txdone) >>>>> - chan->txdone_method |= TXDONE_BY_ACK; >>>>> + chan->txdone_method = TXDONE_BY_ACK; >>>>> >>>> It has to be restored back in mbox_free_channel() >>> >>> Ah right, will fix and repost. >>> >> >> I was too fast to response, I see we already take care of that in free >> channel. So I think it should be fine as is. >> >> if (chan->txdone_method == (TXDONE_BY_POLL | TXDONE_BY_ACK)) >> chan->txdone_method = TXDONE_BY_POLL; >> > You were too fast this time :) >
Indeed, my understanding was correct before. > chan->txdone_method == (TXDONE_BY_POLL | TXDONE_BY_ACK) > > is not same as > > (chan->txdone_method == TXDONE_BY_POLL || chan->txdone_method == > TXDONE_BY_ACK) > Ah crap, yes and I have no idea why I thought so when I was about to change it :( -- Regards, Sudeep
