Re: [RFC PATCH v2 0/5] physmem: Have flaview API check bus permission from MemTxAttrs argument

[Peter Xu <peterx@redhat.com>]

Hi, Peter, Gerd,

On Tue, Aug 24, 2021 at 02:01:53PM +0200, Gerd Hoffmann wrote:
>   Hi,
> 
> > I was vaguely tossing an idea around in the back of my mind
> > about whether you could have a flag on devices that marked
> > them as "this device is currently involved in IO", such that
> > you could then just fail the last DMA (or qemu_irq_set, or
> > whatever) that would complete the loop back to a device that
> > was already doing IO. But that would need a lot of thinking
> > through to figure out if it's feasible, and it's probably
> > a lot of code change.

(Thanks for the write-up, Peter; it helps a lot)

> 
> Quick & dirty hack trying the above.  Not much code, it is opt-in per
> MemoryRegion (so less overhead for devices which already handle all DMA
> in a BH), tracks state in DeviceState.  Adds a check to a rather hot
> code path though.  Not tested yet (stopped investigating when I noticed
> Philippe tries to fix the same thing with another approach).  Not
> benchmarked.
> 
> Maybe it helps ...
> 
> take care,
>   Gerd
> 
> From 80e58a2cd2c630f0bddd9d0eaee71abb7eeb9440 Mon Sep 17 00:00:00 2001
> From: Gerd Hoffmann <kraxel@redhat.com>
> Date: Tue, 17 Aug 2021 07:35:37 +0200
> Subject: [PATCH] allow track active mmio handlers
> 
> ---
>  include/exec/memory.h  |  1 +
>  include/hw/qdev-core.h |  1 +
>  softmmu/memory.c       | 24 ++++++++++++++++++++++--
>  3 files changed, 24 insertions(+), 2 deletions(-)
> 
> diff --git a/include/exec/memory.h b/include/exec/memory.h
> index c3d417d317f0..b1883d45e817 100644
> --- a/include/exec/memory.h
> +++ b/include/exec/memory.h
> @@ -265,6 +265,7 @@ struct MemoryRegionOps {
>           */
>          bool unaligned;
>      } impl;
> +    bool block_reenter;
>  };
>  
>  typedef struct MemoryRegionClass {
> diff --git a/include/hw/qdev-core.h b/include/hw/qdev-core.h
> index bafc311bfa1b..4cf281a81fa9 100644
> --- a/include/hw/qdev-core.h
> +++ b/include/hw/qdev-core.h
> @@ -191,6 +191,7 @@ struct DeviceState {
>      int instance_id_alias;
>      int alias_required_for_version;
>      ResettableState reset;
> +    bool io_handler_active;
>  };
>  
>  struct DeviceListener {
> diff --git a/softmmu/memory.c b/softmmu/memory.c
> index bfedaf9c4dfc..5eb5dd465dd2 100644
> --- a/softmmu/memory.c
> +++ b/softmmu/memory.c
> @@ -437,7 +437,18 @@ static MemTxResult  memory_region_read_accessor(MemoryRegion *mr,
>  {
>      uint64_t tmp;
>  
> -    tmp = mr->ops->read(mr->opaque, addr, size);
> +    if (mr->ops->block_reenter) {
> +        DeviceState *dev = DEVICE(mr->owner);
> +        if (!dev->io_handler_active) {
> +            dev->io_handler_active = true;
> +            tmp = mr->ops->read(mr->opaque, addr, size);
> +            dev->io_handler_active = false;
> +        } else {
> +            tmp = MEMTX_OK;
> +        }
> +    } else {
> +        tmp = mr->ops->read(mr->opaque, addr, size);
> +    }
>      if (mr->subpage) {
>          trace_memory_region_subpage_read(get_cpu_index(), mr, addr, tmp, size);
>      } else if (trace_event_get_state_backends(TRACE_MEMORY_REGION_OPS_READ)) {
> @@ -489,7 +500,16 @@ static MemTxResult memory_region_write_accessor(MemoryRegion *mr,
>          trace_memory_region_ops_write(get_cpu_index(), mr, abs_addr, tmp, size,
>                                        memory_region_name(mr));
>      }
> -    mr->ops->write(mr->opaque, addr, tmp, size);
> +    if (mr->ops->block_reenter) {
> +        DeviceState *dev = DEVICE(mr->owner);
> +        if (!dev->io_handler_active) {
> +            dev->io_handler_active = true;
> +            mr->ops->write(mr->opaque, addr, tmp, size);
> +            dev->io_handler_active = false;
> +        }
> +    } else {
> +        mr->ops->write(mr->opaque, addr, tmp, size);
> +    }
>      return MEMTX_OK;
>  }

Can I read this as a better approach if it still allows P2P so things that
Paolo and Qiang used to mention will still work?

https://lore.kernel.org/qemu-devel/7e4fd726-07e9-dc09-d66b-5692dd51820f@redhat.com/
https://lore.kernel.org/qemu-devel/CAKXe6S+v4z_PYbZ6MMzEZk7Q0Qc+q9tzL+a8918U_-XR=aj7RA@mail.gmail.com/

Can we do that similarly for qemu_set_irq() and friends but based on IRQState?

Thanks,

-- 
Peter Xu