Maurice, It’s doable 😊 I prefer to accept today’s below 4G limitation to maintain simple implementation. Thanks, Ray From: Ma, Maurice Sent: Tuesday, February 23, 2021 3:26 PM To: Ni, Ray ; devel@edk2.groups.io; fanjianfeng@byosoft.com.cn; Patrick Rudolph ; Dong, Guo ; Dong, Eric Cc: You, Benjamin Subject: RE: [edk2-devel] [PATCH] UefiPayloadPkg/UefiPayloadEntry: Remove 4GB memory WA Could we decouple BSP and AP GDT in early waking up stage ? For example, AP uses temporary GDT (below 4GB) just for mode switching. Once AP is in the final stage, AP can reload GDT to match BSP. In this way, we don’t have assumption on the final GDT location. Thanks Maurice From: Ni, Ray > Sent: Monday, February 22, 2021 21:22 To: devel@edk2.groups.io; fanjianfeng@byosoft.com.cn; Ma, Maurice >; Patrick Rudolph >; Dong, Guo >; Dong, Eric > Cc: You, Benjamin > Subject: RE: [edk2-devel] [PATCH] UefiPayloadPkg/UefiPayloadEntry: Remove 4GB memory WA “But we could allocate room under 4G for GDT table directly in CpuDxe.” The GDT pre-allocated is re-used by AP. Why do you suggest CpuDxe allocate GDT? Thanks, Ray From: devel@edk2.groups.io > On Behalf Of Jeff Fan Sent: Tuesday, February 23, 2021 11:43 AM To: devel@edk2.groups.io; Ni, Ray >; Ma, Maurice >; Patrick Rudolph >; Dong, Guo >; Dong, Eric > Cc: devel@edk2.groups.io; You, Benjamin > Subject: Re: [edk2-devel] [PATCH] UefiPayloadPkg/UefiPayloadEntry: Remove 4GB memory WA Ray, Yes. You are right. Acutally, x64 IDT table cannot work correctly on protected mode. :-) But for GDT location, I agree it should be located under 4G space to support AP mode changing. But we could allocate room under 4G for GDT table directly in CpuDxe. Thanks, Jeff From: Ni, Ray Date: 2021-02-23 10:52 To: fanjianfeng@byosoft.com.cn; devel; Ma, Maurice; Patrick Rudolph; Dong, Guo; Dong, Eric CC: devel@edk2.groups.io; You, Benjamin Subject: Re: [edk2-devel] [PATCH] UefiPayloadPkg/UefiPayloadEntry: Remove 4GB memory WA Jeff, You are right that BSP’s GDT and IDT tables are under 4G memory. It’s because when AP wakes up, it needs the GDT for entering protected mode. AP cannot access above 4G memory without entering to long mode. I do agree that the 64bit IDT is not proper for AP when entering protected mode. As long as there is no exception in the short time frame (load 64bit IDT, before entering long mode), it’s still ok. Thanks, Ray From: fanjianfeng@byosoft.com.cn > Sent: Tuesday, February 23, 2021 8:50 AM To: devel >; Ma, Maurice >; Patrick Rudolph >; Dong, Guo >; Dong, Eric >; Ni, Ray > Cc: devel@edk2.groups.io; You, Benjamin > Subject: Re: Re: [edk2-devel] [PATCH] UefiPayloadPkg/UefiPayloadEntry: Remove 4GB memory WA we will save the current BSP's GDT and IDT for APs at first time APs are waken by BSP as below. APs will start from real mode to protected mode and then to long mode. During protected mode, BSP's GDT/IDT table are working on APs. In UefiCpuPkg\Library\MpInitLib\MpLib.c, // // Get the BSP's data of GDT and IDT // AsmReadGdtr ((IA32_DESCRIPTOR *) &ExchangeInfo->GdtrProfile); AsmReadIdtr ((IA32_DESCRIPTOR *) &ExchangeInfo->IdtrProfile); It seems to be one bug we have assumption on GDT table and IDT table located under 4G memory space. Could Ray&Eric help me to confirm it? Jeff From: Ma, Maurice Date: 2021-02-23 00:49 To: Patrick Rudolph; Dong, Guo; Dong, Eric; Ni, Ray CC: devel@edk2.groups.io; You, Benjamin Subject: Re: [edk2-devel] [PATCH] UefiPayloadPkg/UefiPayloadEntry: Remove 4GB memory WA Hi, Ray and Eric, Is there any reason why the GDT base was typecast to UINT32 in CpuDxe driver ? In x64 long mode, the GDT base is actually 64bit. Typecasting will zero out the high 32bit address. To me the correct code seems to be something like: gdtPtr.Base = (UINTN)(VOID*) gdt; Thanks Maurice > -----Original Message----- > From: Patrick Rudolph > > Sent: Monday, February 22, 2021 7:43 > To: Dong, Guo > > Cc: devel@edk2.groups.io; Ma, Maurice >; You, > Benjamin > > Subject: Re: [edk2-devel] [PATCH] UefiPayloadPkg/UefiPayloadEntry: > Remove 4GB memory WA > > Hi Guo, > I tested on 078400ee15e7b250e4dfafd840c2e0c19835e16b and run it in > QEMU. > The problem seems to be here, as gdt is allocated > 4GiB: > gdtPtr.Base = (UINT32)(UINTN)(VOID*) gdt; > > Regards, > Patrick > > On Mon, Feb 22, 2021 at 3:59 PM Dong, Guo > wrote: > > > > > > Hi Patrick, > > Please make sure you are using latest master when testing this patch. > > That issue should be fix be this patch: > > UefiCpuPkg/CpuDxe: Fix boot error (commit: > > ebfe2d3eb5ac7fd92d74011edb31303a181920c7) > > And there is similar fix in another place as below: > > UefiCpuPkg/MpInitLib: Fix a hang in above 4GB case (commit: > > edd74ad3ad79b855f76d9cf60a96c405cb3e863b) > > > > Thanks, > > Guo > > > > > -----Original Message----- > > > From: devel@edk2.groups.io > On Behalf Of > > > Patrick Rudolph > > > Sent: Monday, February 22, 2021 7:04 AM > > > To: devel@edk2.groups.io; Ma, Maurice > > > > Cc: Dong, Guo >; You, Benjamin > > > > > > > Subject: Re: [edk2-devel] [PATCH] UefiPayloadPkg/UefiPayloadEntry: > > > Remove 4GB memory WA > > > > > > This patch breaks booting on master. > > > In CpuDxe.efi / InitGlobalDescriptorTable as the GDT pointer is > > > casted to 32bits. > > > > > > Regards, > > > Patrick > > > > > > On Fri, Feb 19, 2021 at 3:12 AM Ma, Maurice > > wrote: > > > > > > > > Reviewed-by: Maurice Ma > > > > > > > > > Regards > > > > Maurice > > > > > > > > > -----Original Message----- > > > > > From: Dong, Guo > > > > > > Sent: Sunday, February 14, 2021 21:13 > > > > > To: devel@edk2.groups.io > > > > > Cc: Ma, Maurice >; You, Benjamin > > > > > > > > > > > Subject: [edk2-devel] [PATCH] UefiPayloadPkg/UefiPayloadEntry: > > > > > Remove 4GB memory WA > > > > > > > > > > Previous it would hang in CpuDxe if DXE drivers are dispatched above > 4GB. > > > > > Now remove the work around since the fixed in CpuDxe are merged. > > > > > > > > > > Signed-off-by: Guo Dong > > > > > > --- > > > > > UefiPayloadPkg/UefiPayloadEntry/UefiPayloadEntry.c | 5 ----- > > > > > 1 file changed, 5 deletions(-) > > > > > > > > > > diff --git a/UefiPayloadPkg/UefiPayloadEntry/UefiPayloadEntry.c > > > > > b/UefiPayloadPkg/UefiPayloadEntry/UefiPayloadEntry.c > > > > > index 805f5448d9..c403b0a80a 100644 > > > > > --- a/UefiPayloadPkg/UefiPayloadEntry/UefiPayloadEntry.c > > > > > +++ b/UefiPayloadPkg/UefiPayloadEntry/UefiPayloadEntry.c > > > > > @@ -40,11 +40,6 @@ MemInfoCallback ( > > > > > EFI_RESOURCE_ATTRIBUTE_WRITE_THROUGH_CACHEABLE | > > > > > EFI_RESOURCE_ATTRIBUTE_WRITE_BACK_CACHEABLE; > > > > > > > > > > - if (Base >= BASE_4GB ) { > > > > > - // Remove tested attribute to avoid DXE core to dispatch driver to > > > > > memory above 4GB > > > > > - Attribue &= ~EFI_RESOURCE_ATTRIBUTE_TESTED; > > > > > - } > > > > > - > > > > > BuildResourceDescriptorHob (Type, Attribue, > > > > > (EFI_PHYSICAL_ADDRESS)Base, Size); > > > > > DEBUG ((DEBUG_INFO , "buildhob: base = 0x%lx, size = 0x%lx, > > > > > type = 0x%x\n", Base, Size, Type)); > > > > > > > > > > -- > > > > > 2.16.2.windows.1 > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > >