From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: from mail-it0-x22b.google.com (mail-it0-x22b.google.com [IPv6:2607:f8b0:4001:c0b::22b]) (using TLSv1.2 with cipher ECDHE-RSA-AES128-GCM-SHA256 (128/128 bits)) (No client certificate requested) by ml01.01.org (Postfix) with ESMTPS id 92BA921AE30C1 for ; Sat, 3 Jun 2017 11:32:02 -0700 (PDT) Received: by mail-it0-x22b.google.com with SMTP id m47so44907202iti.0 for ; Sat, 03 Jun 2017 11:33:06 -0700 (PDT) DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=linaro.org; s=google; h=mime-version:in-reply-to:references:from:date:message-id:subject:to :cc; bh=rj9rXzoIT+TJxgbyyfuwU2MbbbbEgRMfHQPc/WiDFf4=; b=hnro+bdDUw8xnLjeI5HJeUJMH1R+iF17hz9igxcG7+Z0cjhC1+pNEWqQfXEtnHrSOC 8DlMvL5IIsxFswcT4We5dx46QYFmrvbX95MqchXC7cK4EAjn25H0r8FDsfnF5LWXRmf6 hOFnQqNjGKkT9AOu9Znz9oAv9ec4JoZm0kvhc= X-Google-DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=1e100.net; s=20161025; h=x-gm-message-state:mime-version:in-reply-to:references:from:date :message-id:subject:to:cc; bh=rj9rXzoIT+TJxgbyyfuwU2MbbbbEgRMfHQPc/WiDFf4=; b=nZAZCBXkIciSQSF4F5p7POetSKovun1FcraJYBnUzhRxbhZKWLJJXYdiPWpZ5MrZIS wZ2phv7AnKZwlsDDvIl9sX3p6rNjVRoBnZxYhYgJ7pONzr9ouYYFxY/c9rlCoqdoIOHI xYOlVflShSJJ0FWteUDGUfVnYgzmz3k8eq3O5BthLbObQO8+fYWLsAFcHpQTw+G/IEfS SAUZrWMkAEP5niwEr0jSK7NsVObSGLfTuBDAIPeJVSMIDS5MN5PDR4H+IIb7oQhiodNy FDma2JsGKCNjqkBB07aG1sVUBTPGkISHWfqH6Db3x7qVJ1s+e/43uyOWy0yzIGQ8Zgdm T7nw== X-Gm-Message-State: AODbwcCOjvKw6CVQ8fCSXnkKhO6q005ShvTGEgEJlPxS1An06BvVlNrh bECXafvXCgMnlgsUnHdQvw/othn7oART X-Received: by 10.36.92.143 with SMTP id q137mr5219314itb.98.1496514785511; Sat, 03 Jun 2017 11:33:05 -0700 (PDT) MIME-Version: 1.0 Received: by 10.107.164.24 with HTTP; Sat, 3 Jun 2017 11:33:04 -0700 (PDT) In-Reply-To: <20170603154203.29907-2-lersek@redhat.com> References: <20170603154203.29907-1-lersek@redhat.com> <20170603154203.29907-2-lersek@redhat.com> From: Ard Biesheuvel Date: Sat, 3 Jun 2017 18:33:04 +0000 Message-ID: To: Laszlo Ersek Cc: edk2-devel-01 , Gerd Hoffmann , Igor Mammedov , Jordan Justen Subject: Re: [PATCH 1/1] OvmfPkg/AcpiPlatformDxe: alloc blobs from 64-bit space unless restricted X-BeenThere: edk2-devel@lists.01.org X-Mailman-Version: 2.1.22 Precedence: list List-Id: EDK II Development List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , X-List-Received-Date: Sat, 03 Jun 2017 18:32:02 -0000 Content-Type: text/plain; charset="UTF-8" On 3 June 2017 at 15:42, Laszlo Ersek wrote: > ... by narrower than 8-byte ADD_POINTER references. > > Introduce the CollectRestrictedAllocations() function, which iterates over > the linker/loader script, and collects the names of the fw_cfg blobs that > are referenced by QEMU_LOADER_ADD_POINTER.PointeeFile fields, such that > QEMU_LOADER_ADD_POINTER.PointerSize is less than 8. This means that the > pointee blob's address will have to be patched into a narrower-than-8 byte > pointer field, hence the pointee blob must not be allocated from 64-bit > address space. > > In ProcessCmdAllocate(), consult these restrictions when setting the > maximum address for gBS->AllocatePages(). The default is now MAX_UINT64, > unless restricted like described above to the pre-patch MAX_UINT32 limit. > > In combination with Ard's QEMU commit cb51ac2ffe36 ("hw/arm/virt: generate > 64-bit addressable ACPI objects", 2017-04-10), this patch enables > OvmfPkg/AcpiPlatformDxe to work entirely above the 4GB mark. > > (An upcoming / planned aarch64 QEMU machine type will have no RAM under > 4GB at all. Plus, moving the allocations higher is beneficial to the > current "virt" machine type as well; in Ard's words: "having all firmware > allocations inside the same 1 GB (or 512 MB for 64k pages) frame reduces > the TLB footprint".) > > Cc: Ard Biesheuvel > Cc: Gerd Hoffmann > Cc: Igor Mammedov > Cc: Jordan Justen > Suggested-by: Igor Mammedov > Suggested-by: Gerd Hoffmann > Contributed-under: TianoCore Contribution Agreement 1.0 > Signed-off-by: Laszlo Ersek Thanks Laszlo This works as expected for me. The ACPI reclaim regions happily reside beyond the 4 GB mark on mach-virt. I haven't tried the 'enterprise' flavor machine, but given that it does not actually exist yet, there are no compatibility concerns, and so the behavior observed on mach-virt should extrapolate to that machine as well. Tested-by: Ard Biesheuvel I am quite pleased with this solution, given that it should seamlessly interoperate with QEMU versions with or without patch cb51ac2ffe36. One realization I did make due to my involvement in this discussion is that ACPI reclaim regions are omitted from the linear mapping on arm64 for no good reason: even though we preserve such regions (for kexec or other 'late' references to their contents), the fact that the firmware allows them to be reclaimed implies that they have no special significance to the firmware, and so there is really no good reason to punch them out of the linear mapping. So while the [modest] TLB footprint concern* does apply to mach-virt currently, I intend to propose a change to the arm64 kernel to deal with ACPI reclaim regions differently, after which this benefit no longer exists. Thanks for all the work. Ard. * The default arm64 defconfig defines 4 levels of stage 1 page tables, and KVM uses 3 levels with concatenation at level one, resulting in a 48-bit VA space and a 40-bit IPA space. Under virt, each level of lookup at stage 1 may result in a full page table walk at stage 2, and so it is generally wise to use as few levels as possible. ARM is finicky about aliased mappings with mismatched attributes, and so we omit 'special' memory regions from the linear mapping, which means the remaining DRAM in the same 1 GB frame cannot be mapped as efficiently. This is why it is preferable to keep all such regions close together. > --- > OvmfPkg/AcpiPlatformDxe/QemuFwCfgAcpi.c | 190 +++++++++++++++++++- > 1 file changed, 183 insertions(+), 7 deletions(-) > > diff --git a/OvmfPkg/AcpiPlatformDxe/QemuFwCfgAcpi.c b/OvmfPkg/AcpiPlatformDxe/QemuFwCfgAcpi.c > index 1bc5fe297a96..7aeb6472188b 100644 > --- a/OvmfPkg/AcpiPlatformDxe/QemuFwCfgAcpi.c > +++ b/OvmfPkg/AcpiPlatformDxe/QemuFwCfgAcpi.c > @@ -132,14 +132,169 @@ PointerCompare ( > } > > > /** > + Comparator function for two ASCII strings. Can be used as both Key and > + UserStruct comparator. > + > + This function exists solely so we can avoid casting &AsciiStrCmp to > + ORDERED_COLLECTION_USER_COMPARE and ORDERED_COLLECTION_KEY_COMPARE. > + > + @param[in] AsciiString1 Pointer to the first ASCII string. > + > + @param[in] AsciiString2 Pointer to the second ASCII string. > + > + @return The return value of AsciiStrCmp (AsciiString1, AsciiString2). > +**/ > +STATIC > +INTN > +EFIAPI > +AsciiStringCompare ( > + IN CONST VOID *AsciiString1, > + IN CONST VOID *AsciiString2 > + ) > +{ > + return AsciiStrCmp (AsciiString1, AsciiString2); > +} > + > + > +/** > + Release the ORDERED_COLLECTION structure populated by > + CollectRestrictedAllocations() (below). > + > + This function may be called by CollectRestrictedAllocations() itself, on > + the error path. > + > + @param[in] RestrictedAllocations The ORDERED_COLLECTION structure to > + release. > +**/ > +STATIC > +VOID > +ReleaseRestrictedAllocations ( > + IN ORDERED_COLLECTION *RestrictedAllocations > +) > +{ > + ORDERED_COLLECTION_ENTRY *Entry, *Entry2; > + > + for (Entry = OrderedCollectionMin (RestrictedAllocations); > + Entry != NULL; > + Entry = Entry2) { > + Entry2 = OrderedCollectionNext (Entry); > + OrderedCollectionDelete (RestrictedAllocations, Entry, NULL); > + } > + OrderedCollectionUninit (RestrictedAllocations); > +} > + > + > +/** > + Iterate over the linker/loader script, and collect the names of the fw_cfg > + blobs that are referenced by QEMU_LOADER_ADD_POINTER.PointeeFile fields, such > + that QEMU_LOADER_ADD_POINTER.PointerSize is less than 8. This means that the > + pointee blob's address will have to be patched into a narrower-than-8 byte > + pointer field, hence the pointee blob must not be allocated from 64-bit > + address space. > + > + @param[out] RestrictedAllocations The ORDERED_COLLECTION structure linking > + (not copying / owning) such > + QEMU_LOADER_ADD_POINTER.PointeeFile fields > + that name the blobs restricted from 64-bit > + allocation. > + > + @param[in] LoaderStart Points to the first entry in the > + linker/loader script. > + > + @param[in] LoaderEnd Points one past the last entry in the > + linker/loader script. > + > + @retval EFI_SUCCESS RestrictedAllocations has been populated. > + > + @retval EFI_OUT_OF_RESOURCES Memory allocation failed. > + > + @retval EFI_PROTOCOL_ERROR Invalid linker/loader script contents. > +**/ > +STATIC > +EFI_STATUS > +CollectRestrictedAllocations ( > + OUT ORDERED_COLLECTION **RestrictedAllocations, > + IN CONST QEMU_LOADER_ENTRY *LoaderStart, > + IN CONST QEMU_LOADER_ENTRY *LoaderEnd > +) > +{ > + ORDERED_COLLECTION *Collection; > + CONST QEMU_LOADER_ENTRY *LoaderEntry; > + EFI_STATUS Status; > + > + Collection = OrderedCollectionInit (AsciiStringCompare, AsciiStringCompare); > + if (Collection == NULL) { > + return EFI_OUT_OF_RESOURCES; > + } > + > + for (LoaderEntry = LoaderStart; LoaderEntry < LoaderEnd; ++LoaderEntry) { > + CONST QEMU_LOADER_ADD_POINTER *AddPointer; > + > + if (LoaderEntry->Type != QemuLoaderCmdAddPointer) { > + continue; > + } > + AddPointer = &LoaderEntry->Command.AddPointer; > + > + if (AddPointer->PointerSize >= 8) { > + continue; > + } > + > + if (AddPointer->PointeeFile[QEMU_LOADER_FNAME_SIZE - 1] != '\0') { > + DEBUG ((DEBUG_ERROR, "%a: malformed file name\n", __FUNCTION__)); > + Status = EFI_PROTOCOL_ERROR; > + goto RollBack; > + } > + > + Status = OrderedCollectionInsert ( > + Collection, > + NULL, // Entry > + (VOID *)AddPointer->PointeeFile > + ); > + switch (Status) { > + case EFI_SUCCESS: > + DEBUG (( > + DEBUG_VERBOSE, > + "%a: restricting blob \"%a\" from 64-bit allocation\n", > + __FUNCTION__, > + AddPointer->PointeeFile > + )); > + break; > + case EFI_ALREADY_STARTED: > + // > + // The restriction has been recorded already. > + // > + break; > + case EFI_OUT_OF_RESOURCES: > + goto RollBack; > + default: > + ASSERT (FALSE); > + } > + } > + > + *RestrictedAllocations = Collection; > + return EFI_SUCCESS; > + > +RollBack: > + ReleaseRestrictedAllocations (Collection); > + return Status; > +} > + > + > +/** > Process a QEMU_LOADER_ALLOCATE command. > > - @param[in] Allocate The QEMU_LOADER_ALLOCATE command to process. > + @param[in] Allocate The QEMU_LOADER_ALLOCATE command to > + process. > > - @param[in,out] Tracker The ORDERED_COLLECTION tracking the BLOB user > - structures created thus far. > + @param[in,out] Tracker The ORDERED_COLLECTION tracking the BLOB > + user structures created thus far. > + > + @param[in] RestrictedAllocations The ORDERED_COLLECTION populated by > + CollectRestrictedAllocations(), naming the > + fw_cfg blobs that must not be allocated > + from 64-bit address space. > > @retval EFI_SUCCESS An area of whole AcpiNVS pages has been > allocated for the blob contents, and the > contents have been saved. A BLOB object (user > @@ -163,9 +318,10 @@ STATIC > EFI_STATUS > EFIAPI > ProcessCmdAllocate ( > IN CONST QEMU_LOADER_ALLOCATE *Allocate, > - IN OUT ORDERED_COLLECTION *Tracker > + IN OUT ORDERED_COLLECTION *Tracker, > + IN ORDERED_COLLECTION *RestrictedAllocations > ) > { > FIRMWARE_CONFIG_ITEM FwCfgItem; > UINTN FwCfgSize; > @@ -192,9 +348,12 @@ ProcessCmdAllocate ( > return Status; > } > > NumPages = EFI_SIZE_TO_PAGES (FwCfgSize); > - Address = 0xFFFFFFFF; > + Address = MAX_UINT64; > + if (OrderedCollectionFind (RestrictedAllocations, Allocate->File) != NULL) { > + Address = MAX_UINT32; > + } > Status = gBS->AllocatePages (AllocateMaxAddress, EfiACPIMemoryNVS, NumPages, > &Address); > if (EFI_ERROR (Status)) { > return Status; > @@ -805,8 +964,9 @@ InstallQemuFwCfgTables ( > CONST QEMU_LOADER_ENTRY *LoaderEntry, *LoaderEnd; > CONST QEMU_LOADER_ENTRY *WritePointerSubsetEnd; > ORIGINAL_ATTRIBUTES *OriginalPciAttributes; > UINTN OriginalPciAttributesCount; > + ORDERED_COLLECTION *RestrictedAllocations; > S3_CONTEXT *S3Context; > ORDERED_COLLECTION *Tracker; > UINTN *InstalledKey; > INT32 Installed; > @@ -833,17 +993,26 @@ InstallQemuFwCfgTables ( > QemuFwCfgReadBytes (FwCfgSize, LoaderStart); > RestorePciDecoding (OriginalPciAttributes, OriginalPciAttributesCount); > LoaderEnd = LoaderStart + FwCfgSize / sizeof *LoaderEntry; > > + Status = CollectRestrictedAllocations ( > + &RestrictedAllocations, > + LoaderStart, > + LoaderEnd > + ); > + if (EFI_ERROR (Status)) { > + goto FreeLoader; > + } > + > S3Context = NULL; > if (QemuFwCfgS3Enabled ()) { > // > // Size the allocation pessimistically, assuming that all commands in the > // script are QEMU_LOADER_WRITE_POINTER commands. > // > Status = AllocateS3Context (&S3Context, LoaderEnd - LoaderStart); > if (EFI_ERROR (Status)) { > - goto FreeLoader; > + goto FreeRestrictedAllocations; > } > } > > Tracker = OrderedCollectionInit (BlobCompare, BlobKeyCompare); > @@ -862,9 +1031,13 @@ InstallQemuFwCfgTables ( > WritePointerSubsetEnd = LoaderStart; > for (LoaderEntry = LoaderStart; LoaderEntry < LoaderEnd; ++LoaderEntry) { > switch (LoaderEntry->Type) { > case QemuLoaderCmdAllocate: > - Status = ProcessCmdAllocate (&LoaderEntry->Command.Allocate, Tracker); > + Status = ProcessCmdAllocate ( > + &LoaderEntry->Command.Allocate, > + Tracker, > + RestrictedAllocations > + ); > break; > > case QemuLoaderCmdAddPointer: > Status = ProcessCmdAddPointer (&LoaderEntry->Command.AddPointer, > @@ -1009,8 +1182,11 @@ FreeS3Context: > if (S3Context != NULL) { > ReleaseS3Context (S3Context); > } > > +FreeRestrictedAllocations: > + ReleaseRestrictedAllocations (RestrictedAllocations); > + > FreeLoader: > FreePool (LoaderStart); > > return Status; > -- > 2.9.3 >