[edk2-devel] [PATCH v4 12/13] OvmfPkg/MptScsiDxe: Report multiple targets
Laszlo Ersek
lersek at redhat.com
Mon Apr 20 18:31:51 UTC 2020
On 04/14/20 19:38, Nikita Leshenko wrote:
> The controller supports up to 8 targets (Not reported by the
> controller, but based on the implementation of the virtual device),
> report them in GetNextTarget and GetNextTargetLun. The firmware will
> then try to communicate with them and create a block device for each
> one that responds.
>
> Support for multiple LUNs will be implemented in another series.
>
> Ref: https://bugzilla.tianocore.org/show_bug.cgi?id=2390
> Signed-off-by: Nikita Leshenko <nikita.leshchenko at oracle.com>
> Reviewed-by: Liran Alon <liran.alon at oracle.com>
> ---
> OvmfPkg/MptScsiDxe/MptScsi.c | 38 ++++++++++++++++++++++++-------
> OvmfPkg/MptScsiDxe/MptScsiDxe.inf | 1 +
> OvmfPkg/OvmfPkg.dec | 4 ++++
> 3 files changed, 35 insertions(+), 8 deletions(-)
>
> diff --git a/OvmfPkg/MptScsiDxe/MptScsi.c b/OvmfPkg/MptScsiDxe/MptScsi.c
> index fcdaa4c338a4..3ea08857df5d 100644
> --- a/OvmfPkg/MptScsiDxe/MptScsi.c
> +++ b/OvmfPkg/MptScsiDxe/MptScsi.c
> @@ -46,6 +46,7 @@ typedef struct {
> EFI_PCI_IO_PROTOCOL *PciIo;
> UINT64 OriginalPciAttributes;
> UINT32 StallPerPollUsec;
> + UINT8 MaxTarget;
> MPT_SCSI_DMA_BUFFER *Dma;
> EFI_PHYSICAL_ADDRESS DmaPhysical;
> VOID *DmaMapping;
> @@ -159,6 +160,7 @@ MptScsiInit (
> UINT32 ReplyWord;
>
> Dev->StallPerPollUsec = PcdGet32 (PcdMptScsiStallPerPollUsec);
> + Dev->MaxTarget = PcdGet8 (PcdMptScsiMaxTargetLimit);
>
> Status = MptScsiReset (Dev);
> if (EFI_ERROR (Status)) {
> @@ -169,7 +171,7 @@ MptScsiInit (
> ZeroMem (&Reply, sizeof (Reply));
> Req.Data.WhoInit = MPT_IOC_WHOINIT_ROM_BIOS;
> Req.Data.Function = MPT_MESSAGE_HDR_FUNCTION_IOC_INIT;
> - Req.Data.MaxDevices = 1;
> + Req.Data.MaxDevices = Dev->MaxTarget + 1;
(1) If Dev->MaxTarget is 255, then the RHS will evaluate to 256. But the
LHS can only fit a UINT8.
Can you introduce the PCD in the INF file under a [FixedPcd] section,
and add:
STATIC_ASSERT (FixedPcdGet8 (...) < 255, "...");
to MptScsiInit()?
(I'm unsure (and I'm too tired to check) whether FixedPcdGet8() will
expand to an integer constant. If it doesn't, then STATIC_ASSERT() won't
work. In that case we should likely use a plain ASSERT().)
> Req.Data.MaxBuses = 1;
> Req.Data.ReplyFrameSize = sizeof (MPT_SCSI_IO_REPLY);
>
> @@ -240,7 +242,7 @@ MptScsiPopulateRequest (
> return EFI_UNSUPPORTED;
> }
>
> - if (Target > 0 || Lun > 0 ||
> + if (Target > Dev->MaxTarget || Lun > 0 ||
> Packet->DataDirection > EFI_EXT_SCSI_DATA_DIRECTION_BIDIRECTIONAL ||
> //
> // Trying to receive, but destination pointer is NULL, or contradicting
> @@ -552,12 +554,22 @@ MptScsiGetNextTargetLun (
> IN OUT UINT64 *Lun
> )
> {
> + MPT_SCSI_DEV *Dev;
> +
> + Dev = MPT_SCSI_FROM_PASS_THRU (This);
> //
> - // Currently support only target 0 LUN 0, so hardcode it
> + // Currently support only LUN 0, so hardcode it
> //
> if (!IsTargetInitialized (*Target)) {
> ZeroMem (*Target, TARGET_MAX_BYTES);
> *Lun = 0;
> + } else if (**Target < Dev->MaxTarget) {
> + //
> + // This device support 256 targets only, so it's enough to increment
> + // the LSB of Target, as it will never overflow.
> + //
> + **Target += 1;
> + *Lun = 0;
> } else {
> return EFI_NOT_FOUND;
> }
(2) Please implement the EFI_INVALID_PARAMETER branch as well, as seen
in PvScsiGetNextTargetLun(). It's a check on input:
if (**Target > Dev->MaxTarget || *Lun > 0) {
return EFI_INVALID_PARAMETER;
}
And then, as a consequence, the "*Lun = 0" assignment is no longer
necessary, near the target increment. (Until you implement multi-LUN.)
To confirm, (Target==MaxTarget) on input should return EFI_NOT_FOUND,
while (Target>MaxTarget) on input should return EFI_INVALID_PARAMETER.
> @@ -573,11 +585,17 @@ MptScsiGetNextTarget (
> IN OUT UINT8 **Target
> )
> {
> - //
> - // Currently support only target 0 LUN 0, so hardcode it
> - //
> + MPT_SCSI_DEV *Dev;
> +
> + Dev = MPT_SCSI_FROM_PASS_THRU (This);
> if (!IsTargetInitialized (*Target)) {
> ZeroMem (*Target, TARGET_MAX_BYTES);
> + } else if (**Target < Dev->MaxTarget) {
> + //
> + // This device support 256 targets only, so it's enough to increment
> + // the LSB of Target, as it will never overflow.
> + //
> + **Target += 1;
> } else {
> return EFI_NOT_FOUND;
> }
(3) Same as (2).
Hmmm. Maybe I should have pointed these out under patch#6
("OvmfPkg/MptScsiDxe: Report one Target and one LUN").
Because, even while we only support Target=0, the EFI_NOT_FOUND and
EFI_INVALID_PARAMETER return values of GetNextTarget() and
GetNextTargetLun(), could be distinguished.
Sorry for missing this under patch#6.
> @@ -595,6 +613,7 @@ MptScsiBuildDevicePath (
> IN OUT EFI_DEVICE_PATH_PROTOCOL **DevicePath
> )
> {
> + MPT_SCSI_DEV *Dev;
> SCSI_DEVICE_PATH *ScsiDevicePath;
>
> if (DevicePath == NULL) {
> @@ -605,7 +624,8 @@ MptScsiBuildDevicePath (
> // This device support 256 targets only, so it's enough to dereference
> // the LSB of Target.
> //
> - if (*Target > 0 || Lun > 0) {
> + Dev = MPT_SCSI_FROM_PASS_THRU (This);
> + if (*Target > Dev->MaxTarget || Lun > 0) {
> return EFI_NOT_FOUND;
> }
>
> @@ -635,6 +655,7 @@ MptScsiGetTargetLun (
> OUT UINT64 *Lun
> )
> {
> + MPT_SCSI_DEV *Dev;
> SCSI_DEVICE_PATH *ScsiDevicePath;
>
> if (DevicePath == NULL ||
> @@ -647,8 +668,9 @@ MptScsiGetTargetLun (
> return EFI_UNSUPPORTED;
> }
>
> + Dev = MPT_SCSI_FROM_PASS_THRU (This);
> ScsiDevicePath = (SCSI_DEVICE_PATH *)DevicePath;
> - if (ScsiDevicePath->Pun > 0 ||
> + if (ScsiDevicePath->Pun > Dev->MaxTarget ||
> ScsiDevicePath->Lun > 0) {
> return EFI_NOT_FOUND;
> }
> diff --git a/OvmfPkg/MptScsiDxe/MptScsiDxe.inf b/OvmfPkg/MptScsiDxe/MptScsiDxe.inf
> index ef1f6a5ebb3a..26aca7f95315 100644
> --- a/OvmfPkg/MptScsiDxe/MptScsiDxe.inf
> +++ b/OvmfPkg/MptScsiDxe/MptScsiDxe.inf
> @@ -37,3 +37,4 @@ [Protocols]
>
> [Pcd]
> gUefiOvmfPkgTokenSpaceGuid.PcdMptScsiStallPerPollUsec ## CONSUMES
> + gUefiOvmfPkgTokenSpaceGuid.PcdMptScsiMaxTargetLimit ## CONSUMES
(4) Please keep the PCD list alphabetically sorted, and also keep the
"## ... " comments lined up.
> diff --git a/OvmfPkg/OvmfPkg.dec b/OvmfPkg/OvmfPkg.dec
> index 7fa1581f2101..7b56998abc9b 100644
> --- a/OvmfPkg/OvmfPkg.dec
> +++ b/OvmfPkg/OvmfPkg.dec
> @@ -273,6 +273,10 @@ [PcdsFixedAtBuild]
> ## Microseconds to stall between polling for MptScsi request result
> gUefiOvmfPkgTokenSpaceGuid.PcdMptScsiStallPerPollUsec|5|UINT32|0x39
>
> + ## Set the *inclusive* number of targets that MptScsi exposes for scan
> + # by ScsiBusDxe.
> + gUefiOvmfPkgTokenSpaceGuid.PcdMptScsiMaxTargetLimit|7|UINT8|0x40
> +
(5) Please keep this near the PcdPvScsi* group of PCDs.
> [PcdsDynamic, PcdsDynamicEx]
> gUefiOvmfPkgTokenSpaceGuid.PcdEmuVariableEvent|0|UINT64|2
> gUefiOvmfPkgTokenSpaceGuid.PcdOvmfFlashVariablesEnable|FALSE|BOOLEAN|0x10
>
Thanks
Laszlo
-=-=-=-=-=-=-=-=-=-=-=-
Groups.io Links: You receive all messages sent to this group.
View/Reply Online (#57652): https://edk2.groups.io/g/devel/message/57652
Mute This Topic: https://groups.io/mt/73015393/1813853
Group Owner: devel+owner at edk2.groups.io
Unsubscribe: https://edk2.groups.io/g/devel/unsub [edk2-devel-archive at redhat.com]
-=-=-=-=-=-=-=-=-=-=-=-
More information about the edk2-devel-archive
mailing list