| Message ID | 20090210095614.GA1995@www.tglx.de |
|---|---|
| State | RFC |
| Headers | show |
On Tue, 2009-02-10 at 10:56 +0100, Sebastian Andrzej Siewior wrote: > I have here a mtd part which is 3 GiB with a flash page size of 256KiB. > The 2GiB limit is at erase block 8192. In mtd_is_bad() the computation > for the MEMGETBADBLOCK ioctl() looks like the following: > > | seek = eb * mtd->eb_size; > > with both eb and mtd->eb_size being a signed int results in seek being a > signed result. Therefore I changed the type from signed to unsigned int. If you make "seek" to be 64 bit, and cast one of the multipliers to off_t, you should be fine. > The _FILE_OFFSET_BITS=64 define is required to switch off_t from 32bit > to 64bit an 32bit systems. This is required in order to keep using > lseek() as lseek64 on 32bit system. Without this change lseek() in > mtd_read() is called with a 32bit value with most significat bit set and > the kernel performs a sign extension for the 64bit value which is used > in the mtd layer. > > The last change also changes the size of the parameter which is passed > to the MEMGETBADBLOCK ioctl() from 32 to 64bit. The counter part in > kernel is also defined as loff_t which is of type __kernel_loff_t and > this is "long long". So this must have been broken for a while unless I > missed something. > > Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de> > --- > common.mk | 2 +- > ubi-utils/new-utils/include/libmtd.h | 10 ++++++---- > ubi-utils/new-utils/src/libmtd.c | 20 +++++++++++--------- > ubi-utils/new-utils/src/libscan.c | 5 +++-- > ubi-utils/new-utils/src/ubiformat.c | 9 ++++++--- > 5 files changed, 27 insertions(+), 19 deletions(-) > > diff --git a/common.mk b/common.mk > index 77d28bf..65fc1cc 100644 > --- a/common.mk > +++ b/common.mk > @@ -2,7 +2,7 @@ CC := $(CROSS)gcc > AR := $(CROSS)ar > RANLIB := $(CROSS)ranlib > CFLAGS ?= -O2 -g > -CFLAGS += -Wall -Wwrite-strings -W > +CFLAGS += -Wall -Wwrite-strings -W -D_FILE_OFFSET_BITS=64 > > DESTDIR ?= /usr/local > PREFIX=/usr > diff --git a/ubi-utils/new-utils/include/libmtd.h b/ubi-utils/new-utils/include/libmtd.h > index d3c6a63..94ccd45 100644 > --- a/ubi-utils/new-utils/include/libmtd.h > +++ b/ubi-utils/new-utils/include/libmtd.h > @@ -61,10 +61,12 @@ struct mtd_info > }; > > int mtd_get_info(const char *node, struct mtd_info *mtd); > -int mtd_erase(const struct mtd_info *mtd, int eb); > -int mtd_is_bad(const struct mtd_info *mtd, int eb); > -int mtd_read(const struct mtd_info *mtd, int eb, int offs, void *buf, int len); > -int mtd_write(const struct mtd_info *mtd, int eb, int offs, void *buf, int len); > +int mtd_erase(const struct mtd_info *mtd, unsigned int eb); > +int mtd_is_bad(const struct mtd_info *mtd, unsigned int eb); > +int mtd_read(const struct mtd_info *mtd, unsigned int eb, unsigned int offs, > + void *buf, int len); > +int mtd_write(const struct mtd_info *mtd, unsigned int eb, unsigned int offs, > + void *buf, int len); I do not think you need to make eraseblock number and offset to be unsigned. In fact, I'd like them to be signed, because this is the same we have in the kernel (in UBI/UBIFS), and I'd like to be more or less consistent. Your patch will be twice as short without this change, right?
Artem Bityutskiy wrote: > On Tue, 2009-02-10 at 10:56 +0100, Sebastian Andrzej Siewior wrote: >> I have here a mtd part which is 3 GiB with a flash page size of 256KiB. >> The 2GiB limit is at erase block 8192. In mtd_is_bad() the computation >> for the MEMGETBADBLOCK ioctl() looks like the following: >> >> | seek = eb * mtd->eb_size; >> >> with both eb and mtd->eb_size being a signed int results in seek being a >> signed result. Therefore I changed the type from signed to unsigned int. > If you make "seek" to be 64 bit, and cast one of the multipliers to > off_t, you should be fine. yup. >> The _FILE_OFFSET_BITS=64 define is required to switch off_t from 32bit >> to 64bit an 32bit systems. This is required in order to keep using >> lseek() as lseek64 on 32bit system. Without this change lseek() in >> mtd_read() is called with a 32bit value with most significat bit set and >> the kernel performs a sign extension for the 64bit value which is used >> in the mtd layer. >> >> The last change also changes the size of the parameter which is passed >> to the MEMGETBADBLOCK ioctl() from 32 to 64bit. The counter part in >> kernel is also defined as loff_t which is of type __kernel_loff_t and >> this is "long long". So this must have been broken for a while unless I >> missed something. >> >> Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de> >> --- >> common.mk | 2 +- >> ubi-utils/new-utils/include/libmtd.h | 10 ++++++---- >> ubi-utils/new-utils/src/libmtd.c | 20 +++++++++++--------- >> ubi-utils/new-utils/src/libscan.c | 5 +++-- >> ubi-utils/new-utils/src/ubiformat.c | 9 ++++++--- >> 5 files changed, 27 insertions(+), 19 deletions(-) >> >> diff --git a/common.mk b/common.mk >> index 77d28bf..65fc1cc 100644 >> --- a/common.mk >> +++ b/common.mk >> @@ -2,7 +2,7 @@ CC := $(CROSS)gcc >> AR := $(CROSS)ar >> RANLIB := $(CROSS)ranlib >> CFLAGS ?= -O2 -g >> -CFLAGS += -Wall -Wwrite-strings -W >> +CFLAGS += -Wall -Wwrite-strings -W -D_FILE_OFFSET_BITS=64 >> >> DESTDIR ?= /usr/local >> PREFIX=/usr >> diff --git a/ubi-utils/new-utils/include/libmtd.h b/ubi-utils/new-utils/include/libmtd.h >> index d3c6a63..94ccd45 100644 >> --- a/ubi-utils/new-utils/include/libmtd.h >> +++ b/ubi-utils/new-utils/include/libmtd.h >> @@ -61,10 +61,12 @@ struct mtd_info >> }; >> >> int mtd_get_info(const char *node, struct mtd_info *mtd); >> -int mtd_erase(const struct mtd_info *mtd, int eb); >> -int mtd_is_bad(const struct mtd_info *mtd, int eb); >> -int mtd_read(const struct mtd_info *mtd, int eb, int offs, void *buf, int len); >> -int mtd_write(const struct mtd_info *mtd, int eb, int offs, void *buf, int len); >> +int mtd_erase(const struct mtd_info *mtd, unsigned int eb); >> +int mtd_is_bad(const struct mtd_info *mtd, unsigned int eb); >> +int mtd_read(const struct mtd_info *mtd, unsigned int eb, unsigned int offs, >> + void *buf, int len); >> +int mtd_write(const struct mtd_info *mtd, unsigned int eb, unsigned int offs, >> + void *buf, int len); > > I do not think you need to make eraseblock number and offset to be > unsigned. In fact, I'd like them to be signed, because this is the same > we have in the kernel (in UBI/UBIFS), and I'd like to be more or less > consistent. That is a point. However those things should never be negative so maybe we could change this in kernel. While we here, I get a couple of "compare between signed and unsigned" warnings from gcc. I tried to clean them up but I end up with huge patches similar to this one. Are you aware of those or you simply don't get them? > Your patch will be twice as short without this change, right? Should be. I form a patch and we will see :) Sebatian
On Tue, 2009-02-10 at 11:17 +0100, Sebastian Andrzej Siewior wrote: > > I do not think you need to make eraseblock number and offset to be > > unsigned. In fact, I'd like them to be signed, because this is the same > > we have in the kernel (in UBI/UBIFS), and I'd like to be more or less > > consistent. > That is a point. However those things should never be negative so maybe we > could change this in kernel. They should not indeed be negative. However, signed numbers are just much more appropriate when you write the code, because you may put an error code there if something goes wrong. For example, the UBIFS Garbage collector returns the freed eraseblock number in case of success or a negative error code in case of failure. Compare: int ubifs_garbage_collect(struct ubifs_info *c) and int ubifs_garbage_collect(struct ubifs_info *c, unsigned int *leb) The first prototype is what we have now. The second is what we would have to have to be able to return LEB number and error code. And vs. changing UBI/UBIFS - that would be really a lot of work. Anyway, the thing is that 31-bit should be more than enough for eraseblock number. > While we here, I get a couple of "compare between signed and unsigned" > warnings from gcc. I tried to clean them up but I end up with huge patches > similar to this one. Are you aware of those or you simply don't get them? I'd rather shut the warnings up by an gcc option - is there such an option? I mean, IMO most of these warnings are not really sane, and cleaning that up just adds mess - you stop understanding why is something signed or unsigned :-)
Artem Bityutskiy wrote: > On Tue, 2009-02-10 at 11:17 +0100, Sebastian Andrzej Siewior wrote: >>> I do not think you need to make eraseblock number and offset to be >>> unsigned. In fact, I'd like them to be signed, because this is the same >>> we have in the kernel (in UBI/UBIFS), and I'd like to be more or less >>> consistent. >> That is a point. However those things should never be negative so maybe we >> could change this in kernel. > > They should not indeed be negative. However, signed numbers are just > much more appropriate when you write the code, because you may put an > error code there if something goes wrong. > > For example, the UBIFS Garbage collector returns the freed eraseblock > number in case of success or a negative error code in case of failure. > Compare: > > int ubifs_garbage_collect(struct ubifs_info *c) > > and > > int ubifs_garbage_collect(struct ubifs_info *c, unsigned int *leb) > > The first prototype is what we have now. The second is what we would > have to have to be able to return LEB number and error code. > > And vs. changing UBI/UBIFS - that would be really a lot of work. Yup, thanks for info. That makes sense. > Anyway, the thing is that 31-bit should be more than enough for > eraseblock number. For now :) > >> While we here, I get a couple of "compare between signed and unsigned" >> warnings from gcc. I tried to clean them up but I end up with huge patches >> similar to this one. Are you aware of those or you simply don't get them? > > I'd rather shut the warnings up by an gcc option - is there such an > option? You are using -W which deprecated and -Wextra is doing the same thing. One of the things -Wextra does is adding -Wsign-compare so you have to add -Wno-sign-compare to stop that. Sebastian
On Tue, 2009-02-10 at 11:49 +0100, Sebastian Andrzej Siewior wrote: > >> While we here, I get a couple of "compare between signed and unsigned" > >> warnings from gcc. I tried to clean them up but I end up with huge patches > >> similar to this one. Are you aware of those or you simply don't get them? > > > > I'd rather shut the warnings up by an gcc option - is there such an > > option? > You are using -W which deprecated and -Wextra is doing the same thing. One > of the things -Wextra does is adding -Wsign-compare so you have to add > -Wno-sign-compare to stop that. A patch is welcome :-)
diff --git a/common.mk b/common.mk index 77d28bf..65fc1cc 100644 --- a/common.mk +++ b/common.mk @@ -2,7 +2,7 @@ CC := $(CROSS)gcc AR := $(CROSS)ar RANLIB := $(CROSS)ranlib CFLAGS ?= -O2 -g -CFLAGS += -Wall -Wwrite-strings -W +CFLAGS += -Wall -Wwrite-strings -W -D_FILE_OFFSET_BITS=64 DESTDIR ?= /usr/local PREFIX=/usr diff --git a/ubi-utils/new-utils/include/libmtd.h b/ubi-utils/new-utils/include/libmtd.h index d3c6a63..94ccd45 100644 --- a/ubi-utils/new-utils/include/libmtd.h +++ b/ubi-utils/new-utils/include/libmtd.h @@ -61,10 +61,12 @@ struct mtd_info }; int mtd_get_info(const char *node, struct mtd_info *mtd); -int mtd_erase(const struct mtd_info *mtd, int eb); -int mtd_is_bad(const struct mtd_info *mtd, int eb); -int mtd_read(const struct mtd_info *mtd, int eb, int offs, void *buf, int len); -int mtd_write(const struct mtd_info *mtd, int eb, int offs, void *buf, int len); +int mtd_erase(const struct mtd_info *mtd, unsigned int eb); +int mtd_is_bad(const struct mtd_info *mtd, unsigned int eb); +int mtd_read(const struct mtd_info *mtd, unsigned int eb, unsigned int offs, + void *buf, int len); +int mtd_write(const struct mtd_info *mtd, unsigned int eb, unsigned int offs, + void *buf, int len); #ifdef __cplusplus } diff --git a/ubi-utils/new-utils/src/libmtd.c b/ubi-utils/new-utils/src/libmtd.c index aab4b0e..b7e2fa0 100644 --- a/ubi-utils/new-utils/src/libmtd.c +++ b/ubi-utils/new-utils/src/libmtd.c @@ -157,7 +157,7 @@ out_close: * This function erases the eraseblock and returns %0 in case of success and * %-1 in case of failure. */ -int mtd_erase(const struct mtd_info *mtd, int eb) +int mtd_erase(const struct mtd_info *mtd, unsigned int eb) { struct erase_info_user ei; @@ -174,12 +174,12 @@ int mtd_erase(const struct mtd_info *mtd, int eb) * This function checks if eraseblock @eb is bad. Returns %0 if not, %1 if yes, * and %-1 in case of failure. */ -int mtd_is_bad(const struct mtd_info *mtd, int eb) +int mtd_is_bad(const struct mtd_info *mtd, unsigned int eb) { int ret; loff_t seek; - if (eb < 0 || eb >= mtd->eb_cnt) { + if (eb >= mtd->eb_cnt) { errmsg("bad eraseblock number %d, mtd%d has %d eraseblocks", eb, mtd->num, mtd->eb_cnt); errno = EINVAL; @@ -212,18 +212,19 @@ int mtd_is_bad(const struct mtd_info *mtd, int eb) * of the MTD device defined by @mtd and stores the read data at buffer @buf. * Returns %0 in case of success and %-1 in case of failure. */ -int mtd_read(const struct mtd_info *mtd, int eb, int offs, void *buf, int len) +int mtd_read(const struct mtd_info *mtd, unsigned int eb, unsigned int offs, + void *buf, int len) { int ret, rd = 0; off_t seek; - if (eb < 0 || eb >= mtd->eb_cnt) { + if (eb >= mtd->eb_cnt) { errmsg("bad eraseblock number %d, mtd%d has %d eraseblocks", eb, mtd->num, mtd->eb_cnt); errno = EINVAL; return -1; } - if (offs < 0 || offs + len > mtd->eb_size) { + if (offs + len > mtd->eb_size) { errmsg("bad offset %d or length %d, mtd%d eraseblock size is %d", offs, len, mtd->num, mtd->eb_size); errno = EINVAL; @@ -263,18 +264,19 @@ int mtd_read(const struct mtd_info *mtd, int eb, int offs, void *buf, int len) * of the MTD device defined by @mtd. Returns %0 in case of success and %-1 in * case of failure. */ -int mtd_write(const struct mtd_info *mtd, int eb, int offs, void *buf, int len) +int mtd_write(const struct mtd_info *mtd, unsigned int eb, unsigned int offs, + void *buf, int len) { int ret; off_t seek; - if (eb < 0 || eb >= mtd->eb_cnt) { + if (eb >= mtd->eb_cnt) { errmsg("bad eraseblock number %d, mtd%d has %d eraseblocks", eb, mtd->num, mtd->eb_cnt); errno = EINVAL; return -1; } - if (offs < 0 || offs + len > mtd->eb_size) { + if (offs + len > mtd->eb_size) { errmsg("bad offset %d or length %d, mtd%d eraseblock size is %d", offs, len, mtd->num, mtd->eb_size); errno = EINVAL; diff --git a/ubi-utils/new-utils/src/libscan.c b/ubi-utils/new-utils/src/libscan.c index da13b1b..59322ca 100644 --- a/ubi-utils/new-utils/src/libscan.c +++ b/ubi-utils/new-utils/src/libscan.c @@ -50,7 +50,8 @@ static int all_ff(const void *buf, int len) int ubi_scan(struct mtd_info *mtd, struct ubi_scan_info **info, int verbose) { - int eb, v = (verbose == 2), pr = (verbose == 1); + int v = (verbose == 2), pr = (verbose == 1); + unsigned int eb; struct ubi_scan_info *si; unsigned long long sum = 0; @@ -81,7 +82,7 @@ int ubi_scan(struct mtd_info *mtd, struct ubi_scan_info **info, int verbose) } if (pr) { printf("\r" PROGRAM_NAME ": scanning eraseblock %d -- %2lld %% complete ", - eb, (long long)(eb + 1) * 100 / mtd->eb_cnt); + eb, (unsigned long long)(eb + 1) * 100 / mtd->eb_cnt); fflush(stdout); } diff --git a/ubi-utils/new-utils/src/ubiformat.c b/ubi-utils/new-utils/src/ubiformat.c index 0074c7a..bc6fed3 100644 --- a/ubi-utils/new-utils/src/ubiformat.c +++ b/ubi-utils/new-utils/src/ubiformat.c @@ -356,7 +356,8 @@ static int read_all(int fd, void *buf, size_t len) static int flash_image(const struct mtd_info *mtd, struct ubi_scan_info *si) { - int fd, img_ebs, eb, written_ebs = 0, divisor; + int fd, img_ebs, written_ebs = 0, divisor; + unsigned int eb; off_t st_size; fd = open_file(&st_size); @@ -462,7 +463,8 @@ out_close: static int format(const struct mtd_info *mtd, const struct ubigen_info *ui, const struct ubi_scan_info *si, int start_eb, int novtbl) { - int eb, err, write_size; + int err, write_size; + unsigned int eb; struct ubi_ec_hdr *hdr; struct ubi_vtbl_record *vtbl; int eb1 = -1, eb2 = -1; @@ -482,7 +484,8 @@ static int format(const struct mtd_info *mtd, const struct ubigen_info *ui, if (!args.quiet && !args.verbose) { printf("\r" PROGRAM_NAME ": formatting eraseblock %d -- %2lld %% complete ", - eb, (long long)(eb + 1 - start_eb) * 100 / (mtd->eb_cnt - start_eb)); + eb, (unsigned long long)(eb + 1 - start_eb) * + 100 / (mtd->eb_cnt - start_eb)); fflush(stdout); }
I have here a mtd part which is 3 GiB with a flash page size of 256KiB. The 2GiB limit is at erase block 8192. In mtd_is_bad() the computation for the MEMGETBADBLOCK ioctl() looks like the following: | seek = eb * mtd->eb_size; with both eb and mtd->eb_size being a signed int results in seek being a signed result. Therefore I changed the type from signed to unsigned int. The _FILE_OFFSET_BITS=64 define is required to switch off_t from 32bit to 64bit an 32bit systems. This is required in order to keep using lseek() as lseek64 on 32bit system. Without this change lseek() in mtd_read() is called with a 32bit value with most significat bit set and the kernel performs a sign extension for the 64bit value which is used in the mtd layer. The last change also changes the size of the parameter which is passed to the MEMGETBADBLOCK ioctl() from 32 to 64bit. The counter part in kernel is also defined as loff_t which is of type __kernel_loff_t and this is "long long". So this must have been broken for a while unless I missed something. Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de> --- common.mk | 2 +- ubi-utils/new-utils/include/libmtd.h | 10 ++++++---- ubi-utils/new-utils/src/libmtd.c | 20 +++++++++++--------- ubi-utils/new-utils/src/libscan.c | 5 +++-- ubi-utils/new-utils/src/ubiformat.c | 9 ++++++--- 5 files changed, 27 insertions(+), 19 deletions(-)