Message ID | 20180321020822.86255-3-bjsdjshi@linux.vnet.ibm.com (mailing list archive) |
---|---|
State | New, archived |
Headers | show |
On Wed, 21 Mar 2018 03:08:20 +0100 Dong Jia Shi <bjsdjshi@linux.vnet.ibm.com> wrote: > This refactors pfn_array_alloc_pin() and also improves it by adding > defensive code in error handling so that calling pfn_array_unpin_free() > after error return won't lead to problem. This mains does: > 1. Merge pfn_array_pin() into pfn_array_alloc_pin(), since there is no > other user of pfn_array_pin(). As a result, also remove kernel-doc > for pfn_array_pin() and add kernel-doc for pfn_array_alloc_pin(). > 2. For a vfio_pin_pages() failure, set pa->pa_nr to zero to indicate > zero pages were pinned. > 3. Set pa->pa_iova_pfn to NULL right after it was freed. > > Signed-off-by: Dong Jia Shi <bjsdjshi@linux.vnet.ibm.com> > --- > drivers/s390/cio/vfio_ccw_cp.c | 84 ++++++++++++++++++------------------------ > 1 file changed, 36 insertions(+), 48 deletions(-) > > diff --git a/drivers/s390/cio/vfio_ccw_cp.c b/drivers/s390/cio/vfio_ccw_cp.c > index 2be114db02f9..3abc9770910a 100644 > --- a/drivers/s390/cio/vfio_ccw_cp.c > +++ b/drivers/s390/cio/vfio_ccw_cp.c > @@ -46,65 +46,32 @@ struct ccwchain { > }; > > /* > - * pfn_array_pin() - pin user pages in memory > + * pfn_array_alloc_pin() - alloc memory for PFNs, then pin user pages in memory > * @pa: pfn_array on which to perform the operation > * @mdev: the mediated device to perform pin/unpin operations > + * @iova: target guest physical address > + * @len: number of bytes that should be pinned from @iova > * > - * Attempt to pin user pages in memory. > + * Attempt to allocate memory for PFNs, and pin user pages in memory. > * > * Usage of pfn_array: > - * @pa->pa_iova starting guest physical I/O address. Assigned by caller. > + * @pa->pa_iova starting guest physical I/O address. Assigned by callee. > * @pa->pa_iova_pfn array that stores PFNs of the pages need to pin. Allocated > - * by caller. > + * by callee. > * @pa->pa_pfn array that receives PFNs of the pages pinned. Allocated by > - * caller. > - * @pa->pa_nr number of pages from @pa->pa_iova to pin. Assigned by > - * caller. > - * number of pages pinned. Assigned by callee. > + * callee. > + * @pa->pa_nr initiated as 0 by caller. s/initiated/initialized/ but see below > + * number of pages pinned from @pa->pa_iova. Assigned by callee. So, basically everything is filled by pfn_array_alloc_pin()? Should we expect a clean struct pfn_array handed in by the caller, then (not just pa_nr == 0)? Would it make sense to describe the contents of the struct pfn_array fields at the struct's definition instead? You could then shorten the description here to "we expect pa_nr == 0, any field in this structure will be filled in by this function". > * > * Returns: > * Number of pages pinned on success. > - * If @pa->pa_nr is 0 or negative, returns 0. > + * If @pa->pa_nr is not 0 initially, returns -EINVAL. > * If no pages were pinned, returns -errno. > */ The change itself looks good to me.
On Tue, 27 Mar 2018 11:00:26 +0800 Dong Jia Shi <bjsdjshi@linux.vnet.ibm.com> wrote: > * Cornelia Huck <cohuck@redhat.com> [2018-03-26 15:28:46 +0200]: > > > On Wed, 21 Mar 2018 03:08:20 +0100 > > Dong Jia Shi <bjsdjshi@linux.vnet.ibm.com> wrote: > > > > > This refactors pfn_array_alloc_pin() and also improves it by adding > > > defensive code in error handling so that calling pfn_array_unpin_free() > > > after error return won't lead to problem. This mains does: > > > 1. Merge pfn_array_pin() into pfn_array_alloc_pin(), since there is no > > > other user of pfn_array_pin(). As a result, also remove kernel-doc > > > for pfn_array_pin() and add kernel-doc for pfn_array_alloc_pin(). > > > 2. For a vfio_pin_pages() failure, set pa->pa_nr to zero to indicate > > > zero pages were pinned. > > > 3. Set pa->pa_iova_pfn to NULL right after it was freed. > > > > > > Signed-off-by: Dong Jia Shi <bjsdjshi@linux.vnet.ibm.com> > > > --- > > > drivers/s390/cio/vfio_ccw_cp.c | 84 ++++++++++++++++++------------------------ > > > 1 file changed, 36 insertions(+), 48 deletions(-) > > > > > > diff --git a/drivers/s390/cio/vfio_ccw_cp.c b/drivers/s390/cio/vfio_ccw_cp.c > > > index 2be114db02f9..3abc9770910a 100644 > > > --- a/drivers/s390/cio/vfio_ccw_cp.c > > > +++ b/drivers/s390/cio/vfio_ccw_cp.c > > > @@ -46,65 +46,32 @@ struct ccwchain { > > > }; > > > > > > /* > > > - * pfn_array_pin() - pin user pages in memory > > > + * pfn_array_alloc_pin() - alloc memory for PFNs, then pin user pages in memory > > > * @pa: pfn_array on which to perform the operation > > > * @mdev: the mediated device to perform pin/unpin operations > > > + * @iova: target guest physical address > > > + * @len: number of bytes that should be pinned from @iova > > > * > > > - * Attempt to pin user pages in memory. > > > + * Attempt to allocate memory for PFNs, and pin user pages in memory. > > > * > > > * Usage of pfn_array: > > > - * @pa->pa_iova starting guest physical I/O address. Assigned by caller. > > > + * @pa->pa_iova starting guest physical I/O address. Assigned by callee. > > > * @pa->pa_iova_pfn array that stores PFNs of the pages need to pin. Allocated > > > - * by caller. > > > + * by callee. > > > * @pa->pa_pfn array that receives PFNs of the pages pinned. Allocated by > > > - * caller. > > > - * @pa->pa_nr number of pages from @pa->pa_iova to pin. Assigned by > > > - * caller. > > > - * number of pages pinned. Assigned by callee. > > > + * callee. > > > + * @pa->pa_nr initiated as 0 by caller. > > > > s/initiated/initialized/ > Ok. > > > > > but see below > > > > > + * number of pages pinned from @pa->pa_iova. Assigned by callee. > > > > So, basically everything is filled by pfn_array_alloc_pin()? > Yes. > > > Should we expect a clean struct pfn_array handed in by the caller, > > then (not just pa_nr == 0)? > The current idea is: > - It is a clean struct that pfn_array_alloc_pin() expects from its > caller. > - pfn_array_alloc_pin() and pfn_array_unpin_free() should be used in > pair. They are the only functions those change the values of the > elements of a pfn_array struct. > - Caller of pfn_array_alloc_pin() should either hand in a new allocated > pfn_array (zeroed out), or a freed-after-used one. > - So using pa_nr == 0, is enough to identify all the good cases. > [We set pa_nr to 0 in pfn_array_unpin_free().] > > Validating all of the elements only helps when there is case that a > caller breaks the usage rule of these interfaces - the caller itself > assigns values for pfn_pa elements directly... I don't think we allow > this to happen. > > So I think the current logic is fine. Yes, I think it is fine -- I was mainly wondering whether we wanted more sanity checks. > > > > > Would it make sense to describe the contents of the struct pfn_array > > fields at the struct's definition instead? You could then shorten the > > description here to "we expect pa_nr == 0, any field in this structure > > will be filled in by this function". > Sounds good! > Do you want a separated patch for this, or I do this change on this > patch? Either will be ok with me. Perhaps as an additional patch in front of this one?
On Wed, 28 Mar 2018 10:36:38 +0800 Dong Jia Shi <bjsdjshi@linux.vnet.ibm.com> wrote: > * Cornelia Huck <cohuck@redhat.com> [2018-03-27 12:01:27 +0200]: > > [...] > > > > > > > > > So, basically everything is filled by pfn_array_alloc_pin()? > > > Yes. > > > > > > > Should we expect a clean struct pfn_array handed in by the caller, > > > > then (not just pa_nr == 0)? > > > The current idea is: > > > - It is a clean struct that pfn_array_alloc_pin() expects from its > > > caller. > > > - pfn_array_alloc_pin() and pfn_array_unpin_free() should be used in > > > pair. They are the only functions those change the values of the > > > elements of a pfn_array struct. > > > - Caller of pfn_array_alloc_pin() should either hand in a new allocated > > > pfn_array (zeroed out), or a freed-after-used one. > > > - So using pa_nr == 0, is enough to identify all the good cases. > > > [We set pa_nr to 0 in pfn_array_unpin_free().] > > > > > > Validating all of the elements only helps when there is case that a > > > caller breaks the usage rule of these interfaces - the caller itself > > > assigns values for pfn_pa elements directly... I don't think we allow > > > this to happen. > > > > > > So I think the current logic is fine. > > > > Yes, I think it is fine -- I was mainly wondering whether we wanted > > more sanity checks. > > > Ok. > Check on (pa->pa_iova_pfn != NULL) could be added. It's easy to do so. > Check on pa->pa_iova doesn't make sense, since its value will be > re-assigned anyway. > Check on pa->pa_pfn doesn't make sense, since we treat it as a pointer > that points to part of the memory area that was pointed by > pa->pa_iova_pfn. And we will re-assign it with new pa->pa_iova_pfn > value. Yeah, so additional checks are probably not very useful. > > > > > > > > > > > > Would it make sense to describe the contents of the struct pfn_array > > > > fields at the struct's definition instead? You could then shorten the > > > > description here to "we expect pa_nr == 0, any field in this structure > > > > will be filled in by this function". > > > Sounds good! > > > Do you want a separated patch for this, or I do this change on this > > > patch? Either will be ok with me. > > > > Perhaps as an additional patch in front of this one? > > > It's doable. I will do that. > Cool, thx!
diff --git a/drivers/s390/cio/vfio_ccw_cp.c b/drivers/s390/cio/vfio_ccw_cp.c index 2be114db02f9..3abc9770910a 100644 --- a/drivers/s390/cio/vfio_ccw_cp.c +++ b/drivers/s390/cio/vfio_ccw_cp.c @@ -46,65 +46,32 @@ struct ccwchain { }; /* - * pfn_array_pin() - pin user pages in memory + * pfn_array_alloc_pin() - alloc memory for PFNs, then pin user pages in memory * @pa: pfn_array on which to perform the operation * @mdev: the mediated device to perform pin/unpin operations + * @iova: target guest physical address + * @len: number of bytes that should be pinned from @iova * - * Attempt to pin user pages in memory. + * Attempt to allocate memory for PFNs, and pin user pages in memory. * * Usage of pfn_array: - * @pa->pa_iova starting guest physical I/O address. Assigned by caller. + * @pa->pa_iova starting guest physical I/O address. Assigned by callee. * @pa->pa_iova_pfn array that stores PFNs of the pages need to pin. Allocated - * by caller. + * by callee. * @pa->pa_pfn array that receives PFNs of the pages pinned. Allocated by - * caller. - * @pa->pa_nr number of pages from @pa->pa_iova to pin. Assigned by - * caller. - * number of pages pinned. Assigned by callee. + * callee. + * @pa->pa_nr initiated as 0 by caller. + * number of pages pinned from @pa->pa_iova. Assigned by callee. * * Returns: * Number of pages pinned on success. - * If @pa->pa_nr is 0 or negative, returns 0. + * If @pa->pa_nr is not 0 initially, returns -EINVAL. * If no pages were pinned, returns -errno. */ -static int pfn_array_pin(struct pfn_array *pa, struct device *mdev) -{ - int i, ret; - - if (pa->pa_nr <= 0) { - pa->pa_nr = 0; - return 0; - } - - pa->pa_iova_pfn[0] = pa->pa_iova >> PAGE_SHIFT; - for (i = 1; i < pa->pa_nr; i++) - pa->pa_iova_pfn[i] = pa->pa_iova_pfn[i - 1] + 1; - - ret = vfio_pin_pages(mdev, pa->pa_iova_pfn, pa->pa_nr, - IOMMU_READ | IOMMU_WRITE, pa->pa_pfn); - - if (ret > 0 && ret != pa->pa_nr) { - vfio_unpin_pages(mdev, pa->pa_iova_pfn, ret); - pa->pa_nr = 0; - return 0; - } - - return ret; -} - -/* Unpin the pages before releasing the memory. */ -static void pfn_array_unpin_free(struct pfn_array *pa, struct device *mdev) -{ - vfio_unpin_pages(mdev, pa->pa_iova_pfn, pa->pa_nr); - pa->pa_nr = 0; - kfree(pa->pa_iova_pfn); -} - -/* Alloc memory for PFNs, then pin pages with them. */ static int pfn_array_alloc_pin(struct pfn_array *pa, struct device *mdev, u64 iova, unsigned int len) { - int ret = 0; + int i, ret = 0; if (!len) return 0; @@ -126,18 +93,39 @@ static int pfn_array_alloc_pin(struct pfn_array *pa, struct device *mdev, return -ENOMEM; pa->pa_pfn = pa->pa_iova_pfn + pa->pa_nr; - ret = pfn_array_pin(pa, mdev); + pa->pa_iova_pfn[0] = pa->pa_iova >> PAGE_SHIFT; + for (i = 1; i < pa->pa_nr; i++) + pa->pa_iova_pfn[i] = pa->pa_iova_pfn[i - 1] + 1; + + ret = vfio_pin_pages(mdev, pa->pa_iova_pfn, pa->pa_nr, + IOMMU_READ | IOMMU_WRITE, pa->pa_pfn); - if (ret > 0) - return ret; - else if (!ret) + if (ret < 0) { + goto err_out; + } else if (ret > 0 && ret != pa->pa_nr) { + vfio_unpin_pages(mdev, pa->pa_iova_pfn, ret); ret = -EINVAL; + goto err_out; + } + + return ret; +err_out: + pa->pa_nr = 0; kfree(pa->pa_iova_pfn); + pa->pa_iova_pfn = NULL; return ret; } +/* Unpin the pages before releasing the memory. */ +static void pfn_array_unpin_free(struct pfn_array *pa, struct device *mdev) +{ + vfio_unpin_pages(mdev, pa->pa_iova_pfn, pa->pa_nr); + pa->pa_nr = 0; + kfree(pa->pa_iova_pfn); +} + static int pfn_array_table_init(struct pfn_array_table *pat, int nr) { pat->pat_pa = kcalloc(nr, sizeof(*pat->pat_pa), GFP_KERNEL);
This refactors pfn_array_alloc_pin() and also improves it by adding defensive code in error handling so that calling pfn_array_unpin_free() after error return won't lead to problem. This mains does: 1. Merge pfn_array_pin() into pfn_array_alloc_pin(), since there is no other user of pfn_array_pin(). As a result, also remove kernel-doc for pfn_array_pin() and add kernel-doc for pfn_array_alloc_pin(). 2. For a vfio_pin_pages() failure, set pa->pa_nr to zero to indicate zero pages were pinned. 3. Set pa->pa_iova_pfn to NULL right after it was freed. Signed-off-by: Dong Jia Shi <bjsdjshi@linux.vnet.ibm.com> --- drivers/s390/cio/vfio_ccw_cp.c | 84 ++++++++++++++++++------------------------ 1 file changed, 36 insertions(+), 48 deletions(-)