subject:"\[edk2\] \[PATCH v2 0\/3\] Put AP into safe hlt\-loop code on S3 path"

Re: [edk2] [PATCH v2 0/3] Put AP into safe hlt-loop code on S3 path

2016-11-14 Thread Fan, Jeff

Laszlo,

That's great!  Thanks your details updating.

Jeff

-Original Message-
From: Laszlo Ersek [mailto:ler...@redhat.com] 
Sent: Tuesday, November 15, 2016 9:28 AM
To: Paolo Bonzini; Fan, Jeff
Cc: edk2-de...@ml01.01.org; Yao, Jiewen
Subject: Re: [edk2] [PATCH v2 0/3] Put AP into safe hlt-loop code on S3 path

On 11/15/16 00:56, Laszlo Ersek wrote:
> On 11/14/16 19:13, Paolo Bonzini wrote:
>>
>>
>> On 14/11/2016 19:07, Laszlo Ersek wrote:
>>> On 11/14/16 13:00, Paolo Bonzini wrote:
>>>>
>>>>
>>>> On 14/11/2016 12:27, Laszlo Ersek wrote:
>>>>> Well...
>>>>>
>>>>> http://lists.nongnu.org/archive/html/qemu-devel/2015-10/msg05658.h
>>>>> tml 
>>>>> http://lists.nongnu.org/archive/html/qemu-devel/2015-11/msg00125.h
>>>>> tml 
>>>>> http://lists.nongnu.org/archive/html/qemu-devel/2015-11/msg00563.h
>>>>> tml
>>>>>
>>>>> Are you suggesting that I resurrect this patch? That would be my 
>>>>> pleasure. Please say yes.
>>>>
>>>> It's hard to say no when someone has written the code already. :)
>>>
>>> Thanks. I refreshed both patches (OVMF and QEMU -- no code changes 
>>> just more precise commit messages). Unfortunately, quite a few 
>>> things seem broken, although these patches worked a year ago.
>>>
>>> My QEMU base commit is current master 83c83f9a5266. My host kernel 
>>> is 3.10.0-514.el7.x86_64.
>>>
>>> *** So, when I test these two patches, based on edk2 master (no 
>>> on-list patches), Ia32 target, my boot hangs (spins) with the log ending in:
>>>
>>>> SmmInstallProtocolInterface: [EdkiiSmmExitBootServicesProtocol] 0
>>>
>>> That is, MpInitChangeApLoopCallback() is entered, but it never finishes.
>>> "info cpus" prints:
>>>
>>> * CPU #0: pc=0x7f1f7763 thread_id=17395
>>>   CPU #1: pc=0x7f2ce01e (halted) thread_id=17396
>>>   CPU #2: pc=0x7f2ce01e (halted) thread_id=17397
>>>   CPU #3: pc=0xfff0 thread_id=17398
>>>
>>> and I've also seen a case where all the APs were stuck at the reset 
>>> vector (0xfff0), *not* halted, like VCPU#3 above. They 
>>> don't spin, they're just stuck. The spinning comes from CPU#0, 
>>> apparently in MpInitChangeApLoopCallback.
>>>
>>> *** I flipped the AP sync mode to traditional (considering the 
>>> relaxed mode shouldn't be required with the broadcast SMIs). This 
>>> time the log ends with:
>>>
>>>> SmmInstallProtocolInterface: [EdkiiSmmExitBootServicesProtocol] 0
>>>> MpInitChangeApLoopCallback() done!
>>>
>>> but then QEMU abort()s:
>>>
>>>> kvm_io_ioeventfd_add: error adding ioeventfd: File exists
>>>> 2016-11-14 17:00:41.405+: shutting down, reason=crashed
>>>
>>> I see some ioeventfd stuff in the recent QEMU history; do you think 
>>> it's related?
>>
>> Yes, just try 2.7 for now or disable vhost.
> 
> (1) I think I have some new results.
> 
> I used the gdbserver built into QEMU, and I (sort of) single-stepped 
> the
> MpInitChangeApLoopCallback() function in 
> "UefiCpuPkg/Library/MpInitLib/DxeMpLib.c", and whatever else it called.
> For this I used the Ia32 1x2x2 configuration. Also, the broadcast SMI 
> patches were applied to both QEMU and OVMF.
> 
> I set a breakpoint on RelocateApLoop(), so that when an AP would start 
> up, gdb would switch to it automatically (and it did in fact).
> 
> I liberally used "info cpus" from a separate terminal, and also "info 
> thread" from within gdb (which gives an incredibly cool insight into 
> the VCPU states!)
> 
> (2) Here's a few interesting results (strictly empirically):
> 
> * when I was stepping through the SendInitSipiSipiAllExcludingSelf()
>   function *real slow*, manually, suddenly things worked
> 
> * I noticed that, while stepping through the INIT-SIPI-SIPI sequence in
>   the above-mentioned BSP function, the APs switched from "halted" to
>   "running" after the *first* SIPI. Not the second SIPI, the first one.
> 
> Then I went to the KVM code, and looked at arch/x86/kvm/lapic.c. My 
> explanation is terribly inexact, but:
> 
> * __apic_accept_irq() translates the LAPIC writes (APIC_DM_INIT and
>   APIC_DM_STARTUP) into pending events (KVM_APIC_INIT and 
> KVM_APIC_SIPI)
> 
> * this function also calls kvm_vcpu_kick()
> 
> * the kvm_apic_accept_e

Re: [edk2] [PATCH v2 0/3] Put AP into safe hlt-loop code on S3 path

2016-11-14 Thread Laszlo Ersek

On 11/15/16 02:19, Fan, Jeff wrote:
> Laszlo,
> 
> Have you tried to decrease  PcdCpuInitIpiDelayInMicroSeconds?
> 
> For PCD PcdCpuInitIpiDelayInMicroSeconds before two SIPIs, we introduced this 
> PCD is just to do customization.
>   MicroSecondDelay (PcdGet32(PcdCpuInitIpiDelayInMicroSeconds));
> 
> Per our experience,  we could decrease this PCD value to 10us (microsecond) 
> on some platforms/processors.  You may try it.

I didn't try to decrease it, but experimenting with it should no longer
be necessary. (See my other email.) Moving to the broadcast SMI pattern
and the traditional AP sync method *really* fixes things (together with
this series of yours).

Thanks!
Laszlo


> -Original Message-
> From: Laszlo Ersek [mailto:ler...@redhat.com] 
> Sent: Tuesday, November 15, 2016 7:56 AM
> To: Paolo Bonzini; Fan, Jeff
> Cc: edk2-de...@ml01.01.org; Yao, Jiewen
> Subject: Re: [edk2] [PATCH v2 0/3] Put AP into safe hlt-loop code on S3 path
> 
> On 11/14/16 19:13, Paolo Bonzini wrote:
>>
>>
>> On 14/11/2016 19:07, Laszlo Ersek wrote:
>>> On 11/14/16 13:00, Paolo Bonzini wrote:
>>>>
>>>>
>>>> On 14/11/2016 12:27, Laszlo Ersek wrote:
>>>>> Well...
>>>>>
>>>>> http://lists.nongnu.org/archive/html/qemu-devel/2015-10/msg05658.ht
>>>>> ml 
>>>>> http://lists.nongnu.org/archive/html/qemu-devel/2015-11/msg00125.ht
>>>>> ml 
>>>>> http://lists.nongnu.org/archive/html/qemu-devel/2015-11/msg00563.ht
>>>>> ml
>>>>>
>>>>> Are you suggesting that I resurrect this patch? That would be my 
>>>>> pleasure. Please say yes.
>>>>
>>>> It's hard to say no when someone has written the code already. :)
>>>
>>> Thanks. I refreshed both patches (OVMF and QEMU -- no code changes 
>>> just more precise commit messages). Unfortunately, quite a few things 
>>> seem broken, although these patches worked a year ago.
>>>
>>> My QEMU base commit is current master 83c83f9a5266. My host kernel is 
>>> 3.10.0-514.el7.x86_64.
>>>
>>> *** So, when I test these two patches, based on edk2 master (no 
>>> on-list patches), Ia32 target, my boot hangs (spins) with the log ending in:
>>>
>>>> SmmInstallProtocolInterface: [EdkiiSmmExitBootServicesProtocol] 0
>>>
>>> That is, MpInitChangeApLoopCallback() is entered, but it never finishes.
>>> "info cpus" prints:
>>>
>>> * CPU #0: pc=0x7f1f7763 thread_id=17395
>>>   CPU #1: pc=0x7f2ce01e (halted) thread_id=17396
>>>   CPU #2: pc=0x7f2ce01e (halted) thread_id=17397
>>>   CPU #3: pc=0xfff0 thread_id=17398
>>>
>>> and I've also seen a case where all the APs were stuck at the reset 
>>> vector (0xfff0), *not* halted, like VCPU#3 above. They 
>>> don't spin, they're just stuck. The spinning comes from CPU#0, 
>>> apparently in MpInitChangeApLoopCallback.
>>>
>>> *** I flipped the AP sync mode to traditional (considering the 
>>> relaxed mode shouldn't be required with the broadcast SMIs). This 
>>> time the log ends with:
>>>
>>>> SmmInstallProtocolInterface: [EdkiiSmmExitBootServicesProtocol] 0
>>>> MpInitChangeApLoopCallback() done!
>>>
>>> but then QEMU abort()s:
>>>
>>>> kvm_io_ioeventfd_add: error adding ioeventfd: File exists
>>>> 2016-11-14 17:00:41.405+: shutting down, reason=crashed
>>>
>>> I see some ioeventfd stuff in the recent QEMU history; do you think 
>>> it's related?
>>
>> Yes, just try 2.7 for now or disable vhost.
> 
> (1) I think I have some new results.
> 
> I used the gdbserver built into QEMU, and I (sort of) single-stepped the
> MpInitChangeApLoopCallback() function in 
> "UefiCpuPkg/Library/MpInitLib/DxeMpLib.c", and whatever else it called.
> For this I used the Ia32 1x2x2 configuration. Also, the broadcast SMI patches 
> were applied to both QEMU and OVMF.
> 
> I set a breakpoint on RelocateApLoop(), so that when an AP would start up, 
> gdb would switch to it automatically (and it did in fact).
> 
> I liberally used "info cpus" from a separate terminal, and also "info thread" 
> from within gdb (which gives an incredibly cool insight into the VCPU states!)
> 
> (2) Here's a few interesting results (strictly empirically):
> 
> * when I was stepping through the SendInitSipiSipiAllExcludingSelf()
>   function *real slow*, manually, suddenly

Re: [edk2] [PATCH v2 0/3] Put AP into safe hlt-loop code on S3 path

2016-11-14 Thread Laszlo Ersek

On 11/15/16 00:56, Laszlo Ersek wrote:
> On 11/14/16 19:13, Paolo Bonzini wrote:
>>
>>
>> On 14/11/2016 19:07, Laszlo Ersek wrote:
>>> On 11/14/16 13:00, Paolo Bonzini wrote:


 On 14/11/2016 12:27, Laszlo Ersek wrote:
> Well...
>
> http://lists.nongnu.org/archive/html/qemu-devel/2015-10/msg05658.html
> http://lists.nongnu.org/archive/html/qemu-devel/2015-11/msg00125.html
> http://lists.nongnu.org/archive/html/qemu-devel/2015-11/msg00563.html
>
> Are you suggesting that I resurrect this patch? That would be my
> pleasure. Please say yes.

 It's hard to say no when someone has written the code already. :)
>>>
>>> Thanks. I refreshed both patches (OVMF and QEMU -- no code changes just
>>> more precise commit messages). Unfortunately, quite a few things seem
>>> broken, although these patches worked a year ago.
>>>
>>> My QEMU base commit is current master 83c83f9a5266. My host kernel is
>>> 3.10.0-514.el7.x86_64.
>>>
>>> *** So, when I test these two patches, based on edk2 master (no on-list
>>> patches), Ia32 target, my boot hangs (spins) with the log ending in:
>>>
 SmmInstallProtocolInterface: [EdkiiSmmExitBootServicesProtocol] 0
>>>
>>> That is, MpInitChangeApLoopCallback() is entered, but it never finishes.
>>> "info cpus" prints:
>>>
>>> * CPU #0: pc=0x7f1f7763 thread_id=17395
>>>   CPU #1: pc=0x7f2ce01e (halted) thread_id=17396
>>>   CPU #2: pc=0x7f2ce01e (halted) thread_id=17397
>>>   CPU #3: pc=0xfff0 thread_id=17398
>>>
>>> and I've also seen a case where all the APs were stuck at the reset
>>> vector (0xfff0), *not* halted, like VCPU#3 above. They don't
>>> spin, they're just stuck. The spinning comes from CPU#0, apparently in
>>> MpInitChangeApLoopCallback.
>>>
>>> *** I flipped the AP sync mode to traditional (considering the relaxed
>>> mode shouldn't be required with the broadcast SMIs). This time the log
>>> ends with:
>>>
 SmmInstallProtocolInterface: [EdkiiSmmExitBootServicesProtocol] 0
 MpInitChangeApLoopCallback() done!
>>>
>>> but then QEMU abort()s:
>>>
 kvm_io_ioeventfd_add: error adding ioeventfd: File exists
 2016-11-14 17:00:41.405+: shutting down, reason=crashed
>>>
>>> I see some ioeventfd stuff in the recent QEMU history; do you think it's
>>> related?
>>
>> Yes, just try 2.7 for now or disable vhost.
> 
> (1) I think I have some new results.
> 
> I used the gdbserver built into QEMU, and I (sort of) single-stepped the
> MpInitChangeApLoopCallback() function in
> "UefiCpuPkg/Library/MpInitLib/DxeMpLib.c", and whatever else it called.
> For this I used the Ia32 1x2x2 configuration. Also, the broadcast SMI
> patches were applied to both QEMU and OVMF.
> 
> I set a breakpoint on RelocateApLoop(), so that when an AP would start
> up, gdb would switch to it automatically (and it did in fact).
> 
> I liberally used "info cpus" from a separate terminal, and also "info
> thread" from within gdb (which gives an incredibly cool insight into the
> VCPU states!)
> 
> (2) Here's a few interesting results (strictly empirically):
> 
> * when I was stepping through the SendInitSipiSipiAllExcludingSelf()
>   function *real slow*, manually, suddenly things worked
> 
> * I noticed that, while stepping through the INIT-SIPI-SIPI sequence in
>   the above-mentioned BSP function, the APs switched from "halted" to
>   "running" after the *first* SIPI. Not the second SIPI, the first one.
> 
> Then I went to the KVM code, and looked at arch/x86/kvm/lapic.c. My
> explanation is terribly inexact, but:
> 
> * __apic_accept_irq() translates the LAPIC writes (APIC_DM_INIT and
>   APIC_DM_STARTUP) into pending events (KVM_APIC_INIT and KVM_APIC_SIPI)
> 
> * this function also calls kvm_vcpu_kick()
> 
> * the kvm_apic_accept_events() function processes the pending events.
>   The KVM_APIC_INIT event, if pending, causes kvm_vcpu_reset() to be
>   called, and the AP to be moved to KVM_MP_STATE_INIT_RECEIVED state.
> 
>   And, in that state, a pending KVM_APIC_SIPI event moves the AP to
>   KVM_MP_STATE_RUNNABLE. This confirms my surprising empirical result
>   that only one SIPI is awaited after the INIT before launching the APs.
> 
> * the kvm_vcpu_kick() function in "virt/kvm/kvm_main.c" boils down to a
>   smp_send_reschedule() call, if the kicker and the "kickee" are
>   different processors (for example, when the BSP kicks the AP)
> 
>   This seemed important because it suggested that host kernel
>   scheduling jitter could delay the delivery (reception) of the INIT on
>   the AP after the BSP sent it. If the BSP sent the first (and second)
>   SIPI really fast after the INIT, then those SIPIs could be missed,
>   and the AP would remain stuck in KVM_MP_STATE_INIT_RECEIVED state.
> 
>   And this state (remember kvm_vcpu_reset() from above!) is consistent
>   with the RIP being stuck at the reset vector address, with the AP
>   neither running nor being halted.
> 
> (3) So, I looked at

Re: [edk2] [PATCH v2 0/3] Put AP into safe hlt-loop code on S3 path

2016-11-14 Thread Yao, Jiewen

HI jeff
The patch looks good and I have one minor suggestion on below comment:

Allocate safe memory in ACPI NVS for AP to execute hlt loop on S3 path
==>
Allocate safe memory in ACPI NVS for AP to execute hlt loop in protected mode 
on S3 path

I recommend we add "in protected mode" statement for the reason to allocate 
below 4GiB memory.

With that comment update, reviewed-by: jiewen@intel.com

Thank you
Yao Jiewen


> -Original Message-
> From: Fan, Jeff
> Sent: Tuesday, November 15, 2016 8:50 AM
> To: Yao, Jiewen <jiewen@intel.com>
> Subject: FW: [edk2] [PATCH v2 0/3] Put AP into safe hlt-loop code on S3 path
> 
> 
> 
> -Original Message-
> From: edk2-devel [mailto:edk2-devel-boun...@lists.01.org] On Behalf Of
> Jeff Fan
> Sent: Friday, November 11, 2016 1:46 PM
> To: edk2-devel@lists.01.org
> Subject: [edk2] [PATCH v2 0/3] Put AP into safe hlt-loop code on S3 path
> 
> On S3 path, we will wake up APs to restore CPU context in
> PiSmmCpuDxeSmm
> driver. In case, one NMI or SMI happens, APs may exit from hlt state and
> execute the instruction after HLT instruction.
> 
> But APs are not running on safe code, it leads OVMF S3 boot unstable.
> 
> https://bugzilla.tianocore.org/show_bug.cgi?id=216
> 
> I tested real platform with 64bit DXE.
> 
> v2:
>   1. Make stack alignment per Laszlo's comment.
>   2. Trim whitespace at end of end per Laszlo's comment.
>   3. Update year mark in file header.
>   4. Enhancement on InterlockedDecrement() per Paolo's comment.
> 
> Jeff Fan (3):
>   UefiCpuPkg/PiSmmCpuDxeSmm: Put AP into safe hlt-loop code on S3 path
>   UefiCpuPkg/PiSmmCpuDxeSmm: Place AP to 32bit protected mode on S3
> path
>   UefiCpuPkg/PiSmmCpuDxeSmm: Decrease mNumberToFinish in AP safe
> code
> 
>  UefiCpuPkg/PiSmmCpuDxeSmm/CpuS3.c | 33
> +-
>  UefiCpuPkg/PiSmmCpuDxeSmm/Ia32/SmmFuncsArch.c | 29 +++-
>  UefiCpuPkg/PiSmmCpuDxeSmm/PiSmmCpuDxeSmm.h| 15 +++
>  UefiCpuPkg/PiSmmCpuDxeSmm/X64/SmmFuncsArch.c  | 63
> ++-
>  4 files changed, 136 insertions(+), 4 deletions(-)
> 
> --
> 2.9.3.windows.2
> 
> ___
> edk2-devel mailing list
> edk2-devel@lists.01.org
> https://lists.01.org/mailman/listinfo/edk2-devel
___
edk2-devel mailing list
edk2-devel@lists.01.org
https://lists.01.org/mailman/listinfo/edk2-devel

Re: [edk2] [PATCH v2 0/3] Put AP into safe hlt-loop code on S3 path

2016-11-14 Thread Fan, Jeff

Laszlo,

Have you tried to decrease  PcdCpuInitIpiDelayInMicroSeconds?

For PCD PcdCpuInitIpiDelayInMicroSeconds before two SIPIs, we introduced this 
PCD is just to do customization.
  MicroSecondDelay (PcdGet32(PcdCpuInitIpiDelayInMicroSeconds));

Per our experience,  we could decrease this PCD value to 10us (microsecond) on 
some platforms/processors.  You may try it.

Thanks!
Jeff

-Original Message-
From: Laszlo Ersek [mailto:ler...@redhat.com] 
Sent: Tuesday, November 15, 2016 7:56 AM
To: Paolo Bonzini; Fan, Jeff
Cc: edk2-de...@ml01.01.org; Yao, Jiewen
Subject: Re: [edk2] [PATCH v2 0/3] Put AP into safe hlt-loop code on S3 path

On 11/14/16 19:13, Paolo Bonzini wrote:
> 
> 
> On 14/11/2016 19:07, Laszlo Ersek wrote:
>> On 11/14/16 13:00, Paolo Bonzini wrote:
>>>
>>>
>>> On 14/11/2016 12:27, Laszlo Ersek wrote:
>>>> Well...
>>>>
>>>> http://lists.nongnu.org/archive/html/qemu-devel/2015-10/msg05658.ht
>>>> ml 
>>>> http://lists.nongnu.org/archive/html/qemu-devel/2015-11/msg00125.ht
>>>> ml 
>>>> http://lists.nongnu.org/archive/html/qemu-devel/2015-11/msg00563.ht
>>>> ml
>>>>
>>>> Are you suggesting that I resurrect this patch? That would be my 
>>>> pleasure. Please say yes.
>>>
>>> It's hard to say no when someone has written the code already. :)
>>
>> Thanks. I refreshed both patches (OVMF and QEMU -- no code changes 
>> just more precise commit messages). Unfortunately, quite a few things 
>> seem broken, although these patches worked a year ago.
>>
>> My QEMU base commit is current master 83c83f9a5266. My host kernel is 
>> 3.10.0-514.el7.x86_64.
>>
>> *** So, when I test these two patches, based on edk2 master (no 
>> on-list patches), Ia32 target, my boot hangs (spins) with the log ending in:
>>
>>> SmmInstallProtocolInterface: [EdkiiSmmExitBootServicesProtocol] 0
>>
>> That is, MpInitChangeApLoopCallback() is entered, but it never finishes.
>> "info cpus" prints:
>>
>> * CPU #0: pc=0x7f1f7763 thread_id=17395
>>   CPU #1: pc=0x7f2ce01e (halted) thread_id=17396
>>   CPU #2: pc=0x7f2ce01e (halted) thread_id=17397
>>   CPU #3: pc=0xfff0 thread_id=17398
>>
>> and I've also seen a case where all the APs were stuck at the reset 
>> vector (0xfff0), *not* halted, like VCPU#3 above. They 
>> don't spin, they're just stuck. The spinning comes from CPU#0, 
>> apparently in MpInitChangeApLoopCallback.
>>
>> *** I flipped the AP sync mode to traditional (considering the 
>> relaxed mode shouldn't be required with the broadcast SMIs). This 
>> time the log ends with:
>>
>>> SmmInstallProtocolInterface: [EdkiiSmmExitBootServicesProtocol] 0
>>> MpInitChangeApLoopCallback() done!
>>
>> but then QEMU abort()s:
>>
>>> kvm_io_ioeventfd_add: error adding ioeventfd: File exists
>>> 2016-11-14 17:00:41.405+: shutting down, reason=crashed
>>
>> I see some ioeventfd stuff in the recent QEMU history; do you think 
>> it's related?
> 
> Yes, just try 2.7 for now or disable vhost.

(1) I think I have some new results.

I used the gdbserver built into QEMU, and I (sort of) single-stepped the
MpInitChangeApLoopCallback() function in 
"UefiCpuPkg/Library/MpInitLib/DxeMpLib.c", and whatever else it called.
For this I used the Ia32 1x2x2 configuration. Also, the broadcast SMI patches 
were applied to both QEMU and OVMF.

I set a breakpoint on RelocateApLoop(), so that when an AP would start up, gdb 
would switch to it automatically (and it did in fact).

I liberally used "info cpus" from a separate terminal, and also "info thread" 
from within gdb (which gives an incredibly cool insight into the VCPU states!)

(2) Here's a few interesting results (strictly empirically):

* when I was stepping through the SendInitSipiSipiAllExcludingSelf()
  function *real slow*, manually, suddenly things worked

* I noticed that, while stepping through the INIT-SIPI-SIPI sequence in
  the above-mentioned BSP function, the APs switched from "halted" to
  "running" after the *first* SIPI. Not the second SIPI, the first one.

Then I went to the KVM code, and looked at arch/x86/kvm/lapic.c. My explanation 
is terribly inexact, but:

* __apic_accept_irq() translates the LAPIC writes (APIC_DM_INIT and
  APIC_DM_STARTUP) into pending events (KVM_APIC_INIT and KVM_APIC_SIPI)

* this function also calls kvm_vcpu_kick()

* the kvm_apic_accept_events() function processes the pending events.
  The KVM_APIC_INIT event, if pending, causes kvm_vcpu_reset() to be
  called, and the AP to be moved

Re: [edk2] [PATCH v2 0/3] Put AP into safe hlt-loop code on S3 path

2016-11-14 Thread Fan, Jeff

Sure. Please add it. :-)

-Original Message-
From: edk2-devel [mailto:edk2-devel-boun...@lists.01.org] On Behalf Of Laszlo 
Ersek
Sent: Tuesday, November 15, 2016 9:03 AM
To: Fan, Jeff; Paolo Bonzini
Cc: edk2-de...@ml01.01.org; Yao, Jiewen
Subject: Re: [edk2] [PATCH v2 0/3] Put AP into safe hlt-loop code on S3 path

On 11/15/16 01:47, Fan, Jeff wrote:
> Laszlo,
> 
> I file https://bugzilla.tianocore.org/show_bug.cgi?id=230 to handle AP lost 
> issue.
> 
> I will push the v2 serial of patches, Paolo has already added r-b signature 
> on the first 2 patches in v1 serials.

Please add my

Tested-by: Laszlo Ersek <ler...@redhat.com>

to the first and the third patch.

If you can wait a bit, I might be able to extend my T-b to the second patch as 
well.

(More details to follow soon.)

Thanks!
Laszlo

> -Original Message-
> From: Laszlo Ersek [mailto:ler...@redhat.com]
> Sent: Tuesday, November 15, 2016 7:56 AM
> To: Paolo Bonzini; Fan, Jeff
> Cc: edk2-de...@ml01.01.org; Yao, Jiewen
> Subject: Re: [edk2] [PATCH v2 0/3] Put AP into safe hlt-loop code on 
> S3 path
> 
> On 11/14/16 19:13, Paolo Bonzini wrote:
>>
>>
>> On 14/11/2016 19:07, Laszlo Ersek wrote:
>>> On 11/14/16 13:00, Paolo Bonzini wrote:
>>>>
>>>>
>>>> On 14/11/2016 12:27, Laszlo Ersek wrote:
>>>>> Well...
>>>>>
>>>>> http://lists.nongnu.org/archive/html/qemu-devel/2015-10/msg05658.h
>>>>> t
>>>>> ml
>>>>> http://lists.nongnu.org/archive/html/qemu-devel/2015-11/msg00125.h
>>>>> t
>>>>> ml
>>>>> http://lists.nongnu.org/archive/html/qemu-devel/2015-11/msg00563.h
>>>>> t
>>>>> ml
>>>>>
>>>>> Are you suggesting that I resurrect this patch? That would be my 
>>>>> pleasure. Please say yes.
>>>>
>>>> It's hard to say no when someone has written the code already. :)
>>>
>>> Thanks. I refreshed both patches (OVMF and QEMU -- no code changes 
>>> just more precise commit messages). Unfortunately, quite a few 
>>> things seem broken, although these patches worked a year ago.
>>>
>>> My QEMU base commit is current master 83c83f9a5266. My host kernel 
>>> is 3.10.0-514.el7.x86_64.
>>>
>>> *** So, when I test these two patches, based on edk2 master (no 
>>> on-list patches), Ia32 target, my boot hangs (spins) with the log ending in:
>>>
>>>> SmmInstallProtocolInterface: [EdkiiSmmExitBootServicesProtocol] 0
>>>
>>> That is, MpInitChangeApLoopCallback() is entered, but it never finishes.
>>> "info cpus" prints:
>>>
>>> * CPU #0: pc=0x7f1f7763 thread_id=17395
>>>   CPU #1: pc=0x7f2ce01e (halted) thread_id=17396
>>>   CPU #2: pc=0x7f2ce01e (halted) thread_id=17397
>>>   CPU #3: pc=0xfff0 thread_id=17398
>>>
>>> and I've also seen a case where all the APs were stuck at the reset 
>>> vector (0xfff0), *not* halted, like VCPU#3 above. They 
>>> don't spin, they're just stuck. The spinning comes from CPU#0, 
>>> apparently in MpInitChangeApLoopCallback.
>>>
>>> *** I flipped the AP sync mode to traditional (considering the 
>>> relaxed mode shouldn't be required with the broadcast SMIs). This 
>>> time the log ends with:
>>>
>>>> SmmInstallProtocolInterface: [EdkiiSmmExitBootServicesProtocol] 0
>>>> MpInitChangeApLoopCallback() done!
>>>
>>> but then QEMU abort()s:
>>>
>>>> kvm_io_ioeventfd_add: error adding ioeventfd: File exists
>>>> 2016-11-14 17:00:41.405+: shutting down, reason=crashed
>>>
>>> I see some ioeventfd stuff in the recent QEMU history; do you think 
>>> it's related?
>>
>> Yes, just try 2.7 for now or disable vhost.
> 
> (1) I think I have some new results.
> 
> I used the gdbserver built into QEMU, and I (sort of) single-stepped 
> the
> MpInitChangeApLoopCallback() function in 
> "UefiCpuPkg/Library/MpInitLib/DxeMpLib.c", and whatever else it called.
> For this I used the Ia32 1x2x2 configuration. Also, the broadcast SMI patches 
> were applied to both QEMU and OVMF.
> 
> I set a breakpoint on RelocateApLoop(), so that when an AP would start up, 
> gdb would switch to it automatically (and it did in fact).
> 
> I liberally used "info cpus" from a separate terminal, and also "info 
> thread" from within gdb (which gives an incredibly cool insight into 
>

Re: [edk2] [PATCH v2 0/3] Put AP into safe hlt-loop code on S3 path

2016-11-14 Thread Laszlo Ersek

On 11/15/16 01:47, Fan, Jeff wrote:
> Laszlo,
> 
> I file https://bugzilla.tianocore.org/show_bug.cgi?id=230 to handle AP lost 
> issue.
> 
> I will push the v2 serial of patches, Paolo has already added r-b signature 
> on the first 2 patches in v1 serials.

Please add my

Tested-by: Laszlo Ersek <ler...@redhat.com>

to the first and the third patch.

If you can wait a bit, I might be able to extend my T-b to the second
patch as well.

(More details to follow soon.)

Thanks!
Laszlo

> -Original Message-
> From: Laszlo Ersek [mailto:ler...@redhat.com] 
> Sent: Tuesday, November 15, 2016 7:56 AM
> To: Paolo Bonzini; Fan, Jeff
> Cc: edk2-de...@ml01.01.org; Yao, Jiewen
> Subject: Re: [edk2] [PATCH v2 0/3] Put AP into safe hlt-loop code on S3 path
> 
> On 11/14/16 19:13, Paolo Bonzini wrote:
>>
>>
>> On 14/11/2016 19:07, Laszlo Ersek wrote:
>>> On 11/14/16 13:00, Paolo Bonzini wrote:
>>>>
>>>>
>>>> On 14/11/2016 12:27, Laszlo Ersek wrote:
>>>>> Well...
>>>>>
>>>>> http://lists.nongnu.org/archive/html/qemu-devel/2015-10/msg05658.ht
>>>>> ml 
>>>>> http://lists.nongnu.org/archive/html/qemu-devel/2015-11/msg00125.ht
>>>>> ml 
>>>>> http://lists.nongnu.org/archive/html/qemu-devel/2015-11/msg00563.ht
>>>>> ml
>>>>>
>>>>> Are you suggesting that I resurrect this patch? That would be my 
>>>>> pleasure. Please say yes.
>>>>
>>>> It's hard to say no when someone has written the code already. :)
>>>
>>> Thanks. I refreshed both patches (OVMF and QEMU -- no code changes 
>>> just more precise commit messages). Unfortunately, quite a few things 
>>> seem broken, although these patches worked a year ago.
>>>
>>> My QEMU base commit is current master 83c83f9a5266. My host kernel is 
>>> 3.10.0-514.el7.x86_64.
>>>
>>> *** So, when I test these two patches, based on edk2 master (no 
>>> on-list patches), Ia32 target, my boot hangs (spins) with the log ending in:
>>>
>>>> SmmInstallProtocolInterface: [EdkiiSmmExitBootServicesProtocol] 0
>>>
>>> That is, MpInitChangeApLoopCallback() is entered, but it never finishes.
>>> "info cpus" prints:
>>>
>>> * CPU #0: pc=0x7f1f7763 thread_id=17395
>>>   CPU #1: pc=0x7f2ce01e (halted) thread_id=17396
>>>   CPU #2: pc=0x7f2ce01e (halted) thread_id=17397
>>>   CPU #3: pc=0xfff0 thread_id=17398
>>>
>>> and I've also seen a case where all the APs were stuck at the reset 
>>> vector (0xfff0), *not* halted, like VCPU#3 above. They 
>>> don't spin, they're just stuck. The spinning comes from CPU#0, 
>>> apparently in MpInitChangeApLoopCallback.
>>>
>>> *** I flipped the AP sync mode to traditional (considering the 
>>> relaxed mode shouldn't be required with the broadcast SMIs). This 
>>> time the log ends with:
>>>
>>>> SmmInstallProtocolInterface: [EdkiiSmmExitBootServicesProtocol] 0
>>>> MpInitChangeApLoopCallback() done!
>>>
>>> but then QEMU abort()s:
>>>
>>>> kvm_io_ioeventfd_add: error adding ioeventfd: File exists
>>>> 2016-11-14 17:00:41.405+: shutting down, reason=crashed
>>>
>>> I see some ioeventfd stuff in the recent QEMU history; do you think 
>>> it's related?
>>
>> Yes, just try 2.7 for now or disable vhost.
> 
> (1) I think I have some new results.
> 
> I used the gdbserver built into QEMU, and I (sort of) single-stepped the
> MpInitChangeApLoopCallback() function in 
> "UefiCpuPkg/Library/MpInitLib/DxeMpLib.c", and whatever else it called.
> For this I used the Ia32 1x2x2 configuration. Also, the broadcast SMI patches 
> were applied to both QEMU and OVMF.
> 
> I set a breakpoint on RelocateApLoop(), so that when an AP would start up, 
> gdb would switch to it automatically (and it did in fact).
> 
> I liberally used "info cpus" from a separate terminal, and also "info thread" 
> from within gdb (which gives an incredibly cool insight into the VCPU states!)
> 
> (2) Here's a few interesting results (strictly empirically):
> 
> * when I was stepping through the SendInitSipiSipiAllExcludingSelf()
>   function *real slow*, manually, suddenly things worked
> 
> * I noticed that, while stepping through the INIT-SIPI-SIPI sequence in
>   the above-mentioned BSP function, the APs switched from "halted" to
>   "r

Re: [edk2] [PATCH v2 0/3] Put AP into safe hlt-loop code on S3 path

2016-11-14 Thread Fan, Jeff

Laszlo,

I file https://bugzilla.tianocore.org/show_bug.cgi?id=230 to handle AP lost 
issue.

I will push the v2 serial of patches, Paolo has already added r-b signature on 
the first 2 patches in v1 serials.

Jeff

-Original Message-
From: Laszlo Ersek [mailto:ler...@redhat.com] 
Sent: Tuesday, November 15, 2016 7:56 AM
To: Paolo Bonzini; Fan, Jeff
Cc: edk2-de...@ml01.01.org; Yao, Jiewen
Subject: Re: [edk2] [PATCH v2 0/3] Put AP into safe hlt-loop code on S3 path

On 11/14/16 19:13, Paolo Bonzini wrote:
> 
> 
> On 14/11/2016 19:07, Laszlo Ersek wrote:
>> On 11/14/16 13:00, Paolo Bonzini wrote:
>>>
>>>
>>> On 14/11/2016 12:27, Laszlo Ersek wrote:
>>>> Well...
>>>>
>>>> http://lists.nongnu.org/archive/html/qemu-devel/2015-10/msg05658.ht
>>>> ml 
>>>> http://lists.nongnu.org/archive/html/qemu-devel/2015-11/msg00125.ht
>>>> ml 
>>>> http://lists.nongnu.org/archive/html/qemu-devel/2015-11/msg00563.ht
>>>> ml
>>>>
>>>> Are you suggesting that I resurrect this patch? That would be my 
>>>> pleasure. Please say yes.
>>>
>>> It's hard to say no when someone has written the code already. :)
>>
>> Thanks. I refreshed both patches (OVMF and QEMU -- no code changes 
>> just more precise commit messages). Unfortunately, quite a few things 
>> seem broken, although these patches worked a year ago.
>>
>> My QEMU base commit is current master 83c83f9a5266. My host kernel is 
>> 3.10.0-514.el7.x86_64.
>>
>> *** So, when I test these two patches, based on edk2 master (no 
>> on-list patches), Ia32 target, my boot hangs (spins) with the log ending in:
>>
>>> SmmInstallProtocolInterface: [EdkiiSmmExitBootServicesProtocol] 0
>>
>> That is, MpInitChangeApLoopCallback() is entered, but it never finishes.
>> "info cpus" prints:
>>
>> * CPU #0: pc=0x7f1f7763 thread_id=17395
>>   CPU #1: pc=0x7f2ce01e (halted) thread_id=17396
>>   CPU #2: pc=0x7f2ce01e (halted) thread_id=17397
>>   CPU #3: pc=0xfff0 thread_id=17398
>>
>> and I've also seen a case where all the APs were stuck at the reset 
>> vector (0xfff0), *not* halted, like VCPU#3 above. They 
>> don't spin, they're just stuck. The spinning comes from CPU#0, 
>> apparently in MpInitChangeApLoopCallback.
>>
>> *** I flipped the AP sync mode to traditional (considering the 
>> relaxed mode shouldn't be required with the broadcast SMIs). This 
>> time the log ends with:
>>
>>> SmmInstallProtocolInterface: [EdkiiSmmExitBootServicesProtocol] 0
>>> MpInitChangeApLoopCallback() done!
>>
>> but then QEMU abort()s:
>>
>>> kvm_io_ioeventfd_add: error adding ioeventfd: File exists
>>> 2016-11-14 17:00:41.405+: shutting down, reason=crashed
>>
>> I see some ioeventfd stuff in the recent QEMU history; do you think 
>> it's related?
> 
> Yes, just try 2.7 for now or disable vhost.

(1) I think I have some new results.

I used the gdbserver built into QEMU, and I (sort of) single-stepped the
MpInitChangeApLoopCallback() function in 
"UefiCpuPkg/Library/MpInitLib/DxeMpLib.c", and whatever else it called.
For this I used the Ia32 1x2x2 configuration. Also, the broadcast SMI patches 
were applied to both QEMU and OVMF.

I set a breakpoint on RelocateApLoop(), so that when an AP would start up, gdb 
would switch to it automatically (and it did in fact).

I liberally used "info cpus" from a separate terminal, and also "info thread" 
from within gdb (which gives an incredibly cool insight into the VCPU states!)

(2) Here's a few interesting results (strictly empirically):

* when I was stepping through the SendInitSipiSipiAllExcludingSelf()
  function *real slow*, manually, suddenly things worked

* I noticed that, while stepping through the INIT-SIPI-SIPI sequence in
  the above-mentioned BSP function, the APs switched from "halted" to
  "running" after the *first* SIPI. Not the second SIPI, the first one.

Then I went to the KVM code, and looked at arch/x86/kvm/lapic.c. My explanation 
is terribly inexact, but:

* __apic_accept_irq() translates the LAPIC writes (APIC_DM_INIT and
  APIC_DM_STARTUP) into pending events (KVM_APIC_INIT and KVM_APIC_SIPI)

* this function also calls kvm_vcpu_kick()

* the kvm_apic_accept_events() function processes the pending events.
  The KVM_APIC_INIT event, if pending, causes kvm_vcpu_reset() to be
  called, and the AP to be moved to KVM_MP_STATE_INIT_RECEIVED state.

  And, in that state, a pending KVM_APIC_SIPI event moves the AP to
  KVM_MP_STATE_RUNNABLE. This confirms my surprising empirical

Re: [edk2] [PATCH v2 0/3] Put AP into safe hlt-loop code on S3 path

2016-11-14 Thread Laszlo Ersek

On 11/14/16 19:13, Paolo Bonzini wrote:
> 
> 
> On 14/11/2016 19:07, Laszlo Ersek wrote:
>> On 11/14/16 13:00, Paolo Bonzini wrote:
>>>
>>>
>>> On 14/11/2016 12:27, Laszlo Ersek wrote:
 Well...

 http://lists.nongnu.org/archive/html/qemu-devel/2015-10/msg05658.html
 http://lists.nongnu.org/archive/html/qemu-devel/2015-11/msg00125.html
 http://lists.nongnu.org/archive/html/qemu-devel/2015-11/msg00563.html

 Are you suggesting that I resurrect this patch? That would be my
 pleasure. Please say yes.
>>>
>>> It's hard to say no when someone has written the code already. :)
>>
>> Thanks. I refreshed both patches (OVMF and QEMU -- no code changes just
>> more precise commit messages). Unfortunately, quite a few things seem
>> broken, although these patches worked a year ago.
>>
>> My QEMU base commit is current master 83c83f9a5266. My host kernel is
>> 3.10.0-514.el7.x86_64.
>>
>> *** So, when I test these two patches, based on edk2 master (no on-list
>> patches), Ia32 target, my boot hangs (spins) with the log ending in:
>>
>>> SmmInstallProtocolInterface: [EdkiiSmmExitBootServicesProtocol] 0
>>
>> That is, MpInitChangeApLoopCallback() is entered, but it never finishes.
>> "info cpus" prints:
>>
>> * CPU #0: pc=0x7f1f7763 thread_id=17395
>>   CPU #1: pc=0x7f2ce01e (halted) thread_id=17396
>>   CPU #2: pc=0x7f2ce01e (halted) thread_id=17397
>>   CPU #3: pc=0xfff0 thread_id=17398
>>
>> and I've also seen a case where all the APs were stuck at the reset
>> vector (0xfff0), *not* halted, like VCPU#3 above. They don't
>> spin, they're just stuck. The spinning comes from CPU#0, apparently in
>> MpInitChangeApLoopCallback.
>>
>> *** I flipped the AP sync mode to traditional (considering the relaxed
>> mode shouldn't be required with the broadcast SMIs). This time the log
>> ends with:
>>
>>> SmmInstallProtocolInterface: [EdkiiSmmExitBootServicesProtocol] 0
>>> MpInitChangeApLoopCallback() done!
>>
>> but then QEMU abort()s:
>>
>>> kvm_io_ioeventfd_add: error adding ioeventfd: File exists
>>> 2016-11-14 17:00:41.405+: shutting down, reason=crashed
>>
>> I see some ioeventfd stuff in the recent QEMU history; do you think it's
>> related?
> 
> Yes, just try 2.7 for now or disable vhost.

(1) I think I have some new results.

I used the gdbserver built into QEMU, and I (sort of) single-stepped the
MpInitChangeApLoopCallback() function in
"UefiCpuPkg/Library/MpInitLib/DxeMpLib.c", and whatever else it called.
For this I used the Ia32 1x2x2 configuration. Also, the broadcast SMI
patches were applied to both QEMU and OVMF.

I set a breakpoint on RelocateApLoop(), so that when an AP would start
up, gdb would switch to it automatically (and it did in fact).

I liberally used "info cpus" from a separate terminal, and also "info
thread" from within gdb (which gives an incredibly cool insight into the
VCPU states!)

(2) Here's a few interesting results (strictly empirically):

* when I was stepping through the SendInitSipiSipiAllExcludingSelf()
  function *real slow*, manually, suddenly things worked

* I noticed that, while stepping through the INIT-SIPI-SIPI sequence in
  the above-mentioned BSP function, the APs switched from "halted" to
  "running" after the *first* SIPI. Not the second SIPI, the first one.

Then I went to the KVM code, and looked at arch/x86/kvm/lapic.c. My
explanation is terribly inexact, but:

* __apic_accept_irq() translates the LAPIC writes (APIC_DM_INIT and
  APIC_DM_STARTUP) into pending events (KVM_APIC_INIT and KVM_APIC_SIPI)

* this function also calls kvm_vcpu_kick()

* the kvm_apic_accept_events() function processes the pending events.
  The KVM_APIC_INIT event, if pending, causes kvm_vcpu_reset() to be
  called, and the AP to be moved to KVM_MP_STATE_INIT_RECEIVED state.

  And, in that state, a pending KVM_APIC_SIPI event moves the AP to
  KVM_MP_STATE_RUNNABLE. This confirms my surprising empirical result
  that only one SIPI is awaited after the INIT before launching the APs.

* the kvm_vcpu_kick() function in "virt/kvm/kvm_main.c" boils down to a
  smp_send_reschedule() call, if the kicker and the "kickee" are
  different processors (for example, when the BSP kicks the AP)

  This seemed important because it suggested that host kernel
  scheduling jitter could delay the delivery (reception) of the INIT on
  the AP after the BSP sent it. If the BSP sent the first (and second)
  SIPI really fast after the INIT, then those SIPIs could be missed,
  and the AP would remain stuck in KVM_MP_STATE_INIT_RECEIVED state.

  And this state (remember kvm_vcpu_reset() from above!) is consistent
  with the RIP being stuck at the reset vector address, with the AP
  neither running nor being halted.

(3) So, I looked at the Intel SDM. It says in the sysprog book (yes,
yes, I should be using the humongous fused edition),

8.4.4.1 Typical BSP Initialization Sequence

  15. Broadcasts an INIT-SIPI-SIPI IPI sequence to the APs to

Re: [edk2] [PATCH v2 0/3] Put AP into safe hlt-loop code on S3 path

2016-11-14 Thread Paolo Bonzini



On 14/11/2016 19:07, Laszlo Ersek wrote:
> On 11/14/16 13:00, Paolo Bonzini wrote:
>>
>>
>> On 14/11/2016 12:27, Laszlo Ersek wrote:
>>> Well...
>>>
>>> http://lists.nongnu.org/archive/html/qemu-devel/2015-10/msg05658.html
>>> http://lists.nongnu.org/archive/html/qemu-devel/2015-11/msg00125.html
>>> http://lists.nongnu.org/archive/html/qemu-devel/2015-11/msg00563.html
>>>
>>> Are you suggesting that I resurrect this patch? That would be my
>>> pleasure. Please say yes.
>>
>> It's hard to say no when someone has written the code already. :)
> 
> Thanks. I refreshed both patches (OVMF and QEMU -- no code changes just
> more precise commit messages). Unfortunately, quite a few things seem
> broken, although these patches worked a year ago.
> 
> My QEMU base commit is current master 83c83f9a5266. My host kernel is
> 3.10.0-514.el7.x86_64.
> 
> *** So, when I test these two patches, based on edk2 master (no on-list
> patches), Ia32 target, my boot hangs (spins) with the log ending in:
> 
>> SmmInstallProtocolInterface: [EdkiiSmmExitBootServicesProtocol] 0
> 
> That is, MpInitChangeApLoopCallback() is entered, but it never finishes.
> "info cpus" prints:
> 
> * CPU #0: pc=0x7f1f7763 thread_id=17395
>   CPU #1: pc=0x7f2ce01e (halted) thread_id=17396
>   CPU #2: pc=0x7f2ce01e (halted) thread_id=17397
>   CPU #3: pc=0xfff0 thread_id=17398
> 
> and I've also seen a case where all the APs were stuck at the reset
> vector (0xfff0), *not* halted, like VCPU#3 above. They don't
> spin, they're just stuck. The spinning comes from CPU#0, apparently in
> MpInitChangeApLoopCallback.
> 
> *** I flipped the AP sync mode to traditional (considering the relaxed
> mode shouldn't be required with the broadcast SMIs). This time the log
> ends with:
> 
>> SmmInstallProtocolInterface: [EdkiiSmmExitBootServicesProtocol] 0
>> MpInitChangeApLoopCallback() done!
> 
> but then QEMU abort()s:
> 
>> kvm_io_ioeventfd_add: error adding ioeventfd: File exists
>> 2016-11-14 17:00:41.405+: shutting down, reason=crashed
> 
> I see some ioeventfd stuff in the recent QEMU history; do you think it's
> related?

Yes, just try 2.7 for now or disable vhost.

Paolo

> *** My last attempt was even more strange. I applied Jeff's v2 (this
> series), returned to the relaxed (= currently in-tree) sync mode, and
> (of course) the broadcast SMI patches on both sides. This time I didn't
> even boot an OS, I just entered the setup TUI, and selected the Reset
> option. QEMU crashed again with:
> 
>> kvm_io_ioeventfd_add: error adding ioeventfd: File exists
>> 2016-11-14 17:00:41.405+: shutting down, reason=crashed
> 
> I don't know what to look at, honestly. I think I'll check the reflog
> for my local QEMU master branch, and return to one of my earlier pulls,
> or else use v2.7.0 for testing.
> 
> FWIW, the broadcast SMIs work just fine as long as I'm in the firmware
> (not booting an OS and not resetting, just browsing around); I verified
> with GDB that the broadcast SMI branch was taken in QEMU repeatedly.
> 
> Thanks
> Laszlo
> 
___
edk2-devel mailing list
edk2-devel@lists.01.org
https://lists.01.org/mailman/listinfo/edk2-devel

Re: [edk2] [PATCH v2 0/3] Put AP into safe hlt-loop code on S3 path

2016-11-14 Thread Laszlo Ersek

On 11/14/16 13:00, Paolo Bonzini wrote:
> 
> 
> On 14/11/2016 12:27, Laszlo Ersek wrote:
>> Well...
>>
>> http://lists.nongnu.org/archive/html/qemu-devel/2015-10/msg05658.html
>> http://lists.nongnu.org/archive/html/qemu-devel/2015-11/msg00125.html
>> http://lists.nongnu.org/archive/html/qemu-devel/2015-11/msg00563.html
>>
>> Are you suggesting that I resurrect this patch? That would be my
>> pleasure. Please say yes.
> 
> It's hard to say no when someone has written the code already. :)

Thanks. I refreshed both patches (OVMF and QEMU -- no code changes just
more precise commit messages). Unfortunately, quite a few things seem
broken, although these patches worked a year ago.

My QEMU base commit is current master 83c83f9a5266. My host kernel is
3.10.0-514.el7.x86_64.

*** So, when I test these two patches, based on edk2 master (no on-list
patches), Ia32 target, my boot hangs (spins) with the log ending in:

> SmmInstallProtocolInterface: [EdkiiSmmExitBootServicesProtocol] 0

That is, MpInitChangeApLoopCallback() is entered, but it never finishes.
"info cpus" prints:

* CPU #0: pc=0x7f1f7763 thread_id=17395
  CPU #1: pc=0x7f2ce01e (halted) thread_id=17396
  CPU #2: pc=0x7f2ce01e (halted) thread_id=17397
  CPU #3: pc=0xfff0 thread_id=17398

and I've also seen a case where all the APs were stuck at the reset
vector (0xfff0), *not* halted, like VCPU#3 above. They don't
spin, they're just stuck. The spinning comes from CPU#0, apparently in
MpInitChangeApLoopCallback.

*** I flipped the AP sync mode to traditional (considering the relaxed
mode shouldn't be required with the broadcast SMIs). This time the log
ends with:

> SmmInstallProtocolInterface: [EdkiiSmmExitBootServicesProtocol] 0
> MpInitChangeApLoopCallback() done!

but then QEMU abort()s:

> kvm_io_ioeventfd_add: error adding ioeventfd: File exists
> 2016-11-14 17:00:41.405+: shutting down, reason=crashed

I see some ioeventfd stuff in the recent QEMU history; do you think it's
related?

*** My last attempt was even more strange. I applied Jeff's v2 (this
series), returned to the relaxed (= currently in-tree) sync mode, and
(of course) the broadcast SMI patches on both sides. This time I didn't
even boot an OS, I just entered the setup TUI, and selected the Reset
option. QEMU crashed again with:

> kvm_io_ioeventfd_add: error adding ioeventfd: File exists
> 2016-11-14 17:00:41.405+: shutting down, reason=crashed

I don't know what to look at, honestly. I think I'll check the reflog
for my local QEMU master branch, and return to one of my earlier pulls,
or else use v2.7.0 for testing.

FWIW, the broadcast SMIs work just fine as long as I'm in the firmware
(not booting an OS and not resetting, just browsing around); I verified
with GDB that the broadcast SMI branch was taken in QEMU repeatedly.

Thanks
Laszlo
___
edk2-devel mailing list
edk2-devel@lists.01.org
https://lists.01.org/mailman/listinfo/edk2-devel

Re: [edk2] [PATCH v2 0/3] Put AP into safe hlt-loop code on S3 path

2016-11-14 Thread Paolo Bonzini



On 14/11/2016 12:27, Laszlo Ersek wrote:
> Well...
> 
> http://lists.nongnu.org/archive/html/qemu-devel/2015-10/msg05658.html
> http://lists.nongnu.org/archive/html/qemu-devel/2015-11/msg00125.html
> http://lists.nongnu.org/archive/html/qemu-devel/2015-11/msg00563.html
> 
> Are you suggesting that I resurrect this patch? That would be my
> pleasure. Please say yes.

It's hard to say no when someone has written the code already. :)

Paolo

> Also, no special handling would be necessary on the S3 resume path,
> because after resume, SMM clients like the variable driver would
> continue calling the Trigger() method. SmmControl2Dxe is a Runtime
> DXE Driver. The OVMF side patch looks like this (I think I never
> posted it, but I preserved it):
___
edk2-devel mailing list
edk2-devel@lists.01.org
https://lists.01.org/mailman/listinfo/edk2-devel

Re: [edk2] [PATCH v2 0/3] Put AP into safe hlt-loop code on S3 path

2016-11-14 Thread Laszlo Ersek

On 11/14/16 12:09, Paolo Bonzini wrote:
> 
> 
> On 14/11/2016 11:39, Laszlo Ersek wrote:
>> You've tried that:
>>
>> https://www.mail-archive.com/edk2-devel@lists.01.org/msg02840.html
>> https://www.mail-archive.com/edk2-devel@lists.01.org/msg02923.html
> 
> Uh, right. :)
> 
>> Do you suggest to make the LocalApicLib instances usable at runtime?
>> For that I think we'll need to cover the LAPIC address range with a
>> runtime-marked EfiMemoryMappedIO area. This can be done in
>> "OvmfPkg/SmmControl2Dxe".
>>
>> Also, we'll need a LocalApicLib instance that registers a callback for
>> SetVirtualAddressMap() and converts the LAPIC base address pointer.
>>
>> Currently BaseXApicX2ApicLib.c's GetLocalApicBaseAddress() function uses
>> the MSR_IA32_APIC_BASE register if it's available -- based on CPUID --,
>> and falls back to PcdCpuLocalApicBaseAddress otherwise. And only
>> PcdCpuLocalApicBaseAddress is what we could replace with the virtual
>> pointer. We can't accommodate a guest OS that reprograms the LAPIC base
>> address.
>>
>> Jeff, what do you think?
>>
>> Anyway, I believe KVM doesn't support moving the LAPIC window; is that
>> right?
> 
> No, it doesn't.  Let's add a backdoor in QEMU, where writing a given
> value to port 0xb2 would start generating SMIs to all CPUs.  Then you
> can write this somewhere in the initialization code, and in the S3 boot
> script.

Well...

http://lists.nongnu.org/archive/html/qemu-devel/2015-10/msg05658.html
http://lists.nongnu.org/archive/html/qemu-devel/2015-11/msg00125.html
http://lists.nongnu.org/archive/html/qemu-devel/2015-11/msg00563.html

Are you suggesting that I resurrect this patch? That would be my pleasure. 
Please say yes.

Also, no special handling would be necessary on the S3 resume path, because 
after resume, SMM clients like the variable driver would continue calling the 
Trigger() method. SmmControl2Dxe is a Runtime DXE Driver. The OVMF side patch 
looks like this (I think I never posted it, but I preserved it):

> From 2d0cf255858c265d8b0af36ab586b0dd6a9140fa Mon Sep 17 00:00:00 2001
> From: Laszlo Ersek 
> Date: Fri, 23 Oct 2015 17:10:38 +0200
> Subject: [PATCH] (SQUASH) OvmfPkg: SmmControl2Dxe: select broadcast SMI
>
> Contributed-under: TianoCore Contribution Agreement 1.0
> Signed-off-by: Laszlo Ersek 
> ---
>  OvmfPkg/Include/IndustryStandard/Q35MchIch9.h |  6 --
>  OvmfPkg/SmmControl2Dxe/SmmControl2Dxe.c   | 12 +++-
>  2 files changed, 15 insertions(+), 3 deletions(-)
>
> diff --git a/OvmfPkg/Include/IndustryStandard/Q35MchIch9.h 
> b/OvmfPkg/Include/IndustryStandard/Q35MchIch9.h
> index 18b34a3d4f4e..3fa26395adc7 100644
> --- a/OvmfPkg/Include/IndustryStandard/Q35MchIch9.h
> +++ b/OvmfPkg/Include/IndustryStandard/Q35MchIch9.h
> @@ -83,8 +83,10 @@
>  //
>  // IO ports
>  //
> -#define ICH9_APM_CNT 0xB2
> -#define ICH9_APM_STS 0xB3
> +#define ICH9_APM_CNT0xB2
> +
> +#define ICH9_APM_STS0xB3
> +#define QEMU_ICH9_APM_STS_BROADCAST_SMI   'Q'
>
>  //
>  // IO ports relative to PMBASE
> diff --git a/OvmfPkg/SmmControl2Dxe/SmmControl2Dxe.c 
> b/OvmfPkg/SmmControl2Dxe/SmmControl2Dxe.c
> index e895fd638d48..4f60ca48bdea 100644
> --- a/OvmfPkg/SmmControl2Dxe/SmmControl2Dxe.c
> +++ b/OvmfPkg/SmmControl2Dxe/SmmControl2Dxe.c
> @@ -95,10 +95,20 @@ SmmControl2DxeTrigger (
>// report about hardware status, while this register is fully governed by
>// software.
>//
> +  // On the QEMU platform, we use the status register to request a 
> "broadcast"
> +  // SMI; i.e., to bring all VCPUs into SMM at once. Edk2 handles the case 
> when
> +  // the SMI is raised only on the processor that calls Trigger(), but that
> +  // case has much worse performance.
> +  //
> +  // Note that we exploit the fact that the SMM_CORE never passes a non-NULL
> +  // DataPort pointer (that is, we exploit that it doesn't care about the
> +  // status register value).
> +  //
>// Write to the status register first, as this won't trigger the SMI just
>// yet. Then write to the control register.
>//
> -  IoWrite8 (ICH9_APM_STS, DataPort== NULL ? 0 : *DataPort);
> +  ASSERT (DataPort == NULL);
> +  IoWrite8 (ICH9_APM_STS, QEMU_ICH9_APM_STS_BROADCAST_SMI);
>IoWrite8 (ICH9_APM_CNT, CommandPort == NULL ? 0 : *CommandPort);
>return EFI_SUCCESS;
>  }
> --
> 2.9.2
>

(I guess there's a teeny-tiny window between the two IoWrite8() calls, but I 
don't think any guest OS would permit (or try) an S3 suspend while some 
firmware call is in flight.)

If you wish, I can apply these patches (both QEMU and OVMF) locally, and then 
retest Jeff's v2 and Jiewen's v3 patch sets.

Thank you!
Laszlo

___
edk2-devel mailing list
edk2-devel@lists.01.org
https://lists.01.org/mailman/listinfo/edk2-devel

Re: [edk2] [PATCH v2 0/3] Put AP into safe hlt-loop code on S3 path

2016-11-14 Thread Paolo Bonzini



On 14/11/2016 11:39, Laszlo Ersek wrote:
> You've tried that:
> 
> https://www.mail-archive.com/edk2-devel@lists.01.org/msg02840.html
> https://www.mail-archive.com/edk2-devel@lists.01.org/msg02923.html

Uh, right. :)

> Do you suggest to make the LocalApicLib instances usable at runtime?
> For that I think we'll need to cover the LAPIC address range with a
> runtime-marked EfiMemoryMappedIO area. This can be done in
> "OvmfPkg/SmmControl2Dxe".
> 
> Also, we'll need a LocalApicLib instance that registers a callback for
> SetVirtualAddressMap() and converts the LAPIC base address pointer.
> 
> Currently BaseXApicX2ApicLib.c's GetLocalApicBaseAddress() function uses
> the MSR_IA32_APIC_BASE register if it's available -- based on CPUID --,
> and falls back to PcdCpuLocalApicBaseAddress otherwise. And only
> PcdCpuLocalApicBaseAddress is what we could replace with the virtual
> pointer. We can't accommodate a guest OS that reprograms the LAPIC base
> address.
> 
> Jeff, what do you think?
> 
> Anyway, I believe KVM doesn't support moving the LAPIC window; is that
> right?

No, it doesn't.  Let's add a backdoor in QEMU, where writing a given
value to port 0xb2 would start generating SMIs to all CPUs.  Then you
can write this somewhere in the initialization code, and in the S3 boot
script.

Paolo
___
edk2-devel mailing list
edk2-devel@lists.01.org
https://lists.01.org/mailman/listinfo/edk2-devel

Re: [edk2] [PATCH v2 0/3] Put AP into safe hlt-loop code on S3 path

2016-11-14 Thread Laszlo Ersek

On 11/14/16 09:50, Paolo Bonzini wrote:
> 
> 
> On 14/11/2016 09:17, Laszlo Ersek wrote:
>> On 11/13/16 13:51, Fan, Jeff wrote:
>>> Laszlo,
>>>
>>> Thanks your testing. It seems that there is still some unknown issue 
>>> existing.
>>>
>>> I suggest to push this serial of patches firstly, because they have
>>> big progress to solve the AP crashed issue in
>>> https://bugzilla.tianocore.org/show_bug.cgi?id=216.
>>
>> Sounds good to me.
>>
>>> I could submit another bug to handle "AP lost" issue.
>>
>> I hope that Paolo can continue to help us with the KVM trace analysis.
> 
> I will, but it will take a few days.  In the meanwhile it would be nice
> if you could take a look at using SendSmiIpiAllExcludingSelf() to bridge
> the difference between 0xb2 on QEMU and on real hardware.

You've tried that:

https://www.mail-archive.com/edk2-devel@lists.01.org/msg02840.html
https://www.mail-archive.com/edk2-devel@lists.01.org/msg02923.html

Do you suggest to make the LocalApicLib instances usable at runtime?

For that I think we'll need to cover the LAPIC address range with a
runtime-marked EfiMemoryMappedIO area. This can be done in
"OvmfPkg/SmmControl2Dxe".

Also, we'll need a LocalApicLib instance that registers a callback for
SetVirtualAddressMap() and converts the LAPIC base address pointer.

Currently BaseXApicX2ApicLib.c's GetLocalApicBaseAddress() function uses
the MSR_IA32_APIC_BASE register if it's available -- based on CPUID --,
and falls back to PcdCpuLocalApicBaseAddress otherwise. And only
PcdCpuLocalApicBaseAddress is what we could replace with the virtual
pointer. We can't accommodate a guest OS that reprograms the LAPIC base
address.

Jeff, what do you think?

Anyway, I believe KVM doesn't support moving the LAPIC window; is that
right? (Independently, I seem to recall an attack that stole SMRAM
accesses by hiding SMRAM with the LAPIC window.)

Thanks
Laszlo


>>> Thus, JIewen's
>>> or others' patches could be push as long as they have no additional
>>> issue except for "AP Lost:".
>>
>> I haven't gotten around testing Jiewen's v3 series yet. I think it would
>> be best if I could test Jiewen's v3 after this v2 series of yours is
>> committed. I'll report back with results.
>>
>> Thanks
>> Laszlo
>>
>>>
>>> I could follow up to fix "AP Lost" issue.
>>>
>>> Thanks!
>>> Jeff
>>>
>>>
>>> -Original Message-
>>> From: Laszlo Ersek [mailto:ler...@redhat.com] 
>>> Sent: Saturday, November 12, 2016 3:49 AM
>>> To: Fan, Jeff
>>> Cc: edk2-de...@ml01.01.org; Yao, Jiewen; Paolo Bonzini
>>> Subject: Re: [edk2] [PATCH v2 0/3] Put AP into safe hlt-loop code on S3 path
>>>
>>> On 11/11/16 06:45, Jeff Fan wrote:
>>>> On S3 path, we will wake up APs to restore CPU context in 
>>>> PiSmmCpuDxeSmm driver. In case, one NMI or SMI happens, APs may exit 
>>>> from hlt state and execute the instruction after HLT instruction.
>>>>
>>>> But APs are not running on safe code, it leads OVMF S3 boot unstable.
>>>>
>>>> https://bugzilla.tianocore.org/show_bug.cgi?id=216
>>>>
>>>> I tested real platform with 64bit DXE.
>>>>
>>>> v2:
>>>>   1. Make stack alignment per Laszlo's comment.
>>>>   2. Trim whitespace at end of end per Laszlo's comment.
>>>>   3. Update year mark in file header.
>>>>   4. Enhancement on InterlockedDecrement() per Paolo's comment.
>>>>
>>>> Jeff Fan (3):
>>>>   UefiCpuPkg/PiSmmCpuDxeSmm: Put AP into safe hlt-loop code on S3 path
>>>>   UefiCpuPkg/PiSmmCpuDxeSmm: Place AP to 32bit protected mode on S3 path
>>>>   UefiCpuPkg/PiSmmCpuDxeSmm: Decrease mNumberToFinish in AP safe code
>>>>
>>>>  UefiCpuPkg/PiSmmCpuDxeSmm/CpuS3.c | 33 +-
>>>>  UefiCpuPkg/PiSmmCpuDxeSmm/Ia32/SmmFuncsArch.c | 29 +++-
>>>>  UefiCpuPkg/PiSmmCpuDxeSmm/PiSmmCpuDxeSmm.h| 15 +++
>>>>  UefiCpuPkg/PiSmmCpuDxeSmm/X64/SmmFuncsArch.c  | 63 
>>>> ++-
>>>>  4 files changed, 136 insertions(+), 4 deletions(-)
>>>>
>>>
>>> Applied this locally to master (ffd6b0b1b65e) for testing. I tested the 
>>> series with a suspend-resume loop -- not a busy loop, just manually. (So 
>>> there was always one second or so between adjacent steps.)
>>>
>>> No crashes or emulation failures, but the "AP going lost" issue remains 
>>> present -- sometimes Linux cannot bring up one of the four VCPUs after 
>>> resume.
>>>
>>> In the Ia32 case, this "AP lost" symptom surfaced after the 6th resume.
>>>
>>> In the Ia32X64 case, I experienced the symptom after the 89th resume.
>>>
>>> Thanks
>>> Laszlo
>>>

___
edk2-devel mailing list
edk2-devel@lists.01.org
https://lists.01.org/mailman/listinfo/edk2-devel

Re: [edk2] [PATCH v2 0/3] Put AP into safe hlt-loop code on S3 path

2016-11-14 Thread Laszlo Ersek

On 11/13/16 13:51, Fan, Jeff wrote:
> Laszlo,
> 
> Thanks your testing. It seems that there is still some unknown issue existing.
> 
> I suggest to push this serial of patches firstly, because they have
> big progress to solve the AP crashed issue in
> https://bugzilla.tianocore.org/show_bug.cgi?id=216.

Sounds good to me.

> I could submit another bug to handle "AP lost" issue.

I hope that Paolo can continue to help us with the KVM trace analysis.

> Thus, JIewen's
> or others' patches could be push as long as they have no additional
> issue except for "AP Lost:".

I haven't gotten around testing Jiewen's v3 series yet. I think it would
be best if I could test Jiewen's v3 after this v2 series of yours is
committed. I'll report back with results.

Thanks
Laszlo

> 
> I could follow up to fix "AP Lost" issue.
> 
> Thanks!
> Jeff
> 
> 
> -Original Message-
> From: Laszlo Ersek [mailto:ler...@redhat.com] 
> Sent: Saturday, November 12, 2016 3:49 AM
> To: Fan, Jeff
> Cc: edk2-de...@ml01.01.org; Yao, Jiewen; Paolo Bonzini
> Subject: Re: [edk2] [PATCH v2 0/3] Put AP into safe hlt-loop code on S3 path
> 
> On 11/11/16 06:45, Jeff Fan wrote:
>> On S3 path, we will wake up APs to restore CPU context in 
>> PiSmmCpuDxeSmm driver. In case, one NMI or SMI happens, APs may exit 
>> from hlt state and execute the instruction after HLT instruction.
>>
>> But APs are not running on safe code, it leads OVMF S3 boot unstable.
>>
>> https://bugzilla.tianocore.org/show_bug.cgi?id=216
>>
>> I tested real platform with 64bit DXE.
>>
>> v2:
>>   1. Make stack alignment per Laszlo's comment.
>>   2. Trim whitespace at end of end per Laszlo's comment.
>>   3. Update year mark in file header.
>>   4. Enhancement on InterlockedDecrement() per Paolo's comment.
>>
>> Jeff Fan (3):
>>   UefiCpuPkg/PiSmmCpuDxeSmm: Put AP into safe hlt-loop code on S3 path
>>   UefiCpuPkg/PiSmmCpuDxeSmm: Place AP to 32bit protected mode on S3 path
>>   UefiCpuPkg/PiSmmCpuDxeSmm: Decrease mNumberToFinish in AP safe code
>>
>>  UefiCpuPkg/PiSmmCpuDxeSmm/CpuS3.c | 33 +-
>>  UefiCpuPkg/PiSmmCpuDxeSmm/Ia32/SmmFuncsArch.c | 29 +++-
>>  UefiCpuPkg/PiSmmCpuDxeSmm/PiSmmCpuDxeSmm.h| 15 +++
>>  UefiCpuPkg/PiSmmCpuDxeSmm/X64/SmmFuncsArch.c  | 63 
>> ++-
>>  4 files changed, 136 insertions(+), 4 deletions(-)
>>
> 
> Applied this locally to master (ffd6b0b1b65e) for testing. I tested the 
> series with a suspend-resume loop -- not a busy loop, just manually. (So 
> there was always one second or so between adjacent steps.)
> 
> No crashes or emulation failures, but the "AP going lost" issue remains 
> present -- sometimes Linux cannot bring up one of the four VCPUs after resume.
> 
> In the Ia32 case, this "AP lost" symptom surfaced after the 6th resume.
> 
> In the Ia32X64 case, I experienced the symptom after the 89th resume.
> 
> Thanks
> Laszlo
> 

___
edk2-devel mailing list
edk2-devel@lists.01.org
https://lists.01.org/mailman/listinfo/edk2-devel

Re: [edk2] [PATCH v2 0/3] Put AP into safe hlt-loop code on S3 path

2016-11-13 Thread Yao, Jiewen

Yes. I think it is good to check in because it did fix an existing issue.
I suggest we describe the remaining issue honestly in our checkin log.

Thank you
Yao Jiewen

From: Fan, Jeff
Sent: Sunday, November 13, 2016 8:52 PM
To: Laszlo Ersek <ler...@redhat.com>
Cc: edk2-de...@ml01.01.org; Yao, Jiewen <jiewen@intel.com>; Paolo Bonzini 
<pbonz...@redhat.com>
Subject: RE: [edk2] [PATCH v2 0/3] Put AP into safe hlt-loop code on S3 path

Laszlo,

Thanks your testing. It seems that there is still some unknown issue existing.

I suggest to push this serial of patches firstly, because they have big 
progress to solve the AP crashed issue in 
https://bugzilla.tianocore.org/show_bug.cgi?id=216.

I could submit another bug to handle "AP lost" issue.  Thus, JIewen's  or 
others' patches could be push as long as they have no additional issue except 
for "AP Lost:".

I could follow up to fix "AP Lost" issue.

Thanks!
Jeff

-Original Message-
From: Laszlo Ersek [mailto:ler...@redhat.com]
Sent: Saturday, November 12, 2016 3:49 AM
To: Fan, Jeff
Cc: edk2-de...@ml01.01.org<mailto:edk2-de...@ml01.01.org>; Yao, Jiewen; Paolo 
Bonzini
Subject: Re: [edk2] [PATCH v2 0/3] Put AP into safe hlt-loop code on S3 path

On 11/11/16 06:45, Jeff Fan wrote:
> On S3 path, we will wake up APs to restore CPU context in
> PiSmmCpuDxeSmm driver. In case, one NMI or SMI happens, APs may exit
> from hlt state and execute the instruction after HLT instruction.
>
> But APs are not running on safe code, it leads OVMF S3 boot unstable.
>
> https://bugzilla.tianocore.org/show_bug.cgi?id=216
>
> I tested real platform with 64bit DXE.
>
> v2:
>   1. Make stack alignment per Laszlo's comment.
>   2. Trim whitespace at end of end per Laszlo's comment.
>   3. Update year mark in file header.
>   4. Enhancement on InterlockedDecrement() per Paolo's comment.
>
> Jeff Fan (3):
>   UefiCpuPkg/PiSmmCpuDxeSmm: Put AP into safe hlt-loop code on S3 path
>   UefiCpuPkg/PiSmmCpuDxeSmm: Place AP to 32bit protected mode on S3 path
>   UefiCpuPkg/PiSmmCpuDxeSmm: Decrease mNumberToFinish in AP safe code
>
>  UefiCpuPkg/PiSmmCpuDxeSmm/CpuS3.c | 33 +-
>  UefiCpuPkg/PiSmmCpuDxeSmm/Ia32/SmmFuncsArch.c | 29 +++-
>  UefiCpuPkg/PiSmmCpuDxeSmm/PiSmmCpuDxeSmm.h| 15 +++
>  UefiCpuPkg/PiSmmCpuDxeSmm/X64/SmmFuncsArch.c  | 63
> ++-
>  4 files changed, 136 insertions(+), 4 deletions(-)
>

Applied this locally to master (ffd6b0b1b65e) for testing. I tested the series 
with a suspend-resume loop -- not a busy loop, just manually. (So there was 
always one second or so between adjacent steps.)

No crashes or emulation failures, but the "AP going lost" issue remains present 
-- sometimes Linux cannot bring up one of the four VCPUs after resume.

In the Ia32 case, this "AP lost" symptom surfaced after the 6th resume.

In the Ia32X64 case, I experienced the symptom after the 89th resume.

Thanks
Laszlo
___
edk2-devel mailing list
edk2-devel@lists.01.org
https://lists.01.org/mailman/listinfo/edk2-devel

Re: [edk2] [PATCH v2 0/3] Put AP into safe hlt-loop code on S3 path

2016-11-13 Thread Fan, Jeff

Laszlo,

Thanks your testing. It seems that there is still some unknown issue existing.

I suggest to push this serial of patches firstly, because they have big 
progress to solve the AP crashed issue in 
https://bugzilla.tianocore.org/show_bug.cgi?id=216.

I could submit another bug to handle "AP lost" issue.  Thus, JIewen's  or 
others' patches could be push as long as they have no additional issue except 
for "AP Lost:".

I could follow up to fix "AP Lost" issue.

Thanks!
Jeff

-Original Message-
From: Laszlo Ersek [mailto:ler...@redhat.com] 
Sent: Saturday, November 12, 2016 3:49 AM
To: Fan, Jeff
Cc: edk2-de...@ml01.01.org; Yao, Jiewen; Paolo Bonzini
Subject: Re: [edk2] [PATCH v2 0/3] Put AP into safe hlt-loop code on S3 path

On 11/11/16 06:45, Jeff Fan wrote:
> On S3 path, we will wake up APs to restore CPU context in 
> PiSmmCpuDxeSmm driver. In case, one NMI or SMI happens, APs may exit 
> from hlt state and execute the instruction after HLT instruction.
> 
> But APs are not running on safe code, it leads OVMF S3 boot unstable.
> 
> https://bugzilla.tianocore.org/show_bug.cgi?id=216
> 
> I tested real platform with 64bit DXE.
> 
> v2:
>   1. Make stack alignment per Laszlo's comment.
>   2. Trim whitespace at end of end per Laszlo's comment.
>   3. Update year mark in file header.
>   4. Enhancement on InterlockedDecrement() per Paolo's comment.
> 
> Jeff Fan (3):
>   UefiCpuPkg/PiSmmCpuDxeSmm: Put AP into safe hlt-loop code on S3 path
>   UefiCpuPkg/PiSmmCpuDxeSmm: Place AP to 32bit protected mode on S3 path
>   UefiCpuPkg/PiSmmCpuDxeSmm: Decrease mNumberToFinish in AP safe code
> 
>  UefiCpuPkg/PiSmmCpuDxeSmm/CpuS3.c | 33 +-
>  UefiCpuPkg/PiSmmCpuDxeSmm/Ia32/SmmFuncsArch.c | 29 +++-
>  UefiCpuPkg/PiSmmCpuDxeSmm/PiSmmCpuDxeSmm.h| 15 +++
>  UefiCpuPkg/PiSmmCpuDxeSmm/X64/SmmFuncsArch.c  | 63 
> ++-
>  4 files changed, 136 insertions(+), 4 deletions(-)
> 

Applied this locally to master (ffd6b0b1b65e) for testing. I tested the series 
with a suspend-resume loop -- not a busy loop, just manually. (So there was 
always one second or so between adjacent steps.)

No crashes or emulation failures, but the "AP going lost" issue remains present 
-- sometimes Linux cannot bring up one of the four VCPUs after resume.

In the Ia32 case, this "AP lost" symptom surfaced after the 6th resume.

In the Ia32X64 case, I experienced the symptom after the 89th resume.

Thanks
Laszlo
___
edk2-devel mailing list
edk2-devel@lists.01.org
https://lists.01.org/mailman/listinfo/edk2-devel

Re: [edk2] [PATCH v2 0/3] Put AP into safe hlt-loop code on S3 path

2016-11-11 Thread Laszlo Ersek

On 11/11/16 06:45, Jeff Fan wrote:
> On S3 path, we will wake up APs to restore CPU context in PiSmmCpuDxeSmm
> driver. In case, one NMI or SMI happens, APs may exit from hlt state and
> execute the instruction after HLT instruction.
> 
> But APs are not running on safe code, it leads OVMF S3 boot unstable.
> 
> https://bugzilla.tianocore.org/show_bug.cgi?id=216
> 
> I tested real platform with 64bit DXE.
> 
> v2:
>   1. Make stack alignment per Laszlo's comment.
>   2. Trim whitespace at end of end per Laszlo's comment.
>   3. Update year mark in file header.
>   4. Enhancement on InterlockedDecrement() per Paolo's comment.
> 
> Jeff Fan (3):
>   UefiCpuPkg/PiSmmCpuDxeSmm: Put AP into safe hlt-loop code on S3 path
>   UefiCpuPkg/PiSmmCpuDxeSmm: Place AP to 32bit protected mode on S3 path
>   UefiCpuPkg/PiSmmCpuDxeSmm: Decrease mNumberToFinish in AP safe code
> 
>  UefiCpuPkg/PiSmmCpuDxeSmm/CpuS3.c | 33 +-
>  UefiCpuPkg/PiSmmCpuDxeSmm/Ia32/SmmFuncsArch.c | 29 +++-
>  UefiCpuPkg/PiSmmCpuDxeSmm/PiSmmCpuDxeSmm.h| 15 +++
>  UefiCpuPkg/PiSmmCpuDxeSmm/X64/SmmFuncsArch.c  | 63 
> ++-
>  4 files changed, 136 insertions(+), 4 deletions(-)
> 

Applied this locally to master (ffd6b0b1b65e) for testing. I tested the
series with a suspend-resume loop -- not a busy loop, just manually. (So
there was always one second or so between adjacent steps.)

No crashes or emulation failures, but the "AP going lost" issue remains
present -- sometimes Linux cannot bring up one of the four VCPUs after
resume.

In the Ia32 case, this "AP lost" symptom surfaced after the 6th resume.

In the Ia32X64 case, I experienced the symptom after the 89th resume.

Thanks
Laszlo
___
edk2-devel mailing list
edk2-devel@lists.01.org
https://lists.01.org/mailman/listinfo/edk2-devel

[edk2] [PATCH v2 0/3] Put AP into safe hlt-loop code on S3 path

2016-11-10 Thread Jeff Fan

On S3 path, we will wake up APs to restore CPU context in PiSmmCpuDxeSmm
driver. In case, one NMI or SMI happens, APs may exit from hlt state and
execute the instruction after HLT instruction.

But APs are not running on safe code, it leads OVMF S3 boot unstable.

https://bugzilla.tianocore.org/show_bug.cgi?id=216

I tested real platform with 64bit DXE.

v2:
  1. Make stack alignment per Laszlo's comment.
  2. Trim whitespace at end of end per Laszlo's comment.
  3. Update year mark in file header.
  4. Enhancement on InterlockedDecrement() per Paolo's comment.

Jeff Fan (3):
  UefiCpuPkg/PiSmmCpuDxeSmm: Put AP into safe hlt-loop code on S3 path
  UefiCpuPkg/PiSmmCpuDxeSmm: Place AP to 32bit protected mode on S3 path
  UefiCpuPkg/PiSmmCpuDxeSmm: Decrease mNumberToFinish in AP safe code

 UefiCpuPkg/PiSmmCpuDxeSmm/CpuS3.c | 33 +-
 UefiCpuPkg/PiSmmCpuDxeSmm/Ia32/SmmFuncsArch.c | 29 +++-
 UefiCpuPkg/PiSmmCpuDxeSmm/PiSmmCpuDxeSmm.h| 15 +++
 UefiCpuPkg/PiSmmCpuDxeSmm/X64/SmmFuncsArch.c  | 63 ++-
 4 files changed, 136 insertions(+), 4 deletions(-)

-- 
2.9.3.windows.2

___
edk2-devel mailing list
edk2-devel@lists.01.org
https://lists.01.org/mailman/listinfo/edk2-devel

Re: [edk2] [PATCH v2 0/3] Put AP into safe hlt-loop code on S3 path

Re: [edk2] [PATCH v2 0/3] Put AP into safe hlt-loop code on S3 path

Re: [edk2] [PATCH v2 0/3] Put AP into safe hlt-loop code on S3 path

Re: [edk2] [PATCH v2 0/3] Put AP into safe hlt-loop code on S3 path

Re: [edk2] [PATCH v2 0/3] Put AP into safe hlt-loop code on S3 path

Re: [edk2] [PATCH v2 0/3] Put AP into safe hlt-loop code on S3 path

Re: [edk2] [PATCH v2 0/3] Put AP into safe hlt-loop code on S3 path

Re: [edk2] [PATCH v2 0/3] Put AP into safe hlt-loop code on S3 path

Re: [edk2] [PATCH v2 0/3] Put AP into safe hlt-loop code on S3 path

Re: [edk2] [PATCH v2 0/3] Put AP into safe hlt-loop code on S3 path

Re: [edk2] [PATCH v2 0/3] Put AP into safe hlt-loop code on S3 path

Re: [edk2] [PATCH v2 0/3] Put AP into safe hlt-loop code on S3 path

Re: [edk2] [PATCH v2 0/3] Put AP into safe hlt-loop code on S3 path

Re: [edk2] [PATCH v2 0/3] Put AP into safe hlt-loop code on S3 path

Re: [edk2] [PATCH v2 0/3] Put AP into safe hlt-loop code on S3 path

Re: [edk2] [PATCH v2 0/3] Put AP into safe hlt-loop code on S3 path

Re: [edk2] [PATCH v2 0/3] Put AP into safe hlt-loop code on S3 path

Re: [edk2] [PATCH v2 0/3] Put AP into safe hlt-loop code on S3 path

Re: [edk2] [PATCH v2 0/3] Put AP into safe hlt-loop code on S3 path

[edk2] [PATCH v2 0/3] Put AP into safe hlt-loop code on S3 path

20 matches

Site Navigation

Mail list logo

Footer information