Calvin
Tue Feb 07 12:24:55 CST 2006
You can synchronously complete an IRP which had been marked pending as long
as you return pending from the dispatch routine.
--
Calvin Guan (Windows DDK MVP)
NetXtreme Longhorn Miniport Prime
Broadcom Corp. www.broadcom.com
"ning" <ning@newsgroup.nospam> wrote in message
news:e5Zhv9ALGHA.344@TK2MSFTNGP11.phx.gbl...
> But what if I happen to queue the irp and then complete the IRP
> synchronously? Wouldn't that violate the "IoMarkIrpPending() and
> STATUS_PENDING go together" rule?
>
> Ning
>
> "Skywing" <skywing_NO_SPAM_@valhallalegends.com> wrote in message
> news:%23cPu3Z5KGHA.3272@tk2msftngp13.phx.gbl...
>>I just sent a mail to the DDK doc feedback alias pointing this little
>>omission out; perhaps we shall see it clarified in the official
>>documentation.
>>
>> "Mark Roddy" <markr@hollistech.com> wrote in message
>> news:va4gu19k1afjf7t4dq3mo2veh46d13jt75@4ax.com...
>>> On Mon, 6 Feb 2006 21:51:56 -0500, "Skywing"
>>> <skywing_NO_SPAM_@valhallalegends.com> wrote:
>>>
>>>>I'm curious too, so I did a bit of looking around.
>>>>
>>>>A bit of disassembly shows that IoCsqInsertIrp does in fact do this. It
>>>>was
>>>>easy to determine this through the use of WinDbg in LKD mode against an
>>>>x64
>>>>Win2003SP1 machine, and dumpbin /disasm csq.lib with _NT_SYMBOL_PATH set
>>>>appropriately.
>>>>
>>>>I am of the opinion that this is a documentation omission and that csq
>>>>was
>>>>always supposed to do IoMarkIrpPending(Irp).
>>>>
>>>
>>> Nice work.
>>>
>>> I'd still not rely on this behavior until it gets documented.
>>>
>>>>From the 3790.1830 DDK for the x64 csq.lib:
>>>>
>>>>WdmlibIoCsqInsertIrp:
>>>> 0000000000000000: 45 33 C9 xor r9d,r9d
>>>> 0000000000000003: E9 00 00 00 00 jmp
>>>> WdmlibIoCsqInsertIrpEx
>>>>
>>>>; .
>>>>; .
>>>>; .
>>>>
>>>>WdmlibIoCsqInsertIrpEx:
>>>> 0000000000000000: 48 83 EC 38 sub rsp,38h
>>>> 0000000000000004: 48 89 5C 24 40 mov qword ptr
>>>> [rsp+40h],rbx
>>>> 0000000000000009: 48 89 6C 24 48 mov qword ptr
>>>> [rsp+48h],rbp
>>>> 000000000000000E: 48 89 74 24 50 mov qword ptr
>>>> [rsp+50h],rsi
>>>> 0000000000000013: 48 89 7C 24 58 mov qword ptr
>>>> [rsp+58h],rdi
>>>> 0000000000000018: 33 ED xor ebp,ebp
>>>> 000000000000001A: 4D 85 C0 test r8,r8
>>>> 000000000000001D: 4C 89 64 24 30 mov qword ptr
>>>> [rsp+30h],r12
>>>> 0000000000000022: 49 8B F0 mov rsi,r8
>>>> 0000000000000025: 48 8B D9 mov rbx,rcx
>>>> 0000000000000028: 4D 8B E1 mov r12,r9
>>>> 000000000000002B: 48 8B FA mov rdi,rdx ; rdi = Irp
>>>>
>>>>; .
>>>>; .
>>>>; .
>>>>
>>>> 00000000000000A9: 48 8B 87 B8 00 00 mov rax,qword ptr
>>>> [rdi+0B8h]
>>>>; Irp+0xb8 == Irp->Tail.Overlay.CurrentStackLocation (x64)
>>>> 00
>>>> 00000000000000B0: 80 48 03 01 or byte ptr [rax+3],1 ;
>>>>IO_STACK_LOCATION+3 == Control, 0x1 == SL_PENDING_RETURNED (x64)
>>>>
>>>>;
>>>>; And this is what IoMarkIrpPending() corresponds to in ntddk.h
>>>>;
>>>>; #define IoMarkIrpPending( Irp ) ( \
>>>>; IoGetCurrentIrpStackLocation( (Irp) )->Control |=
>>>>SL_PENDING_RETURNED )
>>>>;
>>>>;
>>>>; And here is IoGetCurrentIrpStackLocation()
>>>>;
>>>>; #define IoGetCurrentIrpStackLocation( Irp ) (
>>>>(Irp)->Tail.Overlay.CurrentStackLocation )
>>>>;
>>>>
>>>>; The prototype for IoCsqInsertIrpEx is as follows:
>>>>
>>>>; NTSTATUS
>>>>; IoCsqInsertIrpEx(
>>>>; IN PIO_CSQ Csq, // rcx
>>>>; IN PIRP Irp, // rdx
>>>>; IN PIO_CSQ_IRP_CONTEXT Context, // r8
>>>>; IN PVOID InsertContext // r9
>>>>; );
>>>>
>>>>
>>>>
>>>>
>>>>"Gerald S." <anonymous@discussions.microsoft.com> wrote in message
>>>>news:e1%23jc14KGHA.964@tk2msftngp13.phx.gbl...
>>>>> Thanks for the reply, however, I'd like to know the exact answer -
>>>>> whether
>>>>> the cancel sample contains a horrific bug or IoCsqInsertIrp calls
>>>>> IoMarkIrpPending? In other words, should I rush to update my old
>>>>> drivers
>>>>> or this can wait?
>>>>>
>>>>> --
>>>>>
>>>>> "Mark Roddy" <markr@hollistech.com> wrote in message
>>>>> news:3nufu1962s13pe0rdnmgmr13dcf7tipqut@4ax.com...
>>>>>> On Tue, 7 Feb 2006 12:30:48 +1300, "Gerald S."
>>>>>> <anonymous@discussions.microsoft.com> wrote:
>>>>>>
>>>>>>>Hi, I had an impression that I don't need to call IoMarkIrpPending
>>>>>>>before
>>>>>>>IoCsqInsertIrp, as the latter internally marks the inserted IRP
>>>>>>>pending
>>>>>>>anyway. Is this correct?
>>>>>>>
>>>>>>>Thanks
>>>>>>
>>>>>> That is actually an excellent question. The documentation doesn't say
>>>>>> anything. The cancel sample implies that the IRP is marked pending by
>>>>>> a call to IoCsqInsertIrp. So we have undocumented behavior with a
>>>>>> sample as the only guide. As far as I know IoMarkIrpPending can be
>>>>>> called repeatedly and do no harm. I would not rely on undocumented
>>>>>> behavior and just mark the IRP pending.
>>>>>>
>>>>>>
>>>>>> =====================
>>>>>> Mark Roddy DDK MVP
>>>>>> Windows Vista/2003/XP/2000 Consulting
>>>>>> Device and Filesystem Drivers
>>>>>> Hollis Technology Solutions 603-321-1032
>>>>>> www.hollistech.com
>>>>>
>>>>>
>>>>
>>>
>>>
>>> =====================
>>> Mark Roddy DDK MVP
>>> Windows Vista/2003/XP/2000 Consulting
>>> Device and Filesystem Drivers
>>> Hollis Technology Solutions 603-321-1032
>>> www.hollistech.com
>>
>>
>
>