[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: [PATCH 2/3] optionrom: add a DMA-enabled multiboot ROM
From: |
Philippe Mathieu-Daudé |
Subject: |
Re: [PATCH 2/3] optionrom: add a DMA-enabled multiboot ROM |
Date: |
Thu, 21 Oct 2021 00:08:04 +0200 |
User-agent: |
Mozilla/5.0 (X11; Linux x86_64; rv:91.0) Gecko/20100101 Thunderbird/91.1.0 |
On 10/20/21 16:02, Paolo Bonzini wrote:
> From: Marcus Hähnel <marcus.haehnel@kernkonzept.com>
>
> Add a new option rom for the multiboot loader, using DMA transfers to copy
> data instead of "rep insb".
>
> This significantly lowers QEMU's startup latency by a factor of about 40,
> for example, going from 30sec to 0.8sec when loading modules of 120MB
> in size.
>
> Signed-off-by: Marcus Hähnel <marcus.haehnel@kernkonzept.com>
> Signed-off-by: Adam Lackorzynski <adam@l4re.org>
> [Modified to keep the non-DMA code depending on #ifdef USE_FW_CFG_DMA;
> do not write below stack. - Paolo]
> Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
> ---
> pc-bios/meson.build | 1 +
> pc-bios/multiboot_dma.bin | Bin 0 -> 1024 bytes
> pc-bios/optionrom/Makefile | 4 +-
> pc-bios/optionrom/multiboot.S | 4 +-
> pc-bios/optionrom/multiboot_dma.S | 2 +
> pc-bios/optionrom/optionrom.h | 66 ++++++++++++++++++++++++++++++
> 6 files changed, 72 insertions(+), 5 deletions(-)
> create mode 100644 pc-bios/multiboot_dma.bin
> create mode 100644 pc-bios/optionrom/multiboot_dma.S
> diff --git a/pc-bios/multiboot_dma.bin b/pc-bios/multiboot_dma.bin
> new file mode 100644
> index
> 0000000000000000000000000000000000000000..c0e2c3102a3358207c61d3ae113524fb6007abc3
> GIT binary patch
> literal 1024
> zcmd^-v1=1i9LIlmUTlcNU1}{N5`u&{DAHSmq6iKdix;GHRxaNnR0A<Lc978Fk#IvN
> zaS+5wQE+fL&*iQWa#czc94ZyL>R{3q;@~8~^LMo;4s!p158nI!-tWur_ul*P<{!&%
> z=%3>xm5f`4Bq!!)`Rkwf;(oFd*qcAb=mhXX1)X>Bw-tylkUpR_ETXkH1AnO4L$nJ|
> zWJui_M1kmKmTID)P`mI^=HM`~{VB1t1l_Ye=Lor4X5{8Gd)zzw_o4<6zLQJ!dr<pW
> z;7%~ysSHL?*HHx*O%k}?fCyRL<7_RWt(4;Aqn}X}S>MolrKNZG2Z|kTOyaJIhYFc^
> zFhhuQ9`r5fQLCrm#T4^_QylQ>8kgs;ZX9bIEiXb`8AIxu;?h~d%9izBkKht%WM1G*
> z^E{W9(OT9dt5in2qF}|dPH>dL>{=sV#-U1<quUb@YkIW%niI?8-7fCz#695e<V=WZ
> zrVCY?W~`3Hw@^=cHALr#9F2GOTYJ;??9d*-Vbsi+;lz|<$jMX#;lr6ov3qKNLH7(W
> z+>yFoWnOtw14D$&k*SUtsT%wS`ki@#&rUrnmtuDv`PtJm(Kg?H$Pdc0CL@YCy4R<D
> pT|LnIbg(BnP0#tqRx5M!bkkaD-nd?`H;YU4Yi6yHwD{k({tHC8FAx9#
Ideally this should be generated on CI as artifact, archived
and we commit the CI generated file.
> diff --git a/pc-bios/optionrom/optionrom.h b/pc-bios/optionrom/optionrom.h
> index a2b612f1a7..8d74c0ddf3 100644
> --- a/pc-bios/optionrom/optionrom.h
> +++ b/pc-bios/optionrom/optionrom.h
> @@ -37,6 +37,17 @@
> #define BIOS_CFG_IOPORT_CFG 0x510
> #define BIOS_CFG_IOPORT_DATA 0x511
>
> +#define FW_CFG_DMA_CTL_ERROR 0x01
> +#define FW_CFG_DMA_CTL_READ 0x02
> +#define FW_CFG_DMA_CTL_SKIP 0x04
> +#define FW_CFG_DMA_CTL_SELECT 0x08
> +#define FW_CFG_DMA_CTL_WRITE 0x10
> +
> +#define FW_CFG_DMA_SIGNATURE 0x51454d5520434647ULL /* "QEMU CFG" */
> +
> +#define BIOS_CFG_DMA_ADDR_HIGH 0x514
> +#define BIOS_CFG_DMA_ADDR_LOW 0x518
> +
> /* Break the translation block flow so -d cpu shows us values */
> #define DEBUG_HERE \
> jmp 1f; \
> @@ -62,6 +73,61 @@
> bswap %eax
> .endm
>
> +
> +/*
> + * Read data from the fw_cfg device using DMA.
> + * Clobbers: %edx, %eax, ADDR, SIZE, memory[%esp-16] to memory[%esp]
> + */
> +.macro read_fw_dma VAR, SIZE, ADDR
> + /* Address */
> + bswapl \ADDR
> + pushl \ADDR
> +
> + /* We only support 32 bit target addresses */
> + xorl %eax, %eax
> + pushl %eax
> + mov $BIOS_CFG_DMA_ADDR_HIGH, %dx
> + outl %eax, (%dx)
> +
> + /* Size */
> + bswapl \SIZE
> + pushl \SIZE
> +
> + /* Control */
Indent off.
> + movl $(\VAR << 16) | (FW_CFG_DMA_CTL_READ |
> FW_CFG_DMA_CTL_SELECT), %eax
> + bswapl %eax
> + pushl %eax
> +
> + movl %esp, %eax /* Address of the struct we generated */
> + bswapl %eax
> + mov $BIOS_CFG_DMA_ADDR_LOW, %dx
> + outl %eax, (%dx) /* Initiate DMA */
> +
> +1: mov (%esp), %eax /* Wait for completion */
> + bswapl %eax
> + testl $~FW_CFG_DMA_CTL_ERROR, %eax
> + jnz 1b
> + addl $16, %esp
Indent off.
Reviewed-by: Philippe Mathieu-Daudé <philmd@redhat.com>