On 1/10/2024 3:41 AM, Hillf Danton wrote:
> On Tue, 9 Jan 2024 11:37:45 -0800 Elliot Berman <quic_eberman@xxxxxxxxxxx>
>> +
>> +static irqreturn_t gunyah_rm_tx(int irq, void *data)
>> +{
>> + struct gunyah_rm *rm = data;
>> +
>> + complete_all(&rm->send_ready);
>> +
>> + return IRQ_HANDLED;
>> +}
>> +
>> +static int gunyah_rm_msgq_send(struct gunyah_rm *rm, size_t size, bool push)
>> + __must_hold(&rm->send_lock)
>> +{
>> + const u64 tx_flags = push ? GUNYAH_HYPERCALL_MSGQ_TX_FLAGS_PUSH : 0;
>> + enum gunyah_error gunyah_error;
>> + void *data = &rm->send_msg[0];
>> + bool ready;
>
> lockdep_assert_held(&rm->send_lock); instead of __must_hold
>> +
>> +again:
>> + wait_for_completion(&rm->send_ready);
>> + /* reinit completion before hypercall. As soon as hypercall returns, we could get the
>> + * ready interrupt. This might be before we have time to reinit the completion
>> + */
>> + reinit_completion(&rm->send_ready);
>
> Given wait_for_completion(&rm->send_ready) with rm->send_lock held,
> complete(&rm->send_ready) works for you with the bonus of cutting this
> reinit off.
>
Good point, thanks! Applied both suggestions
- Elliot
