Re: [Qemu-devel] [PATCH 1/2] hw: add isa-debug-exit device v3

["Hervé Poussineau" <hpoussin@reactos.org>]

Andreas Färber a écrit :
> Am 13.12.2012 15:48, schrieb Lucas Meneghel Rodrigues:
>> From: Gerd Hoffmann <kraxel@redhat.com>
>>
>> When present it makes qemu exit on any write.
>> Mapped to port 0x501 by default.
>>
>> Without this patch Anthony doesn't allow me to
>> remove the bochs bios debug ports because his
>> test suite uses this.
>>
>> Changes from v2: Fixed rebase mistake
>>
>> Changes from v1: Add access_size property to the
>> device, needed for kvm-unit-tests.
>>
>> Signed-off-by: Gerd Hoffmann <kraxel@redhat.com>
> 
>> ---
>>  hw/debugexit.c        |   73 +++++++++++++++++++++++++++++++++++++++++++++++++
>>  hw/i386/Makefile.objs |    2 +-
>>  2 files changed, 74 insertions(+), 1 deletion(-)
>>  create mode 100644 hw/debugexit.c
>>
>> diff --git a/hw/debugexit.c b/hw/debugexit.c
>> new file mode 100644
>> index 0000000..dfe1b2d
>> --- /dev/null
>> +++ b/hw/debugexit.c
>> @@ -0,0 +1,73 @@
>> +/*
> 
> Add a one-line description plus some copyright line?
> 
>> + * This program is free software; you can redistribute it and/or
>> + * modify it under the terms of the GNU General Public License as
>> + * published by the Free Software Foundation; either version 2 or
>> + * (at your option) version 3 of the License.
> 
> Why not "any later version"?
> 
>> + */
>> +
>> +#include "hw.h"
>> +#include "isa.h"
>> +
>> +#define TYPE_ISA_DEBUG_EXIT_DEVICE "isa-debug-exit"
>> +#define ISA_DEBUG_EXIT_DEVICE(obj) \
>> +     OBJECT_CHECK(ISADebugExitState, (obj), TYPE_ISA_DEBUG_EXIT_DEVICE)
>> +
>> +typedef struct ISADebugExitState {
>> +    ISADevice parent_obj;
>> +
>> +    uint32_t iobase;
>> +    uint8_t access_size;
>> +    MemoryRegion io;
>> +} ISADebugExitState;
>> +
>> +static void debug_exit_write(void *opaque, hwaddr addr, uint64_t val,
>> +                             unsigned width)
>> +{
>> +    exit((val << 1) | 1);
>> +}
>> +
>> +static const MemoryRegionOps debug_exit_ops = {
>> +    .write = debug_exit_write,
>> +    .valid.min_access_size = 1,
>> +    .valid.max_access_size = 1,
>> +    .endianness = DEVICE_LITTLE_ENDIAN,
>> +};
>> +
>> +static int debug_exit_initfn(ISADevice *dev)
>> +{
>> +    ISADebugExitState *isa = ISA_DEBUG_EXIT_DEVICE(dev);
>> +
>> +    memory_region_init_io(&isa->io, &debug_exit_ops, isa,
>> +                          TYPE_ISA_DEBUG_EXIT_DEVICE, isa->access_size);
>> +    memory_region_add_subregion(isa_address_space_io(dev),
>> +                                isa->iobase, &isa->io);
>> +    return 0;
>> +}
>> +
>> +static Property debug_exit_properties[] = {
>> +    DEFINE_PROP_HEX32("iobase", ISADebugExitState, iobase, 0x501),
>> +    DEFINE_PROP_UINT8("access-size", ISADebugExitState, access_size, 1),
>> +    DEFINE_PROP_END_OF_LIST(),
>> +};
>> +
>> +static void debug_exit_class_initfn(ObjectClass *klass, void *data)
>> +{
>> +    DeviceClass *dc = DEVICE_CLASS(klass);
>> +    ISADeviceClass *ic = ISA_DEVICE_CLASS(klass);
>> +    ic->init = debug_exit_initfn;
>> +    dc->props = debug_exit_properties;
>> +}
>> +
>> +static TypeInfo debug_exit_info = {
> 
> static const, if resubmitted
> 
>> +    .name          = TYPE_ISA_DEBUG_EXIT_DEVICE,
>> +    .parent        = TYPE_ISA_DEVICE,
>> +    .instance_size = sizeof(ISADebugExitState),
>> +    .class_init    = debug_exit_class_initfn,
>> +};
>> +
>> +static void debug_exit_register_types(void)
>> +{
>> +    type_register_static(&debug_exit_info);
>> +}
>> +
>> +type_init(debug_exit_register_types)
>> diff --git a/hw/i386/Makefile.objs b/hw/i386/Makefile.objs
>> index 0d3f6a8..56aaa9d 100644
>> --- a/hw/i386/Makefile.objs
>> +++ b/hw/i386/Makefile.objs
>> @@ -3,7 +3,7 @@ obj-y += apic_common.o apic.o kvmvapic.o
>>  obj-y += sga.o ioapic_common.o ioapic.o piix_pci.o
>>  obj-y += vmport.o
>>  obj-y += pci-hotplug.o smbios.o wdt_ib700.o
>> -obj-y += debugcon.o multiboot.o
>> +obj-y += debugcon.o debugexit.o multiboot.o
>>  obj-y += pc_piix.o
>>  obj-y += pc_sysfw.o
>>  obj-y += lpc_ich9.o q35.o pc_q35.o
> 
> Otherwise looks good, CC'ing Hervé who did an earlier patch like this.

I see some differences with my patch available at 
http://patchwork.ozlabs.org/patch/169542/

Your patch breaks behaviour for access sizes > 1.
- Current code in QEMU accepts access sizes of 2, which means that if 
you write value 0x1234 to port 0x501, you'll exit with (0x1234 << 1) | 1.
- After your patch, as you're explicitly stating that 
valid.max_access_size = 1, a write of size 2 will be considered as invalid.

I never got any answer from Anthony if he is using access size of 1 or 2 
for his test suite... (see http://patchwork.ozlabs.org/patch/169543/)

I can easily revive my patch if Anthony wants something else that 
iobase=0x501 and access-size=1 by default.

Once again, putting Anthony in Cc to have an answer.

Regards,

Hervé