That's a better idea. Maybe we don't to free them but just reserve it for future uses. The reserved 2MB page table can manage 1GB memory. I think in BIOS time typical memory usage I observed is far less than that.
> -----Original Message----- > From: Yao, Jiewen > Sent: Thursday, November 30, 2017 9:59 AM > To: Wang, Jian J <jian.j.w...@intel.com> > Cc: edk2-devel@lists.01.org > Subject: RE: [edk2] [PATCH 0/2] Enable page table write protection > > Clarify my word below: > > Can you just predict and pre-allocate additional pages for future split in > worst > case? > If new one need split, you can just use the additional pages. > If not, you can free them later. > > For example, you can allocate 2M aligned memory for new pages, for make sure > they are in same page directory. > > I think it can resolve the problem. > The key is to pre-calculate how many pages are involved in the worst case. > > Thank you > Yao Jiewen > > > > -----Original Message----- > > From: Wang, Jian J > > Sent: Thursday, November 30, 2017 9:47 AM > > To: Yao, Jiewen <jiewen....@intel.com> > > Cc: edk2-devel@lists.01.org > > Subject: RE: [edk2] [PATCH 0/2] Enable page table write protection > > > > Good idea. I think it will highly reduce the possibility but not solve it > > still. > > The dilemma here is: do we need to take care of it? This issue exists in > > theory > > but has not yet been encountered in practice. > > > > > -----Original Message----- > > > From: Yao, Jiewen > > > Sent: Thursday, November 30, 2017 9:36 AM > > > To: Wang, Jian J <jian.j.w...@intel.com> > > > Cc: edk2-devel@lists.01.org > > > Subject: RE: [edk2] [PATCH 0/2] Enable page table write protection > > > > > > Can you just allocate 1 more page for split? > > > If new one need split, you can just use the additional page. > > > If not, you can free it later. > > > > > > Thank you > > > Yao Jiewen > > > > > > > > > > -----Original Message----- > > > > From: Wang, Jian J > > > > Sent: Thursday, November 30, 2017 9:17 AM > > > > To: Yao, Jiewen <jiewen....@intel.com> > > > > Cc: edk2-devel@lists.01.org > > > > Subject: RE: [edk2] [PATCH 0/2] Enable page table write protection > > > > > > > > When you split page tables, you need to allocate new pages for new page > > > table. > > > > Since you have new page tables added, you need to mark them to be > > read-only > > > > as well, right? When you do this, the page table for the memory newly > > > allocated > > > > might still needs to be split. This is the worst case but there's still > > > > chance of > it, > > > > right? We cannot guarantee, theoretically, the page table for the newly > > > allocated > > > > page tables is in the same as just split ones. So, in worst case, once > > > > we > want > > to > > > > mark new page table to be read-only, we need to split the page table > > > managing > > > > those memory, and if we need to do split, we need to allocate more new > > > pages. > > > > This might fall into a recursive situation until all the smallest page > > > > table are > > > > created. > > > > In practice it's almost impossible to let it happen but the chances are > > > > we > will > > > fall > > > > into > > > > such recursive situation more than one level. > > > > > > > > > -----Original Message----- > > > > > From: Yao, Jiewen > > > > > Sent: Thursday, November 30, 2017 8:52 AM > > > > > To: Wang, Jian J <jian.j.w...@intel.com> > > > > > Cc: edk2-devel@lists.01.org > > > > > Subject: RE: [edk2] [PATCH 0/2] Enable page table write protection > > > > > > > > > > -- whenever you're trying to mark one page used as page table to be > read- > > > only, > > > > > you need to split its page table in advance > > > > > [Jiewen] Sorry, I do not quite understand the problem statement. > > > > > To set a page to be ReadOnly, you just flip the R/W bit in the page > > > > > entry. > > > > > The step I expect is: > > > > > 1) You split page table. > > > > > 2) Fill the data structure in the new entry. > > > > > 3) Flip R/W bit (<== This is the new step). > > > > > 4) FlushPageTable. > > > > > > > > > > Thank you > > > > > Yao Jiewen > > > > > > > > > > > -----Original Message----- > > > > > > From: Wang, Jian J > > > > > > Sent: Thursday, November 30, 2017 8:44 AM > > > > > > To: Yao, Jiewen <jiewen....@intel.com> > > > > > > Cc: edk2-devel@lists.01.org > > > > > > Subject: RE: [edk2] [PATCH 0/2] Enable page table write protection > > > > > > > > > > > > I did think of protecting new page table in cpu driver. But I think > > > > > > there's > > > risks > > > > > > to do it, which is that there might be a chance that, whenever > > > > > > you're > > trying > > > to > > > > > > mark one page used as page table to be read-only, you need to split > > > > > > its > > > page > > > > > > table in advance, and again and again, until all page tables are > > > > > > for 4KB > > > pages. > > > > > > > > > > > > Although this is a rare case, or the page table "split" will just > > > > > > happen for > a > > > few > > > > > > times, it will slow down the boot process a little bit anyway. So I > > > > > > though > > > it's > > > > > > safer not to protect new page tables created after DxeIpl. > > > > > > > > > > > > Maybe it's just over worrying about it. Maybe we just need to add a > PCD > > to > > > > > > turn on/off it just in case. Do you have any ideas in mind? > > > > > > > > > > > > > -----Original Message----- > > > > > > > From: Yao, Jiewen > > > > > > > Sent: Wednesday, November 29, 2017 9:35 PM > > > > > > > To: Wang, Jian J <jian.j.w...@intel.com> > > > > > > > Cc: edk2-devel@lists.01.org > > > > > > > Subject: Re: [edk2] [PATCH 0/2] Enable page table write protection > > > > > > > > > > > > > > I do think we need update CPU driver to protect new allocated > > > > > > > split > > page. > > > > > > > > > > > > > > If you verified in shell env, I think it will exposed, please add > > > > > > > a test to > > > trigger > > > > > > > page split in CPU driver. > > > > > > > > > > > > > > I recommend to write some unit test to parse page table in shell. > > > > > > > > > > > > > > thank you! > > > > > > > Yao, Jiewen > > > > > > > > > > > > > > > > > > > > > > 在 2017年11月29日,下午8:15,Wang, Jian J > > > <jian.j.w...@intel.com> > > > > > > 写 > > > > > > > 道: > > > > > > > > > > > > > > > > It's in the DxeIplPeim. > > > > > > > > > > > > > > > > By the way, there's an issue in this patch. I forgot to protect > > > > > > > > page > table > > > for > > > > > > 32- > > > > > > > bit mode. > > > > > > > > So this patch works only for 64-bit mode. I'll add it in v2 > > > > > > > > patch. > > > > > > > > > > > > > > > >> -----Original Message----- > > > > > > > >> From: Yao, Jiewen > > > > > > > >> Sent: Wednesday, November 29, 2017 6:56 PM > > > > > > > >> To: Wang, Jian J <jian.j.w...@intel.com> > > > > > > > >> Cc: edk2-devel@lists.01.org > > > > > > > >> Subject: Re: [edk2] [PATCH 0/2] Enable page table write > > > > > > > >> protection > > > > > > > >> > > > > > > > >> Is this code in CPU driver? > > > > > > > >> > > > > > > > >> thank you! > > > > > > > >> Yao, Jiewen > > > > > > > >> > > > > > > > >> > > > > > > > >>> 在 2017年11月29日,下午6:24,Wang, Jian J > > > > > <jian.j.w...@intel.com> > > > > > > > 写 > > > > > > > >> 道: > > > > > > > >>> > > > > > > > >>> Yes, I validated them manually with JTAG debug tool. > > > > > > > >>> > > > > > > > >>> if ((L3PageTable[Index3] & IA32_PG_PS) != 0) { > > > > > > > >>> // 1G page. Split to 2M. > > > > > > > >>> L2PageTable = AllocatePages (1); > > > > > > > >>> ASSERT (L2PageTable != NULL); > > > > > > > >>> PhysicalAddress = L3PageTable[Index3] & > > > > > > PAGING_1G_ADDRESS_MASK_64; > > > > > > > >>> for (Index = 0; Index < EFI_PAGE_SIZE/sizeof (UINT64); > > > > > > > >>> ++Index) > > { > > > > > > > >>> L2PageTable[Index] = PhysicalAddress | AddressEncMask | > > > > > > > >>> IA32_PG_PS | IA32_PG_P | > > > > IA32_PG_RW; > > > > > > > >>> PhysicalAddress += SIZE_2MB; > > > > > > > >>> } > > > > > > > >>> L3PageTable[Index3] = (UINT64) (UINTN) L2PageTable | > > > > > > AddressEncMask | > > > > > > > >>> IA32_PG_P | > > > > > > IA32_PG_RW; > > > > > > > >>> SetPageReadOnly (PageTableBase, > > > > > > > >> (EFI_PHYSICAL_ADDRESS)(UINTN)L2PageTable); > > > > > > > >>> } > > > > > > > >>> > > > > > > > >>> The newly allocated page table is set in the SetPageReadOnly() > itself > > > > > > > >> recursively, like > > > > > > > >>> above code in which L2PageTable is allocated and then set it > > > > > > > >>> to > be > > > > > > read-only > > > > > > > >> after > > > > > > > >>> initializing the table content. > > > > > > > >>> > > > > > > > >>>> -----Original Message----- > > > > > > > >>>> From: Yao, Jiewen > > > > > > > >>>> Sent: Wednesday, November 29, 2017 5:16 PM > > > > > > > >>>> To: Wang, Jian J <jian.j.w...@intel.com> > > > > > > > >>>> Cc: edk2-devel@lists.01.org > > > > > > > >>>> Subject: Re: [edk2] [PATCH 0/2] Enable page table write > > protection > > > > > > > >>>> > > > > > > > >>>> Thanks. > > > > > > > >>>> > > > > > > > >>>> May I know if this is validated in uefi shell, that all page > > > > > > > >>>> table is > > > > > readonly? > > > > > > > >>>> > > > > > > > >>>> I did not find the code to set new allocated split page to be > > > readonly. > > > > > Can > > > > > > > you > > > > > > > >>>> give me a hand on that? > > > > > > > >>>> > > > > > > > >>>> thank you! > > > > > > > >>>> Yao, Jiewen > > > > > > > >>>> > > > > > > > >>>> > > > > > > > >>>>> 在 2017年11月29日,下午4:47,Jian J Wang > > > > > > <jian.j.w...@intel.com> > > > > > > > >> 写 > > > > > > > >>>> 道: > > > > > > > >>>>> > > > > > > > >>>>> Write Protect feature (CR0.WP) is always enabled in driver > > > > > > > >>>> UefiCpuPkg/CpuDxe. > > > > > > > >>>>> But the memory pages used for page table are not set as > > read-only > > > > in > > > > > > the > > > > > > > >>>> driver > > > > > > > >>>>> DxeIplPeim, after the paging is setup. This might > > > > > > > >>>>> jeopardize the > > > page > > > > > > > table > > > > > > > >>>>> integrity if there's buffer overflow occured in other part > > > > > > > >>>>> of > > system. > > > > > > > >>>>> > > > > > > > >>>>> This patch series will change this situation by clearing > > > > > > > >>>>> R/W bit > in > > > > page > > > > > > > >> attribute > > > > > > > >>>>> of the pages used as page table. > > > > > > > >>>>> > > > > > > > >>>>> Validation works include booting Windows (10/server 2016) > and > > > > Linux > > > > > > > >>>> (Fedora/Ubuntu) > > > > > > > >>>>> on OVMF and Intel real platform. > > > > > > > >>>>> > > > > > > > >>>>> Jian J Wang (2): > > > > > > > >>>>> UefiCpuPkg/CpuDxe: Check CR0.WP before changing page > table > > > > > > > >>>>> MdeModulePkg/DxeIpl: Mark page table as read-only > > > > > > > >>>>> > > > > > > > >>>>> MdeModulePkg/Core/DxeIplPeim/X64/VirtualMemory.c | 166 > > > > > > > >>>> +++++++++++++++++++++++ > > > > > > > >>>>> MdeModulePkg/Core/DxeIplPeim/X64/VirtualMemory.h | 14 > > ++ > > > > > > > >>>>> UefiCpuPkg/CpuDxe/CpuPageTable.c | 65 > > > > > > ++++++++- > > > > > > > >>>>> 3 files changed, 241 insertions(+), 4 deletions(-) > > > > > > > >>>>> > > > > > > > >>>>> -- > > > > > > > >>>>> 2.14.1.windows.1 > > > > > > > >>>>> > > > > > > > >>>>> _______________________________________________ > > > > > > > >>>>> 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