<html xmlns:v="urn:schemas-microsoft-com:vml" xmlns:o="urn:schemas-microsoft-com:office:office" xmlns:w="urn:schemas-microsoft-com:office:word" xmlns:m="http://schemas.microsoft.com/office/2004/12/omml" xmlns="http://www.w3.org/TR/REC-html40"> <head> <meta http-equiv="Content-Type" content="text/html; charset=utf-8"> <meta name="Generator" content="Microsoft Word 15 (filtered medium)"> <style></style> </head> <body lang="EN-US" link="#467886" vlink="#96607D" style="word-wrap:break-word"> <div class="WordSection1"> <o:p> </o:p> <div> <div style="border:none;border-top:solid #E1E1E1 1.0pt;padding:3.0pt 0in 0in 0in"> From: Zhang, Jianxun <jianxun.zhang@intel.com> Sent: Tuesday, March 4, 2025 12:38 PM To: Cavitt, Jonathan <jonathan.cavitt@intel.com>; intel-xe@lists.freedesktop.org Cc: Gupta, saurabhg <saurabhg.gupta@intel.com>; Zuo, Alex <alex.zuo@intel.com>; joonas.lahtinen@linux.intel.com; Brost, Matthew <matthew.brost@intel.com>; Lin, Shuicheng <shuicheng.lin@intel.com>; dri-devel@lists.freedesktop.org Subject: Re: [PATCH v5 6/6] drm/xe/xe_vm: Implement xe_vm_get_property_ioctl<o:p></o:p> </div> </div> <div> <o:p> </o:p> </div> <div class="MsoNormal" align="center" style="text-align:center"> <hr size="2" width="98%" align="center"> </div> <div id="divRplyFwdMsg"> From: Cavitt, Jonathan <<a href="mailto:jonathan.cavitt@intel.com">jonathan.cavitt@intel.com</a>> Sent: Tuesday, March 4, 2025 9:08 AM To: <a href="mailto:intel-xe@lists.freedesktop.org">intel-xe@lists.freedesktop.org</a> <<a href="mailto:intel-xe@lists.freedesktop.org">intel-xe@lists.freedesktop.org</a>> Cc: Gupta, saurabhg <<a href="mailto:saurabhg.gupta@intel.com">saurabhg.gupta@intel.com</a>>; Zuo, Alex <<a href="mailto:alex.zuo@intel.com">alex.zuo@intel.com</a>>; Cavitt, Jonathan <<a href="mailto:jonathan.cavitt@intel.com">jonathan.cavitt@intel.com</a>>; <a href="mailto:joonas.lahtinen@linux.intel.com">joonas.lahtinen@linux.intel.com</a> <<a href="mailto:joonas.lahtinen@linux.intel.com">joonas.lahtinen@linux.intel.com</a>>; Brost, Matthew <<a href="mailto:matthew.brost@intel.com">matthew.brost@intel.com</a>>; Zhang, Jianxun <<a href="mailto:jianxun.zhang@intel.com">jianxun.zhang@intel.com</a>>; Lin, Shuicheng <<a href="mailto:shuicheng.lin@intel.com">shuicheng.lin@intel.com</a>>; <a href="mailto:dri-devel@lists.freedesktop.org">dri-devel@lists.freedesktop.org</a> <<a href="mailto:dri-devel@lists.freedesktop.org">dri-devel@lists.freedesktop.org</a>> Subject: [PATCH v5 6/6] drm/xe/xe_vm: Implement xe_vm_get_property_ioctl <o:p></o:p> <div> <o:p></o:p> </div> </div> <div> Add support for userspace to request a list of observed failed pagefaults from a specified VM. v2: - Only allow querying of failed pagefaults (Matt Brost) v3: - Remove unnecessary size parameter from helper function, as it is a property of the arguments. (jcavitt) - Remove unnecessary copy_from_user (Jainxun) - Set address_precision to 1 (Jainxun) - Report max size instead of dynamic size for memory allocation purposes. Total memory usage is reported separately. v4: - Return int from xe_vm_get_property_size (Shuicheng) - Fix memory leak (Shuicheng) - Remove unnecessary size variable (jcavitt) Signed-off-by: Jonathan Cavitt <<a href="mailto:jonathan.cavitt@intel.com">jonathan.cavitt@intel.com</a>> Suggested-by: Matthew Brost <<a href="mailto:matthew.brost@intel.com">matthew.brost@intel.com</a>> CC: Jainxun Zhang <<a href="mailto:jianxun.zhang@intel.com">jianxun.zhang@intel.com</a>> CC: Shuicheng Lin <<a href="mailto:shuicheng.lin@intel.com">shuicheng.lin@intel.com</a>> --- drivers/gpu/drm/xe/xe_device.c | 3 ++ drivers/gpu/drm/xe/xe_vm.c | 75 ++++++++++++++++++++++++++++++++++ drivers/gpu/drm/xe/xe_vm.h | 2 + 3 files changed, 80 insertions(+) diff --git a/drivers/gpu/drm/xe/xe_device.c b/drivers/gpu/drm/xe/xe_device.c index 9454b51f7ad8..43accae152ff 100644 --- a/drivers/gpu/drm/xe/xe_device.c +++ b/drivers/gpu/drm/xe/xe_device.c @@ -193,6 +193,9 @@ static const struct drm_ioctl_desc xe_ioctls[] = { DRM_IOCTL_DEF_DRV(XE_WAIT_USER_FENCE, xe_wait_user_fence_ioctl, DRM_RENDER_ALLOW), DRM_IOCTL_DEF_DRV(XE_OBSERVATION, xe_observation_ioctl, DRM_RENDER_ALLOW), + DRM_IOCTL_DEF_DRV(XE_VM_GET_PROPERTY, xe_vm_get_property_ioctl, + DRM_RENDER_ALLOW), + }; static long xe_drm_ioctl(struct file *file, unsigned int cmd, unsigned long arg) diff --git a/drivers/gpu/drm/xe/xe_vm.c b/drivers/gpu/drm/xe/xe_vm.c index 6211b971bbbd..1e78103fb0f6 100644 --- a/drivers/gpu/drm/xe/xe_vm.c +++ b/drivers/gpu/drm/xe/xe_vm.c @@ -3234,6 +3234,81 @@ int xe_vm_bind_ioctl(struct drm_device *dev, void *data, struct drm_file *file) return err; } +static int xe_vm_get_property_size(struct xe_vm *vm, u32 property) +{ + switch (property) { + case DRM_XE_VM_GET_PROPERTY_FAULTS: + return MAX_PFS * sizeof(struct drm_xe_pf);<o:p></o:p> <o:p> </o:p> </div> <div> > I still think returning the number of faults back to user space is better and clearer.<o:p></o:p> > But the intention is to use this ioctl on any properties, so perhaps it is why a size in byte is chosen.<o:p></o:p> <o:p> </o:p> This is a workaround for the issue you pointed out in the previous revision that caused the get<o:p></o:p> property ioctl to fail if the number of faults in the system changed between the first and second<o:p></o:p> calls. Getting the precise size of the data field would reintroduce that issue, and keeping the lock<o:p></o:p> between the first and second calls to prevent the list from changing between calls would violate<o:p></o:p> several design principles (E.G., what happens if the user only calls the ioctl once?)<o:p></o:p> <o:p> </o:p> </div> <div> ><o:p> </o:p> </div> <div> > But first of all, should we regard faults as a property?<o:p></o:p> <o:p> </o:p> I think this question would be better answered by Matthew Brost, as this was at least<o:p></o:p> in part his suggestion. Though I suppose we could make this an extension of xe_query<o:p></o:p> and just ask userspace to pass in additional information W.R.T. the VM ID as a part of<o:p></o:p> the query structure. On the other hand, we’ll probably want the property ioctl extended<o:p></o:p> in the future to be able to report additional information, which has no precedence in<o:p></o:p> the query ioctl and thus would bloat the query ioctl with additional queries.<o:p></o:p> </div> <div> + default: + return -EINVAL; + } +} + +static int fill_property_pfs(struct xe_vm *vm, + struct drm_xe_vm_get_property *args) +{ + struct drm_xe_pf __user *usr_ptr = u64_to_user_ptr(args->ptr); + struct drm_xe_pf *fault_list; + struct drm_xe_pf *fault; + struct xe_vm_pf_entry *entry; + int ret, i = 0; + + fault_list = kzalloc(args->size, GFP_KERNEL); + if (!fault_list) + return -ENOMEM; + + spin_lock(&vm->pfs.lock); + list_for_each_entry(entry, &vm->pfs.list, list) { + struct xe_pagefault *pf = entry->pf; + + fault = &fault_list[i++]; + fault->address = pf->page_addr; + fault->address_type = pf->address_type; + fault->address_precision = 1; + } + args->value = vm->pfs.len; + spin_unlock(&vm->pfs.lock); + + ret = copy_to_user(usr_ptr, &fault_list, args->size); + kfree(fault_list); + + return ret ? -EFAULT : 0; +}<o:p></o:p> <o:p> </o:p> <o:p> </o:p> </div> <div> > How about this to get rid of fault_list? The assumption is put_user is sufficient on types of data to return.<o:p></o:p> > It also deletes the copied data though KMD may still want to keep them.<o:p></o:p> </div> <div> > <o:p></o:p> </div> <div> > This is Just an idea, It makes copied size is unclear so more refactor would be needed.)<o:p></o:p> </div> <div> > <o:p></o:p> </div> <div> > static int fill_property_pfs(struct xe_vm *vm,<o:p></o:p> </div> <div> > struct drm_xe_vm_get_property *args)<o:p></o:p> </div> <div> > {<o:p></o:p> </div> <div> > struct drm_xe_pf __user *usr_ptr = u64_to_user_ptr(args->ptr);<o:p></o:p> </div> <div> > struct xe_vm_pf_entry *entry;<o:p></o:p> </div> <div> > struct xe_vm_pf_entry *tmp_entry;<o:p></o:p> </div> <div> > int ret = 0;<o:p></o:p> </div> <div> ><o:p> </o:p> </div> <div> > spin_lock(&vm->pfs.lock);<o:p></o:p> </div> <div> > list_for_each_entry_safe(entry, tmp_entry, &vm->pfs.list, list) {<o:p></o:p> </div> <div> > ret = put_user(entry->pf->page_addr, &usr_ptr->address);<o:p></o:p> </div> <div> > if (ret)<o:p></o:p> </div> <div> > break;<o:p></o:p> </div> <div> > ret = put_user(entry->pf->address_type, &usr_ptr->address_type);<o:p></o:p> </div> <div> > if (ret)<o:p></o:p> </div> <div> > break;<o:p></o:p> </div> <div> > ret = put_user(1, &usr_ptr->address_precision);<o:p></o:p> </div> <div> > if (ret)<o:p></o:p> </div> <div> > break;<o:p></o:p> </div> <div> > usr_ptr++;<o:p></o:p> </div> <div> > list_del(entry->list);<o:p></o:p> </div> <div> > }<o:p></o:p> </div> <div> > spin_unlock(&vm->pfs.lock);<o:p></o:p> </div> <div> ><o:p> </o:p> </div> <div> > return ret;<o:p></o:p> </div> <div> > }<o:p></o:p> <o:p> </o:p> I’ll consider it once we can get a finalization on whether this needs to be a separate ioctl<o:p></o:p> or an extension of xe_query. Though either way I’ll have to pass on clearing the list on query.<o:p></o:p> The information should stay persistent until the VM is closed or until it’s lost due to hitting<o:p></o:p> the list size limit.<o:p></o:p> -Jonathan Cavitt<o:p></o:p> </div> <div> + +int xe_vm_get_property_ioctl(struct drm_device *drm, void *data, + struct drm_file *file) +{ + struct xe_device *xe = to_xe_device(drm); + struct xe_file *xef = to_xe_file(file); + struct drm_xe_vm_get_property *args = data; + struct xe_vm *vm; + int size; + + if (XE_IOCTL_DBG(xe, args->reserved[0] || args->reserved[1])) + return -EINVAL; + + vm = xe_vm_lookup(xef, args->vm_id); + if (XE_IOCTL_DBG(xe, !vm)) + return -ENOENT; + + size = xe_vm_get_property_size(vm, args->property); + if (size < 0) { + return size; + } else if (args->size != size) { + if (args->size) + return -EINVAL; + args->size = size; + return 0; + } + + switch (args->property) { + case DRM_XE_VM_GET_PROPERTY_FAULTS: + return fill_property_pfs(vm, args); + default: + return -EINVAL; + } +} + /** * xe_vm_bind_kernel_bo - bind a kernel BO to a VM * @vm: VM to bind the BO to diff --git a/drivers/gpu/drm/xe/xe_vm.h b/drivers/gpu/drm/xe/xe_vm.h index 4d94ab5c8ea4..bf6604465aa3 100644 --- a/drivers/gpu/drm/xe/xe_vm.h +++ b/drivers/gpu/drm/xe/xe_vm.h @@ -184,6 +184,8 @@ int xe_vm_destroy_ioctl(struct drm_device *dev, void *data, struct drm_file *file); int xe_vm_bind_ioctl(struct drm_device *dev, void *data, struct drm_file *file); +int xe_vm_get_property_ioctl(struct drm_device *dev, void *data, + struct drm_file *file); void xe_vm_close_and_put(struct xe_vm *vm); -- 2.43.0<o:p></o:p> </div> </div> </body> </html>