On Mon, Oct 14, 2013 at 1:21 AM, Paul Menzel <paulepanter@users.sourceforge.net> wrote:
Dear Evgeny,


thank you for the patch iterations!

A minor nitpick, I prefer having a verb in the commit summary.

    Add pvscsi boot support

Also you could add your reply (the things you tested on) to the commit
message.


done

Am Sonntag, den 13.10.2013, 20:07 +0300 schrieb Evgeny Budilovsky:
> Signed-off-by: Evgeny Budilovsky <evgeny.budilovsky@ravellosystems.com>
> ---
>  Makefile          |    2 +-
>  src/Kconfig       |   12 ++
>  src/block.c       |    1 +
>  src/block.h       |    1 +
>  src/hw/blockcmd.c |    3 +
>  src/hw/pci_ids.h  |    3 +
>  src/hw/pvscsi.c   |  365 +++++++++++++++++++++++++++++++++++++++++++++++++++++
>  src/hw/pvscsi.h   |    8 ++
>  src/post.c        |    2 +
>  9 files changed, 396 insertions(+), 1 deletion(-)
>  create mode 100644 src/hw/pvscsi.c
>  create mode 100644 src/hw/pvscsi.h
>
> diff --git a/Makefile b/Makefile
> index 3218970..6c6302e 100644
> --- a/Makefile
> +++ b/Makefile
> @@ -34,7 +34,7 @@ SRCBOTH=misc.c stacks.c output.c string.c x86.c block.c cdrom.c mouse.c kbd.c \
>      hw/usb-hid.c hw/usb-msc.c hw/usb-uas.c \
>      hw/blockcmd.c hw/floppy.c hw/ata.c hw/ahci.c hw/ramdisk.c \
>      hw/virtio-ring.c hw/virtio-pci.c hw/virtio-blk.c hw/virtio-scsi.c \
> -    hw/lsi-scsi.c hw/esp-scsi.c hw/megasas.c
> +    hw/lsi-scsi.c hw/esp-scsi.c hw/megasas.c hw/pvscsi.c
>  SRC16=$(SRCBOTH) system.c disk.c font.c
>  SRC32FLAT=$(SRCBOTH) post.c memmap.c malloc.c pmm.c romfile.c optionroms.c \
>      boot.c bootsplash.c jpeg.c bmp.c \
> diff --git a/src/Kconfig b/src/Kconfig
> index c40cc61..6e6814e 100644
> --- a/src/Kconfig
> +++ b/src/Kconfig
> @@ -161,6 +161,18 @@ menu "Hardware support"
>          default y
>          help
>              Support boot from virtio-scsi storage.
> +    config PVSCSI
> +        depends on DRIVES && QEMU_HARDWARE
> +        bool "pvscsi controllers"

Should it be capitalizied?

done  
> +        default y
> +        help
> +            Support boot from Paravirtualized SCSI storage. This kind of storage
> +            is mainly supported by VMware ESX hypervisor. It is commonly used
> +            to allow fast storage access by communicating directly with the
> +            underlying hypervisor. Enabling this type of boot will allow
> +            booting directly from images that were imported from ESX platform,

»ESX platform*s*« or »from an ESX platform«? (I am no native speaker
though.)

To make it shorter remove the »that were«: … imported from images
imported from … (as above).

done
 
> +            without the need to use slower emulation of storage controllers
> +            such as IDE.


[…]


> diff --git a/src/hw/pvscsi.c b/src/hw/pvscsi.c
> new file mode 100644
> index 0000000..df5faeb
> --- /dev/null
> +++ b/src/hw/pvscsi.c
> @@ -0,0 +1,365 @@
> +// QEMU VMWARE Paravirtualized SCSI boot support.
> +//
> +// Copyright (c) 2013 Ravello Systems LTD (http://ravellosystems.com)
> +//
> +// Authors:
> +//  Evgeny Budilovsky <evgeny.budilovsky@ravellosystems.com>
> +//
> +// This file may be distributed under the terms of the GNU LGPLv3 license.
> +
> +#include "biosvar.h" // GET_GLOBAL
> +#include "block.h" // struct drive_s
> +#include "blockcmd.h" // scsi_drive_setup
> +#include "config.h" // CONFIG_*
> +#include "malloc.h" // free
> +#include "output.h" // dprintf
> +#include "pci.h" // foreachpci
> +#include "pci_ids.h" // PCI_DEVICE_ID_VMWARE_PVSCSI
> +#include "pci_regs.h" // PCI_VENDOR_ID
> +#include "std/disk.h" // DISK_RET_SUCCESS
> +#include "string.h" // memset
> +#include "util.h" // usleep
> +#include "pvscsi.h"
> +#include "virtio-ring.h" // PAGE_SHIFT, virt_to_phys
> +
> +#define MASK(n) ((1 << (n)) - 1)
> +
> +#define SIMPLE_QUEUE_TAG 0x20
> +
> +#define PVSCSI_INTR_CMPL_0                 (1 << 0)
> +#define PVSCSI_INTR_CMPL_1                 (1 << 1)
> +#define PVSCSI_INTR_CMPL_MASK              MASK(2)
> +
> +#define PVSCSI_INTR_MSG_0                  (1 << 2)
> +#define PVSCSI_INTR_MSG_1                  (1 << 3)
> +#define PVSCSI_INTR_MSG_MASK               (MASK(2) << 2)
> +#define PVSCSI_INTR_ALL_SUPPORTED          MASK(4)
> +
> +#define PVSCSI_FLAG_CMD_WITH_SG_LIST       (1 << 0)
> +#define PVSCSI_FLAG_CMD_OUT_OF_BAND_CDB    (1 << 1)
> +#define PVSCSI_FLAG_CMD_DIR_NONE           (1 << 2)
> +#define PVSCSI_FLAG_CMD_DIR_TOHOST         (1 << 3)
> +#define PVSCSI_FLAG_CMD_DIR_TODEVICE       (1 << 4)
> +
> +enum PVSCSIRegOffset {
> +    PVSCSI_REG_OFFSET_COMMAND        =    0x0,
> +    PVSCSI_REG_OFFSET_COMMAND_DATA   =    0x4,
> +    PVSCSI_REG_OFFSET_COMMAND_STATUS =    0x8,
> +    PVSCSI_REG_OFFSET_LAST_STS_0     =  0x100,
> +    PVSCSI_REG_OFFSET_LAST_STS_1     =  0x104,
> +    PVSCSI_REG_OFFSET_LAST_STS_2     =  0x108,
> +    PVSCSI_REG_OFFSET_LAST_STS_3     =  0x10c,
> +    PVSCSI_REG_OFFSET_INTR_STATUS    = 0x100c,
> +    PVSCSI_REG_OFFSET_INTR_MASK      = 0x2010,
> +    PVSCSI_REG_OFFSET_KICK_NON_RW_IO = 0x3014,
> +    PVSCSI_REG_OFFSET_DEBUG          = 0x3018,
> +    PVSCSI_REG_OFFSET_KICK_RW_IO     = 0x4018,
> +};
> +
> +enum PVSCSICommands {
> +    PVSCSI_CMD_FIRST             = 0,
> +    PVSCSI_CMD_ADAPTER_RESET     = 1,
> +    PVSCSI_CMD_ISSUE_SCSI        = 2,
> +    PVSCSI_CMD_SETUP_RINGS       = 3,
> +    PVSCSI_CMD_RESET_BUS         = 4,
> +    PVSCSI_CMD_RESET_DEVICE      = 5,
> +    PVSCSI_CMD_ABORT_CMD         = 6,
> +    PVSCSI_CMD_CONFIG            = 7,
> +    PVSCSI_CMD_SETUP_MSG_RING    = 8,
> +    PVSCSI_CMD_DEVICE_UNPLUG     = 9,
> +    PVSCSI_CMD_LAST              = 10
> +};
> +
> +#define PVSCSI_SETUP_RINGS_MAX_NUM_PAGES        32
> +struct PVSCSICmdDescSetupRings {
> +    u32    reqRingNumPages;
> +    u32    cmpRingNumPages;
> +    u64    ringsStatePPN;
> +    u64    reqRingPPNs[PVSCSI_SETUP_RINGS_MAX_NUM_PAGES];
> +    u64    cmpRingPPNs[PVSCSI_SETUP_RINGS_MAX_NUM_PAGES];
> +} PACKED;
> +
> +struct PVSCSIRingCmpDesc {
> +    u64    context;
> +    u64    dataLen;
> +    u32    senseLen;
> +    u16    hostStatus;
> +    u16    scsiStatus;
> +    u32    pad[2];
> +} PACKED;
> +
> +struct PVSCSIRingsState {
> +    u32    reqProdIdx;
> +    u32    reqConsIdx;
> +    u32    reqNumEntriesLog2;
> +
> +    u32    cmpProdIdx;
> +    u32    cmpConsIdx;
> +    u32    cmpNumEntriesLog2;
> +
> +    u8     pad[104];
> +
> +    u32    msgProdIdx;
> +    u32    msgConsIdx;
> +    u32    msgNumEntriesLog2;
> +} PACKED;
> +
> +struct PVSCSIRingReqDesc {
> +    u64    context;
> +    u64    dataAddr;
> +    u64    dataLen;
> +    u64    senseAddr;
> +    u32    senseLen;
> +    u32    flags;
> +    u8     cdb[16];
> +    u8     cdbLen;
> +    u8     lun[8];
> +    u8     tag;
> +    u8     bus;
> +    u8     target;
> +    u8     vcpuHint;
> +    u8     unused[59];
> +} PACKED;
> +
> +struct pvscsi_ring_dsc_s {
> +    struct PVSCSIRingsState *ring_state;
> +    struct PVSCSIRingReqDesc *ring_reqs;
> +    struct PVSCSIRingCmpDesc *ring_cmps;
> +};
> +
> +struct pvscsi_lun_s {
> +    struct drive_s drive;
> +    struct pci_device *pci;
> +    u32 iobase;
> +    u8 target;
> +    u8 lun;
> +    struct pvscsi_ring_dsc_s *ring_dsc;
> +};

Should these go into the header file?

I saw that other storage controllers add constants and structs to the *.c file (lsi-scsi.c, pvscsi.c).
So I assumed this was a convention in seabios code base.
> +
> +static void
> +pvscsi_write_cmd_desc(u32 iobase, u32 cmd, const void *desc, size_t len)
> +{
> +    const u32 *ptr = desc;
> +    size_t i;
> +
> +    len /= sizeof(*ptr);
> +    pci_writel(iobase + PVSCSI_REG_OFFSET_COMMAND, cmd);
> +    for (i = 0; i < len; i++)
> +        pci_writel(iobase + PVSCSI_REG_OFFSET_COMMAND_DATA, ptr[i]);
> +}
> +
> +static void
> +pvscsi_kick_rw_io(u32 iobase)
> +{
> +    pci_writel(iobase + PVSCSI_REG_OFFSET_KICK_RW_IO, 0);
> +}
> +
> +static void
> +pvscsi_wait_intr_cmpl(u32 iobase)
> +{
> +    while (!(pci_readl(iobase + PVSCSI_REG_OFFSET_INTR_STATUS) & PVSCSI_INTR_CMPL_MASK))
> +        usleep(5);
> +    pci_writel(iobase + PVSCSI_REG_OFFSET_INTR_STATUS, PVSCSI_INTR_CMPL_MASK);
> +
> +}
> +
> +static void
> +pvscsi_init_rings(u32 iobase, struct pvscsi_ring_dsc_s **ring_dsc)
> +{
> +    struct PVSCSICmdDescSetupRings cmd = {0,};
> +
> +    struct pvscsi_ring_dsc_s *dsc = memalign_low(sizeof(*dsc), PAGE_SIZE);
> +    if (!dsc) {
> +        warn_noalloc();
> +        return;
> +    }
> +
> +    memset(dsc, 0, sizeof(*dsc));

That memset is not needed, right?

right. I've removed it 
> +
> +    dsc->ring_state =
> +        (struct PVSCSIRingsState *)memalign_low(PAGE_SIZE, PAGE_SIZE);
> +    dsc->ring_reqs =
> +        (struct PVSCSIRingReqDesc *)memalign_low(PAGE_SIZE, PAGE_SIZE);
> +    dsc->ring_cmps =
> +        (struct PVSCSIRingCmpDesc *)memalign_low(PAGE_SIZE, PAGE_SIZE);
> +    if (!dsc->ring_state || !dsc->ring_reqs || !dsc->ring_cmps) {
> +        warn_noalloc();
> +        return;
> +    }
> +    memset(dsc->ring_state, 0, PAGE_SIZE);
> +    memset(dsc->ring_reqs, 0, PAGE_SIZE);
> +    memset(dsc->ring_cmps, 0, PAGE_SIZE);
> +
> +    cmd.reqRingNumPages = 1;
> +    cmd.cmpRingNumPages = 1;
> +    cmd.ringsStatePPN = virt_to_phys(dsc->ring_state) >> PAGE_SHIFT;
> +    cmd.reqRingPPNs[0] = virt_to_phys(dsc->ring_reqs) >> PAGE_SHIFT;;
> +    cmd.cmpRingPPNs[0] = virt_to_phys(dsc->ring_cmps) >> PAGE_SHIFT;;

1. Just one semicolon is needed at the end.

done
 
2. I am no C professional, but why are the `memset()` calls above needed
when the content does not seems to be used, when the pointer is casted
to `unsigned int`?
The pointers I memset, point to physical pages of memory. I need to zero them so that the pvscsi controller on qemu won't see some garbage values such as invalid requests count.
 
> +
> +    pvscsi_write_cmd_desc(iobase, PVSCSI_CMD_SETUP_RINGS,
> +                          &cmd, sizeof(cmd));
> +    *ring_dsc = dsc;
> +}

[…]


Thanks,

Paul



--
Best Regards,
Evgeny