| Message ID | 20190425195643.7104-2-atish.patra@wdc.com |
|---|---|
| State | Superseded |
| Delegated to: | Andes |
| Headers | show |
| Series | [U-Boot,v2,1/2] RISCV: image: Add booti support. | expand |
On 4/25/19 9:56 PM, Atish Patra wrote: > Add gz parsing logic so that booti can parse both Image > and Image.gz. > > Signed-off-by: Atish Patra <atish.patra@wdc.com> > --- > arch/riscv/lib/image.c | 28 +++++++++++++++++++++++++++- > 1 file changed, 27 insertions(+), 1 deletion(-) > > diff --git a/arch/riscv/lib/image.c b/arch/riscv/lib/image.c > index e8802007c446..73ebd0da3885 100644 > --- a/arch/riscv/lib/image.c > +++ b/arch/riscv/lib/image.c > @@ -9,6 +9,8 @@ > #include <common.h> > #include <mapmem.h> > #include <errno.h> > +#include <bootm.h> > +#include <malloc.h> > #include <linux/sizes.h> > #include <linux/stddef.h> > > @@ -16,6 +18,8 @@ DECLARE_GLOBAL_DATA_PTR; > > /* ASCII version of "RISCV" defined in Linux kernel */ > #define LINUX_RISCV_IMAGE_MAGIC 0x5643534952 > +#define GZ_HEADER_0 0x1f > +#define GZ_HEADER_1 0x8b > > struct linux_image_h { > uint32_t code0; /* Executable code */ > @@ -32,9 +36,31 @@ int booti_setup(ulong image, ulong *relocated_addr, ulong *size, > bool force_reloc) > { > struct linux_image_h *lhdr; > + uint8_t *temp; > + void *dest; > + ulong dest_end; > + int ret; > + /* TODO: Is there a way to figure out length of compressed Image.gz ? > + * Otherwise, set it to SYS_BOOTM_LEN which should be sufficient. > + */ Presumably this is a RFC patch then ? What happens if you have two (or more) gzip-ed files back-to-back ? Wouldn't you then decompress both ? That might lead to all kinds of problems. > + int len = CONFIG_SYS_BOOTM_LEN; > + > + temp = (uint8_t *)map_sysmem(image, 0); Is the type cast really needed ? > - lhdr = (struct linux_image_h *)map_sysmem(image, 0); > + if (*(temp) == GZ_HEADER_0 && *(temp+1) == GZ_HEADER_1) { Wrap the image in some fitImage or so contained, mark the image as gzip compressed there and all this goes away. > + /* Need a temporary location to copy the uncompressed image */ > + dest = (void *)map_sysmem(image + 8 * CONFIG_SYS_BOOTM_LEN, 0); > + ret = bootm_decomp_image(IH_COMP_GZIP, 0, image, IH_TYPE_KERNEL, > + dest, (void *)image, len, > + CONFIG_SYS_BOOTM_LEN, &dest_end); > + if (ret) > + return ret; > + /* dest_end contains the uncompressed Image size */ > + memmove((void *) image, dest, dest_end); > + unmap_sysmem(dest); > + } > > + lhdr = (struct linux_image_h *)temp; > if (lhdr->magic != LINUX_RISCV_IMAGE_MAGIC) { > puts("Bad Linux RISCV Image magic!\n"); > return -EINVAL; >
On 4/25/19 10:33 PM, Marek Vasut wrote: > On 4/25/19 9:56 PM, Atish Patra wrote: >> Add gz parsing logic so that booti can parse both Image >> and Image.gz. >> >> Signed-off-by: Atish Patra <atish.patra@wdc.com> >> --- >> arch/riscv/lib/image.c | 28 +++++++++++++++++++++++++++- >> 1 file changed, 27 insertions(+), 1 deletion(-) >> >> diff --git a/arch/riscv/lib/image.c b/arch/riscv/lib/image.c >> index e8802007c446..73ebd0da3885 100644 >> --- a/arch/riscv/lib/image.c >> +++ b/arch/riscv/lib/image.c >> @@ -9,6 +9,8 @@ >> #include <common.h> >> #include <mapmem.h> >> #include <errno.h> >> +#include <bootm.h> >> +#include <malloc.h> >> #include <linux/sizes.h> >> #include <linux/stddef.h> >> >> @@ -16,6 +18,8 @@ DECLARE_GLOBAL_DATA_PTR; >> >> /* ASCII version of "RISCV" defined in Linux kernel */ >> #define LINUX_RISCV_IMAGE_MAGIC 0x5643534952 >> +#define GZ_HEADER_0 0x1f >> +#define GZ_HEADER_1 0x8b >> >> struct linux_image_h { >> uint32_t code0; /* Executable code */ >> @@ -32,9 +36,31 @@ int booti_setup(ulong image, ulong *relocated_addr, ulong *size, >> bool force_reloc) >> { >> struct linux_image_h *lhdr; >> + uint8_t *temp; >> + void *dest; >> + ulong dest_end; >> + int ret; >> + /* TODO: Is there a way to figure out length of compressed Image.gz ? >> + * Otherwise, set it to SYS_BOOTM_LEN which should be sufficient. >> + */ > > Presumably this is a RFC patch then ? > Yeah. I am not very sure if there is a better way to determine the size. Hence the comment. I am hoping somebody here would suggest something ;). > What happens if you have two (or more) gzip-ed files back-to-back ? > Wouldn't you then decompress both ? That might lead to all kinds of > problems. > That will be catastrophic. But this was intended only for booti and the expectation was that only Image.gz must be loaded before this. Having said that, if we can find a reliable way of figuring out the compressed file size here, we can get rid of this hack. >> + int len = CONFIG_SYS_BOOTM_LEN; >> + >> + temp = (uint8_t *)map_sysmem(image, 0); > > Is the type cast really needed ? > Just wanted to be explicit. Will remove it. >> - lhdr = (struct linux_image_h *)map_sysmem(image, 0); >> + if (*(temp) == GZ_HEADER_0 && *(temp+1) == GZ_HEADER_1) { > > Wrap the image in some fitImage or so contained, mark the image as gzip > compressed there and all this goes away. > Yes. FIT image parsing can be done in that way. However, the idea was here to load Image.gz directly. Image.gz is default compressed Linux kernel image format in RISC-V. Regards, Atish >> + /* Need a temporary location to copy the uncompressed image */ >> + dest = (void *)map_sysmem(image + 8 * CONFIG_SYS_BOOTM_LEN, 0); >> + ret = bootm_decomp_image(IH_COMP_GZIP, 0, image, IH_TYPE_KERNEL, >> + dest, (void *)image, len, >> + CONFIG_SYS_BOOTM_LEN, &dest_end); >> + if (ret) >> + return ret; >> + /* dest_end contains the uncompressed Image size */ >> + memmove((void *) image, dest, dest_end); >> + unmap_sysmem(dest); >> + } >> >> + lhdr = (struct linux_image_h *)temp; >> if (lhdr->magic != LINUX_RISCV_IMAGE_MAGIC) { >> puts("Bad Linux RISCV Image magic!\n"); >> return -EINVAL; >> > >
On 4/26/19 7:08 PM, Atish Patra wrote: > On 4/25/19 10:33 PM, Marek Vasut wrote: >> On 4/25/19 9:56 PM, Atish Patra wrote: >>> Add gz parsing logic so that booti can parse both Image >>> and Image.gz. >>> >>> Signed-off-by: Atish Patra <atish.patra@wdc.com> >>> --- >>> arch/riscv/lib/image.c | 28 +++++++++++++++++++++++++++- >>> 1 file changed, 27 insertions(+), 1 deletion(-) >>> >>> diff --git a/arch/riscv/lib/image.c b/arch/riscv/lib/image.c >>> index e8802007c446..73ebd0da3885 100644 >>> --- a/arch/riscv/lib/image.c >>> +++ b/arch/riscv/lib/image.c >>> @@ -9,6 +9,8 @@ >>> #include <common.h> >>> #include <mapmem.h> >>> #include <errno.h> >>> +#include <bootm.h> >>> +#include <malloc.h> >>> #include <linux/sizes.h> >>> #include <linux/stddef.h> >>> @@ -16,6 +18,8 @@ DECLARE_GLOBAL_DATA_PTR; >>> /* ASCII version of "RISCV" defined in Linux kernel */ >>> #define LINUX_RISCV_IMAGE_MAGIC 0x5643534952 >>> +#define GZ_HEADER_0 0x1f >>> +#define GZ_HEADER_1 0x8b >>> struct linux_image_h { >>> uint32_t code0; /* Executable code */ >>> @@ -32,9 +36,31 @@ int booti_setup(ulong image, ulong >>> *relocated_addr, ulong *size, >>> bool force_reloc) >>> { >>> struct linux_image_h *lhdr; >>> + uint8_t *temp; >>> + void *dest; >>> + ulong dest_end; >>> + int ret; >>> + /* TODO: Is there a way to figure out length of compressed >>> Image.gz ? >>> + * Otherwise, set it to SYS_BOOTM_LEN which should be sufficient. >>> + */ >> >> Presumably this is a RFC patch then ? >> > Yeah. I am not very sure if there is a better way to determine the size. > Hence the comment. I am hoping somebody here would suggest something ;). Mark the patch as RFC next time please. >> What happens if you have two (or more) gzip-ed files back-to-back ? >> Wouldn't you then decompress both ? That might lead to all kinds of >> problems. >> > > That will be catastrophic. But this was intended only for booti and the > expectation was that only Image.gz must be loaded before this. That also means it's horribly fragile. > Having said that, if we can find a reliable way of figuring out the > compressed file size here, we can get rid of this hack. See below >>> + int len = CONFIG_SYS_BOOTM_LEN; >>> + >>> + temp = (uint8_t *)map_sysmem(image, 0); >> >> Is the type cast really needed ? >> > > Just wanted to be explicit. Will remove it. > >>> - lhdr = (struct linux_image_h *)map_sysmem(image, 0); >>> + if (*(temp) == GZ_HEADER_0 && *(temp+1) == GZ_HEADER_1) { >> >> Wrap the image in some fitImage or so contained, mark the image as gzip >> compressed there and all this goes away. >> > > Yes. FIT image parsing can be done in that way. However, the idea was > here to load Image.gz directly. Image.gz is default compressed Linux > kernel image format in RISC-V. Sigh, and the image header is compressed as well, so there's no way to identify the image format, right ? And there's no decompressor, so the dcompressing has to be done by bootloader, which would need some sort of very smart way of figuring out which exact compression method is used ?
On 4/26/19 11:05 AM, Marek Vasut wrote: > On 4/26/19 7:08 PM, Atish Patra wrote: >> On 4/25/19 10:33 PM, Marek Vasut wrote: >>> On 4/25/19 9:56 PM, Atish Patra wrote: >>>> Add gz parsing logic so that booti can parse both Image >>>> and Image.gz. >>>> >>>> Signed-off-by: Atish Patra <atish.patra@wdc.com> >>>> --- >>>> arch/riscv/lib/image.c | 28 +++++++++++++++++++++++++++- >>>> 1 file changed, 27 insertions(+), 1 deletion(-) >>>> >>>> diff --git a/arch/riscv/lib/image.c b/arch/riscv/lib/image.c >>>> index e8802007c446..73ebd0da3885 100644 >>>> --- a/arch/riscv/lib/image.c >>>> +++ b/arch/riscv/lib/image.c >>>> @@ -9,6 +9,8 @@ >>>> #include <common.h> >>>> #include <mapmem.h> >>>> #include <errno.h> >>>> +#include <bootm.h> >>>> +#include <malloc.h> >>>> #include <linux/sizes.h> >>>> #include <linux/stddef.h> >>>> @@ -16,6 +18,8 @@ DECLARE_GLOBAL_DATA_PTR; >>>> /* ASCII version of "RISCV" defined in Linux kernel */ >>>> #define LINUX_RISCV_IMAGE_MAGIC 0x5643534952 >>>> +#define GZ_HEADER_0 0x1f >>>> +#define GZ_HEADER_1 0x8b >>>> struct linux_image_h { >>>> uint32_t code0; /* Executable code */ >>>> @@ -32,9 +36,31 @@ int booti_setup(ulong image, ulong >>>> *relocated_addr, ulong *size, >>>> bool force_reloc) >>>> { >>>> struct linux_image_h *lhdr; >>>> + uint8_t *temp; >>>> + void *dest; >>>> + ulong dest_end; >>>> + int ret; >>>> + /* TODO: Is there a way to figure out length of compressed >>>> Image.gz ? >>>> + * Otherwise, set it to SYS_BOOTM_LEN which should be sufficient. >>>> + */ >>> >>> Presumably this is a RFC patch then ? >>> >> Yeah. I am not very sure if there is a better way to determine the size. >> Hence the comment. I am hoping somebody here would suggest something ;). > > Mark the patch as RFC next time please. > Sure. Sorry for not marking RFC in the first time. >>> What happens if you have two (or more) gzip-ed files back-to-back ? >>> Wouldn't you then decompress both ? That might lead to all kinds of >>> problems. >>> >> >> That will be catastrophic. But this was intended only for booti and the >> expectation was that only Image.gz must be loaded before this. > > That also means it's horribly fragile. > >> Having said that, if we can find a reliable way of figuring out the >> compressed file size here, we can get rid of this hack. > > See below > >>>> + int len = CONFIG_SYS_BOOTM_LEN; >>>> + >>>> + temp = (uint8_t *)map_sysmem(image, 0); >>> >>> Is the type cast really needed ? >>> >> >> Just wanted to be explicit. Will remove it. >> >>>> - lhdr = (struct linux_image_h *)map_sysmem(image, 0); >>>> + if (*(temp) == GZ_HEADER_0 && *(temp+1) == GZ_HEADER_1) { >>> >>> Wrap the image in some fitImage or so contained, mark the image as gzip >>> compressed there and all this goes away. >>> >> >> Yes. FIT image parsing can be done in that way. However, the idea was >> here to load Image.gz directly. Image.gz is default compressed Linux >> kernel image format in RISC-V. > > Sigh, and the image header is compressed as well, so there's no way to > identify the image format, right ? And there's no decompressor, so the > dcompressing has to be done by bootloader, which would need some sort of > very smart way of figuring out which exact compression method is used ? > Yes. Image.gz is always gunzip. So checking first two bytes is enough to confirm that it is a gz file. The tricky part is length of the compressed file. I took another look at the gunzip implementation in U-Boot. It looks like to me that compressed header length just to parse the header correctly. It doesn't actually use the "length" to decompress. In fact, it updates the length with uncompressed bytes after the decompression. Regards, Atish
On 4/30/19 3:27 AM, Atish Patra wrote: [...] >>> Yes. FIT image parsing can be done in that way. However, the idea was >>> here to load Image.gz directly. Image.gz is default compressed Linux >>> kernel image format in RISC-V. >> >> Sigh, and the image header is compressed as well, so there's no way to >> identify the image format, right ? And there's no decompressor, so the >> dcompressing has to be done by bootloader, which would need some sort of >> very smart way of figuring out which exact compression method is used ? >> > Yes. Image.gz is always gunzip. So checking first two bytes is enough to > confirm that it is a gz file. What happens once people start feeding it more exotic compression methods, like LZ4 or LZO or LZMA for example ? > The tricky part is length of the compressed file. I took another look at > the gunzip implementation in U-Boot. It looks like to me that compressed > header length just to parse the header correctly. It doesn't actually > use the "length" to decompress. In fact, it updates the length with > uncompressed bytes after the decompression. That's possible.
On 4/30/19 2:52 AM, Marek Vasut wrote: > On 4/30/19 3:27 AM, Atish Patra wrote: > > [...] > >>>> Yes. FIT image parsing can be done in that way. However, the idea was >>>> here to load Image.gz directly. Image.gz is default compressed Linux >>>> kernel image format in RISC-V. >>> >>> Sigh, and the image header is compressed as well, so there's no way to >>> identify the image format, right ? And there's no decompressor, so the >>> dcompressing has to be done by bootloader, which would need some sort of >>> very smart way of figuring out which exact compression method is used ? >>> >> Yes. Image.gz is always gunzip. So checking first two bytes is enough to >> confirm that it is a gz file. > > What happens once people start feeding it more exotic compression > methods, like LZ4 or LZO or LZMA for example ? > booti command help will clearly state that it can only boot kernel from Image or Image.gz. static char booti_help_text[] = "[addr [initrd[:size]] [fdt]]\n" - " - boot arm64 Linux Image stored in memory\n" + " - boot arm64 Linux Image or riscv Linux Image/Image.gz stored in memory\n" (I will update the help text with Image.gz part) Anything other than that, it will fail with following error. "Bad Linux RISCV Image magic!" >> The tricky part is length of the compressed file. I took another look at >> the gunzip implementation in U-Boot. It looks like to me that compressed >> header length just to parse the header correctly. It doesn't actually >> use the "length" to decompress. In fact, it updates the length with >> uncompressed bytes after the decompression. > > That's possible. > David suggested a better idea. 1. User can supply kernel_size and we can store in environment variable. 2. If the size is empty or greater than CONFIG_SYS_BOOTM_LEN, booti fails with appropriate error message. We will update the documents to add the additional step for Image.gz I am fine with either approach. Any preference ? Regards, Atish
On 4/30/19 8:13 PM, Atish Patra wrote: > On 4/30/19 2:52 AM, Marek Vasut wrote: >> On 4/30/19 3:27 AM, Atish Patra wrote: >> >> [...] >> >>>>> Yes. FIT image parsing can be done in that way. However, the idea was >>>>> here to load Image.gz directly. Image.gz is default compressed Linux >>>>> kernel image format in RISC-V. >>>> >>>> Sigh, and the image header is compressed as well, so there's no way to >>>> identify the image format, right ? And there's no decompressor, so the >>>> dcompressing has to be done by bootloader, which would need some >>>> sort of >>>> very smart way of figuring out which exact compression method is used ? >>>> >>> Yes. Image.gz is always gunzip. So checking first two bytes is enough to >>> confirm that it is a gz file. >> >> What happens once people start feeding it more exotic compression >> methods, like LZ4 or LZO or LZMA for example ? >> > > booti command help will clearly state that it can only boot kernel from > Image or Image.gz. > > static char booti_help_text[] = > "[addr [initrd[:size]] [fdt]]\n" > - " - boot arm64 Linux Image stored in memory\n" > + " - boot arm64 Linux Image or riscv Linux Image/Image.gz stored > in memory\n" Obvious question -- does this Image.gz stuff apply to arm64 ? > > (I will update the help text with Image.gz part) > > Anything other than that, it will fail with following error. > > "Bad Linux RISCV Image magic!" Right, so we're implementing file(1)-lite here to detect the format. >>> The tricky part is length of the compressed file. I took another look at >>> the gunzip implementation in U-Boot. It looks like to me that compressed >>> header length just to parse the header correctly. It doesn't actually >>> use the "length" to decompress. In fact, it updates the length with >>> uncompressed bytes after the decompression. >> >> That's possible. >> > > David suggested a better idea. > > 1. User can supply kernel_size and we can store in environment variable. > 2. If the size is empty or greater than CONFIG_SYS_BOOTM_LEN, booti > fails with appropriate error message. You can keep decompressing until you reach $bootm_len, sure . > We will update the documents to add the additional step for Image.gz > > I am fine with either approach. Any preference ? > > Regards, > Atish
On 4/30/19 12:11 PM, Marek Vasut wrote: > On 4/30/19 8:13 PM, Atish Patra wrote: >> On 4/30/19 2:52 AM, Marek Vasut wrote: >>> On 4/30/19 3:27 AM, Atish Patra wrote: >>> >>> [...] >>> >>>>>> Yes. FIT image parsing can be done in that way. However, the idea was >>>>>> here to load Image.gz directly. Image.gz is default compressed Linux >>>>>> kernel image format in RISC-V. >>>>> >>>>> Sigh, and the image header is compressed as well, so there's no way to >>>>> identify the image format, right ? And there's no decompressor, so the >>>>> dcompressing has to be done by bootloader, which would need some >>>>> sort of >>>>> very smart way of figuring out which exact compression method is used ? >>>>> >>>> Yes. Image.gz is always gunzip. So checking first two bytes is enough to >>>> confirm that it is a gz file. >>> >>> What happens once people start feeding it more exotic compression >>> methods, like LZ4 or LZO or LZMA for example ? >>> >> >> booti command help will clearly state that it can only boot kernel from >> Image or Image.gz. >> >> static char booti_help_text[] = >> "[addr [initrd[:size]] [fdt]]\n" >> - " - boot arm64 Linux Image stored in memory\n" >> + " - boot arm64 Linux Image or riscv Linux Image/Image.gz stored >> in memory\n" > > Obvious question -- does this Image.gz stuff apply to arm64 ? > arm64 builds Image.gz but booti for arm64 doesn't use it. I guess Image.gz can be used in FIT image for ARM64 but I am not sure which platform actually uses it. This patch only enables support for RISC-V. >> >> (I will update the help text with Image.gz part) >> >> Anything other than that, it will fail with following error. >> >> "Bad Linux RISCV Image magic!" > > Right, so we're implementing file(1)-lite here to detect the format. > >>>> The tricky part is length of the compressed file. I took another look at >>>> the gunzip implementation in U-Boot. It looks like to me that compressed >>>> header length just to parse the header correctly. It doesn't actually >>>> use the "length" to decompress. In fact, it updates the length with >>>> uncompressed bytes after the decompression. >>> >>> That's possible. >>> >> >> David suggested a better idea. >> >> 1. User can supply kernel_size and we can store in environment variable. >> 2. If the size is empty or greater than CONFIG_SYS_BOOTM_LEN, booti >> fails with appropriate error message. > > You can keep decompressing until you reach $bootm_len, sure . Decompression happens for actual uncompressed length which is encoded into the compressed file footer. So we don't have to worry about how much we decompress. But a correct file_size helps to avoid decompressing a possible corrupt compressed file. I will update the patch as per David's suggestion. Regards, Atish > >> We will update the documents to add the additional step for Image.gz >> >> I am fine with either approach. Any preference ? >> >> Regards, >> Atish > >
On 4/30/19 10:38 PM, Atish Patra wrote: > On 4/30/19 12:11 PM, Marek Vasut wrote: >> On 4/30/19 8:13 PM, Atish Patra wrote: >>> On 4/30/19 2:52 AM, Marek Vasut wrote: >>>> On 4/30/19 3:27 AM, Atish Patra wrote: >>>> >>>> [...] >>>> >>>>>>> Yes. FIT image parsing can be done in that way. However, the idea >>>>>>> was >>>>>>> here to load Image.gz directly. Image.gz is default compressed Linux >>>>>>> kernel image format in RISC-V. >>>>>> >>>>>> Sigh, and the image header is compressed as well, so there's no >>>>>> way to >>>>>> identify the image format, right ? And there's no decompressor, so >>>>>> the >>>>>> dcompressing has to be done by bootloader, which would need some >>>>>> sort of >>>>>> very smart way of figuring out which exact compression method is >>>>>> used ? >>>>>> >>>>> Yes. Image.gz is always gunzip. So checking first two bytes is >>>>> enough to >>>>> confirm that it is a gz file. >>>> >>>> What happens once people start feeding it more exotic compression >>>> methods, like LZ4 or LZO or LZMA for example ? >>>> >>> >>> booti command help will clearly state that it can only boot kernel from >>> Image or Image.gz. >>> >>> static char booti_help_text[] = >>> "[addr [initrd[:size]] [fdt]]\n" >>> - " - boot arm64 Linux Image stored in memory\n" >>> + " - boot arm64 Linux Image or riscv Linux Image/Image.gz stored >>> in memory\n" >> >> Obvious question -- does this Image.gz stuff apply to arm64 ? >> > > arm64 builds Image.gz but booti for arm64 doesn't use it. I guess > Image.gz can be used in FIT image for ARM64 but I am not sure which > platform actually uses it. > > This patch only enables support for RISC-V. Can't this be made generic ? [...]
On 4/30/19 2:42 PM, Marek Vasut wrote: > On 4/30/19 10:38 PM, Atish Patra wrote: >> On 4/30/19 12:11 PM, Marek Vasut wrote: >>> On 4/30/19 8:13 PM, Atish Patra wrote: >>>> On 4/30/19 2:52 AM, Marek Vasut wrote: >>>>> On 4/30/19 3:27 AM, Atish Patra wrote: >>>>> >>>>> [...] >>>>> >>>>>>>> Yes. FIT image parsing can be done in that way. However, the idea >>>>>>>> was >>>>>>>> here to load Image.gz directly. Image.gz is default compressed Linux >>>>>>>> kernel image format in RISC-V. >>>>>>> >>>>>>> Sigh, and the image header is compressed as well, so there's no >>>>>>> way to >>>>>>> identify the image format, right ? And there's no decompressor, so >>>>>>> the >>>>>>> dcompressing has to be done by bootloader, which would need some >>>>>>> sort of >>>>>>> very smart way of figuring out which exact compression method is >>>>>>> used ? >>>>>>> >>>>>> Yes. Image.gz is always gunzip. So checking first two bytes is >>>>>> enough to >>>>>> confirm that it is a gz file. >>>>> >>>>> What happens once people start feeding it more exotic compression >>>>> methods, like LZ4 or LZO or LZMA for example ? >>>>> >>>> >>>> booti command help will clearly state that it can only boot kernel from >>>> Image or Image.gz. >>>> >>>> static char booti_help_text[] = >>>> "[addr [initrd[:size]] [fdt]]\n" >>>> - " - boot arm64 Linux Image stored in memory\n" >>>> + " - boot arm64 Linux Image or riscv Linux Image/Image.gz stored >>>> in memory\n" >>> >>> Obvious question -- does this Image.gz stuff apply to arm64 ? >>> >> >> arm64 builds Image.gz but booti for arm64 doesn't use it. I guess >> Image.gz can be used in FIT image for ARM64 but I am not sure which >> platform actually uses it. >> >> This patch only enables support for RISC-V. > > Can't this be made generic ? > I think so if I just move the gzip detection and decompression code to cmd/booti.c. I will update the v3 patch with this. Regards, Atish > [...] >
On 5/1/19 12:06 AM, Atish Patra wrote: > On 4/30/19 2:42 PM, Marek Vasut wrote: >> On 4/30/19 10:38 PM, Atish Patra wrote: >>> On 4/30/19 12:11 PM, Marek Vasut wrote: >>>> On 4/30/19 8:13 PM, Atish Patra wrote: >>>>> On 4/30/19 2:52 AM, Marek Vasut wrote: >>>>>> On 4/30/19 3:27 AM, Atish Patra wrote: >>>>>> >>>>>> [...] >>>>>> >>>>>>>>> Yes. FIT image parsing can be done in that way. However, the idea >>>>>>>>> was >>>>>>>>> here to load Image.gz directly. Image.gz is default compressed >>>>>>>>> Linux >>>>>>>>> kernel image format in RISC-V. >>>>>>>> >>>>>>>> Sigh, and the image header is compressed as well, so there's no >>>>>>>> way to >>>>>>>> identify the image format, right ? And there's no decompressor, so >>>>>>>> the >>>>>>>> dcompressing has to be done by bootloader, which would need some >>>>>>>> sort of >>>>>>>> very smart way of figuring out which exact compression method is >>>>>>>> used ? >>>>>>>> >>>>>>> Yes. Image.gz is always gunzip. So checking first two bytes is >>>>>>> enough to >>>>>>> confirm that it is a gz file. >>>>>> >>>>>> What happens once people start feeding it more exotic compression >>>>>> methods, like LZ4 or LZO or LZMA for example ? >>>>>> >>>>> >>>>> booti command help will clearly state that it can only boot kernel >>>>> from >>>>> Image or Image.gz. >>>>> >>>>> static char booti_help_text[] = >>>>> "[addr [initrd[:size]] [fdt]]\n" >>>>> - " - boot arm64 Linux Image stored in memory\n" >>>>> + " - boot arm64 Linux Image or riscv Linux Image/Image.gz >>>>> stored >>>>> in memory\n" >>>> >>>> Obvious question -- does this Image.gz stuff apply to arm64 ? >>>> >>> >>> arm64 builds Image.gz but booti for arm64 doesn't use it. I guess >>> Image.gz can be used in FIT image for ARM64 but I am not sure which >>> platform actually uses it. >>> >>> This patch only enables support for RISC-V. >> >> Can't this be made generic ? >> > > I think so if I just move the gzip detection and decompression code to > cmd/booti.c. > > I will update the v3 patch with this. Nice, thanks. But since you're basically implementing file(1)-lite, I wonder whether there isn't similar code somewhere in u-boot already.
On 5/1/19 3:34 AM, Marek Vasut wrote: > On 5/1/19 12:06 AM, Atish Patra wrote: >> On 4/30/19 2:42 PM, Marek Vasut wrote: >>> On 4/30/19 10:38 PM, Atish Patra wrote: >>>> On 4/30/19 12:11 PM, Marek Vasut wrote: >>>>> On 4/30/19 8:13 PM, Atish Patra wrote: >>>>>> On 4/30/19 2:52 AM, Marek Vasut wrote: >>>>>>> On 4/30/19 3:27 AM, Atish Patra wrote: >>>>>>> >>>>>>> [...] >>>>>>> >>>>>>>>>> Yes. FIT image parsing can be done in that way. However, the idea >>>>>>>>>> was >>>>>>>>>> here to load Image.gz directly. Image.gz is default compressed >>>>>>>>>> Linux >>>>>>>>>> kernel image format in RISC-V. >>>>>>>>> >>>>>>>>> Sigh, and the image header is compressed as well, so there's no >>>>>>>>> way to >>>>>>>>> identify the image format, right ? And there's no decompressor, so >>>>>>>>> the >>>>>>>>> dcompressing has to be done by bootloader, which would need some >>>>>>>>> sort of >>>>>>>>> very smart way of figuring out which exact compression method is >>>>>>>>> used ? >>>>>>>>> >>>>>>>> Yes. Image.gz is always gunzip. So checking first two bytes is >>>>>>>> enough to >>>>>>>> confirm that it is a gz file. >>>>>>> >>>>>>> What happens once people start feeding it more exotic compression >>>>>>> methods, like LZ4 or LZO or LZMA for example ? >>>>>>> >>>>>> >>>>>> booti command help will clearly state that it can only boot kernel >>>>>> from >>>>>> Image or Image.gz. >>>>>> >>>>>> static char booti_help_text[] = >>>>>> "[addr [initrd[:size]] [fdt]]\n" >>>>>> - " - boot arm64 Linux Image stored in memory\n" >>>>>> + " - boot arm64 Linux Image or riscv Linux Image/Image.gz >>>>>> stored >>>>>> in memory\n" >>>>> >>>>> Obvious question -- does this Image.gz stuff apply to arm64 ? >>>>> >>>> >>>> arm64 builds Image.gz but booti for arm64 doesn't use it. I guess >>>> Image.gz can be used in FIT image for ARM64 but I am not sure which >>>> platform actually uses it. >>>> >>>> This patch only enables support for RISC-V. >>> >>> Can't this be made generic ? >>> >> >> I think so if I just move the gzip detection and decompression code to >> cmd/booti.c. >> >> I will update the v3 patch with this. > > Nice, thanks. > > But since you're basically implementing file(1)-lite, I wonder whether > there isn't similar code somewhere in u-boot already. > I was hoping to find one as well. But I couldn't. I can add a generic gz detection function(same file(1)-lite approach) to a lib/gunzip.c Regards, Atish
diff --git a/arch/riscv/lib/image.c b/arch/riscv/lib/image.c index e8802007c446..73ebd0da3885 100644 --- a/arch/riscv/lib/image.c +++ b/arch/riscv/lib/image.c @@ -9,6 +9,8 @@ #include <common.h> #include <mapmem.h> #include <errno.h> +#include <bootm.h> +#include <malloc.h> #include <linux/sizes.h> #include <linux/stddef.h> @@ -16,6 +18,8 @@ DECLARE_GLOBAL_DATA_PTR; /* ASCII version of "RISCV" defined in Linux kernel */ #define LINUX_RISCV_IMAGE_MAGIC 0x5643534952 +#define GZ_HEADER_0 0x1f +#define GZ_HEADER_1 0x8b struct linux_image_h { uint32_t code0; /* Executable code */ @@ -32,9 +36,31 @@ int booti_setup(ulong image, ulong *relocated_addr, ulong *size, bool force_reloc) { struct linux_image_h *lhdr; + uint8_t *temp; + void *dest; + ulong dest_end; + int ret; + /* TODO: Is there a way to figure out length of compressed Image.gz ? + * Otherwise, set it to SYS_BOOTM_LEN which should be sufficient. + */ + int len = CONFIG_SYS_BOOTM_LEN; + + temp = (uint8_t *)map_sysmem(image, 0); - lhdr = (struct linux_image_h *)map_sysmem(image, 0); + if (*(temp) == GZ_HEADER_0 && *(temp+1) == GZ_HEADER_1) { + /* Need a temporary location to copy the uncompressed image */ + dest = (void *)map_sysmem(image + 8 * CONFIG_SYS_BOOTM_LEN, 0); + ret = bootm_decomp_image(IH_COMP_GZIP, 0, image, IH_TYPE_KERNEL, + dest, (void *)image, len, + CONFIG_SYS_BOOTM_LEN, &dest_end); + if (ret) + return ret; + /* dest_end contains the uncompressed Image size */ + memmove((void *) image, dest, dest_end); + unmap_sysmem(dest); + } + lhdr = (struct linux_image_h *)temp; if (lhdr->magic != LINUX_RISCV_IMAGE_MAGIC) { puts("Bad Linux RISCV Image magic!\n"); return -EINVAL;
Add gz parsing logic so that booti can parse both Image and Image.gz. Signed-off-by: Atish Patra <atish.patra@wdc.com> --- arch/riscv/lib/image.c | 28 +++++++++++++++++++++++++++- 1 file changed, 27 insertions(+), 1 deletion(-)