On Thu, Nov 23, 2023 at 3:06 PM Ainux Wang <ainux.wang@xxxxxxxxx> wrote:
>
> Huacai Chen <chenhuacai@xxxxxxxxxx> 于2023年11月23日周四 12:05写道：
> >
> > On Thu, Nov 23, 2023 at 8:23 AM Ainux Wang <ainux.wang@xxxxxxxxx> wrote:
> > >
> > > Huacai Chen <chenhuacai@xxxxxxxxxx> 于2023年11月22日周三 22:22写道：
> > > >
> > > > Hi, Yao,
> > > >
> > > > On Wed, Nov 22, 2023 at 4:50 PM <wangyao@xxxxxxxxxx> wrote:
> > > > >
> > > > > From: Wang Yao <wangyao@xxxxxxxxxx>
> > > > >
> > > > > Refer ot UEFI spec v2.10 section 2.3.8 LoongArch Platforms:
> > > > >
> > > > > The processor is in the following execution mode during boot service:
> > > > > ...
> > > > > The memory is in physical addressing mode. LoongArch architecture
> > > > > defines two memory access mode, namely direct address translation
> > > > > mode and mapped address translation mode.
> > > > >
> > > > > So need to change MMU translation mode before config direct mapping.
> > > > From UEFI spec v2.10 section 2.3.8, you can also see "MMU enabled", so
> > > > PG is already enabled in UEFI.
> > > >
> > > > Huacai
> > > >
> > > Hi,Huacai,
> > >
> > > From LoongArch-Vol1-v1.10-EN.pdf , DA or PG both refer to a mode enabled by MMU,
> > > Will the UEFI firmware not use DA mode?
> > > The UEFI spec did not specify the use of DA or PG mode, but the kernel
> > > must use PG.
> > In theory you are right, but "MMU" usually means "TLB", including this
> > case. EFI runtime needs TLB, so UEFI cannot be in DA mode.
> >
> > Huacai
> >
> Is it possible that set_virtual_address_map() only configures the DMW
> under PG mode
> without switching DA to PG mode?
>
> I noticed in efi_get_virtmap() that virt_addr = phys_addr +
> 0x8000_0000_0000_000,
> and I also observed in the kernel's head.S there's a PG mode switch process
> ，which could enable runtime services to use PG mode.
>
> However, this occurs after the STUB,
> so does the STUB phase really require or ensure that the UEFI firmware
> is definitely in
> PG mode?
Anything you think redundant may be not redundant, remember that UEFI
is not the only BIOS and Linux is not the only OS. So Linux kernel
cannot assume any BIOS is the same as UEFI does, UEFI cannot assume
any OS is the same as Linux does, too.
Huacai
>
> > >
> > > Best regards,
> > > Ainux Wang.
> > > > >
> > > > > Signed-off-by: Wang Yao <wangyao@xxxxxxxxxx>
> > > > > ---
> > > > > drivers/firmware/efi/libstub/loongarch.c | 8 +++++++-
> > > > > 1 file changed, 7 insertions(+), 1 deletion(-)
> > > > >
> > > > > diff --git a/drivers/firmware/efi/libstub/loongarch.c b/drivers/firmware/efi/libstub/loongarch.c
> > > > > index 807cba2693fc..4c0a84c58f5b 100644
> > > > > --- a/drivers/firmware/efi/libstub/loongarch.c
> > > > > +++ b/drivers/firmware/efi/libstub/loongarch.c
> > > > > @@ -49,7 +49,7 @@ efi_status_t efi_boot_kernel(void *handle, efi_loaded_image_t *image,
> > > > > struct exit_boot_struct priv;
> > > > > unsigned long desc_size;
> > > > > efi_status_t status;
> > > > > - u32 desc_ver;
> > > > > + u32 desc_ver, val;
> > > > >
> > > > > status = efi_alloc_virtmap(&priv.runtime_map, &desc_size, &desc_ver);
> > > > > if (status != EFI_SUCCESS) {
> > > > > @@ -69,6 +69,12 @@ efi_status_t efi_boot_kernel(void *handle, efi_loaded_image_t *image,
> > > > > priv.runtime_entry_count * desc_size, desc_size,
> > > > > desc_ver, priv.runtime_map);
> > > > >
> > > > > + /* Change address translation mode */
> > > > > + val = csr_read32(LOONGARCH_CSR_CRMD);
> > > > > + val &= ~CSR_CRMD_DA;
> > > > > + val |= CSR_CRMD_PG;
> > > > > + csr_write32(val, LOONGARCH_CSR_CRMD);
> > > > > +
> > > > > /* Config Direct Mapping */
> > > > > csr_write64(CSR_DMW0_INIT, LOONGARCH_CSR_DMWIN0);
> > > > > csr_write64(CSR_DMW1_INIT, LOONGARCH_CSR_DMWIN1);
> > > > > --
> > > > > 2.27.0
> > > > >
