[PATCH v2] MdeModulePkg/NonDiscoverablePciDeviceDxe: Allow partial FreeBuffer


Jeff Brasen
 

Add support for partial free of non cached buffers.
If a request for less than the full size is requested new allocations
for the remaining head and tail of the buffer are added to the list.
Added verification that Buffer is EFI_PAGE_SIZE aligned.
The XHCI driver does this if the page size for the controller is >4KB.

Signed-off-by: Jeff Brasen <jbrasen@...>
---
.../NonDiscoverablePciDeviceIo.c | 53 ++++++++++++++++++-
1 file changed, 51 insertions(+), 2 deletions(-)

diff --git a/MdeModulePkg/Bus/Pci/NonDiscoverablePciDeviceDxe/NonDiscoverablePciDeviceIo.c b/MdeModulePkg/Bus/Pci/NonDiscoverablePciDeviceDxe/NonDiscoverablePciDeviceIo.c
index c1c5c6267c..77809cfedf 100644
--- a/MdeModulePkg/Bus/Pci/NonDiscoverablePciDeviceDxe/NonDiscoverablePciDeviceIo.c
+++ b/MdeModulePkg/Bus/Pci/NonDiscoverablePciDeviceDxe/NonDiscoverablePciDeviceIo.c
@@ -960,12 +960,23 @@ NonCoherentPciIoFreeBuffer (
LIST_ENTRY *Entry;
EFI_STATUS Status;
NON_DISCOVERABLE_DEVICE_UNCACHED_ALLOCATION *Alloc;
+ NON_DISCOVERABLE_DEVICE_UNCACHED_ALLOCATION *AllocHead;
+ NON_DISCOVERABLE_DEVICE_UNCACHED_ALLOCATION *AllocTail;
BOOLEAN Found;
+ UINTN StartPages;
+ UINTN EndPages;
+
+ if (HostAddress != ALIGN_POINTER (HostAddress, EFI_PAGE_SIZE)) {
+ ASSERT_EFI_ERROR (EFI_INVALID_PARAMETER);
+ return EFI_INVALID_PARAMETER;
+ }

Dev = NON_DISCOVERABLE_PCI_DEVICE_FROM_PCI_IO (This);

Found = FALSE;
Alloc = NULL;
+ AllocHead = NULL;
+ AllocTail = NULL;

//
// Find the uncached allocation list entry associated
@@ -976,9 +987,13 @@ NonCoherentPciIoFreeBuffer (
Entry = Entry->ForwardLink)
{
Alloc = BASE_CR (Entry, NON_DISCOVERABLE_DEVICE_UNCACHED_ALLOCATION, List);
- if ((Alloc->HostAddress == HostAddress) && (Alloc->NumPages == Pages)) {
+ StartPages = 0;
+ if (Alloc->HostAddress < HostAddress) {
+ StartPages = (HostAddress - Alloc->HostAddress) / EFI_PAGE_SIZE;
+ }
+ if ((Alloc->HostAddress <= HostAddress) && (Alloc->NumPages >= (Pages + StartPages))) {
//
- // We are freeing the exact allocation we were given
+ // We are freeing at least part of what we were given
// before by AllocateBuffer()
//
Found = TRUE;
@@ -991,7 +1006,41 @@ NonCoherentPciIoFreeBuffer (
return EFI_NOT_FOUND;
}

+ EndPages = Alloc->NumPages - (Pages + StartPages);
+
+ if (StartPages != 0) {
+ AllocHead = AllocatePool (sizeof *AllocHead);
+ if (AllocHead == NULL) {
+ return EFI_OUT_OF_RESOURCES;
+ }
+
+ AllocHead->HostAddress = Alloc->HostAddress;
+ AllocHead->NumPages = StartPages;
+ AllocHead->Attributes = Alloc->Attributes;
+ }
+
+ if (EndPages != 0) {
+ AllocTail = AllocatePool (sizeof *AllocTail);
+ if (AllocTail == NULL) {
+ return EFI_OUT_OF_RESOURCES;
+ }
+
+ AllocTail->HostAddress = Alloc->HostAddress + ((Pages + StartPages) * EFI_PAGE_SIZE);
+ AllocTail->NumPages = EndPages;
+ AllocTail->Attributes = Alloc->Attributes;
+ }
+
RemoveEntryList (&Alloc->List);
+ //
+ // Record this new sub allocations in the linked list, so we
+ // can restore the memory space attributes later
+ //
+ if (AllocHead != NULL) {
+ InsertHeadList (&Dev->UncachedAllocationList, &AllocHead->List);
+ }
+ if (AllocTail != NULL) {
+ InsertHeadList (&Dev->UncachedAllocationList, &AllocTail->List);
+ }

Status = gDS->SetMemorySpaceAttributes (
(EFI_PHYSICAL_ADDRESS)(UINTN)HostAddress,
--
2.17.1


Ni, Ray
 

Deepthi,
Can you please help to review the changes and provide comments?

Thanks,
Ray

-----Original Message-----
From: devel@edk2.groups.io <devel@edk2.groups.io> On Behalf Of Jeff Brasen via groups.io
Sent: Tuesday, February 15, 2022 2:46 AM
To: devel@edk2.groups.io
Cc: Wu, Hao A <hao.a.wu@...>; Ni, Ray <ray.ni@...>; Jeff Brasen <jbrasen@...>
Subject: [edk2-devel] [PATCH v2] MdeModulePkg/NonDiscoverablePciDeviceDxe: Allow partial FreeBuffer

Add support for partial free of non cached buffers.
If a request for less than the full size is requested new allocations for the remaining head and tail of the buffer are added to the list.
Added verification that Buffer is EFI_PAGE_SIZE aligned.
The XHCI driver does this if the page size for the controller is >4KB.

Signed-off-by: Jeff Brasen <jbrasen@...>
---
.../NonDiscoverablePciDeviceIo.c | 53 ++++++++++++++++++-
1 file changed, 51 insertions(+), 2 deletions(-)

diff --git a/MdeModulePkg/Bus/Pci/NonDiscoverablePciDeviceDxe/NonDiscoverablePciDeviceIo.c b/MdeModulePkg/Bus/Pci/NonDiscoverablePciDeviceDxe/NonDiscoverablePciDeviceIo.c
index c1c5c6267c..77809cfedf 100644
--- a/MdeModulePkg/Bus/Pci/NonDiscoverablePciDeviceDxe/NonDiscoverablePciDeviceIo.c
+++ b/MdeModulePkg/Bus/Pci/NonDiscoverablePciDeviceDxe/NonDiscoverablePc
+++ iDeviceIo.c
@@ -960,12 +960,23 @@ NonCoherentPciIoFreeBuffer (
LIST_ENTRY *Entry;
EFI_STATUS Status;
NON_DISCOVERABLE_DEVICE_UNCACHED_ALLOCATION *Alloc;
+ NON_DISCOVERABLE_DEVICE_UNCACHED_ALLOCATION *AllocHead;
+ NON_DISCOVERABLE_DEVICE_UNCACHED_ALLOCATION *AllocTail;
BOOLEAN Found;
+ UINTN StartPages;
+ UINTN EndPages;
+
+ if (HostAddress != ALIGN_POINTER (HostAddress, EFI_PAGE_SIZE)) {
+ ASSERT_EFI_ERROR (EFI_INVALID_PARAMETER);
+ return EFI_INVALID_PARAMETER;
+ }

Dev = NON_DISCOVERABLE_PCI_DEVICE_FROM_PCI_IO (This);

Found = FALSE;
Alloc = NULL;
+ AllocHead = NULL;
+ AllocTail = NULL;

//
// Find the uncached allocation list entry associated @@ -976,9 +987,13 @@ NonCoherentPciIoFreeBuffer (
Entry = Entry->ForwardLink)
{
Alloc = BASE_CR (Entry, NON_DISCOVERABLE_DEVICE_UNCACHED_ALLOCATION, List);
- if ((Alloc->HostAddress == HostAddress) && (Alloc->NumPages == Pages)) {
+ StartPages = 0;
+ if (Alloc->HostAddress < HostAddress) {
+ StartPages = (HostAddress - Alloc->HostAddress) / EFI_PAGE_SIZE;
+ }
+ if ((Alloc->HostAddress <= HostAddress) && (Alloc->NumPages >=
+ (Pages + StartPages))) {
//
- // We are freeing the exact allocation we were given
+ // We are freeing at least part of what we were given
// before by AllocateBuffer()
//
Found = TRUE;
@@ -991,7 +1006,41 @@ NonCoherentPciIoFreeBuffer (
return EFI_NOT_FOUND;
}

+ EndPages = Alloc->NumPages - (Pages + StartPages);
+
+ if (StartPages != 0) {
+ AllocHead = AllocatePool (sizeof *AllocHead);
+ if (AllocHead == NULL) {
+ return EFI_OUT_OF_RESOURCES;
+ }
+
+ AllocHead->HostAddress = Alloc->HostAddress;
+ AllocHead->NumPages = StartPages;
+ AllocHead->Attributes = Alloc->Attributes; }
+
+ if (EndPages != 0) {
+ AllocTail = AllocatePool (sizeof *AllocTail);
+ if (AllocTail == NULL) {
+ return EFI_OUT_OF_RESOURCES;
+ }
+
+ AllocTail->HostAddress = Alloc->HostAddress + ((Pages + StartPages) * EFI_PAGE_SIZE);
+ AllocTail->NumPages = EndPages;
+ AllocTail->Attributes = Alloc->Attributes; }
+
RemoveEntryList (&Alloc->List);
+ //
+ // Record this new sub allocations in the linked list, so we // can
+ restore the memory space attributes later // if (AllocHead != NULL)
+ {
+ InsertHeadList (&Dev->UncachedAllocationList, &AllocHead->List); }
+ if (AllocTail != NULL) {
+ InsertHeadList (&Dev->UncachedAllocationList, &AllocTail->List); }

Status = gDS->SetMemorySpaceAttributes (
(EFI_PHYSICAL_ADDRESS)(UINTN)HostAddress,
--
2.17.1


Jeff Brasen
 

Any thoughts on this patch, now that the stable tag is out?

-----Original Message-----
From: Ni, Ray <ray.ni@...>
Sent: Tuesday, February 15, 2022 11:52 PM
To: Sindhu, Deepthi <deepthi.sindhu@...>
Cc: Wu, Hao A <hao.a.wu@...>; Jeff Brasen <jbrasen@...>;
devel@edk2.groups.io
Subject: RE: [edk2-devel] [PATCH v2]
MdeModulePkg/NonDiscoverablePciDeviceDxe: Allow partial FreeBuffer

External email: Use caution opening links or attachments


Deepthi,
Can you please help to review the changes and provide comments?

Thanks,
Ray

-----Original Message-----
From: devel@edk2.groups.io <devel@edk2.groups.io> On Behalf Of Jeff
Brasen via groups.io
Sent: Tuesday, February 15, 2022 2:46 AM
To: devel@edk2.groups.io
Cc: Wu, Hao A <hao.a.wu@...>; Ni, Ray <ray.ni@...>; Jeff
Brasen <jbrasen@...>
Subject: [edk2-devel] [PATCH v2]
MdeModulePkg/NonDiscoverablePciDeviceDxe: Allow partial FreeBuffer

Add support for partial free of non cached buffers.
If a request for less than the full size is requested new allocations for the
remaining head and tail of the buffer are added to the list.
Added verification that Buffer is EFI_PAGE_SIZE aligned.
The XHCI driver does this if the page size for the controller is >4KB.

Signed-off-by: Jeff Brasen <jbrasen@...>
---
.../NonDiscoverablePciDeviceIo.c | 53 ++++++++++++++++++-
1 file changed, 51 insertions(+), 2 deletions(-)

diff --git
a/MdeModulePkg/Bus/Pci/NonDiscoverablePciDeviceDxe/NonDiscoverable
PciDeviceIo.c
b/MdeModulePkg/Bus/Pci/NonDiscoverablePciDeviceDxe/NonDiscoverable
PciDeviceIo.c
index c1c5c6267c..77809cfedf 100644
---
a/MdeModulePkg/Bus/Pci/NonDiscoverablePciDeviceDxe/NonDiscoverable
PciDeviceIo.c
+++
b/MdeModulePkg/Bus/Pci/NonDiscoverablePciDeviceDxe/NonDiscoverable
Pc
+++ iDeviceIo.c
@@ -960,12 +960,23 @@ NonCoherentPciIoFreeBuffer (
LIST_ENTRY *Entry;
EFI_STATUS Status;
NON_DISCOVERABLE_DEVICE_UNCACHED_ALLOCATION *Alloc;
+ NON_DISCOVERABLE_DEVICE_UNCACHED_ALLOCATION *AllocHead;
+ NON_DISCOVERABLE_DEVICE_UNCACHED_ALLOCATION *AllocTail;
BOOLEAN Found;
+ UINTN StartPages;
+ UINTN EndPages;
+
+ if (HostAddress != ALIGN_POINTER (HostAddress, EFI_PAGE_SIZE)) {
+ ASSERT_EFI_ERROR (EFI_INVALID_PARAMETER);
+ return EFI_INVALID_PARAMETER;
+ }

Dev = NON_DISCOVERABLE_PCI_DEVICE_FROM_PCI_IO (This);

Found = FALSE;
Alloc = NULL;
+ AllocHead = NULL;
+ AllocTail = NULL;

//
// Find the uncached allocation list entry associated @@ -976,9 +987,13 @@
NonCoherentPciIoFreeBuffer (
Entry = Entry->ForwardLink)
{
Alloc = BASE_CR (Entry,
NON_DISCOVERABLE_DEVICE_UNCACHED_ALLOCATION, List);
- if ((Alloc->HostAddress == HostAddress) && (Alloc->NumPages == Pages))
{
+ StartPages = 0;
+ if (Alloc->HostAddress < HostAddress) {
+ StartPages = (HostAddress - Alloc->HostAddress) / EFI_PAGE_SIZE;
+ }
+ if ((Alloc->HostAddress <= HostAddress) && (Alloc->NumPages >=
+ (Pages + StartPages))) {
//
- // We are freeing the exact allocation we were given
+ // We are freeing at least part of what we were given
// before by AllocateBuffer()
//
Found = TRUE;
@@ -991,7 +1006,41 @@ NonCoherentPciIoFreeBuffer (
return EFI_NOT_FOUND;
}

+ EndPages = Alloc->NumPages - (Pages + StartPages);
+
+ if (StartPages != 0) {
+ AllocHead = AllocatePool (sizeof *AllocHead);
+ if (AllocHead == NULL) {
+ return EFI_OUT_OF_RESOURCES;
+ }
+
+ AllocHead->HostAddress = Alloc->HostAddress;
+ AllocHead->NumPages = StartPages;
+ AllocHead->Attributes = Alloc->Attributes; }
+
+ if (EndPages != 0) {
+ AllocTail = AllocatePool (sizeof *AllocTail);
+ if (AllocTail == NULL) {
+ return EFI_OUT_OF_RESOURCES;
+ }
+
+ AllocTail->HostAddress = Alloc->HostAddress + ((Pages + StartPages) *
EFI_PAGE_SIZE);
+ AllocTail->NumPages = EndPages;
+ AllocTail->Attributes = Alloc->Attributes; }
+
RemoveEntryList (&Alloc->List);
+ //
+ // Record this new sub allocations in the linked list, so we // can
+ restore the memory space attributes later // if (AllocHead != NULL)
+ {
+ InsertHeadList (&Dev->UncachedAllocationList, &AllocHead->List); }
+ if (AllocTail != NULL) {
+ InsertHeadList (&Dev->UncachedAllocationList, &AllocTail->List); }

Status = gDS->SetMemorySpaceAttributes (
(EFI_PHYSICAL_ADDRESS)(UINTN)HostAddress,
--
2.17.1





Jeff Brasen
 

Any comments on this patch?

 

Thanks,

Jeff


Jeff Brasen
 

Any thoughts on this patch?

-----Original Message-----
From: Jeff Brasen <jbrasen@...>
Sent: Monday, February 14, 2022 11:46 AM
To: devel@edk2.groups.io
Cc: hao.a.wu@...; ray.ni@...; Jeff Brasen
<jbrasen@...>
Subject: [PATCH v2] MdeModulePkg/NonDiscoverablePciDeviceDxe: Allow
partial FreeBuffer

Add support for partial free of non cached buffers.
If a request for less than the full size is requested new allocations for the
remaining head and tail of the buffer are added to the list.
Added verification that Buffer is EFI_PAGE_SIZE aligned.
The XHCI driver does this if the page size for the controller is >4KB.

Signed-off-by: Jeff Brasen <jbrasen@...>
---
.../NonDiscoverablePciDeviceIo.c | 53 ++++++++++++++++++-
1 file changed, 51 insertions(+), 2 deletions(-)

diff --git
a/MdeModulePkg/Bus/Pci/NonDiscoverablePciDeviceDxe/NonDiscoverable
PciDeviceIo.c
b/MdeModulePkg/Bus/Pci/NonDiscoverablePciDeviceDxe/NonDiscoverable
PciDeviceIo.c
index c1c5c6267c..77809cfedf 100644
---
a/MdeModulePkg/Bus/Pci/NonDiscoverablePciDeviceDxe/NonDiscoverable
PciDeviceIo.c
+++
b/MdeModulePkg/Bus/Pci/NonDiscoverablePciDeviceDxe/NonDiscoverable
Pc
+++ iDeviceIo.c
@@ -960,12 +960,23 @@ NonCoherentPciIoFreeBuffer (
LIST_ENTRY *Entry;
EFI_STATUS Status;
NON_DISCOVERABLE_DEVICE_UNCACHED_ALLOCATION *Alloc;
+ NON_DISCOVERABLE_DEVICE_UNCACHED_ALLOCATION *AllocHead;
+ NON_DISCOVERABLE_DEVICE_UNCACHED_ALLOCATION *AllocTail;
BOOLEAN Found;
+ UINTN StartPages;
+ UINTN EndPages;
+
+ if (HostAddress != ALIGN_POINTER (HostAddress, EFI_PAGE_SIZE)) {
+ ASSERT_EFI_ERROR (EFI_INVALID_PARAMETER);
+ return EFI_INVALID_PARAMETER;
+ }

Dev = NON_DISCOVERABLE_PCI_DEVICE_FROM_PCI_IO (This);

Found = FALSE;
Alloc = NULL;
+ AllocHead = NULL;
+ AllocTail = NULL;

//
// Find the uncached allocation list entry associated @@ -976,9 +987,13 @@
NonCoherentPciIoFreeBuffer (
Entry = Entry->ForwardLink)
{
Alloc = BASE_CR (Entry,
NON_DISCOVERABLE_DEVICE_UNCACHED_ALLOCATION, List);
- if ((Alloc->HostAddress == HostAddress) && (Alloc->NumPages == Pages))
{
+ StartPages = 0;
+ if (Alloc->HostAddress < HostAddress) {
+ StartPages = (HostAddress - Alloc->HostAddress) / EFI_PAGE_SIZE;
+ }
+ if ((Alloc->HostAddress <= HostAddress) && (Alloc->NumPages >=
+ (Pages + StartPages))) {
//
- // We are freeing the exact allocation we were given
+ // We are freeing at least part of what we were given
// before by AllocateBuffer()
//
Found = TRUE;
@@ -991,7 +1006,41 @@ NonCoherentPciIoFreeBuffer (
return EFI_NOT_FOUND;
}

+ EndPages = Alloc->NumPages - (Pages + StartPages);
+
+ if (StartPages != 0) {
+ AllocHead = AllocatePool (sizeof *AllocHead);
+ if (AllocHead == NULL) {
+ return EFI_OUT_OF_RESOURCES;
+ }
+
+ AllocHead->HostAddress = Alloc->HostAddress;
+ AllocHead->NumPages = StartPages;
+ AllocHead->Attributes = Alloc->Attributes; }
+
+ if (EndPages != 0) {
+ AllocTail = AllocatePool (sizeof *AllocTail);
+ if (AllocTail == NULL) {
+ return EFI_OUT_OF_RESOURCES;
+ }
+
+ AllocTail->HostAddress = Alloc->HostAddress + ((Pages + StartPages) *
EFI_PAGE_SIZE);
+ AllocTail->NumPages = EndPages;
+ AllocTail->Attributes = Alloc->Attributes; }
+
RemoveEntryList (&Alloc->List);
+ //
+ // Record this new sub allocations in the linked list, so we // can
+ restore the memory space attributes later // if (AllocHead != NULL)
+ {
+ InsertHeadList (&Dev->UncachedAllocationList, &AllocHead->List); }
+ if (AllocTail != NULL) {
+ InsertHeadList (&Dev->UncachedAllocationList, &AllocTail->List); }

Status = gDS->SetMemorySpaceAttributes (
(EFI_PHYSICAL_ADDRESS)(UINTN)HostAddress,
--
2.17.1


Jeff Brasen
 

Adding Leif/Ard to CC incase they have any comments on this patch.

Thanks
Jeff

-----Original Message-----
From: Jeff Brasen
Sent: Friday, June 17, 2022 9:39 AM
To: devel@edk2.groups.io
Cc: hao.a.wu@...; ray.ni@...
Subject: RE: [PATCH v2] MdeModulePkg/NonDiscoverablePciDeviceDxe:
Allow partial FreeBuffer

Any thoughts on this patch?

-----Original Message-----
From: Jeff Brasen <jbrasen@...>
Sent: Monday, February 14, 2022 11:46 AM
To: devel@edk2.groups.io
Cc: hao.a.wu@...; ray.ni@...; Jeff Brasen
<jbrasen@...>
Subject: [PATCH v2] MdeModulePkg/NonDiscoverablePciDeviceDxe: Allow
partial FreeBuffer

Add support for partial free of non cached buffers.
If a request for less than the full size is requested new allocations
for the remaining head and tail of the buffer are added to the list.
Added verification that Buffer is EFI_PAGE_SIZE aligned.
The XHCI driver does this if the page size for the controller is >4KB.

Signed-off-by: Jeff Brasen <jbrasen@...>
---
.../NonDiscoverablePciDeviceIo.c | 53 ++++++++++++++++++-
1 file changed, 51 insertions(+), 2 deletions(-)

diff --git
a/MdeModulePkg/Bus/Pci/NonDiscoverablePciDeviceDxe/NonDiscoverable
PciDeviceIo.c
b/MdeModulePkg/Bus/Pci/NonDiscoverablePciDeviceDxe/NonDiscoverable
PciDeviceIo.c
index c1c5c6267c..77809cfedf 100644
---
a/MdeModulePkg/Bus/Pci/NonDiscoverablePciDeviceDxe/NonDiscoverable
PciDeviceIo.c
+++
b/MdeModulePkg/Bus/Pci/NonDiscoverablePciDeviceDxe/NonDiscoverable
Pc
+++ iDeviceIo.c
@@ -960,12 +960,23 @@ NonCoherentPciIoFreeBuffer (
LIST_ENTRY *Entry;
EFI_STATUS Status;
NON_DISCOVERABLE_DEVICE_UNCACHED_ALLOCATION *Alloc;
+ NON_DISCOVERABLE_DEVICE_UNCACHED_ALLOCATION *AllocHead;
+ NON_DISCOVERABLE_DEVICE_UNCACHED_ALLOCATION *AllocTail;
BOOLEAN Found;
+ UINTN StartPages;
+ UINTN EndPages;
+
+ if (HostAddress != ALIGN_POINTER (HostAddress, EFI_PAGE_SIZE)) {
+ ASSERT_EFI_ERROR (EFI_INVALID_PARAMETER);
+ return EFI_INVALID_PARAMETER;
+ }

Dev = NON_DISCOVERABLE_PCI_DEVICE_FROM_PCI_IO (This);

Found = FALSE;
Alloc = NULL;
+ AllocHead = NULL;
+ AllocTail = NULL;

//
// Find the uncached allocation list entry associated @@ -976,9
+987,13 @@ NonCoherentPciIoFreeBuffer (
Entry = Entry->ForwardLink)
{
Alloc = BASE_CR (Entry,
NON_DISCOVERABLE_DEVICE_UNCACHED_ALLOCATION, List);
- if ((Alloc->HostAddress == HostAddress) && (Alloc->NumPages ==
Pages))
{
+ StartPages = 0;
+ if (Alloc->HostAddress < HostAddress) {
+ StartPages = (HostAddress - Alloc->HostAddress) / EFI_PAGE_SIZE;
+ }
+ if ((Alloc->HostAddress <= HostAddress) && (Alloc->NumPages >=
+ (Pages + StartPages))) {
//
- // We are freeing the exact allocation we were given
+ // We are freeing at least part of what we were given
// before by AllocateBuffer()
//
Found = TRUE;
@@ -991,7 +1006,41 @@ NonCoherentPciIoFreeBuffer (
return EFI_NOT_FOUND;
}

+ EndPages = Alloc->NumPages - (Pages + StartPages);
+
+ if (StartPages != 0) {
+ AllocHead = AllocatePool (sizeof *AllocHead);
+ if (AllocHead == NULL) {
+ return EFI_OUT_OF_RESOURCES;
+ }
+
+ AllocHead->HostAddress = Alloc->HostAddress;
+ AllocHead->NumPages = StartPages;
+ AllocHead->Attributes = Alloc->Attributes; }
+
+ if (EndPages != 0) {
+ AllocTail = AllocatePool (sizeof *AllocTail);
+ if (AllocTail == NULL) {
+ return EFI_OUT_OF_RESOURCES;
+ }
+
+ AllocTail->HostAddress = Alloc->HostAddress + ((Pages +
+ StartPages) *
EFI_PAGE_SIZE);
+ AllocTail->NumPages = EndPages;
+ AllocTail->Attributes = Alloc->Attributes; }
+
RemoveEntryList (&Alloc->List);
+ //
+ // Record this new sub allocations in the linked list, so we //
+ can restore the memory space attributes later // if (AllocHead !=
+ NULL) {
+ InsertHeadList (&Dev->UncachedAllocationList, &AllocHead->List);
+ } if (AllocTail != NULL) {
+ InsertHeadList (&Dev->UncachedAllocationList, &AllocTail->List);
+ }

Status = gDS->SetMemorySpaceAttributes (
(EFI_PHYSICAL_ADDRESS)(UINTN)HostAddress,
--
2.17.1


Ard Biesheuvel
 

On Thu, 28 Jul 2022 at 13:25, Jeff Brasen <jbrasen@...> wrote:


Adding Leif/Ard to CC incase they have any comments on this patch.
This generally looks ok to me. I just wonder if it wouldn't be simpler
to reuse the existing allocation descriptor if it is not being freed
entirely. Given the [presumably] the most common case is to allocate
and then free some pages at the end, lowering the page count on the
existing descriptor would cover most cases, and we'd only need to
allocate new ones if pages are being freed at the start or in the
middle.


-----Original Message-----
From: Jeff Brasen
Sent: Friday, June 17, 2022 9:39 AM
To: devel@edk2.groups.io
Cc: hao.a.wu@...; ray.ni@...
Subject: RE: [PATCH v2] MdeModulePkg/NonDiscoverablePciDeviceDxe:
Allow partial FreeBuffer

Any thoughts on this patch?

-----Original Message-----
From: Jeff Brasen <jbrasen@...>
Sent: Monday, February 14, 2022 11:46 AM
To: devel@edk2.groups.io
Cc: hao.a.wu@...; ray.ni@...; Jeff Brasen
<jbrasen@...>
Subject: [PATCH v2] MdeModulePkg/NonDiscoverablePciDeviceDxe: Allow
partial FreeBuffer

Add support for partial free of non cached buffers.
If a request for less than the full size is requested new allocations
for the remaining head and tail of the buffer are added to the list.
Added verification that Buffer is EFI_PAGE_SIZE aligned.
The XHCI driver does this if the page size for the controller is >4KB.

Signed-off-by: Jeff Brasen <jbrasen@...>
---
.../NonDiscoverablePciDeviceIo.c | 53 ++++++++++++++++++-
1 file changed, 51 insertions(+), 2 deletions(-)

diff --git
a/MdeModulePkg/Bus/Pci/NonDiscoverablePciDeviceDxe/NonDiscoverable
PciDeviceIo.c
b/MdeModulePkg/Bus/Pci/NonDiscoverablePciDeviceDxe/NonDiscoverable
PciDeviceIo.c
index c1c5c6267c..77809cfedf 100644
---
a/MdeModulePkg/Bus/Pci/NonDiscoverablePciDeviceDxe/NonDiscoverable
PciDeviceIo.c
+++
b/MdeModulePkg/Bus/Pci/NonDiscoverablePciDeviceDxe/NonDiscoverable
Pc
+++ iDeviceIo.c
@@ -960,12 +960,23 @@ NonCoherentPciIoFreeBuffer (
LIST_ENTRY *Entry;
EFI_STATUS Status;
NON_DISCOVERABLE_DEVICE_UNCACHED_ALLOCATION *Alloc;
+ NON_DISCOVERABLE_DEVICE_UNCACHED_ALLOCATION *AllocHead;
+ NON_DISCOVERABLE_DEVICE_UNCACHED_ALLOCATION *AllocTail;
BOOLEAN Found;
+ UINTN StartPages;
+ UINTN EndPages;
+
+ if (HostAddress != ALIGN_POINTER (HostAddress, EFI_PAGE_SIZE)) {
+ ASSERT_EFI_ERROR (EFI_INVALID_PARAMETER);
+ return EFI_INVALID_PARAMETER;
+ }

Dev = NON_DISCOVERABLE_PCI_DEVICE_FROM_PCI_IO (This);

Found = FALSE;
Alloc = NULL;
+ AllocHead = NULL;
+ AllocTail = NULL;

//
// Find the uncached allocation list entry associated @@ -976,9
+987,13 @@ NonCoherentPciIoFreeBuffer (
Entry = Entry->ForwardLink)
{
Alloc = BASE_CR (Entry,
NON_DISCOVERABLE_DEVICE_UNCACHED_ALLOCATION, List);
- if ((Alloc->HostAddress == HostAddress) && (Alloc->NumPages ==
Pages))
{
+ StartPages = 0;
+ if (Alloc->HostAddress < HostAddress) {
+ StartPages = (HostAddress - Alloc->HostAddress) / EFI_PAGE_SIZE;
+ }
+ if ((Alloc->HostAddress <= HostAddress) && (Alloc->NumPages >=
+ (Pages + StartPages))) {
//
- // We are freeing the exact allocation we were given
+ // We are freeing at least part of what we were given
// before by AllocateBuffer()
//
Found = TRUE;
@@ -991,7 +1006,41 @@ NonCoherentPciIoFreeBuffer (
return EFI_NOT_FOUND;
}

+ EndPages = Alloc->NumPages - (Pages + StartPages);
+
+ if (StartPages != 0) {
+ AllocHead = AllocatePool (sizeof *AllocHead);
+ if (AllocHead == NULL) {
+ return EFI_OUT_OF_RESOURCES;
+ }
+
+ AllocHead->HostAddress = Alloc->HostAddress;
+ AllocHead->NumPages = StartPages;
+ AllocHead->Attributes = Alloc->Attributes; }
+
+ if (EndPages != 0) {
+ AllocTail = AllocatePool (sizeof *AllocTail);
+ if (AllocTail == NULL) {
+ return EFI_OUT_OF_RESOURCES;
+ }
+
+ AllocTail->HostAddress = Alloc->HostAddress + ((Pages +
+ StartPages) *
EFI_PAGE_SIZE);
+ AllocTail->NumPages = EndPages;
+ AllocTail->Attributes = Alloc->Attributes; }
+
RemoveEntryList (&Alloc->List);
+ //
+ // Record this new sub allocations in the linked list, so we //
+ can restore the memory space attributes later // if (AllocHead !=
+ NULL) {
+ InsertHeadList (&Dev->UncachedAllocationList, &AllocHead->List);
+ } if (AllocTail != NULL) {
+ InsertHeadList (&Dev->UncachedAllocationList, &AllocTail->List);
+ }

Status = gDS->SetMemorySpaceAttributes (
(EFI_PHYSICAL_ADDRESS)(UINTN)HostAddress,
--
2.17.1


Jeff Brasen
 

-----Original Message-----
From: Ard Biesheuvel <ardb@...>
Sent: Friday, July 29, 2022 9:48 AM
To: Jeff Brasen <jbrasen@...>
Cc: devel@edk2.groups.io; hao.a.wu@...; ray.ni@...;
quic_llindhol@...; ardb+tianocore@...
Subject: Re: [PATCH v2] MdeModulePkg/NonDiscoverablePciDeviceDxe:
Allow partial FreeBuffer

External email: Use caution opening links or attachments


On Thu, 28 Jul 2022 at 13:25, Jeff Brasen <jbrasen@...> wrote:


Adding Leif/Ard to CC incase they have any comments on this patch.
This generally looks ok to me. I just wonder if it wouldn't be simpler to reuse
the existing allocation descriptor if it is not being freed entirely. Given the
[presumably] the most common case is to allocate and then free some pages
at the end, lowering the page count on the existing descriptor would cover
most cases, and we'd only need to allocate new ones if pages are being freed
at the start or in the middle.
There is often freeing at the beginning as well as this is being used to create a 64K aligned section of memory in the case. So it over allocates and the free's some at the beginning and the end. I could probably make it detect and use that but figured this code would support all cases and required less case specific detection.

-Jeff



-----Original Message-----
From: Jeff Brasen
Sent: Friday, June 17, 2022 9:39 AM
To: devel@edk2.groups.io
Cc: hao.a.wu@...; ray.ni@...
Subject: RE: [PATCH v2] MdeModulePkg/NonDiscoverablePciDeviceDxe:
Allow partial FreeBuffer

Any thoughts on this patch?

-----Original Message-----
From: Jeff Brasen <jbrasen@...>
Sent: Monday, February 14, 2022 11:46 AM
To: devel@edk2.groups.io
Cc: hao.a.wu@...; ray.ni@...; Jeff Brasen
<jbrasen@...>
Subject: [PATCH v2] MdeModulePkg/NonDiscoverablePciDeviceDxe:
Allow partial FreeBuffer

Add support for partial free of non cached buffers.
If a request for less than the full size is requested new
allocations for the remaining head and tail of the buffer are added to
the list.
Added verification that Buffer is EFI_PAGE_SIZE aligned.
The XHCI driver does this if the page size for the controller is >4KB.

Signed-off-by: Jeff Brasen <jbrasen@...>
---
.../NonDiscoverablePciDeviceIo.c | 53 ++++++++++++++++++-
1 file changed, 51 insertions(+), 2 deletions(-)

diff --git
a/MdeModulePkg/Bus/Pci/NonDiscoverablePciDeviceDxe/NonDiscoverable
PciDeviceIo.c
b/MdeModulePkg/Bus/Pci/NonDiscoverablePciDeviceDxe/NonDiscoverable
PciDeviceIo.c
index c1c5c6267c..77809cfedf 100644
---
a/MdeModulePkg/Bus/Pci/NonDiscoverablePciDeviceDxe/NonDiscoverable
PciDeviceIo.c
+++
b/MdeModulePkg/Bus/Pci/NonDiscoverablePciDeviceDxe/NonDiscoverable
Pc
+++ iDeviceIo.c
@@ -960,12 +960,23 @@ NonCoherentPciIoFreeBuffer (
LIST_ENTRY *Entry;
EFI_STATUS Status;
NON_DISCOVERABLE_DEVICE_UNCACHED_ALLOCATION *Alloc;
+ NON_DISCOVERABLE_DEVICE_UNCACHED_ALLOCATION
*AllocHead;
+ NON_DISCOVERABLE_DEVICE_UNCACHED_ALLOCATION *AllocTail;
BOOLEAN Found;
+ UINTN StartPages;
+ UINTN EndPages;
+
+ if (HostAddress != ALIGN_POINTER (HostAddress, EFI_PAGE_SIZE)) {
+ ASSERT_EFI_ERROR (EFI_INVALID_PARAMETER);
+ return EFI_INVALID_PARAMETER; }

Dev = NON_DISCOVERABLE_PCI_DEVICE_FROM_PCI_IO (This);

Found = FALSE;
Alloc = NULL;
+ AllocHead = NULL;
+ AllocTail = NULL;

//
// Find the uncached allocation list entry associated @@ -976,9
+987,13 @@ NonCoherentPciIoFreeBuffer (
Entry = Entry->ForwardLink)
{
Alloc = BASE_CR (Entry,
NON_DISCOVERABLE_DEVICE_UNCACHED_ALLOCATION, List);
- if ((Alloc->HostAddress == HostAddress) && (Alloc->NumPages ==
Pages))
{
+ StartPages = 0;
+ if (Alloc->HostAddress < HostAddress) {
+ StartPages = (HostAddress - Alloc->HostAddress) / EFI_PAGE_SIZE;
+ }
+ if ((Alloc->HostAddress <= HostAddress) && (Alloc->NumPages
+ >= (Pages + StartPages))) {
//
- // We are freeing the exact allocation we were given
+ // We are freeing at least part of what we were given
// before by AllocateBuffer()
//
Found = TRUE;
@@ -991,7 +1006,41 @@ NonCoherentPciIoFreeBuffer (
return EFI_NOT_FOUND;
}

+ EndPages = Alloc->NumPages - (Pages + StartPages);
+
+ if (StartPages != 0) {
+ AllocHead = AllocatePool (sizeof *AllocHead);
+ if (AllocHead == NULL) {
+ return EFI_OUT_OF_RESOURCES;
+ }
+
+ AllocHead->HostAddress = Alloc->HostAddress;
+ AllocHead->NumPages = StartPages;
+ AllocHead->Attributes = Alloc->Attributes; }
+
+ if (EndPages != 0) {
+ AllocTail = AllocatePool (sizeof *AllocTail);
+ if (AllocTail == NULL) {
+ return EFI_OUT_OF_RESOURCES;
+ }
+
+ AllocTail->HostAddress = Alloc->HostAddress + ((Pages +
+ StartPages) *
EFI_PAGE_SIZE);
+ AllocTail->NumPages = EndPages;
+ AllocTail->Attributes = Alloc->Attributes; }
+
RemoveEntryList (&Alloc->List);
+ //
+ // Record this new sub allocations in the linked list, so we
+ // can restore the memory space attributes later // if
+ (AllocHead !=
+ NULL) {
+ InsertHeadList (&Dev->UncachedAllocationList,
+ &AllocHead->List); } if (AllocTail != NULL) {
+ InsertHeadList (&Dev->UncachedAllocationList,
+ &AllocTail->List); }

Status = gDS->SetMemorySpaceAttributes (
(EFI_PHYSICAL_ADDRESS)(UINTN)HostAddress,
--
2.17.1


Ard Biesheuvel
 

On Tue, 2 Aug 2022 at 17:32, Jeff Brasen <jbrasen@...> wrote:



-----Original Message-----
From: Ard Biesheuvel <ardb@...>
Sent: Friday, July 29, 2022 9:48 AM
To: Jeff Brasen <jbrasen@...>
Cc: devel@edk2.groups.io; hao.a.wu@...; ray.ni@...;
quic_llindhol@...; ardb+tianocore@...
Subject: Re: [PATCH v2] MdeModulePkg/NonDiscoverablePciDeviceDxe:
Allow partial FreeBuffer

External email: Use caution opening links or attachments


On Thu, 28 Jul 2022 at 13:25, Jeff Brasen <jbrasen@...> wrote:


Adding Leif/Ard to CC incase they have any comments on this patch.
This generally looks ok to me. I just wonder if it wouldn't be simpler to reuse
the existing allocation descriptor if it is not being freed entirely. Given the
[presumably] the most common case is to allocate and then free some pages
at the end, lowering the page count on the existing descriptor would cover
most cases, and we'd only need to allocate new ones if pages are being freed
at the start or in the middle.
There is often freeing at the beginning as well as this is being used to create a 64K aligned section of memory in the case. So it over allocates and the free's some at the beginning and the end. I could probably make it detect and use that but figured this code would support all cases and required less case specific detection.
Ah interesting. Would it help if the allocate routine aligned
allocations to their size?


Jeff Brasen
 

-----Original Message-----
From: Ard Biesheuvel <ardb@...>
Sent: Tuesday, August 2, 2022 10:51 AM
To: Jeff Brasen <jbrasen@...>
Cc: devel@edk2.groups.io; hao.a.wu@...; ray.ni@...;
quic_llindhol@...; ardb+tianocore@...
Subject: Re: [PATCH v2] MdeModulePkg/NonDiscoverablePciDeviceDxe:
Allow partial FreeBuffer

External email: Use caution opening links or attachments


On Tue, 2 Aug 2022 at 17:32, Jeff Brasen <jbrasen@...> wrote:



-----Original Message-----
From: Ard Biesheuvel <ardb@...>
Sent: Friday, July 29, 2022 9:48 AM
To: Jeff Brasen <jbrasen@...>
Cc: devel@edk2.groups.io; hao.a.wu@...; ray.ni@...;
quic_llindhol@...; ardb+tianocore@...
Subject: Re: [PATCH v2] MdeModulePkg/NonDiscoverablePciDeviceDxe:
Allow partial FreeBuffer

External email: Use caution opening links or attachments


On Thu, 28 Jul 2022 at 13:25, Jeff Brasen <jbrasen@...> wrote:


Adding Leif/Ard to CC incase they have any comments on this patch.
This generally looks ok to me. I just wonder if it wouldn't be
simpler to reuse the existing allocation descriptor if it is not
being freed entirely. Given the [presumably] the most common case is
to allocate and then free some pages at the end, lowering the page
count on the existing descriptor would cover most cases, and we'd
only need to allocate new ones if pages are being freed at the start or in
the middle.

There is often freeing at the beginning as well as this is being used to create
a 64K aligned section of memory in the case. So it over allocates and the
free's some at the beginning and the end. I could probably make it detect
and use that but figured this code would support all cases and required less
case specific detection.
Ah interesting. Would it help if the allocate routine aligned allocations to their
size?
The PciIo->AllocateBuffer function doesn't support passing the request in so we would need to know that info beforehand. The current calling in the XHCI driver does a free at the beginning and then the end of the buffer so we could the existing allocation tracker but figured it would be better to correct the function just in case someone called it to free in the middle.


Ard Biesheuvel
 

On Fri, 5 Aug 2022 at 18:56, Jeff Brasen via groups.io
<jbrasen@...> wrote:



-----Original Message-----
From: Ard Biesheuvel <ardb@...>
Sent: Tuesday, August 2, 2022 10:51 AM
To: Jeff Brasen <jbrasen@...>
Cc: devel@edk2.groups.io; hao.a.wu@...; ray.ni@...;
quic_llindhol@...; ardb+tianocore@...
Subject: Re: [PATCH v2] MdeModulePkg/NonDiscoverablePciDeviceDxe:
Allow partial FreeBuffer

External email: Use caution opening links or attachments


On Tue, 2 Aug 2022 at 17:32, Jeff Brasen <jbrasen@...> wrote:



-----Original Message-----
From: Ard Biesheuvel <ardb@...>
Sent: Friday, July 29, 2022 9:48 AM
To: Jeff Brasen <jbrasen@...>
Cc: devel@edk2.groups.io; hao.a.wu@...; ray.ni@...;
quic_llindhol@...; ardb+tianocore@...
Subject: Re: [PATCH v2] MdeModulePkg/NonDiscoverablePciDeviceDxe:
Allow partial FreeBuffer

External email: Use caution opening links or attachments


On Thu, 28 Jul 2022 at 13:25, Jeff Brasen <jbrasen@...> wrote:


Adding Leif/Ard to CC incase they have any comments on this patch.
This generally looks ok to me. I just wonder if it wouldn't be
simpler to reuse the existing allocation descriptor if it is not
being freed entirely. Given the [presumably] the most common case is
to allocate and then free some pages at the end, lowering the page
count on the existing descriptor would cover most cases, and we'd
only need to allocate new ones if pages are being freed at the start or in
the middle.

There is often freeing at the beginning as well as this is being used to create
a 64K aligned section of memory in the case. So it over allocates and the
free's some at the beginning and the end. I could probably make it detect
and use that but figured this code would support all cases and required less
case specific detection.
Ah interesting. Would it help if the allocate routine aligned allocations to their
size?
The PciIo->AllocateBuffer function doesn't support passing the request in so we would need to know that info beforehand. The current calling in the XHCI driver does a free at the beginning and then the end of the buffer so we could the existing allocation tracker but figured it would be better to correct the function just in case someone called it to free in the middle.
What I was wondering is whether such allocations are themselves
multiples of 64k. This is perhaps orthogonal to the issue this patch
addresses, as we'' still need to deal with partial free calls
regardless. But I was curious whether XHCI in particular, and perhaps
more generally, we could streamline this by aligning all allocations
to a log2 upper bound of their sizes.


Jeff Brasen
 

-----Original Message-----
From: Ard Biesheuvel <ardb@...>
Sent: Monday, August 15, 2022 8:42 AM
To: devel@edk2.groups.io; Jeff Brasen <jbrasen@...>
Cc: hao.a.wu@...; ray.ni@...; quic_llindhol@...;
ardb+tianocore@...
Subject: Re: [edk2-devel] [PATCH v2]
MdeModulePkg/NonDiscoverablePciDeviceDxe: Allow partial FreeBuffer

External email: Use caution opening links or attachments


On Fri, 5 Aug 2022 at 18:56, Jeff Brasen via groups.io
<jbrasen@...> wrote:



-----Original Message-----
From: Ard Biesheuvel <ardb@...>
Sent: Tuesday, August 2, 2022 10:51 AM
To: Jeff Brasen <jbrasen@...>
Cc: devel@edk2.groups.io; hao.a.wu@...; ray.ni@...;
quic_llindhol@...; ardb+tianocore@...
Subject: Re: [PATCH v2] MdeModulePkg/NonDiscoverablePciDeviceDxe:
Allow partial FreeBuffer

External email: Use caution opening links or attachments


On Tue, 2 Aug 2022 at 17:32, Jeff Brasen <jbrasen@...> wrote:



-----Original Message-----
From: Ard Biesheuvel <ardb@...>
Sent: Friday, July 29, 2022 9:48 AM
To: Jeff Brasen <jbrasen@...>
Cc: devel@edk2.groups.io; hao.a.wu@...; ray.ni@...;
quic_llindhol@...; ardb+tianocore@...
Subject: Re: [PATCH v2] MdeModulePkg/NonDiscoverablePciDeviceDxe:
Allow partial FreeBuffer

External email: Use caution opening links or attachments


On Thu, 28 Jul 2022 at 13:25, Jeff Brasen <jbrasen@...> wrote:


Adding Leif/Ard to CC incase they have any comments on this patch.
This generally looks ok to me. I just wonder if it wouldn't be
simpler to reuse the existing allocation descriptor if it is not
being freed entirely. Given the [presumably] the most common
case is to allocate and then free some pages at the end,
lowering the page count on the existing descriptor would cover
most cases, and we'd only need to allocate new ones if pages are
being freed at the start or in
the middle.

There is often freeing at the beginning as well as this is being
used to create
a 64K aligned section of memory in the case. So it over allocates
and the free's some at the beginning and the end. I could probably
make it detect and use that but figured this code would support all
cases and required less case specific detection.
Ah interesting. Would it help if the allocate routine aligned
allocations to their size?
The PciIo->AllocateBuffer function doesn't support passing the request in so
we would need to know that info beforehand. The current calling in the XHCI
driver does a free at the beginning and then the end of the buffer so we could
the existing allocation tracker but figured it would be better to correct the
function just in case someone called it to free in the middle.
What I was wondering is whether such allocations are themselves multiples of
64k. This is perhaps orthogonal to the issue this patch addresses, as we'' still
need to deal with partial free calls regardless. But I was curious whether XHCI in
particular, and perhaps more generally, we could streamline this by aligning all
allocations to a log2 upper bound of their sizes.
Xhci code (https://github.com/tianocore/edk2/blob/master/MdeModulePkg/Bus/Pci/XhciDxe/UsbHcMem.c#L604) in allocation requested is greater the EFI_PAGE_SIZE allocates number of requested pages plus pages for the alignment and then frees pages at the beginning and end of the allocation. I am not sure we really could change this without adding an alignment field to the PciIo protocol.

Is there anything else you would like to change on this patch?

Thanks,
Jeff


Ard Biesheuvel
 

On Thu, 8 Sept 2022 at 17:39, Jeff Brasen <jbrasen@...> wrote:



-----Original Message-----
From: Ard Biesheuvel <ardb@...>
Sent: Monday, August 15, 2022 8:42 AM
To: devel@edk2.groups.io; Jeff Brasen <jbrasen@...>
Cc: hao.a.wu@...; ray.ni@...; quic_llindhol@...;
ardb+tianocore@...
Subject: Re: [edk2-devel] [PATCH v2]
MdeModulePkg/NonDiscoverablePciDeviceDxe: Allow partial FreeBuffer

External email: Use caution opening links or attachments


On Fri, 5 Aug 2022 at 18:56, Jeff Brasen via groups.io
<jbrasen@...> wrote:



-----Original Message-----
From: Ard Biesheuvel <ardb@...>
Sent: Tuesday, August 2, 2022 10:51 AM
To: Jeff Brasen <jbrasen@...>
Cc: devel@edk2.groups.io; hao.a.wu@...; ray.ni@...;
quic_llindhol@...; ardb+tianocore@...
Subject: Re: [PATCH v2] MdeModulePkg/NonDiscoverablePciDeviceDxe:
Allow partial FreeBuffer

External email: Use caution opening links or attachments


On Tue, 2 Aug 2022 at 17:32, Jeff Brasen <jbrasen@...> wrote:



-----Original Message-----
From: Ard Biesheuvel <ardb@...>
Sent: Friday, July 29, 2022 9:48 AM
To: Jeff Brasen <jbrasen@...>
Cc: devel@edk2.groups.io; hao.a.wu@...; ray.ni@...;
quic_llindhol@...; ardb+tianocore@...
Subject: Re: [PATCH v2] MdeModulePkg/NonDiscoverablePciDeviceDxe:
Allow partial FreeBuffer

External email: Use caution opening links or attachments


On Thu, 28 Jul 2022 at 13:25, Jeff Brasen <jbrasen@...> wrote:


Adding Leif/Ard to CC incase they have any comments on this patch.
This generally looks ok to me. I just wonder if it wouldn't be
simpler to reuse the existing allocation descriptor if it is not
being freed entirely. Given the [presumably] the most common
case is to allocate and then free some pages at the end,
lowering the page count on the existing descriptor would cover
most cases, and we'd only need to allocate new ones if pages are
being freed at the start or in
the middle.

There is often freeing at the beginning as well as this is being
used to create
a 64K aligned section of memory in the case. So it over allocates
and the free's some at the beginning and the end. I could probably
make it detect and use that but figured this code would support all
cases and required less case specific detection.
Ah interesting. Would it help if the allocate routine aligned
allocations to their size?
The PciIo->AllocateBuffer function doesn't support passing the request in so
we would need to know that info beforehand. The current calling in the XHCI
driver does a free at the beginning and then the end of the buffer so we could
the existing allocation tracker but figured it would be better to correct the
function just in case someone called it to free in the middle.
What I was wondering is whether such allocations are themselves multiples of
64k. This is perhaps orthogonal to the issue this patch addresses, as we'' still
need to deal with partial free calls regardless. But I was curious whether XHCI in
particular, and perhaps more generally, we could streamline this by aligning all
allocations to a log2 upper bound of their sizes.
Xhci code (https://github.com/tianocore/edk2/blob/master/MdeModulePkg/Bus/Pci/XhciDxe/UsbHcMem.c#L604) in allocation requested is greater the EFI_PAGE_SIZE allocates number of requested pages plus pages for the alignment and then frees pages at the beginning and end of the allocation. I am not sure we really could change this without adding an alignment field to the PciIo protocol.

Is there anything else you would like to change on this patch?
No. Thanks for the clarification.

Reviewed-by: Ard Biesheuvel <ardb@...>


Jeff Brasen
 

Anything else needed to get this merged?

-----Original Message-----
From: Ard Biesheuvel <ardb@...>
Sent: Thursday, September 8, 2022 9:55 AM
To: Jeff Brasen <jbrasen@...>
Cc: devel@edk2.groups.io; hao.a.wu@...; ray.ni@...;
quic_llindhol@...; ardb+tianocore@...
Subject: Re: [edk2-devel] [PATCH v2]
MdeModulePkg/NonDiscoverablePciDeviceDxe: Allow partial FreeBuffer

External email: Use caution opening links or attachments


On Thu, 8 Sept 2022 at 17:39, Jeff Brasen <jbrasen@...> wrote:



-----Original Message-----
From: Ard Biesheuvel <ardb@...>
Sent: Monday, August 15, 2022 8:42 AM
To: devel@edk2.groups.io; Jeff Brasen <jbrasen@...>
Cc: hao.a.wu@...; ray.ni@...; quic_llindhol@...;
ardb+tianocore@...
Subject: Re: [edk2-devel] [PATCH v2]
MdeModulePkg/NonDiscoverablePciDeviceDxe: Allow partial FreeBuffer

External email: Use caution opening links or attachments


On Fri, 5 Aug 2022 at 18:56, Jeff Brasen via groups.io
<jbrasen@...> wrote:



-----Original Message-----
From: Ard Biesheuvel <ardb@...>
Sent: Tuesday, August 2, 2022 10:51 AM
To: Jeff Brasen <jbrasen@...>
Cc: devel@edk2.groups.io; hao.a.wu@...; ray.ni@...;
quic_llindhol@...; ardb+tianocore@...
Subject: Re: [PATCH v2] MdeModulePkg/NonDiscoverablePciDeviceDxe:
Allow partial FreeBuffer

External email: Use caution opening links or attachments


On Tue, 2 Aug 2022 at 17:32, Jeff Brasen <jbrasen@...> wrote:



-----Original Message-----
From: Ard Biesheuvel <ardb@...>
Sent: Friday, July 29, 2022 9:48 AM
To: Jeff Brasen <jbrasen@...>
Cc: devel@edk2.groups.io; hao.a.wu@...;
ray.ni@...; quic_llindhol@...;
ardb+tianocore@...
Subject: Re: [PATCH v2]
MdeModulePkg/NonDiscoverablePciDeviceDxe:
Allow partial FreeBuffer

External email: Use caution opening links or attachments


On Thu, 28 Jul 2022 at 13:25, Jeff Brasen <jbrasen@...>
wrote:


Adding Leif/Ard to CC incase they have any comments on this patch.
This generally looks ok to me. I just wonder if it wouldn't
be simpler to reuse the existing allocation descriptor if it
is not being freed entirely. Given the [presumably] the most
common case is to allocate and then free some pages at the
end, lowering the page count on the existing descriptor
would cover most cases, and we'd only need to allocate new
ones if pages are being freed at the start or in
the middle.

There is often freeing at the beginning as well as this is
being used to create
a 64K aligned section of memory in the case. So it over
allocates and the free's some at the beginning and the end. I
could probably make it detect and use that but figured this code
would support all cases and required less case specific detection.
Ah interesting. Would it help if the allocate routine aligned
allocations to their size?
The PciIo->AllocateBuffer function doesn't support passing the
request in so
we would need to know that info beforehand. The current calling in
the XHCI driver does a free at the beginning and then the end of the
buffer so we could the existing allocation tracker but figured it
would be better to correct the function just in case someone called it to free
in the middle.
What I was wondering is whether such allocations are themselves
multiples of 64k. This is perhaps orthogonal to the issue this patch
addresses, as we'' still need to deal with partial free calls
regardless. But I was curious whether XHCI in particular, and
perhaps more generally, we could streamline this by aligning all allocations
to a log2 upper bound of their sizes.

Xhci code
(https://github.com/tianocore/edk2/blob/master/MdeModulePkg/Bus/Pci/Xhci
Dxe/UsbHcMem.c#L604) in allocation requested is greater the EFI_PAGE_SIZE
allocates number of requested pages plus pages for the alignment and then
frees pages at the beginning and end of the allocation. I am not sure we really
could change this without adding an alignment field to the PciIo protocol.

Is there anything else you would like to change on this patch?
No. Thanks for the clarification.

Reviewed-by: Ard Biesheuvel <ardb@...>


Ard Biesheuvel
 

On Wed, 21 Sept 2022 at 18:27, Jeff Brasen via groups.io
<jbrasen@...> wrote:

Anything else needed to get this merged?
That is up to the MdeModulePkg maintainers.

-----Original Message-----
From: Ard Biesheuvel <ardb@...>
Sent: Thursday, September 8, 2022 9:55 AM
To: Jeff Brasen <jbrasen@...>
Cc: devel@edk2.groups.io; hao.a.wu@...; ray.ni@...;
quic_llindhol@...; ardb+tianocore@...
Subject: Re: [edk2-devel] [PATCH v2]
MdeModulePkg/NonDiscoverablePciDeviceDxe: Allow partial FreeBuffer

External email: Use caution opening links or attachments


On Thu, 8 Sept 2022 at 17:39, Jeff Brasen <jbrasen@...> wrote:



-----Original Message-----
From: Ard Biesheuvel <ardb@...>
Sent: Monday, August 15, 2022 8:42 AM
To: devel@edk2.groups.io; Jeff Brasen <jbrasen@...>
Cc: hao.a.wu@...; ray.ni@...; quic_llindhol@...;
ardb+tianocore@...
Subject: Re: [edk2-devel] [PATCH v2]
MdeModulePkg/NonDiscoverablePciDeviceDxe: Allow partial FreeBuffer

External email: Use caution opening links or attachments


On Fri, 5 Aug 2022 at 18:56, Jeff Brasen via groups.io
<jbrasen@...> wrote:



-----Original Message-----
From: Ard Biesheuvel <ardb@...>
Sent: Tuesday, August 2, 2022 10:51 AM
To: Jeff Brasen <jbrasen@...>
Cc: devel@edk2.groups.io; hao.a.wu@...; ray.ni@...;
quic_llindhol@...; ardb+tianocore@...
Subject: Re: [PATCH v2] MdeModulePkg/NonDiscoverablePciDeviceDxe:
Allow partial FreeBuffer

External email: Use caution opening links or attachments


On Tue, 2 Aug 2022 at 17:32, Jeff Brasen <jbrasen@...> wrote:



-----Original Message-----
From: Ard Biesheuvel <ardb@...>
Sent: Friday, July 29, 2022 9:48 AM
To: Jeff Brasen <jbrasen@...>
Cc: devel@edk2.groups.io; hao.a.wu@...;
ray.ni@...; quic_llindhol@...;
ardb+tianocore@...
Subject: Re: [PATCH v2]
MdeModulePkg/NonDiscoverablePciDeviceDxe:
Allow partial FreeBuffer

External email: Use caution opening links or attachments


On Thu, 28 Jul 2022 at 13:25, Jeff Brasen <jbrasen@...>
wrote:


Adding Leif/Ard to CC incase they have any comments on this patch.
This generally looks ok to me. I just wonder if it wouldn't
be simpler to reuse the existing allocation descriptor if it
is not being freed entirely. Given the [presumably] the most
common case is to allocate and then free some pages at the
end, lowering the page count on the existing descriptor
would cover most cases, and we'd only need to allocate new
ones if pages are being freed at the start or in
the middle.

There is often freeing at the beginning as well as this is
being used to create
a 64K aligned section of memory in the case. So it over
allocates and the free's some at the beginning and the end. I
could probably make it detect and use that but figured this code
would support all cases and required less case specific detection.
Ah interesting. Would it help if the allocate routine aligned
allocations to their size?
The PciIo->AllocateBuffer function doesn't support passing the
request in so
we would need to know that info beforehand. The current calling in
the XHCI driver does a free at the beginning and then the end of the
buffer so we could the existing allocation tracker but figured it
would be better to correct the function just in case someone called it to free
in the middle.
What I was wondering is whether such allocations are themselves
multiples of 64k. This is perhaps orthogonal to the issue this patch
addresses, as we'' still need to deal with partial free calls
regardless. But I was curious whether XHCI in particular, and
perhaps more generally, we could streamline this by aligning all allocations
to a log2 upper bound of their sizes.

Xhci code
(https://github.com/tianocore/edk2/blob/master/MdeModulePkg/Bus/Pci/Xhci
Dxe/UsbHcMem.c#L604) in allocation requested is greater the EFI_PAGE_SIZE
allocates number of requested pages plus pages for the alignment and then
frees pages at the beginning and end of the allocation. I am not sure we really
could change this without adding an alignment field to the PciIo protocol.

Is there anything else you would like to change on this patch?
No. Thanks for the clarification.

Reviewed-by: Ard Biesheuvel <ardb@...>




Jeff Brasen
 

MdeModulePkg maintainers, Any comments on this?

-----Original Message-----
From: Ard Biesheuvel <ardb@...>
Sent: Wednesday, September 21, 2022 10:32 AM
To: devel@edk2.groups.io; Jeff Brasen <jbrasen@...>
Cc: hao.a.wu@...; ray.ni@...; quic_llindhol@...;
ardb+tianocore@...
Subject: Re: [edk2-devel] [PATCH v2]
MdeModulePkg/NonDiscoverablePciDeviceDxe: Allow partial FreeBuffer

External email: Use caution opening links or attachments


On Wed, 21 Sept 2022 at 18:27, Jeff Brasen via groups.io
<jbrasen@...> wrote:

Anything else needed to get this merged?
That is up to the MdeModulePkg maintainers.

-----Original Message-----
From: Ard Biesheuvel <ardb@...>
Sent: Thursday, September 8, 2022 9:55 AM
To: Jeff Brasen <jbrasen@...>
Cc: devel@edk2.groups.io; hao.a.wu@...; ray.ni@...;
quic_llindhol@...; ardb+tianocore@...
Subject: Re: [edk2-devel] [PATCH v2]
MdeModulePkg/NonDiscoverablePciDeviceDxe: Allow partial FreeBuffer

External email: Use caution opening links or attachments


On Thu, 8 Sept 2022 at 17:39, Jeff Brasen <jbrasen@...> wrote:



-----Original Message-----
From: Ard Biesheuvel <ardb@...>
Sent: Monday, August 15, 2022 8:42 AM
To: devel@edk2.groups.io; Jeff Brasen <jbrasen@...>
Cc: hao.a.wu@...; ray.ni@...;
quic_llindhol@...;
ardb+tianocore@...
Subject: Re: [edk2-devel] [PATCH v2]
MdeModulePkg/NonDiscoverablePciDeviceDxe: Allow partial
FreeBuffer

External email: Use caution opening links or attachments


On Fri, 5 Aug 2022 at 18:56, Jeff Brasen via groups.io
<jbrasen@...> wrote:



-----Original Message-----
From: Ard Biesheuvel <ardb@...>
Sent: Tuesday, August 2, 2022 10:51 AM
To: Jeff Brasen <jbrasen@...>
Cc: devel@edk2.groups.io; hao.a.wu@...;
ray.ni@...; quic_llindhol@...;
ardb+tianocore@...
Subject: Re: [PATCH v2]
MdeModulePkg/NonDiscoverablePciDeviceDxe:
Allow partial FreeBuffer

External email: Use caution opening links or attachments


On Tue, 2 Aug 2022 at 17:32, Jeff Brasen <jbrasen@...>
wrote:



-----Original Message-----
From: Ard Biesheuvel <ardb@...>
Sent: Friday, July 29, 2022 9:48 AM
To: Jeff Brasen <jbrasen@...>
Cc: devel@edk2.groups.io; hao.a.wu@...;
ray.ni@...; quic_llindhol@...;
ardb+tianocore@...
Subject: Re: [PATCH v2]
MdeModulePkg/NonDiscoverablePciDeviceDxe:
Allow partial FreeBuffer

External email: Use caution opening links or attachments


On Thu, 28 Jul 2022 at 13:25, Jeff Brasen
<jbrasen@...>
wrote:


Adding Leif/Ard to CC incase they have any comments on
this patch.
This generally looks ok to me. I just wonder if it
wouldn't be simpler to reuse the existing allocation
descriptor if it is not being freed entirely. Given the
[presumably] the most common case is to allocate and
then free some pages at the end, lowering the page count
on the existing descriptor would cover most cases, and
we'd only need to allocate new ones if pages are being
freed at the start or in
the middle.

There is often freeing at the beginning as well as this is
being used to create
a 64K aligned section of memory in the case. So it over
allocates and the free's some at the beginning and the end.
I could probably make it detect and use that but figured
this code would support all cases and required less case specific
detection.
Ah interesting. Would it help if the allocate routine
aligned allocations to their size?
The PciIo->AllocateBuffer function doesn't support passing the
request in so
we would need to know that info beforehand. The current calling
in the XHCI driver does a free at the beginning and then the end
of the buffer so we could the existing allocation tracker but
figured it would be better to correct the function just in case
someone called it to free
in the middle.
What I was wondering is whether such allocations are themselves
multiples of 64k. This is perhaps orthogonal to the issue this
patch addresses, as we'' still need to deal with partial free
calls regardless. But I was curious whether XHCI in particular,
and perhaps more generally, we could streamline this by aligning
all allocations
to a log2 upper bound of their sizes.

Xhci code
(https://github.com/tianocore/edk2/blob/master/MdeModulePkg/Bus/Pci/
Xhci
Dxe/UsbHcMem.c#L604) in allocation requested is greater the
EFI_PAGE_SIZE allocates number of requested pages plus pages for the
alignment and then frees pages at the beginning and end of the
allocation. I am not sure we really could change this without adding an
alignment field to the PciIo protocol.

Is there anything else you would like to change on this patch?
No. Thanks for the clarification.

Reviewed-by: Ard Biesheuvel <ardb@...>