From mboxrd@z Thu Jan 1 00:00:00 1970 Received: from outbound.baidu.com (mx24.baidu.com [111.206.215.185]) by smtp.subspace.kernel.org (Postfix) with SMTP id 2F485538D85; Tue, 8 Sep 2026 12:10:16 +0000 (UTC) Authentication-Results: smtp.subspace.kernel.org; arc=none smtp.client-ip=111.206.215.185 ARC-Seal:i=1; a=rsa-sha256; d=subspace.kernel.org; s=arc-20240116; t=1788869436; cv=none; b=DqO/PITgUpmYfvkQ1c/n+QlBG3qD9G6VHnp3eV45q2QN6I3zeknrEy4NLMhAmT93cVGJlwH5WPOsDJmW3tjdY8J8YNCxaBglDkkqp4n8LiMpaK6bsoaDKmzaWUuFdb6QYD7wM2FAEam7x4H0tc1XoBrZd/5j2ywUWEEXh9U10rs= ARC-Message-Signature:i=1; a=rsa-sha256; d=subspace.kernel.org; s=arc-20240116; t=1788869436; c=relaxed/simple; bh=kiBE1+g76z4qwzHwHU+V4b7I2l+21nG8jl2+rGPzT8c=; h=From:To:CC:Subject:Date:Message-ID:MIME-Version:Content-Type; b=NymHPVkzXIPc5fu5++aP+NSuFXGdJExgD0jaHSPrrwMAhvmMgLp8G/egLAft8ytE0LzAT9xZmmo4RGdy+Yi/WG72GDQsmpDQWkQWZlaWguDLmB73sowuLlmSlUDPXoZyHYJ0QLZbDZ5rPbqIDbKAt+f0B92AvIc9LaZI8bbERRw= ARC-Authentication-Results:i=1; smtp.subspace.kernel.org; dmarc=pass (p=quarantine dis=none) header.from=baidu.com; spf=pass smtp.mailfrom=baidu.com; dkim=pass (2048-bit key) header.d=baidu.com header.i=@baidu.com header.b=ixmrfraU; arc=none smtp.client-ip=111.206.215.185 Authentication-Results: smtp.subspace.kernel.org; dmarc=pass (p=quarantine dis=none) header.from=baidu.com Authentication-Results: smtp.subspace.kernel.org; spf=pass smtp.mailfrom=baidu.com Authentication-Results: smtp.subspace.kernel.org; dkim=pass (2048-bit key) header.d=baidu.com header.i=@baidu.com header.b="ixmrfraU" X-MD-Sfrom: lirongqing@baidu.com X-MD-SrcIP: 172.31.50.47 From: lirongqing To: Paolo Bonzini , Sean Christopherson , , CC: Li RongQing Subject: [PATCH] KVM: irqchip: allocate routing entries in chunks Date: Tue, 8 Sep 2026 20:01:33 +0800 Message-ID: <20260908120133.2381-1-lirongqing@baidu.com> X-Mailer: git-send-email 2.17.1 Precedence: bulk X-Mailing-List: kvm@vger.kernel.org List-Id: List-Subscribe: List-Unsubscribe: MIME-Version: 1.0 Content-Type: text/plain X-ClientProxiedBy: bjkjy-exc6.internal.baidu.com (172.31.50.50) To bjkjy-exc3.internal.baidu.com (172.31.50.47) DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=baidu.com; s=selector1; t=1788868908; bh=2xDETEACWh44ozaK44g8QqrbA8/wL3drFTvFddCxvNg=; h=From:To:CC:Subject:Date:Message-ID:Content-Type; b=ixmrfraUUjUBCru0y4x5Cvf5tknI2ABrUwcfCoyTG0ADp+DgqHb4anlqd8GtWyM2f q7C0+3iB54dLvBTq1nVUu84Xax5vXbZ2Gl/zskETRY56GyiibGGeBz7wliI/r++Suo V7HnZ+/OGUtGAYN15V3XuOowDRd3aO9kgiV0hQPwiA5e3zTeZBY8Q+XDWVV8Zymidh 9vnRDWSNxF8vvaOaa7HRR/oD8vPeE5dN204fZbRxDnuIxl3m+RvMDLultG83cW2T6H 9o3dq0N3QVAZinTSuwlcES076IAw/QUuMKj8kzFoibnbvRWwLHJHT06D4RuIhoeJH0 JodYb1f0ph2dA== From: Li RongQing kvm_set_irq_routing() allocates each routing entry separately, so a routing table with thousands of GSIs needs thousands of small allocations and frees, adding significant allocator overhead. Allocate the entries in chunks instead: each chunk holds up to PAGE_SIZE / sizeof(struct kvm_kernel_irq_routing_entry) entries, and the chunk pointers are kept in the routing table so that all entries are freed together when the table is released. Each chunk is capped at PAGE_SIZE instead of allocating one array for the whole table: with nr up to KVM_MAX_IRQ_ROUTES (4096), a single array would be a multi-page contiguous request, which is what tends to fail once memory is fragmented. Page-sized chunks stay on the normal kmalloc path, and a failed allocation only costs one chunk. The chunk pointer array uses kvzalloc_objs() and can fall back to vmalloc. The last chunk is sized to the number of entries actually left, so a table smaller than one chunk - the common case - allocates only what it needs. Measured with an eBPF probe on kvm_set_irq_routing() on an Intel EMR CPU: when a VM has a 2000+ entry routing table, the time spent in the function drops from about 700us to about 300us. Signed-off-by: Li RongQing --- include/linux/kvm_host.h | 2 ++ virt/kvm/irqchip.c | 62 +++++++++++++++++++++++++++++++++++++++--------- 2 files changed, 53 insertions(+), 11 deletions(-) diff --git a/include/linux/kvm_host.h b/include/linux/kvm_host.h index 03bfc92..83848d7 100644 --- a/include/linux/kvm_host.h +++ b/include/linux/kvm_host.h @@ -693,6 +693,8 @@ struct kvm_kernel_irq_routing_entry { struct kvm_irq_routing_table { int chip[KVM_NR_IRQCHIPS][KVM_IRQCHIP_NUM_PINS]; u32 nr_rt_entries; + u32 nr_entry_chunks; + struct kvm_kernel_irq_routing_entry **entry_chunks; /* * Array indexed by gsi. Each entry contains list of irq chips * the gsi is connected to. diff --git a/virt/kvm/irqchip.c b/virt/kvm/irqchip.c index 462c706..044b831 100644 --- a/virt/kvm/irqchip.c +++ b/virt/kvm/irqchip.c @@ -18,6 +18,9 @@ #include #include +#define KVM_IRQ_ROUTING_ENTRIES_PER_CHUNK \ + (PAGE_SIZE / sizeof(struct kvm_kernel_irq_routing_entry)) + int kvm_irq_map_gsi(struct kvm *kvm, struct kvm_kernel_irq_routing_entry *entries, int gsi) { @@ -107,12 +110,14 @@ static void free_irq_routing_table(struct kvm_irq_routing_table *rt) struct kvm_kernel_irq_routing_entry *e; struct hlist_node *n; - hlist_for_each_entry_safe(e, n, &rt->map[i], link) { + hlist_for_each_entry_safe(e, n, &rt->map[i], link) hlist_del(&e->link); - kfree(e); - } } + for (i = 0; i < rt->nr_entry_chunks; ++i) + kfree(rt->entry_chunks[i]); + kvfree(rt->entry_chunks); + kfree(rt); } @@ -170,9 +175,11 @@ int kvm_set_irq_routing(struct kvm *kvm, unsigned nr, unsigned flags) { + struct kvm_kernel_irq_routing_entry **chunks = NULL; struct kvm_irq_routing_table *new, *old; struct kvm_kernel_irq_routing_entry *e; u32 i, j, nr_rt_entries = 0; + u32 nr_chunks; int r; for (i = 0; i < nr; ++i) { @@ -183,6 +190,13 @@ int kvm_set_irq_routing(struct kvm *kvm, nr_rt_entries += 1; + /* + * The chunks hold the routing entries, so they are sized by the number + * of entries passed in by the caller, not by nr_rt_entries, which is + * the size of the GSI map. + */ + nr_chunks = DIV_ROUND_UP(nr, KVM_IRQ_ROUTING_ENTRIES_PER_CHUNK); + new = kzalloc_flex(*new, map, nr_rt_entries, GFP_KERNEL_ACCOUNT); if (!new) return -ENOMEM; @@ -192,26 +206,54 @@ int kvm_set_irq_routing(struct kvm *kvm, for (j = 0; j < KVM_IRQCHIP_NUM_PINS; j++) new->chip[i][j] = -1; + r = -ENOMEM; + if (nr_chunks) { + chunks = kvzalloc_objs(*chunks, nr_chunks, GFP_KERNEL_ACCOUNT); + if (!chunks) + goto out; + + new->entry_chunks = chunks; + new->nr_entry_chunks = nr_chunks; + } + for (i = 0; i < nr; ++i) { + u32 idx = i / KVM_IRQ_ROUTING_ENTRIES_PER_CHUNK; + u32 off = i % KVM_IRQ_ROUTING_ENTRIES_PER_CHUNK; + r = -ENOMEM; - e = kzalloc_obj(*e, GFP_KERNEL_ACCOUNT); - if (!e) - goto out; + if (!chunks[idx]) { + struct kvm_kernel_irq_routing_entry *chunk; + /* + * A chunk is only entered at its first entry, so nr - i + * is the number of entries left for this chunk; the last + * chunk is short. + */ + u32 cnt = min_t(u32, nr - i, + KVM_IRQ_ROUTING_ENTRIES_PER_CHUNK); + + chunk = kzalloc_objs(*chunk, cnt, GFP_KERNEL_ACCOUNT); + if (!chunk) + goto out; + + chunks[idx] = chunk; + } + + e = chunks[idx] + off; r = -EINVAL; switch (ue->type) { case KVM_IRQ_ROUTING_MSI: if (ue->flags & ~KVM_MSI_VALID_DEVID) - goto free_entry; + goto out; break; default: if (ue->flags) - goto free_entry; + goto out; break; } r = setup_routing_entry(kvm, new, e, ue); if (r) - goto free_entry; + goto out; ++ue; } @@ -228,8 +270,6 @@ int kvm_set_irq_routing(struct kvm *kvm, r = 0; goto out; -free_entry: - kfree(e); out: free_irq_routing_table(new); -- 2.9.4