[PATCH v4 07/12] efi: passing kexec necessary efi data via setup_data

[Matt Fleming]

On Tue, 26 Nov, at 01:57:52PM, Dave Young wrote:
> Add a new setup_data type SETUP_EFI for kexec use.
> Passing the saved fw_vendor, runtime, config tables and
> efi runtime mappings.
> 
> When entering virtual mode, directly mapping the efi
> runtime ragions which we passed in previously. And skip
> the step to call SetVirtualAddressMap.
> 
> Specially for HP z420 workstation it need another variable
> saving, it's the smbios physical address, the HP bios
> also update the SMBIOS address after entering virtual mode
> besides of the standard fw_vendor,runtime and config table.
> 
> Tested on ovmf+qemu, lenovo thinkpad, a dell laptop and an
> HP z420 workstation.
> 
> v2: refresh based on previous patch changes, code cleanup.
> v3: use ioremap instead of phys_to_virt for esdata
> 
> Signed-off-by: Dave Young <dyoung at redhat.com>
> ---
>  arch/x86/include/asm/efi.h            |  12 +++
>  arch/x86/include/uapi/asm/bootparam.h |   1 +
>  arch/x86/kernel/setup.c               |   3 +
>  arch/x86/platform/efi/efi.c           | 161 ++++++++++++++++++++++++++++++----
>  4 files changed, 160 insertions(+), 17 deletions(-)

[...]

> +void __init parse_efi_setup(u64 phys_addr, u32 data_len)
> +{
> +	int size;
> +	struct setup_data *sdata;
> +	u64 esdata_phys;
> +
> +	if (!efi_enabled(EFI_64BIT)) {
> +		pr_warn("skipping setup_data on EFI 32BIT!");
> +		return;
> +	}

Hmm... this warning could be more informative. Perhaps something along
the lines of,

	"SETUP_EFI not supported on 32-bit\n"

because the reason we skip is because it isn't supported.

> +
> +	sdata = early_memremap(phys_addr, data_len);
> +	if (!sdata)
> +		return;
> +
> +	size = data_len - sizeof(struct setup_data);
> +
> +	esdata_phys = phys_addr + sizeof(struct setup_data);
> +
> +	nr_efi_runtime_map = (size - sizeof(struct efi_setup_data)) /
> +			sizeof(efi_memory_desc_t);
> +	early_iounmap(sdata, data_len);
> +
> +	/* iounmap esdata in function efi_enter_virtual_mode */
> +	esdata = early_memremap(esdata_phys, size);
> +}
>  
>  static efi_status_t virt_efi_get_time(efi_time_t *tm, efi_time_cap_t *tc)
>  {
> @@ -504,8 +531,12 @@ static int __init efi_systab_init(void *phys)
>  		}
>  
>  		efi_systab.hdr = systab64->hdr;
> -		efi_systab.fw_vendor = systab64->fw_vendor;
> -		tmp |= systab64->fw_vendor;
> +
> +		if (esdata)

Could we name this something more explicit? 'efi_setup' perhaps?

> @@ -631,6 +670,41 @@ static int __init efi_memmap_init(void)
>  	return 0;
>  }
>  
> +static int __init efi_reuse_config(u64 tables, int nr_tables)
> +{
> +	void *p, *tablep;
> +	int i, sz;
> +
> +	if (!efi_enabled(EFI_64BIT))
> +		return 0;
> +
> +	sz = sizeof(efi_config_table_64_t);
> +
> +	p = tablep = early_memremap(tables, nr_tables * sz);
> +	if (!p) {
> +		pr_err("Could not map Configuration table!\n");
> +		return -ENOMEM;
> +	}
> +
> +	for (i = 0; i < efi.systab->nr_tables; i++) {
> +		efi_guid_t guid;
> +
> +		guid = ((efi_config_table_64_t *)p)->guid;
> +
> +		/*
> +		HP z420 workstation smbios will be convert to
> +		virtual address after enter virtual mode.
> +		Thus in case kexec/kdump the physical address
> +		will be passed in setup_data.
> +		*/

This isn't the correct multi-line comment format,

		/*
		 * This is the preferred way to to write a multi-line
		 * comment.
		 */

> +		if (!efi_guidcmp(guid, SMBIOS_TABLE_GUID))
> +			((efi_config_table_64_t *)p)->table = esdata->smbios;
> +		p += sz;
> +	}
> +	early_iounmap(tablep, nr_tables * sz);
> +	return 0;
> +}
> +
>  void __init efi_init(void)
>  {
>  	efi_char16_t *c16;
> @@ -676,6 +750,9 @@ void __init efi_init(void)
>  		efi.systab->hdr.revision >> 16,
>  		efi.systab->hdr.revision & 0xffff, vendor);
>  
> +	if (esdata && esdata->smbios)
> +		efi_reuse_config(efi.systab->tables, efi.systab->nr_tables);
> +

Would it make sense to move the ->smbios check inside efi_reuse_config()
in case we need to extend the number of tables in the future?

>  	if (efi_config_init(arch_tables))
>  		return;
>  
> @@ -886,6 +963,43 @@ ret:
>  }
>  
>  /*
> + * map efi regions which was passed via setup_data
> + * the virt_addr is a fixed addr which was used in
> + * 1st kernel of kexec boot.
> + */
> +static void __init efi_map_regions_fixed(void)
> +{
> +	int i;
> +	unsigned long size;
> +	efi_memory_desc_t *md;
> +	u64 end, systab;
> +	void *p;
> +
> +	efi_runtime_map = kzalloc(nr_efi_runtime_map * memmap.desc_size,
> +				GFP_KERNEL);
> +	if (!efi_runtime_map)
> +		pr_err("Out of memory, EFI runtime on nested kexec non-functional!\n");
> +
> +	for (i = 0, p = efi_runtime_map; i < nr_efi_runtime_map; i++) {
> +		md = esdata->map + i;

Heh, I read that as 'esdata->map + 1' the first few times.

> +		efi_map_region_fixed(md);
> +		size = md->num_pages << PAGE_SHIFT;
> +		end = md->phys_addr + size;
> +
> +		systab = (u64) (unsigned long) efi_phys.systab;
> +		if (md->phys_addr <= systab && systab < end) {
> +			systab += md->virt_addr - md->phys_addr;
> +			efi.systab =
> +				(efi_system_table_t *) (unsigned long) systab;
> +		}
> +		if (efi_runtime_map) {
> +			memcpy(p, md, memmap.desc_size);
> +			p += memmap.desc_size;
> +		}

Is this if () needed? Is it possible to enter the loop and have
'efi_runtime_map' be NULL?

> +	}
> +}
> +
> +/*
>   * This function will switch the EFI runtime services to virtual mode.
>   * Essentially, we look through the EFI memmap and map every region that
>   * has the runtime attribute bit set in its memory descriptor into the
> @@ -901,6 +1015,10 @@ ret:
>   * so that we're in a different address space when calling a runtime
>   * function. For function arguments passing we do copy the PGDs of the
>   * kernel page table into ->trampoline_pgd prior to each call.
> + *
> + * Specially for kexec boot efi runtime maps in previous kernel should
> + * be passed in via setup_data. In that case runtime ranges will be mapped
> + * to fixed virtual addresses exactly same as the ones in previous kernel.
>   */
>  void __init efi_enter_virtual_mode(void)
>  {
> @@ -919,12 +1037,15 @@ void __init efi_enter_virtual_mode(void)
>  		return;
>  	}
>  
> -	efi_merge_regions();
> -
> -	new_memmap = efi_map_regions(&count);
> -	if (!new_memmap) {
> -		pr_err("Error reallocating memory, EFI runtime non-functional!\n");
> -		return;
> +	if (esdata)
> +		efi_map_regions_fixed();
> +	else {
> +		efi_merge_regions();
> +		new_memmap = efi_map_regions(&count);
> +		if (!new_memmap) {
> +			pr_err("Error reallocating memory, EFI runtime non-functional!\n");
> +			return;
> +		}
>  	}
>  
>  	BUG_ON(!efi.systab);
> @@ -932,11 +1053,17 @@ void __init efi_enter_virtual_mode(void)
>  	efi_setup_page_tables();
>  	efi_sync_low_kernel_mappings();
>  
> -	status = phys_efi_set_virtual_address_map(
> -		memmap.desc_size * count,
> -		memmap.desc_size,
> -		memmap.desc_version,
> -		(efi_memory_desc_t *)__pa(new_memmap));
> +	if (esdata) {
> +		status = EFI_SUCCESS;
> +		early_iounmap(esdata, sizeof(struct efi_setup_data) +
> +			      nr_efi_runtime_map *
> +			      sizeof(efi_memory_desc_t));
> +	} else
> +		status = phys_efi_set_virtual_address_map(
> +			memmap.desc_size * count,
> +			memmap.desc_size,
> +			memmap.desc_version,
> +			(efi_memory_desc_t *)__pa(new_memmap));
>  
>  	if (status != EFI_SUCCESS) {
>  		pr_alert("Unable to switch EFI into virtual mode "

Please fold the pr_alert() into the else branch so that you don't have
to do the 'status = EFI_SUCCESS' dance.

-- 
Matt Fleming, Intel Open Source Technology Center