| Message ID | 20190523025113.4605-4-scott.branden@broadcom.com (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
| Series | fs: add partial file read support | expand |
On Wed, May 22, 2019 at 07:51:13PM -0700, Scott Branden wrote: > Adjust request_firmware_into_buf API to allow for portions > of firmware file to be read into a buffer. mdt_loader still > retricts request fo whole file read into buffer. > > Signed-off-by: Scott Branden <scott.branden@broadcom.com> > --- > drivers/soc/qcom/mdt_loader.c | 7 +++++-- > 1 file changed, 5 insertions(+), 2 deletions(-) > > diff --git a/drivers/soc/qcom/mdt_loader.c b/drivers/soc/qcom/mdt_loader.c > index 1c488024c698..ad20d159699c 100644 > --- a/drivers/soc/qcom/mdt_loader.c > +++ b/drivers/soc/qcom/mdt_loader.c > @@ -172,8 +172,11 @@ static int __qcom_mdt_load(struct device *dev, const struct firmware *fw, > > if (phdr->p_filesz) { > sprintf(fw_name + fw_name_len - 3, "b%02d", i); > - ret = request_firmware_into_buf(&seg_fw, fw_name, dev, > - ptr, phdr->p_filesz); > + ret = request_firmware_into_buf > + (&seg_fw, fw_name, dev, > + ptr, phdr->p_filesz, > + 0, > + KERNEL_PREAD_FLAG_WHOLE); So, all that work in the first 2 patches for no real change at all? Why are these changes even needed? And didn't you break this driver in patch 2/3? You can't fix it up later here, you need to also resolve that in the 2nd patch. thanks, greg k-h
Hi Greg, On 2019-05-22 10:52 p.m., Greg Kroah-Hartman wrote: > On Wed, May 22, 2019 at 07:51:13PM -0700, Scott Branden wrote: >> Adjust request_firmware_into_buf API to allow for portions >> of firmware file to be read into a buffer. mdt_loader still >> retricts request fo whole file read into buffer. >> >> Signed-off-by: Scott Branden <scott.branden@broadcom.com> >> --- >> drivers/soc/qcom/mdt_loader.c | 7 +++++-- >> 1 file changed, 5 insertions(+), 2 deletions(-) >> >> diff --git a/drivers/soc/qcom/mdt_loader.c b/drivers/soc/qcom/mdt_loader.c >> index 1c488024c698..ad20d159699c 100644 >> --- a/drivers/soc/qcom/mdt_loader.c >> +++ b/drivers/soc/qcom/mdt_loader.c >> @@ -172,8 +172,11 @@ static int __qcom_mdt_load(struct device *dev, const struct firmware *fw, >> >> if (phdr->p_filesz) { >> sprintf(fw_name + fw_name_len - 3, "b%02d", i); >> - ret = request_firmware_into_buf(&seg_fw, fw_name, dev, >> - ptr, phdr->p_filesz); >> + ret = request_firmware_into_buf >> + (&seg_fw, fw_name, dev, >> + ptr, phdr->p_filesz, >> + 0, >> + KERNEL_PREAD_FLAG_WHOLE); > So, all that work in the first 2 patches for no real change at all? Why > are these changes even needed? The first two patches allow partial read of files into memory. Existing kernel drivers haven't need such functionality so, yes, there should be no real change with first two patches other than adding such partial file read support. We have a new driver in development which needs partial read of files supported in the kernel. > > And didn't you break this driver in patch 2/3? You can't fix it up > later here, you need to also resolve that in the 2nd patch. I thought the driver changes needs to be in a different patch. If required I can squash this driver change in with the request_firmware_into_buf change. But the 2nd patch won't work without the 1st patch either. So that would mean you now want all 3 patches for different subsystems squashed together? Please let me know how you would like the patch series submitted. > > thanks, > > greg k-h Regards, Scott
On Thu, May 23, 2019 at 09:41:49AM -0700, Scott Branden wrote: > Hi Greg, > > On 2019-05-22 10:52 p.m., Greg Kroah-Hartman wrote: > > On Wed, May 22, 2019 at 07:51:13PM -0700, Scott Branden wrote: > > > Adjust request_firmware_into_buf API to allow for portions > > > of firmware file to be read into a buffer. mdt_loader still > > > retricts request fo whole file read into buffer. > > > > > > Signed-off-by: Scott Branden <scott.branden@broadcom.com> > > > --- > > > drivers/soc/qcom/mdt_loader.c | 7 +++++-- > > > 1 file changed, 5 insertions(+), 2 deletions(-) > > > > > > diff --git a/drivers/soc/qcom/mdt_loader.c b/drivers/soc/qcom/mdt_loader.c > > > index 1c488024c698..ad20d159699c 100644 > > > --- a/drivers/soc/qcom/mdt_loader.c > > > +++ b/drivers/soc/qcom/mdt_loader.c > > > @@ -172,8 +172,11 @@ static int __qcom_mdt_load(struct device *dev, const struct firmware *fw, > > > if (phdr->p_filesz) { > > > sprintf(fw_name + fw_name_len - 3, "b%02d", i); > > > - ret = request_firmware_into_buf(&seg_fw, fw_name, dev, > > > - ptr, phdr->p_filesz); > > > + ret = request_firmware_into_buf > > > + (&seg_fw, fw_name, dev, > > > + ptr, phdr->p_filesz, > > > + 0, > > > + KERNEL_PREAD_FLAG_WHOLE); > > So, all that work in the first 2 patches for no real change at all? Why > > are these changes even needed? > > The first two patches allow partial read of files into memory. > > Existing kernel drivers haven't need such functionality so, yes, there > should be no real change > > with first two patches other than adding such partial file read support. > > We have a new driver in development which needs partial read of files > supported in the kernel. As I said before, I can not take new apis without any in-kernel user. So let's wait for your new code that thinks it needs this, and then we will be glad to evaluate all of this at that point in time. To do so otherwise is to have loads of unused "features" aquiring cruft in the kernel source, and you do not want that. > > And didn't you break this driver in patch 2/3? You can't fix it up > > later here, you need to also resolve that in the 2nd patch. > > I thought the driver changes needs to be in a different patch. If required I > can squash this You can NEVER break the build with any individual kernel patch, that should be well known by now as we have been doing this for over a decade. thanks, greg k-h
On Thu 23 May 09:56 PDT 2019, Greg Kroah-Hartman wrote: > On Thu, May 23, 2019 at 09:41:49AM -0700, Scott Branden wrote: > > Hi Greg, > > > > On 2019-05-22 10:52 p.m., Greg Kroah-Hartman wrote: > > > On Wed, May 22, 2019 at 07:51:13PM -0700, Scott Branden wrote: > > > > Adjust request_firmware_into_buf API to allow for portions > > > > of firmware file to be read into a buffer. mdt_loader still > > > > retricts request fo whole file read into buffer. > > > > > > > > Signed-off-by: Scott Branden <scott.branden@broadcom.com> > > > > --- > > > > drivers/soc/qcom/mdt_loader.c | 7 +++++-- > > > > 1 file changed, 5 insertions(+), 2 deletions(-) > > > > > > > > diff --git a/drivers/soc/qcom/mdt_loader.c b/drivers/soc/qcom/mdt_loader.c > > > > index 1c488024c698..ad20d159699c 100644 > > > > --- a/drivers/soc/qcom/mdt_loader.c > > > > +++ b/drivers/soc/qcom/mdt_loader.c > > > > @@ -172,8 +172,11 @@ static int __qcom_mdt_load(struct device *dev, const struct firmware *fw, > > > > if (phdr->p_filesz) { > > > > sprintf(fw_name + fw_name_len - 3, "b%02d", i); > > > > - ret = request_firmware_into_buf(&seg_fw, fw_name, dev, > > > > - ptr, phdr->p_filesz); > > > > + ret = request_firmware_into_buf > > > > + (&seg_fw, fw_name, dev, > > > > + ptr, phdr->p_filesz, > > > > + 0, > > > > + KERNEL_PREAD_FLAG_WHOLE); > > > So, all that work in the first 2 patches for no real change at all? Why > > > are these changes even needed? > > > > The first two patches allow partial read of files into memory. > > > > Existing kernel drivers haven't need such functionality so, yes, there > > should be no real change > > > > with first two patches other than adding such partial file read support. > > > > We have a new driver in development which needs partial read of files > > supported in the kernel. > > As I said before, I can not take new apis without any in-kernel user. > So let's wait for your new code that thinks it needs this, and then we > will be glad to evaluate all of this at that point in time. > The .mdt files are ELF files split to avoid having to allocate large (5-60MB) chunks of temporary firmware buffers while installing the segments. But for multiple reasons it would be nice to be able to load the non-split ELF files and the proposed interface would allow this. So I definitely like the gist of the series. > To do so otherwise is to have loads of unused "features" aquiring cruft > in the kernel source, and you do not want that. > Agreed. I'll take the opportunity and see if I can implement this (support for non-split Qualcomm firmware) based on the patches in this series. Regards, Bjorn
Hi Bjorn, On 2019-05-26 10:36 p.m., Bjorn Andersson wrote: > On Thu 23 May 09:56 PDT 2019, Greg Kroah-Hartman wrote: > >> On Thu, May 23, 2019 at 09:41:49AM -0700, Scott Branden wrote: >>> Hi Greg, >>> >>> On 2019-05-22 10:52 p.m., Greg Kroah-Hartman wrote: >>>> On Wed, May 22, 2019 at 07:51:13PM -0700, Scott Branden wrote: >>>>> Adjust request_firmware_into_buf API to allow for portions >>>>> of firmware file to be read into a buffer. mdt_loader still >>>>> retricts request fo whole file read into buffer. >>>>> >>>>> Signed-off-by: Scott Branden <scott.branden@broadcom.com> >>>>> --- >>>>> drivers/soc/qcom/mdt_loader.c | 7 +++++-- >>>>> 1 file changed, 5 insertions(+), 2 deletions(-) >>>>> >>>>> diff --git a/drivers/soc/qcom/mdt_loader.c b/drivers/soc/qcom/mdt_loader.c >>>>> index 1c488024c698..ad20d159699c 100644 >>>>> --- a/drivers/soc/qcom/mdt_loader.c >>>>> +++ b/drivers/soc/qcom/mdt_loader.c >>>>> @@ -172,8 +172,11 @@ static int __qcom_mdt_load(struct device *dev, const struct firmware *fw, >>>>> if (phdr->p_filesz) { >>>>> sprintf(fw_name + fw_name_len - 3, "b%02d", i); >>>>> - ret = request_firmware_into_buf(&seg_fw, fw_name, dev, >>>>> - ptr, phdr->p_filesz); >>>>> + ret = request_firmware_into_buf >>>>> + (&seg_fw, fw_name, dev, >>>>> + ptr, phdr->p_filesz, >>>>> + 0, >>>>> + KERNEL_PREAD_FLAG_WHOLE); >>>> So, all that work in the first 2 patches for no real change at all? Why >>>> are these changes even needed? >>> The first two patches allow partial read of files into memory. >>> >>> Existing kernel drivers haven't need such functionality so, yes, there >>> should be no real change >>> >>> with first two patches other than adding such partial file read support. >>> >>> We have a new driver in development which needs partial read of files >>> supported in the kernel. >> As I said before, I can not take new apis without any in-kernel user. >> So let's wait for your new code that thinks it needs this, and then we >> will be glad to evaluate all of this at that point in time. >> > The .mdt files are ELF files split to avoid having to allocate large > (5-60MB) chunks of temporary firmware buffers while installing the > segments. > > But for multiple reasons it would be nice to be able to load the > non-split ELF files and the proposed interface would allow this. > > So I definitely like the gist of the series. > >> To do so otherwise is to have loads of unused "features" aquiring cruft >> in the kernel source, and you do not want that. >> > Agreed. > > I'll take the opportunity and see if I can implement this (support for > non-split Qualcomm firmware) based on the patches in this series. I'm back from my leave now. Have you managed to utilize my partial firmware read in your driver yet? > > Regards, > Bjorn
Hi Greg, On 2019-05-23 9:56 a.m., Greg Kroah-Hartman wrote: > On Thu, May 23, 2019 at 09:41:49AM -0700, Scott Branden wrote: >> Hi Greg, >> >> On 2019-05-22 10:52 p.m., Greg Kroah-Hartman wrote: >>> On Wed, May 22, 2019 at 07:51:13PM -0700, Scott Branden wrote: >>>> Adjust request_firmware_into_buf API to allow for portions >>>> of firmware file to be read into a buffer. mdt_loader still >>>> retricts request fo whole file read into buffer. >>>> >>>> Signed-off-by: Scott Branden <scott.branden@broadcom.com> >>>> --- >>>> drivers/soc/qcom/mdt_loader.c | 7 +++++-- >>>> 1 file changed, 5 insertions(+), 2 deletions(-) >>>> >>>> diff --git a/drivers/soc/qcom/mdt_loader.c b/drivers/soc/qcom/mdt_loader.c >>>> index 1c488024c698..ad20d159699c 100644 >>>> --- a/drivers/soc/qcom/mdt_loader.c >>>> +++ b/drivers/soc/qcom/mdt_loader.c >>>> @@ -172,8 +172,11 @@ static int __qcom_mdt_load(struct device *dev, const struct firmware *fw, >>>> if (phdr->p_filesz) { >>>> sprintf(fw_name + fw_name_len - 3, "b%02d", i); >>>> - ret = request_firmware_into_buf(&seg_fw, fw_name, dev, >>>> - ptr, phdr->p_filesz); >>>> + ret = request_firmware_into_buf >>>> + (&seg_fw, fw_name, dev, >>>> + ptr, phdr->p_filesz, >>>> + 0, >>>> + KERNEL_PREAD_FLAG_WHOLE); >>> So, all that work in the first 2 patches for no real change at all? Why >>> are these changes even needed? >> The first two patches allow partial read of files into memory. >> >> Existing kernel drivers haven't need such functionality so, yes, there >> should be no real change >> >> with first two patches other than adding such partial file read support. >> >> We have a new driver in development which needs partial read of files >> supported in the kernel. > As I said before, I can not take new apis without any in-kernel user. > So let's wait for your new code that thinks it needs this, and then we > will be glad to evaluate all of this at that point in time. I have submitted all the necessary patches you requested here. These include first adding tests for existing API that never had a kernel selftest: https://lkml.org/lkml/2019/8/22/1367 Followed by API enhancement, tests updated, and a new driver requiring enhanced API: https://lkml.org/lkml/2019/8/22/1404 > > To do so otherwise is to have loads of unused "features" aquiring cruft > in the kernel source, and you do not want that. > > thanks, > > greg k-h Thanks, Scott
diff --git a/drivers/soc/qcom/mdt_loader.c b/drivers/soc/qcom/mdt_loader.c index 1c488024c698..ad20d159699c 100644 --- a/drivers/soc/qcom/mdt_loader.c +++ b/drivers/soc/qcom/mdt_loader.c @@ -172,8 +172,11 @@ static int __qcom_mdt_load(struct device *dev, const struct firmware *fw, if (phdr->p_filesz) { sprintf(fw_name + fw_name_len - 3, "b%02d", i); - ret = request_firmware_into_buf(&seg_fw, fw_name, dev, - ptr, phdr->p_filesz); + ret = request_firmware_into_buf + (&seg_fw, fw_name, dev, + ptr, phdr->p_filesz, + 0, + KERNEL_PREAD_FLAG_WHOLE); if (ret) { dev_err(dev, "failed to load %s\n", fw_name); break;
Adjust request_firmware_into_buf API to allow for portions of firmware file to be read into a buffer. mdt_loader still retricts request fo whole file read into buffer. Signed-off-by: Scott Branden <scott.branden@broadcom.com> --- drivers/soc/qcom/mdt_loader.c | 7 +++++-- 1 file changed, 5 insertions(+), 2 deletions(-)